ElectricSystems,Dynamics,

andStabilitywithArtificial
IntelligenceApplications
JarnesA.Mornoh
Howard University
Washington, D.C.
MohamedE.El-Hawary
DalhousieUniversity
Hali&r,NovaScotia,Canada
M A R C E L
m
MARCEL INC. NEWYORK DEKKER, BASEL
D E K K E R
ISBN:0-8247-0233-6
Thisbook is printed on acid-freepaper.
Headquarters
Marcel Dehher, Inc.
770 Madison A\enue, N ~ N Yorh, NY 10016
tel:212-696-9000: f ay: 212-685-35-10
Eastern Hemisphere Distribution
Marcel Dekker AG
Hutga\w -1. Postfach 8I ? . CH--1001Bawl. Switzerland
tel: -11-61-261-8482; f ax: 31-61-261-8896
World Wide Web
http : / / w m~ ~ .dekker.coiii
The publisher offers discounts on this book when ordered i n bulk quantities. For more
information. writc to Special Sales/Profehsional Marketing at the headquarters d d r i : ~ ~
;I bove.
Copyright 02000 by Marcel Dekker, Inc.All Rights Reserved.
Neither thih book nor any part may be reproduced or transmitted i n any formor by i111>'
iiieans, electronic or mechanical, including photocopying, microfilming. and recording,
or by any information storage and retriet.al system, without permission in ~ ~ i t i n g f r o i n
the publisher.
Current printing (lastdigit):
1 0 9 8 7 6 5 4 3 2 I
PRINTED IN THE UNITEDSTATESOFAMERICA
To Felicia, forthe amazing grace,and Omeiza,
Ojirese, Eshovo, andToyin,joy of life.
-JAM
To Ferial, thanks for your company.
Bob,Ron, and Betty, you make me happy.
-MEE
This page intentionally left blank
SeriesIntroduction
Powerengineeringis the oldestandmost traditionalof the various areas within
electrical engineering,yet nootherfacet of modern technologyis currently un-
dergoing a more dramaticrevolutionin both technologyand industrystructure.
Among the most excitingchanges arethose where new solutionsare being ap-
plied to classicalproblemareas.
System dynamics and stability engineering have been crucial elements of
powersystemengineeringsinceearlyin thetwentiethcentury.Smooth,continu-
ousoperationof modern powersupplysystemsdependsgreatlyon the accurate
anticipationof interconnectedequipment,dynamicbehavior,and correctidenti-
fication of the systems operating limits. Proper engineering requires precise
methodsthatcanmanagethat knowledgeanddirecti t tothedesignof economi-
calandsecurepowersystems.Artificialintelligenceoffersan excitingnew basis
for performing dynamic modeling and stability analysis, one that can provide
considerable value and new insight to these often difficult aspects of power
system performance.
Electric Systems,Dynamics,and StabilivwithArtificial IntelligenceAppli-
cations isanexceedinglycomprehensiveandpracticalguidetobothpowersys-
tem dynamicsand stabilityconcepts, and to the use of artificial intelligencein
theirmodelingandengineering.Drs.MomohandEl-Hawaryprovideacompre-
hensiveintroductiontopower systemdynamicsand stability,alongwith athor-
ough discussion of recently developedconcepts such as transient energy func-
tions. Theirbook is rich i n its appreciation of the intricate operating constraints
and issuesthat real-world power systemengineers andoperators must faceevery
day. But wtiat sets this book apart is its application of artificial intelligence to
these long-recognized power system engineering challenges.Chapters 7-9 ex-
pIain ho 11ei i raI tietuwrks, espert syste111s usiingknowIedge-based fraineu'orks.
and t ' u~zylogic can be applied to the solution of sotlieof the thorniest problems
in power system dynamics.
Like all books in Marcel Dekker's Power Engineering series,Elecstr-ic.SJ.Y-
tettr. v, ~>j wt t i i c *s , App/ic.trliotis present? t i t i d SttihilitjvIl'itli Arlificitrl Itit~~ligutic*ci
tnodern power technology i n ;I context of proven. practical applications: iiset'u
;is a reference book iis well iis forself-study andclassroom use.Marcel Dekker':,
Pourer Engineering series bille\~entually include booksco\w-ingthe entire field
of ponwengineering,i nallot'itsspecialties and sub-genres,allainned atprovid-
ing practicing pouwengineersu,iththe knou~ledgeand techniyiies they need to
tneet the electric industrlf'schallenges i n the twenty-first century.
Preface
Theintention of thisbookistoofferthereader afirmfoundationforunderstand-
ing and analyzing power system dynamics and stability problems as \yell asthe
application of artificial intelligence technology to these problems. Issues i n this
area are extremely important not only for real-time operational considerations,
but also in planning, design, and operational scheduling. The significance of
dynamics and stability studies grows as interconnected systems evolve to meet
the requirements of acompetitive andderegulated operational environment. The
complexities introduced give rise to new types of control strategies based on
advances i n modeling and simulation of the power system.
Thematerial presented in this book combines the experience of the authors
i n teaching and research at a number of schools and professional developtnent
venues. The work reported here draws on experience gained in conducting re-
search sponsored bythe Electric PowerResearch Institute. the National Science
Foundation, the Department of Energy, and NASA for Dr. Momoh. Dr. EI-
Hawary's work was supported by the Natural Sciences and Engineering Rc-
search Council of Canada and Canadian Utilities Funding.
This book is intended to meet the needs of practicing engineers invol\.edi n
the electric power utility business, aswell as graduate students and researchers.
I t provides necessary fundamentals, by explaining the practical aspects of artifi-
cial intelligence applications and offering an integrated treatment of the evolu-
tion of modeling techniques and analytical tools.
Chapter I discusses the structure ofinterconnected power systems, founda-
tions of system dynamics,and definitions for stability and security assessment.
Chapter 2 deals with static electric network models and synchronous machine
representation and itsdynamics. Limits foroperationsof asynchronous machine
and static load models are discussed as well. Chapter 3 deals with dynamic
models of the electric network including the excitation. and prime mover and
governing system models. Thechapter concludes with a discussion of dynamic
load models.
Chapter 4 covers concepts of dynamic security assessment based on tran-
sient stability evaluation. Thischapter includes both conventional and extended
formulations of the problem. Chapter 5, a complement to Chapter 4, treats the
more recent approachof anglestability assessment viathetransient energy func-
tion idea. Chapter 6 introduces the idea of voltage stability and discusses tech-
niques for its assessment.
Chapters 7 through 9 are devoted to an expose of artificial intelligence
technology and its application to problems of system stability, from both the
angle and the voltage sides. In Chapter 7,we introduce basic concepts of artifi-
cial neural networks, knowledge-based systems, and fuzzy logic. In Chapter 8.
we deal with theapplication of artificial intelligence to angle stability problems.
and the extension to voltage stability is presented in Chapter 9. Chapter I C
offers conclusions and directions for future work in this field.
I n developing this book. we have benefited from input from many of 0111'
students.colleagues, and associates. While they are too many tocount, we wish
totnention specifically encouragement by H.LeeWillis,theeditor of the Powei-
Engineering Series for Marcel Dekker, Inc. The continual counsel and prodding
of B. J. Clark was extremely helpful. We acknowledge the able administrative
support of Linda Schonberg and the assistance of our respective deans.
Wearegrateful toDr.Chieh forthegreat inspiration andgenerouscontribu-
tions, and to many others, whose names are not included, i n the development
ot'this volume. Our students, both present and former. contributed their time
and many valuable suggestions. Many thanks to them and especially to the
young research assistants at the Center for Energy Systems and Control for
putting up with the burdensome challenge of producing this bookjust in time.
Finally. the book would not have been published without the help of our
Creator and the support of our families.
Contents
Series 1titi.odiic.tiotz H.LeeWillis
Prefiice
1 Introduction
1 . 1 HistoricalBackground
1.2 Structureat aGenericElectricPower System
I .3 Power SystemSecurityAssessment
2 StaticElectricNetwork Models
10
2.1 ComplexPowerConcepts
1 1
2.2 Three-PhaseSystems
14
2.3 SynchronousMachineModeling
21
2.4 ReactiveCapabilityLimits
31
2.5 StaticLoad Models
32
Introduction
10
Conclusions
35
3 DynamicElectricNetworkModels
36
Introduction
36
3.1 ExcitationSystemModel
36
3.2 Prime Moverand GoverningSystemModels
40
3.3 Modeling of Loads
COnc1LIsi ons
4 Philosophyof SecurityAssessment
Introduction
4.1 The Swing Equation
4.2 Some Alternative Forms
4.3 Transient and Subtransient Reactances
4.4 Synchronous Machine Model i n Stability Analysis
4.5 SUbtransient Equations
4.6 Machine Models
4.7 Groups of Machines and the Infinite Bus
4.8 Stability Assessment
4.9 Concepts in Transient Stability
4.10 A Method for Stability Assessment
4.11 Matheinatical Models and Solution Methods i n Transient
StabiIity Assessment for General Networks
4.12 Integration Techniques
4.13 The Transient Stability Algorithm
Conc1usions
S AssessingAngle Stability via TransientEnergy Function
Introduction
5. I StabilityConcepts
5.2 System Model Description
5.3 Stability of a Single-Machine System
5.4 Stability Assessment for ri-Generator System by the
TEFMethod
5.5 Application to ;I Practical Power System
5.6 Boundary of the Region of Stability
Conclusion
6 VoltageStability Assessment
Introduction
6.I Worhing Definition of Voltage Collap\e Study Terms
6.2 Typical Scenarioof Voltage Collapse
6.3 Time-Frame Voltage Stability
6.4 Modeling for Voltage Stability Studie\
6.5 Voltage Collapse Prediction Methods
6.6 Clas\ification c)f Vo1tage Stab i1i t y ProbIe111s
6.7 Voltage Stability As\es\ment Techniques
Col1 ter 1f.\
xi
6.8 Analysis Techniques for Steady-State Voltage
Stability Studies 135
6.9 Parameterization
IS 1
6.10 The Technique of Modal Analysis 156
6.11 Analysis Techniques for Dynamic Voltage Stability Studies 157
Conclusion 169
Modal Analysis: Worked Example 170
7 Technologyof IntelligentSystems 175
Introduction 175
7.1 Fuzzy Logic and Decision Trees 177
7.2 Artificial Neural Networks 177
7.3 Robust Artificial Neural Network 183
7.4 Expert Systems 191
7.5 Fuzzy Setsand Systems 206
7.6 Expert Reasoning and Approximate Reasoning 213
Conclusion 220
8 Applicationof ArtificialIntelligenceto AngleStabilityStudies 221
Introduction 22I
3 3 3
eh& 8. I ANN Application i n Transient Stability Assessment
8.2 A Knowledge-Based System for Direct Stability Analysis 238
Conclusions 257
9 Applicationof ArtificialIntelligenceto VoltageStability
Assessmentand EnhancementtoElectricalPowerSystems 259
Introduction 259
9.1 ANN-Based Voltage Stability Assessment 260
9.2 ANN-Based Voltage Stability Enhancement 265
9.3 A Knowledge-Based Support System for Voltage
Collapse Detection and Prevention 272
9.4 Implementation for KBVCDP 278
9.5 Utility Environment Application 287
Conclusion 287
10 Epilogueand Conclusions 289
298
31I
332
35I
This page intentionally left blank
ElectricSystems,Dynamics,
andStabilitywithArtificial
IntelligenceApplications
This page intentionally left blank
1
Introduction
1.1 HISTORICALBACKGROUND
Electric power has shaped and contributed to the progress and technological
advances of humans over the past century. I t is not surprising then that the
growth of electric energy consumption in the world has been nothing but phe-
nomenal. I n the United States, forexample, electric energy sales have grown to
well over 400times in the period between the turn of thecentury and the early
1970s. This growth rate was 50times as much as the growth rate i n all other
energy forms used during the same period.
Edison Electric of New York pioneered the central station electric pouter
generation by opening of the Pearl Street station in 1881.This station had a
capacity of four250-hp boilers supplying steamto sixengine-dynamosets.Edi-
son's system used a I 10-dcunderground distribution network with copper con-
ductorsinsulated with ajute wrapping.Thel o b i t \ dt ci geof tlic-, cii-mitlimited the
service area of a central station, and consequently central stations proliferated
throughout metropolitan areas.
The invention of the transformer, then known asthe "inductorium." made
ac systems possible. The first practical ac distribution system in the United
States was installed by W.Stanley at Great Barrington, Massachusetts. i n 1866
forWestinghouse, whoacquired the American rights to thetransformer from its
British investorsGaulard and Gibbs.Early acdistribution utilized 1000V over-
head lines.
By 1895, Philadelphia had about twenty electric companies with distribu-
tion systems operating at 100V and 500 V two-wire dc and 220 V three-wire
dc: single-phase, two-phase, and three-phase ac;with frequenciesof 60,66. 125,
and 133cycles per second;and feeders at 1000-1200 V and 2000-2300 V.
The consolidation of electric companies enabled the realization of econo-
mies o f scale in generating facilities. the introduction of a certain degree of
equipment standardization, and the utilization of the load diversity between
areas. Generating uni t sizes of up to 1300MW are i n service, an era that w;i?
started in I973 by the Cumberland Station of the Tennessee Valley Authority.
Underground distribution of Lwltages up to 5 kV was made possible by thc
de\.elopment of rubber-base insulated cables and paper insulated, lead-co\.erec
cables in the early 1900s.Since that time higher distribution voltages hi1L.e beer
necessitated by load growth that would otherwise overload low-voltage circuit!,
and by the requirement to transmit large blocks of power over great distances.
Coninion distribution voltages in today's systems are in 5 , IS, 25, 35. and 6'1
kV \vltage classes.
The growth in size of power plants and in the higher voltage equipment
u ' a s accompanied by interconnections of the generating facilities. These inter-
connections decreased the probability of service interruptions, made the utiliza-
tion of the most economical units possible, and decreased the total reserve ca-
pacity required to meet equipment-forced outages. This growth was ;11so
accompanied by the use of sophisticated analytical tools. Central control of thc
interconnected systems was introduced for reasons of economy and safety.Th,:
advent of the load dispatcher heralded the dawn of power systems engineering.
\+!hose objective is to provide the best system to meet the load demand reliablq.
safe1y,and economical1y,Ut i 1izing state-of-the-art computer faci1ities.
Extra high voltage (EHV)has become the dominant factor in the transmi>-
sion of electric power over great distances. By 1896,an 1 1 kV three-phase line
M . ~ Stransmitting 10 MW from Niagara Falls to Buffalo over a distance of 2 3
miles. Today, transmission voltages of 230 kV, 287 kV, 345 kV, 500 kV, 735
kV, and 765 kV are commonplace, with the first I100 kV line scheduled fclr
energization i n the early 1990s. One prototype is the I200 kV transmission
tower. The trend is possible. resulting i n more efficient use of right-of-wa;!.
lower transmission losses, and reduced environmental impact.
Thepreference for acwas first challenged in 1954when the Swedish Stale
Power Board energized the 60-mile, 100 kV dc submarine cable utilizing I!.
Larnm's Mercury Arc FAves at the sending and receiving ends of the world s
first high-voltage direct current (HVDC)l i nk connecting the Baltic island o f
Gotland and theSwedish mainland.Today, numerous installations with voltages
up to 800kV dchave become operational around theglobe. Solid-state technol-
ogy advances havealsoenabled theuseofthesilicon-controlledrectifiers (SCR)
of thyristor for HVDC applications since the late 1960s.Whenever cable trans-
mission isrequired (underwater orin a metropolitan area),HVDC is more eco-
nomically attractive than ac.
Protecting isolated systems has been a relatively simple task, which iscar-
ried out using overcurrent directional relays with selectivity being obtained by
time grading. High-speed relays have been developed to meet the increased
short-circuit currents due to the larger size units and the complex interconnec-
tions.
For reliable service, an electric power system must remain intact and be
capable of withstanding a wide variety of disturbances. It is essential that the
system be operated so that the more probable contingencies can be sustained
without loss of load (except that connected to the faulted element)and so that
the most adversepossible contingencies donot result in widespread and cascad-
ing power interruptions.
The November 1965blackout in the northeastern part of the United States
and Ontario had a profound impact on the electric utility industry. Many ques-
tions were raised and led to the formation of the National Electric Reliability
Council in 1968.The name was later changed to the North American Electric
Reliability Council (NERC).Its purpose is to augment the reliability and ade-
quacy of bulk power supply in theelectricity systemsof North America. NERC
is composed of nine regional reliability councils and encompasses \Tirtually all
the power systems in the United States and Canada.Reliability criteria for sys-
tem design and operation have been established by each regional council. Since
differences exist in geography, loadpattern, and power sources, criteria for the
various regions differ to some extent.
Design and operating criteria play an essential role in preventing major
system disturbances following severe contingencies. The use of criteria ensures
that, forall frequently occurring contingencies, the system will, at worst, trans-
mit from the normal state to the alert state, rather than to a more severe state
such as the emergency state or the in extreenzis state. When the alert state is
entered following a contingency, operators can take action to return the system
to the normal state.
1.2 STRUCTUREOF A GENERICELECTRIC POWERSYSTEM
While no two electric power systems are alike, all share some common funda-
mental characteristics including:
1 . Electric powerisgenerated using synchronous machinesthat aredriven
by turbines (steam,hydraulic, diesel, orinternal combustion).
2. Generated power istransmitted from thegenerating siteso\rer longdis-
tances to load centers that are spread over wide areas.
3. Three phase ac systems comprise the main means of generation. trans-
mission and distribution of electric power.
4. Voltage and frequency levels are required to remain within tight toler-
ance lekrels to assure a high quality product.
The basic elements of ii generic electric poww system are displayed i n
Figure I . 1 . Electric power is produced at generating stations(GS)and transmit-
ted to consumers through an intricate network of apparatus including transmis-
sion lines,transformers. and switching devices.
Transmission network is classified as the folloufing:
I . Transmission system
2. Si i btransmission system
3. Distribution system
,s~~.strin The t i . i ~i i ~s i ~i i . s . s i ~) i i interconnects aI1 major generating st;i tic)ns ;I nd
main load centers in the system. I t forms the backbone of the integrated p o \ \ ~r
system and operates at the highest voltage levels (typically,230 kV and ab0F.e1 .
The generator \vltagesare usually i n the range of 11-35 kV. These are steppej
up to the transmission cultage level, and power is transmitted to transmission
hubstations where the voltages are stepped down to the subtransmission I t . \ , c. l
(typically,69 kV to I38 kV) . The generation and transmission subsystems ;ite
often referred to as the hi i l k po\r.ei.systerii.
The . viihtr. iiri. siiii, s. siori sy.stew transmits power at a lower voltage and i n
sm ii1ler qitantities froi n the transiiiission si i bstation to the distribi i tion subst;I-
tions. Large industrial customers are commonly supplied directly from the s u b
transmission system. I n somesystems,there expansion and higher \,ohagel e \ ~ l s
becoming necessary fortransmission, the oldertransmission lines areotten re12-
gated to si i btransmission function.
The di.stri/mtioii.sj:steiri is the final stage in the transfer of'pourer to tlie
individual customers. The primary distribution voltage is typically betwwn 4. 0
kV and 34.5 kV. Small industrial customers are supplied by primary feedcrs
at this \.oltagelevel. The secondary distribution feeders supply residential and
commercial customers at I20/240V.
The function of an electric power system is t o con\.ert energy from one of
the naturally available forms to the electrical form and to transport i t to the
points of consumption. Energy is seldom consumed i n electrical form but is
rather converted to other forms such as heat, light, and mechanical energy. The
rid\rantage of the electrical form of energy is that i t can be transported and
controlled with relative easeand with ahigh degreeof efficiency and reliability.
A properly designed and operated power system should. therefore. meet lie
following fitndament aI reqiiirements:
--
--
5
22kV
-
500kV 500kV 230kV
-
20 kV
Tielinetoneigh-
boringsystem
I
I 1 1
I
Trans-
I
Transmissionsystem
I
mission (230kV) Tieline
distribution
Subtransmission
Residential Commercial
Figure1.1 Basic elements of H power system.
1. The systemmust be abletomeet thecontinually changing load demand
for active and reactive power. Unlike other types of energy. electricity
cannot be conveniently stored in sufficient quantities. Therefore, ade-
quate spinning reserve of active and reactive power should be main-
tained and appropriately controlled at all times.
2. Thesystem should supply energy at minimum cost and with minimum
ecological impact.
3. The quality ofpower supply must meet certain minimum standards
6
with regard to the following factors: ( I ) constancy of frequency; (l )
constancy of voltage; and (3)level of reliability.
Several levels of controls it~vol\~ing acomplex array of devices are used to
meet the above requirements. These aredepicted in Figure I .2, which identifit*s
the various subsystems of a power system and the associated controls. In this
overall structure, there are controllers operating directly on individual system
elements. In agenerating unit these consist of prime mover controls and excih-
tion controls. The prime mover controls are concerned with speed regulaticm
andcontrol of energy supply systemvariables such asboiler pressures, temperi-
tures, and tlows. The function of the excitation control is to regulate generatlir
bultage and reactive power output. The desired MW outputs of the individual
generating units are determined by the system-generation control.
Theprimary purpose of thesystem-generation control istobalance the total
system generation against system load and losses so that the desired frequency
and power interchange with neighboring systems (tieflows)is maintained.
The transmission controls include power and bdtage control devices. sw:h
as static viir compensators, synchronous condensers, switched capacitors ;I id
Frequency Tie Generator
flows power
Supplementary
,-- - ------- - ------------------ ------- -- -,Cmd-. ---------I
I
I
I
I
I
I t
I
I Excitation I
I
I
I
I
I
+Systemand
control
4-
-
II
I
I
I
I
I
I
I
I
II
I
I
I
I
I
1I
I
I1
I
W
Rime
mover
i
Voltage
Shaft
Power
-
Generator
power
speed/
-
Power
;
i
I *
i TransmissionControls
;
I
;
I
i
Reactivepowerarid
Voltagecontrol,INDC
associatedcontros
transmissionand
Figure 1.2 Subsystems of a power system and associated controls.
I ntroductio ri 7
reactors, tap-changing transformers, phase-shifting transformers, and HVDC
transmission controls.
These controls described above contribute to the satisfactory operation of
the power system by maintaining system voltages and frequency and other sys-
tem variables within their acceptable limits. They also have a profound effect
on the dynamic performance of thepower system andon its ability tocopewith
disturbances.
The control objectives are dependent on the operating state of the power
system. Under normal conditions, the control objective is to operate as effi-
ciently as possible with voltages and frequency close to nominal values. When
an abnormal condition develops, new objectives must be met to restore the
system to normal operation.
Major system failures are rarely the result of a single catastrophic disturb-
ance causing collapse of an apparently secure system. Such failures are usually
brought aboutby acombination of circumstances that stressthe network beyond
its capability. Severe natural disturbances (such as a tornado, severe storm, or
freezingrain),equipment malfunction,human error,and inadequatedesign com-
bine to weaken the power system and eventually lead to its breakdown. This
may result in cascading outages that must be contained within a small part of
the system if a major blackout is tobe prevented.
1.3 POWERSYSTEMSECURITYASSESSMENT
The term Po\ivr.SysfeniStability isused todefine theability of the bulk power
electric power system to withstand sudden disturbances such as electric short
circuits or unanticipated loss of system components. In terms of the require-
ments forthe proper planning and operation of the power system, i t means that
following the occurrence of a sudden disturbance, the power system will:
1 . Survive the ensuing transient and move intoan acceptable steady-state
condition, and
2. In this new steady state condition, all power system components are
operating within established limits.
Electric utilities require securityanalysis toensure that,foradefined set of
contingencies, the above two requirements are met. The analysis required to
survive a transient iscomplex, because of increased system size,greater depen-
dence on controls, and more interconnections. Additional complicating factors
include theoperation of the interconnected system with greater interdependence
amongitsmember systems,heavier transmission loadings,andtheconcentration
of the generation among few large units at light loads.
The second requirement is verified using steady state analysis in what is
PowerSvsternStablllty 1
. 4
T
+ + +
Strwrural Volruge(Slow) OherTvpes(SSR, Lo6
I I Frequencb Oscillatiw s, etc
I
4 + +
Small Small Static
Disturbance Disturbance (Power
(Contingency) (Quasi-Static)
+- Cntical Sensitirity Bifurcation Bifurcation
Eigenvaluc AMIYSIS
I
Theory(Hopf.
Analyv\ et c. )
Bifurcation TapChanger Singular Value
Decompoiition
Techniques
Frequency
M Y Energy
Domi n Functions
Analysis Angle
Stability Linear
Selective Rogramrmng
Vdtage Ma Nmm Jacobian
AM I ysi s Stability Lnadability
~~~~-~~~~~~~~
4
Unlned
Useof theEnergyFunctionand h e Bifurcalion Analytical Studies
Approach
I
Linear Rogrammngfor DecompositionOpimal NonlinearRagramming
GermationRescheduling dnd PowerHow(OPRwth (NLP)withquadratic
Enhancement
(Prevcatlve
Load Control TransieruEnergy Marginand objccti~cfunctionsand
Coatrol)
VdcagcEnergy Marginds LinearRogramrmng(LP)fu
GenerationandLoadControl
Control using Exper(System Control Selections.output ArttndalNeuralNet worb
IFS)Pnmiizatim ClassifierandAnalyzer (ANN),and/ aExpert
intelligent
Systemsfasolutionoutput
SupportSystem
rel'erred to :is "s t ;itic seci i rit y iissessment."The t'i rst reqUirenit'11t is the hi i bject
deii1t \vith i11 "dynami c seci i ri ty assessme11t." Dynamic sec11 ri t y sti i di es ;I re
broadly classified ;is being either "angle stability studies" or "voltage stabil; t),''
;is depicted i n Figure 1.3.
In "angle stability studies." problems areclassified aseither "large distitrb-
mce.' tix transient e\'aluiitioii, or "small disturbance" for steady state stability
t'\,:ilurition. A similar classification for \soltage probleiix is indicated i n Figure
1..3. Solution techniqites tor transient unglc stability e\,aluationincluck:
Time domain sim ii1at ion
Direct methods
Hybrid methods
ProbabiIistic niethods. patt e1-11 recognition
COnnpitt;itiona1 int e1I i gence
Time domain simulation techniques invol\Ve judicious use of integration
methods such as Runge-Kutta. trapezoidal rule of integration, and ~~a\ . et c~r m
re1axat i on.These methods are partic41ar1y usef u1 for off-1ine transit 11t stabi1i t y
anaIysis.
On the other hand,in voltagestability studies. the problems areclassified
as large disturbance in some contingency cases, small disturbance i n quasi-
static cases. and static, which requires solutions to the general pouw tlo\v
(algebraic)equations only. I t isasa result of this classification that the solution
techniques and requirements are derived. The bifurcation theory. Linear Pro-
gramming applications. and the use of the energy fiinction are but ii few such
techniqLies. Again, the timedomain simuIations thiit ;ire i11\YII \red takead\,;in tage
of various numerical integration methods mentioned earlier. (Themachine dj8-
namics lead to differential equations that are inherently nonlinear.)
The unified approach as indicated on Figure 1.3is aimed at encompassing
the similarities and differences that distinguishes the Lrarious techniques used i n
assessing the stability of the electric power system. This is \4,hetheror not the
problems are a result of voltage or angle instability.The resulting enliancenient
that isbrought forth by thisapproach, measurable ;is a benefit-to-cost index. lies
in the de\ ~l opment and use of more robust tools for solLting present ~ind long-
range problems. I n this light, various programming and optimization schemes
that are applicable include decomposition Optiiutrl Po\t*orFlo\tp (OPF).Liiiorrr
Pi . oSr . cr i i i i i i i r i , ~(LP),and Qir(rcltntic.Pi . oRt . cri i i i i i i i i S (QP), with the necessary and
suffi cient system and network constraint s.
Finally. this book introducesthree fundamental types of intelligence support
systems that truly adds the rigor, value, and robustness to the desired enhance-
ment schemes. These support systems include expert systems (ES), fuzzy logic
(FL).and artificial neural network (ANN).Each have their unique characteris-
tics (decision-support, classifiers, learning capabilities, etc.) and are ridaptable
i n providing viable solutions to a variety of voltage/angle instability problems
associated with the electric power system. The discussion on this area of artifi-
cial intelligence applications to power system stability rind dynamics is pre-
sented i n the final few chapters of this book.
&
StaticElectricNetwork Models
INTRODUCTION
The power industry in the United States has engaged in a changing busincss
environment for some time, by moving away from a centrally planned s ys t m
to one in which players operate in adecentralized fashion with little knowleclge
of the full-stateof the network, and where decision-making is likely to be mar-
ket driven ratherthan through technical considerations alone.This newenviron-
ment is quite different from the one in which the system operated i n the past.
Thisleads to the requirement of new techniques and analysis methods for func-
tions of system operation,operational planning, and long-term planning.
Electrical power systems vary in size, topography and structural compo-
nents. However, what is consistent is that the overall system can be diLiried
into three subsystems, namely, the generation, transmission, and distribution
subsystems. System behavior is affected by the characteristics of each of the
major elements of the system. The representation of these elements by means
of appropriate mathematical models is critical to successful analysis of sys;em
hehavior. Due to computational efficiency considerations for each diffe-ent
problem, the system is modeled in adifferent way.This chapter describes some
system models for analysis purposes.
We begin in Section 2.1 by introducing concepts of power expressed as
active, reactive, and apparent. This is followed in Section 2.2 by a brief reLiew
of three phase systems. Section 2.3 deals with modeling the synchronous ma-
chine from an electric network standpoint. Reactivecapabilitycurves areexam-
ined in Section 2.4. Static and dynamic load models are discussed in Section
2.5to conclude the chapter.
2.1 COMPLEX POWERCONCEPTS
In electrical power systems one is mainly concerned with the flow of electrical
power in the circuit rather than the currents. As the power into an element is
basically theproduct ofthevoltage acrossandcurrentthrough it, it isreasonable
to exchange the current for power without losing any information. I n treating
sinusoidal steady-state behavior of an electric circuit. some further definitions
are necessary. To illustrate, we use a cosine representation of the sinusoidal
waveforms involved.
Consider an impedance element 2=ZL$.For a sinusoidal voltage, v ( r ) is
given by
v(t ) =tlCOSOt
The instantaneous current in the circuit shown in Fig. 2.1 is
i(r)=/,,,cos(or- Q)
where the current magnitude is:
The instantaneous power is given by
p( r)=i r( t ) i ( t )=XI I,,,[cos(or)cos(or- $)]
Using the trigonometric identity
1
cosclcosp=-[cos(a- p)+cos(cl+p)]
2
we can write the instantaneous power as
+
V
Z
-
Figure2.1 Instantaneouscurrent in a circuit.
The average powerp( , is seen to be
Since through 1 cycle. the a\ erageof co\(2ot - @) i \ zero. this term contrib-
ut e\ nothing to the acerage of / I .
I t i \ more convenient touse the effectiLre( r i m) calue\of boltageand curt-cnt
than the iiiiixiinuni ~al ues . Sub\tituting xi=v% Ci,,,,). and I, , , =-\/'?(/,,,,,). \+eget
Thusthe p o ~ wentering any network is the product of the effectilre v:iIies
of'terminal iaoltage and current and the cosine ol'the phase angle betLveen the
\,oltageand ciirrent which is called thei ~ o ~ ~ i ~ ~ ) . , t ~ / ~ . t o ) . (PF).Thisappliestosiniis-
oidal Lroltages and currents only. For ii purely resistive load. cos$ =1. and I he
current i n the circuit is fully engaged in con\ qi ng power from the soiirce to
the load resistance. When reactance (inductiveor capacitilte)as t+,ell;is mi s -
[;inccarepresent,ii componentof thecurrent in thecircuit isengaged i n con\ ~r: ~ -
ingenergy that is periodically stored in and discharged from the reactance. This
storedenergy,being shuttled into and out of the rnagnelic field of ;it1 inductaiice
or the electric field of acapacitance. addsto the magnitude of the current i n the
circuit but does not add to the a\.eragepower.
Thea\ wagepower i n ii circuit iscalled i1ctiL.e power. and loosely speaking
the po\+rerthat suppliesthe storedenergy i n reacti\re elementsis called reacliite
po\i.er.Acti1.cpou.cr is denoted bjfP, and the rci1ctii.epourer. is designattx ;i\
Q.Thej'areexpressed as
I n both equations. 1' and 1 are rim calues of terminal Lfoltageand current.
and Q, is the phase angle by mrhich the current lags the \,oltage.
Bot h P and Q arcof the samedimension,that is i n (Joules/s)Watts.Hou-
eb'er,toemphasizethe factthat Q represents the nonactive power,i t is measiired
i n reiictiLpe k~oltampere units (\'at-).Larger and tiiore practical units arc ki 1or . m
and megavars. related to the basic unit by
As\urnt. that 1'. 1' cos$, and 11sin@,a11 shown in Fig. 2.2, are each multi-
plied bqr /.the r i mcralue of thecurrent. When the componentsof iultage 1' , : OS@
vv
I 3
Static' Electric Ne t uv r k Models
Isin $
I
Figure 2.2 Phasor diagrams leading topower triangles.
and V sin$ are multiplied by current, they become P and Q respectively. Sinii-
larly, if I,Icos$, and I sin$ are each multiplied by V, they become V I , P. and
Q respectively. This defines a power triangle.
We define a quantity called the complex or apparent poufer,designated S.
of which P and Q are orthogonal components. By definition,
S = P +j Q = i/*
= V/ cos@+j V 1 sin@
= V/ (cos@+, j sin@)
Using Euler's identity, we thus have
s= VIC"*
or
s =VIL$
If we introduce the conjugate current defined by the asterisk ( *' )
I* = l l ( L @
i t becomes obvious that an equivalent definition of complex 01- apparent POW-
er is
s= VI" ( 2. 5)
We can write the complex power i n two alternati\,ef or m by using the
relationships v=Z7 and 7 =Y v
This leads to
s=ZI I" =ZJ I J ? (2.6)
or
S=VY*V" = Y" , VI " (2. 7)
V
I
Figure2.3 Seriescircuit of n impedances.
Consider the series circuit shown in Fig. 2.3. Here the applied Lvltage i:,
cqual to the sum of the voltage drops:
\'=I(& +z, +...+Z,:)
Multiplying both sides of.this relation by /*results in
s =\'I*=I / * ( Z ,+22 +...+Z, ! )
or
uith
s,= ,II?z,
being the individual element's complex power. Equation (2.8)is knourn as tke
sunimiition rule forcomplex powers.The summation rulealsoappliesto paral1i:l
circuits. The use of the summation rule and concepts of complex pourer ai.e
ai\rantageousi n solving problems of power system analysis.
The phasor diagrams shown in Fig. 2.2 can be converted into compltmx
pourer diagrams by simply following the definitions relating complex power .o
tzoltage and current. Consider the situation with an inductive circuit i n urhic-h
thecurrent lagsthe voltage by theangle$.Thecomplexconjugate ofthecurrent
M i l l be i n the first quadrant in the complex plane as shown i n Fig. 2. 3(a).
Multiplying the phasors by V, we obtain the complex power diagram i n
Fig. 2 . 4b).Inspection of the diagram as well as previous development lexis to
ii relation for the power factor of the circuit:
P
cos$ =__
1sI
2.2 THREE- PHASE SYSTEMS
A significant portion of all the electric power presently used is generated. trails-
mitted. and distributed using balanced three-phase bultage systems. The single-
Static Electric Ne t wwkModels 1.5
Figure2.4 Complex powerdiagram showingthe relationship among voltage. current,
and power components.
phase voltage sources referred to in the preceding section originate in many
instances as part of the three-phase system. Three-phase operation is preferable
to single-phase because athree-phase winding makes moreefficient use of gen-
erator copper and iron. Power flow in single-phase circuits is known to be pul-
sating. This drawback is not present in a three-phase system as will be shown
later. Also, three-phase motors start more conveniently and, having constant
torque, run more satisfactorily than single-phase motors. However, the compli-
cations of additional phases are not compensated for by the slight increase of
operating efficiency when polyphase systems of order higher than three-phase
are used.
A balanced three-phase voltage system consists of three single-phase volt-
ages having the same magnitude and frequency but time-displaced from one
another by 120".Figure 2.5(a) showsa schematic representation where the sin-
\120
van
120Pnce 120
Figure 2.5 (a)A Y-connected three-phase system and (b)the corresponding phasor
diagram.
gle-phase \,()Itagesources appearin ii wyeo r Y-connection;adeltaor A configu-
ration is also possible, as discussed later. A phasor diagram shouring each
the phase voltages is also given in Figure 2.S(b).As the phasors rotate at the
iiiigular t'reqiiency cu with respect to the refkrence line in the counterclockc~,isc
(ciesignatedas positive)direction, the positive maximum value first occurs t ' c r
phase( I and then i n siiccession for phases hand (*.Stated i n ;I different ~t'aq'. to
;in ohser\~erin the phasor space. the voltage of phase ( I arri\,est i n t follo\zred1'4,
that of h and then that of (*.For this reitson the three-phase Lroltage of Fig. 2 5
i h said to have the phase sequcncc t r bc . (order,phase sequence. or rotation a11
mean the ss;itiiething).This is important t or certain applications. For es~irnple.
in three-phase induction motors. the phase seqitence dcterniines u,hethertlic
i i i ot or rotates clock\i.iseor coiinterclock~ise.
2.2.1 Current andVoltage Relations
Balanced three-pha+e \y\tem\ c;in be \tudied u\ing technique\ de\reloped t or
$ingle-phawcircuit\.Thearrangement o f the three \ingle-phaw Ioltagc\ i nt o ;I
Y o r ;i A configuration require\ \ome modification\ in dealing U ith the o\er.iIl
\>\tern.
2.2.2 Y-Connec tion
With reference to Fig. 2.6. the cotiimon terminal I I I \ called the neutral or <tar
(k')point. The ~ol t agesappearing between any t uo of the line terminal\ 1 1 . h,
md ( h a b e different relation\hips in magnitude and phawtothe\oltage\ apptar-
ing bet\$eenany one line terminal and the neutral point 1 1 . The \et of colta;e\
\',.1; ,,itid itrecalledthe line coltage\,and the \et of coltage+li,. and \' \'(
c i ~ c referred toii\ the phaw coltage\.Con\ideration o f phawr diagram\ pro\ide\
the rcqiiI red reI;itionship\.
The etfectikekaliie\ o f the phase \wlt:ige\ ;ire \hohn III Fig. 2.6i i \ \,',, \;,.
and l' !.Each ha\ the wi l emagnitude, andeach 15di\placed 120"fromtheolher
tuo ph;i\or\. To obtain the magnitude and phare angleof the line \oltnge froin
( I to h (i.e.\ ' , , I . L+C appl y Kirchhoff'\ voltage l a ~ :
This equatioii state\ that the ~ol t ageexisting from ( I to h i h equal to the
coltage from ( i to 11 (i.e..\,',) piu\ the voltage f rom I I to h.Thus Eq. ( 2. 10) can
be re~irittenas
(. !. I I )
I7
Qn =vpL-120O
Figure2.6 Illustration of the phase and magnitude relation\ beiueen thc phaw and
line coltageof'a Y-connection.
Since for a balanced system, each phase Lroltage has the same magnitude.
let UI set
I K,!I =' &,I I = I YllI =1;
( 2. 12)
where 1;: denotes the effective magnitude of the phase voltage. Accordinglj u e
may write
K,,=y,o0 (2. 13)
Kll=y - - 120" (2.14)
yl,=y L- 230"=y L120" (2.1s)
Substituting Eqs. (2.13)and (2.14)in Eq. (2.1I ) yields
K,?=y (1 - 1 i- 120")
=f i p 3 0 0 (2.16)
Sinii1arly we ob tain
X( =fi y- - 90" (3. 17)
(2.18)
Theexpressions obtained above for the line boltages shou that they consti-
tute a balanced three-phase voltage system Mhose magnituctes are v? time\
thow of the phase voltages. Thuswe write
1.: = fi y (2.111)
A current tlowing out of a line terminal ci (or11 or i \ the same a\ thi\t ( a )
tloning through the phase source Lroltageappearing between terminal\ 11 and ( I
(or1 1 and h or I I and c) . We can thus conclude that for a Y-connected t hree
phase wurce, the line current equalsthe phase current. Thu\
l, = f,, ( 2. 2) )
I n the aboceequation, l, denotes the effective \ due of the linecurrent ai d
/,, denotes the effectike tfaluef or the pha\e current.
2.2.3 A-Connection
We now consider the case when the three single-phase sources are rearrangd
to forrn ;I three-phaseA-connection asshown i n Fig.2.7.I t isclear from in5ptc-
tion of the circuit shown that the line and phase \oltage\ haw the wine magiii-
tude:
The phase and linecurrents. howecerare not identical, and the relation\liip
betkveen them can be obtained by using Kirchhoffs current lau at one of he
1ine terminals.
I n a manner similar to that adopted for the Y-connected source,let U\ con-
\icier the pha\or diagram shoun i n Fig. 2.8.Assume the phaw currents to tw
/ n \b bb
ob
C
Ibc
I,, C
Figure 2.7 A A-connected three-phase s wr w.
I
Stntic Electt-icNet UY)t-k ModeIs
Figure2.8 Illustrationof therelationbetweenphasearid linecurrentsin a A-connection.
/ < I / , =I/ ,
I/,[=I,,i - 120"
I ,,,=I,,i120"
The current that flows in the line joining [ I to ci' is denoted by l$,,,, and is
given by
As aresult, we have
I,,,,, =I,,[I 120"- 1i01 L!
which simplifies to
I,,,,, =&,L I 50"
Similarly.
Note that a set of balanced three-phase currents yields a corresponding set
of balanced line currents whose magnitudes are fitimes the rnagnitudes of the
phase values:
20
\+hxl, denotes the m:ignitude 01' any o f the three line currents.
2.2.4 Power Relationships
Arrumethat ii three-phare generator is supplqinga balanced loud \+,iththc t h r w
rintiroidal phase voltagcs:
\' ( f 1 = \ '7 1;: \ l Il ( l ) t
\ ' , ( f ) = \ 21,:\111((1)/ - 120- )
\' ( I ) = \ 21,:\ l l l ( ( I) f + 120 )
~chcre@ is the phase angle bt.t\swn thc current iind voltage i n c d i ph;iw. ?'he
total po\ser i n the load is
This t i msout to be expanded US:
L:sing ;I trigonometric identit,~,ive get
Yote that the last three terms in the abo\.e equation add tip to ~ r o . l'hus
n' c obtain
When referring to the voltage level of'a three-phase s\'stem.by con\'ention.
one in\.ariably tinderstands the line l-oltages.From the iibo\.e cliscussion thc
relationship bet\\,t.cnthe linc and phase \.oltagesi n ;i Y-connectetl s ~~s t em s
Thepower equation thus reads in t er mof line quantities:
p10= f i , l f l 11, l C O 4
We note that the total instantaneous power is constant. hairing a magnitude
of three times the real power per phase.
We may be tempted to assume that the reactiLre power is of no importance
i n athree-phase system sincethe Q terms cancel out. Houte\zer,this situation is
analogous to the summation of balanced three-phase currents and \ dt agesthat
also cancel out. Although the sum cancels out, these quantities are still \very
much i n evidence within each phase.
We extend the concept of complex or apparent power ( S) to three-phase
systems by defining
SIO=3y1;
where the L1ctik.eand reactijrepowers are obtained from
I n terms of line Lralues, we can assert that
sl,= fiv1:
and
2.3 SYNCHRONOUSMACHINEMODELING
I n po\s'er system stability analysis. there are several types of models used for
representing the dynamic behavior of the synchronous machine. These models
arededuced by using some approximations tothe basic machine equations.This
section gives a brief introduction to synchronous machine equations.
2.3.1 Stator and Rotor Voltage Equations
In de\.eloping performance equations of a synchronous machine, the follou,ing
assumptions are made:
I . The statorurindings aresinusoidally distributed along the air-gapits far
a s the mutual inductance effects with the rotor are concerned.
2. The stator slots cause no appreciable variation of the rotor inductances
Lvith rotor position.
3. Magnetic hysteresis is negligible.
1. Magnetic saturation effects are negligible.
Based on these assumptions, a synchronous machine can be represented b j
six Lvindings as shown in Fig. 2.9. The stator circuit consists of three-phast-
;irniatiire uindings carrying alternating currents. The rotor circuit consists o.
field and iiniortisseur windings. The positive direction of ;I stittor winding cur.
rent is rissumed to be into the machine.
The Lwltage equations of the three-phase armature uindings are:
(),, = - N, , i , ,
t l t
Rotation
0, elec.rad/s
c
Figure2.9 Stator mcf rotor circuits of ;I synchronous rnachinc. ci. h.c:Stator ph: w
windings:jil:field cvinding: k d : d-axis armature circuit;k q : q-axis armature circuit;1 =
I . 2, ....r l ; I I =number of xinature circuits;0=angle by Lvhich cl - axi 5 leads the iiiag-
rietic axis of phase winding. electrical radiaii~;: ( I CO,=rotor angular \elocitj.clectri1:al
rad/x.
Static Electric Net\t*orkModels
23
where
L,,,,.L,,,,,L,,are self-inductance of nhc windings.
L,,/,. L,,,,LI,,,,LI,,,L,,,,LIIl are mutual inductances between two stator winding
(cih,hc,NC' )
L,,,,/,L/,,,/,L,,,/are mutual inductances between stator and fields windings
LlIAll, L,lL,I, LlL,/are mutual inductances between stator winding and d-axis
armature circuit
L,,A,,, LhAll, L,A,lare mutual inductances between stator windings and q-axis
armature circuit.
Therotor circuit voltage equations are given by:
(2. 24)
Therotor circuit flux linkages aregiven by
v/d=Lff,/;/,/ +L[?.diA[/ - Ldd;/>- LhAili/i- L,kqi ,
vkd=Llldild +LALdilrl - &i Adi , i - LRAdili - L,A,/;(
vAq=LAAqikt, - LiiAyio- Lldqih- LcAq;c
( 2. 25)
Equations (2.22)and (2.23)associated with the statorcircuits together with
equations(2.24)and(2.25)associatedwiththerotorcircuitscompletelydescribe
the electrical performance of a synchronous machine.
The fact that mutual and self inductances of the stator circuits vary with
rotor position q which in turn varies with time, complicates the synchronous
machine Eqs. (2.22)to (2.25).The variations in inductances are caused by the
variations i n the permeance of the magnetic flux path due to nonuniform air
gap. This is pronounced in a salient pole machine in which the permeances
along the two axes are significantly different. Even i n a round rotor machine
there are differences between the two axes due mostly to the large number of
slotsassociated with the field winding.
The self and mutual inductances of the stator circuits are given by
I,,=I,( , +1, !CO\ 78
1, =I. ,,+ +
3 '
[ :]=:
4!!- K ,;,,
r.1.28,
tit
where
All the inductancesexpressedasdqOcomponentsareseento be constant, i.e..
they are independent of the rotorposition. It is interestingto note that i ,,does not
appear i n the rotor f l ux linkageequation. This is because the zerosequence com-
ponents of armature currentdonot produce net mmf :icross the air-gap.
While the dqO transformation has resulted i n constant inductances i n Eqs.
(2.28)to (2.30).the mutual inductances between stator and rotor quantities are
not. For example, the mutual inductance associated with the f l ux linking the
field ufindingdue to current i,, flowing in the d-axis stator winding from equa-
tion (2.30)is (3/ 2)L,,,,,.whereas from equation (2.29)the mutual inductance
associated with fl ux i n thed-axis stator winding due to field current isL,,,,,.This
difficulty is overcome by an appropriate choice of the per uni t sqrstcm for the
rotor quantities.
2.3.3 Per Unit Representation
I t is usually conivenient to use a per uni t system to normalize system irariables.
to offer computational simplicity by eliminating units and expressing sqrstem
quantities as dimensionless ratios. Thus
A we11-chosen per unit systeni can mini111 ize comp i t ;itionaI effort , x i mp1ify
e\'aluation. rind facilitate understanding of system characteristics. Some baw
quantities may be chosen independently and quite arbitrarily, Lvhile others t'ol-
low automatically depending on the fundamental relationships between system
\.ariables. Normally, the base values are chosen so that the principal ttariable?
mrill be equal to one per unit under rated operating conditions.
I n the case of a synchronous machine, the per unit system inay be used t cI
remoF.earbitraryconstants and simplify mathematical equations s o that they rnq
be expressed in terms of equivalent circuits. The basis for selecting the per uni
system forthe stator is straightforward, while it requires carefulconsideration for
the rotor. TheL,,,,-basereciprocal per uni t system will be discussed here.
The following base quantities for the stator are chosen (denoted by sub-
scripts)
f?,h,l,r. =peak ~( al ue of rated line-to-line voltage, v
=peak value of rated line-to-linecurrent, A
,f\.,,,<=rated frequency, Hz
The base \ due of each o f the remaining quantities ;ire automatically st't
and depend on the above :is follows:
CO,,,~,,=271 .fh,l,c. electrical rad/s
2
o,,,,,.,,, mechanical rad/s =\clh.l,, -
TFf '
~ \ t 7 . l W
Z+lre=-:-,ohms
1\h,i\r.
The stator voltage equations expressed in per unit notations are gi \ mb j
_ _
27
Thecorresponding fluxlinkage equations may be written as
_ _ - - - -
-
w,/ =--&/id +LJ/ </+L<lL,/~L/
-
_ - - -
wq =-L& +Ld,/i</
-
WO =-MO (2. 32)
The rotor circuit base quantities will be chosen so as to make the fl ux
linkage equations simple by satisfying the following:
I . The per unit mutual inductances between different windings are to be
reciprocal.This will allow the synchronous machine model tobe represented by
simple equivalent circuits.
2. All per uni t mutual inductances between stator and rotor circuits i n
each axis are to be equal.
3. The following base quantities for the rotor are chosen, i n view of the
L,,,/-base per uni t system choose,
( 2. 33)
The per uni t rotor flux linkage equations are given by
(2. 34)
Since all quantities in Eqs. (2.31)to (2.34)are i n per unit, we dropthe overbar
notation in subsequent discussion.
If the frequency of the stator quantities is equal o the base frequency, the
per unit reactance of a winding reactance is numerically equal to the per unit
inductance. Forexample:
xl/= 2zfLl/ (Q)
=2nfhd,rL\hd\r., Dividing by Z\hd\c iff = f b d , c , then the per unit values of X, / and L,,
areequal.
2.3.4 Classical RepresentationoftheSynchronousMachine
The per unit equations completely describe the electrical and dynamic perfor-
mance of a synchronous machine. However, except for the analysis of very
\mall \ j \tern\,theseequation\ cannot be used directly for \}stem \tability \tud-
I C\ . Some \implification\ and approxitiiationsare required to reprewit the \ j n-
chronoii\ machine i n stabilitj \tudie\. For large \ ystem\. i t i \ nece\wrq to neg-
1ec.t the tr:in\tornier voltage term\ \ir<,arid \ir, atid the et'f'ttct o f \peed \ariation\.
Therefore,the rnachine equation tie\cribed bjfEys.(2.33)and ( 2. 33)become
( 2. 36,
By defining the follou~ing variables
the riiachine eyuationx becomes
( 2. 37 )
Mhere Et: i \ the q-a\ii\ c-omponent of the \oltage behind transient rcactanc\e
\'. I ", , i \ the open-circuit tran\ient time con\tant. E,i \ the \folt:ige proportiond
to I , ,and oloi \ the ~ol t ageproportional to E,<,.Sincc per uni t \ =I_ , l'rom
l !c~u~l tl on(2. 37)me hace
For studies i n \i.hich the period of analj'sisis sinall i n comparison c+,itli I-,:,
thc riiachine inodel i x often simplified bj'assuming t hat E';is constant throiigh-
o i i t the stlidy period. This ;issiinipti on e1i m inates the on1j , difterentia1 eyi i ;ition
asxociiited M i t h the electrical chxiicteristicsof the machine. A further approui-
rriation is to ignore transient saliency by assuniing that .v(; =.v(; and to :issiiriie
that the t'lux linkage al wremains constant.With theseassuinptions.the \,oIt;ige
behind the transient impedance R,,+j x$has a constant magnitude. The equi\.a-
lent circuit is shown i n Fig. 2.10.The machine terminal voltage phasor is repre-
sented by
q=Ei 6- (R(, +.j.~;)i,
The machine dynamic model is represented by
T, @4! =M, , ,- M,
(It
P,=R, ( V, I )
v,=EJ - (R,,+j x; ) i, (2.39)
where V, is the machine terminal voltage phasor and can be calculated from
power flowconsiderations.Then can becalculated.The machine scl?ing equa-
tion can then be solved.
Equation (2.39)isthe so-called classical model of the synchronous machine
and is widely used i n power system stability studies. This classical model is
often used for three different time frames: subtransient, transient. and steady-
state. Figure 2.1I surnmarizesthese three simple synchronous machine models.
The subtransient and transient assume constant rotor fl ux linkages. and the
steady-state model assumes constant field current. These models neglect sali-
ency effects and statorresistance andofferconsiderable structural andcomputa-
tional sim plicity.
Figure2.10 Eqiii\~alcntcircuit synchronous urith x,;=xi:.
t t
(a)Subtransientmodel
EL6
(b)Transientmodel
EqL6
( c)Steady-statemodel
Figure2.11 Simple hjmchronous machine model.
Stcrtic Electric Netwwrk Models
2.4 REACTIVECAPABILITY LIMITS
It isimportant in voltage stability and long-term stability studiesto consider the
reactive capability limits of synchronous machines. Synchronous machines are
rated in terms of maximum MVA output at specified voltage and power factor
(usually 0.85or 0.9 lagging) which they can carry continuously without over-
heating. The active power output is limited by the prime mover capability to a
value within the MVA rating. Thecontinuous reactive power output capability
islimited by three considerations: armature current limit,fieldcurrent limit,and
end region heating limit.
Figure 2.12 demonstrates a family of reactive capability areas for three
different values of hydrogen coolant pressure. Note that the higher the pressure.
the larger the capability curve. In Fig. 2.12, the region AB is the field current
limited while the region BC is due to armature heating constraints.
t
0.6p.f.lag
t
Figure2.12 Reactive capability curvesof a hydrogen-cooled generator at rated \ dt age.
2.5 STATIC LOADMODELS
Con\re11 ti onit1 transient-stabi1i t y stL i d ies werei n Ivedmain1y withgenerator stii.
bilit),.and little importance wits attached to loads. Recently, significant atteiitioii
has beeng i \ wto loadmodeling. Much ot'the domestic load andsomeindustrial
load consist o f heating and lighting. particularly i n the winter. and i n earl)'loati
111ode1s these \+'ereconsiderediis constant im pedances.Rotii tingeclLIipinent
often inodeled ;IS ii siiiiple form o f synchronousmachine and coniposite load
u'eresimulated by a mixture of these two types of load.A lot of uurk hits gon:
i nt o the development ot'more accurate load models.These include somecorn-
ples niodels of specific large loads. Most loads. howe\ter. consist o f ;I I xgc
c l i i antit y of cli\ wse eqiiipment of v;iry ing I c \ ~l s atid coinposition and some
cq11 ivaIenImodc1 is iiecessary.
A static l o d model expresses the churacteristic of the load at m y instarit
ot'tirric in terms o f algebraic functions of the bus \.()Itageniagnitude and t'w-
qi1cnc.y at that instant. The active power component P and the reacti\,cpo\f.crr
c0111pone11t Q are c011 sidered separateIy.
,A general load chxacteristic may be adopted such that the MVA loadii~g
Lit :I particulx bus is ;Ifunction of'lroltagemd fi-eclucncj~:
Q =K,/\'" f
$1here A',, atid A'(, are constant\ \+hich depend upon the nominal alue o f t ie
\:triable\ P and Q.For constant frequencq operation. u e write:
~+hmP and Q are acti\,eand reacti\re coinponetits of the load \+(henthe hiis
\'oltageniagnitude is 1'. The siibscript 0 identifies the Fdiics o f the rcspccri\~
\ , x iii bIcs ;it the iiiiti11I or iiominal c i peratitig condition.
Static loads are reliiti\Jeljfiiniiffected by frecliicncj' change\, i.e..ni,=1 1 , =
0. and with constant impedance loads i i i , =1 1 , =2. The iniportance of' acciiiate
load models has been deriionstrated f or \coltagesensitive loads.Figure 2. I3dtbiii-
oiistratcs the po~verand current characteristics 01' constant poLver, constant c iir-
rent, and constant inipedance loads.
Manq reseiirchers identified the characteristic load parameters f:x \,iirioiis
homogeneous loads.iypical \,aluesareshoLvii i n Table2.1.Thesecharac*terihtic*\
t
IVI
Nominal voltage Nominal voltage
(a) (b)
Figure2.13 Characteristic4 of different load models. ( a ) ActiLte and rcacti\c po\\er
\ersus voltage. ( b) Current \ er\ u\ \oltage.
inay be combined to @\rethe overall load characteristic at a bus. For esample,
agroupof homogeneous loads,eachwith acharacteristic ofN.j , and a nominal
power of P,inay be combined to give an overall characteristic of:
,=I
Table2.1 Typical Values of Characteristic Load Parameters
Filament lamp I .6 0 0 0
FIUoresct'nt Iamp 1.2 3.o - I .o -2.8
Heater 2.0 0 0 0
Induction motorhalf load 0.2 1.6 I .s -0.3
Induction motor f ul l load 0. I 0.6 2.8 I .8
Reduction furnace 1.9 2.I -03 0
A1iiininumplant I . 8 -.- 3 3 -0.3 0.6
An iilternatii,e model which has been widely used to represent the Lroltagc:
dependence of loads is the polynomial model:
This nwdel iscorninonly referred to as theZIPmodel. since i t iscomposed
of constant impedance ( Z ) ,constant current (11,and constant power ( P )conipo-
nents. The parameters of the model are the coefficients P, to P, and Q, to
which define the proportion of each component.
When the load parameters m , and 1 1 , are less than or equal t o uni t y, a
problem canoccurwhen the voltage dropstoa low value.As the voltage magni-
tilde decreases, the current magnitude does not decrease. In the limiting c;i!,e
wi t h zero cultage magnitude, ii load current flows m*hichis clearly irration:1.
gikren the nondynamic nature of the load model. From a purely practical poii t
of Lieut. then the loadcharacteristics areonly valid forasrnall Ldt agedeviation
from nominal. Further, i f the voltage is sinall. small errors in magnitude a i d
phase produce large errors in current rnagnitiide and phase. This results i n lo.;s
of accuracy and with iterative solution methods of poor convergence. Theie
effects ciin be overcome by using ii constant impedance characteristic to rep-
resent loads where the voltage is below some predefined ~~a l ue , for example
0. 8p11.
The parameters o f this model areexponents ci and h.With these exponerrts
equal to 0, 1 . or 2, the model represents constant power. constant current, or
constant iIn pedancechar ;ict eris tics ,respectic'e1y.For composite Ioads , their v i1-
lies depend on the aggregate characteristics of load components.
The frequency dependence of load characteristics is represented by ni ul t i -
plying the exponential model or the polynomial niodel by ;I f'actoras follow>:
or
where Af is the frequency derivation (.f-.f;,).
CONCLUSIONS
I nthischapter weofferedareview of powerconcepts forsingleand three phase
systems. We alsotreated the fundamentals of synchronous machine models for
stability evaluation including the idea of a reactive capability curkeand static
load models. The per unit system was reviewed and extended to quantities not
frequently encountered, such as time and frequency. Also. some modeling as-
pects of staticloads,such asfrequency dependent loads i n typical electricpo\tw
systems, were also discussed.
The reader is referred to the bibliography section for references dealing
with materials in the chapter and contributions made by many other pioneers i n
the field. Please note that an annotated glossary of terms is given to summarize
the key definitions and terminology employed in this chapter.
3
Dynamic ElectricNetworkModels
INTRODUCTION
Chapter 2 focused on steady-state models to represent power system elemcr'is
for the s o ~.alledstatic analysis studies. We nou turn O L I ~ attention to inodtIs
fortransi ciit or dyniiiiiicoperationa1 stl i dies.Thischapter describes soiiiesyst en i
niodclh for analytical purposes. A model of an excitation system is studied i n
Section 3.1 and Section 3.2gikres ii discussion of a model of'the prime nio\er
and go\~eriiorsystem. Dyii;iinic load models arc discussed i n Section 3.3 to
concludethechapter.
3.1 EXCITATIONSYSTEM MODEL
The basic function of an excitation system is t o probride ;idirect current to lit.
synchronousmachine field Lvinding. I n addition,the excitation system performs
control and protecti\,efunctions essential t o the secure operation of the s>rstm
by controlling the field Lmoltage and hence the field current 10 be within Licccpt-
;ibIe le\.e1s 11nder different operatirig conditions.
The control tunction~ include the control o f iroltage and reiictiire po'i'er
tlow,therebyenhancing pourer syste111 stabiIit y.The protective functionsensi i re
th;i t the ciipi1bi1i t y 1i111its ot the syiichronous iiiiichine,escit ation s ~ ' s te111, Ind
otherecliiipmcntare not exceeded.
37
i
Power Power i i Power
source
(regulator)
source
(regulator)
;
i
i
j
'iource
(exciter)
w
1 r i i Excitation
4 - 1
Reamplitier
Power
i
: power
Amplifier i SWrCe 7 machine
I + i (exciter)
- 1 I : I : I
I I
I L - I
I
I
II
I
I I I
I
I
I
I
I
I
II
I -
I I I
I : I I
I
~ , ~
7
4 Regulator i M a n u a L E ~ S+Sy nc hr o no us ~ ~
'iConlrol: ; machine
4 Excitation *
system Excitation
controlsystem
*:
Figure3.1 Functional block diagram of a synchronousexcitation control sy\tein.
Figure 3.1 shows the functional block diagram of a tj'pical excitation con-
trol system forii synchronous generator. The following is ;ibrief description of'
the various subsysteins identified in the figure. The exciter pro\sides dc p o ~ ' e r
to the synchronous machine field winding and constitutes the pouw stage of'
the excitation system.Usually an exciter is modeled by the first-order system
as shown in Fig. 3.2. The effect of saturation is considered by introducing S,
and K, , , and thus the exciter model is given by
1
(3.11
E,=
1 +S,, +K,+T, s
v,
The \.ohage regulator processes and amplifies the input control \ign;il\ to ii
leiel and form appropriate for the control of the exciter. This include\ both
regulating and excitation system stabilizing function (ratefeedbach or lead-lag
compensation).Nornially. the regulator is inodeled by a fir\t-order \ j stem a\
sho\+ni n Fig. 3.3.The regulator model is given by
Figure3.2 Block diagram model of exciter
Theterminal coltagetransducer sensesthe generator terminal voltage. recti-
fie\ and filters it t o dc, and compares i t with a reference which repre\ent\ the
h i r e d terminal coltage. Figure 3.3 shows the terminal voltage tran\duc-er
model given by:
Thepower system stabilizer provides an additional input signal to the r e p -
lator to further damp power system oscillations. Some commonly used i n w t
signals iire rotor speed deviation, accelerating power, and frequency debiatirn.
A pouwsystem stabilizer ismodeled asshown in Fig.3.5.The PSS model
is gikwi by
AV=K, , G' ( s) Au+Ki,G(.\)At;' (.$.4)
U'here
Figure3.3 Block diagram of the voltage regulator.
V
S v F
Figure3.4 Block diagram of aterminal voltage transducer model.
Limiters and protective circuits ensure that the capability limits of excitor and
synchronous generator are not exceeded.
The full excitation system is modeled as shown in Fig. 3.6.The system is
represented by the following set of linear differential equations:
T, &V- I j
dt
?; C' VX=- V, - K( I : : , - y+l ( +y)
dt
T =V, - (S,+K, +I )E,
dr
dV
T, --t=-v+&(& - ( S, +K, +1 ) E, )
d r T
X 2
1+T1S '"s
-
KPSS I + T5S l+T 2s 1+T2S
Figure3.5 Block diagram of a typical Power SystemStabiliter (PSS)model.
40
3.2 PRIME MOVER AND GOVERNINGSYSTEM MODELS
The prinie soiircesofelectrical energy supplied byutilities itre thekinetic enctg>'
o f Uater and the thermul energ~' deri\wl frorn fossil fuels md nucleiir fission.
The prirnc rnoi.erscot1Lw-tthese soiirces of energj'into niechanicd entrgj' t l - at.
i n t ur n. is con\,ertedto electrical form by the synchronous generator.The priiiie
m0i . er go1.erning syst e111 provides a means of c011t1-0ng power and t'rccliicIIL'1'. I 1i
The !.Utictioti;i 1 reIiiti onship betwee11the biisic eIcme11ts ;issociiited wit11 PO\IcI
gcner:itioii and control i x shokkrn i n Fig. 3.7.This section introduces t he i i i occls
I' oi- 111draiilicturbines and go\,crningsystems;is well iissteam turbines atid tlicir
go\~crnitig s>~stcms.
3.2.1 Hydraulic Turbinesand Governing System Model
Theh>draulicturbine model describes the characteristics of'gate opening p:incl
o11 tp i t rnccha11ical power. I11 power systemdynamic ;I 11a1ysis.the h>,driiLII ic tLI r-
biiic is ~i sual l ymodel edby an ideal lossless turbine alongUith theconsideration
ot'"wiitcr hiitiinicr" effect caused by the uwer inertia. is gic,enbl,
3.h)
M here T, )i \ \+raterstarting time.
BCC;ILIWo f the "LI. ater h;itiiiiier" effect,;I change i n gate position produce\
L
EnergySupply
System
t
Speed Valve 1_1) Turbine
Generator
Governor or
~
Speed
Figure3.7 Functionalblock diagram of pobver generationand control.
an initial turbine pouer change which is opposite that \shich I \ de\ired. For
stable control performance, a large transient (temporari\) droop M it11 a long rc-
setting time I \ therefore required. This is accomplisht.d by introducing a tran-
sient gain reduction compensation i n the governing \i\ \tern. The coiiipen\ation
retard\ or limit\ the gate niovement unt i l the mater flouand poueroutput h a ~ c
time to catch up.The gmerning system model i \ \ho\+ni n Fig. 3.8.
Pilot Valve
maxgate
and
Figure3.8 Bloch dingram of gojwning system for ;I hjdraulic rurbinc.
42
The goirerning system model is given by
7(-,4! =-p +x,
t f t
( 3 . 7 )
=pilot v a l \ ~and ser\mi(itortime constant
K , =servogain
7;7 =main servo time constant
R,, =permanent droop
R, =temporary droop
&=reset time
ql,,,,, ,,,, =maximum gate opening rate
q,ll,i, il,,rC =maximum gate closing rate
p=gate position
3.2.2 SteamTurbinesandGoverningSystemModel
A \team turbine convert\ stored energy of high preswre and high teniperatiire
jteani into rotating energy.The input of the steam turbine i \ control \, al\epvsi-
tion (Ayl).nthile its output is torque (AT, ). In power \tability analj\is. a
order rnodel is used for \team turbine, i.e.,
ufherer,,=time constant
Comparing the turbine models for hydraulic turbine and \team turbine, I t I \
clearthat the re\ponseof a \teain turbine has no peculiarity wch a\ that extiib-
1tt.dby ;I hydraulic turbine d~i e to water inertia. Thegoverning requiremen('to f
\team turbine\, i n this re\pect, are more straightforward. There i \ no need for
triin4ie11t droopcoinpen\ation.
Thego\erning \ j \tern model i \ given by
-
-
max gate
band
1
r x l -
'
@ KS - l +s TG
S
Figure3.9 Typical block diagramof a steam turbine
cx=K, (U,, - U,- R,,X,) ( 3. 9)
lit
Atypical governing model for steam turbine is shown i n Fig. 3.9.
3.3 MODELINGOF LOADS
Load models are traditionally classified intotwobroad categories: static models
and dynamic models. Earlier, we considered the static load models (Chap.2) .
I n this section, the dynamic load model isdiscussed.
3.3.1 DynamicLoad Models
Typically, motors consume 60to 70% of the total energy supplied by a power
system. Therefore, the dynamic effects dueto motors are usually the most sig-
nificant aspects of dynamiccharacteristics of system loads. Modeling of motors
is discussed in this section.
An induction motor can be represented by the equivalent circuit shomm i n
Fig. 3.10.which accounts forquantities in one phase.
In the equivalent circuit all quantities have been referred to the stator side.
The directions of current shown are positive when operating as a motor, i n
which case the slips is positive. Therotor equation of motion is given by
(3.10)
Thetorque (r)is slip-dependent,
44
Figure 3.10 f:qiiiwlc.iit circuit ol' ;I ihrw-pli;isc. iiiductioiiin;icliiiic..
(3.1 I I
where /: is the nuinher of polesatid s is the slip. def'incd ;IS:
,< = !'C! I !
11,
bvi t h 1 1 , being the synchronousspeedof the machine. where P is the number of'
pdes. aiicl,/'isthe frequency. I n Eq. (3.I I ) we have
(K,+j .U,)
U,+,; x,
' .U,,,
=J -
K.+.; IX. + x,..,
I t is noted that Ecl. (3.10)represents ;I stendy state pcrformaiice nioclcl 0 1 the
intluction motor. with all qiiantities referrecl to the statorsiclc.There;ire inodcls
that represent the trnnsicnt performance of the niotor t h; i t are bnsccl on I l ux
liiikngcs. volt;iges. and torque variutions.
CONCLUSIONS
In this brief chapter we concentrated on models ibr power ~ystciii stahiliiy i n
the tiiiie domain.Wediscussed electric excitation model as well asprime iiIover
and governor system models. We concluded with a brief introduction to dy-
namic load model. For fiirther reading on the topic. ii reference list ar d an
annotated glossary of teriiis is provided ;it the hack of the book.
Philosophyof SecurityAssessment
INTRODUCTI 0N
We areconcerned with the implications of il major network disturbance such as
a short circuit on a transmission line, the opening of a line or the switching 011
of a major load to namejust a few. Here, we will consider the behaiviorof the
system immediately following such a disturbance. Studies of this nature are
called transient stability analysis. The tendency of a power system to react to
disturbances in such a manner as to maintain its equilibrium (stayi n synchro-
nism)is referred to as stability. Oneway of classifying disturbances is through
the categorization of small versus large. A disturbance is assumed to be small
if the behavior of the system can be adequately represented through a linsariza-
tion ofthe nonlinear system of dynamic equations of the system.
Stability considerations have been recognized to be among the essential
tools inelectricpowersystemplanning.Thepossible consequences of instabilitj,
i n an electric power system were dramatized by the northeast pobrer failure of
1965.This is an example of a situation that arises when il seLFere disturbance is
not cleared away quickly enough. Theblackout began with a loss of;I transmis-
sion corridor. which isolated a significant amount of generation from its lo:id.
More recently. a transmission tower in the Consolidated Edison sj~stem u' a s hit
by a severe lightning stroke in July 1977.The events that follo\s.ed led to the
45
shutdown of power in New York City.Both eventsdramatized theconsequences
o f iiii instability in an interconnected electric power system.
Our intention is to give ;in introduction to transient stability in electric
power sjrstems. We treat the ciise of a single machine operating to supply iir
infinite bus.Theanalysis of the tiiore complex problem of large electric po~vei'
riet\\.orks1t.ith the ititerconnections taken into consideration is treated ;is istell.
4.1 THESWING EQUATION
I n the p o ~ s er \j' stem\ engineer' \ terminology, the dynamicequation relating t h t s
inertial toryue to the net accelerating torque o f the s j nchrotious machine rotor
i \ called the s~ ing equation.Thi\ \imply \tate\
( 1%
,-
-
T',,
i 4.1t
tlt-
The left-hand-side i \ the inertial torque uhich is the product of the inerti,i
( i n hg In') of all rotating tii;i\se\ attached to the rotor shaft arid the angul:r
:icccler:ition. The accelcrating torque T, , is in Nc\\ton-meter\ and can be c'\ -
pres\ect as:
I n the abol e, T,,,i \ the dritirig mechanical torque and i \ the retarding or
load electrical torque. The angular po4tion of the rotor 0 may be eupre\\ed ii\
the f o l l o ~ing win ofangle\
The angle cc is a constant Nhich is needed if the angle 0 is mea\ured from
;in different from the iingular reference. The angle w,?tI \ the re\ult of the
rotor angularmotion at rated \peed. The angle 8 i \ time \ a r j ing and repre\erts
de\iation\ from the rated angular displacement\. This gi\es the ba\i\ f-or our
1s ~ i e I-c1;iti on
We find i t iiiore conLtenient to substitute the dot notation
Therefore we have
Jii'=T,,,- T,, (4.5)
4.2 SOME ALTERNATIVE FORMS
Some useful alternative formsof Eq.(4.5)have been developed. The first is the
power form which is obtained by multiplying both sides of by 03 and recalling
that the product of the torque T and angular velocity is the shaft power. This
results i n
Jog =P,,,-P, ,
The quantity J o ~is called the inertia constant and is truly an angular mo-
mentum denoted by M (Jdrad. ):
Thus the power form is:
M8 =P,,,-P , ( 3. 7)
A normalized form of the swing equation can be obtained by dividing Eq.
(4.5)by the rated torque THto obtain thedimensionless equation.
Theleft-hand sideoftheaboveequationcanbefurthermanipulated toyield
a form frequently used. Recall the definition of the kinetic energy of a rotating
body. This gives the kinetic energy at rated speed as
1
Wi=-J o ~
2
then
L 2 W , -
TH o?,T,
We know further that the rated power is
-I8
Thus
Consequently we huvc
A constant which has proved very useful is c-noted by H. which is equ.11
to thc kinetic energy itt rated speed dividedby the rntcd power PK
The units of H ;ire i n sec. As il result we write the per unit or iiortiia1izi:d
suingequation its
Obscrvingthiit .,,, = P,,,,, wc cat1 then write
ZH 6 = p,,,- p , (4.0)
On
\vlicrcthccquution is i npi i .
4.2.1 Machine Inertia Constants
Thc iltigttlitr iiionicntuiii inertia constant M as defined by Eq. (1.6)cu11be oh-
tainctl from mnnufacturcr supplicd machine datil. The machine kinetic ~11crgy.
,V. iiiay be written i n ter111sofM i1S follows:
where wKis the angular speed in electrical degrees per second.This i n tttrtl is
rclutccl to the frequency by
We c mtherefore conclude that
The value of N is obtained from the moment of inertia of the machine
Lisually denoted by WR and traditionally given i n Ib-f. The con\-ersionformu-
la is:
The relation between H and Mcan be obtained using Eq. ( 3. 8)re\s<rittenas
H = -
N
(4.14)
G
Here G is the machine rating. Thus
(4.15)
The quantity H doesnot vary greatly with the rated power and speed of the
machine, but instead has a characteristic value or set of \ due\ foreach cla\sof
machine. I n theabsenceof definiteinformation typical \aluesof H maybe used.
The cur\,esi n Figs. 3.I . 3.2. and 4.3give the general characteristic \ariation of
H for exi\ting and future large turbo generators.
5
- -
#S
Figure4.1 Inertia conat ant bfor large turbo generator rated 500MVA and below.
C' ht rpt er I
4 -
3.5
- -
A
CrJ
.
i 3 - -
s
3 2.5
- -
E
I
cn
2 2
--
-3600 RPMFOSSIL
c
cn
c 0
-C-1800RPMNUCLEAR
1.5
- -
2
-
C
1
- -
0.5
- -
, 1 1 1 ,
, - 0 I I
Figure4.2 Expected inertia constants f or liitiire large turbo generators.
In systern studies Uhere several machines ha\ing different ratings are wed,
the H constant foreach machine, given to a base of the machine rating. rriu\t
beconberted to the coiiimon system base by multiplying H i n Eq. (4.14) bj the
ratio (niachinebase MVA/system base MVA).
4.3 TRANSIENTANDSUBTRANSIENT REACTANCES
In order to understand the concept of transient and \ubtran\ient reactance\ of a
\ynchronous generator, let U\ consider the transient behavior during a balariced
t'ault. The dependence of the talue of the short circuit current in the ele:tric
p o ~ e r sj \ t em on the in\tant i n the cycle at which the \hart circuit occur\ can
be verified using ;I simple model. The inodel is ;I generaror uith wrie\
re\istance R and inductance L \ hown i n Fig.4.4.The xroltage o f thegene-ator
I \ a\\uriied to Lary a\:
5I
4
- -
I
3 3.5
- -
F
$ 3 - -
3
z
I 2.5
- -
v)
c.
c
tj 2 - -
- A 450-514rpm
0
0
.E 1.5
- -
+B 2Ul-400rpm
r
+C 133-180rpm
-
t
1
- -
+D 80-lXIrpm
0.5
- -
. . . . . . . . . . . . . . . . . . . . 0 -
Figure4.3 Inertia constantsoflarge water wheel generator\.
r ( t )=E,,, sin(cot+00
With a balanced fault placed on the generator terminals at t =0. then we
can show that a dc term will in general exist. Its magnitude at t= O may be
equal to the magnitude ofthe steady-state current term.
The transient current i(r) is given by
where
The worst possible case occurs for the value of agi \ mby:
j wL R
In t hi \ casetheciirrent magnitude illapproach titicethe\tcadq-\talc IIa \ -
i ni i i i i i \alue iinmcdiiitely after the \hart circuit. The tran\ient current I \ pi\-
Thus:
E,,,
i ( t )=-1
z
I t isclear that the maximum of i ( r)
is twice that ofE,,,/Z.This n~a\~efortii is
shown in Fig. 4%
E,,,
For thecase tana=or we have i ( t )=-sinot.This a.a\.eformis shourn i n
z
Fig. 3.Sb.
I t is clear from inspection ofeither the expression for the short circuit cur-
rent or the response u'ajseform given i n Fig. 4.6that the reactance of the ma-
chine appears to be time varying. This is soif we assume afixed Lfoltagesource
E. For our power system purposes we let the reactance vary i n a step-wise.
fashion X:, Xg;and X, , as shown i n Fig. 3.6.
The current history i ( t ) can be approximated i n three time zones by three
different expressions. In the first, denoted the subtransient inten-al.lasting up
to 2 cycles. the current is I". This defines the direct axis subtransient reactancc
The second denoted the transient gives rise to
Figure4.5 Short circuit current waLCeforms.
/
E
x;=
I'
kchere I' is the transient current and X:, is the direct axistran\ient reactance. Ti e
transient intenfa1last5forabout 30cycles.The steady-state condition gi\ses t76'
direct iixi\ synchronous re;ictance.
Note that the subtran\ientreactance can be as low as7%of t he\ynchronc U\
re;ict;i nce.
4.4 SYNCHRONOUSMACHI NEMODELIN
STAB1LlTYANALYSIS
A brief outline of equations to account for fl ux changes in a synchronous ma-
chine is given to define various electrical quantities and to construct phasor
diagrams. Thefollowing approximations are involved i n the models discussed:
I . The rotor speed is sufficiently near 1.0pi( and may be considered a
constant.
2. All inductances are independent of current. The effects due to satura-
tion of iron are not considered.
3. Machine winding inductances can be represented as constants plus si-
nusoidal harmonics of the rotor angle.
4. Distributed windings may be represented as concentrated windings.
5. The machine may be represented by a voltage behind an impedance.
6. There are no hysteresis losses in the iron, and eddy currents are only
accounted forby equivalent windings on the rotor.
7. Leakage reactance only exists in the stator.
Under these assumptions, classical theory allows constructing a model of
the synchronous machine i n the steady-state, transient, and subtransient states.
The per unit system adopted is normalized, although the term propor-
tional should be used instead of equal when comparing quantities. Note that
onep i i field voltage produces 1.O p i fieldcurrent and 1 .O p i open-circuit ternii-
nal voltage at rated speed.
4.4.1 Steady-State Equations
Figure 4.7 shows the flux and voltage phasor diagram for a cylindrical rotor
synchronous machine ignoring all saturation effects.
The following commentsexplain the construction:
I . The fl ux @, is proportional to the field current 1,and the applied field
voltage and acts in the direct axisof the machine.
2. The stator open-circuit terminal voltage E,is proportional to @, which
is located on the quadrature axis.
3. The voltage E,is proportional to the applied field voltage and may be
referred to asE/.
4. When the synchronous machine is loaded,a flux Q> proportional toand
in phase with the stator current I and when added vectorially to the
field fl ux@, gives an effective flux a,..
5. Theeffective internal stator voltage E,is due to @, and lags i t by 90.
6. The terminal voltage V is foundfrom the voltage E,by considering the
cf,
DirectAxis
t
I
I
I
I!
--
b
QuadratureAxis
\oltage d r o p due to the Ieahage reiictance X and iiriiiatiirc rt" i+-
tatice K .
7 . B) +iniilaritj01' triangle+.the dit'terencc betbeen E,and E i \ i n ph i\e
U ith the / X \oltagedropand i \ proportional to /.Ther et i mthe \ol ~i ge
difference ni aj bu treated ;i\ ;i koltage drop ;icro++;in iirniatiire rcict-
ancex,.
The'riitii o f X,, and X i is termed the synchronous reactance, x>.
For the salient pole synchronous machine. the phasordiagram is more coin-
pies. Bec~iii'rethe rot or is syninietrical about both the d and c/ ;ixes i t is cot- \.u-
nient to rusol\~enianj' phasor quantities into coniponents i n these ;ises. 'The
stator currunt niajc be trtxited i n this manner. Although a,, ujill bc proportional
to I,:iind a,, will be proportional to /,,, because the iron paths i n the t wo :,xes
;ire diftcrent. the total iirniatiire reaction tlus0\ +. i l l not beproportional t o /iior
necessarily be i n phase \+rith i t . Retaining our earlier normalizing ;tsutiipti)ns.
i t inay be assumed that the proportionality between !*, and @</, is i i n i t y but tlic
proportionality between and @(, is less than uni t y and is a function of the
saliency.
4.4.2 Salient PoleSynchronousMachine
Figure4.8showsthephasordiagramofthe salientpolesynchronousmachine.The
dandqaxes armature reactance4aredeveloped a4 inthec)lindricalrotorcaw. Di-
rectandquadrature4ynchronousreactance4X, , andX, , canbeestabli\hed. i.e..
From the phasor diagram we have:
where V+ and V,. arethe axial componentsof the terminal voltage V.
In steady-stateconditions i t is reasonable to use the field voltage E, or tht!
voltage equivalent to fieldcurrent E, behind the synchronousreactancesas thc
machine model.In thiscase the rotor position (quiidratureaxis)with respect t o
the synchronouslyrotating frame of reference is given by the angular position
of E,.
As thecylindricalrotormodelmay be regardedasii specialcaseofnsalient
machine ( X, ,= X, , ) , we will consideronly the salient pole machine.
4.4.3 Transient Equations
For fasterchanges in the conditions external to the synchronousmachine. the
steady-statemodel is no longerappropriate.Due to the inertiaof the flux l i n l -
ages these changes cannot be introduced throughout the whole of the modcl
inimedintely.I t isessentialtoestablishnew fictitiousvoltagesE,; and E,;. reprc-
sentingthc t l ux linkagesof thc rotor windings.Thesetransient voltagescall t-e
showntoexist hehind the transient reactancesX,; and X,;,
E:,= v,,+ R,,I,,- la,x<;
E*;= V,,+ RJcl+ l , , X;
ThevoltageE, isnow consideredasthesumof twovoltages.E,, and E,, aid
is the voltagebehind synchronousreactance.I n steady-state.current flows only
in the field windingand hence, in that case,E,,= 0and E, = 0.
Allowing for the rotor t l ux linkages change with time requires using tie
followingordinarydifferentialequations:
-E ,
SE,;= -
r,:..
Thephasordiagramof the machineoperatingin thetransientstateiss how
in Fig.4.9.
DirectAxis
El
i E,
Figure 4.9 Phasor diagram of a salient pole synchronous machine in the transient
state.
4.5 SUBTRANSIENTEQUATIONS
Other circuits exist in the rotor, either intentionally, as in the case of damper
windings, or unavoidably. These circuits are taken into account if a more exact
model is required. The reactances and time constants involved are small and
can often be practically disregarded. When required, the development of these
equations is identical to that for transients and yields:
Theequations are developed assuming that the transient time constants are
large compared with the subtransient time constants. A phasor diagram of the
synchronous machine operating i n the subtransient state is shown i n Fig. 4.10.
4.6 MACHINEMODELS
I t is feasible to expand the model even further than the subtransient level but
this is rarely done in multi-machine stability programs. Investigations using a
Direct Axis
generator inodel with up to seven rotor \+~indiiigs. ha\,e shown that using the
\tandard niachine data.the I~IOI-e complex models donot necessu-ilyj'ieldaccti-
ixte results. Ho~. t . \ . er , improved results can be obtained if the data, cspeciall>.
t tie tinie constant s. are appropriate1y modified.
The mo s t contwiientmethod of treating sq'nchronous machines of differir-g
coniplcxit>'is to allow e x h machine the I ~ I ; I S ~ I I ~ L I I ~ possible number of ec1u.t-
tioritr a i dthen let the actual model used be cleterniined autoinatically accordir g
to r hedata presented. Thus,fibre models are possible tot ii four-\s~indingr ot or
4.6.1 Model 1
4.6.2 Model2
4.6.3 Model3
cl- and q-axis transient effects requiring two differential equations (SE:.and ,SE).
Thefollou~ing equations are used.A block diagram is shown i n Fig. 3.11.
E:,=v,,+R,,I,,- l(/x:/
E:, =\'(, +RJ, , +1,,x,;
U
Figure4.11 Block diagramrepresentation formodel 3.
4.6.4 Model4
ti- and q-axis subtransient effects requiring three differential equation\ (
$E::, and \E::). The followingequations are used.
4.6.5 Model5
tI- ai d (1- axis si i btra11sie11t effects reqi i iring f our clifferentia1 eyii;I ti oris (.;E;.
.SE,:. and .SE:;). Thefollouingequations are used. .SE:;,
The following mechanical equations need to be solFred forall these models
s6=O - OH
4.7 GROUPSOF MACHINESANDTHE INFINITE BUS
Groups of synchronous machines or parts of the system may be represented by
a single synchronous machine model. An infinite busbar, representing a large
stiff system, may be similarly modeled as a single machine represented by
model 1 , with the simplification that the mechanical equations are not required.
This sixth model is thus defined as:
4.7.1 Model0
Infinitemachine-constant voltage (phaseand magnitude) behind ci-axistransient
reactance ( X: , ) . Only the following equations are used.
4.8 STABILITY ASSESSMENT
In this section, we discuss the conventional approach to stability assessment
applicable to a single machine against an infinite bus. The method leads to the
equal area criterion.
We concluded that a simple representation of the salient pole machine is
offered by the model 0given by:
U'e \+i I 1 make the to1lowing additio nal a \ s 11111ptions:
A\ ;I result:
I:' =I' - I X'
0=\ ' +I X '
The output po\ier of ttic niachine is g i \ wbj :
\ ' =I ' c o d
I ' , =-\' , 4i i i i 5
The phasor diagram is st1ou.nin Fig. 4.12.
Theelectric P OM' C ~output ofthe salient pole machine is thereti)regiven b j :
The Lariation o f thc output for- salient pole rnachinc L+ i t h the torcliie o r
po\srer angle6is \ho\s,n in Fig.4.13.
Network
Imaginary Axis
Machine
Quadrature Axis
DirectAxis
Real Axis
"d
Figure4.12 Synchronous niachine and network frame\ of rc!crence tor dcvclopiiig
electric po\ver oiitpiit formula.
=Pe
Figure4.13 Power anglecharacteristicsfora salient pole machint..
66
output
Power,P
Angle, S
(radians)
Figure4.14 Power angle characteristicsfor ;I round rotor machine.
I n the case of ii round rotor machine, u'ehave Xf:=X( ; and hence
E V , .
P, =-- sin6
x:,
The ~ariation of the output power fora round rotor machine uith the angle
6,(torqucor powwangle)is shown i n Fig. 3.14.
Ewmple 1
A \ynchronou\ machine is connected to an infinite bus through a transformer
ha\ingii reactance of 0.I p i and a double-circuit transmission line uith 0.45pi'
reactance for each circuit. The \ystem is shown in Fig.4.IS. All reactance\ are
giken to a ba\e o f the machine rating.The direct-axi\ transient reactance of t ie
rnachine i \ 0.15p i t . Determine the \sariation of the electrical pomw uith :ink le
6.Assullle v=I .op.
SoILitio17
An equitralent circuit of the LtboLre system is shown in Fig. 3.16.From this ~s' e
ha\-ethe following: Xc( ,=0.475p i .
.
Figure4.15 System for example 1 ,
Changes in the network configuration between the t uosides (sendingand
receiving) will alter the value of Xcyand hence the expression for the electric
power transfer. The following example illustrates this point.
Example2
Assume for the system of Example 1 that only one circuit of the transmission
line is available. Obtain the relation between the transmitted electric powrer and
the angle 6.Assume othervariables to remain unchanged.
Solution
The network configuration presently offers an equivalent circuit as shoikw i n
Fig.4.17.
For the present we have
Figure4.16 Equiident circuit for example 1
E L 6
Figure4.17 Ecliii\dcnt circuit tor cuainplc 2
Thcretore.
0bserl.e that the ii1;ixiiiiiitii alue of the i i m cline i \ l o ~ v r than the OI c
corrc\ponding to the precious example.
4.9 CONCEPTS INTRANSIENT STABILITY
111order to gain ;in under\t;iiicling of the concept\ i n\ ol\cd iri transient \tahility
prediction. u e k v i l l concentrate or1 the \implil'icd netuorh con\isting of';i\eri:\
reactance connecting the machine and the int'inite hi \ . LTiider thew condition\
oiit.pouer cxpr t xi on recliice\ to
For.\iniplicityof notation u' e\ \ ' i l l ;issiiiiiestcadJr-state \.alues.An iiiiport;int
assumption that we adopt is that the electric changes iii\.olc,ed;ire much faster
th;in the resulting mechanical changes produced b), the gencrator/tirrbine spced
coiitrol.Thus \i'e;Issiiiiie that the rnechanical p o u wis ii constant torthe purpose
o f triinsieiit stabilitj,cdculations. The functions P,,,iiiid P, are plotted i n Fig.
-4.18.
The intersection o f the t Li ' o functions dcfincs t n' o values for 6.The l o~r er
\~:ilueis denoted by 6,,.Consequently,the higher is n: - 6,,according to t he \>m-
iiietrj of the ciir\'e. At both points P,,,=P, , t hat is ti16/dt'=0and L\,C s a j ' that
thc s)'stciii is in ecliiilibi.iitiii.
Power,P(P.u.)
?
(radians)
Figure4.18 Pouer-angle cun' e.
Assume that a change in the operation of the system occurs such that 6
is increased by a small amount A& Now for operation near 6,,.P,>P,,,and
c126/c/t'becomes negatiLreaccording to the swingequation. Thus 6is decreased.
andthesystem respondsby returning toitsstableoperatingorequilibrium point.
We refer to this as a stable operating point. On the other hand, operating at n-
6,,results in a system response that will increase 6and mo\re further from n-
6,).For this reason, we call x - 6, )an unstable equilibrium point.
I f the system is operating in an equilibrium state supplying an electric
pourer P,,,with the corresponding mechanical power input P,,,,,,then the corre-
sponding rotor angle is 6(,.Suppose the mechanical power P,,,is changed to P, , , ,
at a fast rate, which the angle 6cannot follow as shown in Fig. 3.19.I n this
case P;,l>P, and acceleration occurs sothat 6increases. This goes on unt i l [he
point 6, where P,,,=P,,and the acceleration is zero. The speed, houtvw, is not
zero at that point, and 6continues to increase beyond 6, . I n this region P,,,<P,
and rotor retardation takes place.
The rotor will stopat 6where the speed is zero and retardation uill bring
6down. This process continues on as oscillations around the new equilibriiini
point 6, . This serves to illustrate what happens when the system is sub-jectedto
a sudden change i n the power balance of the right-hand side of the sutingequa-
70
Power,P(P.u.)
(radians)
Figure4.19 Po\scr angle curve.
tion.The \ituation described abo\fewill occur for 5uddenchange\ i n P,;i\MI : I I
The \ j st em discussed in Examples 1 and 2 \er\re\ to illustrate t hi \ point. nhich
u e discii\\ further i n the next example.
E\i,?niplc.3
The \j'sterii of example 1 i \ deliirering iin apparent power of 1 . I p i ( at 0.85 PF
lagging with twocircuits of the line i n ser\ice.Obtain the \ource voltage ( ewi -
t a t i o n koltage)E and the angle 6under these conditions. With the \econd cir(*uit
open a\ i n Example 2. ;i new equilibrium 1 1 1 angle can be reached. Shetch the
power mgle curves for the t w' o conditions. Find the angle 6,,and the electric
POMer thatcan be tran\terredimmediately following the circuit opening, ;I\ \ L ell
i i ) 6, . A\ \ ume that the excitation \oltage remain\ unchanged.
S oILIti on
The power delivered is P, ,=S cos$, P,,=I . I x0.85=0.94pi
1
Using P=VICOS @, then the current i n the circuit is I= -~ cos ' 0.85
1 . 1
Thus we can Urite
71
E = V + j X l
1 +(1.1i - 31.79")(O.37SL90")
=1.28+j0.44=I .3S i 19.20"p i r
Therefore, E =1.35P.u.,6,,=19.20"
The power anglecurve for the line with two circuits according to Example
1 , is
P,,,=2.1053x 1.3Ssin6=2.83sin6
With onecircuit open,thenew poweranglecurveisobtainedas in Example
2, thus
3 P, , =I .43x 1.3Ssin6=1.93sin6
Thetwo power angle curves are shown in Fig.4.20.
From inspection of the curves, we can deduce that the angle 6,.can be
obtained from
P,,,=P,sinS,(curveB)
0.93=1.93sin6,
6, =29.15"
We can obtain the value of electric power corresponding to with one
line open as
P,.,o=1.93sin19.2"
=0.63pi4
4.10 A METHODFORSTABILITY ASSESSMENT
Topredict whetheraparticularsystemisstableafteradisturbance i t isnecessary
to solve the dynamic equation describing the behavior of the angle 6immedi-
ately following an imbalance oradisturbance to the system. The systemis said
to be unstable if the angle between any two machines tends to increase without
limit. On the other hand if under disturbance effects, the angles between e\.erq'
possible pair reach maximum value and decrease thereafter, the system is
deemed stable.
Angle. 13
(radians)
Assuming as we hwc :ilrcady done that the input is constant, ncglipi3le
doniping iuid coiist;itit source \ dtagc behind the tnitnsient reactance. the angle
hetwccn t wo machines eitherincrease indefinitely or oscillates iif'tcr all disturh-
;iiicch h;i\.coccurred. Thcrcforc'. i n the c;isc of tw:o machines, these \ \ i l l either
f d l out of' step on the first swing or never. Here the obscrvution that the iiia-
chinehangulardifferetices stay constaltcan be tiikt.tl ;IS ill1 indication 01' hystctii
\tahility. A simple method lor determining htiibility knowii iis the c'cluiil-; re;i
tiicthotl is nvnilable. We will discussthis here.
Theh\vitig ccluation for ;I machine connected to ;in infinite busci111he writ-
tell a s
We obtain an expression for the variation of the ringirlar speed o wi t h ?.
by noting the alternative form
P
0tlo,= (16
M
Integrating, assuming cc)=0and integrating the abo\,eequation, \IT obtain
or
The above equation gi\ves the relative speed of the machine M i t h r-eclpcctto
a reference frame niming at a constant speed (bydefinition of the angle 6).I f
the \ystem is stable,then the speed must be zero when the acceleration i \ either
zero or i \ opposing the rotor motion. Thus for a rotor n4iich i \ accelerating. the
condition for stability i \ that a value of 6,exists such that
This condition is applied graphically in Fig. 4.2I where the net area under
the P,, - 6ci r r ~~e reaches zero at the angle 6as s how. Obsenve that at 6,,.P, is
negative and conseqirently the system is stable. Observe that the area A equals
A! as indicated.
The accelerating power need not be plotted to assess stabilitj,.Instead. the
same information can be obtained from aplot of electrical and mechanical pout-
ers. The former is the power angle curve and the latter is assumed constant. 111
this case the integral iiiay be interpreted as the area between the P, c i r r \ ~ and
thec~rr~re of P,,,both plotted versus 6.Thearea tobe equal to zero. must consist
of a p0sitii.e portion A I.for which an equal and opposite negatiLre portion of,4
must exist, for which P,,,<f,,. This explains the term equal-area criterion for
stability.This situation is shown in Fig.4.22.
I f the accelerating power reverses sign before the t ~ oareas A and .4?;ire
equal. synchronism is lost.This situation is illustrated i n Fig.3.23.The arekiA:
is smaller than A , and as 6increases beyond the value \!,here P(, re\erxessign
again, the areaA j is added toA , .
71
A
Figure4.21 S t;Ihi Ii t J condition f or accelerating rot or.
Power,P
P m
-
0 6,
b
Power
Angle,8
(radians)
Figure4.22 Equal -areac*r.itcrioritor \tabilit!,.
75
Accelerating
Power,P,
I
Figure4.23 Accelerating powrer as a function of the torque angle.
Example4
Consider the system of the previous three examples. Determine whether the
system is stable for the fault of an open circuit on the second line.If the system
is stable, determine 6,,the maximum swing.
Solution
From the examples given above we have
6,,=19.2"
6r=29.50'
The geometry of the problem is shown in Fig. 4.24. We can calculate the
areaA immediately:
?UIS
AI =0.94[29.15- 19.20In - 1.93sin6t / S
1*0 I02 0
Observe that the angles 6, and 6,)are substituted for i n radians. The re-
sult is:
AI =0.0262
The angle 6,.is
76
L
Power, P (P.u.
2.84
1.93
0.94
0
6,= 1x0 - 6 , = 15o.xf;
This clcurly gives
;init the system is stable.
The iingl~. 6,. i s obtained hq solving torAI = A, . Here we get:
77
This gives some algebra
I .93c od, +0.01646,=2.1376
The solution isobtained iteratively as 6,=39.39"
This example shows the application of the equal area criterion to the case
of a generator supplying power to an infinite bus overtu'oparallel transmission
lines. For the loading indicated above the system is stable.The opening ofone
of the lines may cause the generator to lose synchimism e\'enthough the load
couldbe supplied overasingleline.Thefollowingexample illustrates thispoint.
Example5
Assume that the system in Example 3 is delivering an actiLre power of 1.8pi 1
using the same source, voltage E, asbefore. Determine urhetherthe sqrstem u?ill
remain stable after one circuit of the line is opened.
Solution
We have for the initial angle 6,,
I . 8=2.84sin&,
6, ,=39.33"
The angle 6,.is obtained from
I .8=1.93sin6,
6 , =68.85"
The areaA , is thus:
OX x s
A , =I .8(68.85- 39.33jn - 1 1.93sin6(16=0.13
I 8o 31) 73
The areaA_.is obtained as:
ii
It
A, =j 3.86\inti(16- 1.816,- 6,1- =o.o6
180
We note thatAI >A: , and the system is therefore unstable.
I f ii three-phase \hart circuit took place at ;i point on the extreme end o.
the line. there i \ some impedance bet\veen the generator bus and the load ( i n f i -
tiite)h i \ . Therefore, \ome p o ~ e ri \ tran\mitted cvhile the fault i \ \ t i l l on.Tht.
\ituation I \ \iiiiiIiir to the one\ anal ) l edabo\ eand \ie ii\e the follouinge'iamplt:
to illu\trate the point.
Ewmple6
.A generator is deli\,ering 25%- o f PI,,,,to an infinite bus through ;i transmission
line. A f a ul t occurs such that the reactance bet uwn the generator and the bii4
i \ incrcased to tuvtimes its prefuult value.
I . Find the 6,,before the faiilt.
2. Shoci,graphicallq,c i , hat happens when the f i i ul t is sustained.
3. Find the niiisiriiiiin \raliie of 6 s\\tingi n case of ;I sustained l'ault.
Solution
Figure4.25illustratesthe situation forthis example.Theamplitudeof thepo\i t r
anglc c i i r ~e tttith the f;iiiIt su\tained is hal f o f the original \-alue.
Before the t'itiilt\+eha t e
At the t ' ui i l t instant,\4fe get
A\ before. the \tability condition yields
Hence
0 . 5 COb 6 , +--<6,=0.5473 K
7'
Bq trial and error
6, =46.3"
A
Power,P(P.u.)
0.50
0.25
Torque
0
Angle, S
(radians)
Figure4.25 Power anglecurves for example6.
The following example illustrates the effects of short circuits on the net-
work from a stability point of view.
Example7
Thesystemof the previous examples deliversapower of 1.Op i when subjected
to a three-phase short circuit in the middle of one of the transmission circuits.
This fault is cleared by opening the breakers at both ends of the faulted circuit.
If the fault is cleared for6,=50, determine whether the system will be stable
ornot. Assume the same sourcevoltage E ismaintained asbefore. I f the system
is stable, find the maximum angle of swing.
Solution
Thepoweranglecurves havebeen determined fortheprefault network i n Exam-
ple I and for the postfault network in Example 2. In Example 3 we obtained
E =1.35plr
j0.225
Figure4.26 NctLforh coni'iguration ciuring rhe faitlr.
Theret' rm
During the fault the netnvrk offers ;I different cont'i~uratioti.L+ hic.t- i \
\ho\\11 i n Fig. 4.26. We \ + i l l need to reduce the netnorh in wch ii u ; i j ;I<, to
obtain ;I clear path l'rotn the +oiirct'to the infinite h i \ . We dot hi \ bj u\iiig ;I1'
- A tr~iii\t'ortnationii\ indicated in Fig.4.27.
)(0. 25)=I , 2
x=0.45+0.25+
(0.45
0.235
and fri i i l t pourerangle curve isgiven by
f =1, I3sin6
The three power angle curl'esare shown i n Fig. 1.28.
The initial angle i h gilren by the equation
1.O =2.84sit&
6,,=20.62"
The clearing angle is 6,=SO"
Power
P(P.U.
2.a
/
\
I .93
1.13
I .a
y i i i II I
I I , I I
. Torque
6 f ~ n g ~ e . S
(radians)
Figure4.28 Pre-fault. during fault, and post-fault poww angle CLII'VL'St or t.xii11113lc 7
S2
The areaA I can thus be crilculated as:
The maximum area A?is obtained using the angle 6,
I =1.93siii6:
6;'=31.21"
6,=I80- 6;=148.79'
NOW
We note that A I >A?,and the system is theretore stable.
Tocalculate the angle o f iiiiisiiiiurn swing we ha\fe
Hence
Bjr trial and error
6,=66.3''
4.11 MATHEMATICALMODELSANDSOLUTI ON METHODS
INTRANSIENT STABILITY ASSESSMENT FOR
GENERAL NETWORKS
I t is coiiinioii practice to model static equipment in the transmission sj .!ten1
by lumped equicralent pi parameters independent of the changes arising i n the
generati ng and load eqUi pment, This approach i s enip1oyed i n mU1t imiichine
stabilitj programs because the inclusion o f time Lrarying parameters ~voulcl pro-
duce major computational difficulties. Moreover, frequency, the most obtrious
variable in the network, usually varies by only a small amount and thus, the
errors involved are insignificant. Additionally. the rates of change of network
variables areassumed to avoid the introductionof differential equations intothe
network solution.Thetransmission network can thusbe represented i n the same
manner as in the load-flow or short-circuit programs, that is, by a square com-
plex admittance matrix.
Thebehavior of the network is described by the matrix equation:
I, , , =YV
where I,,, is the vector of injected currents into the neturork due to generators
and loads Y isthe admittance matrixofthe network, and V isthe vector of nodal
voltages.
Any loads represented by constant impedances inay be directly included in
the network admittance matrix with the injected currents due to these loads set
to zero. Their effect is thus accounted for directly by the network solution.
4.11.1 System Representation
Two alternative solution methods are possible. The preferred method uses the
nodal matrix approach. while the alternative is the mesh matrix method. Matrix
reduction techniquescanbe used ifspecificnetwork information isnot required.
but this gives little advantage as the sparsity of the reduced matrix is usually
very much less.
Nodal Matrix Method
I n this method, all network loads are converted into Norton equi\-alentsof i n-
jected currents i n parallel with admittances. The admittances can be included i n
the network admittance matrix to form a modified admittance matrix which is
then inverted, or preferably factorized by some technique so that solution at
each stage is straight forward.
The following solution process applies:
I . For each network load, determine the injected currents into the modi-
fied admittance matrix by solvingtherelevant differential andalgebraic
equations.
2. Determine network voltages from the injected currents using the Z-
matrix or factors.
As the network voltages affect the loads, an iterative process is often re-
quired, although good approximations can be used to avoid this.
With the Nodal Matrix method, bus voltages are available directly and
branch currents can be calculated if necessary.
4.11.2 SynchronousMachineRepresentationinthe Network
The equations representing ;I )syiichroiioiis miichine ;ire gi\ren i n the form of
The\,cninvoltages behind its impedances. This must be modified to ii currint
source in parullel urith ;in admittmce using Norton's theorem. The xhi i t t mce
of the machine thus formed may be added to the shunt admittance of the 11i;i-
chine bus and treated as a net uwk parameter-. The \ector /,!/ thus contains the
Norton equiktalent ciirrcii ts ofthe sj'iichroiious iiiachines, The sj'iichronous 11I a-
chine equations are kvritten i n a frame OF reference rotating ivith its o\ vn rot or.
Thc real anci imaginary coniponenth o f the iietwork equations, iis giLrt.11 i n Fig.
4.29. art' obrai11ed !I-( )I I I tht' fo1Io c i ' iiig transtorniation
The in\,erserelation is
Network
t
imaginary axis
Machine Machine
direct axis
The transformation also applies to currents.
When saliency isaccounted for,the subtransient and transient reactances i n
direct andquadrature axesframes aredifferent, and theNorton shunt admittance
will have adifferent \ due in each axis, and when transformed into the netLivrh
frame of reference. will be time varying. To circum\rent this difficult). a con-
stant impedance is used while modifying the injected current.
4.11.3 Load Representationinthe Network
To be suitable for representation i n the overall solution method, loads niust be
transformed into injected currents intothe transmission netM-orkfrom kvhich the
terminal voltages can be calculated. A Norton equivalent model of each load
niust therefore be introduced. I n a similar way to that adopted for synchronous
machines. theNorton admittancemay be included directly i n thenet1iu-kadmit-
tance matrix.
A constant impedance load is therefore included i n the network admittance
matrix and its injected current is zero. This representation is extremely simple
to implement. causes no computational problems, and impro\.esthe accurricyof
the netuforksolution by strengthening the diagonal elements i n the :idmittatice
matrix. Nonimpedance loads may be treated similarly. I n this case the stead>,-
statevaluesof voltage and complex powerobtained from the load t l oware uscd
to obtain a steady-state equivalent admittance (7,)) which is included i n the net-
urork admittance matrix [Y ] . During the stability run, each load is s o l k ~ dse-
quentially along with the generators, etc.,to obtain a new admittance (r). i.e.:
The current injected into the network thus represents the de\siation of the
load characteristic from an impedance characteristic.
By converting the loadcharacteristic to that of aconstant impedance, M hen
the \toitage drops below some predetermined value (I/,,,,,,),the irijected current
i \ kept relatikrely small. An example of a load characteristic and itscorrespond-
ing injected current is shown in Fig.4.30.
In an alternative model the low-voltage impedance is added to the net\sorh
and the injected current compensates forthedecriationfrom theactual character-
istic. In this case, there is a nonzero injected current i n the initial steady-state
operational condition.
86
I I )
power
Figure 4.30 Load and injecled currents for a con4tant type load Uith l o ~ ' iolta;e
adju\tiiient. ( a ) Load current. ( h ) Injected current.
4.11.4 System FaultsandSwitching
I n general most power system disturbances to be studied will be caused hy
changes in the network, normally caused by faults and subsequent switchiig
action. Occasionally the effect of branch or machine switching will be consid-
ered.
Although faultscanoccur anywhere inthe system, i t ismuch easiercomru-
tationally to apply afault toabus. In this case, only the shunt admittance at the
bus need be changed, that is. a modification to the relevant self-admittance of
the Y matrix. Faults on branches require the construction of ;i dummy busat the
f aul t location and suitable modification of the branch data unless the distarce
between the f wl t position and the nearest bus is small enough to be ignored
87 Philosophy of Secirr-ityAssessriierir
Theworstcase isathree-phase zero-impedance faultandthis involvesplac-
ingan infinite admittance inparallel with theexisting shuntadmittance. In prac-
tice, a nonzero but sufficiently low-fault impedance is used so that the bus
voltage iseffectively brought tozero.This isnecessary tomeet therequirements
of the numerical solution method.
The application orremoval of a fault at an existing bus doesnot affect the
topology of the network and where the solution method is based on sparsity
exploiting ordered elimination, the ordering remains unchanged and only the
factors required for the forward and backward substitution need be modified.
Alternatively the factors can remain constant and diakoptical techniques can be
used to account for the network change.
4.11.5 BranchSwitching
Branch switching can easily be carried out by either modifying the relevant
mutual- and self-admittances of the Y matrix or by using diakoptical techniques.
In either case, the topology of the network can remain unchanged, as an open
branch is merely one with zero admittance. While this does not fully exploit
sparsity, the gain in computation time by not reordering exceeds the loss by
retaining zero elements, in almost all cases.
The only exception is the case of a branch switched into a network where
nointerconnections existed prior tothat event. In this case,eitherdiakoptical or
reordering techniques become necessary. To avoid this problem, a dummy
branch may be included with the steady-state data of sufficiently high imped-
ance that the power flow is negligible under all conditions, oralternatively, the
branch resistance may be set negative to represent an initial open circuit. A
negative branch reactance should not be used asthis is a valid parameter where
a branch contains series capacitors.
Where a fault occurs on a branch but very close to a bus, non-unit protec-
tion at that bus will normally operate before that at the remote end.Therefore,
there will be a period when the fault isstill being supplied fromthe remote end.
There are two methods of accounting for this type of fault.
The simplest method only requires data manipulation. The fault is initially
assumed toexist at the local bus rather than on the branch. When the specified
time forthe protection and local circuitbreaker tooperate has elapsed, the fault
is removed and the branch on which the fault is assumed to exist is opened.
Simultaneously, the fault is applied at the remote bus, but in this case,with the
fault impedance increased by the faulted branch impedance, similarly the fault
is maintained until the time specified for the protection and remote circuit
breaker to operate has elapsed.
Thesecond method isgenerally more involved but i t is better when protec-
tion schemes are modeled. I n this case, a dummy bus is located at the fault
position. (eventhough i t is close to the local bus) and a branch with a \U-:,!
small impedance is inserted between the dummy bus and the local bus. Th:
faulted branch then connects the dummy bus to the remote bus and the branci
shunt susceptance originally associated with the local bus is transferred to the
dummy bus.This may al l be done computationally at the time when the fault i s
being specified.Thetwo branches can now be controlled independently by sui,-
able protection systems.An advantage of this scheme is that the fault durativn
need not be specified iis part of the input data. Opening both branches et'!'el:-
ticely isolates the fault, \+rhichcan remain permanently attached to the dunin
bus. or if auto-reclosing is required. i t can be reiiiowd automatically after ;I
suitabIe deioni~ation period.
I f the network is not being soliredby ii direct method, the second method
will probably fail. During the iterati1.esolution of the network, slight volta,;e
errors nrill ciiuse large currents to flow through ii branch wi t h ;I 1w-ysmill
impedance. This will slow convergence and i n extreme cases uill c~iuse diwr-
gence. With ii direct method, based on ordered elimination, ;in exact solution o f
the bus Lwltagesis obtained forthe iiijected currents specified at that particuhr
iteration.Thus,pro\.idedthat the impedance is not s o small that numerical prob-
lems occur L+,hen calculating the admittance, and the subsequent fwtors for 1he
forward m c l backward substitution, then convergence of the ow- al l soluti oii
betu.een machines mc l netL+x)rk will be unaffected.Thek-aluc. o f the lo\+!-impcd-
ance branch between the cluriiriiy and local bus may be set at II fraction of the
total branch impedance, sub-jectto a minimum value. I t this fraction is u n k r
0.001.the change i n branch impedance is \ w y small compared to the iiccuric>r
of the netuwk data input and i t is unnecessary to niodify the iiiipedance of'the
branch from the remote to the dummy bus.
4.11.6 MachineSwitching
Machine su,itchingmay be considered, either as a network or machine opera-
tion. I t is ~i netu.ork operation if a duiiiiiiy bus is created to which the machine
is connected. The dummy bus is then connected to the original machine hub b!,
;I I OU*- impedance branch.
Alternati\rely. i t may be treated as a machine operation by retaining the
original ns t Nvr k topology. When a machine is sw.itchedout, it is necessaiy to
reniove its iri-jected currcnt froni the network solution. Also a ny shunt admit-
tance included i n the network Y matrix, which is due to the machine mu!,t be
ren10ved.
Although adisconnectedmachine canplay nodirect parti nsystemstability,
its response should still be calculated asbefore, nrith the machine statorcurrent
set to zero. Thus machine speeds,terminal \.oltages.etc.. can be obser\xd ci'eii
when disconnected from the system and in the ek'ent of reconnection, sensible
results are obtained.
When an industrial system isbeing studied many machines may be discon-
nected and reconnected at different times as the ~ ~ ~ l t a g e le\rel changes. This
process will require many recalculations of the factors in\-ol\,edi n the forumd
and backward substitution solution method of the netmwk.However, these can
be avoided by using the method adopted earlier to account for synchronous
machine saliency. That is, an appropriate current is iiijected at the relt'\mt
buses, which cancels out theeffect of the shunt admittance.
4.12 INTEGRATIONTECHNIQUES
Many integration methods have been applied to the power \ysteni tran\ient \ta-
bility problem, and the principal methods are discus\ed nou.
4.12.1 PredictorCorrectorMethods
These methods for the solution of the differential equation
.sY =F ( Y . X ) with Y ( 0 )=Y, ,
and X ( 0 )=X, ,
hakre all been detreloped from the general k-step finite difference equation:
Basically the methods consist of a pair of equations, one being explicit
( PI ,=0)to give a prediction of the solution at t(rl +I ) and the other bein,0 1111- .
plicit ( PI ,f0)which corrects the predicted value. There are a great Larietj of
method\ aLrailable, wch as hybrid methods, the choice of which being made bq
the requirements of the solution.
I t isusual for simplicity to maintain aconstant step length with these meth-
od\ if X >2. Each application of a corrector method inipro\e\ the accurac~ of
the method by one order, up to a maximum given by the order of acc~iracy of
the corrector. Therefore, if the corrector is not to be iterated. i t i \ common to
use a predictor with an order of accuracy one less than that of the corrector.
The predictor is thus not essential asthe value at the preLriousstep may be u\ed
as a first crude estimate,but the number of iterations of the corrector may be
1arge.
While for accuracy, there is a fixed number of rele\mt iteration\, i t is
desirable forstability purposes to iterate tosome predetermined levelofconver-
gence.Thecharacteristic root ( z , )of apredictor orcorrector when applied tothc
single variable problem
may be found from
Applying acorrector to the problem defined and rearranging gives:
the solution to the problem becomes direct. The predictor is now not necessaiy
as the solution only requires information of 1' at the previous steps. i.e.. .it
\v(ii =i +1 ), for i =1 , 2,.....k.
Where the problem contains two variables, one nonintegrable, such that:
then
where
Although and ~ i , , , ~ areconstant ata particular step, thesolution is itera-
tive.
Strictly in this simple case, x( i i +1 ) could be eliminated but in the general
multivariable case this is not so.Theconvergence of the method is now a f uIC-
tion of the nonlinearity of the system. Provided that the step length is suffi-
ciently small. a simple Jacobian form of iteration gives convergence i n only a
few iterations. It isalsopossible to formaJacobian matrix and obtain asolution
by a Newton iterative process, although the storage necessary is much larger
and the step length must be sufficiently small to ensure convergence.
For a multivariable system, the following twoequations are coupled
.sY =F( Y , X) with Y ( 0 )=Y,)andX ( 0 )=XII
and the solution of the integrable variables is given by the matrix equation.
Theelements of the vector L' , , +~aregiven by the vector form of
and the elements of the sparse matrix are given by
The iterative solution may be started at any point in the loop, if Jacobian
iterations are used.
4.12.2 The Euler Method
Consider the following ordinary differential equation:
Let x(t)be thestate vector of this nonlinear differential equation, which is tobe
solved by an appropriate integration technique. The Euler method utilizes a
predictor function based on the Taylor seriesexpansion of .r(t +A[), uthereAt is
the step size.As such,by neglecting the higherorder terms in the seriesexpan-
sion, we obtain the generalized Euler's formula:
d.r(t )
s ( t +At ) =x( t ) +x(t). Ar where i ( r ) =__
at
The method is not often used for real-time applications in power systems, as i t
iscomputationally burdensome. Also,the accuracy of this model for integration
is sacrificed by the truncation of O[At)']and higher order terms done in the
Taylor series expansion. Figure 4.3I outlines an example of Euler's integration
technique to power system dynamic stability assessment.
4.12.3 The Modified Euler Method
Reconsider the following problem tobe solved:
i ( t )=f I ( s ) , t ]
SetLoopControl time:T,, &&
SetIntegrationStepsize: At
FlowCalculationsandobtainP,(t),
themachineinjectionin p.u.MW
Solvethesystemof differential
equationsfor allmachines:
?f
w( t ) =--(& - 4"
H
I
SetFaultTime:tf=t
s = o - w
R'
' w
R
=2nf
Perform'duringfault'PowerFlow
Calculations
Figure4.31 The transient stability algorithm
@ Q
UsingEulerSMethod,Solvethe
machineangles:
6( t )=u(t)-2 q
6(t+Ar)=&r ) + 8(f)Ar
YesI
SaveDisplay
Final Results:
~~
I
~ ~ ~ ~ ~ ~
SetFaultClearingTime:&=t
I
TorqueAngle,
MachineSpeed,
Generated
Obtainnewstateof themachine
Power,Times(t,
andorthenetworkperturbations
tf.tmarretc.) and
i
soon.
Figure4.31 Continued.
where x ( f ) and.f'(x,t ) are scalar orvector quantities. Expanding. d f ) to the riglit
and left using the Taylor expansion yields:
If the O[(At)'Jand higher order terms are truncated for a small step size, i t c;in
beshown that the modified Euler formula is obtained asfollows:
Alternatively, we may bfrite:
Note that the modified Euler inethod is not self*-startingand thus requires an
initial prediction on the state variables in the \rector solution.
4.12.4 Trapezoidal Method
The trapezoidal rule is anonself-starting integration technique that is somewhat
related to the modified Euler method and is based on the geometric interpreta-
tion of the problem. The order of truncation i n the Taylor series expansion of
. v( t kAr) is O[(A?)']. Thetrapezoidal formula is:
The merits of the trapezoidal method isevident in the accuracy brought forward
iis a result of the truncation of the O[(At)'] and higher order terms.
4.12.5 Runge-KuttaMethods
Runge-Kutta methods are able to achieve high accuracy while remaining sirgle
step methods. This is obtained by making further evaluation of the functions
within the step. Here we present a class of self-starting prediction formulae,
which are applicable i n the assessment of transient analysis of rotating na-
chines, in which case we are solving the swing equation of the generators.
The numerical integration of the ordinary differential equation given b).:
by a predictor generally involves calculation of xn+1 as ii function of x,,, Y,, I,
x,,-:, ....fll. f,,-l,fl,-., ...,and tl,. I f the predictor is self-starting, i t must be free
of terms in xn-l ,x,,-?,...,f,, fll-l, fIl-? and soforth.That is,the next statesolution
sould be of the form: x,,+~=F(x,,,fl,,t"). Now, the equation below summarizes
this requirement and is called the Runge-Kutta predictor formula.
where k( i ) =h f
i
r ( n ) +c,h, ! ( / I ) +
111
/=I
and the sum of all the weights equal to unity, i.e., w,=1
I= I
The coefficients are uniquely determined, giving rise to various orders and
associated approximate or accurate models of the Runge-Kutta predictor for-
mula. Table 4.1 summarizes the 2"d,3rd,and 4IhorderRunge-Kutta formulae.
Being single-step these methods are self starting and the step length need
not be constant. Ifjis restricted sothatj <i then the method is explicit and cI
must be zero. Whenj is permitted to exceed i, then the method is implicit and
an iterative solution is necessary.
Table4.1 Summary of the RK-2. RK-3, and RK- 4Predictor Forniulae
Runge-Kutta Predictor Formulae Coefficients
RK-2
.Vil+] =.Vll +-
h
( k"' +k' ?' ]
2
RK-3
.Vfl+l =.Vfl +-
It
[k"'+ +k'?']
6
RK- 3 k' '=F(.vl,,t1,)
=.vil +-
It
(k"'+2k'!'+2k'" +kf4']
h
6
k'?'=F! . Vl t +- k " h +-
i
2 2
Explicit Runge-Kutta methods have been used extensibdy in transient sta-
bility studies. They have the adkmtage that a packaged integration method is
iisually available or quite readily constructed and the differential equations at-:
incorporated with the method explicitly. I t has only been wi t h the introduction
of nore detailed systemcomponent models with very small tiiiie constants, th:,t
the problems of stability has caused interest i n other methods.
Again. theadvantage of the Runge-Kutta techniques is that theyareall seli'-
starting. The propagated error from one iteration to the next does not increa>e
rapidly, thus the inethod is said to be stable. The exception is the solution 10
"stiff' prohleins, &(here the solution may diverge unlessa small step size is used.
(Thestiffnessof theordinary differentialequationsot'the systemcan be iiie;isiirLd
iis ii functiuono f the ration between the smallest to the largest eigeinduesof the
linearized system or eigenvector analysis of the Jacobian matrix.) Ne\wtheless.
they arenot ~ e r y attractivetopourersystemengineersiisk powwflowcalculatioris
must bedone at every iteration.where k is theorder of the Runge-Kutta forinula.
The tollou,ingesnmple demonstrates the RK-4 application.
Exc?niple
Application o f the 1"' er \ J \ t cin orderRunge-Kutta integration techniqiie to p o ~
dynamic \tability asws\iiient.The algorithm is giLen belou.
I . Start the RK-4 Subroutine.
2. Initialize all RK- 3dependencies.
COllIlt, d=0: StepSize.At
Time. t =0: Miiximum Time,t,11,1,
Maximum iteration=itniax
3. Sol\e the initial pouer flow and the machine equation\, and obt.iin
the generator torcllie angle (6(0)). terminal voltage\. bu\ in.jection\
P;"'(o).etc.
4. InitialiLe the RK-4 Coefficients: k , =I, =0 for all i E { l , 4 }
5. Set the e\tirritrfc iirc/c.\ to: i =d+I .
6. Perform RK- 4calculations (obtainingthe estimate\).
97
7. Increment count: d=d+I .
8. Increment the estinicitr i1ii1e.vto: i =i +I .
I
9. Update the torque angle delta to:6( t )=& ( I ) +5k, and sol\^ the pom'er
flow for I v , l,Py"(t),etc.
10. Compute the new coefficients.
I
k, =( ~ ( 0 ) +- I , - ) - 2Ff)At
2
I , =w( f i l i - f : "' ( t ) ) At
2H
I I . If d<3. then goto step7,otherwise continue.
12. Compute the final value of the power angle and the machine speed at
t =t +At
1
6(t+At ) =6(t )+-(ki +2k:+2 k , +k, )
6
I
CO(t+At)=CO( t ) +-( 1, +21, +21,+1 4 )
6
13. Reset the count: d=0
14. Compute the final power flow forthis time interkal toget 1 V,1 . P?( t1.
etc.
IS. Increment the timer ( t =t+At )and make the following comparisons:
-If ( t <t,,,,,,)and ( N 2itrncix) then flag the user: "Maximum
Number of iteration reached!" and goto step 16.
Elseif ( t <t,,,,,,)and ( N <itmax)then increment the iteration count
to: N =N +1 and goto step4.
Else Flag the user: "Run time value of tmax reached!" and con-
tinLie,
16. Display allresults, power angle,machinespeed.generatorp o ~ ~ e r . gen-
erator terminal voltage, etc.
17. End of subroutine.
4.12.6 TheSamplingMethod(James J.Ray;JamesA. Momoh, etal.)
Another promising development in this area is the sampling method. I t u'as
developed at Howard University with collaboration from the Energy Systems
Network Laboratory (ESNL)in 1989.The sampling method, also known as the
"theta" method. takes advantageoftheexpandedTaylorSeries in itsformulatior
and approximations. It was incorporated in a stand-alone program capable oi'
performing stability studies and fault analysis on a wide variety of electric,
power systems. The sampling method has been tested against the classical inte-
gration methods. A 4:1 speed advantage was observed, without sacrificing thr:
accuracy brought forward in the results. Extended application of this method i r i
power systemstabilitystudies isbeing developed forcommercial purposes,both
as a design tool and asa computational support system.
4.13 THETRANSIENT STABILITY ALGORI THM
An overt'iew of the structure of a transient stability program is given i n Fik.
4.32. Only the main parts of the program have been included, and as can be
seen,the same system may have several case studies performed on i t by repear-
edly specifyingswitching data when nofurtherswitchingdata isavailable. Cor-
trol returns to the start to see if another system is to be studied. With care,the
program can be divided into packages of subroutines each concerned with only
one aspect of the system.This permits the removal of component models when
not required and theeasy addition of new models whenever necessary. Thusfor
example,the subroutines associated with the synchronous machine, the AVRs,
speed governors, etc.,can be segregated from the network. Figure 4.32showsii
more detailed block diagram of the overall structure where this segregation is
indicated. The diagram is subdivided into the five sections indicated in Fi;.
4.33. While the block diagrams are intended to be self-evident several logic
codes need to be explained. These are as follows:
KASE
This is the case study number for a particular system. It is initially set to zero
and increniented by I at the end of the initialization and at the end of each case
study.
KBl FA7
The sparse vectored inverse of the nodal network matrix isobtained using three
bifactorization subroutines. The first and secondsubroutines are integerroutires
which determine bus ordering and nonzeroelement location.The code KBIFA1
is set to unity if i t is necessary to enter these two subroutines, otherwise i t is
set to zero.
r+=? Read in steady-state system data
Section1
Section2
Section3
r
Either Solveformachinesand network
or recalculatenonintegrablevariables
(initiallyandwhenswitchinghasoccurred)
Section4-
1.Print outresults (if necessary)
2.Makepower balance checkof
initialconditions (if necessary)
NO
1
YES NO
Section5,
- Store initial Reset initial
conditions conditions
Figure4.32 The transient stability algorithm.
I 0 0
Synchronousmachine AVR calculations wed governor
Network calculations
calcuyns ,~ ,~A ,f c al c ul j y~ns ,
KASE =0
KBlFAI =0
KBIFA3=0
input
YES
NO
V
Figure4.33a Section I
KBIfA3
Theelements of the spar\e-vectored inverse are evaluated i n the third bifactor-i-
zation subroutine. The code KBIFA3 is set to unity if i t i \ necessary to en er
t hi \ wbroutine, otherwise i t i \ set to Lero. When KBIFA3 i \ unity, i t indica e\
that a netiborh discontinuity ha\ occurred and hence it i \ al w u\ed for t bi \
purpcx.
Time
The integration time.
H
The integration step length. Like KBIFA3,it is also used t oindicate adisconti-
nui t y when i t is set to zero.
4
Switching
Determine bus order
and nonzero element
locationfor bifurcation
KBIFA3=1
TIME=O
+ +
Is KASE2I Is KASE2I
NO NO
andist haeno andist haeno
switchingat TI ME=O? switchingat TI ME=O?
YES YES
**
I
I
K B I F G P syncmachine *AVR initial
KBIFm Syncmachine
t,conditions.
initial initial
YES
conditions
conditions
+
Speedgov.
1
T initial
initial
conditions conditions
I
4
numericalpart
ofbifurcation
Print out initial
conditions
whenU S E > 1)
Figure4.33b Section 2.
I 02
v
A
-7-T Determine
switching ops.
if branch change
set KBIFA3 = 1
if new branch
set KBIFAI = 1
I
ES
-
Determine bus
order and non zero
element location
for bifactorization
1 KBIFA3 = 0
part of
KBIFA3 = 1
Figure4. 33~Section 3.
Philosophy of SecirrityAxsessnierit
I
Solve for network
KBIFA3 = 1
<>-
TIME =
E- PRINT TIME
YES
Print out
Print out busbar
Print out AVR results
and branch results
sync. machine
if required
results
4 speed gov.
a results if required
Set flag for
step doubling
KBIFA3 =0
I
Figure4.33d Section 4.
I04
A
v
YES YES
r 1
NO
Perform power balance
check to confirm
NO
'End of case '
TIME =0
KASE = KASE + 1
1 KASE = 1 ?
r-? Store initial
steady- state)
(steady-state)
conditions
Figure4.33e Section 5.
PKlNTTlME
The integrationtime :it which the next printout ot'result\ is required.
MAXTIME
The predet'ined mixiniiiiii integrationtime f or the case study
Start solution
Lr'
Calculateconstants
nonintegrable variables
,-,-,-,,,,-,,---'
+I speedgov. calc
Inot usuallyrequired
a- ----------------1
Evaluateintegrable
Same forcach A
varaibleusingalgebraic b
Sameforeach
formof mpezoidalmethod4
speed gov.
I
[HALF= 0
Figure4.34a Section I
ITMAX
Maxiinuin number ot'iterationsper stepsince l ast printout of results.
Note that iiiany data errorchecksarerequired i n a program of this t j p but
they ha1.ebeen omitted from the block diagram forclarity.
i
106
Figure4.34b Section 2
I
107
v
A YES
NO
Re-evaluate conditions
at beginning of step
H = H / 2
IHALF =IHALF +I
NO
'Not converging '
TIME = MAXTIME
I
Figure4 . 3 4 ~ Section 3.
I ox
StrcictiireotMachine'2ndNetivork IterativeSoI~iti017
The~- uc t ur e of this partof theprogram requires furtherdescription.Tuoforris
o f \olution arepossible depending on whether an integration stepis beingetralu-
ated orif the nonintegrable Lariablesarebeing recalculated afterii discontinuitq,
A bloch diagram is gi\en in Fig. 4.33.The ridditional logic codes uwd i n this
part of the program are:
ERROR
The niiiXiniuin dit'ference betueen an) integrable 1 ariable from one iteration to
;in() ther.
ITK
Nu inber of iterations reqiiired for solution.
IHALF
Number ol' inimediate stephalving required for the solution.
TCILERANCE
Specified niiiYifnuni \alueof ERROR for concergence.
I f coil\erpence ha\ not been achieved after a \pecified number of iteration\
the ca\e study is tei-ininated.Thi\ is done by \etting the integration time ccual
to the iiia\iiiiuiii integration time. The latest rewlts are t hu\ printed out ;iild ;i
neu c;i\e \tudj i \ atternpted.
CONCLUSION
This chapter dealt uriththe philosophy of securit), assess~nent based on fre-
quency domain models arid equal area criterion concepts. I n particular. we de-
fine the conventional ingredients for power systeni stability including appica-
tions of the swing equation and its alternate forms.
Frequency domain models of synchronous rnachines introduced the idca of
subtransient. transient, and steady statereactances. Models forasalient poleand
round wound synchronous machines were discussed.The equal area criteria and
its applications were discussed. The chapter concludes with the treatnieiit of
transient stability for the general network case, including ;I floMxhart of ii tran-
sieri t stiibi1it4' assessment prograni.
Again, the reader is reminded that additional information inay be obti.ined
from the list o f references and the annotated glossary of terms.
5
AssessingAngle Stabilityvia
TransientEnergy Function
INTRODUCTION
I n the actual operation of an electric power system, the parameters and loading
conditions are quite different from those assumed at the planning stage. As a
result. to ensure power system security against possible abnorinal conditions
due to contingencies (disturbances), the system operator needs to simulate con-
tingencies i n advance, assess the results,and then take preventiLrecontrol action
if required. This whole process is called dynamic security assessment ( DSA)
and preventive control.
Simulation studies (calledtransient stability studies)can take up to an hour
fora typical system with detailed modeling fora500-bus. 100-machinesystem.
Since i t takes a long time to conduct a transient simulation even for a single
contingency, direct methodsof stability assessment such asthose based on Lya-
punov or energy functions offer attractive alternatives.
It should be noted that a transient stability study is often more than an
investigation of whether the synchronous generators, following the ~ccurr-ence
ofdisturbance, will remain i n synchronism. It can be ageneral-purpose transient
analysis. i n which the "quality" of the dynamic system behavior is in\,estigated.
The transient period of primary interest is the electromechanical transient. LISLI-
ally lasting up to a few seconds in duration. If growing oscillations are of con-
cern,orif thebehavior of special controls isof interest. a longer transient period
may be covered in the study.
For transient stability analysis, a nonlinear system model is used. The
system is described by a set of differential equations and a set of algebraic
eq11ations. Genera11y,the differentia1 equations are milchine equations,contro
system equations. etc. The algebraic equations are system \,ohage equation:*
involLtingthenetwork admittance matrix.Thetime simulation method anddirec
method are often used for transient stability analysis. The former method deter.
mines transient stability by solvingthe systemdifferentialequation stepby step.
while thedirect method determines thesystem transient stabiliry without explic.
i t l y solving the system differential equations. This approach is appealing and
has receiLved considerable attention. Energy-based methods are a special case o f
the moregeneral Lyapunokrssecond ordirect method. theenergy function being
the possible Lyapiinov function.
Thischapterdealswith transient stability by aspecificdirect method rnainl;,,
the transient energy function (TEF)method. We begin by co\wing some basic
concepts from the theory of nonlinear system stability.
5.1 STABILITY CONCEPTS
Consider anautonomous systemdescribed by theordinary differential equatior.
where i =. t(t), and F( A)are n-vector\. F(.t)is generally a nonlinear function ( ~ f
t . Stability in the sense of Lyapunob i \ referred to an equilibrium state of Eq.
( 5 .I ). The equilibrium \tate is defined as the stage i rat which t ( t ) remaii \
unchanged for all f . That is,
The solution for -trfrom Eq. (5.2)is a fixed state since F(.t)is not an explicit
tunction of t . For convenience, any nonzero i ris to be translated to the origin
(t =0).That is. to replace t by .I +i t in Eq.( 5 .I ) to have
which gives
15.3)
I
*fit /--
I.. .. ................................. ...........................*........................ .......................+
6 .
It0
Figure5.1 Illustration of local stability.
Note that the current s differs from the old one by 4.As can be seen later,
from the definitions, this translation does not affect the stability of the system.
Thus,the origin of Eq. (5.3)is always an equilibrium state. I t should be noted
that t in Eq.( 5. 3)may be any independent variable, including time.
5.1.1 Definitionsanda Lemma
Stability(Local)
The origin of the system described by Eq. (5.3)is said to be stable if for any
given E >0,there exists a 65E such that ll.r,~ll <6implies ll.r(t)ll<E for all t
where soisan initial state.The origin iscalled unstable if i t is not stable.
The concept is illustrated i n Fig. 5.1where the initial statex,)has a magni-
tude less than 6.and the trajectory ofs remains within thecylinderof radius E.
AsymptoticStability(Local)
The origin of the system described by Eq. ( 5. 3) is said to be asymptotically
stable if it is stable and also if given ~ ~ . x l ~ ~ ~ <6 implies s -+0, (IS t -+00. The
arrow is used to mean approach,
Figure 5.2demonstrates theideaofasymptoticstability where thetrajectory
tends to 0as time tends to infinity.
Figure5.2 Illustration of asymptotic stability.
--
Glolm11~~ Asjmptotic- Stnbi l i t)/
Theorigino f the \ystem describedby Eq.( 5. 3)is said to be globallyasymptoti-
call>stable i f i t is \table and also implie\ -+0cis I+00, forany .v,, i n t h?
\v hoIe space.
Posi t i ve Deti'nite Function
A uniquely defined. scalar and continuous function V(.v) is said to be positi\,c
definite i n ii region K if V(.v)>0 for.v #0and V ( 0 )=0.
A space surt'ace formed by all .v satisfying W. v ) =0 is called a contour.
Ob\.iousl>r. contours with different \.aluescannot intersect one another. I t ' the>
do. \ ' (. v) has tM'o \ ~al ~i es at the intersection. We need the follou.ing lenimii r t b -
qiiirecl for the proof o f ii Ljrapunov theorem.
LcYl117lC?
Thereexists a spheredefined by l[.vll=N i n which V( . v) increasesiiionotoiiical'4,
along radical Lwtors emanating from theorigin.That is, V(PI,)increasesniono-
tonicully L b i t h pi n 0 5pIN for any unit \rector I I started from the origin.
Thisciin be shownusingthe assumptionof positi\.edefiniteness.First,coil-
t i nui t y. V(.Y)>0 and V ( 0 )=0, assiiiiie that V(pI1)increases monotonically \bri . h
pi n an interval 0IpIpIIand begins to decrease after p=pII.GiLfenii I I .there
i h ;in associated pIlLbrhich may be unbounded pll(=00). Let \ t - among all the U ' S
that has the sniallest pi,.then IIp\,n*II=pill\1t.11=PI,5pII.Since \'( PI,)increas$:\
monotonically nith pi n the inter\,al05P I pi,5p,,.\+'eare able to identify the
positi\,enumber tobe N =pi,.
Li ' ci pi ~~i ov Theoreni
There are three important theorems on 5tability de\eloped by Lyapunov. V/e
includethew here to form the theorem gi\ten below. In the theorem, V( i)i \ t i e
total dui \ atikc of \'( t ) on the trajectory \pecified by Eq.(5.3).That i \ ,
\'( \ ) =tll'=\' \( \) \ =1'( 1 I / ( t
tlt
Uhere \',(i)i \ the r ow \ector formedby the partial deri\ati\res of V(.v).
Regi017~R, RI, R,
A11 the regioris ;ire assuincd to contain the origin ;is an interior point. R: i < , ;I
\Libregionot'K ,\+,hictiis ;I siibi-egionof R:K: I K ,IK.
Theorem
Let V( x )be a positiise definite function with continuous partial deri\rriti\ves i n a
region R, then
The origin of the system described by Ey.(5. 3)is stable if i ' (. v) I0 i n
a subregion R, IR.
The system isasymptotically stable in theregion if i t isstableand i'(.v)
=0(identically zero)takes place only at the region i n a subregion R.
IRI.
The origin is globally asymptotically stable i f the sqstem is asymptoti-
cally stable. R2isthe whole space and V(.v) +00 as 11.v11 -+0.
I * Let be the srnaller one between a given E and N specified i n the
lemma. That is. I - =Mill [ E. NI .
Continuity of V(.\-)assures that there is a mi ni mum of 1T. v) on the
sphere ll.vll =t*. Let j*be among all the 11's that yield the mi ni mum.
V( I - , ) =I U . then V(I*, , ) 2111 must hold for any 1 1 . Monotonicity tells that
V(PI,)=111 for a pi n the interval 0IPII - , and hence \'(.v =PI,)=i i i is
enclosed by the sphere ~ ~ . r ~ ~ =I", and also i t is a closed contour since I I
is any uni t vector.
Let 6be the minimum norm of the points on the closed contour \/(.I-)
=1 1 1 , then since ~ I x , ~ ~ ~ <6is enclosed by the contour, V(.v,,) <1 1 1 ~ O I I O M ~ S
by monotonocity. Thus, any trajectory initiated from .v,) cannot possi-
bly crossthe sphereII.vll =t - 5E duetothefactthat \'(.v) is non-increas-
ing and V(.v) 2111 on the sphere II.YII =I - .
This completes the proof of part I of the theorem.
Part 2 can be shown by observing V(.\-)can be identically zero onl y at the
origin. Hence, V(.v) keeps on decreasing except at acountable number of points
at which i t stops decreasing momentarily. This implies that .v +0 L \ 'L' I - + -
since V( 0)=0 only when .v =0in R2.
I t seemsob\.ious t ohave part 3verified by the same reasoning used i n part
2.This is true except forthe case when V( x)approaches a finite value asll.v]l +
00 when .v,) is allowed to be any point in the whole space. The assumption that
V(.v) 4 a s \l.vll+-excludes this possibility which completes the proof of the 00
theoreni.
Except for relying on experience, there is no systematic method to find the
Lyapunov function asrequired by thetheorem. It hasbeen shownthat any stable
and constant linear system has ii V( x )but no one has yet shonm itsexistence i n
genera1 for nonIinear systems.
I14 Chciptrr5
It can be hhown that instability and asymptotic stability of the system dt:-
scribed by Eq. (5.3)are the same asits linearized system at the origin, which is
whereA =J; ( O) is a constant and n-square matrix. That is, the system describd
by Eq. (5.3)is unstable if at least one eigenvalue of A has ii positive real pxt
and is asymptotically stable if all eigenvalues have negative real parts.
Asymptotic stability asjudged from the linearized system is simple but 0 1
less practical use. I t is valid only in a sufficiently small region which is not
easily known. The Lyapunov function contains more information on stability i n
the regions R, and R2. For instance. global asymptotic stability tells us that the
trajectory initiated fromanywhere in the whole space, converges to the origiii.
5.1.2 Applicationof Lyapunov'sMethodtothe Simple Pendulum
We consider the dynamicsof a pendulum as a prototype for exploring stability
of an electric power system.
The motion of a pendulum with friction is described by
where -n: <n: <rlis theangle,CI isthedampingconstantand dhistheundamped
angular velocity. We regard the problem as a mathematical one without rest-ic-
tion on 8.Toconvert the system to the standard form of Eq. ( 5. 3) . we detine
that . vI=8and .i,=.Y? to get:
First, we want to check if i t is possible to find a L Y ~ ~ U I WV function for the
problem. The matrix A for the linearized system is:
Both eigenvalues ofA have negative and real parts when N >0.This sugge;tsa
possibility of finding a Lyapunov function.
The Lyapunov function is sometimes referred to as a generalized energy
function. Thename comes from the fact that i t has an initial positive kralueand
does not increaseas time goeson.This is thecase of physical systems that iiiok'e
without interference from theoutside such as the pendulum system.Tosee this,
let M. J , p and I be the mass. inertia, damping constant and the length of the
pendulum, then one dynamic equation becomes
J 6+pi 6+M ~ I sine=o
Multiply by i v and then integrate with respect to t :
The total energy of the pendulum (sum of kinetic and potential energjr)is
The time derivative is
Therefore, the proposed Lyapunov function isgiven by
with
V( x)=-ari
Thus,the Lyapunov function is actually the total energy per uni t inertia of
the pendulum system.
Now. we have
1 7
V(.r)=h( 1 -cosx,)+-. rj
2
in the region
R={-271: <x<271:;free.r2)
The derivative of 11is
1 . Theorigin isstablesincetheconditionof part 1 is satisfied bychoosiiig
RI =R. This is also true for the choice ( I =0.which is the case of ii
frictionless pendulum.
2. V( . v) =-(i.vi =0 implies that .v2 0and hence .i: =0.This makes .i-, =
-u.v2 - h sinv,=0 which yields sinvl=0 or .vl =/in.Therefore. \f(.v) =
0only at the origin by choosingR,={ - n<.vl <n:,ftw. v2) . Thus, t ie
origin is asymptotically stable sincethe condition o f part 2 of the thco-
rein is satisfied i n R2.
3. Part 3is not applicable forthe follcnving t M' o reasons:
( i )
\/(.I-) =0at . vl =21mi n the whole space,
( i i ) .vl +00 Mfith.v2=0makes V(.v)=h( I - cos.^,) 52bwhich doesi r o t
approach infinity iis required.
For demonstration purposes, M'L'employ the proposed \'(.v) to find ;I 65 E
iis required by the definition of stabilit}!.Tothis end.let I ( =[ l r , , I ( : ] ' be any u n i t
i wt or started from the origin, then
3 7 7
onthe circle.I +. v> =1.- or \l.vll=I - clV/t/.v, =h si i ul -.II <0forhII and.vl > 0.
Since \ / ( . I) is a decreasing function, i t ha\ a niinuinum at .tl =+r - and .I =0.
Therefore,ui =h(I - CO\/-).The expression for the minimuin becomes coirpli-
cated forh>1 .
on the contour \'(.v) =111. c/ / A/ . vI =2(.vl - h sin.\l)>0for - .vl >0and hII .
Sincethe minimuin occ~trslit .vl =0and .v2=&U, we hai e the rninitwrn
norm
The inequality holds for the reason that
117
For the case b>I , we may choose an independent \variable a=wf \ii t h CO' =1 7 .
Thus. the original dynamic equation becomes
where 0=&/daandCI' =cdw. Hence,the results of( b) and( c ) are \Valid because
h=I . Probably. there exists a better V(.t-)to yield the same results of ( b) and
( c) without considering h5 1 andh>1 separately. Although different Lyapunov
functions may serve the same purpose on stability. they may be different from
other points of view such as estimating 6and other control applications.
3. The condition of part 2 guarantees that the closed contour
keeps on decreasing in R,, astime goeson unt i l C=0asa l i mi t . Being
nonnegatiLte.each term of V(.r)must be zero siniultaneouslj \idien C=
0.This suggests that xI40andx2+0as C=0and hence asjiniptotic
stability for the origin isestablished.
5.2. SYSTEM MODEL DESCRIPTION
5.2.1 RealPowerSuppliedby a Generator
For a power network consisting of 12-generatorsconnected together by mutual
admittances as shown i n Fig. 5.3,we may write in matrix form:
Figure5.3 A general rt-Generator system.
I 18
where [ I ] is the injected current vector, [ E ]isgenerator internal voltage vector,
and [V ] is the system admittance matrix in which the generator impedances art:
included.
for all i, j=1.2. .,.n. The matrix element andcomplex voltage are specified by
-
Y, ,=G,,+jR,, and E, =E, 6, (5.f1
The real power supplied by generator i to the network is
( 5 . 7 ,
where
PI,=C,,sin(6,- 6, )+D,,cos(6,- 6, )
with C, ,=E, E,H, , and Ill, =E,E,G,,
The power P,, is the real power delivered by generator i toj ; i t may 3e
positive ornegative.P, ,=E; G,,is thepowerdelivered tothe local loadat gencr-
ator I .
From Eq. (5.8)it is clear that f,,depends on difference\ between phiw
angles rather than individual phase angles. This result suggests that one may
choose an arbitrary reference for the angles without affecting the resulting q, .
Indeed, we will first choose a reference rotating at synchronous speed to cle-
\tribe rotor dynamic\ and then a centerof inertia to minimize the kralue of rotor
hinetic energy.
5.3 STABILITYOF A SINGLE-MACHINESYSTEM
Consider ii generator connected to an infinite bus with voltage V L 0" throirgh
a pure reactance X as shown in Fig. 5.4.If the internal reactaxeand EMF of
Assessitig Ai i gl eStcihility \in Transiertt Eiiergy Fiirictiori
I I9
2
1
vL0
Infinite
Bus
Figure5.4 Single generator and infinite bus system.
the generator are given by X , and E L 6respectively, we have from Eq. (4.7)
that
P,.=A( s ) sin6
with
E V
A( x) =-
x,+x
Multiply on both sides (4-7)by w6t=d6gives
M O=(P,,,- PJd6 (5.9)
Let S,,=(6,,,0),S,=(&,a,)and S,,,=(S,,,, 0)be the start. fault clearing and
maximum statesof the fault respectively asshown in Fig.5.5. Let S=(6, w)be
any state on the P-6 curve generated by X , then integration of Eq. (5.9)from S
to S, yields
-
1
M d - -
1
MO$=P,,,(6- 6, )-A(X)(COSG, - ~0.46)
( 5. 10)
2 2
(a) During the hult: X = X ,
For S =S,,,we obtain from the above equation
-
1
Mw,!=PJ6,)- 6,)-A(X,)(COSG,- COS^,,) AI
( 5. 11 )
2
(b)
After the clearance: X =X ,
For S=S,,,,we have from Eq. (5.1I ) that
I 30
Power
P(P.U.:
A-Prefault
B- Post-fault
C-During
fault
I
I I
I
I
p , i i
--Lx
! I
I
I
II II
+Torque Torque
n Angle, S
(radims)
Thus,~ . ' e conclude froin ( a) and ( b) that
Therefore. 6,,,can be found by \olving the nonlinex algebraic Eq. ( 5 .I 21
o r graphically from Fig. 5.5 tojudge stability. Thi\ i \ hno\\n ;is the eyual-.iic.a
criterion for \tability \ t i d y of power system\.
I t i \ important to note that the excessive energy A, created during fault to
con\eer-tecito rolor kinetic energy al clearance. Thi\ re\ult in\pire\ the ii\e o 'the
transient t. ~it. rg~f function (TEF)afterclearing.That i \ . toobtain f r om Eq.(f .12)
U ith X =X ,
I ' I
where
E =-1 MO2 +P J 6 - 6, )
(5.13)
2
and
V =A( X, COS^, - COS^)
5.3.1 TransientSwing
Let II =do/clt denote the acceleration of the system, then Eq. (5.13)together
with Fig. 5.5 showthat CI <0abovetheP,,,lineand that ( I >0belour i t after friult
clearance. There are two angles: 6, =sin-' P, , , / A( X, ) and 6,,=TI: - 6,that corre-
spond t oII =0. Figure ( 5. 5) reveals that 6increases from 6,,with O)=0 due t o
II >0 until reaching 6,,,<6,, at which o =0. Since ( I <0 at 6,,,, 6 begins to de-
crease until o =0 at an angle less than 6,and then comes back because ( I >0.
As such. the poclrer angle swings back and forth around 6,.This is the case for
E - 1' <0at 6,,because 6,, is unreachable (0is imaginary).Ho\be\,er. delayed
fault clearance may result in largeA, which makes 6cross 6,,uith 03 20.Then.
6 increases further without return dueto CI >0.This is thc case for ( E - V) 20.
Wewill consider the transient tobe stable if the pouwangle s\\Iings around
6, and is otherwise unstable. This definition makes i t possible to ;isscss the
stability by means of the ( TEF) asfollows:
( a) The transient is stable if ( E- V) <0 at 6,,;large magnitude yields better
stability.
( b) Thetransient is unstable if ( E - V)20at 6,,.
5.4 STABILITYASSESSMENT FORIFGENERATOR SYSTEM
BY THETEF METHOD
Consider a power system consisting of ri-generators. The dynamics of each is
described by Eq. (5.11) . Wehave
de,=
(It
for i =I , 2,...,1 1 . The dynamics are second order and nonlinear differential
equations coupled together through the phase anglescontained i n theexpression
I 22 Chcipter .i
of e,.Although the numerical solution can be determined, a closed form of
solution is impossible toobtain. We need not discuss the solution of the system
of equations asstability that does not require the exact solution.
Another reference with velocity CO,,istobe chosen to minimize the integral-
square error
forany t , and t:. The necessary and sufficient condition for this purpose is
M, = M, o, , with M, = cM,
1- I 1- I
which indicates that the reference is the center of inertia (COI).TheCO1has a
phase angle 6,,satisfying
With respect to 6,,, all generators have phase angles
0, = 6,- 6,)
In terms of 8,,Eq.(5.14)is now expressed by
M, ~-+ M,
(10
( ! ! ! ! U = U,,,, - c,= p, - P:, (5.15)
tlt t l t
with
and
wherej f i signifies summation o f j from 1 to ti except i.All the angles 6, - 6,
in P, , may be replaced by 8,,= 8,- 8,because
Multiply Eq. (5.11 ) by 8,and then sum i from 1 to U :
I23
(5.17)
The second term of Eq.(5.17)is zero because
Thestatesat clearance angle 6,and unstable equilibrium 6,,are tobe speci-
fied by 11 pairs of (angle,velocity),that is.
Integration of Eq. (5.15)with clt from S,to S,, gives
( 5. 18)
where
and
I1 1'
v=cjP:,e,dt (5. 30)
,=I ,
with c' and 14 denoting S, and &.
Weknow fromEq.(5.19)thatE isthetotal energy input pluskinetic energy
of rotors. The electrical energy stored and dissipated in the system is V given
by Eq. (5.20).The energy here is referred to power integration with respect to
phase angle (nottime).
The stored energy in V is path independent but the dissipated energy i n V
is path dependent. To show this, we multiply Eq. (5.17)by 6,and then sum
from i =1 to 11 to obtain:
(5. 2I )
Theprecedingequation canbe verified by carrying outthe summation. Substitu-
tions of 6,from Eq. (5.8)gives
c,e,+~ , , 6 ,C, sine,, e,,+D,, cos(8,+8,) ( 5. 22) =
Therefore, the integration of Eq. (5.2 ) results in
- 8 , )- co\ ( 8 - 8/11+ I ( 5. 33)
We can see from Eq. (5.23)that the first part is the stored energy and I \
independent of thepathof integration.But,thesecond part denoted by I dcpenci\
OII the path of 8,. Sonie kind of approximation has to be used to evaluatc sincc
O , ( t ) cannot be found analytically.
Lct 0, be qqxoximated for all i= I . 2,. . . . 1 2 , by
behere C, and K, areconstants but may change with i andf7t)is the only one f ) r
all I . For con\mience, we use a parameter I / to make
where t , and I , , are the times at the clearance and unstable equilibrium. Then
0, = C + K f ( t ) = C' + K, f l t + ( I , - t ) ' I
= c' + K g( " )
f or t, Ir It,, and 05 I I I I . The two constants are required to meet the bound:q
conditi ons
Solving the twro constants gives
Based on this approximation, we haw
and hence i t follows that
with
This result makes i t possible to integrate I as
I t is interesting to note thatfit)org(rr)need not be known and that anjJ t )
yields the same approximation for I as indicated by Eq. (5.28).Thus. one may
perceive.f(r)to be the Col:
which naturally minimizes the integral-square error /M,(O,- f' fdtThis isproba-
I ,
bly the best choice of theJs. For example, if g( u) is chosen as the combination:
then g( 0) =g, g(1)=a+p and
I n concept,one may regard a systemasstableif the kinetic energy accuniu-
lated at the instant of clearance can be absorbed by the electrical components
of the system. Thus. i t is the kinetic energy that determines the stabilitjr. I t is
usually during the fault that some generators are affected and tend to separate
from the rest that arecoherent with the system Col . Sofarasthe kinetic energqr
is concerned. the gross motion of the separating generators (saythe first k ) may
be considered asa single generator with inertia and velocity the sameasthat o f
their COI.That is,
i i
M, =CM, and MR =
M, 6,
I I I - I
Tobe coherent with the system's COI,the rest of the generators must ha1e
zero velocity; 8,=0 for all i=k +I , k +2,...1 1 . Coherence suggests that the
rest of the generators rnay be regarded as an infinite bus. As such, the whole
system beha\res like a single generator and an infinite bus and hence the restilt
obtained before may be applied t othe multiple generator system.
We modify Eqs. (5.18)and (5.19)according to the conclusion and updit e
of Eq. (5.19)below.
1
E - V =- M,n:,
-
where
and
Note that R,,a,,,(I,,), and (I,,),, denote the velocities and angles at theclearance
and unstable state.
Comparing Eq. (5.26)with Eq. (5.1I ) enables us to draw the same con(l u-
+ionasbeing made for single generator systems. That is. the system is stab12 i f
E <V and unstable if E>V. It is inconclusive to talk about E=V \ince n e
have made approxiniations in evaluating E and V.
5.5 APPLICATIONTOA PRACTICAL POWER SYSTEM
The application of the direct method to actual power systems is quite difficult.
A number of simplifying assumptions are necessary. To date, the analysis has
been mostly limited to power system representation with generators represented
by classical models and loads modeled asconstant impedances. Recently. there
have been several attempts to extend the method to include more detailed load
models.
Inamulti-machine powersystem,theenergy function Vdescribing thetotal
system transient energy for the postdisturbance system is given by:
( 5. 30)
where
(3: =angle of bus i at the postdisturbance SEP
J, =2H, o, ,=per uni t moment of inertia of the itgenerator
The transient energy function consists of the following four terms:
1 . 1/2Z1,~:: change i n rotor kinetic energy of all generators i n the CO1
reference frame
2. CP,:,(e, - (3:): change in rotorpotential energy of all generators relatiLre
to CO1
3. CC C,/(cos@,-cos0,;):change in stored magnetic energy of all branches
4. ZZ D,, cos0,,d(O,+(3,): change in dissipated energy of all branches
The first term is called the kinetic energy (En,,)and is a function of only
generator speeds. The sum of terms 2, 3, and 4 is called the potential energy
(E,,,,)and is a function of only generation angles.
Thetransient stability assessment procedure involves the following steps:
Step 1 Calculation of the critical energy V,,.
Step2 Calculation of the total systemenergy at the instant of fitult-clear-
ing Vd
Step3 Calculation of stability index: V, ,- V,,. The system is stable if the
stability index is positive.
Time-domain simulation is run up to the instant of fault clearing to obtain
the angles and speeds of all the generators. These are used tocalculate the total
system energy (V, , )at faultclearing.The flowchartof TEFfortransient stability
analysis is shown in Fig. (5.6).
5.6 BOUNDARYOF THE REGI ONOF STABILITY
The calculation of the boundary of the region of stability, V( ?, is the most diffi-
cult step in applying the TEF method. Three different approaches are briefly
described here,
Input system data
b
~~~~~~~ ~
I Power now calculation
Con ting ency apecifica tion
Form during fault Y matrix
and reduced Y matrix
Form post-fault Y matrix
and reduced Y matrix
C alcglate post fault SEP
C alcolate critical energy V .I
Calculate .yatem total energy
V, at clearing time
Hes 2
1. TheClosestUnstc?ble EquilibriumPoint(UEPIApproxh
Earlypapers on theapplication oftheTEFmethod fortransient stability analysis
used the following approach to determine the smallest V, ,
Step 1 Determine all the UEPs. This is achi ei dby solving the postdis-
turbance system steady-st ateequations with different initia1 va1lies
of bus angles.
Step2 Calculate system potential energy at each of the UEPsobtained i n
step 1 . The critical energy V( ris gi\fenby the system at the UEP.
which results i n the minimum potential energy.
This approach computes the critical energy by implicitly assuming the \I or\t
fault location. hence, the results are very conservative.
2. TheControllingUEPApproach
Thedegreeof conservatism introduced by theclosest UEPapproach is such that
the results are usually of little practical value. The controlling UEP approach
removes much of this conservatism by computing the critical energy depending
on the friult location.This approach is based on the obser\ation that the system
trajectories for all critically stablecases get close to those UEPsthat are closelj,
related to the boundary of system separation. The UEPs are called the control-
ling or relevant UEPs.
The essence of the controlling UEP method is to use the constant energy
surfrice through the controlling UEP to approximate the rele\mt part of the
stability boundary (stablemanifold of the controlling UEP)to which the f i i ul t -
on trajectory is heading.
For any fault-on trajectory q ( t ) starting from a point p ~ A ( . v , ) with V( p )<
V(.i-), if the exit point of the fault-on trajectory lies i n the stable manifold of .i-,
the fault-on trajectory must pass through the connected constant energy surfiice
AV((.?)before it passes through the stable manifold of .t(W(.i-)) (thusexits the
stability boundary AA(-t,J).Therefore,theconnected constantenergy surfuceA\,
(.P) can be used to approximate the part of the stability bondary AA(,?,)for the
fault-on trajectory .v,( I). The computation process in this approach consists of
the following steps:
Step1 Determine the controlling UEP,A-,,, forthe fault-on trajectory .v,(I).
Step2 The critical energy V, is the value of the energy function V ( * )at
the controlling UEP,that is,V, =V( x, , , ) .
Step3 Calculate the value of the energy function \ I(. ) at the time of fiiult
clearance (say,I,,) using the fault-on trajectory I{,=V( . v, ( I ( , ) ) .
Step4 If V, <V, , then the postfault system is stable. Otheritrise, i t is un-
stable.
The key element of the controlling UEP method is how to find the controllinp
UEP for a fault-on trajectory. Much of the recent work in the controlling UEP
method is based on heuristics and simulations. A theory-based algorithm to
find the controlling UEP for the classical power system model with transfer
conductance G,is presented now.
The energy function is of the form:
JI
where M, = M, , .Y' =( 6' ,0)isthe stableequilibrium point (SEP)under consij-
eration. ' = I
Algorithmto FindtheControllingUEP
The reduced system is
The algorithm for finding the controlling UEP consists of the following steps:
Step 1 Fromthe fault-ontrajectory ( 6( f ) , w( r ) ) , detect thepoint 6*at which
the projected trajectory 6(r)reaches the first local maximum : ) t
E, , (*). Also, computethe point 6- that isone stepaheadof 6" along
6(r).and the point 6' that isone step after 6*.
Step 2 Use the point 6" as initial condition and integrate the postfault
I1
reduced system Eq. (5.31)to find the first local minimum of
I : I
I.f;(s>), say at 6:.
Step 3 Use 6- and 6'as initial conditions and repeat Step 2 to find the
corresponding points, say 6,and 6;respectively.
Step4 Compare the values of If(6)i at 6,, 6:, and 6;.The one with the
smallest lralueis used a s the initial guess to solve Eq. (5.3I ),.f;(6)
=0,say the solution is 6,,,.
Step5 Thecontrolling UEP with respect to the fault-on trajectory is (6,(#.
0).
The proposed algorithm finds the controlling UEP Lria the controlling UEP
of thereduced system Eq.(5.31)with respect totheprojected frlult-ontrajectory
6(r).Steps 1-4 findthecontrollingUEPof thereduced system and step5 relates
thecontrolling UEPof thereduced systemtothecontrolling UEPof theoriginal
system.Theoreticaljustification of theproposed algorithm can be found i n work
done by Chiang.
3. TheBoundaryofStability-Region-BasedControlling
UEP(BCU)Method
Earlier UEP methods faced serious convergence problems when solving for the
controlling UEP, especially when the system is highly stressed or highly un-
stressed, or when the mode of system instability is complex. These problem\
usually arise if the starting point for the UEP solution is not sufficiently close
to the exact UEP. Someof the convergence problems can be otm-comeby the
BCU method which has the capabilityof producing a much better \tarting point
for the UEP solution.
CONCLUSION
One of the major innovations in stability assessment is based on the energy
function concept. which isan offshootof Lyapunov stability criteria.Thischap-
ter introduced the fundamental Lyapunov stability thought and the procedure of
constructing Lyapunov stability function.
The main thrust of this chapter isto utilize concepts of system modeling to
evaluate system stability using the energy function method. The reader is re-
ferred to the list of references and the annotated glossary of terms for further
information on the subject matter.
Voltage StabilityAssessment
INTRODUCTION
Voltage stability studies evaluate the ability of a power system to maintain ac-
ceptable voltages at all nodes under normal conditions and afterbeing subject:cl
to contingency conditions. A power system is said to have entered ii state of
voltage instability when a disturbance causes a progressive and uncontrollable
decline i n voltage values. Inadequate reactive power support from generators.
reactive sources, and transmission lines ciiri lead to kdt age instability or e ~ w i
\.oltage collapse, which have resulted i n several major system fiiilures (b1ac.k-
outs)such as:
I . September I910, New York Power pool.
2. Northern Belgium System and Florida System disturbances of I982
3. Swedish system disturbance in December 1983.
3. French and Japanese system disturbance5 i n 1987.
5. Recently, i n the late nineties, i n the U.S.and other parts of the word.
The literature and background studies reLriewed indicate that \roltage i n<ta-
bility or collapse are characterized by a progressike fall of i dt age which can
take several forms. The main factor is the inability of the network to meet the
demand of reactive power. The process of instability may be triggered by seine
fornio f di\turbance,resulting i n changes i n thereactikre poucrrequirement.'-he
disturbance may either be small or large changes i n essential load. The conse-
quence of voltage instability may, however. have uidespread impact on the
system.
Power systemvoltage stability involvesgeneration.transmission, anddistri-
bution.Voltage stabilityisclosely associatedwithother aspectsof poM-ersystem
steady-stateanddynamic performance. Voltage control,reacti1.epowercompen-
sation and management, rotor-angle stability, protective relaying. and control
center operations all influence voltage stability.
Voltage stability studies involve a wide range of phenomena. Because of
this, voltage stability means different things to different people. I t is a fast phe-
nomenon for engineers involved with the operation of induction motors. air
conditioning loads, or HVDC links. I t is a slow phenomenon (iri\.ol\.ing,for
example. mechanical tap changing) for others. (Appropriate analysis methods
have been discussed with the debate centering on \%!hether the phenomena of'
voltage stability are static or dynamic).Voltage instabilitjr or collapse is a dqr-
namic process. The term "stability" implies that a dynamic system is being
discussed. A power system isadynamic system.I n contrast to rotor angle ( s j w
chronous)stability, the dynamics involve mainly the loads and the means for
voltage control. Voltage stability has alternatively been called load stability.
The lossof lines orgenerators can sometimes cause \.ohage quality degra-
dation. This phenomenon has equally been attributed to the lack of sufficient
reactikre reserve when the power system experiences a he;\\,} load or a se\.ere
~ contingency.Thus, voltage instability ischaracterized i n such;i I J that \ dt age
magnitude of the power system decreases gradually and then rapidly near the
collapsing point. Voltage stability is classified as either static voltage stability
or dynamic voltage stability. The latter is further classified into small signal
stability and large disturbance stability problems. A unified frame\%.ork related
to voltage stability problems will be shown in a proceeding section.
In dynamic voltage stability analysis, exact models of transformers. Static
Voltage Compensating devices (SVCs),induction motors and other types ot'
loads are usually included in problem formulations i n addition to models 01'
generators, exciters. and other controllers. Small signal voltage stabiliti, prob-
lems are formulated as a combination of differential and algebraic equations
that are linearized about an equilibrium point. Eigen analysis methods are used
to analyze system dynamic behavior. Small signal analysis can provide uset'ul
information on modes of voltage instability and is instructik'e i n locating VAR
compensations and in thedesign ofcontrollers. Largedisturbance \wltage stabil-
ity is approached mainly by using numerical simulation techniques. Since s ~ r s -
tem dynamics are described by nonlinear differential and algebraic equations
that cannot be linearized in nature. Voltage collapse is analyzed based on ii
center manifold voltage collapse model.
I24
6.1 WORKING DEFINITIONOF VOLTAGE
COLLAPSE STUDY TERMS
Voltage stability has been viewed asa steady-state viability problem suitable
for static analysis techniques. The ability to transfer reactive power from pro-
duction sources to consumption sinks during steady operating conditions is a
major aspect of voltage stability. The following definitions are often used i n
voltage stability studies.
Voltage collapse incidentsi n the U.S.,Europe,andJapan ha\x led todiffer-
ent explanations. interpretations. and concerns. To achieve a unifying franre-
work, a working definition for detection and prevention of Ldtage collapse I-iis
been constructed (EPRIRP 2473-36).
6.1.1 ClassificationofVoltageCollapseDetection
Detection of VC isbased ondetermination of imminence of power system volt-
age violating its limits which value may lead to system instability and conse-
quently voltage collapse. Thephenomenon i n many instances isdue to a def cit
in reactikte power generation, loss of critical lines or degradation of control on
key buses.
6.1.2 ClassificationofVoltageCollapsePrevention
Voltage collapse prevention is any action taken to reduce the likelihood of
power system degradation due to the violation of operating limits. Prevention
scheme includes the use of optimal power tlow strategy and other measure:*to
minimize voltage deviations.
Otherdefinitions of voltage collapseorstability aredocumented in the l i ter-
ature. For example,according to the IEEE working group in voltage stability.
voltage collapse/stability is defined asfollows:
Voltage stability is thecihility of a system t o maintain kroltage sothat when
load admittances are increased, load power will increase, so that t o t h
power and voltage are controllable.
Voltage collapse is thepr.oc.t..s.s by which voltage instability leads to 10s.; of
voltage in a significant part of the system. Voltage degradation may
also lead to angle stability as well, especially if the preventive mea-
sures are not enforced quickly enough. Sometimes only careful post
incident analysis can discover the primary cause of voltage collapse.
Voltcige Stcrhility A.s.ses.sment 1-35
6.2 TYPICAL SCENARIOOF VOLTAGE COLLAPSE
Assume that a power system undergoes a sudden increase of reactive power
demand following a system contingency, the additional demand is met by the
reactive power reserves carried by the generators and compensators. It is possi-
ble, because of a combination of events and system conditions. that the addi-
tional reactive power demand may lead to voltage collapse, causing a major
breakdown of part or all of the system. A typical sequence of events leading to
a voltage collapse can be as follows:
1. The power system is experiencing abnormal operating conditions with
large generating units near the load centers being out of service. Some EHV
lines are heavily loaded and reactive power resources are low.
2. A heavily loaded line is lost which causes additional loading on the
remaining adjacent lines. This increases the reactive power losses in the lines
causing a heavy reactive power demand on the system. (Reactive power ab-
sorbed by a line increases rapidly for loads above surge impedance loading).
3. Immediately following the loss of the line,aconsiderable reduction of
voltage takes place at adjacent loadcenters duetoextra reactive powerdemand.
Thiscauses a load reduction, andthe resulting reduction in power flow through
the lines would have astabilizing effect.Thegenerator AVRs would, however,
quickly resolve terminal voltages by increasing excitation. The resulting addi-
tional reactive power flow through the inductances associated with generator
transformers and lines would cause increased voltage drop across each of these
elements. At this stage, generators are likely to be within the limits of P-(2
outputcapabilities, i.e.,within thearmature and field current heating limits. The
speed governors would regulate frequency by reducing MW output.
4. TheEHVlevel voltage reduction at loadcenters wouldbe reflected into
the distribution system.The ULTCs of substation transformers would restore
distribution voltages and loads to prefault levels in about 2 to 4 minutes. With
each tap change operation, the resulting increment in load on EHV lines would
increase the line XI and RI losses, which in turn would cause a greater drop
in EHV lines.If the EHV line isloaded considerably above the SIL,each MVA
increase in line flow would cause several MVAsof line losses.
5. As a result, with each tap-changing operation, the reactive output of
generators throughout the system would increase. Gradually. the generators
would exceed their reactive power capability limits (imposed by maximum al-
lowable continuous field current)one by one. When the first generator reached
its field current limit, its terminal voltage would drop. At the reduced terminal
voltage for a fixed MW output, the armature current would increase. This may
furtherlimitreactive output tokeepthearmaturecurrentwithin allowable limits.
Itsshareof reactive loading would be transferred toother generators, leadingto
overloading of more and more generators. With fewer generators on automatic
excitation control,the system would be much more prone to voltage instability.
This \.+wild likely be conipounded by the r:duced effectiLmess of shunt coni-
pensators at low voltages. The process will eventually lead to voltage collap;e
or avalanche, possibly leading to loss of synchronism of generating.
Sometimes. the term 'voltage security' is used. This means the ability of ;I
systemnot only to operate i n a stable inode but also toremain stable follohiiig
credible contingencies or load increases. I t often means the existence o f a con-
siderable margin from an operating point tothe voltage instability point fc)l l o\ \ , -
ing credible contingencies.
6.3 TIME-FRAME VOLTAGE STABILITY
Voltage instability incidents in \.arious places around the world has been c.c-
scribed iis ii function o f time.The timeranges fromsecondstotenths ofminutcs.
Three time frames and sconarios described by Carson Taylor are sumniari;red
;is follo\+fs:
I . Transient cultage stability: This occurs bet\+.een0 to 10 seconds.
2. Classical Voltage Stability: This occurs betureen I to 5 minutes. This
is the c1iissi csccnari oin1'0I1,ing aiit0111 aticon-Ioad tapchanger,distribution \,(1It-
age regi i 1ation.xi cl gener;itor cU rre11t Iimit ing.This scenarioin\'oI\res high 1o;i Js.
high po~verimports from ii remote generator, and a large distribution. Tap
changcr action is alsosignificant,i t i t canbe beneficial or detritnental depend,ng
on load characteristics and location of the tapchanger. Collapse i n one area c-an
at'tect ;I much larger area, thus, leading to amt i j or black-out. This occurrenct' i \
tj.pical on the East Coast. the pacific West, and in Japan.
3. Long-term Voltage Stability: This in\rol\res se\feriilminutes. Se\.t ral
VC incidents (theTokyo I977 blackout. the Shiedenblackout)are cxperien:cs
o f VC.The factors may include overload time, limit of transmission, loss load
di\,ersitJ,dueto low Lroltage (theoreticallycontrolled loads),timeliness ofappI>'-
ing reuc*ti\.epower. and other operating interventions such iis load shedding.
Reported idtage instabiIi t y iticidents \v ith and L+i i thoii t voItageare su111Ina-
r i xd i n Figure 6. I .For classical \,()Itageinstability. the phenomenon \ +r i l l occur
at the onset o f the \,oltagccollapse. For long-term stabilitjr, the shorter-t.me
frame phenoniena wi l l occur once \ dt agebegins to sag leading to \ultagec:ol-
lapse.
6.4 MODELINGFORVOLTAGE STABILITY STUDIES
Voltage stability studies involve the solution to algebraic and differential equa-
tionsthat mapthesystembehavior under steady-stateandtransient state\.Belo\v
are tjpical \ector\ encountered and the notation used.
--- -11
Classical Long-Term
(LargeDisturbance) (LargeDisturbance) (LoadBuildup)
GeneratorExc.Dynamics LTCs
i
-
RimeMoverControl Load DiversityJrhermostat
0 0 0
Max.ExciterLimiter
0
MechanicallySwitchedCapacitors Linnransformer Overload
-
;-y- Gen.ChangdAGC
0 4
InertiaDynamics Boiler Dynamics
0 0 1
OperationIntervention
I
DC ConverterLTCs GasTurbines
F-
- 1
I min 10min I hr
1 I I I 1 1 1 1 1 I 1 I I I I 1 I1 I I I 1
Timein seconds
Figure6.1 Time-iramc ofvoltage stability (courtesyof C a r ~ nTa)rlor).
1 . Dynamic state \rector.~ ( t )
\+!here
8(t) =Rotor angle
E(t) =Voltage components of synchronous machine
S( r) =Dynamics i n load bus
p( t ) =Other dynamic states (exciter,governor)
2. Algebraic state vector, y ( t )
where
~ ( t ) . 6( t)=Bus voltage niagnitude and angle
Chtrprrr 6
Q( t ) =Nonscheduled reactive power
q( r )
=Other algebraic variables
3. Parameter vector,p( t )
where
P,(r 1 =Turbine shaft power
P, ( t 1, Qr ( t )=Scheduled load power
6 d ( t) =Controlled voltage or set points
CT(t1 =Other similar parameters
6.5 VOLTAGE COLLAPSE PREDICTION METHODS
The framework for voltage stability studies can be simplified to fit the time
span of the analysis. Thecategories of interest are as follows:
6.5.1 StaticStability
Assume all time derivatives equal zero at some operating point H( . v , j - . p)=0
6.5.2 DynamicStability
At some operating point, small perturbations (local)
6.5.3 ExtendedStability
Simulation through time (upto hours)
6.6 CLASSIFICATION OFVOLTAGE STABILITY PROBLEMS
Voltage problems are distinguished in three categories:
1 . Primary phenomena related tosystemstructure.Theseretlect theauton-
oinous response of the system to reactive supply/demand imbalanccs.
139 VoltugeStcrhility Assessrtierit
2. Secondary phenomena related to control actions. These reflect the
counterproductive nature of some manual orautomatic control actions.
3. Tertiary phenomena resulting from interaction of the above.
This classification of voltage quality problems implies that the problems
involve both staticanddynamicaspectsof systemcomponents. Voltage collapse
dynamics span a range in time from a fraction of a second to tens of minutes.
Time frame charts are used to describe dynamic phenomena which show time
responsesfromequipment thatmay affect voltage stability.Thetimeframechart
isshown i n Table 6.I , where.q isastatevector representing transient dynamics.
I,is a state vector representing long-term dynamics.
Y is a state vector representing very fast transient dynamics related to net-
work components and P is a system parameter vector. Then the time frames to
be considered become very fast transient, transient, and long term. The main
characteristics of the three time frames are asfollows:
1. A very fast transient voltage collapse involLres network RLC compo-
nents having very fast response. The time range is from microseconds to milli-
seconds.
2. A transient voltage collapse involvesalargedisturbance and loads ha\.-
ing arapid response. Motor dynamics following a fault are often the main con-
cern. The time frame is one to several seconds.
3. A long-term voltage collapse usually involves a load increase or a
power transfer increase. Within this time frame, a voltage collapse shows load
restoration by tap-changer and generator current limiting. Manual actions by
systemoperators may be important.Thetime frameisusually 0.5to 30 minutes.
Sincevoltage stability is affected by various system components in a wide time
range, i n order to tackle this problem, one must consider proper modeling and
analysis methods. Currently, voltage stability approaches mainly include static
and dynamic,i.e.,transient voltage collapse and long-term voltage collapse.
Table6.1 Time Frame and Relevant Models in Voltage
Stability Assessment
Voltage stability models and time scale
Micro to milli seconds A few seconds Minutes
I40
6.7 VOLTAGESTABILITY ASSESSMENTTECHNIQUES
Thelosso f linesor generators can sometimescausedegradation i n voltage.This
phenomenonhiis equally been attributedtothe lack of sufficientreacti\,eresetve
when the pouer systemexperiences a heavy load o r severe contingency. Thus.
\-oltagestability is characterized in such a way that voltage magnitude of the
p o ~ wsystem decreasesgradiially and then rapidly in the neighborhood o f I he
collapsing point. Voltage stability is classified as static voltage stability and
dq'naiiiic \.oltagestability.The latter is furtherdi\ridt.d into srnallsignal stabilitJr
ancl large disturbance stability problems.
I n dynai n ic vo1tage st 11biIi t y ana1ysis, exact tnodeIi ng of transformcrs,
SVCs. induction motors, and other types of loads are usually included i n prob-
lem formulations i nadditiontomodelsof generators,exciters.andothercontrol-
lers. Small signal voltage stability problems are fimiiulated asa combination ot
differential rind algebraic equations that are linearized about an equilibri.in1
point. Eigen analysis methods are used to analyze system dynamic beha\rior.
Small signal rinalysis can provide useful informationon modes of \voltage in:,ta-
bility and is instructikre i n locating VAR compensations and i n the design of
controllers.Ontheotherhand, large disturbance voltage stability is mainly dorilt
\+,ithbj.numericul simulation techniques. since system dynamics are descrilxx!
by nonlinear differential and algebraic equations that caniiot be linearized i n
nature. The mechanism of \,()Itagecollapse has been explained as saddle node
bifurcation i n some literature. Voltage collapse is anrilyzed based on ii ceriter
manifold \ dt a ge collapse model.
Static Lroltage stability analysis is basedon po\+wsystem load tlow e q u -
tions. Indices characterizing the proximity of ;in operating state to the collapse
point are cie\~eloped. Thedegeneracy of the load tlo\ss Jacobian matrix has bcen
used ;is ;in iritlex of po\ +wsystem steady-state stabilitjr. I:nder certain coiidi-
tions, ;I change i n the sign of the determinant o f the Jacobian niatris diiiing
c*ontinuoiis\rariations of pnrametersmeans that ;I real eigeiivalue of the liii1:ar-
i d sb+ringequations crosses the imaginary axis i nt o the right hal f o f the coin-
ples plane and stability is lost. Various researchers ha\.e considered th;it ;I
changei n the sign o f the Jacobian matrix may probubljfnot indicate the losjof
steady-state stabilitj, urhen e\winiirnber eigen\Aues \+!hosereal part cross the
imaginury axis. Voltage stability is alsorelated to iiiiiltiple load flow soluti~ms.
A proximity indicator t or Lroltage collapse ( VCPI ) was defined for ;I bus, ;in
area. or the complete system ;is ii Lectorof ratios of the incremental gener.ited
reactive powcr at agenerator toagiven reactive load d e n i d increase.Adil'ter-
ent indicator ( Lindex)is calculated from normal load tlow results with reaxm-
;IbI e coinput iitions. The 111 ini inu111 singu1ar va1ue of the Jiicobian was pr op)sed
;is ii \coltagesecurity index. since the magnitude of the rninimiim singular \ d u e
coincides Lsrith the degreeof Jacobian ill-conditioning and the proximity to col-
lapse point. Based on a similar concept, the condition number of the Jacobian
is also applied as an alternative voltage instability indicator by pioneers i n the
field.
Bifurcation theory is used to analyze static stability and voltage collapse.
Staticbifurcation of powerflowequationswere associated Lvith eitherdi\wgent-
type instability or loss of casualty. Researchers ha\.edescribed necessary and
sufficient conditions for steady-statestability based on the concept of feasibility
regions of power flowmaps andfeasibility margins but with high computational
efforts.A security measure isderived to indicate system \~ulnerability to Lwltage
collapse using an energy function for system models that include i dt age\,aria-
tion and reactiireloads. I t isconcluded that the key to applicationsof theenergy
method is finding the appropriate T\pe-l low voltage solutions.
I n addition to the above methods for direct coinputation of stability index.
some indirect approaches, based on either the continuation method oroptimiza-
tion methods haire been developed to compute the exact point of collapse. I n
a p pIying the co ntinuation methods,assumptions about 1oad changi11g patte1-17s
are needed.
In summary, the methods for static voltage instability analysis are based on
multiple load flow solutions (voltageinstability proximity indicator [VIPI],en-
ergy method),load flow results ( L index, VCPI),oreigen\Auesof theJacobian
matrix (iiiinimum singular value and condition number).While studies on djr-
namic voltage collapse shed light on control strategy design (off-lineapplica-
tions),static \.oltage stability analysis can provide operators with guideline i n-
formation on the proximity of the current operating state to the collapse point
(on-lineapplications). I n this case, an index, which can @\read\wice utarning
about the proximity to the collapse point, is useful.
The next discussion will be that of the formulation of selected \x)lta, ce sta-
bility indices. Of the wide range of techniques available. we shall discuss the
VIPI method. a method based on singular value decomposition. condition nuin-
ber ofthe Jacobian, and the method based on the Energy Margin.
6.7.1 Voltage InstabilityProximityIndicator Method
The \.oltageinstability proximity indicator (VIPI)was developed by Y. Tamura
et al. based on the concept of multiple load flow solutions. A pair of load f l o ~
solutions.vl and.v2 are represented by two vectors ci and 17 as followfs:
which are equivalent to:
where . vl is the normal (high)power tlow solution and.v2 its corresponding IOYA.
voltage power tlow solution;LI isasingular vector i n the spaceof node voltages
and h is a margin vector in the same space.
We now define twoother vectors Y, and Y( ci ) .called singular vectors in t k
space of node specifications. The relationship between these \rectors is shoun
in Figure 6.2.
VIPI is defined by the following equation:
where vector Y, , consists of bus injections computed with respect to. vl but the
injection kAues corresponding to reactive powers of PV buses are replaced by
the squared values of voltage magnitudes. Y(cc) consists of bus injections with
respect to vector ( I : I(.v ( 1 is the /,-norm of vector, .v. Thecomputation of VIPI is
easy once therelevant low voltage power tlow solutionsareobtained. Generallj,
speaking. finding all the relevant low voltage solutions are time-consuming tor
practical size systems.
6.7.2 MinimumSingular Value (U,,,,,,) Method
When an operating state approaches the collapse point, the Jacobian matrix of
the power tlon equations ( J ) , approaches singularity. The minimum singu ;ir
Figure6.2 Concept o f VIP1 in the node specification space.
value of the Jacobian matrix expresses the closeness of Jacobian singularity.
Thesingularvaluedecompositionmethod isusedtosolvetheminimum singular
value for static voltage stability analysis.
Accordingtothe theoryofsingularvaluedecomposition.power flowJacob-
ian can be decomposed as:
J = LEV' (6.11 )
where: J E R'J1*'J1 is the power flow Jacobian matrix; C=diag(o,,(J?, - ,oJJ)
with(J,,,,, =(J,2o22 2(J,~=(J,,,,,, 20.If matrixJ hasrank ) - ( I " 5211).itssingu-
lar values are the square roots of the I" positive eigenvalues of A7A(orAA' ) .
U and V are orthonormal matrices of order 211,and their columns contain the
eigenvectors of AA' and ATA respectively. From Eq. (6.9).i t can be obtained
that
AV, =CJ,~(,
(6.13)
A' U, =(3,V, (6.13)
We define
E,=u,V: (6.14)
Then Eq. (6.I 1)can be written as:
(6.16)
then, as far as the /?-normof the Jmatrix is concerned, J' is a matrix of rank
11 - 1 nearest to the J matrix of rank n. This means that the smallest singular
value of a matrix is a measure of the distance between matrices J andJ'. As for
the power flow equations, its minimum singular value expresses the proximity
of the Jacobian to singularity. It can be used as an index for static voltage
stability.
6.7.3 ConditionNumberof theJacobianMethod
The condition number is used in numerical analysis to analyze the propagation
of errors i n matrix A or vector 6 in solving variable vector x for the linear
equation Ax =h. If matrix A is ill-conditioned, even very small \xiations i n
vector 17 (orA ) may result in significant changes i n solution \rector.v.
Forthe Iinearized load flowequations, thecondition numberoftheJacobi.in
matrix canbe iisedto measure itsconditioning and whetherany small variations
i n i'ector11 ( or A ) inay result i n significant changes i n solution vector .r.
Forthe Iinearized load tlowequations, thecondition number of theJacobi,in
matrix can beused to measure itsconditioning andwhetherany small variations
i n loads niay lead to large changes i n bus voltages. If the condition number is
greater than a specifiedthreshold, this will Iiieiin that the current operating state
is close to the collapsing point.
A precise measure ofthe sensitivity ofalinear sq'stemsolution with resptxt
to matrix A or vector h ciiiibe defined as:
For pobrer t l ocb Jacobian iiiatrix J . the \alue of Cond, (J ) can gile an
indication of the condition of J "with respect to inversion." A small ~ ~ a l u e of'
Cond2( J ) (1 - 10)refer\ to a bell-conditionedJacobian matrix (relatitrely large
voltage stability margin):a large value of Cond2(J ) (>100)mean\ thattheopi*ra-
ting 5tute is \ cr yclose tothe point of Jacobian \ingiilarity and hasa I ON coltige
\tiibilitj iiiargin. The extreme condition is that J is singular and Cond,(Jr is
infinite. Hence. the condition nuniber Cond?(J ) ciin be used to iiieii\ure the
proxiiiiitj of the operating \tates to voltage collap\e.
6.7.4 EnergyMargin-BasedMethod
The energy method uses an energy function, dericed from a clo\ed form \o:tor
integration of the real iiiid reactive mismatch equations betueen the oper;ihle
pouerflou \olution and ;i I w oltage power tlou \olution, to prokide ii yuiinti-
tatice tiieasiire ol'I ~ O M clo\e the \y\tem i \ to toltage instability. The point of
toltage in\tability correspond\ to the \addle node bifurcation point definec hq
;i \iiigiiliir potver tlou Jacobian Ni t h ~ e r ocnergj margin.
The encrgq function t'or boltage \tabilit> analysi\ i \ defined as:
1-45
(6.19)
with real and reactive mismatches defined as:
where: .I-'=(a' , V' )is the normal operable power tlow solution ( or the stable
equilibrium point, SEP): xf' =(a",V") is the relevant low Yoltage po\\.er ilow
solution with respect tox' (orunstable equilibrium point, UEP).
A large energy value indicates a high degree of \rollage stabilitjr ndiilc ;i
small \yalue indicates a low degree of voltage stability. I n applj'ingthe c ne r g~~
method. the key is finding the relevant UEPs.Since the number of rele\rant l o ~ i p
\.o~tage power tlow solutions is very large (Y'- I for a practical syteni. the
exhausti\veapproach is not feasible.There is an i mpr o\ 4technique to compute
all the Type-I UEPs based on the results that sho\i' for tjpical po\i'ersj'stems.
the system always loses steady-state stability by a saddle node bifurcation bc-
tureen the operable solution and a Type- 1 low-voltage solution. That condition
restricts the computation of relevant UEPs only corresponding to sj'sterii PQ
buses, or practically PQ load buses. After finding all thc rele\mt UEPs. the
buses corresponding to which the energy function has the lowest \ dues arc
buses \sulnerable to voltage instability. Similar to the VIP1method. the energy
methods depend on the low-voltage power flow solutions. urhere the Ne\s,ton-
Raphson method with the optimal multiplier can be used.
6.8 ANALYSIS TECHNIQUES FORSTEADY-STATE
VOLTAGESTABILITYSTUDIES
6.8.1 Introductionto the Continuation Method
I n its early stages, \ dt agecollapse studies were mainly concerned \\!it11 stead>.-
state voltage behavior. The voltage collapse isoften described ;is aproblcm that
results when a transfer limit is exceeded. The transfer l i mi t of an elcctrical
pom'er network isthe tnaximal realorreactive power that the system c;in deli\w
from the generation sources t o the load area. Specit'icallqr.the transfer l i mi t i h
136 Clttipter 6
the maximal amount of power that corresponds to at least one power-tlow solJ-
tion. From the well-known P-V orQ-V curves,onecanobserve that the volta;;e
graduallydecreasesasthe power transfer amount isincreased.Beyond themaxi-
mum powertransfer limit,thepower-tlow solution doesnotexist,which implies
that the system has lost its steady-state equilibrium point. From an analytical
point of view,thecriteria fordetecting the point ofvoltage collapse isthe point
where Jacobian of power-flow equations become singular.
The steady-state operation of the power system network is represented ~ y
power-tlow equations given in equation (6.20).
where 8represents the vector ofbusvoltage angles and V represents the vec or
of bus voltage magnitudes. h is a parameter of interest we wish t o b a y . I n
general thedimension of F will be 211,,,+t i / )u, where upsand / I / \ are the number
ofPQ and PV buses, respectively.
From equation (6.20)one obtains the fundamental equation of sensitiii t )
anaIysis
Let .v =[e, V].From Eq. (6.2 1 ), one can obtain an ODE system
( 622)
For a specific variation of the parameter h,the corresponding variatioIi t o
the solution x is calculated by evaluating the Jacobian (dF/d.4-].I t should be
emphasized that the singularity of the power flow Jacobian dF/d.v is necesaary
but not a sufficient condition to indicate voltage instability. The method pro-
posed to observe the voltage instability phenomenon isclosely related to tn111ti-
ple power flow solutions. which are caused by the nonlinearity of power flow
solutions.
The drawback of the method is that i t relies on the Newton-Raphson
method of power flow analysis, which isunreliable i nthe vicinity of the vollage
stability limit. As such, researchers have developed a technique knourn as the
continu ation method.
137 VoltcigeStcthility Assessriierlt
6.8.2 ContinuationMethodandItsApplicationto
Voltages StabilityAssessment
Consider the power flow equation defined in Eq. (6.20).The vector function F
consists of \I scaler equations defining a curve in the I I +1 diniensional (.v,h)
space. Continuation means tracing this curve. For aconvenient graphical repre-
sentationof the solution (.v.h)of Eq.(6.20)we need aone-dimensional measure
of x. Thefrequently used measures are:
( i )
1 .v 1 =E','=,xf(square of the Euclidean norm),
( i i )
j s I =max 1 .I-, (maximum norm), 1
I=II f
( i i i ) 1 s 1 =xi for some index k , 1 Ik I n .
In power systems generally we use the measure of ( i i i ) . As can be seen
from Fig. 6.3 we have a type of critical solution for h=k*,where for h>h*
there arenosolutions. Forh<h* wehavetwosolutions(oneisthe high voltage
state
variable
parameter
Figure6.3 The fold type curve including predictor-corrector step.
solution andthe other is the low voltage solution).When happroaches h' ~0. <
h+).both solutions merge. At this point theJacobian of the power f l o ~ , solution
is singular. In the mathematical literature these points are called turning points.
fold points. or bifiircation points. An algebraic featiire of the turning point is
given by F, below
F, ( . V' ~ ' . ~ ' ~ ) is singular for rank <11.
F (p:,A:!:)/F.\( ,.v* .A:]:hiis ;I f ul l rank I I and satisfies some nondegenericitj, 1 - cl: )
conditions.
Several techniques ha\e been proposed to calculate these point\. Thew
methods based their iinaIy\is ont\voapproache\ referred toa\ direct and indiicct
inethods.
6.8.3 Detectionof VoltageCollapsePointsUsing
theContinuationMethod
Direct Methods
This approach tries to find the maximum allowdAe wriation of h:that is. an
operating point (. P, A' ; : ) of the equation:
such that theJiicobian at this point is singulur. I t solvesthe tolluct~ing \>stemo f
eq11ations
This procedure basically augments the original set of pobrer tlow eqiiations
= F(.v,h)=0by F,(.v,)L)/I0where his;in u-\fectorn,ith/ I , =1 , Thedisadewmges
of this approach are:
The dimension of the nonlinear set of equations to be solcved is tnice that
for the concmtional power flow.
The approach requires good estimate of the vector h.
The advantage is that. convergence of the direct method is \,ery fast i 'the
initial operatingpoint isclosetotheturning point.Theenlarged systemissol\wl
in such ii way that i t requires the solution of four I I x/ I ( 1 1 is the dimensicln of
the Jacobian F, (. v, h))linear systems, each wi t h the same matrix. requiring only
one LU decomposition.
indirect Method (ContinuationMethods)
Assuming that the first solution (.q,,&,)of F(.u.h)=0,is available. the continua-
tion problem is to calculate further solutions, ( . ~, . h, ) , unt i l one reaches a (.v2.h>).
target point. say at h=A*. The ith continuation step starts from an approsima-
tion of (x,,h,) and attempts to calculate the next solution. However, there is an
intermediate step in between. With predictor, corrector type continuation, the
step i 3i+I is split into two parts. The first part tries to predict a solution.
and the second part tries to make this predicted part toconi'ergeto the required
solution:
Continuation method\ differ among other thing\, i n the follo\Jing: (1)
choice of predictor,( 2) type of the parameterization \trategy. ( 3 ) type of correc-
tor method. ( 3 ) step length control. All four aspects uill be explained through
the formulation of the power flow equations.
I n order to apply the continuation method to the poner tlou problem, the
power flow equation\ must be reformulated to include a load parameter (A).
Thiscan be doneby expressingthe loadand thegeneration at abusas;I function
of the load parameter ( h) .The general form of the ne\+equation\ a\sociated
with each bus i is:
where the subscripts L,,G,,and T,denote bus load, generation, and power out
of a bus respectively. The voltage at bus i is V, 8, and Y,/ a,/is the (i,j)th
element of the system admittance matrix [Yut , 5) . P, l ( h) and Q,,(h) terms depend
on the type of load model. For example for the constant power load:
(6.26)
For the nonlinear model
I so
I n addition, for any type of load model. the active power generation term
can be modified to obtain
where the following definitions are made
PI ,Ql, , , =Original load at bus i, active and reactive respectively
r <
=Multiplier to designate the rate of load change at bus i as hchanges
=Power factor angle of load change at bus i
=Apparent power which is chosen to provide appropriate scaling of h
=Active generation at bus i in the base case
=Constant to specify the rate of change in generation as hvaries
=Initial voltage at the bus
=Frequency dependent fraction of active power load
=Voltage exponent for frequency-dependent active power load
=Voltage exponent for nonfrequency-dependent active power load
=Ratio of uncompensated reactive power load to active power load
=Voltage exponent for uncompensated reactive power load
=Voltage exponent for reactive power compensation
Now if F is used to denote the whole set of equations, then the problem
can be expressed as il set of nonlinear algebraic equations given by Eq. (6.20).
The predictor,corrector continuation process can then be applied to those equa-
tions.
The fir\( task in the predictor \tep is to calculate the tangent vector. Tlii\
tectorcan be obtained from factorizing Eq. (6.21). i.e.,
( 6. 18)
On the left side of the equation is a matrix of partial derivatives multipl ed
by vector of differentials. The former s the conventional power flow Jacobian
VoltageStcihilityAssrsstnerit 151
augmented by one column ( Fj , ) ,while the latter t =[de,dV,dh]'is the tangent
vector being sought. A normalization has to be imposed in order to give t a
nonzero length. One can use forexample
e:t =tL=I (6.29)
where ekis an appropriately dimensioned row vector with allelements equal to
zero except the kfhone, which equals one. If the index k is chosen properly.
lettingtk=+ I .O imposes anonzero norm on the tangent vector which guarantees
that the augmentedJacobian willbe nonsingular atthepoint of maximum possi-
ble system load. Thus the tangent vector is determined as the solution of the
1inear system
Once the tangent vector has been found by solving Eq. (6.30),the prediction
can be made as follows:
(6.31)
= [E][ ;]+ U [i]
where "*" denotes the predicted solution and (3 is a scalar that designates the
step size.
6.9 PARAMETERIZATION
The branch consisting of solutions of Eq. (6.20)forming a curve i n the (.LA)
space has to be parameterized. A parameterization is a mathematical way of
identifying each solution on a branch. A parameterization is a kind of measure
along the branch. There are many different kinds of parameterization. For in-
stance, by looking at a PV curve, one sees that the voltage is continually de-
creasing as the load nears maximum. Thus,the voltage magnitude at some par-
ticular bus could be changed by small amounts and the solution is found for
each given value of the voltage. Herethe load parameter would be free to take
on any value i t needed tosatisfy theequations.Thisiscalled localparameteriza-
tion.I n local parameterization the original setof equations isaugmented by one
equation that specifies the value of oneof the state variables. In equation form
this can be expressed asfollows:
1-52
\%=[q =o, (6.31I
where q is an appropriate \~aluefor the kth element of J*.Now once ii suitablc
index k and the value of 11 itre chosen. a slightly modified Ne~\'ton-Rapliioii
( N- R) power flow method (alteredonly i n that one additional equation and oiic
iiddirionalstate sariable are involved)can be iised tosol\.ethe set of equatior-s.
This pro\.idesthecorrector needed to modify the predicted solution found i n tlic
prc\,ious section.
The algorithm t or static assessnient is shourn i n Figure 6.4.We ciin use ;I
simple example to explain the static ~inulysis proccdure.
6.9.1 StaticAssessment: A Worked Example
Consider ;i \j'steiii is represented by
Uhere h is ;I irariation parameter from h,,=0toh,=A,,,
To begin, s,ol\.ethe s j stem equations at h=0, ~e ha\re
InputSystemData 1
f
SelectContingencies
*
SelectContinuationParameter 1
&
Solvebaseloadflow
.
I I Choosestepsize o I
*
CalculateStabilityindex
A p OT *Q
Figure6.4 The algorithm for static assessment.
2-r -1 0 d.r
2 I - 1 1 [&
=0
[
0 1 0 clh
Sinces,,=1. substituting intothe aboveequation we have:
2 - I 0 c1.r
2 1 4]["
= O
[
0 1 0 dh
and
t1.r 0.5
[."];[ 1.0
dh 2.0
Therefore,
Choosing0=0.I , one gets
.?, =.c*=1.o+0.05=1 .OS
f,=j'" =-2 +0.1=- I .9
h, =h*=0+0.2=0.2
where.t,y,and are theapproximatedsolutions.
In order to find the solutionof F(.u,~.,h) =0. we need to solvetheequation
where r\ is an appropriate value of y.
Chooser\ =y* =-1.9.we havethe solutionof
' 1
VoltcigeStcihility Assessriieiit
XI =1.OS
=-1.9
h, =0.2
Based on the solution of ( x, y, L) , we can get the solution of ( . I - ~ . ~ ! ~ , X~ ) , we
have
Choose CJ =0. I , one gets
, ;, =[I:i: +11
0.2+9
21
Choose q=-1.8, we have the solution of
XI =fl
=-1.8
h, =2 0- 1.8
Using the same procedure until the target system is reached. The modal
analysis procedure is given in the following. System linearization equation is
given by
wherep represents the variation parameter.
At . Y, ~=I , y,,=-2, pi ,=0,the above equation can be reduced to
I56
6.10 THE TECHNIQUEOF MODALANALYSIS
The inodal or eigen\ralue analy\is method i \ a kind ot' sensitivity analy\is but
the modal \eparation provides additional insight. The \ystem partitional niatri-,
equations of the Newton-Raphson method canbe reuritten as
( 6. 33)
Mhere the partitioned Jacobian retlects ;I solb~edpower tlow condition and i n-
clude\ enhanced de\.icemodeling. By letting AP=0.we can write
Uhere J K, is a reduced Jacobian matrix ofthe system.J Kdirectly relates the hi \
voltage magnitude and bus reactive power injection.
Let h,be the ith eigenwlue of J , with cland q, being the corresponding
colurnn right eiget n ~ ~ tor and rowleft eigen\.ector, respecti\re1y.
The ith modal reactise power Lwiation is
where A': cci,=I with c,l thej t h element o f 5,.The corresponding ith mocal
~ v l tage \wiittion is
The magnitude ofeach eigenvalue h, determines the weakness of the corie-
sponding modal voltage. The smaller the magnitude of h, the uteaker the cor ~e-
sponding modal voltage. If h, =0. the ith modal voltage will collapse because
any change in that modal power will cause infinite modal voltage irariation.
If all eigenkraluesare positive,the system is considered voltage stable.This
is a dift'erent dynamic system where eigenvalues with negative real parts i re
stable.The relationship between system bdtage stability and eigen\ralues of the
J , matrix is best understood by relating the eigen\dues with Q-\' sensiti\.ityo f
each bus. J , can be taken as ;t syminetric matrix and therefore the eigenvulies
of J , are close to being purely real. I f all the eigenc~alues are positiLe J Kis
positi\re definite and the V-Q sensitivities are also positiLre. indicating that the
system is voltage stable.
Thesystem isconsidered voltage unstable if at least one of the eigenvalues
is positive. A zero eigenvalue of J , means that the system is on the \yerge of
voltage instability.Furthermore,smalleigenvalues ofJ , determinetheproximity
of the system to be voltage unstable.
The participation factor of bus k to mode i is defined as
For all the small eigenvalues, bus participation factors determine the areas
close to cdtageinstability. In addition to the bus participations, modal analysis
also calculates branch and generator participations. Branch participations indi-
cate which branches areimportant i nthestability ofagi \ mmode.ThisproLides
insight intopossible remedial actionsaswell ascontingencies. which inay result
in lossof voltage stability.Generator participations depict which machines niust
retain reactit,e reserves to ensure stability of a given mode. Figure 6.5depicts
the technique static voltage stability assessment using modal analysis.
For a practical system with several thousand buses it is impractical and
unnecessary to calculate all the eigenvalues. Calculating only the minimum ei-
genvalue of J K is not sufficient because there are usually more than one \ \ ~ a k
modes associated with different parts of the system. and the mode associated
with the minimum eigenvalue may not be the most troublesome mode as the
system is stressed.TheI I I smallest eigenvalues of J , aretheI I I least stablemodes
of the system. I f the biggest of the vz eigen\dues, say mode I I I . is a strong
enough mode, the modes that are not computed can be neglected because they
are known to be stronger than mode ni.An implicit inixrse lopsided simultane-
OLISiteration technique is used to compute the I I I smallest eigenvalues of J , and
the associated right and left eigenvectors.
Similar to sensitivity analysis, modal analysis (seethe worked example at
the end of this chapter)is only valid for the linearized model. Modal analj~sis
can, for example. be applied at points along P-V cur\.esor at points i n lime of
adynamic simUIation.
6.11 ANALYSIS TECHNIQUESFORDYNAMIC
VOLTAGE STABILITY STUDIES
I t isonly recently that theeffectsof systemand loaddynamics arebeing in\.esti-
gated in thecontext of voltage collapse.Thedynamics that arebeing considered
are:
1.58
r
Obtain system architectureand
networkdata
t
I . SolveBaseCasepowerflow
+-
2.Do ContingencyAnalysis
3.Selectadesiredset
+
Detailed
I . FonnFull JacobianMatrix.J
Analysis?
Determineweakervoltageareas
basedon eigenvalues
I
Performparticipation factor
analysis I
I
Compute
A V =g/\-'qAQ
Plot P-V andP-QCurves
L--l_--l
Figure6.5
Static voltage stability assessment using modal analysis.
1. Machine andexcitation systemdynamics includingpower systemstabi-
lizer (PSS).
2. Load dynamics.
3 . Dynamics ofSVCcontrols and FACTS devices.
4. Tap-changer dynamics.
5. Dynamics due to load frequency control. AGC, etc.
While 1, 2, and 3 involve fast dynamics, 4and 5 represent slow dynarrics.
A classification process of dynamic voltage stability vis-a-vis static stability is
shown in Figure 6.6.Here "load" implies demand and "U" represents set pcints
of LFC, AGC, and voltage/VAr controls at substations. .Y, represents the slow
1
VolfugeSfcJhi1if.YAssessriierif
,Subsystems: SubsystemF
x&(x *x , u* I - d)
k f + x $xFut Load)
k$ ( x *XF"U*Load)
N (Voltage Collapse)
Type.11
Instability
Both SubsystemsSandFareStable
I I
Figure6.6 Classification of voltage instabilities.
variables such as the state variables belonging to tap-changing transformers,
AGC loopand center of angle variables i n the case of a multi-area representa-
tion. .rf.represents the fast variables belonging to the generating unit including
PSS andgovernor,inductionmotor loaddynamics, SVCdynamics, and soforth.
Theoverall mathematical model is of the form:
(6.38)
(6.39)
(6.40)
(6.411
Ignoring the more slower AGCdynamics and the faster network transients
(60Hz)we can categorize the variables appearing in Eq. (6.38)-(6.41).
x\ =[ n, ]
i =1, ...,I?
where
1 1 , = transfortner tap ratio
rectangular \ ariables of ith bii4 \ oltage or [ Of ]
\:j
/, =
niachinc terminal currents in machine rcference franie i = I , . . . . 111
de\ired real power of ith generator
cle\ired \wltage at ith gcneratoi bu\
de~tredI oltagc at the bu\ controlled b j tap-changer i
i = 1, 2. . . . . 111
i = 1, 2. . . . . I I I
pr = c'ectorof load parameters to be defined.
The state \w-iablesof the static VAr system ( SVC) control and induct,011
motor \ + r i l l appear i n .v, i f included i n the o\rerall model. As an example \%re
giifebelour the equations for;I I I I machine 11 bus s qwmhaltingp tap-changing
transformers. OnIy the sytichronoiis machine 11nd t ap-changer dy naniics arc in -
c1Uded.
6.1 1.1 Equations of Slow and Fast Subsystems
Foran ur-rnachine,11-bus systemha\ingI - tap-changing transformerh. thefollou~-
ing equations are applicable
Sl o\. v Subsystem
f ast Subsystem
~ , , , c ' % =-E;, - (x(/, - x;,) +E,,,,
i = I .. . ..
tit
- T\,'& =-vRl+K \J,, E,,,, +K ,,( v,,.,,- v,)
i = I . ....111
rlt TlI
KII T / , S=- RI, +~ E , , , , i = I , ....111 (6.43)
tlt TI I
The algebraic equations for the stator and network can be used to andjrze
the system
6.11.2 loadFlow and EquilibriumPoint
Theequilibrium point is calculatedforagivensetof reference points, \',',, ,.Tl,,,,l',,
and agiven demand PI ,and QL,and then solvingthe follonfingequations forthe
\w-iables8?, .8,,.V,,,+], ,V,,.
I62 Cticipter- ri
We may alternatively combine Eq. (6.44)in acompact way as
, I
P:"' =~ v , v , Y , , c o s ( ~ , - 8,- a,k)=o i = I , . ..,I I
i-I
Q:"'
(1
=CV, V, Y, , ~~ - e,- ~( O, a,h)=o i = I , . ..,11 (6.3: )
i-I
and
The parameter vector pl can be defined in terms ofPI,,,,Q,l,l,tt,,,,IZ,,,, etc.
Theequilibrium point is calculated for a given set of reference points V, , ,
,.T,,,. V,,,,and a given demand PI , and Ql, .Theload flow equations areextracted
from Eqs.(6.45)and (6.46)as follows
1 . Specify bus voltage magnitudes numbered 1 to m.
2. Specify bus voltage angle number 1 (slackbus).
3. Specify net injected real power P'",'=PI , and Q'"/'= Ql , at allbuses num-
bered 111 +1 to 1 1 .
Solve the following equations for the variables
e?,...,elf. v,,,, ,.....v,,.
The standard load-flow Jacobian matrix involves the linearization of Eq.
(6.47)with respect to 6:.....611rVlll+lr ...,V,,.After the solutionusing Newtcm's
method. compute
(6.48)
In the above load flow problem one can include inequalities on Q genera-
tion at P-V buses, switching Var sources, etc.From the load flow solution, the
initialconditionsof statevariables inEq.(6.48)canbecomputed systematically.
Theinitial value of V, is V,,,.
Linearization
Define 9' =[ 9:0;] corresponding to generator and load buses. Alsodefine
.v' =[s: 1 .4=[.v;,.Y;, ...J,!]
where
s:= [HI, ...,l l , , ]
and
.I-;=[G,,o,,E;,.E,,,E I,,,, VR,*R,,]i =I . ...1 111
and the algebraic variables asI,, Vq, V,.Alsolet
s,:=(Pi,,(V,l.Qi,(V, ) )
The linearized equations corresponding to Eqs. (6.32)-(6.43)can be ex-
pressed as
A, ,
0
0
0
0
In Eq.(6.48)the variationscorresponding AVlin the nonlinear load charac-
teristic is contained in ASLeand ASL,.
6.11.3 StaticStability(TypeI Instability)
In Eq. (6.46),suppose that both Ais=AtI.=0. Then we have a static situation
with all equations being algebraic. Let all the voltage deviations i n AO?and At!,
be denoted by AP. then the rest of the algebraic variables can be eliminated
(assumingcon\tant power load) to express A P = J I HAp, . I f det (1,) -+0 ;I\
load is increased i t is referred to as Type I static in\tability. i.e.,the \y\tem i \
not ableto handle the increased load.
6.11.4 DynamicStability (Type IIInstability)
Eliminating the algebraic ~ariables i n Eq.(6.39)atid assuming Ail E0. i t can bt.
expressed as
6.11.5 Slow Instability
Theoretically i t shoiild bepossible toeliminateAY,i n Eq.(6.50)using thesingii-
lar perturbation theory and obtain the linearized slow system as At,=A,A.\.,.
The time scale of the phenomena is so large that linearized results may ii8.)t
retlect the true picture. For such a time intensi\rephenomena. nonlinear siii1ul.i-
t i on is recolnmended.
6.11.6 Fast Instability
Fint ue rearrange the Lrar.iable\ [Al,.Av,,A$',]a r [AZ,.&.AV,,....AV,,l1 A02.AOj.
....AO,,,AV,,,+,, ... =[A:.Avl. Next b e a\\ume I , a\ con\tant and load pa-
r:inieter\ a\ con\tant which implie\ Ap, =0. We get
For the constant power case, both AS, and AS2are=0.Otherwise. AS, =
AS,,(V, ) and AS2,=AS,,(V,). For a given voltagedependent load, AS,,and
can be computed. Only the appropriate diagonal elements of B:, C,,and C; Mi l l
be modified and we obtain the system
Now cqis the load flow Jacobian JLI and B2 8'=J , / . The system matrix. A
c. c,I
is obtained as
Atl=A,,,Avl+E h
Usingdrastic assumptions about voltage control and loadcharacteristic\ that the
steady-state stability associated with the system matrix, A,,, can be determined
by examining the load flow Jacobian, J , / .
6.11.7 VoltageStabilityAssessment
The algorithm for Lwltage collapseholtage stability assessment includes static
and dynamic assessment. The algorithm for dynamic stabilitl- assessnient is
shown in Fig. 6.7.
6.11.8 VSTAB-Voltage StabilityAssessment(EPRI)
A more promising method with the trade name VSTAB. uses po~7el- tlow and
modal analysis techniques. I t provides assessment of the proximity to Lvltage
instability and determines the mechanism of voltage instability. I n this method.
the proximity to \voltage instability isevaluated by conducting a series ofp o ~' e r
tlow solutions with load increase until load tlow diLwgsiice is encountered.
When load flow divergence is encountered, the step size for load increase i x
reduced and thepower flowsarecontinued.Thevoltage stability limit is consid-
ered to have been reached when the step sizereaches the cutoff duespecified
by the user. The load level at this point isthe maximum loadabilitjf. This proce-
dure is carried out simultaneously forthe intact systemaswell asforcontingen-
cies. Load increase can be carried out with or without generation scaling. The
slackbusgeneration is not scaled. Loading can be by area or by zonc.
The mechanism of voltage instability is studied i n VSTAB by using modal
analysis. Modal analysis employing V-Q sensitivities can identifqr areas that
have potential problems and provide information regarding the mechanism of
Lroltage collapse. The method is briefly discussed as follows.
The usual power tlow equations can be expressed i n the linearized form.
( 6. 53)
A\I Ael
where
AP=incremental change in bus real power
I66
*
SelectContingencies
+
SelectContinuationParameterP(Q)
*
Sdvebaseloadflow
rk
Increase P(Q)
+
Runthepowerflow
I
<Converge?
Yes
I I
Computeinitial conditionsof statevariable
-
I Calculateeigenvalue A 1
Figure6.7 Summary of the dynamicvoltage stability assessment technique.
AQ=incremental changein bus reactive power injection
A0=incremental change in bus voltage angle
AV=incremental changei n bus voltage magnitude
The system dynamic behavior can be expressed by the first order diffei-en-
tial equation,
167 Voltage Stability As.se.ssineiit
where
X=state vector of the system
V = bus voltage vector
Forthe steady-statecondition X=0,usingtheenhanceddevice models used
in Eq. (6.54).Equation (6.53)can be rewritten as:
( 6. 55)
where
Af,,=incremental change in device real power output
AQ,, =incremental change in device reactive power output
AV,,=incremental change in device voltage magnitude
A0(,=incremental change in device voltage angle
The terms Al l , AI?, and AZ2represent a modified form of J, w. J\>,,JVH.
J,-, in the terms associated with each device. Wecan study the Q-V sensititrity
while keeping f constant. For this analysis we can substitute Af = O i n Eq.
(6.53)togive us upon simplification,
AV=Jk
A0 (6.56)
where
By analyzing the eigenvalues and eigenvectors of the reduced Jacobian JR.
we arrive at
1)=A-xq (6.58)
or
1
v, =-9,
A,
(6.59)
where
his the iIhmodal voltage.
I hS
U\ing modal analy\is.these relative bu\ participation and branch particip.1-
(ionfactor\ c mbecomputed forthe i"' mode.Thecomplete procedure for \tatic
\oltage \tabilit> a\se\snient L ia rnodal analy\i\ is outlined i n Fig. 6.8.
6.11.9 PreventiveControlof VoltageStability
There are t ~ o letelsof lroltagestability enhancement, the fir\tle\reluith delic:e
ba\eci control. the \econd level i \ i n the form of operation-based control. The
Obtain base case
Setfor pre-contingency
solution
Solvetheload
flow
0: cont ingenc
GenerateQVcurves 1 b + ,
-
Newloadlevelor change
Figure6.8 Thc VSTAB algorithm (ClEPRI ) .
voltage stability is improved by optimal systemoperation conditions. The static
analy\is method is used for the determination of prcbmitive control scheme.
System operation conditions are determined by F( 8 . 1 . h)=0.The design ofa
broltage stability preventive control scheme includes the \teps outlined i n Fig.
6.9.
CONCLUSION
Pourer syste111 i.01tage stabiIity involves generation, transniission and distrih i -
tion. So to maintain the voltage stabilitr is crucial t or thc normal operatioti
InputSystemData
4
SelectCritical
Contingencies
1
UseOptimalPower Flow
todoContingency
4
IdentifyandRank contingencieswith
lowstabilitylimitusingVSTAB
+
I Selectthefirstcontingency 1
Anyother
contingencies?
Selectthenext contingencyfromthelist
4
Outputresults
Incorporatetheselectedcontingenciesin
theContingencyConstrainedOPFfor
expandingthelowest stabilitylimit
c
Adjustcontrolparameterstoreflect
optimizedvalues
a I
Figure6.9 VoItage 5t abi1ity pre\.entive controI \cheiii e.
I 70 Chccprer 6
of a power system. In adequate reactive power support from generators arid
transmission lines lead to voltage instability or voltage collapse which halie
resulted in several major system failures (blackouts)such asthe massive Tokyo
blackout in July 1987.
In order to prevent the stability limit being reached or exceeded during a
given contingency, remedial actions need to be recommended. I t is well knov,Tn
that in all cases,voltage instability is caused by inadequate transmission capac-
ity at agivenoperating condition duetoacontingency, which the systemcannot
withstand. Based on contingencies that occur, the distribution of plant gene1a-
tors, transmission tlows and load to meet given stability criteria is usually done
by using effective/economical control actions.
Future work in the determination of adequate remedial measures for stabil-
i t y enhancement have been proposed in past publications, where the correctiLre
control action ishandled asan optimization problem.Thetwo-stage formulati.In
to achieve the desired stability enhancement utilizes the concepts of Chapta S
and this chapter. Thefirst stagehandles voltage stability enhancement while [he
second-stage optimization scheme deals with angle stability enhancement. The
process will leadto aunified index.Hence,whencarrying out stability enhance-
ment based on a selected list of contingencies, only enhancement of the app-o-
priate problem (either voltage or angle) needs to be carried out, thus saving
labor and computational time. Future work in unifying the indices while inc,.>r-
porating the irarious available controls is still a challenge. The reader is intrired
to research further literature in selected references located at the end of . he
book. Also. the annotated glossary of terms supports the chapter.
MODALANALYSIS: WORKED EXAMPLE
Consider the SO0 kV. 322 km (200miles) lines transmission system shown in
Fig. I O(a)below supplying power toaradial load from a 'strong' power system
represented by an infinite bus. The line parameters, as shown i n Fig. 10(b).are
expressed in their respective per unit values on a common system base of 00
MVA and 500 kV.
1 . 1 Compute the full admittance matrix of the two-bus system and write the
power flow equations from the sending end to the receiving end i n the form:
I .2 Hence orotherwise, write down the expressions forthe four ( 4) sub-matri-
ces of the Jacobian i n the linearized load tlow equations as defined by:
VoltageStcihility Assswiierit I7i
(a) Infinite Bus Load Bus
Bus I Bus2
pZ-JQz,
TransmissicmLine
(3c1Load
Shunt Load
Qsh
-7 j
(b) Infinite Bus LoadBus
Bus I Bus2
v,=1. OLOO
"2 =I v, I
Y, =2.142-J24.973
Figure 6.10 The SS0 kV, 370 km (230miles) line tranwii\\ion \ystem \uppl>4ng a
radial load: (a)schematic diagram of the transmission system and (b)the equi\alt.nt
WYEcircuit repmentationof the transmission line.
1.3 WhenP2=1500MW,calculatetheeigenvalues of thereduced Q-V Jacob-
ian matrix and the V-Q sensitivitieswith the following different reactive power
injectionsforeach of the corresponding two voltages on the Q-V curve.
a. Q,=500MVAR.
b. Q, =400MVAR.
c. Values of Q,close to the bottom of the V-Q curve
1.4 Determine the voltage stability of the system by computing the eigenval-
ues of the reduced V-Q Jacobian matrix for the following cases:
a. P =1500MW, Q, =450MVAR.
b. P=1300MW. Ql =950MVAR.
( Aswni ethat the reactiLne pouwQii \ wpplied by ii \ hunt capacitor).
Solution
From the figure. the admittance matrix of the 2-bus system is
2.142 - j22.897 -2.142 + j24.973
-2.142 + i24. 973 2.142 - j23.897
Theexpression tor P and (2 at any bus k are gibfenby:
where
Hence. are interested in only P1and Q2b v i t h V,=0p i i .
Hence the expressions t or the Jacobian terms aregive by:
aP+
J, l l =
=-2.132cose +24.973sine +4.284\>
av,
(a)The linearized power flowequations are
with
The expression for J,,. J,,, Jpo. and Jl, , were given before. For this simple
system, JKis a I x 1 matrix. The eigenvalue lambda of the matrix is the same
as the matrix itself.The Q-V sensitivity isequal to the inlverseoftheeigenkdue.
For each of the Q,s,there are two solutions for the recei\,ing end \*oltage.
Table 6.2summarizes the V, 8, h,and dV/dQwith P =1500MW and Q=500.
400, 306, and 305.9 MVAr. For each case the eigenvalue and W Q sensiti\,ity
are both negatikre at the low voltage solutions, and are both p0sitii.e at high
jroltagesolutions. With Q=305.9 MVAr close to the bottom of the Q-\ c~11-k~
tlVIdQ is large and his very small.
Table6.2 Results for ModalAnalysis Worked Example
High Voltage Solution Lou Voltage Solution
500.0 1.024 -37.3 17.03 0.059 0.671 -66.7 -39.87 - 0. 02S
400.0 0.956 -40.1 12.4I 0. 081 0.706 -60.3 -20.96 -0.048
306.0 0.820 -48.2 0.52 1.923 0.812 -48.8 -0.9SO - I . OS3
305.5 0.184 -48.7 0.02 50.10 0.815 -48.6 -0.700 -1.134
( b) With the shunt capacitor connected at the receiving end of the line, the
self admittance is:
Y,, =2.142-j(22.897- H , )
with P =IS00 MW and A 450 MVAr reactive shunt capacitor. V, =
0.98I , 8=-39.1 degrees. Since B,=4.5 pu., then Y2:=2.132-
j(22.897-4.5)=2.132-j18.397. With this new value of Y2:,the r t s -
duced Q-V Jacobian matrix isJ , =5.348,andJ , is positive indicating
that the system is voltage stable.
with P =I900 MW and a 950 MVAr reactive shunt capacitor, V, =
0.995, 8=-52.97 degrees. Since B,=9.5 pi'.. then YJ2=2.142 -
j(22.897- 9. 5)=2.142-jl. 397. With t h i h new value of Y?:.the re-
duced Q-V Jacobian matrix isJ , =-13.683,andJ , isnegative indicat-
ing that the system is voltage iinstable.
7
Technologyof IntelligentSystems
INTRODUCTION
The previous chapters on voltage and angle stability assessment represent rich
techniques for the analytical solutions of large interconnected systems. Several
important and notable successes have been achieved in solving general simula-
tion based calculations of stability for off-line studies. There remains a large
amount of problems in power systems that are largely solved by human experts
either in conjunction with results from numerical analysis and decision support
systems. Thefollowing features frequently characterize these problems:
1. Inadequate model of the real world.
2. Complexity and size of the problems which prohibit timely computa-
tion.
3. Solution method employed by the human is not capable of being ex-
pressed inan algorithm ormathematical form.It usually involvesmany
rules of thumb.
4. The operator decision making is based on fuzzy linguisticsdescription.
5. Analysis of security such asvoltage orangle isbased on humanjudge-
ment and experience.
These drawbacks have motivated the power system community to seek
alternativesolutionstechniques through theuse of artificial intelligence systems
and variants of its techniques. In this chapter, we present a brief summary of
I75
such techniques including expert systems decision trees. artificial neural t i c a t -
works, fuzzy logic systems,and their hybrids.These rechniqiies have been ein-
ployed for solving various p o ~ ' e r system operation and planning probleins and
especially tor the assessment of transient and i dt a gc stabilityprediction, insla-
bi1ity preveiit i on. and conti i i gency ranking,etc.
Soiiie cornriional i t y exists among ii i t e1Iige11 t systems approaches. The ~ 1 ' s -
temreq11 iremeiits forde ve1oping or;issessirig iiiteI1igent systemsapproaches L re
;Is tollou~s:
1 . Ability to identify system states.
2. Selecticityo f controls.
3. Learningability to update knowledge.
4. Coordination of tasks.
5. Flexibility.
6. Ability to handle uncertainty.
An expert system (ES). also referred to ;is Knowledge Based System. em-
bodies human expertise i n ii narrow t'ield or domain i n a machine implement:i-
tion form. I t utilizes elementsof human knowledge to providedecision support
comparablc tothe humanexpertand iscapableofjustifyin,(7 ' at ;I l e ~ d Its reason-
ing. I t separates the inference mechanism f'rom the kno~vletlgeand uses one o r
niore knowledge structuressuchas productionrules frames.semanticnets. pr xl -
icatc calculus. and objects to represent knowledge.
The function o f the expert system consistsof its ability t o collect ui cl store
an expert's ability to solve a problem sothat a lionexpert can use i t . The fuiic-
tional componentsof an expert system are representedand defined here.
I . Knou.ledge Base: Contains all relc\iant informaticm about the s>'slem
11nderstudy.
2. User intertxe: Input/output or so called nian-machine interface gi\,es
the necessary informationand decision rules tothe user.
3. Inference Engine: Analyzes the system using i t then rules based on
go;i I/cl11t;IoI.-i cnt ed striitegy cii11ed foru.ard/back\+(ard ch;i i11ing.
Other modules such iis control iiiechanism and niodification loop lire iisii-
iiI 1y t:xc1iided frotn expert systemsto iichiebt robustness.
I n the knowledge base, forexample i n a power system. the database gves
the operating characteristics of the system through stateestimation.The re5ults
from off-line studies are iist'd ah setso f rules in the knoufledgebase.
Three special ti'atures distinguish Expert Systems from traditional pvuer
s> r stem aniiIysis:
1. The expert system allon~seasy tlexibility ot' mrinipulation ot' doi-iain
specific kno~~l e dge without hai.ingto \\.:itch forthe impact o f chaiigeh
on the u ' a ~n'e;ire reasoning i t .
2. The expert system is concerned with manipulating symbolic informa-
t i on rather than the numerical information directly.
3. Theexpert system addresses problems where knowledge may be deter-
ministic and more imprecise and allows forcertainty i n reasoning.
7.1 FUZZY LOGICAND DECISIONTREES
Fuzzy set theory deFdoped by Prof. Zadeh i n 1965 as a mathematical
means of describing vagueness i n linguistics. A fuzzy set is ;i generalization of'
ordinary sets that allows assigning ii degreeof membership for each element to
range from (0,I ] iiiterLral.
A fuzzy set differs from a crisp set which has a unique binary membership
fraction i n that the fuzzy set has an infinite number of membership function3
that inay represent it. Fuzzy reasoningoffers away tounderstand system beha\,-
i or through int erpoIati on approxiinate1y between input and o i l t pi1t situations.
Fuzzy logic isbased simplyon the way thebrain deals urith inexact intimnation.
A fuzzy system combines and applies sets with fuzz). rules to problcnis \ +, i t h
overall complex nonfinear behavior. As a structured normal estiniator, i t ex-
presses fuzzy if the rules of someexpert knowledge.
Fuzzy set theory offers a new method for niodeling the inexactness and
uncertaint y conce rning deci sion-making.Fuzzy Iogi c im pro~ ' e s the potenti a1 tor
modeI ing huinan reasoning and for presenting and uti1izing 1ingU istic descrip-
tions in a cornput erized i nferencing enviroilment.
The two methods of deiveloping fuzzy models arebased on:
1. Laws of cause and effect which use rules of relations described hqr
reasoning i n variable sets theory.
2. Laurs of transition. which use ordinary equations to express. ciiiisc' and
effect reIationshi p.
Fuzzy set theory uses the concept of possibility defined as a number be-
t u wn one (completelypossible) and zero (totally iriipossible)instead of proba-
biIit y whi ch nieasures the appropriate uncertainty of avai 1ab1e st atisti c;i1 i nfor-
mation. Where probability fails without statistical data. fuzzy set theorj,doesa
betterj obthan other intelligent systemssuch asneural nets and expert systems.
7.2 ARTIFICIAL NEURAL NETWORKS
While f i i u ~ , expert 5yrtem. and deci\ion tree technique\ are rule-bii\ed. there
arecertain AI techniques that use Artificial NeLird Netuforh(ANN) pattern rcc-
ognitionclu5tering \trategies. An ANN isconsidered i i machine that i \ dc\igiicd
to model the M orh the brain performs in dealing ufitha particular t a 4 of intcr-
est.The neural network is usually implemented using electronic components or
simulated in a software computation. A neural network performs computatim
through learning.It isamassively parallel distributedcomputation that istrained
with the appropriate data and reproduces trained strategiesand adapts according
to charging situations.
Pattern recognition techniques on the other hand, act asclassifiers depeiid-
ingon the pattern of interest.Clustering isalsooneof the unsigned ANN lea-n-
ingtechniques.Instead ofselectingatraining set,weconstruct asetof unlabelcd
trial factors. These techniques use Euclidean distance to estimate the member-
ship in a pattern groupor to classify according to some prevailing rates.
Of the three intelligent systems approaches discussed here neural netwo*ks
are best suited to problems where voluminous data and a complex nonlin:ar
relationship exists between the input and output patterns, such as instability
assessment and load forecasting problems. Artificial neural network schemes
share the ability to identify the model structure of unknown systems with input/
output data and efficiently solve combinational problems, which define their
relationships. An overview of artificial neural networks and some significant
neural learning algorithms is provided in this chapter.
The evaluation of intelligent systems has shown that neural-based ap-
proaches are most beneficial forpower system stability assessment, due to their
ability to generalize and "learn" from historical information. A basic otferve w
of neural computing practices and methods is protided i n the next section I n
comparison to neural nets, expert system approaches are rather system specific
and rigid. Fuzzy set theory is relatively new as a power system analysis tool.
Further validation of membership functions and the use of network opera1ion
support using real fault situations warrants further investigation of fuzzy logic
as a power system analysis tool. A review of the salient points defining the
three intelligent systems discussed here is provided in Table 7.1.
ANNs are parallel distributed processing systems composed of nonliiiear
processing elements that perform in a manner similar to the most elementary
functions of biological neurons. For instance, ANNs possess the ability to Icarn
fromexperience,generalize frompreviousexamples tonewones,and toabsfract
pertinent information from examples containing irrelevant or incomplete data.
ANNs are not suited to simple mathematical tasks, such as computing t vl tage
drops along ;I feeder. However, ANNs are beneficial i n solving a great number
of pattern recognition problems, that are either computationally burdensome or
impossible for conventional iterative programs to solve.
Neural networks offer the following advantages:
ANNs have the ability to learn and construct acomplex nonlinear mapping
through a set of input/output examples. The network architec-ture
allowsfor easy training without a need for a structured model.
Table7.1 Features of Intelligent SystemsApproaches
Intelligent AppIi caIion
systems Systemcomponents Functionality areas
Artificial neural Neurons Learning a mapping (rela- C1assificati on
networks Weighted connections tionship)from exam- Estimation
between neurons ples
Learning algorithm SeIf-organi zation into
classesor clusters
Associatilre memory
Optimization
General
Expert systems Knowledge base Learning from inference Estimat i (3i i
User interface Data structure Decisi(3n
Inference engine Linkage to external appli- making
cations
Distributed kno\+.ledge
Specific
Fuzzy logic Crisissets Handling of human sub- Decision
Membershipfunctions jectivity making i n
Fuzzy sets Human Knokvledge in casesof in-
"thinking" machine exact or
Handling of human ambi- ambi gU ous
guity in communi- data sets
cation
Input variables can be easily added or deleted. Correlated or uncorrelated
datacan be utilized.
Neural networks have a superior noise rejection capability that can effec-
tively deal with uncertainties of the actual process.
Neural networks execute very quickly. Most of the calculation overheads
occur during the initial off-lines training.
Neural networks consist of a large number of parallel processing units
which can be implemented using general orspecial purpose hardware.
Hence neural networks can relieve the burden of computation from the
Energy Management System computers.
Before neural networks can gain the necessary recognition as useful prob-
lem solving tools in the power industry. certain fundamental issues have to be
addressed. Some are associated with neural network fundamentals and others
are problem dependent and are listed below:
Determining the proper consistency of the training and testing data slits
including. the number of patterns. input dimensions. and statistical
properties. i n order to pro\.ide adequate generalization and kiiowlecge
retention.
Neural network decomposition in order tofind the optimal stageorlocation
for the use of neural netm'orks.This is an important consideration in
dealing with large-scale systems.
The use o f feature selection and clustering techniques for data prepro-
cessing. This can help achieirereduction i n both dimensional and c(m-
binatoriaI coinpIexity.
Neural net uwk must not be allowed to niemorize the training data nor
becoiii e sati i rate d.
ANNs ha\re been developed i n numerous configurations. Despite their di-
krersity, some cominonality exists among the various network paradigms. In the
f o11 o\ving sections sonie fii ndamenta1s o1' netU'ork c(1inpUtirig are presented,
7.2.1 FundamentalsofANN
We review ;I \pecific neural network architecture, i t \ actikation function and
element\ of training. Thi\ is followed by LI discu\sion of some of the factor\ of
1earning ii1gorithnis itic1lidirig bach-propagation.coiinterpropagation andc1L \ter
Iearning ;i 1gorit hms.
Architecture
The artificial neuron, also known as the processing element, is riiodeled to re-
flect the configuration of the biological neuron (Fig.7.1). A set of input? are
applied. each representing the output of another neuron. Each input to a nctiron
is multiplied by ;I weight, corresponding to the input's connection strenglh o r
importance to that neuron.
The relationship described above is illustrated i n Fig. 7.2. Here, ;i sct of
inputs, are applied to the neuron. The inputs, collectively defined ;is the \rector
X,are niultiplied by their associated weights (W, , )before being applied t o the
suriiriiationblock, E.The inputs X, now multiplied by the weights. collectiitely
defined asthe vector W.are summed at the summation block. Zto product. the
activation levelA, of the./"'neuron in the next layer.subject to athreshold \alue,
8.This relationship c mbe stated i n vector notation as follows:
Activation Functions
The activation level, A,, is further processed by an acti\,ation function. F. to
produce the output signal of thej " ' neuron, Out,.The activation functior rnay
be ii simple linear function, such as again. K:
Figure7.1 Component4of biological neurons.
Figure7.2 Components of an artificial neuron.
IN2
Out,=KA,
Where, K is aconstant. Or i t can be a threshold function, such as:
Out=0 i t A, <8
Where 8 is a constant threshold value. Alternately, F' may be a function that
more accurately simulates the nonlinear characteristics of a biological neuron.
and permits more general network functions. One such function is the sigmoid
activation function. which is the activationof choice forthe majority of conteni-
porary neural systems.
The sigmoid function is known as a "squashing function," since i t con-
presses the range of A,,sothat the valueof Out,neverexceeds somelow boun3-
aries.Theboundaries aredefined as0and 1 in thecaseof the sigmoid function.
For a large value of A,, the denominator in Equation 7.4would approach I , and
thus produce an output close to 1. Likewise, for a value of A,equal to the --oo.
denominator i n Equation 7.4would approach +=. and thus produce an output
approaching zero.
In order to provide a clearer understanding of' the role of the activation
function in neural computing,we think of the activation function asdefining a
nonlineargain in ananalog electronic system.Thegain iscalculated by cotnpirt-
ing the ratio of the change in Out,to a small change inA,. Thus. the gain is the
slope of the curve at a specific activation level. The gain may vary front a I OU,
value at large negative activation levels, to a high \.dueat the zero point, and
back to a low value at large positive levels. The central high gain region of the
sigmoidal activation function solves the problem of processing small sign:Is,
while its regions of lower gains at positive and negative extremes are app-o-
priate for large activation lelrels.Thus a neuron ciin perform adequately olrer ;I
wide range of input levels.
TrainingtheNeural Network
A neural network istrained sothat agiven stimulus (setof inputs)will produce
a desired response (output set). Each input or output set is referred to a; a
vector. Training is the process by which input vectors are continuously applied
to a network, while adjusting the unknown network components. until e;ich
input vector produces the desired output vector.
Training algorithms can be classified assupervised or unsupervised. Super-
vised training requires pairing each input vector with a desired output, ortarget
vector, together these are called a training pair. Usually, neural networks are
trained with a large number of these training pairs. A rule-of-thumb is to have
at least 100 such training pairs for each input neuron, in order to represent a
wide variety of system conditions. In the supervised training process an input
vector is applied, the actual output of the network is determined and compared
to the target value. The difference between the actual and target values is the
error that is then fed back through the network to update the interconnection
weights, and thus, minimize the error. The entire process is repeated until the
error for the entire training set has met an acceptable tolerance level.
For unsupervised training, the training set consists solely of input vectors.
These types of training processes group similar input setsintoclasses according
to the statistical properties of the training set.There is no way to predetermine
which output classes will be produced by a given input vector. Therefore, unsu-
pervised training generally requires some postprocessing of data to transform
the output classes into some meaningful format. Usually, all that is required is
a simple identification of the input-output relationships defined by the netu-ork.
Some significant training algorithms are presented and discussed in the next
section.
7.2.2 Back Propagation LearningAlgorithm
Of the existing neural network paradigms, the back propagation learning algo-
rithm is an interesting approach to the solution of classification problems. The
back propagation learning algorithm is atechnique foroptimizing the intercon-
nection weights in an ANNby minimizing the global errorbetween the desired
and actual output of all the cases evaluated during the training session.
The procedure begins with the presentation of the input stimuli and the
desired output of all the training casesand the initialization of all the intercon-
nection weights to random values. The input stimuli are fed forward through
the network to compute the actualoutput values. Theerror between the desired
and actual outputs are computed and propagated back through the network to
adjust the weights. This process is repeated until the global error satisfies a
prespecified tolerance value.
7.2.3 Counter Propagation LearningAlgorithm
The counter propagation network is not asrobust asthe back propagating tech-
niques, but i t can provide quick solutions for applications that cannot tolerate
long training sessions. The network functions as a look-up table capable of
generalization. The training process associates input vectors with their corre-
sponding output vectors. that may be either, in binarjfor continuous form. Tlic
generaIization capabi1i t y of thecouriter propagation network a1lows it toext r;i2t
;i desired output response e \ wi urhen presented u!ith purtially incomplete or
incorrect input data. This makes the counter propagation network useful fl)r
pattern recognition,patt e1-11CO nip1etion,and signa1 enhancement rippI ication s.Iii
its simplest form the counter propagation network is ;i combination of two MY I -
known training algorithms. the Kohonen, and the Grossbet-gOutstar.
The Kohonen layer functions in a winner take all fashion. That is, for ;I
giiteii input vector only one output neuron in the Kohonen layer is acti\rat:cl
(oiitputs;i logical one),all of remaining outputs of the Kohonen layer are h c Id
at zero. The net output of each Kohonen neuron is the cveighted sum of'the
inputs. therefore, the net output wi t h the highest \ d u e is the "winner."
The Grossberg layer functions in a similar manner with its net output eqiial
to the sum o f the u,t.ighted values from the Kohonen layer. Since there is oril~,
oiie nonzero term from the Kohonen layer the calculution is simple. In fact,Ihc
onl y action of each iieuron in the Grossberg layer is to output the value of rlie
weight that connects i t to the sing1e lionzer o kohonen tieU r o11.
The limitations of the counter propagation network mike i t inttriorto bi ch
propagating techniques for most mapping net"ark applications. Ho\ve\rer, 1lic
counter propagation IictLvork has been found to be useful for r;ipid prototj' piig
of systems that can e~witually yield to the greater accuriicy and rohustncss ot'
back propagati11g itIgorithms.
7.2.4 ClusteringlearningAlgorithm
A clustering-based ANN M ' ~ S implemented for applications to pou.er sys1ciii
problems.Theclustering ANNissimilar in nature to the ART1 ANNdeveloped
byCarpenter and Grossbergand Kohonen's self-organizing map.Thealgorilhm
iiic1ides 11 feature extriiction scheme for st u tistical clerivii tion of ii moi-e rot111st
set of tcaturc \rariables than those that could beprokrided through purely het ris-
tic methods.
The clustering ANN utilizes both the superczised and unsuper\sised traiiiing
methodologies. The supervised module was used to synthesize the mapping hc-
t c c w n the input stimuli atid the desired output clusters forriicd bq' the tinsuper-
\,isedinodule. Continuous handshaking het\\.t.cn the tcvo modules rakes place
unt i l a prcspecified tolerancc is satisfied.
7.3 ROBUSTARTIFICIAL NEURALNETWORK
Back propagation ANNI iire well suited to enhancement of the conc~entional
Tran\icnt Energj Function (TEF)proccdure Iijr many re;isoii\. First. the) can
be trained by the results of off-line TEF studies. Second, they can be quickly
called on-line toclassify asystem subject todisturbance ;IS secureor potentidly
insecure. Third, they can generalize from cases that MI X used i n training to
classifycases that were not previously evaluated. Fourth, they arerobust enough
to handle a variety of systems. And finally,the multi-layer perceptron net\zurks
ciin be used to overcome nonlinearly separable problems.
The de\.elopment of the robust ANN invol\.es the selection of ;in appro-
priate network architecture todetermine how the neurons ure connected. and the
selection of an appropriate learning algorithin to determine how neurons proccss
their information and how the connection strengths betnfeenlayers \ \ weopti-
mized. and the application of training data acquired from off-line.
7.3.1 Network Configuration
Theba\ic neuron model for the back propagation algorithm contain\ input\. t/.
that are adjusted by weights, and summed.After ufhich,a bia\ or thrcshold \ t y l ,
value. 0, is added. and the total is passed through the acti\ation function F( i1
;is an output. Mathematically this process can be rcprcwnted by:
Where I!$, is the weight applied between i"' neuron of input layer and,j"'ne~iro~i
of hidden layer and /,11,1, represents the total number of input neurons.
Before starting the training process, all of the Lveight must be initializcd to
small random numbers (-0.5 <\t'/,, <+OS ) . This ensures that the net\s.orkis not
satiirated by large values of the weights. The random nature of the initial
Lireights also allows the network to learn unequal final values of the \+!eightsif
that is what is required for optimal performance.
7.3.2 Overview of Training
Training the robust artificial neural network requires the following step-bj-step
procedure:
I . Apply all the input vectors from the training set to the netuwh.
2. Calculate the output of the network foreach input \rector.
3. Compute the error between the actual network output and the desired
output for each input vector.
4. Calculate the global errorasthe averaged sumof the squared errorhfor
the entire training set.
IN6 Clttrpter 7
5. Adjust the weightsand thresholds i namanner that minimizes theerror.
6. Repeat steps 1-5 until the global error is acceptably low.
Upon completion of the training procedure the network is used for recogni-
tion and the weights arenot changed. Itmay be seenthatsteps 1 and2constitute
ii "forward pass" through the network. in that the signal propagates from the
network input to its output. Steps 3-5 are a "reverse pass," where the globiil
error propagates backwards through the network and is used to adjust the
weights and thresholds.
The first two steps in the training procedure can be expressed as follow;:
an input vector,X,is applied and an output vector,Out. isproduced. Theinput-
target vector pair, X and T,comes from the training set.Theoutput vector, Out.
is produced as a result of the application of X to the network. Calculations n
multi-layer neural nets are performed on a layer by layer basis, beginning wi h
the hidden layer closest to the input layer. The weighted values of the inp.it
vector are summed in the form:
Where
A, =active lektel of neuronjin the hidden layer
X, =iIhinput from the input vector X
W,,,=interconnected weight applied between i"' neuron of input layer and j " '
neuron of hidden layer.
The nonlinear sigmoidal activation function is applied to the activati.In
level, A, in Eq. (7.7).to dri\.ethe value of hidden neurons toward one of the
two states (0or I ;stable or unstable).
( 7. 7)
Where:H, =output of neuronj in the hidden layer
8,=dynamic threshold value of neuronjin the hidden layer
The addition of the threshold value, 8,increases training efficiency. The
same process is applied for networks with additional hidden layers with the
output of the preceding layer serving as inputs to the next layer in the netwcrk.
In the output layer the output nodes aredefined asthe weighted sum of the
outputs from the preceding layer. No activation function isapplied at the output
node. Theoutput neurons arecomputed as follows:
Where:Out,=output of kIhneuron in the output layer
W$,,a =interconnected weight applied between j"' neuron in the hidden
layer and kt hneuron of the output layer.
Therefore,thecalculationoftheoutputsof thefinal layerrequires theappli-
cation of Eq. (7.6)to each layer,from the network's input to its output. In the
reverse pass, defined by steps 3-5, two tasks must be accomplished. First, the
weight of theoutput layer must be adjusted. And second,the weight and thresh-
olds of the hidden layers must be adjusted.
In order to adjust the weights in the output layer an update factor for each
weight must be computed. First, the global error, E, for the entire training set is
computed using Eq.(7.9)astheaveragesumof thesquared differences betureen
the actual network outputs {OutL] and the target, ordesired values, {T, ].
The error signal is multiplied by the derivative of the signioid function
[Out,(I - Out,)]for the kIhneuron of the output layer, thereby producing the 6
value for the neuron.
the 6is then multiplied by the output value from the previous layer that corre-
sponds to the weight being adjusted.
=6, H, (7. 11 )
where:A H , ~ , ~ =update factor for interconnection weight between thej"' neuron
in the hidden layer and the k"' neuron in the output layer.
An identical processisperformed foreachweight proceeding from aneuron
in the hidden layer to a neuron in the output layer.
Adjustment of the weights between the input layer is slightly different,
since the hidden layers have no target vectors. Therefore,the undate process
described above cannot be used forthis set of weights. Equation (7.11 ) is used
forallof the weights, regardlessof location, however, the 6term forthe weight
between the input layer and hidden layer must be calculated without the benefit
of the target vector.
Instead, the 6 values from the output layer are used to adjust the 6values
i n the first hidden layer :incl the 6's are propagated backisrurds to :ill of thi:
preceding layers.
Consider a neiiron i n the hidden layer immediately behind theoutput layer.
During the for\\w-dpass the output values of each layer are processed forward
to the next layer via the interconnection weights, i n ternis of the 6. values ar:
passed i n the re\'ersedirection from the output layer to the hidden layer,during
the training period. Each weight between the hidden and output layers is multi-
plied bjr the 6\ d u e of the output neuron to which i t is attached. The ~ ~ ~ l i i e (+!'
6for the weights in the hidden layer preceding the output l aj wi s produced bj ,
summing all of the products from theoutput layers and niultiplying them by the
deri\.ati\resof the sigmoidal function, Lit that particular hidden node. Thus,
The6values are calculated foreach neiiron i n ;i given hidden laqrer,and i I 1
of the ueights i n that l a ~ wareatljusted according11 uhing theexpression belou:
for;i multi-layer networh this process is repeated layer by layer. unt i l the i nput
layer is reached and 2111 of the unknowns ha1.ebeen adjusted.
7.3.3 Summaryof Learningvia Robust Back PropagationANN
I n SLII11111iiry. the learning period consists of ii gibm network uith ;Irandom w t
of \ifeight\faluesbeing simulated by r i l l of the inpur vectors in a training set.A
global system error is computed, which is iisually quite large initially. thus -e-
quiring rici-jijustmentof the interconnection weights via a process known iis train-
ing. A irarietyof training algorithmsexist. ofwhich ii Robust Back propagation
has significant advantages o \ ~ r classical Back propagation algorithms. Using
the robust procedure, update factors for all of the weights in the network .ire
computed based on the global system error between the actual netuork outputs
and the target values. A gradient descent method is used to track the global
error bvi t hi n an acceptable tolerance level by optirnizing the interconnection
weights.Theconjugate gradient procedure possesses the ad\,antagesof simplic-
ity and minimal storage requirements i n mapping ii given input to its coi-re-
sponding output state.A successt'iil learning exercise will yield a stable sel 01'
utights. which exhibit only minor fluctuations in the \alue if further learning
is attempted.
Precautions should betaken against "over training," uhich occurs tirhen the
network has become too specific to the training data and is no longer ablc to
generalize to previously unseen data. Topre\,entthis occut-rencethe networL is
periodically presented with test data during the training procedure. If the global
system er r or increasesfromonecheckpoint to the next then the training proce\+
was halted. A comparison of the robust tool to classical Back propagation para-
digms is pre\ented in Table 7.2.
Table7.2 Classical Back Propagation Versus Robust ANN
Cla\s1cal Ad\antage\ of
Fe;iti i re bach propagation Robu\t ANN robu\ t ANN
Update scheme Arbitrary threshold Dynamic thres- StreilmIines Ieiir11it1g ~ I * o -
hold cess by updating dqm-
namic threshold during
e\.ery iteration of learn-
ing session
Indi\,id ual ci>w amic
threshold f o r eacli liid-
den node increases tic-
cClr3c4'
Data prexenta- Stocha st ic presentati on Batch present;I- Batch data presentati on
tion tion order can remain se-
quential \ahicli is much
lesscomputationall>,in -
tensi\,e than st ocliil~ tic
data present ation.Ihat
must be random
Error function Local error function Globalerror Globalerror tunclion rep-
function resents true s)'steiii
wide errorforeach iter-
ation i n the learning
sexsi on
Weights ilrc llpdiited f or
entire training data at
each iteration
Speed Neural Ware Profes- 152iteration for Significant speed en-
sional II/Plus: the XOR pro- hanceinent overcIa s si -
Over 3000 iterations blem cal methods
for the XOR
proble111.
Rumelhart, Hinton &
Williams 2-15 itera-
t i o n forXOR
probIeni
Thebenefits in speed and accuracy of the Robust ANN overclassical Ba:k
propagation methods can be traced back to three main characteristics. First, the
Robust ANN utilizesadynamic (adaptive)threshold value foreach hidden node
during the training session that is updated along with the interconnecti,)n
weightsasopposed tothearbitrary thresholdsthatallowforgreaterindividual t y
i n the learning process for each hidden node and thus have a positive impact on
the accuracy of the final output.
The second and third beneficial traits of the Robust ANNarecoupled.Sirct'
the Robust ANN utilizes a global error function that updates the weights and
thresholds based on the average errorof the entiredata set,the training datacan
be presented in an identical sequentialorderforeachpass. I ncontrast, thecla>si-
cal Back propagation method utilizes alocal errorfunction that updates the data
after each training vector and thus the data must be presented in a differxit
random order for each pass, which increases the computational complexity md
thus, the required time forconvergence.
7.3.4 Conclusions
Several different types of neural nets have been discussed in this section. This
is by no means an exhaustive list. A selection has been made based on applca-
bility to power system problems. While these have been inspired by biological
neurons. the issue is not their ability to exactly model biological systems. hut
their importance as another solution paradigm for power system problems.
The area of artificial networks is a very active field of research. For i n-
stance, i t includes neural smithing for optimal architecture and learning design.
Pruning techniques are used to determine the optimal number of neurons; to
solve a specific problem. and adaptive learning algorithms t o avoid retrairiing
forlargedata sets.Training with noisecanbe used i norder-toovercome o\ wf i (-
ting problems.
There are continuing efforts to combine and enhance ANN methods *.s,ith
regularization theory. stochastic, Bayesian,orother statistical and pattern ret.wg-
nition techniques in order to define optimality criteria for ANN perforinancc. I n
fact,ANNs might be considered a set of highly adaptive statistical tools.I t itor-
mation-theoretic concepts are introduced i n order to measure how ANNsgcnt'r-
alize, i.e.,how they will perform on unknown test data.
Finally, the combination ofneural networks mrith other AI techniques like
symbolic structures from expert systems, fuzzy logic,and genetic algorithr-isis
being explored for many technical applications. Here complex problem: are
described with fuzzy techniques or expert rules and the subsequent classitica-
tion, regression, or optimization tasks are solved with ANNs.
7.4 EXPERT SYSTEMS
It is not easy to give a precise definition of an eqwrt system. because the con-
cept of the expert system itself is changing as technological advances i n com-
puter systems take place and new tasks are incorporated into the old ones. In
simplewords, i t canbe definedasacomputer systetn that models the reasoning
andaction processesof ahuman expert inagiven problem area.Expert systems,
like human experts, attempt to reason within specific knowledge domains.
An expert system allows the knowledge and experience of one or more
experts to be captured and stored in a computer. This knowledge can then be
used by anyone requiring i t . The purpose of an expert system is not to replace
the experts, but simply to make their knowledge and experience more widely
available. Typically there are more problems than there are experts aF&lable to
handle them. The expert system permits others to increase their productivity.
improve the quality of their decisions, or simply to solve problems when an
expert is not available.
Valuable knowledge is a major resource and i t often lies with only a fen
experts. It is important to capture that knowledge so others can use i t . Experts
retire,get sick,moveontootherfields,andotherwise becomeunavailable.Thu\
the knowledge is lost. Books can capture some knowledge. but they lea1.e the
problem of application to the reader. Expert systems provide direct means of
applying expertise.
An expert system has three main components: aknowlecigc~Imw.an irfkr-
erice erigirie, and a rrtnri-machineiriterfclce.The knowledge base is the set of
rules describing the domain knowledge for use in problem solving. The prime
element of the man-machine interface is a working memory which serves to
store information from the user of the system and the intermediate results of
knowledge processing. The inference engine uses the domain knowledge to-
gether with the acquired information about the problem to reason and provide
an expert solution.
7.4.1 A WorkingDefinition
An expert system is an artificial intelligence (AI) program incorporating a
knowledge base and an inferencing system. It is a highly specialized piece of
software that attempts to duplicate the function of an expert in some field of
expertise.Theprogram actsasan intelligent consultant oradvisor inthedomain
of interest, capturing the knowledge of one or more experts. Nonexperts can
then tap the expert system to answer questions, solve problems, and make deci-
sions in the domain.
The expert system is a fresh new, innovative way to capture and package
knowledge. Its strength lies i n its ability to be put to practical use when ;in
expert is not :i\xilable. Expert systems riiake knowledge more widely a\,ailable
and help overcome the age-old problem of translating knowledge into practical
results. I t is one more way the technology is helping us to get a handle o n 11-e
o\.ersupplyof information. All AI software is knowledge-based ;IS i t contairrs
useful fricts,data. and relationships that are applied to a problem.
Expert systems, howe\rer, are LI special type of knowledge based systeii
theycontain heuristic knowledge. Heuristics areprimarily fromreal world exp8:-
rience. not from textbooks. I t is knowledge that comes directly from those peo-
ple-the experts-who have worked for many years within the domain. I t is
knowledge derived from learning by doing. I t is perhaps the most useful kirid
o f knowledge, specifically related to everyday problems. I t has been said that
knowledge is power.Certainly there istruth i nthat but ina more practical senw.
knonrledge becomes power only when i t isapplied.Thebottom line i n any ficld
of endeavor isRESULTS,somepositive benefit or outcome.Expert systems i re
one niore \+'ayto achieFre results fiisterand easier.
7.4.2 Characteristicsof Expert Systems
One advantage of the expert system is its permanence. Human expertise c;in
quickly fade. regardless of whether i t involves mental or physical activity. In
expert must constantly practice and rehearse to rnaintain proficiencjc i n sono
probleni area. An significant period o f disuse can seriously affect the expel1' s
perforrnance. Once expertise is acquired i n an expert system. i t is around f or -
ever, barring catastrophic accidents related to memory storage. Its perinanelice
is not related to its use.
Another ad\.antageof the expert system is the ease c+.ith which i t ciin be
transferred or reproduced.Transferring from one human to motheristhe labi)ri-
ous, lengthy. and expensiLVeprocess called education ( or i n someciises. knoufl-
edgeengineering).Transterring an expert system is the tri\.ialprocess of cop>'-
ing o r cloning a program or data file. An expert system is also much easic.to
document. Documenting human expertise is extremely difficult and time con-
suming. Documenting :in expert sj'stem is ;I straight forwud mapping bet\+ ecn
the way in that the expertise is represented i n the system and the naturiil 1;in-
gUage description of that re presentiition.
An expert system produces niore consistent. reproducible results than does
the human expert. A human expert may make different decisions in idenlical
sitiiations because of emotional factors. For exaniple. 11 human ma ~ ' forgct to
use a n important rule i n ii crisis situation because of time pressure or stress An
expert system is not susceptible to these distractions.
A final advantage of the expert system is its low cost. Human experts,
especially the highly skilled ones, itre very sc;irce and hence very expena9i\fo.
Expert systems, in contrast, are relatively inexpensi\re. They are costl}!to de-
velop but relatively inexpensivetooperate.Theiroperatingcost is.just thenomi-
nal computer cost of running the program. Their high de\relopment cost ( j'ears
of effort of high-priced knowledge engineers and domain experts)is offset b!,
their low operating cost and the ease of making new copies of the system.
Although expert systenis tend to perform well, there are important are;is i n
which hurnan expertise is clearly superior to the artificial kind. This does not
reflect a fundamental limitation of expert systems or AI , just the ci~rrent state-
of-the-art.
One such area is creativity. People are much more creative and i nno\ . at i \ ~
than even the smartest programs. A human expert can reorganize information
and use i t to synthesize new knowledge. while an expert system tends tobeha\-e
i n a somewhat uninspired routine manner. Human experts handle unexpected
ei'entsby using iniaginati\reand novel approaches toproblem sol\.ing,including
drawing analogies to situations in complete1y different prob1 e111 dom;iins. Pro-
grams have had little success in doing this.
Another area where human expertise excels is learning. Human experts
adapt to changing conditions: they adjust their strategies to conform to n e ~ '
situations. Expert systems arenot particularly adept at learning new concepts or
rules. probably because i t is a very difficult task that has ;il\+aysbeen made i n
dekrelopingprogrms that learn, but these programs tend to ~+wr ki n cxtremely
simple domains and do not do well when confronted \+.it11the complexit!^ and
detaiIs of rea1-wor Id prob1ems.
Human experts can make direct use of complex sensory input, Likther i t
be sight. sound. taste, or smell. But expert systems nianipulrite sqwibols that
represent ideas and concepts, sosensory data must be transformed into s ~mbol s
that can be understood by the system. Quite a bit of information niay be lost in
translation, especially when visual scenes are mapped into sets of ob-jectsand
the relations between them,
Human experts can look at the big picture, examine all aspects of ;i prob-
lem. and see how they relate to the central issue. Expert systems. on the othcr
hand. tend to focus on the problem itself, ignoring issues rele\mt to. but sepii-
rate from. the basic problem. This happens because i t takes ;i huge amount of
expertise just to handle the basic problem, and i t would take almost ;is much
expertise tohandle each of the hundreds of tangential probleins that could arise.
I n the future, when faster and cheaper techniques for acquiring expert kno~vl -
edge are developed. this situation may change.
7.4.3 Applications in Power SystemOperations
The focus of expert system research in the power systems operations area has
been to help the system operator to function more effecti\~ely. Specific ob-jec-
tives include:
Chcr],te,.:7
Understanding the interface requirements to integrate databases, vari-
ouscomputer architectures, fullgraphics systems,andapplications soft-
ware expected in future control centers;
Defining the appropriate applications of expert systems (e.g.,asa rou-
tine aid to operators, providing action plans during emergencies, orthe
rebuilding of a system after a major disturbance);
Building domain specific knowledge for general purpose expert sys-
terns (e.g.,rules and techniques used by today's experienced operators
to handle the daily operation of a power system):
Developing and evaluating specific expert systems (e.g.,replacing tc-
diousjobsdone by humans);
Demonstrating expert systems that aid the operators in the most co>t
effective areas.
Control center operators benefit from using expert systems because sucn
systcrns not only identify a problem. but they can also provide the underlyins
reasoning used to define the problem, and a set of recommended actions t:)
ameliorate the problem. The best applications of expert systems have been i11
diagnostic situations where information about the system being studied is held
i n static tables. For application i n power systerns, information about the systern
can be taken from the real time database or a simulator.
7.4.4 Rule-BasedSystem
There is alot of confusion aboutterminology in the artificial intelligence litera-
ture because there are few standardizationefforts. Sointhe following paragraph
we present the definitions used in this chapter. A knowledge-based system is 'a
program i n which the domain knowledge is explicit and separate from the pro-
gram's other knowledge." I t consists of the following fi\re components: tte
knowledge base, the knowledge acquisition module, the inference engine, the
explanation module. and the user interface (Fig.7.3).
TheKnowledge BaseContains Facts and Rules
Like any database, the knowledge base contains the data, called facts, whkh
describe fixed properties of the problem. I n addition, aknowledge base contairis
instructions describing how to process these data. If these instructionsare given
i n an IF...THEN ...rule format, then we are talking about a rule-based
systein.
For the rule IF A THEN B, A is called the left-hand side ( LHS) and B the
right hand side ( RHS) of the rule. If A and Bare logical expressions which ciln
be true or false. A is called the premise and B the conclusion. But as we w I I
-
KNOWLEDGE
ACQUISITION
KNOWLEDGE
BASE
FACTS&RULES
ENGINE
t
Figure7.3 Knowledge-based system.
see later, we will alsoinclude actions like closing an element in the LHSor the
RHS of a rule. These actions will change the initial facts.They cannot be con-
sidered assimple logic expressions.
These facts and instructions represented as rules may already hakre been
formulated and written down explicitly because they are common inkvariable
knowledge of the domain. In ourcase there exist manuals describing the possi-
ble actions which can be performed on different elements and normally ekrery-
one familiar with these manuals can manipulate these elements. I n particular
these manuals describe the problem completely. Nevertheless this work i n con-
trol centers is done by operators who are certainly experts i n this domain and
who know their domain by heart.
I n other cases, there is a human expert who uses heuristic or complex
knowledge acquired during years of work. Often this kind of knowledge is in-
complete and the reasoning mechanism applied by the human expert cannot be
replaced by pure logic rules. For instance, intuition or so called common sense
cannot be formalized this way. The main reason why we call our systcm a rule-
based system and not an expert system is due to the fact that in principle our
knowledge is complete. It can easily be formulated as rules. Also we do not
want to contribute the misunderstanding that i n any domain where a human
expert explains his knowledge by I F. ..THEN ,..rules, this reasoning can be
efficiently implemented by an expert system. This misunderstanding leads to
hundreds of unsatisfying feasibility studieson expert system applications which
can be found in the literature today. In short, a rule-based system is a special
case of a knowledge-based system.
I96
Knorvledge BaseAcquisitionModule2ndI/O Intertke
<ireUserFriendly
Theknowledge base is built up using a program called aknowledge acquisition
component. Depending on the nature of facts and rules this component may b:
asophisticated ruleeditor generating ASCII or binary filesorii program extract-
ing relevant data from another database or e\wi l'rom a real-time process a n J
storing i t in the predefined format on nussstorage medium.
The knowledge acquisition module as well as the user interface should bt.
user t'riendly.Menu-driven and/ormouse-driwi dialogue, graphicoutput multi-
nrindoLving ul l help to achieve this goal. These techniques are already n' c, I
knon'11.
TheInterferenceEngineProvides Control
Whereas rules describe how to process data, the instructions of the inferencc
engine describe how to process these rules. They f i x for instance the order i n
Lvhich turo rules concerning the same element areexecuted or fired.One possi-
bility concerns the enumeration ofthe rules, where the numbers f'ix the prioril),
of the rule. Another possibility is to process the youngest mentioned rule f i r >t .
The processing protocol may be specificfortheproblem.Then i t iscalled metal-
knowledge with meta-rules. In order to keep the inference engine ;IS general ;is
possible i t is sometirnes convenient to store this tneta-knowledge in the knom1-
edge base.
The instructions alsodefine thegeneral problem solving strategy.There are
two important techniques. backward chaining and forward chaining. The go: 1-
dri\.enstrategy backward chaining tries to prove a goal, kvhile data-dt-ihrencon-
trol forward chaining deduces new data from existing facts.
TheExpl,iri,ition ModuleExplnnntion ottheInferenceChain
The explanation module provides ;I trace facility for the end user i n order to
comment on the reasoning process.
Shells2ndProgrminiingLanguages arenot ldentic-a1
An "empty" hnowledge-ba\ed systemwhich provides control anddatastructute\
but no hnowledge iscalleda shell.Any shelliswritten in acertain programming
language for in\tance Parcal, LISP,OPS83, o r PROLOG. AI1 these langua: e\
ha be their ad\antage\ anddi\advantage\. We note that OPS83provide\ thed; i t a
type "rules" and the forward chaining control strategy, whereas PROLOG I \
e\pecially \uited to implement backward chaining control, because of i t \ ba1.h-
traching facilities. Neither procedural languages like Pa\cal nor functional oiie\
lihe LISP pro\ide thew facilities that therefore habe to be programmcd explic-
itly. Although they may be more efficient for a final inipleiiientatioii they are
not suited for rapid prototyping i n order to test concepts and methodologies.
We should emphasize however that PROLOG does not fit i n our definition
of a knowledge-based system shell because neither a control strategy nor an
explanation facility are not provided but have to be programmed.
Using PROLOG factsas knowledge base factsand PROLOG predicates as
rules does not permit the explicit separation of the inference engine and knou.1-
edge base. Although it may be tempting to quickly implement a prototype i n
pure PROLOG in order to test the concept, this prototype cannot be enhanced
without changing the code of the program (except for those cases u'hereonl y
pure predicate calculus is used and no extra control is needed i n order t o en-
hance efficiency). Note however that the PROLOG trace facility does not auto-
matically provide an explanation module. Depending on the PROLOG-inter-
preter,theoutputof this facility ismore orless unreadable for nonpro~rr\mniers
(e.g.,variables are simply denoted by numbers).Furthermore, predicates used
for building up strategies have to be hidden to the trace procedure because
showing them will produce a lot of redundant information. The user-interf;ice
provided by standard CPROLOG is rather poor for our application. Similar ar-
guments concerning control. explanation, and user-friendliness are iralid f or
OPS83.
Let us have a look at the reasoning process of the human substation opera-
tor. The operator asks whether i t is possible to close an element or to s ~i ppl ~'
power to a certain load and how to do i t . He therefore pr o~~! s hqpothesis a
instead of looking at the substation configuration and collecting all possible
conclusions and actions based on the initial stateof the station. For this re;ison.
many power oriented rule base systemshave been written i n PROLOG because
this goal-driven strategy reflects the reasoning process of the end user better.
7.4.5 Reasoningwith Uncertainty inRule BasedSystems
One of the important features of expert systems is their ability to deal u-ith
incorrect or uncertain information. There will be times when an expert system,
in gathering initial inputs, will ask a question for which there is no iinsnw. I n
such acase, simply say so.Expert systems aredesigned to deal urith cases such
as this. Because one may not have a particular fact. the search process \{,ill
undoubtedly take adifferent path. i t may take longertoconicup\\rith an ans~ver.
but the expert system will.
Traditional algorithmic software simply cannot deal with incomplete infor-
mation. If one leaves out apiece of data, one may not receive an ans\ver at all.
I f the data is incorrect, the answer will be incorrect. This is u9here artificial
inte1 1 igence prograins, parti cu1ar1y expert systems. are partic111ar1y Usef uI \\.hen
the inputs are ambiguous or completely missing. the progi-ammay still fi nd a
solution to the problem. The system may qualify that solution, but at least i t is
an answer that can i n many casesbe put topractical use. This isconsistent with
expert level problem solving where one rarely has all the facts before making ii
decision. Corninon senseorknowledgeof theproblem tells us what is importmt
to know and what is less important. Experts almost always work with incoin-
plete or questionable information. but this doesnt prevent them from solving
the problem.
Thus, increasingly in the design of expert systems, there been a focus on
inethods of obtaining an approximate solution to a problem when there is no
clear conclusion from the given data. Logically, a s expert system problems he-
come inore complex,the difficulty of reaching a complete conclusion certainty
increases, s o in some cases, there must a method of handing uncertainty. Re-
searchers report that a classical expert system gave incorrect results due to ;he
sharpness of the boundaries created by the if-then rule of the system: howe\er,
once amethod fordealing with uncertainty (i n these twocases fuzzy set theory)
was used. the expert system reached the desired conclusions.
The successful performance of expert systems relies heavily on human 2x-
pert knowledge derived from domain experts based on their experience. :he
other forins of knowledge include causal knowledge and information from case
studies, databases, etc. Knowledge is typically expressed i n the form of high
level rules.Theexpert knowledge takes the form of heuristics, procedural rules.
and strategies. I t inherently contains vagueness and imprecision because expxts
are not able to explicitly express their knowledge. The process of acquiring
knowledge is also quite imprecise, because the expert is usually not aw;irv of
allthe tools used i n the reasoning process. The knowledge that onereasons \b,ith
inay itself contain uncertainty. Uncertain data and incomplete information are
other sources of uncertainty in expert systems.
Uncertainty i n rule based expert systems occurs in two forms. The first
form is lingiiistic uncertainty whichoccurs ifan antecedentcontains vague state-
ments such as the level is high or the value is near 20. The other fori-iof
uncertainty. called evidential uncertainty. occurs if the relationship betfieiian
obserkration and a conclusion is not entirely certain. This type of uncertain1y is
most corninonly handled using conditional probability which indicatesthe likeli-
hood that a particular observation leads to a specific conclusion. The study of
making decisions under either of these types of uncertainty will be referred to
iis plausible or approximate reasoning. Several inethods of dealing with uricer-
tainty i n expert syst emhave been proposed, including:
Subjectikreprobability
Certainty facrors
Fuzzy measures
Fuzzy set theory
The first three methods are generally used to hand evidential uncertainty,
while the last method, fuzzy set theory is used to incorporate linguistic uncer-
tainty.
As expert assessments of the indicators of the problem may be imprecise,
fuzzy sets may be used for determining the degree to which a rule from the
expert system applies to the data that is analyzed. When applying a method of
reasoning with uncertainty to arule based expert system.there must be a method
of combining or propagating uncertaintybetween rules.A method of propagating
uncertainty for the method of reasoning with uncertainty will be discussed next.
SubjectiveProbability andStatistics
One method of dealing with uncertainty is to use conventional statistics and
probability. For example, with the use of statistics,sufficient data may be avail-
able to compute the mean (average),median, and standard deviations. These
new figuresderived from original dataprovide additional knowledge which will
help in making adecision. Recall thatprobability issimply aratioof thenumber
of times that a particular action will occurfora given number of attempts. I t is
really a ratio asshown below:
P(.u) =Nuniber of occurrence of an event/total number of e\'ent\ that take place
The probability of ,v occurring, stated asP( x) , is the ratio of the number of
times s occurs to the total number of event that takes place. For example, i n
rolling a standard die,the probability isone-sixth that any one of numbers, one
through six,will comeup. This may alsobe expressed as afraction 0.16667,or
asapercentage, 16.67%.I n many knowledge representation cases.the probabil-
ity foracertain condition oraction may be known orcan be estimated. For the
probability of a certain even taking place is 70%,then one may initiate some
action if the probability is equal or greater than 70%. If the probability is less
than 70%.then perhapsanaction may notbe taken. Forexample,the production
rule below uses the probability:
IFthe stone isclear,without color
THEN it isdiamond (probability 60%)
An example will illustratethis. Suppose we ask ten engineers whether they
can program in the BASIClanguage.Outof the ten, three saythey can.We can
use these figures to compute the probability:
P(BAS1C)=-J 10=0.3
What this \ays is that the probability of anengineerto be able to progra11
i n BASICis0.300.Wec;in alsoexpres\ thi\-a percentageby\imply niultipl:,-
ing the probubilitj b j 100.We \ ay that the probabilitj of engineer\ being able
to prograni i n BASIC i \ 3O'k.Probability figure\ lihe thi\ can be used to deter-
mine rule strength if the) f i t the problem.
Multiple probability lralues Ui l l occuri n manj \jf\tem\. Forexample,a riile
may haLe three part\ t o i t \ antecedent,each \+ithaprobabilitjcrilue. Theo ~ e r d l
probability o f the c aluethen become\the product o f the indiiidual probabilitic.\,
i f the part\ o f the antccedent are independent o f one another. I n ;I three p;irt
antecedent,the probabilitie\ may bc 0.90.0.70,and 0.65.Theo\erall prohabil-
i t ) i \
Thecombinedprobability i \ about4I c/c .But thi\ i \ true onl y i f the indi\id-
ual part\ of'the antecedent donot affect or depend c m one another.
Sometime\ one rule reference\ another. Here the indiLidual rule probabili-
t i t ' \ can propagate from one to another. There i \ ;i need to e ~a l ua t e the tc t i l l
probabilitj o f ii ~eq11enceof rules or a path through the \earthtreetodetermine
i t ii \pecific rule fires. Or one maj be able to use the combined probabilitj to
prcclict the be\t path through the warch trce. I n other ~+o r d \ , the probabilirit.\
become the "cost\" of'the indiLidual arc\ i n the tree.
Thcre are niitiierwi\ methods o f computing combined probabilitie\. If' thc
r ~ i h are independent, ;I \iiiiple product can be uwd ;I\ dewi bedbefore. HOM-
e\cr, mo\t ekent\ iind rule\ are dependent upon one another. I n that cii\(. ;I
\peci;il procedure ciilled Baye\' RLIICor Theorem c;in compute the probabilitj
of'e\ ent A occurring gi\en that e\ ent B Iia\ alrcadq occurred.Thi\ i \ e\ pr e~\ ed
;i\ P( A111). Baj e\ ' Theorem i\:
We Mi l l not attempt to explain thi\ rule, but one \hould be ;i~+are ot the
t'ac%tthat manq expert \y\tems iise Baye\' Theorem in\tead of certaint) factot-\
to deal N i t h uncertainty. Se\ era1 iiiii-ior expert \j \tern c k \ elopment tool\ ii\e
Ba) e\ ian probabi1ity.
hlcmitresotBeliet,ind Disbeliet
Mca\ure\ o f belief' and disbelief arose from the desire that ekvidence \ hml d
incrementally increa\e the belief o r di\belief in ii h j pothe\i\. Theformaldet'ini-
tion ot the tiie;i\Lire o f belief w;i\ ba wl on the idea that if ;I prior probabilitJf,
P ( / I )i \ defined, then the maximum amount of belief that can be added 10 P ( / I )
f rom anew pieceof e\fidence is 1 -P(I1).I f a piece of e\sidence confirms R I 1 1 0 ) .
then this would amount to. adding P( hic)- P( h) to the pre\.ioii\ belief, so the
belief i n 11 has been increased by
Ap(11)=
All e ) - All)
I - p(11)
The mea\ureof increaseddi\belief can be defined \iniilarly. No\! u r i t h t h i \
idea, let the measure of increase belief ( MB ) giken sonic t'\ idence about ;I ( 1
hypothesis / I be defined as:
and let the measure of increased disbelief ( MD )be defined as
Note that when evidence P is assigned to a hypothesis I, . onl y one otthc
MII orMBfunctions will be greater than zero sothat ;I single piece 01' c.\.idencc
cannot be iised as both ameasureoftheconfirmation and negation of;I Iijpttie-
sis.As the nieasuresofbelief and disbelief were used i n the design of' ;in expert
system. i t \vas found that a representation of the uncertainty i n terms of ;I single
measU re nwU Id be n i oreconven i ent i n inakingcoinparisons of differcn t hypotli -
eses.
Certni nt~/ Fnctors
Thereare se\-eralmethodsof dealing with uncertain information. I n rule based
expert syst em. numerical factorsindicatingthetrutho r probability of a premise
or conclusion are used as a measure for uncertainty. These numerical factors
are kno\vn ascertainty factors (CF)and probability.
I n ii high percentage of expert system rules, there L i ' i l l be no ambifuit!. or
m r i l l uncertainty. We know with confidence whether or not ;I p;trticular preniisc
o r conclusion is true or false. If the information is not kno\ i . n at all. then ttic
rule requesting it will not fire. In cases where there is the possibility that t i e
information is not known, special rules canbe created todeal with this problem.
The rule might state that if a particular piece of information is not available,
then acertain action will be initiated.
Still,there are many cases where the information is known but we hakfe
less than IOO% confidence i n its truthfulness. Just iis weather forecasters LI SC
a number to predict the likelihood of rain, so can a confidence number be
used with production rules. A forecaster may say that there is a 80%probabil-
ity of rain. They are saying that they dont know for sure whether or not i t is
going to rain. On the other hand, they have enough information to be able to
say that 80%of the time under similarcircumstances i t rains. While a certaiiity
or confidence factor is not really a probability, i t is a number that helps to
represent the uncertainty. A certainty factor is simply a measure of the corifi-
dencc that ii particular fact or rule is true or not true. I t is irsually a numl)t.r
between 0 and I , where zero indicates no confidence and one means f ul l or
complete confidence. We alsohear certainty factorscalled confidence factors or
rule strength.
Certainty factors are used with both the premise (IF) and conclusion
(THEN)portions of ii rule. The two examples gikren below show hou, corrfi-
dence factors are used.
IF the patient has the fever, CF=0.6
THEN prescribe rest i n bed
IF the patient is sneezing
AND has arunny nose
AND has watery eyes
THEN the patient has acold, CF=O.S
Figure 7.4shows several ways to use certainty factors. The scale is up to
the progranimer. In Fig. 7.4A, a scale of 0 to I is used where 1 =absolute
certainty; that is. 100%truthfulness or validity of the premise or conclusior of
a rule. The 0,of course. indicates absolute uncertainty or falsity. Intermed ate
\palues have varying degrees of truthfulness or uncertainty. A scale of 0 to I 0
or 0 to 100can also be used with the same result.The +and - scale shourri i n
Fig. 7.4B is another approach. A +5 indicates absolute certainty while ;I - 5
indicates 100% contradiction. The 0 in the center of the scale indicates un-
known. One could also use a- I / O/ +I scale as well.
Determining whetheraparticular rule istofire requiresthe inferenceeng.ine
t o look at the confidence factor and evaluate i t . For example, if one is using the
0to 1 scale. we might want the rule to fire i f the confidence factor is abow ii
certain threshold level. say at 02. you may assign a threshold of + I or - I
CONFIDENCE
0 .25 .5 .75 1
FALSE UNKNOWN TRUE
(8)-5 0 +5
Figure7.4 Confidence and certainty factor scales.
depending upon the circumstances at the minimum acceptable level for deter-
mining whether something is true or false. Other levels may be set depending
upon the problem.
In rules with compound premise clauses connected by AND or OR,each
clause may have its own CF. For such situations, there must be a way to com-
pute a CFfor the rule. This is done by using the minimum CFof all clauses
connected by AND or the maximum CFof all clauses connected by OR.Some
examples will illustrate this.
RULE 1:
IF X(0.4)
AND Y( 0. 75)
THEN Z
Composite CF=0.4
RULE2:
IF D(0. 3)
AND E(0. 8)
THEN F
Composite CF=0.8
I f each rule in ;I reasoning chain has a CF. each wi l l . of course, affect rhe
other. The outcome has to be decided based upon some composite e\'aluation.
One way this is done is with a special formula.
This \ ay\ that the CFtor ruleX is added to the CFforrule Y and from that
is wbtracted the product of the CF\ for rule\ X and Y. Belou are t uurules that
illustrate this point:
RULE3:
IF P
AND Q
THEN R(O.65)
RULE 4:
IF R
THEN S(O.20)
The ccmipositeCFis then:
0.65 +-2.00 - ( 0. 65)(0.20)=0.85 - 0.I3=0.72
Of coitrse. there will usua11y be more than two rules i n ;I chain. The a h e
formula can be u\ed by taking the composite CFof two rules and combiniiig i r
Lvith the CFof ;I third rule. That ne\v compo\ite CF is then combined t h ;I
fourth, and s o on.
Fuzz\.,1ogic
F u / q \et-ba\ed techniques can provide an excellent tramworh t or \jf\teinati-
tally representing the impreci\ion inherent i n an expert'\ hno\tledge. W t . UW
the foII O W i11g example to iIIU\ trate,
IFthe teriiperature is high (0.8)
ANDsystem i \ operating in hea\!y load period (0.9)
THEN system i \ high stressed
The parameters i n the premise and consequent temperature, load pcriod.
and \y\tetii \tress can be represented using simple fuzzy inembership function\.
In t hi \ case. the f i i z ~ jinformation is contained i n the terms high arid healj .
Although, there are situations where membership grades probabilities can
take on similart'alues,theyare not the same.Onedistinguishing factor bet\b.een
probability fuzzy membership grades is that the summation probabilities on a
finite universal set must equal 1 .
The main drawback of nonfuzzy methods i n dealing with uncertainty is
their handling of linguistic terms. Fuzzy theory provides a natural frame\sa-k
for dealing with linguistic terms used by experts. Imprecision in numeric data
can be easily dealt with by expressing i t as a fuzzy number. Fuzzy sets can
be conveniently incorporated i n expert systems to better deal uith uncertainty
imprecision.
StructureoftheExpert StaDility AssessmentSysteni
The fuzzyexpert system structure isshown i n Fig.7.5.Itsdatabase contains the
power system topology.
The knowledge base of the fuzzy expert system contains all the data of
the contingencies. The information is based on known statistics of protection
performance used in the system. If these data are not a\railable\ t k n a fiiult
occurs, the fuzzy expert system asks the dispatcher to provide them and then
saves them in adatabase for future use. Models for estimation of possible con-
tingencies, and heuristic rules about the relay characteristics for actual fiiult
determination are alsoinclude.
On-lineDatafrompowersystem
I
Dispatcher Database
Networkidentifier
interface
A
Knowledge Hypothesesand
-+ Base Calculations
InferenceEngine
Determinationofstates
Figure7.5 Fuzzy expert system structure.
206
7.5 FUZZYSETS AND SYSTEMS
In 1965,L.A. Zadeh laidthe foundation of fuzzy settheory a s amethod todeal
with the imprecision of practical systems. Bellman and Zadeh write: "Much
decision making in the real world takes place in an environment in which the
goals. the constraints and the consequences of possible actions are not precisely
known." This "imprecision" orfiizziness isthecoreof fuzzy setsorfuzzy logic:.
Fuzzy sets were proposed as a generalization of conventional set theory. Pal.-
tially a s a result of this fact,fuzzy logic remained the purview of highly special-
ized and mathematical technicaljournalsfor many years.Thischanged abruptly
in the late 1980s.
7.5.1 Fuzzy Sets
In ;Iconventional (nonfuzzy,hard,orcrisp)set,anelementof theuniverse eithcr
belongs to or does not belong to the set.That is, the membership of an element
is crisp-it is either yes ( i n the set)or no (not in the set).A fuzzy set is ;I
generalization of an ordinary set in that i t allows the degree of membership for
each element to range over the uni t interval [ O, I ] . Thus, the membership func-
tion of a fuzzy set maps each element of the universe of discourse to its range
of space. which, in most cases, is assumed to be the unit interval.
One major difference between crispand fuzzy sets is that crisp setsa1wa:;s
hi1t.eunique mernbership functions,whereas ebvery fuzzysethasan infinite nur-i-
ber of possible membership functions that may represent it. This enables fuz;!~;
systems to be adjusted for maximum ut i l i t y to agiven situation.
7.5.2 Fuzzy Systems, Complexity,andAmbiguity
Zadeh's principle of incompatibility was given in 1973to explain why there is
a need for ;I fwzy systems theory. The principle states. in essence, that as t i e
complexity of ;I system increases,ourability to make precise and yet significant
statements about its behavior diminishes until ;I threshold is reached beyorid
mrhich precision and significance (orrelevance) become almost mutually exclu-
sive characteristics. This suggests that complexity and ambiguity (imprecision
are correlated: "The closer one looks at a real-world problem, the fuzzier t' e-
conies its solution."
I t is characteristic of the way a human thinks to treat problems involvi,ig
complexity and ambiguity i n a subjective manner. Complexity generally steins
from uncertainty in the form of ambiguity: these are features of most social.
technical, and economic situations experienced on adaily basis. In considering
aconiplex system,humans reason approxirnately about itsbehatrior(acapabilrty
that computers do not have)and thus maintain only ;I generic understanding o f
theproblem.Thisgenerality andambiguity areadequateforahuman toperceive
and understand complex systems.
As one learns more and more about a system, its complexity decreases and
understanding increases. As complexity decreases, the precision afforded by the
computational methods becomes more useful in modeling the system. For less
complex systems, thus involving little uncertainty. closed-form mathematical
expressions offer precision descriptions of the systems behavior. For systems
that are slightly more complex but for which significant data exist, model-free
methods,such ascomputational neural networks,pro\.idepowerful andeffectiLre
means to reduce some uncertainty through learning based on patterns i n the
available data.
Basic statistical analysis is founded on probability theory or stationary ran-
dom processes, whereas most experimental results contain both random (typi-
cally noise)and nonrandom processes. Oneclassof random processesorstation-
ary processes exhibits the following three characteristics:
1 . The sample space on which the processes are defined cannot change
from one experiment to another, that is, the outcome space cannot
change.
2. The frequency of occurrence, or probability, of an event uithin that
samplespace isconstant andcannotchange from trial totrial orexperi-
ment toexperiment.
3. The outcomes must be repeatable from experiment to experiment. The
outcome of one trial does not influence the outcome of a pre\ious or
future trial.
However, fuzzy sets are not governed by these characteristics.
Thefundamental differencebetween fuzzinessand probability is that fuzzi-
ness dealswith deterministic plausibility, while probability concerns the likeli-
hood of nondeterministic, stochastic, events. Fuzziness is one aspect of uncer-
tainty. It is the ambiguity (vagueness) found in the definition of a concept or
the meaning of a term such as confortable tenipemtirrc or \ i d 1 cwokod. Hob-
ever,the uncertainty of probability generally relates to the occurrence of phe-
nomena, as symbolized by the concept of randomness. In other word5, ;1state-
ment is probabilistic if it expresses some kind of likelihood or degree of
certaintyorif i t istheoutcomeofclearly definedbutrandomly occurringevents.
For example, the statements There is a50-50chance that he will be there. It
will be sunny tomorrow, and Roll the dice and get a six demonstrate the
uncertainty of randomness.
Hence, fuzziness and randomness differ in nature: that is,they are different
aspects of uncertainty. The former conveys subjective human thinking. feel-
ings, or language and the latter indicates an objective statistic i n the natural
sciences.
From the modeling point of view. fuzzy riiodels and statistical modelsal;o
possess philosophically different kinds of information: Fuzzy membershipsrep-
resent similaritiesof objects otimprecisely defined properties, while probabili-
ties conveyrelative frequencies. Thequest fora method of quantify nonrand(~ri1
uncertainty (imprecision,vagueness, fuzziness)i nphysical processes istheha:,ic
premise of fuzzy system theory, for to understand uncertainty i n ;I system is to
understrind the system itself. As understanding impro\ses,the fidelity i n mod~:l-
ing iiiiproLW.
Whene\ferprecision iselrident,forexample, fuzzy systemsarelessefficient
than iiiorc precise algorithins in offering ;I better understanding of the problem.
Req11i r ing precision in cngineering mode1s and prod11 cts trans1ates to rey11iri11g
high cost and long lead times i n production and dc\~elopment.For other tl- ;in
siinple systeins. expense is proportional to precision: More precision entiiils
higher cost. When considering the use of fuzzy logic for ;I giLm problc111.
;in engineer or scientist should ponder the need for exploiting the tolerance for
imprecision. Not only does high precision dictate high costs but i t also ent;iils
l ow tractability i n ;I problem.
On the other hand. t ' u u y systemscan focus on modeling problems chariic-
terized by iinprecise o r ;inibiguoiis information. The following iire situatir is
\\,here i t is appropriate to formulate system problems cvithin ;I fuzzy sj' slcm
frallle\4fork:
I . In processes involbing hunian interaction (e.g.. hunian descripti\.c or
int11i t ive thi nkirig)
2. When a n expert is iivailable w'ho can specify the rules underljring the
sj'steiii beha\.iorand the fuzzy sets that represent the characteristic, o f
each \ # ariable
3. When ii mathematical model of the process doesnot exist,or exists but
istoodifficult toencode. or is toocomplex t o be e\~aluatedfast eiic.ugh
forreal-tinie operation,oi.in\wlvestoomuch niemory oil the design.itecl
chiparchitecture
4. In processes concerned w'ith continuous phenomena(e.g..oneo r i-iore
of thecontrol \w-iablcsarecontin~ious) that arenot easily broken d o u ~ i
i nt o discret e segments
5. Whcn high arribient noise leifelsmust be dealt t+.ithor i t is i mpot i nt
to iise iiicxpensi\~csenwrsand/orlou~-precision microcontrollers.
The ability to use f u ~ z q 'systcni tools \ +. i l l allow one to addresh the \rast
iiiii.jori tj , of' probIenis t hat ha\c the preceding charact eristics.Fuzzy formuI;I .ions
ciiii help to achic\,etractabilitj,,robustness, anti l o w w solution costs.
Any field c;in be fu//,itied ancl hence gencralized by replacing the coiicept
o f ;i crispset i n the target field bj.theconcept ot';i l'iizzj' set. Therefore. WOc;iii
t ' i i u i f soiiicbasic fieldssuchasgraph theor),.arithiiictic.andprobabilitj,t l icory
to develop fuzzy graph theory, fuzzy arithmetic, and f u n y probability theor).
respectiLrely,moreo\'er,we can also fuzzify some applied field\ such a\ neural
networks. pattern recognition, and mathematical programiiiing to obtain f u ~ q
neuraI neturorks,fuzzy pattern recognition,and fuzzy matlieniatica1 p rograni-
ming,respectiveIy. The advantages of f u zzifcation incILIde greater genera1ity.
higher expressive power, anenhanced ability to model real-world problems,and
a methodology for exploiting the tolerance for imprecision.
7.5.3 FuzzyTheory inElectricPowerSystems
With the remarkable and successful penetration of fiizqr sj'stenis into manuflic-
turing, appliances, and computer products, their applications in power systenis
are beginning to mature and receive wider acceptance i n the electric po\+w
community. The application of fuzzy set theory to po~j ' ersystems is a rela[i\~el~,
new area of research.
Concepts of fuzzy set theory were first introduced i n solLingpo\sw sj'steni
long-range decision-making problems i n the late 1970s. Houe\ter. substantial
interest in its applications to power areas is fairly recent. While con\~entional
analytical solution methodsexist for many problems i n po\s.ersystem operation,
planning. and control. their formation of real-world problems suffers from re-
strictive assuinptions. Even U'ith these assuniptions. so1ving 1arge-scaIe po\\.e r
system probleins is not trivial. Moreover. many uncertainties exist i n a signit'i-
cant number of problems because power systems are large. coniplex. \J-idelJr
spread geographically. and influenced by unexpected events.These factorsmake
i t difficult todeal effectively with many power systemproblems through strictl),
conventional approaches alone.Therefore, areas of computational (artificial)i n-
telligence emerged in recent years in power systems to complement convcn-
tional mathematical approaches and proved to be effecti\.e when properly cou-
pled together.
I n conceptualizing power system problems, the expert's empirical kno\ \ . l -
edge is generally expressed by language containing ambiguous ort ' uzqdescrip-
tions. As ii result, classical Boolean logic may not be a \ di d [ool to represent
suchexpertise. Fuzzy logic,on theotherhand, isanaturalchoice forthispu~-pose.
Thegrowing number of publications on applications of fuzzy-set-based ap-
proaches to power systems indicates its potential role i n sol\ingpo\\.ersystem
problems. Results obtained thus far are promising, but fuzzy set theory is not
widely accepted. The reasons for its lack of acceptance include the follouins:
Misunderstanding of the concept
Exce\si\e claims of some researchers
Lach of implemented and available systems
I t \ {tatus;I\ ii n e ~ theory
Unexpected events and their uncertainties are traditionally represented by
probability. However, i t hasrecently been made clearthat someof the uncertain
facts are intrinsically of a fuzzy nature and are difficult to manage properly
using probabilistic approaches.
There are problems in power systems that contain conflicting objectives.
I n power system operation, economy and security, maximum load supply, and
minimum generating cost are contlicting objectives. The combination of thcse
objectives by weighting coefficients is the traditional approach to such prob-
lems. Fuzzy set theory offers a better compromise and obtains solutions tnat
cannot be easily found by weighting methods. The benefits of fuzzy set theory
over traditional methods are as follows:
Provides alternatives for the many attributes of objectiLVes selected
Resolves contlicting objectives by designing weights appropriate to a sc-
lected objective
Prokiies the capability for handling ambiguity expressed in diagnostic pro-
cesses, which involves symptoms and causes
Power system components have physical and operational limits that are
usually described as hard inequality constraints in mathematical formulaticms.
Enforcing minor violationsof someconstraints (practicallyacceptable)increases
the computational burden and decreases the efficiency and may even prelent
finding a feasible solution. In practice,certain slight violations of the inequality
constraints arepermissible.Thismeans thatthere isnot aclearconstraintbound-
ary and the constraints can be made soft. Traditionally. this problem has b x n
managed by modifying either the objective function or the underlying iteraiicte
process.The fiizzysetapproach inherently incorporatessoftconstraints and thus
simp1ifies im p1enientation of such considerations.
The following steps are recommended when fuzzy set theory is used to
solve power system problems:
Decription of original problem: The problem to be solved should firs1 be
\tated mathematically and linguistically.
Defining thresholds for variables: For a given variable, there is a specific
\alue with the greatest degree of \atisfaction evaluated from empirical
hnohledge. and a certain deviation is acceptable with decreasing de-
gree of sati\faction until there is a value that is completely unaccqt-
able. The two \alues corresponding to the greatest and lea\t degrc: of
satist i ction are termed threshold\.
Fuzzy quantizittion: Based on the threshold values already determined,
proper foriiis of member\hip functions are constructed. The funct on\
4hould reflect the change in degree of satidaction with the change I I I
\ariables e\duatedby experts.
Selectionof the fuzzy operations: In terms of the practical decision-making
process by human experts, a proper fuzzy operation is selected so that
the results obtained are like those obtained by experts. The interpreta-
tion of results using fuzzy systems isbased on domainexperts' reason-
ing.Therefore, at this level a hybrid fuzzy set-expert system scheme is
desirable. I t helps to remove any ambiguity that may occur i n problem
solving.
7.5.4 Membership Functions:Definitions
Let X be a set of objects, called the universe, whose element are denoted A-.
Membership i n a subset A of X is the membership function, MA from X tothe
real interval ( O, I ] . The universe is all the possible elements of concern in the
particular context.A is called thefuzzy setand isasubset of X that has no sharp
boundary. p,,!is the grade of membership .Y in A. The closer the LAue of p.\is
to 1 , the more .r belongs to A. The total allowable uni\rerseof values is called
the doinniri of the fuzzy set. The domain is a set of real numbers. increasing
monotonically from left to right where the values can be both positive and
negative.A is completely characterized by the set of pairs
Support of a fuzzy set A in the universal set X is the crisp set that contains all
the elements of X that have a nonzero membership grade i nA. That is
SuppA =( xE XIp,,(s)>0) (7.15)
Fuzzy sets with a finite support assume that .Y, is an element of the support of
fuzzy setA and that p1is its grade of membership inA. Then A is written as
(7.16)
When X is an interval of real numbers a fuzzy setA is expressed as
(7.17)
An empty fuzzy set has an empty support which implies that the member-
ship function assigns 0to all elements of the universal set.
A technical concept closely related to the support set is the alpha-level set
or the "a-c-irt." An alpha-level is a threshold restriction on the domain of the
' I ' C'llt1ptc.I' 7
fuzzy set based on the membership grade of each domain value. This set,A . is
the a-cut of A which contains all the domain \faluesthat are part of the f ~ i ; ~ y
set at a minimum membership value of a . There are two kinds of a-cuts: n ~ a k
and strong. The weak a-cut isdefined asA,, =( X . p,,(.r(2U }and the strong a-
cut as A,, =(X,p,.,(.Y( >a}.Also, the alpha-level set describes a poi +wor
strength function that is used by fuzzy models to decide whether or not ;i t r ut h
~ ~ a l u e to zero. This is an important facilit).that should be considered equi~~alent
controls theexecution o f fiizzy rules;is well as the intersection o f niultiple fiiI LJ '
sets.
The degree of membership is known iis the membership or truth function
since i t establishes ;I one-to-one correspondence between a11element in the do-
main and ii truth value indicating its degree o f membership i n the set. I t takes
the forrn,
The triangular membership function is the most freqiiently iised funciion
and the most practical. but other shapes are also used. One is the trapei,oid
ivhich contains more information than the triangle. A fuzzy set can d s o be
represented bq' a quadratic equation (invol\ring squares. I I -. or numbers to the
second power)to product. ;I continuous curve. Three additional shapes ncliich
are named for their appearance ;ire: the S-function, the PI-function, and tht: I_-
functio11.
7.5.5 Set Operations
Unionmd Intersection of F L I Z Z ~Sets
The cluxsical union ( U ) and intersection (n)of ordinary subsets of X iire t's-
tended by the following formulas for intersection. A nB and union. A U H:
w,herept CI N and p are respectively the rnembership functions of AUB md
AnB.
For each element .v i n the ~iniversal set, the function in Eq. (7.19)t a h ;is
itsiirgiimentfrom the pairconsisting of theelement's menibershipgrades i 1 set
A and i n setB and yields the membership gradeof theelement i n the setconsti-
tuting the union of A and B. The disjunction or union of two sets means that
any element belonging to either of the sets is included i n the partnership which
expresses the maximum \ due for the two fuzzy sets in\wlved.
The argument to the function in Eq. (7.20)returns the membership grade
of theelement i n the setconsistingof the intersection ofA andB. A conjunction
or intersection makes use of only those aspects of set A and set B that appear
in both setswhich expresses theminimum value forthetwofuzzy sets in\.ol\-ed.
7.5.6 Complementofa FuzzySet
The complement of A, - A, which is the part of the domain not i n a set, can
also be characterized by Not-A.This is produced by inverting the truth function
along each point of the fuzzy set and is defined by the membership function
v.vE x.p \(.v) =I - p, (.\)
( 7. 71)
Thecomplement registers thedegreetowhich an element iscomplementary
to the underlying fuzzy set concept. That is, how compatible is an elements
value I.\-] with theassertion,x isNOTJ, where.\-is an element from the domain
and .v is a fuzzy region. A fuzzy complement is actually a metric. I t measures
the distance between two points in the fuzzy regions at the same domain.The
linear displacement between the complementary regions of the fuzzy regions
determines the degree to which one set is a counter example of the other set.
We can also view this as a measure of the fuzziness or information entropji n
the set.
DefiningFuzzy Sets
The steps below gi\re general guidelines i n defining fuzzy sets.
1. De)te)i-iiiiiw the typeofj k y r?iecisicreiiieiit. Fuzzy setscan define:
Orthogonal mappings between domain values and their member\hip
i n the set (ordinaryfuzzy set);
Differential surfaces which represent the first derivative of some ac-
tion, degree of change between model states, or the force of con-
trol that must be applied to bring a system back toequilibrium:
A proportional metric which reflects a degree of proportional com-
patibility between a control state and a solution state;
A proportionality set which reflects a degree of proportionality be-
tween a control state and a solution state.
2. Choosc~theslzcipe (orsu$cice i wi - p i i o l o g ~~) of tlw.fii:qq set. Theshape
maps the underlying domain back to the set membership through a
correspondence between the data and the underlying concepts. Some
possible shapes are triangular, trapezoidal, PI-curve, bell-shaped. S-
curves, and linear. Every base fuzzy set must be normal.
3. Selwt ciii cippropricite degree of m-erlap. The series of indi\idual
Chtrprrt- 7
fuzzy sets, associated with the same solution variable, are converted
into one continuous and smooth surface by overlapping each fuzzy set
with its neighboring set.Thedegree of overlap depends on the concept
modeled and theintrinsicdegreeof imprecision associated with thetwo
neighboring states.
-1. Eii.siirt~thcit theclorritiiris mi o r i g thefi rxj sutscissoc-ieitetl btvitli tl i cs t i r r i p
. soli r tiori \Uricih1e.sJherre tli iJscitiie irt ii\vr-sr of'cl i s coi rI-se.
7.6 EXPERTREASONI NGA N D APPROXIMATEREASONI NG
7.6.1 Fuzzy Measures
Thefuzzy measure assigns a value to eachcrispset of the universal set signifl-
ing the degree of evidence or belief that a particular element belongs in the set.
For example,we might want to diagnose an i l l patient by determining whethttr
this patient belongs to the set of people with, pneumonia. bronchitis, emph1.-
sema, or a common cold.A physical examination may provide us with helpfril
yet inconclusiveevidence. Therefore, we might assign ahigh value,0.75. toour
best guess, bronchitis, and a lower value to theotherpossibilities, such as0.45
to pneumonia. 0.30to a common cold, and 0 to emphysema.These values rc-
flect the degree to which the patient's symptoms provide evidence for one dis-
ease rather than another, and the collection of these values constitutes a fuz;ty
measure representing the uncertainty or ambiguity associated with several well-
defined aIternatives.
A fuzzy measure is a function
where pCP( x) is a family of subsets of X such that:
I . $ E p a n d X ~p.
2. If A E pthen E p.
3. I f p is closed under the operation of set union, that is if A E p arid
B E p.then alsoA UB E p.
The set p is called a Bore1 field, since A UB 2A and A UB 2B. We ha1e
max[g(A)., q( B) ]5g(AUB) due to the required monotonicity. Similarly, since
A nB C ,4andA f l B CB. we have (?(AflB ) 5min[g(A),g(B)].
Twolargeclassesof f uz ymeasuresarereferred toasbelief andplausibilitj
nieasures which are complementary (ordual)in the sense that one of them cqin
be uniquely derived from the other.Given a basic assignment 1 1 1 . a belief mea-
sure and ii plausibility measure are uniquely determined by the formulas
215
which areapplicable forallA E p(x).Alsom( A) refers tothedegreeofevidence
or belief that a specific element of X belongs to the set A alone. The beliyf
i?ze~i.sure, Bel(A),represents the totalevidence orbelief that theelement belongs
to the set A as well as to the various special subsets of A. The p l a i t s i h i l i f ~
nieasur-e,Pl ( A) , represents not only the total evidence or belief that the element
in question belongs to the setA orto any of its subsets but alsothe additional
evidence orbelief associated with sets that overlap with A. There are also three
important special typesof plausibility andbelief measures,prohcihility i ?~~w. wi - e. s
and a pair of complementary measures referred to as possihility cri i cf riecx).ssir!,
measures.
7.6.2 ApproximateReasoning
The root mechanism in a fuzzy model is the proposition. These are statements
of relationshipsbetween mode variables andoneormorefuzzy regions.A series
of conditional and unconditional fuzzy associations or propositions isevaluated
for its degree of truth and all those that have some truth contribute to the final
output state of the solution variable set. The functional tie between the degree
of truth in related fuzzy regions is called the method of implication. The func-
tional tie between fuzzy regions and the expected value of a set point is called
the method of defuzzification. Taken together these constitute the backbone of
approximate reasoning. Hence an approximate reasoning system combines the
attributesof conditional and unconditional fuzzy propositions,correlation meth-
ods,implication (truthtransfer) techniques, proposition aggregation, and defuz-
zification.
Unlike conventional expert systems where statements areexecuted serially,
the principal reasoning protocol behind fuzzy logic is a parallel paradigm. I n
conventional knowledge-based systems pruning algorithms and heuristics are
applied to reduce the number of rules examined, but in a fuzzy system all the
rules are fired.
7.6.3 The Roleof LinguisticVariables
Fuzzy models manipulate linguistic variables. A linguistic variable is the repre-
sentation of a fuzzy space which is essentially a fuzzy set derived from the
evaluation of the linguistic variable. A linguistic variable encapsulates theprop-
ertiesof approximate orimpreciseconcepts in asystematicandcomputationally
useful way.
The organization of a linguistic variable is:
Clltr/t~l- 7
L, ,/ t ((1,.......q,,}( I l l .......I f , } f ; (7.2: I
where predicate q represents usuality or frequency yualifiers, I i represents ;I
hedge andj : is the core fuzzy set. The presence of qualifier(s)and hedge(s)i s
optional. Hedges change the shape of fuzzy sets i n predictable ways and func:-
tion in the saine fashion as adverbs and adjectiires i n the English languitgc.
Frequency and usuality qualifiers reduce thederi\red fuzzy set by restricting t t e
truth niembership function to ;I rangeconsistent uith the intentional meaning of
the qualifier. Although 11 linguistic variable may consist of many separate ierm\.
i t is consictered ii single entity i n the fuzzy proposition.
7.6.4 Fuzzy Propositions
A f u z q modelconsistsofii seriesof conditional and unconditional t uzzy prop^
sitions. A proposition or statement establishes a relationship betweena \-alueI I I
the underlJring domainand ;I fiizzy space.Propo.sitiori is onethat is qualified
an i t statement. The proposition following the if term if. is the antecedent l)r
predicate md is an arbitrary fuzzy proposition. The proposition follom~ing tlie
t l i or i term is the consequent and is alsoany arbitriiry fuzzy proposition
interpreted ii\ t i \ a member of Y to the degree that \ I is ii member of 2.An
unconditional f u z q proposition is one that i \ not yualified by an if \tatement.
X i s k
c\.hercX is ii sciilar frwii the domain and Y is ;I linguistic variable.
Llncondi tional statements are alwaysapp Iied within the inode1 and depend-
ing on how they ;ire applied. serveeitherto restrict theoutput space ortodefiiie
;t clefnult solution space.We interpret an unconditional fuz7y proposition ;is .Y
is the minimum subset o f Y m.hen the output fuzzy set X i \ empty. then X is
restrictcd to 1. othercvisc. for the clotiiain o f 1; X becomes the nziri(X, k). The
solutioii fuzzy spaceis updated by taking the intersection of the solution set anc l
the target fuzzy set.
I f ii modelcontains ;i mixture of conditional or unconditional propositioiis,
then the orderof execution becomes important. Unconditional propositions are
gent.rally used to establish the defnult support set for a model. I f none o f the
conditional rules executes, than ii b-aluefor the solution krariable is deterrnincd
f rom the space bounded by the cinconditionals. For this reason they must he
executed bcfore anyo f the conditionals. Theeffect o f ei.:iluating ii fuzz>. pr ol ~) -
sition is ;I clegree or gradeo f iiieiiibershipderic,eclfrom the trmsfer functioti
where .r is a scalar from the domain and Y is ;I linguistic \variable. This is
the essence of an approximate statement. The derived truth membership \Value
establishes acompatibility between .r and the generated fuzzy space.This truth
value is used i n the correlation and implication transfer functions to create or
update fuzzy solution space.The final solution fuzz),space is created by aggre-
gating the collection of correlated fuzzy proposition.
7.6.5 Fuzzy Implication
The r r z or i ot or i i c method is a basic fuzzy implication technique for linking the
truth of t wogeneral fuzzy regions.When two fuzzy regions arerelated through
a simple proportional implication function.
if.\- is Y then :is 1.1'
functionally represented by the transfer function
then under a restricted set of circumstances, a fuzzy reasoning sq'stemcan de-
\relopan expected \valuewithout going through composition and decomposition.
The value of the output is estimated directly from a corresponding t r ut h mem-
bership grade i n the antecedent fuzzy regions. While the antecedent fuzzy ex-
pression might be complex,the solution is not produced by any formal method
of defiizzification. but by a direct slicing of the consequent, fuzzy set at the
antecedent's truth level. Monotonic reasoning acts asa proportional correlating
function between twogeneral fuzzy regions.The important restriction on mono-
tonic reasoning isits requirement that the outputforthe model be a singlefuzzy
variable controlled by a singlefuzzy rule (withan arbitrary complex predicate1.
The multiplication space generated by the general composition r - i r l c. . ~ of' i r z -
fi~r-rric-e is derived from the aggregated and correlated fuzzy spaces produced b),
the interaction of many statements.In effect allthe propositionsarerun i n paral-
lel tocreate an output space that contains information from all the propositions.
Each conditional proposition whose evaluated predicate truth is abo\re the cur-
rent a-cut threshold contributes to the shape of the output solution Lrariable's
fuzzy representation. There are two principal methods of inference i n fiizzy
systems: the min-max method and the fuzzy additive method. These methods
differ in the way they update the solution variable's output fuzzy representation.
For theriziti-1w.virzjkr-rrzcrmethod theconsequent fuzzy region is restricted
to the minimum of the predicate truth. The output fuzzy region is updated by
2IN Chtrprer 7
taking the maximum of these minimized fuzzy sets. The consequent fuzzy sct
is modified before i t is used to set each truth function element to the minimum
of eitherthe truth function ortruth of the propositions predicate. The solution
fuzzy set is updated by taking, for each truth function value, the maximum c ~ f
eitherthe truth valueof thesolution fuzzysetorthe fuzzysetthat wascorrelated
to produce the minimum of consequent. These steps result in reducing the
strengthof the fuzzy setoutput toequal the maximum truthof the predicate and
then, using this modified fuzzy region, applying i t to the output by using the
OR (union)operator. When all the propositions have been evaluated, the output
contains a fuzzy set that reflects the contribution from each proposition.
The fuzzy additive compositional inference method updates the soluticn
variables fuzzy region in a slightly different manner. The consequent fuzzy
region is still reduced by the maximum truth value of the predicate, but the
output fuzzy region is updated by a different rule, the bounded-sum operatioi.
Instead of taking the ma~( p, ~) [ . r , ] , p,,[y,]) at each point along the output fuzi.4.
set, the truth membership functions are added. The addition is bounded by [ I ,
01 so that the result of any addition cannot exceed the maximurn truth value of
a fuzzy set. The use of the fuzzy additive implication method can provide a
better representation of the problem state than systems that rely solely on the
min-max inference scheme.
7.6.6 CorrelationMethods
The process of correlating the consequent with the truth of the predicate s t em
from the observation that the truth of the fuzzy action cannot be any greater
than the truth of the propositions premise. There are two principal methods ot
restricting the height of the consequent fuzzy set: correlation minimum and
correlation product. The most common method of correlating the consequent
with the premise truth truncates the consequent fuzzy region at the truth of the
premise. This iscalledcorrelation minimum, sincethefuzzy set isminimized 3y
truncating i t at the maximum of the predicates truth. Thecw-rlelcrtionn i i n i v i i ! u i
mechanism usually creates a plateau since the top of the fuzzy region is sliced
by the predicate truth value. This introduces a certain amount of informatim
loss.I f the truncated fuzzy set is multi-modal orotherwise irregular,the surfrice
topology above the predicate truth level is discarded. The correlation method,
however, is often preferred over the correlation product (whichdoes preserc-e
the shape of the fuzzy region) since i t intuitively reduces the truth of theconx-
quent by the maximum truth of the predicate, involves less complex and faster
arithmetic, and generates an aggregated output surfacethat iseasier todefuzzify
using the conventional techniques of composite moments (centroid)or compos-
ite maximum (centerof maximum height).
While correlation minimum is the most frequently used technique, the tor-
relation product offers an alternative and, i n many ways, better method of
achieving the correlation. With correlatioti product, the intermediate fuzzy re-
gion is scaled instead of truncated. The truth membership function is scaled
using the truth of the predicate. Thishastheeffect of shrinking the fuzzy region
while sill retaining the original shape of the fuzzy set. The correlation product
mechanism doesnot introduce plateaus into the output fuzzy region. although i t
does increase the irregularity of the fuzzy region and could affect the results
obtained fromcomposite moments orcomposite maximum defuzzification.This
lack of explicit truncation has the consequence of generally reducing informa-
tion loss. If the intermediate fuzzy set is multimodal, irregular,orbifurcated i n
other ways this surface topology will be retained when the final fuzzy region is
aggregated with the output variables under generation fuzzy set.
7.6.7 Aggregation
The evaluation of themodel proposition ishandled through an aggregation pro-
cessthat produces the final fuzzyregions foreach solution variable.This region
is then decomposed using one of the defuzzification methods.
MethodsofDefuzzification
Using the general rule of fuzzy inference, the evaluation of a proposition pro-
duces one fuzzy set associated with each model solution variable. Defuzzifica-
tion ordecomposition involves finding avalue that best represents the informa-
tion contained in the fuzzy set. Thedefuzzification process yields the expected
value of the variable foraparticular execution of a fuzzy model. In fuzzy mod-
els, there are several methods of defuzzification that describe the ways we can
derive an expected value forthe final fuzzy state space.
Defuzzification meansdropping aplumb line tosomepoint on the under-
lyingdomain. At the point wherethis linecrosses thedomain axis,theexpected
value of the fuzzy set isread. Underlying all the defuzzification functions isthe
process of finding the best place along the surface of the fuzzy set to dropthis
line. This generally means that defuzzification algorithms are a compromise
with a tradeoff between the need to find a single point result and the loss of
information such asa process entails.
The two most frequently used defuzzification methods are composite mo-
ments (centroid)and composite maximum. The crtitrwicl or center or graLrity
technique finds thebalance point of the solution fuzzy region by calculating the
weighted mean of the fuzzy region. Arithmetically, for fuzzy solution region A,
this is formulated as
( 7. 25)
where tl is the /'"domain value and u(t1)is the truth membership value for that
domain point. A centroid or composite moment defuzzification finds a point
representing the fuzzy set's center of grak4y.A i i i t i . vi i i i i i i i i c~~c. oni pos i t i t , r i find.;
the domain point with the maximum truth.There are three closely related kind\
of composite maximum techniques: the average maximurn, the center of maxi-
niurii, and the simple composite maximum. I f this point is ambiguous (thati >.
i t liesalongaplateau),then these methodsemploy aconflictresolution approacI
such as averaging the values orfinding the center of the plateau.
Also there are other techniques for decomposing a fuzzy set into an ex-
pected value. The ti\'ei-ti,qe of'iiiii. u'iiiiiiii i ~ i l i r c )defuzzit'ication method finds the
mean niaxiniuni value of the fuzzy region. I f this is a single point, then this
\Aue is returned; otherwise, the value of the plateau is calculated and returned.
7'110 t i \ vi -i i gt l of'the ~iori:ororegion is the same as taking the average of the
support set for the output fuzzy region. The, f i i i - t i i i t l iitwr e~l ~qe of't he siippo 't
w t technique selects the value at the right fuzzy set edge and is of most w e
when the output fuzzy region is structured as a single-edge plateau. Tlir c wi t t ' r
( ? ~ ' i ~ i t i . v ; i i i i i i i ~ . ~ technique, i n a inultirnodal or multiplateau fuzzy region, finds ttt'
highest plateau and then the next highest plateau. The midpoint betnmeen ttlc
centers of these plateaus is selected.
CONCLUSION
I n this chapter, we offer the reader ii glimpse of the main set of tools from the
emerging area of intelligent systems. While there are excellent books and oth:r
forms of literature on the subject treated, the attempt of the coverage is to oftrr
a brief summary of the ingredients of artificial neural networks (A"). exptart
systems (ES). and fuzzy logic (FL)systems.
The coverage is intended to provide the necessary background for the fol-
lowing t u u chapters. Once again, the reader is inirited to consult the list ot'
references and the annotated glossary of terms in the back of'the textbook. ;I+ it
source of further information.
Applicationof Artificial Intelligence
toAngleStabilityStudies
INTRODUCTION
The computational requirements associated with dynamic security analysis
(DSA)by conventional methods aretwoorthree ordersof magnitude more than
the requirements for static security analysis. Therefore comprehensi\~e on-line
DSA is infeasible in present power system control centers. The exploitation of
novel techniquestosolvethe problem of DSA isessential foron-lineimplemen-
tation. The interest i n application of artificial intelligence ( AI ) to dynamic secu-
rity analysis has been increasing steadily. The following advantages are ad-
dressed through using these methods:
I . A decomposition of the on-line DSA problem into manageable sub-
prob1eins.
2. The need for an integrated environment for expert systems, neural net-
works and conventional programs working together.
3. Identification ofthe subproblemforwhich theneural networks aremost
suitable.
4. Feature variables of the power system and its dynamic security attri-
butes to be used as the training inputs tothe neural network.
5. Procedure to select appropriate contingency and check security ofsys-
tem using expert system.
22I
8.1 ANNAPPLICATION INTRANSIENT
STAB1 LlTY ASSESSMENT
The comprehensive on-line DSA is infeasible i n present power system control
centers. The exploitation of novel techniques to solve the problem of DSA is
essential for on-line implementation. Neural network ( NN) is one of the state-
of-the-art methods used to perform dynamic security analysis in power systeins.
The interest in NN methods has been increasing i n \carious fields because they
can work well where rule-based expert systemsand conventional algorithms .ire
inadequate. For example, quantitative rules based on subjective inputs can be
better implemented as ;I neural network classifier which can be trained on pre-
cisely defined training examples.
8.1.1 RobustANN-Based TransitEnergy Function
Earlier in this book, the formulation of thedynamic stabilitjrproblem for mul t i -
power syst emwas presented. As previously stated,the process by which opr n-
tors and planners monitor the behavior of power systems, screen the imious
random and scheduled contingencies that may occur, and recommend corrective
control action(s ) to maintain or regain system stability, is known as dynamic
security assessment (DSA).A hybridized tool, that incorporates the beneficial
aspects of a structured preserving energy function arid arobust artificial neriral
network. has been implemented to solve the contingency screening problenI of
DSA. I t is more accurate than conventional transient energy function ( T 5F)
methods. since the load buses will be modeled as nonlinear voltage dependent
components instead of being absorbed into the admittance matrix. while m;iin-
taining the robust ANN properties of both speed and tlexibility. Developncent
of the ANN-enhanced TEFtool involves theconstruction of twomodules,corre-
sponding to the trainingksting mode of one operation and the recall modt: of
the other operation respectively.
An implementation scheme for both modules of the integrated tool is pro-
vided here. Specific emphasis is placed on the training of the robust ANN, the
handling of the"over-training" issueandtheevaluationof network performaice.
DescriptionotLearning Processtor theRobustANN-Based TEF
Every neural system is subject to a learning process, through which a relation-
ship is established between a set of known inputs and 11 known result. The
learning algorithm for the robust ANN tool is shown i n Fig. 8.1.The sunis of
the weighted inputs at each node of the hidden layer of the net were compited.
The sigmoidal activation was then applied to the sum of each hidden node i n
orderto represent a better representation of theeffects of the input stimuli. The
weighted sum of the hidden nodes was then computed as the initial output of
Atiglr Smbi1it.vStirdies
0 ApplyinputstimuliX
Sumweightedinputsathiddenlayer
A = C X WI
Applysigmoidalactivationhnctionathidden
layer H =1/(I+
Calculatetheactual output
out =cH W H
ComputetheglobalerrorforNtraining
cases E =( 1 / 2 ~ ) C ( ~ a r g e t -0ut)2
Adjusttheweights andthresholdsinamanner
thatminimizes E
E <Tolerance?
Figure8.1 Learning algorithm forrobust ANN tool.
the ANN. The error for each case was computed as the sum of the squares of
the differences between the desired and the actual outputs. For each interation,
the global error was computed as the average of the sum of the errors for the
cases evaluated.
The mininiization routine, shown i n Fig. 8.2,was then used to determitie
thegradient of the unknowns (weightsandthresholds) with respect totheglobal
error. These \Aues were then used to update unknowns and the input stimuli
u'ereprescnted tothenetwork overand overagain. unt i l the tolerance l i mi t u'as
satisfied. The final values ot'the weights and thresholds were then used by t'ie
recall module of the ANN enhanced energy function (TEF)tool to map t ie
relationship between components of the energy of the system and the stabililqr/
instability of the system.
The data generated by the TEF program was then presented to the roblist
ANN, u n t i l the specified criteria \+riissatisfied. Approximately I000cases 1t.tre
ewluated. \ f r i t h close to 60%of those dedicated to training the ANN tool. The
reiis011for the \.List number of training cases was t o ensure that ii dikrerse repi-e-
sentation of the sqsteni, under \.ariousconditions. u'asprokrided to the ANN.
Again. the learning algorithm utilized aconjugate gradient search procediir-t.
t o compute the minimum value of the error function and thus optimize the
interconnection weightsand threshold values. The initial ureightsand thethrt':,h-
old values ~' e r t ' set to random \,aluesbetween -03 and +O. S.
The goal of any ANN application is to pro\side an accurate response to
ii given stimulus i n ii timely and efficient t1i;inner. Therefore, simple netnrork
architecture isutilixd :is abase case and more complex topologies \+!eree\ al-
uated i n t urn. The accuracy of the training session was defined by a user-spcci-
ficd tolerunce o f 10.'. Once the conLwgence criteria had been realized the ret-
work uas tested using data not presented during training. The results o f h e
testing phase demonstrated the ability of the ANN togeneralize.
An additional routine was included to prek'ent the network from "o\er-
training" or becoming too specific to the training set. This \+';isaccomplislied
ty t.\~aluating the global error function wi t h respect t o the test data (data not
used for training)subject to the weights and thresholds at ei'erq' t\i.cntj,-t't'th
iteration of the learning process. If the global error of the test data i ncre; i d
from one check point to the next then the training process \ v x stoppcd.
Kc ~ c a l lModule
For the recall mode of operation the TEF program and the robust ANN \\ere
integrated into a unified program to compute the security state of the g i \ w
sj'sternforii selectedcontingency. The initial prefriultconditions erecotiipt,tet!
using ii generic Newton-Raphson load tlow method, as before. The clearing
energy were calculated in the traditional niiinner to represent the condition:;of
the pouw system at f r i ul t clearing time. These values u ~ e then applied to the
Robust ANN, using the weights and thresholdsobtained i n the traininig session.
t o determine the security state of the power system. A description of the ;Igo-
rithiii for the recall riiodule of the robust ANN-based TEFtool is presentet1 in
Fig. 8. 3.
Begin minimization routine
I I
+
Input length ofvector, total #ofcases, andinitial data
&
Definemaximum # ofiterations
I
I
Definethecurrenterror
+
I
I
Computethegradientoftheerrorfunction
+
Savethecurrentvalueofthegradient
Initialize thecounter,setn=l
1
Searchalongalinedefined by thegradienttofindtheminimum value of
theerrorfunction forthecurrentunknown vector,as thegradientis
updated with respecttodisplacement
Yes
I
Globalerror<previous
Computeglobal error
Evaluatetestdatawith
currentunknown
t
Updatethegradientwithafunction
ofthetrainingmomentum
N=1
Stop;Return
Figure8.2 Minimization roiltine.
The results obtained by the training procedure were presented to the robust
ANN as training data,with the clearing energy functioning as the input stimuli
and the stability assessment computed by the TEF method functioning as the
desired output value.The inputstimuli were weighted and sunimed at thehidden
layer of the net. The sigmoidal activation function "as then applied to each
226
Begin Step I
7
Step2
Computesteadystatecondition
Step3
.
Computepre-faultsystemconditions
Step4
w . t . CO1referenceframe
Computecomponentsofenergy
Step5
atfaultclearingtime
4
Providestimuli(componentsof
Step6
clearingenergy) toRobustANN
Applyweights&thresholdsdetermined
Step7
viatrainingsession
4
Computestabilityindexofpower
step8
system
S1ep9
Figure8.3 Recall Module of the robust ANN-based TEFtool.
hidden node, and the resulting values at each hidden node were weighted ard
summed as the stability index of the system. A block diagram of the recall
module 3of the robust ANN-based TEFtool is shown in Fig. 8.3.
Steps 1-4 of Fig. 8.3follow the procedure for computation of the clearirg
energy. I n step5 the six components of the clearing energy were applied to the
input layer of the robust ANN as shown i n Fig. 8.4. The inputs were then
multiplied by the interconnection weights computed during and summed at the
hidden layer.The sigmoidal activation function was then applied to summation
at each hidden node, subject to the threshold value determined by the training
session. The final values of each hidden node were then weighted and summt:d
to compute the stability index of the power system for the given contingency
Angl e Sttrhi1it.v Stiidic)., 227
Figure8.4 Robust ANN-based TEFnetwork architecture.
The number of iterations required fortraining andthe percentage of correct/
incorrect classifications were thecontributing factorsi nthedetermination of the
optimal network architecture and significant input features for the solution of
the robust ANN-based TEF.
8.1.2 Results
The ANN-enhanced TEFprogram defined in Sec. 8.2.2was tested on the IEEE
39-bus I 0-generator power system. As previously stated, simulation by the
ANN-enhanced TEF program involved two stages of work. In the first stage
simulation studies were conducted on the TEFprogram to collect data defining
the effects of close to 1000 different contingencies on the IEEE 39-bus test
system.Arobust learningalgorithm wasthen used tosynthesize therelationship
between the faulted systems,caused by the various contingencies. and thepost-
fault system. by optimizing the interconnection weights and thresholds i n the
network. In the second stage the robust ANN was integrated into the TEFtool
to alleviate thecomputational burden caused by the deterinination of the critical
energy.
In this section the ability of a robust ANN-based TEF tool to sol\re the
DSA problem forelectric power systems isdemonstrated. The results of studies
on the IEEE 39-bus test system using robust ANN-based TEFtool, "ere Lrali-
dated via comparison to results from an industrial grade TEFprogram and con-
ventional time domain simulation studies.Thecomparison of results is preceded
by a description of the IEEE 39-bus system.
The analysis of the results presented here clearly demon\trates the abilit)
of the ANN-enhanced scheme to significantly decrease the time necessary f o.
dynamic security assessment.
Input DataandSystem Diagrcmtor theIEEE 39-BusPower System
In order to verify the accuracy of the TEF tool, the results of case studies 011
the IEEE 39-bus system using the robust ANN-based TEF were compared to
results generated by anindustrial gradeprogram onthe samesystem.The pouret-
system generation and load data for the IEEE 39-bus network are presented i n
Tables8. I and 8.2.Aschematicdiagramofthe39-buspower system isprokrideci
i n Fig. 8.5.
Con7pDarisonotRoOustANN-Based TEF to Classicnl TEF
The robu\t ANN-based TEF procedure &as used to \ o l ~ ethe contingerick
w-eening problem for DSA. Validation of the security a\w\mient determineJ
bythe robwt ANN-bawdTEF&as achieved through compariwn to the rewlt\
generated u\ing the indu\try grade TEF3.0program, deleloped by Ontario H j -
Table8.1 39-BusS p t m Load Data
I 3 322.00 2.40
2 4 5( ) O.oo I84.00
3 7 233.80 84.00
4 8 522.00 176.00
5 12 8.50 88.00
6 IS 320.00 153.00
7 16 329.40 32.30
8 I8 I 58.00 30. 00
9 20 680.oo IO3.00
I0 21 274.00 1 15.00
I I 23 247.50 84.60
12 24 308.60 -92.20
13 ' 5 '44.00 47.20
I4 26 139.00 17.00
1s 27 281.oo 75.50
16 '8 206.00 27.60
17 29 283.30 26.90
I8 31 9.20 4.60
I9 39 1 lO4.00 250.oo
2
Table8.2
Gen
no.
I
2
3
1
5
6
7
8
9
I0
39-BusSystem Generator Data
Bus p,I IVI
no. ( MW) (Volt)
39 IOOO.00 I .0300
31 N/A* 0.9820
32 650.00 0.9831
33 632.oo 0.9972
34 508.oo I .o123
35 650.00 1.0493
36 560.00 1.0635
37 540.00 I .0278
38 830.00 1.0205
30 250.00 1.0475
I 1 t
26 28 29
J
38
s7
18
I17 a4
8 I
I
I 8 La
1 4
I 21
56
6
12
10
+a 10
Figure8.5 Diagram o f 39-buspower system.
drounder EPRI project RP2206- 1, and classical time domain simulation studies
using the trapezoida1 nu meri ca1 integrati on techniqlie. A COm parative ana1y:;is
of the results of simulation tests on the IEEE 39-bus system with respect to
accuracy is presented here.
A three-phase fault occurring at generator 30, was used as a basis for
cornparsion. The fault clearing time and assessment for each case are listed
below in Table 8.3.
Evaluation of the results in Table 8.3clearly demonstrate that the security
assessments proc,idedby the robust ANN-based TEFtool aregenerally i n good
agreement with the security assessiiients provided by the results of con\mtiotial
energy function and time domain simulation methods.
TestingotRobustANN-BasedTEF onIEEE 39-BusSystem
In order togenerate enough trainingksting data for the robust ANNthe effe:ts
of okw 1000contingencies were tested on the IEEE 39-bus system and ecdu-
ated using theTEFmethod.Close to60%of these caseswere randomly seleced
for training with the remaining 40+% reserved for testing. The contingencies
represented three-phase faults ;it the terminal of each generator i n the system.
with f nul t clearing times Igarying between .000I seconds and 0. 5seconds. Each
input filecontained theenergycomponents at clearingtime and thedesired fiial
output for each case.
In complex pattern classification problems, such as this one,the success 01'
the training session is highly dependent on the initial weights and thresholds
and the topology of the network. For all of the studies conducted the weights
and thresholds were initialized at random values between -03 and +0.5.Since
there is no general rule for the determination of the optimal number of node:' in
the hidden layer many topologies were evaluated.
Table8.3 Comparisonof Results for 39 Fault at Bus 30
CritIcaI Robust Robu\t
clearing TEF3. 0 TEF3. 0 ANN- b a d ANN-ba\ed
tinie, t, AW a\\e\\ment TEFAW TEF;i\\e\\iiient
0.1 5.9047 Stable 0.102538 Stable
0.I2 5.3756 Stable 0.102539 Stable
0.I4 4.7045 Stable 0.102540 Stable
0.16 3.8742 Stable 0.102542 Stable
0.18 2.8703 Stable 0. I02543 Stable
0.20 1.683I Stable 0.102543 Stable
0. 22 0.1702 Stable 0.I02544 Stable
0.24 - I .2373 U11stable 0.867565 U11 stable
Table8.4 Description of Test Cases
No. of
variables i n
Case input vector Description of test case
A 6 Inputvector contains all six variables
B 5 Kineticenergy componentexcluded from the input vector
C 5 Component of potential energy w.r.t. position excluded from
the input vector
D 5 Component of potential energy consumed in the real pober
load excluded from the input vector
E 5 Component of potential energy conwned in the reactive
power loadexcluded from the input vector
F 5 Component of potential energy consumed in the generator re-
actance excluded from the input i'ector
G 5 Component of potential energy consumed i n the tranmission
lineexcluded from the input vector
SelectionofRelevant FeaturesofRobustANN-Based TEF
In all, twenty-nine different network architectures were trained and tested to
determine the optimal topology for the ANN-enhanced TEF tool. I n order to
determine which of the six components of the energy of the faulted system
featured significantly in the assessment of system security,each of the variables
was alternately deleted from the traininghesting set. The studies are defined as
cases A-G forsimplicity (seeTable 8.4).Thetraining performance (numberof
presentations of input data)and accuracy (testing error percentage) were used
to evaluate the studies.
A decrease in the number of presentations fortraining coupled with similar
or improved classification of the test data would give a good indication that ii
certain component did not feature prominently in the learning process, and t hus
in the assessment of security.
The traininghesting results (Tables 8.5-8.1 1 ) clearly showed that the net-
Table8.5 Case A
Architecture No. of presentations Testing error c/ c
6-2-1 504 0.35
6-3-I 18096 5.86
6-4-1 252 19.14
6-5- 1 5014 4.95
Table8.6 CW B
5-7- I 387 0.68
5-3- I 600()O 0.68
5-4- 1 2464I 3.83
5- 5- 1 40669 -0
~ ' o r k s with tLi'oneurons i n the hidden layer generallyconverged with the fe\vcst
number of presentations. In niost cases. testing accuracy uras significantlq, ; i f-
tectecl bj, the exclusion o f any one of the input \winbles. Ilourecw. u,hen the
kinetic energy component was excluded from the training set (Case B [Ta1)le
8.61)there it'iis ii clear inipro\wiient i n the speed of the network. while testing
iiccur;icy u';is maintained at close to 100%.Thus,the optimal network architcc-
ture l i ~ the IEEE 39-bus system \\';is the configuration \irith the kinetic eneIg4'
coniponerit deleted from the training set and onl y tn' o neurons i n the hidden
Iaq'er .
Therefore. i t ~vi i sestablished that the kinetic energy component "as not ;I
niiijor contributor to the ;issessnient of the dynamic stabilitj of the 39-bus S: ~ ' S-
tom. Lkrhen using the robust ANN-based TEFtool. E\Auation of the results of
the clearing energy module on a 3-bus 2-generator system. as well as the - W-
bus I O-generator IEEE test system provided sonie interesting obser\ations. I n
the 2-generator system the kinetic energy term "as one o f three components
that dominated the SPEFat f i i ul t clearing time. For ;I 3-phasefault, lasting for
ii duration of 0.2 seconds. on the first machine of the system the magnitude of'
the kinetic etiergy term comprised close to 2XCd o f the magnitude of the total
energy. I n comparison, for ;I 3-phase fault lasting 0.2 seconds on the largest
machine i n the 39-bus system the magnitude o f the kinetic energy componcnt
of the SF'EF at fault clearing time onl y constituted0.002% of the magnitude of
the total clearingenergyterm (Table8.12).Thisinay be partially duetothe Iact
that the kinetic energy component is computed as ii function of the ratio o f the
Table8.7 Cilse C
,Arc-hitecturc No. o f prcwntations Testing error '4
5- 2- I 754 17.79
5- 3- I 3960 19.14
5-4- I 1359 9.91
5- 5- I 136'15 -0
Table8.8 Case D
Architecture No. of presentations Testing ei-ror(1:
5-2-I 846 16.44
5-3-I 36247 30.05
5-4- 1 3I20 6.53
5- 5- I 573 6.98
Table8.9 Case E
Architecture No. of presentations Testing error (,l
5-2-1 I069 16.22
5-3- 1 I678 5.63
5-3-I I0373 13.5I
5-5- 1 756 19.82
Table8.10 CW F
Architecture No. of presentations Tedngerror C/ r
5-2- 1 60000 3.15
5-3-1 60000 3.38
5-4- 1 A 0 0 0 0 5.86
5-5- I 58810 -0
Table8.11 Case G
Architecture No. of presentations Testing error r/c
5-2-1 1304 23.77
5-3-I 60000 20.05
5-4- I 410 12.16
5-5- I 3I33 0
Table8.12 hliignitude ofComponents ofTEFat Fault Clearing Time
System
W, W'J
3-bu\ 8.72E-2 0.2001 9.78E-6 2.73E-2 1.26E-3 1.79E-4
2-generator
39-bu\ 1.90E-5 3. SO E-5 9.54E-2 2.38E-? 0.9338 6.35E-2
10-generator
moment of inertia of the critical generator, MA, to the moment of inertia of ihe
entire \y\tem,M,.Therefore,forsystemscontainingagreaternumber of genera-
tor\,the hinetic energy term will decrease in magnitude.
Thedominance of each component in theTEFterm had asignificantef f xt
on the \election of the feature variables for an assessment of system \tablit>.
Therefore. a niore exten\i\e \tudy is needed, with detailed machine models L nd
a ~ar i et y of te\t systenis, before ageneral statement can be made on the featlire
\ariables for DSA of electric power \ystenis by energy function technique\.
8.1.3 ANN-Based CriticalClearingTimeAssessment
Critical clearing time (CCT)- t, is defined as maximal fault duration for which
the system remains transiently stable. Mathematically, CCT is acomplex func-
tion of prefault system conditions (operating point, topology, system pararne-
ters),fault structure (typeand location)and postfault conditions that themselves
are dependent on the protective relaying policy employed. It would be highly
desirable to define this relation analytically but the diversity of variables i n-
~ d v e dmakes this task extremely complicated.
I n practice, CCT can be obtained in one of two ways: ( 1 ) by trial .ind
error analysis of system postdisturbance equations or(2)by integrating faull-on
equations and checking the \!dueof the Lyapunov or energy function unt I i t
reaches it previously determined critical level V, , . I n either approach, a numeri-
cal integration process is involved. From the pointof viewof on-line impleri-en-
rations of CCTassessment, this presents a major difficulty.
Application of ANNs is a promising alternative. Thecapabilities of neural-
nets enable them to readily synthesize the complex mappings that transform the
niinierous input attributes or features of a power system into the single valued
space of CCTs.
We examine the generalization capabilities of neural nets, fixusing on their
ability todeal with a large range of operatingconditions andchanges i n netvwrk
topology. For CCT assessment, each synchronous generator is representec. by
three features derived from the measurable parameters of the po\+,er system.
The first feature,AI , ,istherotor angle relative tothe inertialcenterof angle
(COA)measured at the instant of fault initiation.
The second feature, A?,, represents the acceleration parameter of the syn-
chronousgeneratorduringthefault.Becausethemechanical output i ntheperiod
immediately following the disturbance remains unchanged. the machine fault
dynamics may be described by:
where P,, is the output of the i'hgenerator during the fault, when i t rernains
almost constant.
The third feature, A.,,.is proportional to the kinetic energy each generator
stores during the fault.
(8.3)
I t is important to observe that all three attributes can be e\duatedwithout
solvingany differential equations.They areexpressed i n terms of the operating
parameters of the system prior tothe fault and include the structure of the fault.
Selected features are related to the fault on dynamics, however, and thus can
provide a good representation of system behavior during the first transient
swing.
Theexample four-machine power system shown in Fig. 8.6will be used to
illustrate an ANN approach to CCTassessment. Three-phase short-circuit faults
are simulated at line 2-3 near bus 3. The fault-clearing policy is to restore the
prefault system topology. Twodifferent system topologies are used to generate
the training set patterns: topology 1 with all seven lines i n seririce: topology 2
with the line 3-4removed. Fifteen different loading conditions are selected for
each topology, all with the loading level of the system in the range 0.6-2.0
relative tothe nominal operating point.Thepower factorsof the loads are main-
tained at their nominal values.
The training set consists of 30 twelve-dimensional patterns (4generators
times 3features)labeled with corresponding CTvalues. The latterare obtained
by numerically integrating the posdisturbance system equations.
The neural network used for this study is shown in Fig. 8.7.I t consists of
an input layer with twelve units and a single hidden layer with six units. After
training, the mapping learned by the net is represented by the weights and
2.M
3
7
Figure8.6 One-linediagram o f the four-machine pouer sy\tem.
Input
Pattei
LJ
Figure8.7 Multi-lajwxl neural-net bawd a\wswient of critical clearing time.
thresholds; the least squares error was reduced to 2x IO-' after SO10 presenta-
tions of the training set. The other parameters of the learning process \%.ere the
learning rate r\ =0. 3and the momentum 5=0.4.
The performance of the neural network i n new test cases. i.e.,cases not
seenduringthetraining session,isillustrated inTable8.13.Examples 1-7 corre-
spond to power system topology 2, a topology used i n the learning session.
Examples 8-20correspond to a new system topology obtained by rertmring
lines 2-5.
The critical clearing times estimated by the neural netn-orkagree with the
analytical values. Provided that afew different topologies appear i n the training
session, the network is able to operate generally, handling new topologies as
well as new operating conditions.This performance impro\.eso\'ercon\mtional
nonadaptive pattern recognition in which every new topology requires ii com-
pletely new set of discrimination rules.
Table8.13 Compari\on of CCT5ObtainedMith Nunierical
IntegrationofDifferential Equation\ (ActualCCT)and
E\timated Uith a Neural Net (E\timated CCT)
Load 1 ~ 3 ~ 1 Actual CCT Estimated CCT
Example /"( sec sec
1 0.65 0.59 0.59
3
- 0.85 0.49 0.49
3 0.9s 0.46 0.45
4 1.15 0.39 0.39
5 I .45 0.33 0.33
6 1.65 0.29 0.29
7 1.95 0.25 0.25
8 0.80 0.53 0.53
9 0.90 0.49 0.19
10 1.oo 0.46 0.46
I I 1.10 0.43 0.43
12 1.20 0.40 0.40
13 1.30 0.38 0.38
14 1.40 0.36 0.36
1s 1. so 0.34 0.34
16 I .60 0.32 0.32
17 I .70 0.3I 0.31
18 1.80 0.29 0.29
19 I .90 0.28 0.25
20 2.00 0.23 0.22
8.2 A KNOWLEDGE-BASED SYSTEM FOR DIRECT
STABILlTY ANALYSIS
Identifying the safe operating regimes requires analysis of a large number of
combinations of outages and the associated derivation and n-ianipulationof t h?
massive data and related sensitivity information. I t is an extremely difficult a d
time consuming process when conventional computation methods are used.
Such methods are implemented in the form of traditional computer programs
uith no direct provision for exploiting the operator knowledge of the system
and without utilizing proper reasoning techniques and inferring logics (expel-t
systems)to process such operator knowledge. The use of expert systems i n
TEF method applications would therefore enable the security assessment prct-
grains to perform the required cognitive tasks.
The expert system methodology appears to be uniquely suited to take acl-
vantage of the following important attributesof theTEFmethod: ( I ) The abilitjr
to generate, organize, store. and retrieve alarge amount of information required
for ponver system transient analysis to be used for planning and operating ii
stability-limited power network, and ( 2) to use this information for intellige~it
decision-making toassessthedynamic security statusof agiven power network.
8.2.1 ExpertSystem Development inDSA
In order to dec~l op a TEF-expert sytem, the following information is needed.
( a)
General formation and application concepts
( b)
TEFknowledge base structure and configuration
( c) Application rules
I n dynamic security assessment. the knowledge base and applicaticm
rules should include thecontingency data and operating security infor-
mation. Therefore,the rules can be constructed as follows:
( 1 )
Rules derived based on theoretical concepts
( 2 )
Rules and data constructed from detailed off-line computation
( 3 )
Rules established from operator knowledge and experience
( 4)
Rules and data collected from previous on-line computed see-
narios
( 5 )
Rules established from analyst knowledge and experience in the
analysisof multiple TEFcasesandcomparisons withcornparal.le
numerical integration cases.
8.2.2 Roleof the ExpertinDSA
At present. most of the murk in DSA isperformed off-line. It requires an exten-
\ive amount of interaction froni the expert who use\ a time-domain sirnulatil) n
program to identify the response of the power system to anumber of postulated
disturbances. The role of the expert involves:
(a)
Significant amount of human interaction
(b)
Use of human intuition, judgement and expertise and cognitive skills
repetitively
(c)
Routine actions by the engineer or system operator
SelecttheBase Case
In the operations planning environment, the base system configuration is ob-
tained from real-time operating conditions or by manual adjustments to a full
system base case. The need for the study is identified by the system operator
whenever the system configuration reaches a state for which operating limits
are not prespecified by previous studies. The operator may also initiate a study
before approving scheduled maintenance on major equipment.
Of allthepossible study cases,theengineer selects only afew asbase cases
worthy of detail study. The base case selected represents a credible worst case
scenario during the period of interest.
SelectingtheRelevantContingencies
For a base case obtained i n the previous section, the engineer identifies appro-
priate credible outages which are potentially dangerous to the system based on
past experience. and system configuration deviations from the full system and
base case power flow solution. In general, the contingencies most commonly
studied involve the following:
Lossof a single generator
Three-phase-to-ground fault cleared by the primary protection by opening
the faulted line
Single-phase-to-ground faultclearedbybackup protection by delayedopen-
ing resulting in multiple equipment outages
Three-phase-to-ground fault cleared by the primary protection by opening
multiple transmission lines sharing the same transmission towers.
I t is impossible toenumerate all possible outages. Therefore, only the most
severe credible outages of important components of the system are considered.
The dynamically important components in the system are identified by the ex-
plicit or implicit use of rules.
OperatorActions
After examining the results foranumberof selected case studiesandcontingen-
cies the engineer determines the operating constraints for the system such that
they will be vaild f or ;i number of operating conditions not explicitly studied.
The system operators are required to operate the system within the sccuriij
constraints.
In the cases o f minor security Lriolations for which timely postcontingenc.
correctiLreactions area\xilable,the operator may choose to continuethe oper.i-
tion of the system in its current condition.In the cases ofother security t.io1.i-
tions, the operator may choose to dothe following:
Reschedule generation
Reschedule interchanges
Arm or disarm remedial actions such as generation rejection. load shedding.
a11d s j 'st eiii separatioi i .
8.2.3 Rule Base Structure
Bawdon the acquired hnowledge, i t i \ found that xolution oi the dynamic WCLI -
r i t j a\w\wietit problem f i t \ naturally with ;I rule-bawd \y\tem combined M i t h
procedure\. Therefor-e,the rule-based technique is applied to demonstrate t ie
tea\ibilitj o f ;in expert \ J \tern approxh. I n [heproce\\ o f Ano~ledgeba\e ce-
Lelopiiient.the attempt u;i\ made to iii;iirit;iiri generalit4 o f rule+Uhile i i i i r i i i i i 1-
tng the \)\tern \peciftc feature\.
hfodulesotKul t ) B ~ s eStructure
~ ~ There iire f i modules in the de\eloped rule base. Each mociule contain<ii
number o f rule\ performing ii yxcifictask i n the dynamicwcurity a\se\\iiicnt.
For con\enience,each module o f the rule haw \ +i l l be r e f e r i ~ ~ I to a s a "tash. '
FirstTask: MonitorSystemConditions
AtialyLe s>.steiiioperation conditions for g i \ w system status. Check i t I he
changes i n sj'steiiiconditionssince the last xsessiiierit are sigiit'icantenough t o
uwr a nt ii new ;isscssiiieiit. It' there ;ire significant changes then DSA is necdccl.
Thedecisionprocesses i i i \ wl \ d here ha\.eto relj,completely on the prccontin-
gency steady-state condition of the pow'er sj'steiii and its relationships \+(it11pre-
\,ioitslyriiialq~zed base case(s).
SecondTask:ContingencyAnalysisandSelection
Select contingencies rc'leLrant to the current system state. Rules i n this task i n-
clucleesiiiiiiiiirig the iiiipoi-tmceo f each comporient to the dynamicsecuritj o f
the sj'stemm c l selecting the outage o f the inore important components for ttir-
ther studj,.Check \\.hether[ he current sq'steiii slate is secui.e i c , i t h respect [ ( ) ;i
: i \ r : i givc'11 coiitingeticy.This t;isli iii ay use i1;ib1e iiif or mati oil regarding the b.ise
c;ise.gi\,en contiii geticy.and the transie11t condition of'the system im mediii te1j
ti , I Io\+'i11g l'au1t-cIeariiig.
ThirdTask:AcceptabilityJudgement
Using results of security and invulnerability assessment, rules i n this task can
make a judgement on the overall acceptability of the operating conditions. An-
other function of rules in this task isto peform adirectjudgement of acceptabil-
ity based on qualitatikre assessment of the sysem operating point (generation.
load, and transmission), the external condition (weather,etc.),and disturbance
(outages.etc.),and reactions (linetripping, etc.).Rules forthedirect acceptabil-
ityjudgement i n the first task provide a sufficient condition for acceptability.
FourthTask:SecurityandInvulnerabilityAssessment
If the direct acceptability assessment task does not give a positive mslver. then
the next task will be initiated. This task contains rules. which conipare security
margins to the specified thresholds. The security margins are obtained bjfcom-
paring the current operating value of a critical parameter such as line tlow to
the maximal f l ow \,a1ue iit which instabi1i ty occLIrs.In\ ' U1nerabiIi t y ;is ses s tncnt
is also performed i n this task. At present, invulnerability refers to it "sensiti\'c"
operating condition that i nj dves a severe drop i n security margin i n ternis of'
the transient energy function (TEF)if a perturbation of the operating condition
occLI rs.
FifthTask: RemedialActions
This is a collection of rules, which can be used to suggest change i n thc operat-
ing conditions i n order to regain system security. I f the security margin drops
toanegative value(orbelow a specific threshold),then controlsmust bechosen
to eliminate the abnormality. The selection of control is based on sensiti\,it>f of
the control with respect to the security margin and the amount of a\,ailable
controls. The controls necessary to fully correct the insufficiency of security
margin are suggested.
I n addition to the tasks mentioned above, i t is necessary to modit'). the
securit y knowledge base through adding significant new information about the
contingency and the initial operating conditions to the existing security :irchi\re
in the knowledge base. Modification occurs only if the present analysisrequires
ii change i n the security region of the system as implied by the knoutledge
already a\sailable i n the knowledge base. The structure of DSA F i i t h rule base
is presented i n Fig. 8.8.
Task 1: SystemStateMonitor
This task consists of rules I .. I . I , and 1.2. I t senres as a fast monitor of'the
s ~~s t em base state. This task can only give a "yes" annvcr or ;t "not-sure" ;in-
swer. If the answer is "not-sure," further analysis is needed, such ;is iinaIj,sisof
Tasks 2 and 3.
c
Nosignificant
change SystemStateMonitor
(Systemconditionchange?)
-
No,analyze
r
nextoutage
ContingencyAnalysisandSelection(Is
-
importantoutage?)
t
Security&Invulnerability
RemedialActions
Assessment
Modifysecurity
v v *
knowledgebase
1
Figure 8.8 DSA structureuith rule base.
1 . IF the system Sis in the base case.
THEN S certainly satisfies the dynamic security requirement.
Data: System base case data.
OutpU t :S-d ynamic security.
I . 1 IFsystemcondition hasonlyasmallchangearound thebasecme.
THEN system can maintain dynamic security.
Data: System base case data with smallchange.
Output: S-dy namic security.
1.2 IF system condition has significant change.
THEN systemmay not satisfy thedynamic securitycriterion. and
further analysis is needed.
Data:System base case datawith big change.
Output: System may have the problem of dynamic security.
Task 2:ContingencyAnalysisandSelection
This task also consists of rules I . , 1 . 1 , and 1.2. This task is in\.ol\Ved if the
systemcondition analysisin Task 1 givesanot-sure answerunder the signifi-
cant change of system condition, especially under the outage cases. It offers a
fast analysis of the system contingency scenarios. Only the important contin-
gency scenarios are selected forfurther dynamicsecurity analysis.
I . IFthe outage scenario OCis not important.
THEN OCwill not be selected.
Data:System data nd outage parameters.
Output:OC-neglected.
1 . I IFthecontingency performance index (PI)is lessthan the thresh-
old (PI,) for a given outagescenarioOC.
THEN OC will not be selected.
Data: system data,outage parameters, PI. and PIT.
Output:OC-neglected.
1.2 IF the contingency performance index (PI) is greater than the
threshold (PI,) foragiven outage scenarioOC.
THEN OCwill be selected for further dynamic security analysis.
Data: system data,outage parameters. PI, and PI7.
Output:OC-selected.
Task 3: Acceptabilityjudgement
This task can be considered as having 2 subtasks.One subtask consists of the
rule numbered 2 and its subordinates (i.e., rules 2.1, 2.1.1 and 2.1.2, 2.2 and
2.2.1).I t servesasafasteranalyzer of thesystem conditions (e.g.,weather.load
trend, etc.),and tries to draw a conclusion about acceptability. This task can
only give a yes answer or a not-sure answer. If the answer is not-sure,
the other subtask (representedby rules 1., I . 1 , and I . 1 . 1 ) will be imroked.This
taskgivesa yeshoanswertotheacceptable orunacceptablequestion depend-
ing on the output of Task 4.
I . IFe ~ w y operatingcondition i nCsatisfiesthesecurityand nvi i 1nerabi1-
ity constraintsand policies,
THEN C is acceptable.
Data: security and vulnerability euluation results (satisf.:d- or-not 1,
all cases in C. policy descriptions
Output:C-acceptability
I . I IFSF(C) =yesandCF-SF(C) is greaterthan athreshold forevery
operatingcondition i n C.
THEN the system satisfies the security and in\~ulnerability c(In-
striiints.
Data: SF(C),CF-SF(C)forall cases i n C. threshold.
Output:sec11rity,in\ . i i 1nerabi1i t y constraints satisfiedor not.
1.2 IF every operating condition in C is secure and in\ulneral)le
THEN SF(C) =yes.
Data: "seciire" o r "\rulnerability" status of all cases in C.
Output: SF(C).
2. IF (Modeo f Operation M is NORMAL,and Disturbance Dis mediiim
or large).o r ( Mis High-Risk and Disturbance is Small).
THEN C is acceptable.
Data: Mode of Operation,Disturbance.
Output:CacceptabiIity.
2.1 IF System S is SUSCEPTIBLE and External condition E is
THREATENING THEN the Modeof Operation Mis High-Risk.
Othenvise,M is NORMAL.
Data: S,E.
Output: M.
2.1.1 IF LoaD (LD) is STEADY and Transmi\sion ( TR) i \
NORMAL,or LoaD is PICK UP and GeneratioN ( GN 14
LOWandTR i \ Normal,or LoaD is STEADYandGeiier-
atioN (GN)is LOW and TR i \ WEAK and Generation
Inertia (GI)i \ LARGE.
THEN the System S i \ ROBUST.
0ther wise,it is SUSCEPTIBLE.
Data: LD. TR,GN,GI.
output:s.
2.1.2 IF WEather (WE)is OFCONCERN,
o r WEather is OF NO CONCERN and INterconneciion
( I N) ha\ INSUFFICIENT SUPPLY.
THEN External Condition E i \ THREATENING.
Data: WE. IN.
Output: E.
2.2 IF the UPSET is MINOR and the REACTION is ADEQUATE.
THEN the Disturbance is Small.
Date: UPSET, REACTION.
Output: D.
IFthe UPSET is SEVEREand the REACTION is ADEQUATE.
THEN the Disturbance D is Medium.
Data: UPSET, REACTION.
Output: D.
IFtheUPSETisMINOR and theREACTION isINADEQUATE.
THEN the Disturbance D is Medium.
Data: UPSET, REACTION.
Output: D.
IF the UPSET is SEVERE and the REACTION is INADE-
QUATE.
THEN the Disturbance D is Large.
Date: UPSET. REACTION.
2.2.I IFtheCorrectiveAction isLINETRIPPING andtheTIM-
ING isnot FAILED, orthe Correctii-eAction isGENER-
ATION REJECTION and TIMING isON TIME.
THEN the Reaction R is ADEQUATE.
Otherwise, i t is INADEQUATE.
Data: Corrective Action,Timing.
Output: R.
Task 4:Securityand VulnerabilityAssessment
Rulesfor Security Assessment
This task is invoked if the system condition analysis i n Task 3 gi \ m :I "not-
sure" answer. As the name implies, this task consists of two parts. I n one the
security is assessed based on the use of energy margin (orsecurity margin).I n
the other part. the sensitivity of the energy margin with respect to parameter
bw-iationsis considered, i.e.,how vulnerablethe systemistoparameter changes.
1. IF a family of regimes in a class have not crossed zero at a parameter
\alue I. and the total probability of this family is higher than or equal
to p.
THEN the system is secure with probability p for this class of contin-
gencies as long as the parameter isbounded by the value I .
Data: probabilities of the given class of contingencies, I, p. f r t mi l y of
regimes.
Output: secure w Prob p.
1 . 1 IF an operating regime crosses the parameter (horizontal)asi\ at
246
a value IMAX.
THENthe system operating point issecureaslong asthe parame-
ter is less than IMAX.
Data: IMAX,zero-crossing of the given regime.
Output:Operating point secure or not.
2. IF the normalized energy margin (NEM)is greater than a threshold
(zero).or the Strength SG is HIGH and Stressis LOW or MEDIUM,
or the Strength SG is Medium and Stress is LOW.
THEN the operatingcondition C is secure.
Otherwise,it is insecure.
Data: NEM. Threshold,SG,Stress.
Output:C secure or not.
3. IF(SecurityMargin >Threshold).
THEN (Systemis Secure).
Otherwise,Systemis Insecure.
Data: Security Margin, Threshold.
Output:Secure(YESINO).
3.1 IF (Decisionon updating Security Margin is YES).
THEN (CallProcedure Update Security Margin).
(seeTask 5 for procedure).
Data: Decision on updating Security Margin, Mismatches be-
tween base case and current generationdinterchanges, sensitivily
of each generationhnterchangeto Security Margin. base case Sc-
ciirity Margin
Output:(current)Security Margin.
3. I . 1 IF(currentgenerations areNOTequal to base casegener-
ations) AND (current tie interchanges are NOT equal .o
base case interchanges).
THEN(Update=YES).
Data: current (real-time) generation/interchange profile.
base casegeneration/interchange profile.
Output: Decision on updating Security Margin ( YES/NCl).
Rulesfor VulnerabilityAssessment
1 . IFthe degradation of energy margin (EEM)is larger than a threshold
for any of the specified variations v, or the IMPRESSIONABILITY
(IM) is HIGH, or the IM is LOW and UNSETTLENESS (US)is
LARGE.
THEN the operatingcondition C is vulnerable. Otherwise,i t is i nl r ul -
nerable.
Angle Stcihilitj*Stiidirs 247
Data: DEM, Threshold, Variations, IM, US.
Output:C-vul nerable-or-not .
I . I IFSensitivityofEnergy Margin (SEM)ishigher thanathreshold.
THEN IMPRESSIONABILITY (IM)is high.
Otherwise, it isLOW.
Data: SEM,Threshold.
Output: IM.
IFDelta Parameter Change (DPA)is larger than a threshold.
THEN the UNSETTLENESS is LARGE.
Otherwise, i t is SHALL.
Data: DPA,Threshold.
output: us.
Task5:Remedia/-Actions
ThisTask is invoked only if the security assessment of Task 4gives a negative
answer. The objective here is to adjust the available controllers in such a way
that the system becomes secure.
1 . IF (Corrected-Security-Margin <Threshold)AND (thereare available
control actions).
THEN (Select next control action in priority list) AND (Determine
amount of threshold violation).
Data:Corrected-Security-Margin, Threshold, available controlsordered
in priority.
Output: Highest priority control c, Amount of Security Margin Viola-
tion.
1.1 IF (control action c is available) AND (control action c has the
highest priority).
THEN(Determine thenecessary amountofcforcorrection)AND
(Calculate the Corrected-Security Margin).
Data: availability of control c,the current control c selected from
rule 2.
Output: amount of control c,needed, corrected-security-margin.
I .2 IF(control action c is atis limit).
THEN (set status of c to UNavailable) AND (removec from the
priority list).
Data: control c and its limit.
Output: updated availability statusof control c
1.3 IF(control action c isthe current remedial control)AND (neces-
sary amount of control c is not at its limit).
THEN (nomore remedial action is needed)
Data: control c. necessary amountof c.
Out put:stopchoosirig remediaI iictions.
2. IF(SecurityMargin<Threshold)AND(thereare NOa\railable control
;icti ons.
THEN (noremedial action is possible).
8.2.4 Summary
I n this sectionfundamentalconceptsandbasic requirementsfor;idynamicsecu-
rity ;issessiiieiitexpert system halte been introduced tor a stability-limitedpo\ser
net\\x)rk. The computerizedprocedure uses power system data pro\Vidcd bq, t ic
TEFmethod ;is input data to the expert system.
For ii gi cm contingencythe TEFmethod prolrides energy margins and cii-
crgy margin sensitiLrities to operating parameters. The energy margins pro\.i:Ic
;I measure o f the system security. The energy margin sensiti\,ities proifidt.;I
~iie:isureof system \fulnerabilityt o changes i n operating paranieters. This data
is suitable for pro\-idingthe logic needed by an expert system for dynaniic
security assessment.
A fraiiie\sw-k for identifying and organizing the knowledge base recltii Y-
tiients for the \.artoilstasks incul\vd in dynarnic security ;issessiiieiit has hccn
presented. A multilevel tree structure fortheTEF-expertsqrstem is proposecl 'or
the analq'siisassociated u i t h decisions concerning the pouwsystem's securitj/
\xilnerability status. I t is important to note that the procedure outlincd ahc1i.c
pi-ocrides;i suitable franienwk i n Lvhich to implcliicnt ii much needed dqwt-iic
contingency classification and selection algorithm.
8.3 FUZZY-LOGIC-BASED POWER SYSTEM STAB1 LlZATlON
De\relopment o f ii fuuy-logic-based power system stabilizer to maintain stal~il-
i t y and enhanceclosed-loopperfi)rriianceof ii powersystem is described in this
chapter. Simulation studies on ;i single-machine inf'inite-bus system and oi ;I
111tiIti-machilie pouvrs)'stern model show itcry sat isfactory performance.
The fii~zq'-logic-basedPSS (FLPSS)h:is been implemented on ;I lo\i--co\t
micro controller and tested i n the laboratory on a physical model ot';I single-
11i;ic hitie ini t'i 11 i t e bussi r stem.Experiment~ i ltests and resuIts ;ire a1sodescribcd.
8.3.1 FLCStructure
In concentional control. the mo u n t of control i \ determined i n relntion t o ;I
number 01' data inputs using a \et of equation\ to expre\\ the entire control
proces*\.E\pre*\\inghuriian experience i n the torni ol';I mathematicril t' ormula I \
a very difficult task, if not an impossible one. Fuzzy logic pro\lides a simple
tool to interpret this experience into reality.
Fuzzy logic controllers are rule-based controllers.The structureof the FLC
resembles that of aknowledge-based controller except that the FLC utilizes the
principles of fuzzy set theory in its data representation and its logic. The basic
configuration of the FLC can be represented simply i n four parts. as sho\s.ni n
Fig. 8.9:
Fi t z, i fi (ut i oi i riiotlirlt), the functions of which are. first. to read. me^' w e .
and scalethe control variable (e.g.,speed,acceleration) and.second, to
transform the measured numerical values tothe corresponding linguis-
tic (fuzzy)\wiables with appropriate membership \Aues.
Kiio, i, lcdgc~ht rsc~,which includes the definitions of the fuzzy niembersliip
functions defined foreach control variable and the necessary rules that
specify the control goals using linguistic \.ariables.
Ii;fi. i-cvic. c) i i i d i m i s i i i . which is the kernel of the FLC. It should be capable
of simulating human decision making and influencing the control x-
tions based on fuzzy logic.
Dc.fir,7,7;~ictitioii i ~ i o d i r l c ,which converts the inferred decision from the l i n-
guistic \,ariablesback to numerical values.
8.3.2 FuzzylogicController (FLC) Design
Thedesign process of an FLC may be split into the five steps described belo\ir.
Selectionoi ControlVariables
Theselection ofcontrol variables (controller inputs and outputs)dependson the
nature of the controlled system and the desired output. I t is more coninion to
I
KnowledgeBase
Controller Crisp Controller
input
b Fuzzifier Interface -+
Defuzzifier
-b
output
+
Figure8.9 Schematic diagram of the FLC building blocks.
use :he output error ( e )and the rate o i derivative of the output (e) ascontroller
inputs. Some investigators have also proposed the use of error and the integral
of error asan input to the FLC.
Mendx~shi pFunction Detinition
Eacn of the FLC input signals and output signals. fuzzy variables ( X , = ( e . e .
U ) ) . has the real line R iis the universe of discourse. I n practice. the universe cf.
discourse is restricted to a comparatively sinall interval ( Xmi n, , Xmax,l. The
universe of discourse of each fuzzy variable can be quantized into a number c f
overlapping fuzzy sets (linguistic variables). The number of fuzzy sets for each
fuzzy b.ariable varies according to the application. A common and reasonable
nutnber is an odd number (3. 5. 7 . . .). Increasing the number of fuzzy SCIS
results i n ;I corresponding increase in the number of rules.
A niernbership function is assigned to each fuzzy set. The membership
functions map thecrispvalues in!o fuzzy values.A setof membership functiors
defined for seven linguistic variables NB. NM. KS, 2. PS. PM. and PB, which
stand for Negative Big, Negative Medium, Negative Small. Zcro, PositiLc
Small. Positive Medium. and Positive Big. respcctively. Membership functiom
can be of a variety of shapes, thc most unusual being triangular, ii-apezoidal.or
a bell shape.The triangular shape is used for the controller.
Forsimplicity, i t is assumed that the membership functions aresymmetricail
:md each one overlaps with the adjacent functions by 50%. In practice. t t t
membership functionsarenormalized in the interval [-L. L ] , which is symmetri-
cal around zero. Thus. control signal iimplitudes (I uzzy variables)are expresscd
in !ernis of controller parameters (gains).
Rule Creation and Inference
In general. fuzzy systems. iis functim estimators. map an input fuzzy set to iln
output fuzzy set S:I -l. Fuzzy rules are the relations between the fuzzy sets.
They usually are i n the form if A. then B, where A is the rule antecedent a!!d
B is the rule consequence. Each rule defines a fuzzy patch in the Cartesian
product A x B (systemstate space).Theantecedents of each fuzzy rule descri!w
a fuzzy input region in the state space.Thisenables one to effectively quanti.w
continuous state space so that i t can cover a finite number of these regions. In
terms of associative memory definition (FAM),each rule represents an associa-
tion (Al: Bl).Thc structure proposed firing all the rules at the same time (analo-
gous to neural networks).This enables easier, faster. and very large scale intc-
grated (VLSI) analogue and digital designs. A fuzzy system using t v v f o
antecedents and one consequence ( A, , B,. C,)is shown in Fig. 8.10.
The association ( Al 9B,. C, )or the rule of A; , and B, then C, maps inputs 4.
25I
FAMrule 1
Figure8.10 FAM system architecture fortwofuzzy antecedent and oneconsequence.
B, to C, a partially activated version of C. The corresponding output fuzzy set
Ccombines the partially activated sets C,...,C,that is,
Fuzzy Inference
The well-known inference mechanisms in fuzzy logic are the correlation-mini-
mum encoding and the correlation-product encoding. Consider fuzzy setsA and
B tobe fuzzy subsetsofX , Y.Thegeometric set-aspoints interpretation of finite
fuzzy setsA and B as points in unit cubes allows the representation of the sets
of vectors. Thus, A and B can be represented by numerical fit vectors A = ( ( I , .
...,q,)and B=( h, ,...,b,,,),\there al =rnA(x,)arid b, =i l l 8 ( U , ) . Let therelation
between A and B be governed by the FAM rule ( A, B) . Using the definitions.
the correlation-minimum encoding and the correlation-product encoding can be
expressed follows.
1 . The correlation minimum encoding isbased on the fuzzy outer product
notation. Expressed in matrices, the pairwise multiplication is replaced
by thepairwise minima and thecolumn sumswith thecolumn maxima.
Thismax-mincomposition relation isdenotedby thecompositionoper-
ator 0.Thus the fuzzy outer product of the fit row vectors A and B,
which forms the FAM matrix M, can be given by
252
M = A O H
Where
and the capoperator indicates the pairwise minima. A
2. The correlation-product encoding use\ the \tandard mathematical o
product of the f i t vector\ A and B to form the FAM matrix M:
urhere
Correlation-miriitnumencoding produces amatrix of clipped B sets,while corr:-
lation-product encoding produces ii matrix of scaledR sets.
Det~izziticationStrntegy
Defuuification is a process of converting the FLC inferred control actions from
f i w y \values to crisp values. This process depends on the output f i 1 7 ~ y\ t - t .
Mhich isgenerated fromthe fired rules. The output furLy\et I \ formed bq either
;I corre1atio11-minimU in encoding or the corre1ation-product encoding.
8.3.3 Fuzzy Rules
Fuzzy rules play a mujor role in the FLCs and have been increstigated exten-
sicrely. However, rules usually can be generated using knoufledgeand operatiig
experience with the system orthrough the understanding of system dynaiiiic?.
I n most cases, fuzzy rules map two input fuzzy variables, forexample. the
error e and the derivative of error 6 * ,intoone output fuzzy cw-iablethe control
signal u. I n such a system each fuzzy Lwiable can be easily quantized t c ;I
number of fuzzy sets.
8.3.4 FLC ParameterTuning
Parameter tuning forthe FLCplays an important roleinachievingthecontrolltr
goals. Prectiousexperience with the controlled system is helpful in selecting 1 he
initial cdues of the FLC parameters. If sufficient information is not a\.aila>I t .
about the controlled system. the selection of witable FLC parameters can Iw-
come a tedious trial-and-error process. Someefforts have been reported i n the
literature to automate the tuning of the FLC parameters at the design stage to
get an optimal or near optimal system performance.
Another algorithm to tune the FLC parameters off-line is proposed i n this
section.The objective of the proposed parameter tuning algorithm is to change
the controller gains in an organized manner to achievedesired system response.
The tuning algorithm tries to minimize three system performance indices (PIS).
These indices arethe system overshoot (OS)and the performance indices J , , J:,
are gi\-enas
(8.8)
where r is the system reference, Jis the system output, P is the system error,
and t is the time.
An FLC with two inputs and a single output has two input parameters K, .
K,,, andoneoutputparameterK,,. Thethree parametersKc, K:, and K,,, aretuned
using the guided search algorithm.
The algorithm changes the three parameters i n overlap loops, simulates the
system with the new parameters, and calculates the performance index. I t also
detects if one of the parameters degrades the performance indices or leads to
instability.I n thiscasei t stopsincrementing thisparameter. If thedesired perfor-
mance indices are achieved, the search stops. Otherwise, i t continues ()\!erthe
specified search range of the FLC parameters.
In case of a complete lack of information about the parameters. the search
for the best parameters may require a large number of iterations i n searching
for a proper minimum. Using some practical information about signal l e ~ ~ l s , i t
is easy to set an operating range to the FLC parameters.
8.3.5 Automatic RuleGeneration
I n some cases the dynamic behavior of the controlled system is unpredictablc
and difficult to understand. This situation imposes the need toautomate the rule
generation process. A lot of effort has been devoted to achieve this goal. I n
some cases an upper hierarchical set of rules has been chosen to generate the
controller rules. This technique requires some understanding of the system dy-
namics to build these types ofsupervisory rules.
I n some other cases, the rules were generated using an artificial neural
network. The generation of rules is achieved by training the neural net\vorh
using sampled data sets. Although effective. this method requires sonis newiil
network background and long training time.
In this section t wo effective automatic rule generation ( ARG) methods arc
proposed.These methodsarevery similarto the neural network technique.Th,:),
also use sampleddataensernble but employ ii fuzzy system to generate the rules
instead of a neural network. The rules can be obtained easily from thc kno\r-n
desired perforrnance (input-output data pairs) of the system when contro1l:d
by another well-designed controller. for example. ;in adaptive controller 01 ;i
proportional-integral-derivative (PID).
I n the case o f a two-input. one-output FLC. the controller requires t \ r o
f uz ~y \*ariables.and using ;I fuzzy mapping function (fuzzy rules). i t gencraic.3
one-output t u~z y control ac*tion. Instead. the ARG is ;I re\erse process. I t ii?cs
the t hr w f u ~ z y\.ariables ( c. 0'. 1 1 ) and generates the miipping function ( t i i i ~ y
rules).
ARC' Using Highest Mdtch Method
The highest niatch ( HM) algorithm properly tuzzifies the three input variabIc.5
(1'. ( c. 1 1 ) . Each fuzzy \-ariable matches in two tuzzy subsets. The algoritliiii
considers only the subsets wi t h highest match (highestmembership value).Tt ~us
i t c iiiatches i n both X and PS with membership values 0.75 and 0.25. resplx-
tively. i t will consider that o matches only i n the 2 subset. Similarly the 0' ;.lid
1 1 subsets will be PS and NS. respectively. Thus tor this sample data. the r.ilc
hii\*ing the antecedents o = Z and Y' = PS will be assigned to;r consequence I ' =
NS: that is. the generated rule will be:
This process is repeated for the whole set of saiiipled dataensemble. At the
endof the abo\,c-mcntioned process. the sainerule may be assigned todifferciit
coiiscquences. Torcsolvt. this problem. the final rule consequence is generated
by \\,cighingall thcconsequences ;issignedtothis ruleby thcir number otoc ~ur -
rences.
A RG Using FLIZZ)' 111i ~~r ~~t 1c e Revcr se Engineering
The irlgorithnitiientioned in Sec.5.6.I focusesthe inforniation derived from the
input-output data set on ;I single rule (thehighest niatch) and ignores the e f t c t
on the other fired rules. This may lead to ;I considerable number of unassigled
(empty)rulcs. Thc f u z ~ yinference reverse cngineering (FIRE)technique i i \ ~~i c I s
this drawback considering the aggregated effect o f all fired rules.
2.55 Aiigle StabilityStirdirs
8.3.6 Fuzzy-Logic-BasedPower SystemStabilizer
A FLPSS with a set of FAM rules and continuous membership functions as
discussed earlier has been designed. An organized method is followed togener-
ate the rule base and to tune the parameters of the FLPSS.
The performance of the FLPSS is first studied using a simulation model of
asynchronous machineconnectedtoan infinite bus.The FLPSS hasbeen tested
on the power system under various fault and load disturbances to ensure its
effectiveness, robustness, and reliability asa power system stabilizer.
Single-MachinePowerSystemModel
A power system model consisting of a synchronous machine connected to a
constant voltage bus through a double-circuit transmission line is used in the
simulation studies. A schematic diagram of the model is shown in Fig. 8.11 .
The system is represented by a ninth-order nonlinear model including the
governor andtheautomatic voltage regulator (AVR)exciter.Thestateequations
representing the power system and the synchronous machine, governor. and
AVR model parameters are given in the referenced literature. The solution of
the model differential equations wasobtained using afourth-order Runge-Kutta
method with a simulation time stepof 1 ms.
I
Governor
TransmissionLines
Figure8.11 A schematic diagram of a synchronous machine model with stabilizers.
The control signal generated by the PSS is injected as a supplementary
\tabiliing signal to tile the AVR siimming point. The sy\teni model has bee1
designed to support both the CPSS and the FLPSS mdcan be ea\ily extendel
to support other types of PSS.
FLIZZ,VLogic Power SystemStnbilizer
Llsing the description given in Sec. 8.4.3for FLC design, a power system st ab-
lizer based o n the FLC algorithm has been developed. Since the goal of this
application ist o stabilize and itnpro\fethedamping of thesynchronous machitic:.
speed de\iition Ao and actiLre power de\riation AP, , ha\re been selected as thc
controller inputs. The controller output is then injected i nt othe AVR siiiiiming
point.
This configuration implies that the FLC has t wf oinput parameters. K,,,anti
K,,,atid one output parameter, K,,,iis seen i n Fig. 8.12.The selection of thew
parameters is usually subjectiire and requires pr e\ ki s knowledge of the fuzi >-
control \cu-iubles(input and output signals).Also pre\,ious experience of' tlic
controlled system dyn~iniics iscorninonly iised In the creation of the fuzzy con-
trol riilt's. Ho~+re\rer, i i n organized approach ;is described in the next section hiis
been adopted forthegeneration o f rules and tuning ofparameters forthe FLPS5.
FLPSS/RLIIc) Genercltion ' IndPJrcimeter Tuning
U\ing the FIRE ARC inethod and ;I sampled data \et generated by using the
CPSS, ii proper \et of rule\ was obtained. The rule\ used in all the tollo~t itig
Fuzzymapping(rules)
is
Figure8.12 Schematic diagram of the FLPSS.
257
ActivePower
PS PM PB ActivePower rNB NM NS
deviation
NB NB NB
Z
1 1NM NS Z
NB NM NM NS Z PS
NM NS NS Z PS PM
NM NS Z PS PM PM
NM NS Z PS PS PM PB
IPM INS Z PS PM PM PB PB
I PB I z PS PM PB PB PB PB
Figure8.13 FLPSS rules generated by fire algorithm.
experiments are shown i n Fig. 8.13.The correlation-product inference niecha-
nism is used togenerate theoutput fuzzy set for the FLPSS.Thedefuzzification
process is based on the center-of-gravity method.
Once the proper rules are obtained, the proper parameter tuning should be
d i n order to achiekre good performance. Tuning of FLPSS parameters can be a
tedious trial-and-error process if not enough information is a\railiibleabout the
range of controller ~w-iables and how they change mrith different disturbances.
The objective of the off-line tuning algorithm is to determine the controller
parameters that provide the desired system response.
The tuning algorithm tries to minimize three system PIS by \.aryiiig the
FLPSS parameters. In the present case, the output parameter is set t o gi1.e the
maximum al l ou~bl e control action,while theother twoparameters K,,,,K,,.\%'ere
tuned using the guided search algorithm.
The selected values of the FLPSS parameters are those that minimize the
use of the guided blind search.Oncethe FLPSS had been tuned,the parameters
were kept unchanged throughout subsequent studies. The tuned parameters of
the FLPSS are
K,,,=100 K,, =0.21 K,, =0.1
CONCLUSIONS
Unlike the classical design approach, which requires a deep understmding of
the system, exact mathematical models, and precise numerical values. ;i basic
feature of the fuzzy logic controller is that a process c;in be controlled urithout
the knowledge of its underlying dynamics. The control strategy learned ttiro~igh
experience can be expressed by a set of rules that describe the behaj-iorof the
2-58 Chccprer- Y
controller using linguistic terms. Proper control action can be inferred from thirS
rule base that emulates the role of the human operator or a benchmark contrcd
action. Thus, fuzzy logic controllers are suitable for nonlinear, dynamic prc-
cesses for which an exact mathematical model may not be available.
Using the principles of fuzzy logic control, a PSS has been designed t.1
enhance the operation and stability of a power system. Results of simulation
and experimental studies look promising.
Thereader isreminded that furtherinformation on thecontents ofthis chap-
ter can be found in the referenced articles and transaction papers. at the end cf
this volume.
Applicationof ArtificialIntelligence
to Voltage StabilityAssessment
andEnhancementof Electrical
PowerSystems
INTRODUCTION
Most of the voltage collapserelated incidentsarebelieved to be related to hear-
ily stressed systems where large amounts of real and reactive power are trans-
ported overlong EHV transmission lines while appropriate power resources are
notavailable tomaintain normal voltage profilesat receiving endbuses. I n some
cases, however, voltage profiles show no abnormality prior to undergoing \yolt-
age collapse because of load variations. Operators may observe no ad\Fance
warning signalsuntil sudden significantchanges in the voltage magnitude result
in actionsof automatic protective equipment tocrash thenetwork.Thisphenom-
enon is caused by large and small disturbances. Large disturbances consist of
the loss of generators, transmission lines, and transformers. Small disturbances
on the other hand consist of slow variation in the system load.
The common analytical techniques used for voltage stability assessment
during these disturbances are:
I . Minimum singular value decomposition which determine the singular-
ity of the Jacobian matrix of the system under study;
2. The concept of multiple load flow solutions which are used for deter-
mining the proximity of a particular voltage collapse point. This
method derives the variation of load change for specified bus Froltage
degradation;
3. The concept of energy margin developed as an indicator to voltaglr
stability by computing the system stable equilibrium points, SEPs,and
i i nst;I b1e eqi i i1ibriuIn points, UEPs:
4. The condition number of power flow Jacobian matrix used to estimat:
the voltage collapse point; and
5. Thecontinuation method, which computes the neighborhood of wltag:
collapse for variation in load. The scheme is based on the solution rf
the load flowequations by suitable modification of the Jacobian using
ii continuation parameter to avoid singularity of the Jacobian near tht.
collapse point.
There are a lot of reservations about the results, accuracy. difficulties, and
the computational burden involved i n using these techniques. Therefore, a tool
Lvhich can probide timely e\raluation of voltage stability of the system i i ndt ~
diirersified operating conditions would be very useful.
In recent years, efforts to irnprokre on speed, accuracy, and ability to handle
stressed/ iI1-conditioned syst enis have 1ed to the deve1opinen of int e11igent sy:i-
tenis-based tools. The potential application of ANN. and ES as alternati\,eap-
proaches forsolvingcertain difficult power system problems, where thecon\'t'ti-
tional techniques ha\re not achiekred the desired speed, acciiracy, or efficicncy
is quite promising.
This chapter deals wi t h the application of ANN and ES to \,oltagestabili-y
iissessment .
9.1 ANN-BASED VOLTAGE STABILITY ASSESSMENT
The purpose of ANN-based voltage stability assessment is to obtain the n i a ~i -
niiini MW loading of the system for a given contingency without conductilig
P1' studies. The proposed scheme for ANN-based wltage stability assessment
iitilizes a multi-layer feed forward network employing back propagation algo-
rithni for the training process. Details of constructing the network architecture,
preparation of thetraining data,andtesting process Nf i l l be done i n the follou.iiig
sections.
9.1.1 Mechanismfor GeneratingTrainingData
Based onModalAnalysis
The mechanism used for the production of the data needed for training the
proposed network is based on modal analysis, employing VQsensitivities that
have the ability to identify areasthat have potential problems and pro\rideinfor-
mation regarding the mechmism of ivltage collapse. The method can be tl e-
scribed as follows:
The system dynamic behavior can be expressed by the first order differen-
tial equation.
x =,fcx.v,
whereX = state vector of the system
V = bus Lroltage Lfector
Under the steady state condition X =0,using enhanced device models. the
linearized power tlow equations can be written as:
where:
AP,/:Incremental change in device real power output:
AQ,/:Incremental change in the device reactive power output;
A@,/:Incremental change in the device voltage angle;
Av/:Incremental change i n the device voltage magnitude;
J,*. J,,l,J,,, and J,, represent a modified form of the ponfertlow Jacobian
elements i n the terms associated with each device.
We can study the QV sensitivity while keeping P constant. and substitute
A P = 0in the linearized power flow equations togive after some simplification.
where
We examine the eigen values and the eigen vectors of the reduced Jacobian .Ih'.
to arrive at:
where A- ' is a diagonal matrix with entries l/h,,h,is the i"' modal tvltage. \!=
q AVand q=qAQ.
The bus participation factors determine the contribution of each h,to 1'0
sensitivity at bus k . They can be expressed in terms of the left and right eigcn
\rectorsof JK, as:
,362
The branch participation factors P,, which give the relative participation )f
branchjin mode I are given by:
AQloss for b r a n c h j
p,,=
max AQfor all branches
The AQ loss can be found by calculating the AVand A0 change at both
endsof the branch.
The generator participation factors c,,,that gic-ethe relative participation of
machine in mode i are given by:
AQfor machine 1~
er,,=
rnax AQforall machines
The machine participation factors can be used to determine the generatclrs
that supply the most reactive power on demand.The reactive reserve at these
generators can contribute heavily to voltage stability. The modal analysis algo-
rithm is utilized in the EPRI VSTAB program.
9.1.2 Suggested Neural NetworksArchitecture
The ANN-based voltage stability assessment proces\ is designed to predict the
critical sy\tem loading, expressed i n MW,fora given contingency without ccm-
ducting PV sirnulations. The network architecture, as shown in Fig. 9.1,is a\
fol10ws:
Theinput layerconsist\of 12neurons that repre\ent the 12input \,ariabl:\.
namely:
1. Q,totallQ,installe~ (ignoringslack bus reactive generation).
L.
3
Q, at most critical generator/Max Q, at that generator. The mo\t crti-
cal generator is the one with the lowest voltage.
3. The Q, reserve at generator with least Q, re\erve.
3. The number of generators \iting at limiting Qq.
5. Lowejt voltage at base case loading.
6. Number of buses with voltage below I p .
7. Total active power demand in MW.
8. Total reactive power demand in MVAR.
9. Total active power loss i n MW.
10. Total reactive power losses in MVAR.
1 1 . Ratio of the mo\t critical branch MVA flow\, to total MVA denxnd
(fromthe power tlow solution for the intact \ystem).
pected Maxi mum
Th
Wa
Hi ddenLayer
Figure9.1 Suggested network design.
12. Ratio of the most critical branch MVA flows to the maximum total
MVA demand before collapse (intact system only).
Thenumber of the neuron in the hidden layer was obtained experimentally
i.e., this number was determined fromstudying the network behavior during the
training process taking into consideration some factors like convergence rate,
error criteria, etc.I n this regard, different configurations will be tested and the
best suitableconfiguration willbe selected based on the accuracy level required.
The output layer consists of one neuron representing the predicted maxi-
mum MW loading in MW.
9.1.3 TrainingProcess
The data needed forthe training process is obtained from the simulation results
using EPRI VSTAB. Next the results from VSTAB areanalyzed and processed
fortraining and testing.The training dataexploited about 75% of the simulation
results.Finally,thedata sets,setaside fortesting,were presented forthe trained
network for the testing process.
The data was modified to satisfy the input requirements of the neural net-
work for better discrimination between data sets.Onethousand cases were pro-
ducedby the VSTAB program forthetraining andtesting process. Theaccuracy
duringthetraining process wassetto 1.OE-03.Theconfiguration with 5 neurons
achieved a convergence tolerance below 1.E-03 in about 1000iterations. The
configuration with 7 Iieiirons i n the hidden layer achiewd the same accuracy
le\,elwi t h ii higher number of training iterat,ons.The configuration with 3neii-
rons i n the hidden layer did not achieye ;in accuracy below 1.OE-3 i n aboiit
600(1 i t t:riitions possibIy reaching a IocaI niinimum .The t raiiiing error shonrt.d
tiiore icariation Liith change i n the architecture than i n the previous studies.
I t \+mobserved that the critical branch loading makes a significant contri-
bution tonz'ard Ifoltagecollapse. The lower the total Q, reserc'e. the lower the
alloudAe niasimum MVA loading on the system. The Q, reserve at the nio.;t
critic;iI gener;it o r is \'cry inst1-11mentiiI i11 maint;i ining high masiini i m IoadahiIi t y v , .
Con\.crselj.,iiiipro\,etiientsi n the voltages and/ orreactive reserve vi acompens.i-
tion at such busesc;m increase thestability margin.Thecases with low cwltagcs
during contingencies at basecase loading hakre alower maximum MVAloaditig
beforecolliipse.The total acti\.e and reacticre losses for the contingencies at tlic
base ciise loading signif)( the degree of change i n the net\\w-k t1on.s due .o
contingcncies.
9.1.4 CaseStudy: NewEngland39-BusSystem
The Lroltage stability studies were conducted on the New England 39-bus sqss-
tem. shoum i n Fig.9.2.wthich is slightly adapted to the study.Theexisting base
c;ise evas used to create ;i \.ariety of cases with different load distributions Lzmith
Iiear-basecase loading. different generation patterns, generator voltage pattenIS.
and transformer tap ratio settings.This was achieved through randomly Lw-yiiig
relecmt pat-metersabout theoriginalbase case values. I n some caseseconon-ic
dispatch and loss minimization were used to make the solution comply Lkci t h
LW1tage and fl ow 1ii n its. Cont ingencies consistirig of different brant h oiit ages
and different generator outages giving rise to different Q, reserves were implc-
menteci on all o f the ;ibo\re data sets. In total, a few thousand different c;ises
\\'ere generated. For each case the 12 parameters used as inputs to the AbN
uwcproduced cc' ith thecorresponding maximum system loadability. 7574 of'tlic
sirnulation results tirere used to train the network ~, h i l e theremaining 25% ~ v c re
iised for the testing process.
Based on the training error statistics, there were 5 neurons in the hiddun
layerarchitecture forthetesting process. A sampletestresult fortheback propa-
gation iilgorithrn using the selected ANN architecture is shown in Fig. 9.3.
Figure 9.3shours the difference between the desired and computed outitit
for the test data expressed in MW. The global nortnalized test error wi s2.07E-
03, which correspond to SO MW. The minimum error is less than 100 M' N.
The difference between the maxinium load at collapse and base load is i n Ihe
neighborhood of 1000MW indicating an accuracy better than 90% i n the ANN
test resLI 1ts.
Area 1
-3 7
29
-
-30
25 26 1
+
I 28 1
1 t
2 I + I + ,. ........... . . . . --38
18
27
9 24 ......,, 6
I
17
+ f ' "
I
"
"\. . . . . . . . .
1 ,.(..................................................... dJJ .
?
16
t 3
/;
, !
.................................................. .....
_,.P.II --35
,_,.....
15
-..,
... 22
T
.......................
I
'-I-: "'i
.......................
5
6 12
23
13
t7 - t 2 1 11 I
--36
+ -
10 /
!
L 4 i 2 i 2 ; 3
8
--a* j
9
Area2 Area3
f
9.2 ANN-BASED VOLTAGE STABILITY ENHANCEMENT
The~ ' o r kofSec.9.I isextended toaccommodate voltage stability enhancemcnt
using switchable shunt compensators. Thebuses sensitiiretoreactiive compcns;i-
tion are identified Lria modal analysis and used ascandidate buses for q~pl j . i ng
compensation. Key physical parameters contributing toward voltage col l qxc
such as reactij'e generation reserve and reactive compensation at base load are
used ;IS key input Lrariables. The ANN output provides the enhanced ni;isiniuni
demand at collapse and the reactive compensation needed to achieixe the en-
hanced stability margin. The New England 39-bus system is used forthe study.
9.2.1 VoltageStabilityEnhancementUsing
Shunt VAR Compensation
The \ T O I t age I*tabi1 i t j r enhancement using shunt VAR cornpcn\ation inoditic\ the
nt'tworh power flow equation\ at base load to the form gi \ en bqr:
,366
100
80
60
40
20
0
-20
-40
-60
-80
-100
J
ContingencyNumbor
ContinwncyNumber
Figure9.3 A \ample test result.
26 7
P : - P ~ , +P , , , =o i = 1 , 2 , . . . .NI? ; E gen ~ = o . I . . . . . ~k
Thecollapse state power flow equations are:
P:- +P,;,=o
i=I , 2,. ...NI? iE gen =0, I , ..., ~k
Q:k- Qkf - (2:: + Q(;,=0
The symbol () represents the state collapse point, Q( represents the shunt
VAR compensation, and k represents the contingent cases. One form of shunt
VAR compensation is via static VAR compensators (SVC).The optimal Lralue
of SVCat collapse point can be obtained through optimization where thegener-
ator voltages and transformer tap ratios can be optimized. An alternative. using
VSTAB, employing power flow without optimization is used i n the current
study.
9.2.2 SuggestedNeural NetworksArchitecture
The ANN-based voltage stability enhancement is designed to predict the en-
hanced maximum demand at collapse and the reactive compensation needed to
achieve the enhanced stability margin. The network architecture. as shoufni n
Fig. 9.4,is asfollows:
er
Hi ddenLayer
Figure9.4 Suggested artificial neural network (ANN)design.
,368
TheIiipcitLayer
Vo1tage stabi1it 4' ;issessment requi res i dent i fic;i ti c)n of the c()1I apse point b;i51: cl
o n load bwiation coitiinonly employing ii property such iis singularity of t i e
load tlow Jacobian matrix at the collapse point. The load at buses participatiiig
i n collapse. the reactive compensation at base load at selected buses, and the
reactit.e reserve at each generator tire used ;is input to the neural network.
The nurnber of neurons in the hidden layer is determined bj,:
A,,+N,+N,
tz8her-e
iV, i \ the numberof buse\ participating i n the collap\e;
N, i \ the numberof VAR site\; and
N , i \ the number of generator\.
Hidden Layer
I n the \tidy, 1-2 hidden layers uith multiple hidden node are tested. For ei ch
proposed nettzwh architecture. the number of the hidden neuron\ varie\ from
2-30.
OcitpLIt L' 7 !/er
I n this study ~e asses\ rc'actiLte compensation at collapse and the maxirnim
poner demand at the col l i pepoint. Hence the number o f the output node. i \
determined by:N, +I .
9.2.3 StudiesConducted
For the New England 39-bus system, S VAR sites were chosen based on the
participation factors. Voltage stability studies were then carried out Lvith t h x
buscs treated as VAR sites containing snitchable shunts. The load at buses
participating in the collapse and the magnitude of the shunts required to m:.in-
tain the voltage within an acceptable range (0.9S-I.OS p i ) at base load and at
the collapse point mw-e noted along with the Q, reserve at base load and the
total rnaximum powerdemandat collapse. The load at buses numbered 12. IS.
16.20, 2I , 23, and 24 were used as part of the input while buses numbered 12.
15.20, 21, and 24 wereselected a s VAR sites based o n their high participation.
A few hundred load configurations were generated using ii random perturbation
around the base load and the preceding procedure u.as repeated for all different
1oad confi gUriitions.
Different ANN urchitectures urere constructed uith hidden nodes ranging
from 2-20, and hidden layers from 1-2. The different ANN were trained l Y, i t h
the 85% of the simulated dataresults. The stability margin was computed asthe
ratio of the maximum power demand at collapse to the base load providing an
index in the range of 1.0-2.0. All other data was comrerted to per unit during
training and testing.
The ANN training statistics are presented i n Table 9.1. This table shoi4.s
the approximate number of presentations required to achie\.ean error below the
tolerance I .0E-04.The training process is terminated if convergence was not
reached with 100,000presentations. The following remarks are made:
I . The different ANN architectures with one hidden layer and nith less
than 6 hidden nodes did not converge beloMy the tolerance error for
o\.er 100.000presentations.
2. All architectures with one hidden layer and with less than 6 hidden
nodes did not converge below the tolerance error for oi'er 100.000pre-
sentations.
3. All architecture with 1 hidden layer and ufithmore than 5 hidden nodes
converged below the tolerant error in less than 4000presentations.
4. Within the scope of the architectiires tested, those with 1 1 . 14. and 30
hidden nodes converged with the least number of presentations. i.e..
600-700 presentations.
Table9.1 Training Statistics for Different
ANN Architectures
ANN architecture
Number of Number of Number of iterations
hidden layers hidden nodes forconvergence
1 3-5 >100,000
6 3800
7 I600
8-9 2200
I 0 I000
I I 600
12-13 1500
14 600
15 1700
20 1100
30 700
7
L 2-5; 2-5 >100,000
6;6 10.000
10; 10 3600
5. Theaddition of an extra hidden layer did not improve the performance
of the ANN.
6. With 2 hidden layers and with less than 6hidden nodes in each laye1.
the ANN did not converge below the tolerance within 100.000presen-
tat ions.
7. For architectiirescomprising 6hidden nodes i n each layer,the number
of presentations required for convergence was about 10.000.
8. For 10 hidden nodes i n each layer. the ANN converged in about 3600
presentations.
A comparison of the test results using the different architectures uith 1
hidden layer are shown in Tables9.1-9.3.
The errors are expressed in MW and MVAR for better readability. Fror-i
these tables the following remarks are made:
1.
The maximum error is i n its maximum \Aue for the network with 3
and 4 hidden nodes.
2. Theerror startsdecreasing at 5 hidden nodes and stabilizesfor6hidden
nodes and higher.
3. The lowest average error is with 7 hidden nodes or uith 30 hidden
nodes.
4. Sincea smaller architecture is naturally preferred, the architecture wi t h
7 hidden nodes is recommended for the study.
Table9.2 MaxitnumTest Error for Various ANN Architectures
Maxi mum
Maximum error in reactiLrecompen\ation error in
at collap\e expre\\ed in MVAR e\timated
Number of inaxinium
hidden node\ Q,12 Qc15 20 Q, 2I Q(24 demand MW
3 37.9 86.0 66.7 58.6 31.3 121.6
4 33.7 47.4 33.2 35.3 23.4 113.1
5 39.7 28.7 15.1 1 0 . 1 16.I 34.8
6 11.2 17.9 9.1 10.5 15.3 29.
7 9.2 13.6 8.2 13.0 11.9 26.9
8 8.4 28.0 10.9 1 1.o 16.9 25.5
9 10.0 13.9 7.8 8.2 14.8 329
10 13.5 22.2 9.0 8.5 15.2 28.9
I I 9.7 28.8 8.2 10.5 16.5 35.0
12 12.0 17.4 10.9 1 1 . 1 10.9 30.3
30 10.0 18.8 8.6 8.1 11.4 298
Table9.3 Average Test Error forVarious ANN Architectures
Average
Average error in reactive compensation
error in
atcollapse expressed in MVAR
e\ timated
Number of niaxiinum
hidden nodes Q,12 Ql15 Q,20 Q,21 Q,24 demand MW
3 19.6 31.9 20.9 19.0 12.2 39.6
4 15.3 18.3 11.0 14.1 10.7 39.I
5 12.0 8.9 5.7 3.9 8.3 15.9
6 5.1 8.7 4.8 5.1 5.2 16.0
7 3.9 8.3 5.5 6.3 5.3 9.8
8 4.8 12.2 5.2 4.5 6.3 12.8
9 5.1 7.9 3.9 4.2 5.7 1 1 . 1
I 0 5.2 9.5 3.7 4.3 6.2 10.3
1 1 4.1 13.1 4.0 5.6 6.9 10.0
12 5.5 8.7 3.8 5.0 6.2 11.5
30 4.9 9.7 2.6 3.5 4.9 7.9
A sample set of results forthe chosen architecture (i.e.,1 hidden layer and
7 hidden nodes)is shown in Tables 9.4and 9.5.Table 9.4shows the difference
between the desired and thecomputed output forthe test data expressed i n MW
and MVAR.
Table 9.5 shows the desired output for reactive power compensation and
Table9.4 Sample ANNTest Errorfor Architecture that Has 7 Hidden Nodes
Maximum
Error in estimated reactive
error in
compensation at collapse expressed in MVAR
estimated
Number of maxiniuni
hiddennodes Q,12 Q,15 Q,20 Q,21 Q,24 demand MW
1 0.28 1.35 1.77 1.52 2.76 2.01
2 9.18 10.85 5.8 2.18 6.86 2.92
3 3.31 4.28 6.41 6.95 2.77 16.46
4 0.46 13.6 7.32 12.98 0.85 1.24
5 0.80 5.32 5.21 12.03 7.74 26.92
6 4.89 2.26 6.13 3.18 11.87 11.96
7 1.85 9.81 8.16 6.99 4.58 17.89
8 5.8 11.52 7.16 0.84 10.53 16.36
9 2.47 10.63 5.04 9.75 1.34 0.66
10 8.79 13.813.18 2.31 6.47 3.82 1.93
Table9.5 Desired Output forTe\t Cases( f r omVSTAB)
De4ired rtlactice compcnwtion at collapw De\ired
expre\\ed in MVAR demand at
collap\e
ca\ e Q I3 Q 15 Q20 Q21 Q,24 in h1M'
I 0.28 I .35 1.77 1.52 2.76 2. 0I
2 9.18 10.85 5.8 2.18 6.86 2.92
3 4.31 4.28 6.41 6.95 2.77 16.46
4 0.46 13.6 7.32 12.98 0.8.5 1.24
5 0.80 5.32 5.21 12.03 7.74 26.92
6 4.89 2. 26 6.13 3.18 11.87 11.96
7 1.85 9.8I 8.16 6.99 4.58 17.89
8 5. 8 11.52 7.I6 0.84 10.53 16.36
9 2.17 t0.63 5.04 9.75 I .31 O.t>6
I0 8.79 13.813.I8 -.. -I 3 1 6.47 3.82 I .93
the poner demand at collapse. For the reacticfecompen\ation. we see that I he
error ranges from about 0.28 MVAR to 13.6MVAR. From Table 9.3me we
that the acerage error is i n the neighborhood of 3-8 MVAR. For the e\tima cd
cicmand at collapse. the error ranges from about 0.66MW to about 26.9MW.
The de\ireci demand being i n the range of 10,000MW. Hence, the relat ce
percentage error i \ Ies\ t han ;ipprouitnatelj 0.3(A indicating good ANN prec'ic-
tion.
9.3 A KNOWLEDGE-BASEDSUPPORT SYSTEM FOR
VOLTAGE COLLAPSE DETECTIONAND PREVENTION
9.3.1 VoltageCollapseUsingExpert SystemTechnology
Thc \theme 1i)rexpert \y\tetn-based coltage collapse detection and prediction
i11cI11de\ the deci si on procedi i re, hnoN!1edgeiicc1LI i\ ition aboLIt hoM.' to \cI ect ,i n d
ranh indicators affecting \oltage collapse (VC), precentict' mea\iire\ to hancile
cari ou\ c iolations. and \election otappropriate models for croltage collap\e dc-
tection and p r e ~ ention.
Specificallq.i t \hould be able to handle the t'ollocving lash\:
I . Selection f or performance index forcontingencie\:
2. Selection of'optitniution models forpreiention;
3. Recommendation o f pre\ention niea\iire\ for cwltage problem\; and
4. Selcction ot.procedi~i*etor conductiiig c\pet-itnent\ that lead to \olt;ige
col I:Ip\e.
9.3.2 KnowledgeAcquisitionand Presentation
The knowledge about the system performance variables and sensiti\?ity paraiiie-
ters may be acquired off-line. The knowledge that needs to be acquired is BS
fo11oWh:
I . Load history:
2. System configurations and critical outages;
3. Control variables: generator taps, VAR, etc.:
4. List of facts on generator data, bus data, capacitor data. line data,and
interchange data: and
5. Solution models and techniques.
Theidentification variables involve the relevant parameters associated \vith
load history, critical line outages, control variables, and other niodeling tech-
niques.From the voltage collapsedetection schemesreviewed, i t was found that
load history is not taken into account, the selection of critical lines depends on
thesystem conditions andthatprioritization of control parametersaffectssystem
behavior. Therefore, currently used mathematical predictive techniques cannot
be used to ascertain adequate voltage collapse detection and preLrenti\.e mea-
sures.
Theproposed work reported here basically considers the static model using
multiple-solution and divergenceevaluation ofpower flowcalculation to predict
VC and the dynamic model for studying the effects of governors. exciters. etc.
The preventative measures used are the VAR planning technique and optimum
power flow(OPF) forcorrecting violationsof voltages and flowsat an optimum
cost.
It is intended that the proposed expert system shouldbe able to identify the
type of data encountered, choose the best model for voltage collapse detection
and use the appropriate model of OPF for prevention. The knowledge/esperi-
ence regarding thedetermination of thebest protection forgiven impact parame-
ters should be appropriately acquired off-line before implementation.
9.3.3 Structural Designof KBVCDP Scheme
Theestimationof voltage collapsephenomena isperformed inthree major tasks.
namely (1 ) power flow base established or divergence control, ( 2) selection of
detection schemes, and (3)correction of violations. This framework de\.eloped
i n numerical methods is designed and enhanced by using KB strate,' c'les to
achieve optimal performance. Figure 9.5 depicts the implementation logic of
interdependent models required to build the knowledge base voltage collapse
detection and prevention (KBVCDP)scheme. It consists of a detection and pre-
vention scheme enhanced by the KB support, as described i n the proceeding
sections.
c' InputData
x Planning
SystemExpansion
Operational I
AutomaticDisplayofDetection:
Run PowerFlowto - P-V,Q-VCurves
obtainBaseCase 4 - Parameteredcauses I
Solutions - Operational&Planning
r-
ResultEvaluationby
Knowledge-
Knowledge-Based
Decision-Making
ofMeasuresof
1
PreventingV.C.
t 4Yes
Knowledge-BasedModeland
Knowledge-basedsupport
MethodSelectionforSophisticate
I
forscreeningofpossible
indicationwhichwill
causeVoltageCollapse
- Contingency
4
- Selection
>
- Load Condition
- EventAnalysis
- Prioritization
PlanningforSiting
I I
SelectionofVCD
NumericalMethods
PerformOptimalPowerFli)w
4 I
ResultEvaluation
I Transient Power
DynamicStability
Stability FlowModel Model
Model
AutomaticDisplayofF'reveiition
Results
- Effectiveness
- Reliability
ResultEvaluation;Model& - Economics
'7
. Method SelectionSuggestions
- B
ofthe KBVCDP implementation. Figure9.5 Fl o ~~c h a r t
4
27-5
Power Flow DivergenceMonitoring(KBPFDM)
The program modules utilize heuristic rules to identify cases of load flow run
that may diverge due to bad data, telemetering error, method of solution. etc.
Suggested corrective measures are also given by the expert system.
IndicatorSelection(KBIS)
For an off-line type of study, VC estimation may be detected by evaluating the
impact of different indicators on the system that will predict limit \.iolations,
thereby causing VC. The KBIS provides prioritized parameters for detection
schemes, which can be any of the numerical methods or improved power flow
techniques such asan automatic power flow technique.
The automatic power flow technique is a self-contained power flow ap-
proach that givesthe sensitivityof voltage measurement with respect toparame-
ter changes and make QVand PV plots toidentify the knee point. Other select-
able detection schemes developed within KBVCDP include other numerical
techniques such as Barbier's and Schlueter's.
KBVCPreventionScheme(KBVCP)
The identified critical VC points and measures for preventing VC are handled
effectively by the KBVCP. I t selects appropriate controls as constraints for the
optimization algorithm, minimizes relevant objective functions and includes
other conditions and models i n seeking an optimum corrective measure that is
economical and effective. The display of VC knee point and the indicators are
given i n graphical form. Using the KB system, the corrective measures and the
rules fixed are tabulated in matrix forms.
9.3.4 StructuralDesignofExpertSystem
A preliminary prototype of the expert system used to verify the feasibility of
the KBVCDPhasbeen designed inPROLOG. Itsstructure isshown in Fig.9.6.
It consists of the following parts: knowledge base, blackboard, inferenceengine
and user interface, and application programs.
KnowledgeBase
A knowledge base in the expert system (ES)is the principal source of knowl-
edge used forthe KBVCDPimplementation. Itconsists of afact base and arule
base.The first baseconsists of fact statementsandcontains thebasic description
and record of all power system components, status and configurations and
knowledge needed for describing power system states, conditions for selecting
measures and methods forinterpretation of results.Theknowledge contained i n
HUPFP
I
Knowledge-
Data BasedSupport
Man-Machine Bridges Systems
Interface
( C )
(OPS83;
%
PROLOG)
Figure9.6 Prototjpe w-uctureo f the hnobledge-based wpport \y\tein.
the rule ba\e I \ u\ed to de\cribe the cau\e-effect relation\hip between fact\ i n
the tact baw and perform the ba\ic deci\ion mahing that bill lead t o a good
deci\ion \upport \chenie.These rule\ are wed i n forming the KBPFDM, KBIS,
KBMS, KBCM, etc.
B/xkDoC?rd
The blackboard is designed for database inanagemetit and communication t c -
t uwn the expert system and other units. Messagesto prompt new rules orst.irt
new evaluation, terminate a study, select relewit data, or improve qualitati\-e
reasoning are stored uithin the blackboard. With results reasoning comple1t.d
output modules and user accessmodules are fetched to complete the study.
Inkrenc-eEngine
The inference engineisdedicated tomaking effecti\,euse o f domain knowlecge
i n the hnowledge base and to perform message interaction on the blackboard.
based o11 diit 11 dric'e11, foruwd-chaining,and tiiotiUIar procedUres.
Rules for Detecting Power Flow Divergence
The rules for detection are deceloped based on the experience of the operator\.
literature review, and interviews with experts. The rules detect bad system d;ita,
checherrors i n telemetering, andevaluate theerroneous results due to nurnerical
methods l i d . Forming those rules and their combination\ i \ designed to i t di-
cate \pecific load tlow runs that lead to divergence.
RLi/esfor Se/ectingVCPCirm?eters(indicators)
Rules decelopment for prioritizing parameters [hat may lead to VC itre con-
\tructed. They consider the impact o f ec'ents \uch as line outage, generation
outage, load transformer taps, excitation effects, and other combinations along
wi t h other sy\teniconditions. The severity of indicators causing VC :ire ranhed.
Rules for Selecting ModelsandApproaches for VCDetection
Rule development for selecting VC detection methods are based on choice of
models, desired options (speed,accuracy, and display),and other niajor e\'ents
causing VC.
The expert system selectsthe indicators that ha\,eahigher priority for VC.
Several system parameters, limits, and violationsare stored i n the database.The
detection system condition and events causing VC are based on the skills and
experience of the operator. Very often the operator's decision \ +r i l l change ac-
cording to the severity of the contingency or other factors.
The detection and prevention of voltage collapse (VC)based on the KB
approach is fast and accurate. The system reads the data, performs manipula-
tions, prioritizes the indicators causing VC and selects appropriate measures.
models, and methods for detection and correction.
9.3.5 KnowledgeOrganization for KBVCDPSystem
The de\dopmentof a KB support system for KBVCDP requires the determina-
tion of a knowledge organization scheme for the VC detection and pre\ention.
Using a matrix format, the heuristic rules describing detection strategies are
formalized. The rules are divided into each of the subtasks. With appropriate
data, the prototype KB system will be tested.
Rules for Preventive Measures Selection
Rules for selecting preventive measures are developed based on the follo\+ring
aspects:
1. Rules forranking correctivemeasures int er mof theeconomic evalua-
tion. The effects of system conditions such as loading, generation.
length of lines, and seasons of the year are employed to select appro-
priate measures.
2. Rules for evaluating reactive power, LTC, and others constrain effec-
tiveness.These rules evaluate the effectiveness of LTC, and other con-
trols as constraints forthe optimization algorithm. The role of LTCas
detrimental or beneficial is evaluated by using system conditions. The
selection of appropriate objective functions is identified by using sys-
temconditions and limit violations based on the typeof \kdation.rank-
ing of indicators causing the violations. and other system conditions.
9.4 IMPLEMENTATIONFORKBVCDP
The algorithm described in Fig. 9.5 was programmed on a VAX I1/780
and IBM PC compatible computer. The numerical programs are coded i n
FORTRAN,while the symbolic computation programs are coded in PROLOG.
The functional descriptions of each of the KB modules are discussed.
Thedata filesandcommunication between these modulesarealsodisplayed
i n Fig. 9.7.The program runs on the IEEE 57-bus test system using the rules
developed for each of the KB modules. These rules are simulated based cn
predetermined experiments.
When theexperience and knowledgeaspectsarefilled inthe matrix accorcl-
ing todifferent selections,different rules canthen bedeveloped.Therule matris
approach makes rule development both easier and faster. At the same time, the
rule matrix method is also very useful forthe new rule developed while the neA'
experiences have been gained. Each column of the rule matrix is a knowledge
uni t for the rule condition or conclusion and suggestions that are identifiell.
Tables 9.6-9.9 summarize some important results based on experience.
Other combinations of system conditions are checked and results will cla+
sify divergence of the power flow as critical, sek'ere,or normal for the po"l:r
flow divergent part of KBVCDP. Other parts of KBVCDParedeveloped by the
same approach.
Rules are capable of prioritizing causes of power flow (PF)divergence.
multiple solutions, and extraneous solutions. Other associated knowledge bases
for kultage collapse detection include the development of rules that study the
indicators fordetecting voltage collapse and selectingthe performance index fq.lr
\!oltagecollapse studies,and rules forselection of measures for \voltage collap;e
periods.
9.4.1 Limitationofthe Operator Assisted ExpertSystem
When the human experts' heuristics are successfully encoded i n the KB, a re-
ipiew ofthe literature indicatesthat the ES was comparable in performance with
human expert to a certain extent. Creating the KB for the VCDP or for any ;lf
the operational problems (e.g.,security-assisted. and voltage control)is ;1dit'i-
cult problem. One must first define the problem to be solved and then encode
i t in a set of rules. However. the KB cannot solve the problem by itself. A
human expert uses past experience together with the background of the whcle
problem to make the correct decision on causes of VC and preventive niethoh
needed. Hence, in order to design an ES that can utilize the expertise of t i e
human, we must find away of integrating the human expert's empirical knovrl-
edge i n one domain.
Successfully encoding the human expert's empirical knowledge into a x t
SYSTEM
I
PONER 1.
COMMlrZJlCATlO
CHANNEL
Operaton
environment
Di spty
I ntcrprewh
of Resutu and
SUggCStCd
actions on
screen alum
Rartincly
recording
TASK MANAGER
4
l n f m
Engine
~ ~~~ ~
NUMERICAL. PROGRAMS
FORTRAh'. C
-
2x0
Table9.6a Matrix of Rules f or Voltugc Collapw:MethodSelcction
of production rules might also pro\feto be a difficult task.The knowledge baw
is likely to be incomplete and inay be inconsistent. These limitations iirejus . i -
fied i n VCDP since the cause of'VC are system-dependent. The model uscd
in\'ol\,esthe interaction of system parameters, system conditions, etc. Thc EiS
designed may have the problem of portability mc i inconsistent knowledge.
The traditionaI nonknow1edge-based approaches for vo1tage co11ii pse detcc-
tion and prevention can be used to s ol k~ the probleiiis i n VCDP. At the s;iiiic
time, the following are common probleriis:
Thedi\ergence reawn forthe power tlwf,automatic seriesrun power tlon.
and optimal pouer t l ow program\ are aluay\ difficult to identify espe-
cially for the les\ experienced power engineer\.Even thejudgement of
experienced engineerscanbe wrong because \x)luminousinput and01 i t -
put data are i n\ wl \ ed.At the wmetime, the divergence of the iteration
procedure ma y be cau\ed by reawn\ i n the program other than i n the
data.
28I
Table9.6b Matrix of Rules for VoltageCollapse:Method Selection
Events IF Then I 1 I I
Table9.7 Corrective Measures and Economical Ranking
Then
If
Econoinica1
No. Measures condition number
3
I Capacitordispatch -
3
- Dispersed generation control 1
3 Demand-\ide management 6
4 Voltage reduction for load relief 5
5 LTC and distribution voltage regulator 4
6 Load shedding 7
7 Series conipensation 8
8 ParraIlel compensation 3
9 Network expan4on 9
I 0 New plants I 0
Table9.8 Matrix of Rules tor Poiver Flow Di\wgenceIndicators
(SolutionsMethods)
If Then
For
Newtori-
Mis- I11itial Ri ~phs ot ~ For GLitlss-
I11;It cheh v;I I11es INR) Seidel ( CS)
Rul u divergence clivcrgencc
niir~iber Iricr- Dec. Ye5 N o cliie to cltic to Concl usion
I X X III-condition
problem
&
7
X X X
-
3 X X X Data
repre\entation
-
J X X X
5 X X X Initial 1 due\
>Zero Deficit Surplus Beneficial
_________-
s X
X
X X
x X X
X X x
x X X
N X X
I:
X
The \voltagecollapse indicatorselection is \.cry complicated. I n po\i'ersq's-
tern \ dt age collapse studies. what causes the \roltage co1l:ipse is the
most important study aspect of the stable problem. Most of the time.
there may be more than one cause, but deteriiiining the original and
most importantones aredifficult.
Becauseof the selectionindicators,the pre\'entiveiiieiisureselection is also
quitecomplicated.The selection of pre\rentivc measures iii\wl\.escon-
siderati on of economi cs, planning strategies. operatioiiaI secU r itj,and
systemteasibilityforthe measures.At thistime. there :ire noacceptable
strategiesthrthis problem.
For certain study problems, the selection of studj, models and nuiiicrical
methodsis;I complexoneevenforresearchers\ 1 7 h o 1iai.ebeen stucijing
voltage collapse phenomena for ;i long time. Interpretation ot'the nu-
merical program results usually requirethe-judgeiiieiitof ;in engineer.
Tosolve the abo\,eproblems for the nonkno\l..lcdgc-hased approaches. the
knOM'1edge-based ;I pproach has the fo11owingadYantuges:
The data bridge can transfer the quantitatiLredata into qualitati\re data di-
rectly.Thus, the KB schemecan directly use the data for logical rea-
soning for determining any possible errors i n the numericd data anti
making suggestions for correcting them. This procedure n,ill s i i \ ~;I
considerable mount of time for indicator identification.
Thek11oLVIedge-basedvo1t ageco 1 1 ap sei ndicator se1ecti on.st idy inodeI.and
method selection scheme can make more sophisticated and faster
choicesforthe decision-makingprocedure.The selectedindicatorsand
study model is more suitable for the study. This is the siiiiic ;is i i i
preventivemeasure selection,the KBVCDPgivesmore choiceof iiitxi-
suresrather t han the rightjudgement at due time.
The KBVCDPalsogives heuristic and clearguidanceon how to detect aiiti
pre\'ent\,olt:ige collapse in a more con\renient ~ z a q ' .
The approach, using knowledge-based technology, is suggested toenhance
the detection and prevention scheme for voltage collapse. The suggested ap-
proach has been prototyped asan expert system program both on the IBM PC/
AT using the PROLOG language and on the VAX I1/780 using OPS83iind C
languages. The data bridge concept has been introduced. and data bridges tor
different 1111merica1 programs havebeen designedand impIeinented.ShorteI' I W -
~ soniiig time and more accurate results by using KBVCDP are \.eritiecf 1 ~ the8
results and the explanation.
9.4.2 Examplesof KBVCDPOutput
The KBVCDPdeveloped has been tested on different \j'\[eiiisunderoper;itional
conditions. The program was initially deLreloped under the VAX 1I /780 and
Table9.10a Test Results Number I : Rule Matrix for Power Flom Divergence
(57-BusSystem)
Test V-6 PV Line\
case Rule position position R/X 105s System
11Uniber fired Tp nuniber number ratio number condition
1 In limits 1 12,39 Low No Se\rere
3 Out Iimits
3
- 12, 39, I High No Severe
x Out Iiini ts I 27, 36,42 High 1-5 CriticaI
I2 In limits 3 27, 36.42 High No Critical
IS Out Iimi ts 12 I , 39 High No Critical
16 Out Iimits
3
- 27, 36,42 Low No Critical
20 I n limits 39 I , 12 High 1-5 Severe
21 Out Iiinits 1 12, 39 L,ow No Severe
IBM PC/AT environment using PROLOG, PASCAL. and FORTRAN. Later,
the program was modified using OPS83, C, DCL, and FORTRAN language\
entirely under the VAX I1/780environment.
9.4.3 Tests of KBVCDPon the 57-BusSystem
Todemon\trate thecapability of the KBVCDP,a seriesof testswereperformtd
on the 57-bus system.Tables 9.10a-9.10f give the rewlts ofthe rule matrix fq)r
Table9.10b Test Results Number 2: Rule Matrix for Power Flow Divergence Indicator\
(SolutionMethod)
Te\t Newton- GaI\\-
ca\e Ruled Initial Data Raphson Seidcl
number fired Mi mat ch \:due\ preci\ion Tolerance method mehod
9 I Increase bus Ok Double Not reached 11I -condi tional
nUmber data
12
10 2 Decrease Ok Single Not reached
I I 4 Increase bus Ok Double Satisfied Data precision
number
29
12 S Increase bus Bus number Double Satisfied Initial \,alue
number 33 hea\'ily
29 loaded
28.5 Voltcigr Stcihilitj~As.srssriieiit/Elec.tr.ic.alPortvrSj~stenis
Table9. 10~Test Results Number 3: Rule Matrix for VC Indicator Selection
Test Line
case outage
number Rule fired number LTC condition Compensation Rank
13 Event load I 1-15 Off limit 4-18 Serie5andparallel on I
14 Event load 3 1-15 Off limit 4-18 Serie\and parallel on 1
15 Generation 1-15 1-15 In limits Serie5and parallel on 2
event 3
16 LTCevent 2 Off limit 4-18 Serie5and parallel on 2
17 Compen5ation I -15 Off limit 4- 18 Serie5 on L7
event 5
Table9.10d Test Results Number 4:Rule Matrix for Detection Method Selection
Test case
number Model selection Speed Accuracy Display Selected method
~ ~ ~ ~~
18 Dynamic X Tamura and Barhier
19 Dynamic X Tamiira
20 Static and dynamic X Tamura
21 Static X Hu and Schlueter
22 Static X Hu and Schlueter
23 Static X Schlueter
24 Static X Hourard Uni
25 Static X BPA
Table9.10e Test Results Number 5:Rule Matrix forPreventive Measures Selection
Test CorrectI \e
ca\e Flow meawre
number Model selected Qlimits V limits over Q(,>Q(,,,l, ranh
26 Dynamic out out out Yes 5
27 Dynamic out out Yes 3
28 Dynamic out out out 1
29 Static and dynamic Out out out 3
Table 9.10f Te\r Re\ults Number 5: Rulc Matrix for OPF
~~ ~
30 X X X X X X X X Voltage and c w t
31 X X X X X X X X Q,,,\,c,
the power flow divergence indicator with respect to power flow data.Test cases
number I and number 2are with correct slack busandcontrol busposition. The
RX ratio anddata format results in the samesystem condition. that is. classified
as "severe."
9.5 UTILITYENVIRONMENTAPPLICATION
The developed KBVCDP isnumerical program-dependent. Fordifferent nuiner-
ical programs, the data bridges from numerical program to knowledge-based
program vary. The application of KBVCDP to different utility companies needs
more implementation of the data bridges for special progranis.
CONCLUSION
The use of artificial neural networks for voltage stabilitj' assessment md en-
hancement has been presented. The method utilizes modal analysis to obtain
suitable buses for use as VAR sites. The ANN-based maxiniurn loadabilitj, of
the power system was demonstrated and the ANN-based maximum loadabilit>r
of the power system with VAR compensation was demonstrated ;IS ~s~el l .
The maximum loadability of the power systemand rei1ctiL.epo~i' erconipen-
sation at collapse used for training the ANN haLre been obtained using the EPRI
VSTAB program. The sensiti\rity of the accuracy of the predicted output from
the ANN has been investigated with different ANN architectures. The trained
ANN has been tested with input data not previously seen by ANN.
I t isconcluded that theback propagation algorithm canbe usedLvith reason-
ableaccuracy forproviding an estimateforthereactive compensation at collapse
aswell asameasure ofthe voltage stability margin in thepresence of sis.itchable
shunts. The appropriate ANN architecture needs to be determined based on the
re1evant syste in data.
The trained ANN provides reasonable results i n an extremely short time,
almost instantaneously. when compared with other existing methods i i tiIizing
succes s ive power flowor optimization.
The inputshutputs for training are easily obtained kriasuccessi\re po~i' er
tlows and do not require extensive computation when compared \s,ith certain
otherenergy-based methods utilizing ANN for assessment of the Lroltage stabil-
ity l i mi t .
Thedesign, analysis, anddevelopment of a knois.ledge-based expert system
to detect and prevent voltage collapse has been demonstrated. The detection
scheme selects parameters that cause voltage collapse and also evaluates a\,ail-
able models of the power system. It uses rule-based techniques for prioritizing
indicators and choosing appropriate numerical methods.Theprevention schemc
identifies measures needed to correct violations of bus voltages and flows due
to a loss of generating units or critical lines. Selected measures are under-load
tap changers, capacitors, line switching, load shedding, and generation adjust-
ment.Theknowledge-based system ranks these measures in terms of theireffec-
tiveness incorrections. Subject toequipment and network constraints. aspecial-
ized optimal power flow scheme based on a reactive power model of a power
system minimizes cost and power loss.
Theexpert system was testedon medium-sized powersystems. Bonneviell:
Power Administration's Puget Sound Area, and New England's Boston Are: .
Results reveal the need for adequate use of under-load tap changers and other
control devices as a short-term measure to prevent voltage collapse. Such de-
vices can otntiate the need for the installation of a new generation of capacity
and the expansion of transmission systems.
10
Epilogue and Conclusions
This book treated broad issues in power systemdynamics with special emphasis
on the twin subjects of angle and voltage stability. Both are essential tools for
dynamic security assessment.
Dynamic security characterizes a power-systems ability to withstand dis-
turbances and ensure continuity of service. In operational planning, dynamic
security analysis encompasses a large class of problems. such as finding the
security statusof a network, the power transfer limit in atransmission corridor,
the worstcontingency inaspecificarea,orperforming somecomplex sensitivity
analyses. In practice, dynamic security isoften measured in terms of adynamic
security limit, defined as the maximum power transfer for which the network
cannot only survive the worst possible normal contingency, but also guarantee
an acceptable level of service quality without lossof load.
Dynamic security analysis is dominated by the use of algorithmic software
for off-line evaluation and control of load flow, transient stability, and other
network characteristics. Even though simulations are readily performed, assess-
ing and ensuring dynamic securityacross all possible topologies and contingen-
ciesremains a formidable challenge.
Power system networks currently operate more and more in a stressed state
where conservative operation often results in significantfinancial consequences.
The process for an on-line dynamic security assessment requires a reasonable
level of built-in intelligence to detect and determine the following:
1. Assess the systemdynamic performance
2. Determine the degreeof stability or instability (margin)
3. Determine the sensitivity of the margin to key \w-iables
4. Iteratively repeat the preceding steps to obtain the stability limit.
In ordertodoii crediblejobof securityassessment, a large number of coti1bin.L-
tionso f nctccwk configurations undercontingency must bedone.Thisis bejvricl
the capability of the typical operational planning uni t i n iiii electric pokcw- s y -
tem. There iire three reasons forthis:
I . Determining the power transfer limit is ;i complex. iterative process
requiring the execution of many stability simulations and considerable
expertise. Toperform a single power transfer limit search, one begi is
byexecuting;I power tlow for ;I gi kwtopology and carrying out ni m-
ual modifications to the input data until ;Isatisfactory steady-state C;ISC
is tound. This is then used to initialize the t1etu.or-kfor a transieiit-
stability simulation, alsomanually initiated by the user.Oncethe simu-
lation is complete. m'emay resort toyet another tool to apply transietit-
stability criteria, determining what the next step u.ill be (i.e.,'incre;ise
or decrease powertransfer i n the faulted corridor),tomodify local-flow
software inputs accordingly and re-enter the process. This is repeated
until the required limit is found.
2. Theprocess of computingtransient stability simulation remains higttl),
time-intensi\~e due to expertise-related tasks. This is true for data cq)l-
Iection, inpi1t \'a1idation, oiit put postprocessing.ancl resii It anal ysis -e-
yuiring ii considerable amount of time.
3. Additional expertise is reyuired to scale the problem clou~ii n such ;I
\+raythat the dynamic security limit of only ii small set of topologies
need to be found explicitly. Thisreduced set forms ii basis forcstimat-
ing the needed dynamic security limits. The price to pay is ii lack of
precision that translates into inore conser\xti\re security limits.
In orderto improve the efficiency and accuracy of off-line intelligence S ~ C L I I i t y
analysis. the application of artificial intelligence techniques has been proposxi.
Alternatively, expert system. fuzzy logic and ANN usually i n combination U ith
energy based methods or simplified algorithms may be used. An interestmg
recent concept involves the L I S ~ of a generalized shell forthe purpose of mecx i -
nizing processes i n dynamicsecurity assessment. MWXLLLI et al. (1993)detii~)n-
strate that semantic networks combined nrith frame-based structures can eftec-
tiirely model the process and provide a framework for constructing a histilj
powerful anduser-friendly en\rironment.Theirresearch carriedouton a produc-
tion level prototype showsthat i t is possible to mechanize routines traditiondly
carried out by hunian experts. This mechanization greatly enhances the reali/a-
t i on of complex processes.
Theexpectationof dramaticincreases in theinterconnection andcomplexity
of power systems has raised concerns about the performance, security.and con-
trol of transmission and distribution networks. These increases are expected to
result primarily from two developments:
1 . New legal requirements allowing greatly expanded Miheeling of p o \ \ ~ r
over existing networks, and
2. The wide-spread application of power-electronics-based acti\res\vitch-
ingandcontrol devicestoraise thereal powercapacity of existing lines
k i i local control of i dt age, impedance, or phase angle.
We have already suggested that power systems are already becoming increas-
ingly stressed. Unexpected behavior has been observed i n many netnarks under
11 nLIsual stress,behakrior suggesting that some important systein d ynaiiiics are
not yet well understood and hence are not accounted for i n simulations con-
ducted for control and security purposes. Enhanced fundamental knontledge
about system stability and nonlinear dynamics may be necessary i n order to
control existing and future power systems reliably and continuously iinder all
contingencies. In addition toensuring power network security under normal and
exceptional stress. such improved understanding could yield substantial eco-
nomic benefits by allouring secureoperation closertoperformance margins t han
is currently considered safe.
Interconnected power networks can be described as stressed. highly nonlin-
ear,noncontinuous systems.Systemsof this type areextremely difficult t o :~ccii-
rately model,eithermathematically orconceptually. Thus,current p o ~' e r sqstem
simulations generally aim to model the slow, quasi-continuous dynamics most
significant to system dispatchers, such asexcessive voltage and apparent pouw
variation with respect to changing demand. The details of transient beha\ior,
with which most computational complexity is associated, are for the most part
neglected.
TypicaI network simu 1ations are cons iderab1y siinp1ified and app1-0xiin;ite
(althoughthey still incorporate hundreds of complex equations).I n addition.
even the most comprehensive models include fundamental approximations to
account for anticipated behavioral nonlinearity or noncontinuity and depend on
critical assumptions about performance because we lack ;I complete understand-
ing of the physical behavior of complex power system. For example, to fncili-
tate computation for conventional simulations, parameters like load \.alues are
held constant, even though these parameters are known to \ my slightly. And
the modeling of systems that incorporate activecontrol devices requires the use
of assumptions because complete descriptions of device behaLrior are not yet
avaiIa bIe.
Observing that all models donot use the sameapproximations and assunip-
tions. their predictions of system behavior may differ strikingly, especially for
conditions of stress. Utility experts thus find it difficult to assess the validity ( ~ f
agiven model's results. Nonetheless, aslong aspredictions indicate that control
measures will maintain generation-load imbalances within a normal operating
range, dispatchersassumethatthesystemwill remain stable,and theyarealmost
always correct.
Recent studies of historical voltage collapse events suggest that traditional
load simulations (suchas those assuming constant load angle, constant imped-
ance, and constant current) may not capture important voltage dynamics. For
example,avoltage collapse may be initiated by the oscillatory behavior of geri-
erator exciters, but models that consider only the slow dynamics of the system
ignore exciter dynamics and other transient behaviors. As a result, unsteady
system states may be incorrectly assessed as stable, with continued operaticn
possibly leading to local or systemwide failure.
Chaos theory, the study of nonlinear dynamics, may provide new ap-
proaches for understanding the transient behavior of complex power systetiis
and for ensuring network stability. Mathematical research during the past 2 0
years indicates that any system involving feedback can exhibit chaos.It can tie
shown that chaos is present in simple power system models over a range of
loading conditions. Recent research results suggest that events such as voltage
collapse, low-frequency electromechanical oscillations, and transient stabiliiy
may be linked to chaotic behavior.
I t has been established that there is arelationship between voltage collapse
and chaos-related bifurcations in voltage-reactive power solutions. Significai t
work has been carried out aimed at increasing our understanding of chaos arid
transient stability. In particular, researchers examined in detail certain unusual,
possibly chaotic behaviors observed in small power systems and are looking at
tools for evaluating system stability under parametric or structural variations.
Efforts focus on improving the characterization of bifurcations in simple arid
complex power systems. Bifurcations, which represent qualitative transforma-
tions in a system's operational behavior (changesfrom stable to unstable states,
for example,or the onset of multiple allowable solutions),can lead to chac-s.
Because fundamental power flow equations have multiple solutions (swingdy-
namics),bifurcations i n power system behavior are always possible.
A three-bus model was enhanced to include generator dynamics, revealing
severaladditional bifurcations,someof which leadtochaoticoscillation through
period-doubling cascades. For these cases, an infinite number of power flew
solutionsoccurin response toonly asmallchange inreactive load,causingrapid
tluctuations in voltage and loadanglethatcouldproduce power flowoscillations
exceeding the thermal limits of transmission lines and lead to system collapsc!.
Simulations focusing on these dynamic-generator-case bifurcations also in-
dicate that voltage collapse may take place before the reactive power demand is
increased to the system's steady-state operating limit, the point at which t i e
static-case, saddle-node bifurcation occurs. Thus,the model system, and proba-
bly actual power systems as well, may be less stable under fluctuating real-
world conditions than under the steady-state cases assumed i n conventional sta-
bility calculations.
Structural stability is a relatively new concept i n power system analysis.
Broadly speaking,asystem isstructurally stable if small variations i n the model
do not change qualitatively the set of trajectories originating from all initial
conditions in the state space. For a given dynamic model. we examine system
behavior subject to small disturbances in the domain of parameter space. This
is different from Lyapunov stability where we wish to know if a perturbation
in the state-of-the-system results in the trajectory returning to the equilibrium
asymptotically. A dynamic system which is unstable in the sense of Lyapunov
is structurally stable since the trajectories do not change for small changes i n
parameter values.
A complete characterization of structural stability for two-dimensional sys-
tems can be related to the nature of the equilibria and limit circles. For higher
order systems such a complete characterization is not possible. Bifurcation the-
ory is directly related to structural stability. Research work has been conducted
that aims to provide dispatchers with a better understanding of the liinits of all
significant structural parameters (such as load angle, current, impedance, and
reactive power demand) so that power systems can be operated to maximize
cost-effectiveness without initiating harmful chaotic behavior.
Structural stability is important forpower system security because key pa-
rameters areknown tofluctuate slightly.For asystem tobe assessedas structur-
ally stable, itsbehavior must return toits steady-state if perturbed. Forexample.
a power network is considered structurally stable if slight changes in current or
load angle donot appreciably affect the interactions between voltage and other
system variables. (Thisdiffers from the more common notion of Lyapunov sta-
bility, which assesses stability on the basis of the initial values of system \ , xi -
ables.)Structural stability defines the parameter range within which system be-
havior is qualitatively similar; multiple ranges may exist fora given parameter.
Little work has been donein this area for nonlinear models of power systems.
Research has revealed that for systems characterized by more than two
parameters, no complete description of structural stability ispossible. But para-
metric limits of structural stability can be related conceptually and mathemati-
cally to bifurcations. Both indicate qualitative changes in system behavior and
can be seen in phase portraits of a system.Thus,bifurcation analysis of system
behavior can partially characterize structural stability ranges.
I t was observed that if control actions are taken into account, the critical
values of parameters that influence voltage collapse may be lower than those
predicted by staticcriteria. They postulate that the range overwhich a dynamic
system isstructurallystableisalwayssmallerthantherange forthesamesystem
considered under static conditionsonly.Thus, parameter margins identified by
con\vntional, static-case pourer system models to ensure safe system beha\,ior
ma ~ ' be obre1.1 y permissive for dyn;i m ic,real-wor1d networks.
Results suggest that knoLvledge of dynamic stability limits is key for reli-
h i e control of modern pourer systems.These margins must be satisfactory < i t
all system buses for the bwying load demands typical over ;i23 hr period. 1'0
identify the iniirgins. an enhanced understanding of active pouw netMwks is
reqUired. Part i cUIarIy ii n portant are physica1 and iliatheniat i ciil clescription5 ()1'
the beha\,ioro f i1ctii.econtrol de\,icesthat f ul l y account for their l'ast-s\t.itchiIg
cIyniiniics.
There is a need for better analytic capabilities i n both the areas of systeii
planning and operations.Promising techniques include the use o f energy t'un:-
tion methods. which are based upon the use of LyapunoLP's direct method f1)r
assessing thestabilitjr of systenis of nonlinear differential equations.After inarrj'
j'earsof research. eriergjffunctions ;ire n o w successfully used to iissess systeiii
transient stability i n the planning context. and are on the ~' cr geof being uscd
on-linetodetermine the transient stability limits. Morerecently. energy metholls
1iac.ealsobeen employedt hr assessing the system'squasi-staticLroltage securitJ.
The relationship between the use of energy functions i n these tct'o different
contexts. ~b' i i sexaminedby Overbye et al. (I991).
The key points forenergy function usage in the transient stability problem
is that the energy is used to determine whether the postfault is b, i t hi nthe region
of iittrxtionof thepostfault stableequilibrium point (SEP).arid thatthe deternii-
nation o f the critical boundary of stability dependso n the f;iulted system trii-jec-
tor)'.
I n contrast. for the voltage security problem the system state is assumedto
\rarl' cliiasi-statically (i.e.,not subject to a large disturbance)i n responw to the
slots,(with;I time scale of min to hr) Lwi at i on i n sq'stem parameters, such ;is
the aggregate l ori ds. Lossof the system stability i nthis context \\wild occur I- o t
through ii large disturbance pro\-fiding the system state with enough"energ)." to
escape the SEPs energy "cvell." but rather through the coalescing of' the S I P
wfith one of the tj'pe-one UEPs contained on the boundary through ii sxltlle
node bifurcation iis the energy "u~ell"tlattens out.Because of the netxi to model
nonimpedance loadsandto retain the network topology. the SPMhas been used
excI Usive1y in voltagesecurity assessment.
I t has been proposed that the UEPenergy,which is ;i measure of the clis-
tancc betu.een the SEPand UEP, be used ;IS a measure for the sj'steni cliiirsi-
static \.oltagesecurity. Numerical testing has shown that the energy iiieasu~w
tend tochange i na manner proportional tothe changei nthe system's maximum
steady loadability. System steady-state voltage security could thus be asses>ed
bq,monitoririg these energjr iiie;isiires, with decreasing energq' \ralues i ncliciiting
increased vulnerability to voltage instability;a \ due of energy approaching zero
would indicate an imminent loss of system operating point.
There is,ho\lfeirer,a significant difference i n the use of these UEPenergies
from their transient stability use. I n transient stability the goal is to determine it
.siri,qlc' cr-itictrloiiorgj*W I I ~ M . which depends upon the fault-on system trti-jectory.
I n contrast with the steady-state Lroltage problem one is concerned urith main-
taining system voltage security while facingan imprecisely kno\bfnfuture system
parameter \*ariation.While a future voltage collapse ~ . o u l dbe characterized by
the coalescing of the SEPwith a particular UEP, there is no need to determine
precisely this UEP a priori. Rather the energies associated Lvith ;I number of
UEPs would be monitored. It is shown that each UEP energ>rcan be used to
quantify the voltage stability i n a particular portion of the system and that onl y
the small subset of UEPs with low associated energies need to be calculated.
This is because i n ii large system,there is a need to simiiltaneoiisl~~ monitor the
voltage security in a number of separate areas.
Tuvmajor similarities in theapplication of energ}'methods to the transient
and voltage problems is that the same energy formulation can be us\c.d i n both.
and that both make useof theenergy of the UEPson the boundary of the region
of attraction. In general for a three bus system, i t is shokvn that the po~verflo\v
equations have an SEP solution and another type-one UEP solution. Energj.
methods could then be used for transient analysis by determining if the energ~'
after a fault is cleared is less than that of a specific threshold.
Whether the system possesses other solutions depends on the load model
used. If the load is modeled as constant impedance then there are no other
solutions to the power flow equations, since the network is linear and could be
replaced by a Thevenin equivalent impedance. However. if the load is modeled
as constant power, then the system has two additional solutions.
Phase portraits can be used to illustrate a number of the mechanisms that
could cause the system to lose its stable operating point: and that \ s~oul dthus
need to be considered i n security analysis. For the transient analysis. stability
would be assessed by determining if the energy at the time the fault is clemxi
is less than 19". The \Aueof IY'depends on the exit point (i.e..the point \ b h x
the critically cleared trajectory would cross the stability boundary &) . Since
this value would typically be by a large generator angle. the exit point \vould
be i n the Lricinity of the transient stability UEP. A new use of the energy func-
tion would be for dynamic voltage security assessment. As i n the case of tran-
sient analysis, stability would be assessed by determining if the energj' after
somedisturbance is lessthan somecritical value. Howe\.er.thedistiirbance here
u~ouldbe one that utould tend to cause the system trajectory to exit the SEPs
UEP.Such a disturbance would be characterized by Io\v \.oltageniagnitu~ies (;is
opposed to large generator angles),and may hmre ;i substantialljr longer time
frame than the transient stability problem (onthe order of tnin asopposed to > ).
Examples of such disturbances are the effects of LTC transformers, generators
hitting reactive power limits, and load dynamics.
Quasi-steady voltage security could be monitored by noting the variation in
the UEP energy as the system state changes in response to the slow (on the
order of min to hr)variation in system parameters, such as the aggregate loacs.
The second segment of this book introduces a relatively new technolopy
and its direct applications and future to power system engineers. Over the p a t
few decades, the topic of artificial intelligence (AI)has gained the interest of
researchers andscientists inmany fields.Theapplications span manydisciplines
including electrical engineering extending its way to electrical power system
analysis, protection, and enhancement. The resulting advantages will inherently
depend on the specific applications of the existing methods that are available to
the engineering sector.
As discussed in prekfiouschapters of the book, the primary branches i n
artificial intelligence are Fuzzy Logic, Artificial Neural Nerworks, and Expcrt
Systems.Theobvious benefits of using Fuzzy Logic and/or Fuzzy Logic Expcrt
Systems include theenhancement oftheassessment criteria used i n the study of
large power systems, and also as decision support systems. This mode of 41
application asa support tool to the existing technique provides clearer solution
alternatives toproblems, inwhich casethere aremany variants with multi-levcls
of uncertainties. Therefore the assessment of the power system is impro\,edand
the appropriate control actions to be exercised in the event of contingencies are
more efficiently allocated, especially in a utility where the resources are either
limited or expensive to use. Furthermore, the use of Fuzzy Logic and/or Expxt
Systems i n power system stability studies can significantly reduce computa-
tional burden i n some algorithms, as the human or engineering judgment is
exploited i n determining redundancies of choice. The classic example iscontin-
gency analysis of a power system, whereby i t may not be necessary to consider
(111 possihle cases, but rather ci l l (sufficiently)pr-ohcihle cases.
The area of Artificial Neural Networks (ANN)has its place as a planning
and operation tool or support tool also, especially i n light of some of the more
recent restructuringdevelopments thathasoccurred inthepower industry.Thcse
changes, such as deregulation of the power industry, have been a result of
changes in the markets for electricity both as a corninodity and service. The
challenge in the application of ANN lies i n the unanswered questions that u'e
have come face to face with as both engineers and power system tools and
policy makers. For instance, what are the impacts of artificial intelligence sup-
port tools on power system reliability and stability'?Can thetechnology impro\.e
the desired goals of the classical integration techniques (Runge-Kutta,Modified
Euler,Trapezoidal, and soforth)making them better assessment methods'?How
can we perform detailed security assessnient of the p o u ~ r system isit11 the intro-
duction of this enhancement method?
I n conclusion. we have realized the capabilities and limitations of the exist-
ing methods utilized i n performing coniprehensi\re analysis on the ponrer s j ' s -
tem. both i n the areas of steady stateand transient stability studies. Maintaining
power system stability is one challenge, but i t is the llip-side of the coin that
motivates us to go to work, which are the problems associated \\!it11 ponwsj rs-
tern instability. The sekrerity of the faults resulting from Lvltage and/or angle
instability has led to the de\celopmentof many methods of monitoring and pre-
\renting these unwanted results. Such methods. some of nthich u w e highlighted
and discussed in detail in this book, arecurrently being enhanced \ ki the appli-
cation of state of the art technologies. The cost-benefit analysisof such tools is
yet to be done for large-scale power system applications. for both planning and
operational purposes. I n this light, continued research and the use of artificial
intelligence asanenhancement tothecomputational anddecisiontool nil1intro-
duce much benefit to the power industry.
Glossary
Acceleratingtorque: Thedifference between the inputtorquetothe rotor ai d
the sum of the load and losstorques; the net torque available for accelerat-
ing the rotating parts.
Activation function: Hidden units that are needed to introduce nonlinearity
into a neural network.
Active power: The average power in an electric power circuit is referred to
as the active power.
Admittance: The term admittance (symbol Y) is used in atransform network
(s-domainor complex, for sinusoidal steady state operation)to denote the
inverse of the impedance parameter. The current through an element is the
product of the element's admittance and the voltage across i t .
Angle stability: Stability of an electric power system evaluated on the ba:;is
of the analysisof thedynamics of therotor angle dynamics andthe couplcd
electric network.
Angularacceleration: The rate of change of angular rotor speed.
Angularspeed: The rate of change of the rotor angle.
Approximate reasoning: A computational rnodeling of any part of the pi.o-
cess used by humans to reason about natural phenomena.
ArtificialNeuralNetwork(ANN): Physical cellular systemsthatcan be used
to acquire, store. and utilize experiential knowledge.
Back propagation learning algorithm: The back propagation algorithm
modifies synapticconnection strengthswith nonlocal error information.The
algorithm propagates the instantaneous squared error backward from 1he
output through the hidden layers to the input at each iteration.
Back propagationor counterpropagationneuralnetworks: A neural net-
Lvork training method that uses the generalized delta rule that uses a gradi-
ent descent inethod to minimize the total squared error of the output coin-
puted by the net.
Capacitor bank: An assernbly at one location of capacitors and necess;irl'
accessories required for a complete operating installation.
GIossNI-! 299
Capacitive reactance: The imaginary part of the impedance of a predonii-
nantly capacitive element.
Certainty factors: Guesses or projections by the domain expert about the
relevance of evidence to support the recommendations passed by the infer-
enceengine of the rule-base system to the user.
Clearingangle: The angle corresponding to the time elapsed from the begin-
ning of overcurrent to the final circuit interruption.
Conductance: The following are not equivalent but are supplementary: (1)
The real part of admittance and (2)the physical property of an element that
is the factor by which the mean square voltage must be multiplied to gi\.e
thecorresponding power lostasheat orasotherpermanent radiation orloss
of energy from the circuit.
Continuationpower flow: The Jacobian matrix of the power flow computa-
tion becomessingularat thevoltage stability limit.Continuation pourerflow
overcomes this problem by using locally the parametrized continuation
method involving predictor and corrector steps.
Convergence: Foran iterative process,convergence means that thesuccessive
differencesbetween subsequent values ofthe unknown values ha\*e progres-
sively become closer and are below acertain predefined convergence toler-
ance threshold.
Counterpropagation: A neural network means thatcombine an unsupenised
Kohonen layer with a teachable output layer.
Crispsets:Thefuzzy set term for traditional set theory.That is. whether ornot
an object belongs to aset.
Criticalclearingangle: Forasystem of one machineconnected to an infinite
bus. and for a given fault and switching arrangement, the critical clearing
angle is that switching angle for which the system is at the edge of insta-
bility.
Current: A generic term used where there is no danger of ambiguitjt to refer
to anyoneormore currents specificallydescribed. The use of certain adjec-
tives before "current" is often convenient, as in conducting current.
Database: A collection of clauseswith eachclauseon adatabase representing
something that is known to be true.
Decisiontrees: Toolsformaking number-based decisionswhere a lot of coin-
plex information needs to be taken into account.
Directmethod: Thedirect method is usually referred to asLyapunoi. second
method of stability assessment, which attempts to determine stabilitj b j
using a suitable Lyapunov function.
Direct-axis: The axis that represents the direction of the plane of symmetry
of the no-load magnetic flux density, produced bq' the main field \+inding.
normally coinciding with the radial plane of symmetry of a field pole.
Direct-axis component of armature current: The component of the arnia-
titre current that produces a magnetomotive force distribution that is sy11-
metrical about the direct axis.
Direct-axiscomponentofarmaturevoltage: Thecomponent of thearmatlire
i.oltage of any phase that is i n time phase with the direct axis coniponc!nt
of current i n the same phase. A direct axis component of t dt age may be
produced by: (I ) rotation of the qiiadrature axis component of rnagnctic
flux, ( 2) w-iation ( i f any)of the direct axis component of magnetic fliix.
and ( 3) resistance dropcaused by the tlow of the direct axis component of
armature current.
Direct-axiscomponent of magnetomotiveforce: The component of rnagiie-
tomotit'eforce that is directed along the direct axis.
Direct-axis current: The cUrrent that prodUc es direct -axis magnetom otI \.c
force.
Direct-axis magnetic-flux component: The magnetic-flux component di-
rected along the directed axis.
Direct-axissubtransient impedance: The magnitude obtained by the vector
addition of the value for the armature resistance and the value tor diicct
axis subtransient reactance. The resistance ~ a l u e to be applied i n this c;iw
will be ;I function of frequency depending on rotor iron losses.
Direct-axis subtransient reactance: The quotient of the initial \Aue of ;I
sudden change in that the fundamental alternating currcnt coniponent of
armature cmoltage,that is produced by the total direct-axis priiiiary flux. vind
the \,alueof this simultaneous change i n fundumental alternating curtcnt
component o f direct axis armature current, the machine running at r; ted
speed.
Direct-axissubtransient voltage: The direct-axis component of the term nal
\,oltagc\+rhichappears iinmediately after the suddenopeningof theextoi'nal
circuit when the machine is running at a specified load, before any i'lux
variation in the excitation and damping circuits has taken place.
Direct-axissynchronous impedance: The magnitude obtained by the \w:tor
addition of the \ d u e for armature resistance and the \lalue for direct ;\xis
sytichronous reactiince.
Direct-axissynchronousreactance: The quotient ofa sustained \Aue of that
fundamental alternating current component of armature \ dt age that is pro-
duced by the total direct axis flux due to direct axis armature current and
the vdueof fundamental alternating current component of this current. the
machine running at rated speed. Unless otherwise specified, the V; IIUC of
synchronous reactance will be that corresponding torated armaturecurient.
For most machines, the arrnature resistance is negligibly small compared to
the synchronous reactance.
Direct-axistransient impedance: The inagnitude obtained by the \rector xi-
30I
dition of the value for armature resistance and the value for direct axis
transient reactance.
Direct-axistransient reactance: Thequotient of the initial dueof asudden
change in that fundamental alternating current component of armature \,()It-
age, which is produced by the total direct axis f l ux, and the value of the
simultaneouschange infundamental alternating current component ofdirect
axisarmaturecurrent.Themachine isassumed to be running at rated speed.
Direct-axis transient voltage: The direct axis component of the armature
voltage that appears immediately after the sudden opening of the external
circuit when running at a specified load. neglecting the components that
decay in the first few cycles.
Direct-axisvoltage: Thecomponentof voltage that would produce direct axis
current when resistance is limited.
Distributionsystem: The distribution system is the final stage i n the transfer
of power to the individual customers.
Dynamic response: An output expressed asafunction of time resulting from
the application of aspecified input under specified operating conditions.
Dynamic security assessment: The process of using tools of transient stabil-
ity analysis in predicting the vulnerability of apower system to contingen-
cies.
Electricpowersystem: Thefunction ofanelectricponwsystem istoconvert
energy from one of the naturally available forms to the electrical form and
to transport i t to the points of consumption.
Equalareacriterion: For arotor that isaccelerating. the condition for stabil-
ity isthat a maximum valueof the rotor angle exists and that thearea under
accelerating power versus rotor angle curve is zero up to that maximum
rotor angle.
Excitationcontrol: Thefunction of excitationcontrol istoregulate thegener-
ator voltage and reactive power output.
Excitation system: The equipment providing field current for a synchronous
machine, including all power, regulating. control, and protective elements.
Expert systems (ES): A computer program that processes problem-specific
information i n the working memory with a set of rules contained i n the
knowledge base system,using an inferenceengine to infernew information.
Fault: Aphysical condition that causes adevice, acomponent, or an element
to fail to perform in a required manner. forexample,a short-circuit. a bro-
ken wire, an intermittent connection.
Feedback: The return of a fraction of the output to the input.
Feeder: A set of conductors originating at a main distribution center and sup-
plying one or more branch circuit distribution centers. or any combination
of these two types of equipment.
Fictitious voltage: Voltage terms introduced i n transient equations to explain
the fl ux inertia linkages created by rapid changes in conditions external to
the machine (seeTransient equations).
Flux linkages: The sum of fluxes linking the terms forming the coil.
Forward chaining: The process used in an expert system for deriving new
information from known information. In forward chaining of rules, allrules
whose pre-conditions are fulfilled by the data set of the given observations
and derived intermediate results are eventually triggered.
Frequency: Thenumberofcompletecyclesof sinusoidalvariationperunittime.
Frequency response: A characteristic, expressed by formula or graph, which
describes the dynamic and steady-state response of a physical system i n
terms of the magnitude ratio and thephase displacement between asinuso -
dally krarying input quantity and the fundamental of the corresponding 011-
put quantity asa function of fundamental frequency.
Fuzzy expertsystem: An expert system that uses acollection of fuzzy me w
bership functions and rules, instead of Boolean logic, to reason about datii.
Fuzzy logic: Fuzzy logic deals with ambiguity i n defining various variabks
involved in describing the operation of a system.
Fuzzy logic systems (FLS): These are systems which mathematically modl:l
complex relationships that are usually handled in a vague manner by 1a11-
gUage or 1inguisti cs.
Fuzzy quantization: A mathematical means of describing vagueness in l i i i -
guistics. The membership functions defined on the input variables are ap-
plied to their actual \ due s in order to determine the degree of truth for the
premise of each rule.
Fuzzy reasoning: A fomal mathematical procedure for the representation ~ ) f
uncertainty that is used in the management of real systems.
Fuzzyset theory: A mathematical formulation developed by Dr. Lotfi Zadvh
of UUBerkeley in the 1960sasameans tomodel the uncertainty of natural
language. The theory is used to solve problems that contain vagueness 'or
uncertainty in the representation of knowledge and Pactual statements.
Fuzzy subset: Fuzzy subset F of a set S can be defined as a set of ordercd
pairs. each with the first element from S,and the second element from the
interval [O. I ] . with exactly one ordered pair present for each element of S.
Governor: Theassembly of fluid,electrical, or mechanical control equipment
used for regulating the tlow of water, steam. or other medium to the pr i m
mover for such purposes as starting, holding speed or load, or stopping.
IF-THEN-ELSERule: A basic knowledge representation by means of w-hich
directional relationships between objects can be represented using a pr e-
condition and anactionthat is often supportedby thecertaintyof theactic'n.
Impedance: The term impedance (symbol 2) is used in a transform netnwrk
(s-doinainor complex, for sinusoidal steady state operation)to relate cur-
rent through an element and the voltage across it, via the extended Ohm's
law v=ZI.
Inductance: The property of an electric circuit by virtue of which a varying
current induces an electromotive force in that circuit or a neighboring cir-
cuit.
Induction motor: An ac motor in which a primary winding on one member
(usuallythe stator)isconnectedtothe power sourceandapolyphase secon-
dary winding, or a squirrel cage secondary winding on the other member
(usually the rotor) carries induced current.
Inertiaconstant: Theenergy stored inthe rotor whenoperatingat rated speed
expressed askilowatt-seconds per kilovoltampere rating of the machine.
Inferenceengine: Thebrain of theexpert system asi t contains the knowledge
that is used to decide on how to apply the rules to infer new knowledge. I t
controls the firing of the rules in the expert system.
Infinitebus: A voltage source that maintains constant voltage and frequency
regardless of the load connected to it.
Intelligent systems: These are man-made systems that emulate functions of
living creatures and human mental faculties.
Iterations: Describesaprocess,which repeatedly executesasequenceof steps
until some condition is satisfied.
Knowledge-based system (KBS): A sub-program for solving problems, in
which the problem-solving methods and the knowledge are kept separate,
which makes i t amenable to modification by exchanging the knowledge
with the sameproblem-solving method and contributes to theexplainability
by enabling the knowledge used for deriving the problem solution to be
given.
Lag: Thedelay between two events.
Laplacetransform: Thequantity obtained by performing the operation:
where s =0+jm.
Learning algorithm: These are computational procedures that are generally
aimed at ill-defined and time-varying processes and use heuristics, adapta-
tion and pattern recognition techniques to construct neural networks and
rule-based systems.
Linearsystem: A system or element with the properties that if is the re-
+p)is the response to (.r,
+. r2), and ky , is the response to ks,.
( ! , then x2, is the response to y2 and .rI, sponse to
Load: Theelectric power used by d e ~ ice\ connected to an electrical generat-
ing sytetii.
Load angle: The angulardi\placement. at a specified load. of the center lire
of ;I field pole froni the axi\ of the armature magnetoriioti\e force pattern.
Load angle curve: A characteri\tic curve giving the relation\hip between tte
rotor di\placementangleand the load, forcon\tant value\ of wmature Iolt-
age, field current. and power factor.
Load center: A point at which the load of ;I gi\en area i \ a\\iinied to tx
concentratcd.
Load model: A mathematical representation ofthe \ariation of the acti\e ar~d
reactiLe power ~-equirement\ of the load at ii certain point i n the \j{teii
ttrith Loltagea nd frequency at that point.
Lyapunov function: A \calar differentiable function V(1 defined i n \oiiic
open region including X , suchthat in that region:
Man-machineinterface: A niedium through which theuserof theexpert s j s-
tem is ableto communicateufiththe computerallouringdata transfer i n t ic
foriii of input data,reports, systetii tiiainten;i 11ce,and resU1t s.
hlechanical power: The power imparted by the prime m o ~ wto the synchio-
nous generator.
hlembership function: I t is a graphical representation ot' the magnitude ot
participation of each input tothe fuzzy system andby computingthc 1ogic.d
product of the membership weights for each acti\re rule, a set of fuc. /j'
011t put respotise iiiagni t iides areproduced.
hleta-knowledge: Thek~iou~ledge that ;I systemhas about its o ~ . ' i i kiiou.lcdgc
that characterizes its ou'ti workings. This could include probabilities ot' w-
chi tectU r;iI moduIe,effici etici es or dependabiIities.
hlonotonic reasoning: A deduction i n the theory of classical logic that c;in
be proven to be valid under all conditions for all time.
Neural network: Neural networks consist of numerous, simple processing
units or "iieurons" that we can globally program f or computation. Nc~ir;il
tiet uvrksare triiitied to store, recognize. and ;issociati\reIy retrie\re patt eims
or database entries, to sol\.eproblems in\~olving estimating sampled t'uiic-
tions when we ci o not know the form of the functions.
Neuron: The neuron is the fundamental unit of the ne r \ ws system,particu-
larlj. the brain. The neuron is ii simple processing uni t that receikw ;nd
c w it1bities sigti;iIs fro111 inanq'othertieuroils throiigh fiIii me11tarj ' i11pitt paths.
Newton's method: An iterative method for finding the solution to a set of
nonlinear equations using first order derivatives.
Nonlinearsystem: Asystem described by asetof nonlinear equations. I n this
case the superposition principle does not apply.
Perception learning rule: An iterative learning procedure that usually con-
verge to the correct weights that will ultimately give the correct output for
all input training patterns.
Phase angle: The measure of the progression of a periodic nave i n time or
space from achosen instant or position.
Positive definite: A quadratic form V( x)=.Y' QX is positive definite if \'(.v) >
0for all values of .Y different from zero.
Powerfactor: Thecosineof theanglebetween the current through ;in electric
circuit element and the voltage across i t is referred to as the element's
power factor.
Powerflow equations: The system of equations relating the active and tmc-
tive power injections at each bus in an electric network i n terms of voltage
magnitudes and phase angles.
Power system stability: The term is used to define the ability of the bulk
power electric system to withstand sudden disturbances such as electric
short circuits or unanticipated loss of system components. For further de-
tails see Sec. 1.3.
Prime mover: The machine used to develop mechanical horsepomw neces-
sary to dri\ve ;I generator to produce electrical pourer.
Primemovercontrol: Thefunction of prime mover control is to regulate the
speed and control energy supply variables such boiler pressures, tempera-
tures, and tlows.
Quadrature axis transient impedance: The operator expressing the relation
between the initial change in armature voltage and a sudden change i n
quadrature axis armature current component, under the following assump-
tions: ( 1 ) Only the fundamental frequency components considered for both
voltage and current. (2)Nochange in the voltage applied to the field Liyind-
ing. ( 3 )The rotor running at steady speed.(4)Considering only the slokt'est
decaying component and the steady statecomponent of the \voltage drop.If
no rotor winding is along the quadrature axis and/or the rotor is not made
out of solid steel, this impedance equals the quadrature-axis synchronous
impedance.
Quadrature axis: The axis that represents the direction of the radial plane
along which the main field winding produces no magnetization. nornially
coinciding with the radial plane midway between adjacent poles. The posi-
tive direction of the quadrature axis is 90"ahead of the p0sitii.e direction
of the direct axis, in the direction of rotation of the field relati1.e to the
armature.
Quadrature axis component of armature current: The component of the
armature current that produces a magnetomotive force distribution that is
symmetrical about the quadrature axis.
Quadrature axis component of armature voltage: The component of the
armature voltage of any phase that isin time-phase with thequadrature axis
component of current in the same phase. A quadrature axis component of
voltage may be produced by: ( I ) rotation of the direct axis component of
magnetic t l ux, ( 2) variation ( i f any) of the quadrature axis component
of magnetic t l ux, (3)resistance dropcaused by the flow of the qiiadratuie
axiscomponent of armature current.
Quadratureaxis current: The current that produces quadrature axis magnc-
tomotive force.
Quadrature axis magnetomotive force: The component of a magnetomo-
tive force that is directed along an axis in quadrature with the axis of tt-e
poles.
Quadratureaxissubtransientimpedance: Theoperator expressing the rela-
tion between the initial change in armature voltage and asudden change . n
quadrature axisarmaturecurrent, under thefollowing assumptions: (I )On y
thefundamental frequencycomponents considered forboth voltage andcur-
rent. ( 2) No change in the voltage applied to the field winding. (3)The
rotor running at steady speed. If no rotor winding is along the qiiadratu-e
axis and/or the rotor is not made out of solid steel, this impedance equals
the quadrature axis synchronous impedance.
Quadratureaxissubtransientreactance: The ratio of the fundamental cori-
ponent of reactive armature voltage, due to the initial value of the funda-
mental quadrature axis component of the alternating current component q.>f
the armature current, to this component of current under suddenly applicd
balanced load conditions and at rated frequency. Unless otherwise stated,
the quadrature axis subtransient reactance is that corresponding t o ratcd
armature current.
Quadrature axis subtransient voltage: The quadrature axis component of
the terminal voltage that appears immediately after the sudden opening of
the external circuit when the machine is running at a specific load, before
any t l ux variation in the excitation and damping circuits has taken place.
Quadrature axis synchronous impedance: The impedance of armature
winding under steadystateconditions where theaxisof thearmaturecurrent
and magnetomotive force coincides with the quadrature axis. I n large syn-
chronous machines where the armature resistance is negligibly small. tne
quadrature axis synchronous impedance isequal tothe quadrature axis syn-
c hronous reactance.
Quadratureaxissynchronousreactance: The ratioof the fundamental coin-
ponent of reactive armature voltage, dueto the fundamental quadrature axis
307
component of armature current, to this component of current under steady-
state conditions and at rated frequency. Unless otherwise stated, the \,alue
of quadrature axis synchronous reactance will be that corresponding to the
rated armature current.
Quadrature axis transient reactance: The ratio of the fundamental compo-
nent of reactive armature voltage, due to the fundamental quadrature axis
component of the alternating current component of the armature current, to
this component of current under suddenly applied load conditions and at
rated frequency. The value of current is determined by extrapolating the
envelope of the alternating current component of the current wave to the
instant of the sudden application of the load,neglecting the high-decrement
current during the first few cycles. Note that usually the quadrature axis
transient reactance equalsthequadrature axissynchronous reactance except
in solid-rotor machines, since in general there is no really effective field
current in the quadrature axis.
Quadrature axis transient voltage: The quadrature axis component of the
armature voltage that appears immediately after the sudden opening of the
externalcircuit when running ataspecifiedload,neglectingthecomponents
that decay in the first few cycles.
Quadrature axis voltage: The component of voltage that would produce
quadrature axis current when resistance is limited.
Reactance, effective synchronous: An assumed value of synchronous reac-
tance used to represent amachine in asystem studycalculation forapartic-
ular operating condition.
Reactivecapabilitycurve: Acurve that identifiesthe reactive capability lim-
its of a synchronous machine based on three considerations: ( 1 ) Armature
current limit; ( 2) field current limit;and (3)end region-heating limits.
Reactive power: The reactive power Q is defined as the square root of the
square ofthe apparent power S minus the square of the active power P.
Reactive power is developed when there are inductive, capacitive, or non-
linear elements in the system. It does not represent useful energy that can
be extracted fromthe systembut itcancause increased lossesandexcessiire
voltage peaks.
Region ofstability: A region in the state space, where the system is stable.
Regulator: An electric machine regulatorthatcontrols theexcitation of asyn-
chronous machine.
Regulator, excitation system: A regulator that couples the output variables
ofthesynchronous machinetotheinputof theexciterthrough feedback and
foruwdcontrolling elements forthe purpose o f regulating the synchronoi~~
machine output variables.
Resistance: The physical property of an element, deLfice,branch, netuwk. o .
system that is the factor by urhich the mean squareconduction current must
be multiplied togive thecorresponding power lost by dissipation asheat o r
asotherpermanent loss of energy from the circuit.
Simulation: The representation of the functioning of one system by another-.
for example to represent a physical system by ;i mathematical model.
Slip: The difference between the synchronous speed and the actual speed of , I
rotor to the synchronous speed.
Soft computing: A term coined by Zadeh to refer to the emerging area of
coinputational intelligence such as fuzzy logic, neural networks. geneti,:
aIgori th111s. and expert syste111s.
Stability: An aspect of system behavior associated with sj,stemsha\!ing th:
general property that bounded perturbations result in bounded perturbation?,
i n the output.
Stability,absolute: Global asymptotic stability maintained for all gikren not-
linearities.
Stability, Lyabunov: For ii solution $(.\-(t , ) ) , t ) ,Lyabunov stability means that
forevery gi\.enE >0there exists a 6>0such that liAv(t,,)liI 6 implies I~ASI]~
5E for t 2t,,.
Stability, powersystem: In a system of two or more synchronous machints
connected through an electric network, the conditions i n which the diffe* -
ence o f the angiihr positions of the rotors of thc machines either rernairI?,
constant whi1e not subjected to a distUrbatice,or becomes constant fo1Iov1 -
ing an aperiodic disturbance.
Stabilizer: An excitation system stabilizer isan element or groupof elenienis
that modify the forward signal by either series or feedback compensatic11
to improve the dynamic performance of the excitation control system.
Stabilizer, powersystem: An element or group of elements that provide ;.II
additional input to the regulator to improve pouwsystem djrriamic perfor-
i i i ~ i nce.
Static: Refers to a state in which a quantity exhibits no appreciable chanlre
within an arbitrarily long time interval.
Securityassessment: I nanoperational environment, securityassessment co1-
sistsof predicting the vulnerability of the system to possible disturbance!,.
Steadystate: That in m!hichsome specified characteristic o f a condition such
asvalue,rate.periodicity,ormplitudeexhibitsonly negligible changeover
:in arbitrarily long time.
Steadystatecontingencyanalysis: Steady statecontingency analysispredict x
power tlowsand bus voltage conditions following events such as transmis-
sion line outages. transformer outages and generator outages.
Steadystatestability: A condition that exists i n apower system i f i t operates
with stability when not subjected to an aperiodic disturbance. I n practice, a
variety of relatively small aperiodic disturbances may be present without
any appreciable effect upon the stability. This is \.did as long as the resul-
tant rate of change in load is relatively slow in comparison mi t h the natural
frequency of oscillation of the major parts of the system or\,it11 the rate of
change in field flux of the rotating machines.
Steadystatestabilitylimit: The maximum power tlow possible through sotne
particularpoint i nthe systemwhen theentire systemorthe part of the sjfstem
to which the stability limit refers is operating with steady state stability.
Subtransient current: The initial alternating component of armature current
following a sudden short-circuit.
Subtransient Reactance: The reactance of a generator at the initiation of a
fault.
Subtransmissionsystem: Incontrast tothetransmission system.the subtrans-
mission systemtransmits power at alower voltage and in sinallerquantities
from the transmission system to the distributions substations.
Susceptance: The imaginary part of admittance.
Synchronous generator: A synchronous alternating current machine that
transforms mechanical power into electrical po~ver.
Synchronous machine: A machine in which the a\rerage speed of normd
operation is exactly proportional to the frequency of the system to which i t
is connected.
Synchronous reactance, effective: An assumed due o f synchronous reac-
tance used to represent a machine in a system study for a particular operat-
ing condition.
Synchronous speed: The speed of rotation of the niagnetic t l u s produced bjr
or linking the primary winding.
System generation control: The function of system generation control is to
balance the total system generation against system load and losses sothat
the desired frequency and power interchange with neighboring systems is
maintained.
Threephasesystem: A combination of circuitsenergized by alternatingelec-
tromotive forces which differ in phase by one third of acycle (120").
Timedelay: A time interval which ispurposely introduced i n thepert'ormance
of afunction.
Torque: The vector product of force and moment arm and is uridelj. desig-
nated by the unit Newton meter.
Transient current: The current under nonsteady conditions. Also. the alter-
natingcomponent of armaturecurrent immediately follontinga sudden short
circuit, neglecting the rapidly decaying component present during the first
few cycIes.
Transient energy function: This is a Lyapunov function used in the direct
method for transient stability analysis.
Transient equations: Used whenever conditions in the modeled system
change faster than astatic representation of voltage andtlow can acconim.)-
date. Under such conditions. t l ux inertia means these rapid changes cannot
establish themselves throughout the machine immediately, and two new
fictitious \dtages, usually referred to as E(;and E(;must be introduced to
represent the tlux linkages in machine rotor winding, and used in the
model equations.
Transient reactance: The reactance of a generator between the subtransient
and synchronous states.
Transmissioncontrol: These include power and voltage control devices s1iC-h
as st atic VAR co111pensators, synchronous condensers. swi tched ca pacitc1rs
and reactors, tap-changing transformers, phase shifting transformers. aid
HVDC transmission controls.
Transmissionline: A line used for power transmission.
Transmissionsystem: Atransmission system interconnects all major gencr-
ating stations and main load centers in an electric power system.
Turbine-generatorunit: An electric generator with its driving turbine.
Voltage: Voltage is synonymous with potential difference between two wn-
ductors.
Voltagecollapse: It is the process by which the sequence of events accompa-
nying voltage instability leads to a low iinacceptable voltage profile i r ;I
significant part of the power system.
Voltage regulator: A synchronous machine regulator that functions to main-
tain the terminal voltage of a synchronous machine at a predeterniircd
\ due. orto vary i t according to a predetermined plan.
Voltage stability: A voltage stability study evaluates the ability of a pov,w
systemtomaintain acceptable voltages at allnodesunder normal conditions
and after being subjected to contingency conditions.
Voltagestability indexing: Many indices characterizing the proximity of an
operating state t o the voltage collapse point have been developed. The 11e-
generacy of the load flow Jacobian matrix is used as an index of the po\b'er
system steady state stability.
Appendix:ChapterProblems
PROBLEMSFORCHAPTER2
Problem2.1
A single-phase sourceisconnectedtoatwo-terminal, passive circuit with equiLt-
dent impedance measured between the terminals given by:
z=(2.03 2 . 0 ) R
The source current is i(t) =4cos o tkA.Determine:
a. The instantaneous power,
b. The real power, and reactive power delivered by the source, and
c. The source power factor.
Problem2.2
Consider a single-phase load with an applied voltage:
\ , ( I ) =120cos ( Wt +10")volts
and load current
i(r) =8cos (cot- S0")A.
Determine the power triangle.
Find the pourer factorandspecify Lvhether i t is laggingor leading.Calculate
the reactive poufersupplied by capacitors i n parallcl u'ith the load that correci4
the pourer flictor to 0.9lagging.
Problem2.3
A circuit consist5 of t uo impedances, Z, =20 30"Q and Z;=30 -4Soi2i n
parallel, supplied by ;i source boltage V =120- 6O"\olts. Determine the poc ~c r
triangle for each of the impedances and for the wiircc.
Problem2.4
An inclustrial plant consisting primarily of induction motor loads absorbs 1000
hW at 0.75power factor lagging.
;i. Compute the required kVA rating of ii shunt capacitor t o improkre the
poner factor to 0.9lagging.
b. I f a synchronou\ motor rated 1000hpu i t h 90%efficiency operating at
rated load and at unity p o ~ e r frictor i \ added to the plant instead of't' ie
capacitor, c;I IcLI1;itc thc res111ti ng pourer I'actor. A\\11me con\ t ;int \oIt-
age.( 1 hp=0.746kW)
Problem2.5
The real power delicered by ii soiirce to t wo impedances. Z,=(3. 0+ j S . O ) i2
and Z:=10.0i 2connectedi n parallel, is IS00 W. Determine:
i i . The red power absorbed by eachof the impedances.
b. The sourcecurrent.
Problem2.6
A \inglt.-phaw \oiirce ha5 a terminal voltage V =120Lwlts rind a current 1 =25
L -W A , Lvhich Iea\e\ the positibe terminal of the \ ~i i r ce.Determine the r:al
and reacti\epower,and\tate 1% hether the sourcei \ deliveringor absorbingu : h .
Problem2.7
A sourcesupplies p ~ ~ e r to the following three loads connected i n parallel: ( 1 )
;Ilighting load draLving I0kW,( 2 ) an induction motor drauing 10kVAat 0.90
power factor lagging, and (3)a synchronous motor operating at 10 lip, 8Sck
efficiency and 0.95power factor leading ( 1 hp=0.746M) .
Determine the real. reactive, and apparent pouferdelivered by the soi~rce.
Also, draw the source power triangle.
Power2.8
Three identical impedances of (26+j l 5) 52 are connected in wye t o a 460 V
balanced three-phase source. Determine:
a. The magnitude of the line currents.
b. The total power dissipated forthe three phases.
Problem2.9
Three identical impedances of ( 18+j 22. 5) 52 are connected i n NI T to a 550 V
balanced three-phase source. Determine:
a. The magnitude of the line currents.
b. The total power dissipated for the three phases.
c. The total reactive power, and
d. The power factor.
Problem2.10
Repeat Problem 2.9 for the case where the three impedances are connected i n
delta.
Problem2.11
Current, voltage, and power to a balanced three-phase circuit are meawred and
found to be 20 A, 550 V, and 10.5kW. respectively. Determine equi\.alent cir-
cuits per phase as follows:
a. Wye-connected, seriescombination of resistance and reactance i n each
phase.
b. Delta-connected, parallel combination of resistance and reactance i n
each phase.
Problem2.12
Voltage, apparent power, and power to a balanced three-phase circuit are mea-
sured and found to be 460 V, 50 kVA, and 48.5 kW respectively. Determine
equivalent circuits per phase as follows:
a. Wye-connected, parallel combination of resistance and reactance i i i
each phase.
b. Delta-connected, seriescombination of resistance andreactance i n each
phase.
Problem2.13
The current, voltage. and power factor of a balanced three-phase circuit are mea-
sured and found to be IS A, 440 V, and 0.75 lagging respectively. Determinlz
equivalent series-connected resistance and reactance circuits per phase if th:
phases are:
a. Wye-connected.
b. De1ta-connected.
Problem2.14
Thevoltage,apparent power, and power factor of abalanced three-phase circuit
are measured and found to be 600 V, 150kVA, and 0.9leading respectivelj.
Determine equivalent parallel connected resistance and reactance circuits per
phase if the phases are:
a. Wye-connected.
b. De1ta-connected.
Problem2.15
Three impedances are connected in delta to a balanced 208 V, three-phase
source of sequence abc-.The impedances are
Z,,ll=I0+j20n
z,,, =2 0 - j l 0 R
z,,, =20+j 10 R
a. What are the three-phase voltages'?
b. Calculate the three-phase currents.
c. Calcualte the three-line currents.
Problem2.16
Three impedances are connected in delta to a balanced 460 V, three-pha?e
source of sequence cdx. The impediances are
3 I5
Z(,/, =2s+j 15 R
Z,,=17-j18 R
z,,, =20+j 20R
a. What are the three-phase voltages?
b. Calculate the three-phase currents.
c. Calculate the three linecurrents.
Problem2.17
Two loads are connected to a 460 V, three-phase balanced source. One is a
three-phase motor connected i n delta and running such that the power is25 kW
with a linecurrent of 35A.Thepower factoris known to be lagging.Theother
is a single-phase 10 kW heater that takes a uni t y factor current of 22 A when
connected between lines b and c. Using V,,,,as a reference. determine the three
line currents.
Problem2.18
Two loads are connected to a 460 V, three-phase balanced source. One is a
three-phase balanced load connected in wye and haLting a line current of 20 A
with a power factor of 0.9(lagging).The other is a single-phase load which has
a current of 15 A at a power factor of 0.7 (leading)when connected between
lines a and c.Using asa reference, determine the three line currents.
Problem2.19
Two loads are connected to a 208 V, three-phase balanced source. One is a
three-phase motor connected in delta and running such that the line current is
10A with a power factor of 0.866(lagging). Theother is a single-phase heater
which takes a current of 15 A at a power factor of 0.98 (lagging)ufhencoti-
nected between lines a and b. Using Vllbasa reference, determine the three line
currents.
Problem2.20
Two inductive loads are connected to a 460 V, three-phase balanced source.
One is a three-phase balanced load of 50kW connected in wye and having a
line current of 125 A. The other is a single-phase load of 5 kW and 10 kVA
connected between lines a and c.Using V,,,?as a reference. determine the three
line currents.
Problem2.21
Athree-phase induction motor isconnected to a balanced S S 0 V, 60Hzsupplq
For a particular mechanical load the input is 100kVA and 80 kW. The pouer
factor is tobe increased to0.95(lagging)bymeans of adelta-connected capaci
tor banh connected to the motor terminal\.Determine thecapacitance per pha\c.
required.
Problem2.22
When a certain three-phase induction motor is operated at its rated load thc
current, voltage,and power are 70A, 550V. and SO kW respectively. A seconcl
motor, when connected to the same source,takes acurrent of SO Aand apower-
of 30 kW. Normally both motors operate simultaneously. Assuming that thc
system frequency is 60Hz.
;I. Determine the delta-connected capacitance per phase required to raiw
the power factor to 0.95(lagging).
b. With thi\ value of capacitance remaining i n the circuit. determine thc
re\ulting power f'actorMhen the second motor is disconnected.
Problem2.23
A certain inductikre,balanced three-phase load dissipates 60hW with a currert
o f 66A L+,hen connected to ;I S S 0 V, 60Hz supply.
;I. Obtain the parameters of theequivalent wye-connected circuit in which
the reactance and resistance are connected in sereies.
b. A set o f three capacitors, each 500 CIF,is connected in \cries with the
load. Determine the current,voltage, and pouwof the original load.
c. Obtain the ow-al l power factor.
Problem2.24
A sq nchronous machitie hasthe following inductance\ associated uiththe xtator
windings:
f., =3.3+0. 05CO\ 2 6 mH
x
L,,,,=- I .6 - 0.05 CO\ [ 217 + 1 I ~ H
3
Use equations 2.26and 2.30to determine L,,and L,,i n Henrqrs
Problem2.25
A500MVA, 24kV, 60Hz three-phase synchronous machine has the f ol l o~~i ng
inductances in Henrys:
L,, =5 mH
L,,=4.5 mH
Determine the base impedance and inductance. and then find the per uni t
value of L,,and L,,.
Problem2.26
Asynchronous machine serves a load with:
v,=I i 10"/"I
1,=0. 5 ,-2O" /"I
Assume that X, , =1.2 and neglect the armature re4stance. Find the \aIue of
E' and 6based on the equivalent circuit of Fig. 2.10.
Problem2.27
Consider a static load represented by the model of Eq. (2.40). I t is hno\\n that
ri 10%increase i n voltage magnitude results i n a 15% increaw i n p o ~ er.Predict
the percentage change in power for a 10%decrease i n Ldtage.
Problem2.28
The model of Eq. (2.41)represents the dependence of the actiLre po\s-erof ;1
static load on voltage. The following measurements are available for incre;ises
i n the voltage magnitude and the corresponding increase i n active ponter:
AV% AP%
10 58.5
15 68.13
20 78.00
Predict the change in active power for a 10%decrease i n \>oltagc magni-
tude.
318
PROBLEMSFORCHAPTER 3
Problem3.1
Consider the exciter model given by Eq. (3.1)with K, =1.00,S,=0.8,and T,=
0.5 s. Assume that the voltage reference is ;i uni t step. Find the exciter voltage
output Y, as a function of time. Assume zero initial conditions.
Problem3.2
Find the steady state \Aueof the exciter voltage for Problem 3.1.
Problem3.3
Use the block diagram of Fig. (3. 5)for a typical stabilizer to establish a state
space model of the system.
Problem3.4
ASSLIIII~ that the input steps f or the stabilizer of Fig. (3.5)are given by AO =
0.1and AP! . Find t he steady state \ due of AV, . Assume that:
T, =0.I 7;=0. 05 7;=0 . 2
K, , =0.6 K,,=0.8
Problem3.5
Consider the model of Fig. (3.8)represented by Eq. (3.7).Write a state space
model for the system in terms of four states. Define the control input as:
Ldt)=CO,,,- CO,
Assume that:
T,, =0. 2 T, =5 K,=0.4
q,=0. 05 K, =5 K,,=0. 04
Find the response to a unit step input.
Appendix
Problem 3.6
Repeat Problem 3.5using a three state model involving xI ,s2,and q.
Problem 3.7
Consider the model of Fig. (3.9).Write a state space model for the system i n
terms of the two states. Define the control input as:
- u ( r ) =a,,,0,
Assume that:
T,;=0.2 K, =4 R,,=0.05
Find the eigenvalues of the system and the response to a unit step input.
PROBLEMSFORCHAPTER4
Problem 4.1
Given a 60 Hz, four-pole turbo generator rated 20 MVA, 13.2 kV, with an
inertia constant of H =7 kWs/kVA:
a. Calculate the kinetic energy stored in the rotor at synchronous speed.
b. Find the acceleration if the net mechanical input is 26,800 hp and the
electric power developed is 16MW.
c. Assume that the acceleration in part b is constant for a period of 10
cycles. Find the change in 6in that period.
If thisgenerator isdeliveringrated MVA at 0.8PFlag when afault reduces
the electric power output by 50%,determine the accelerating torque at the fault
time.
Problem 4.2
A 60 Hz alternator rated at 20 MVA isdeveloping electric power at 0.8power
factor lagging with net mechanical input of 18MW. Assume that acceleration
is constant for a period 15 cycles, in which 6 attains a value of IS" electrical
from zero initial conditions. Calculate the inertia constant H for the machine.
Problem 4.3
Show that the speed of a generator subject to a constant decelerating power of
1 p i 4 will be reduced from rated value to zero in 2 H seconds.
320
Problem4.4
A 20 MVA. 13.8kV, 60 HL.two-pole, Y-connected three phase alternator ha\
an armature winding re4stance of 0.07ohms per phase anda leakage reactancc
of 1.9ohms per phase. The ;irni;itiire reaction EMF for the machine i \ relatcd
to the ;irtii;itiire current by E,,, =- j l O. O 1 I ( / . Assume that the generated EMF i \
related to the field current by E, =60I , .
a. Coiiipiite the field current required to establish rated kwltageacrossthe
terminal\ of a load \+!henrated armature current i \ delikwed at 0.8 F'F
Iagging.
b. Conipiite the field current needed IO probidc rated terminal \ dt age o
;i load that draw\ 10074 of rated current at 0.85lagging.
Problem4.5
A 10 MVA. 13.8kV, 60 HL . two pole, Y-connected synchronous generator i \
deli\wing rated current at rated voltage and uni t y PF. Find the armature resi+
tance a1id \ynchro1ioii\ reactance gicren that the field excitation ~ol t agei +
1 1035.41V md lead\ the tcrminal cxdtage by an angle 47.96".
Problem4.6
A 1500hVA. three-phase, Y-connected,1I60 V. 10-pole,60 Hz synchronoti\
generator hasiin armatlirere4stanceof0.126ohmsper phawand ii synchronou\
reactance of 3 ohms per pha\e. Find the full load generated kroltageper phaw
at 0.8PFlagging.
Problem4.7
The nchronoii\ reactance o f ;i cylindrical rotor synchronou\ generator i \ 0. 90
p i . I f the machine i \ deli\eringactiirepowwof I .OOp i r to;in infinite bus M I N \c
koltage is 1.OO p i , at uni t ) PF, calculate the excitation vollage and the pover
angIe.
Problem4.8
A cylindrical rotor niachine is delivering actiLte ~ O M er of 0. 80pr and reactiLe
pouter of 0.60pi( at ;I terminal i ~l t a g e of 1.OO p i r . I f the power angle is 22'.
com pute the eYCit ;ition b v l t age and the machitie'\ \ynchronoi i s reactance.
Problem4.9
A cylindrical rotor machine is delivering active power of 0.80p i and reacti\,c
poiver of 0.60pi when the excitation voltage is 1.20p i and the p o \ +~r angle is
25". Find the terminal voltage and synchronous reactance of the machine.
Problem4.10
The reactances ~v, /and .v,/ of a salient-pole syiichronou\ generator :ire 0.95and
0.70per uni t . respectiLrely.The armature resistance i \ negligible. The generator
delivers rated kVA at uni t y PFand rated terminal Ltoltage. Calculate the c'wita-
tion iroltage.
Problem4.11
The reactances and .v,/ of a salient-pole synchronous generator are 1.OO and
0.60per uni t respectii.ely.The excitation voltage is I .77pi' and thc infinitebu\
\ dt ageis maintained at 1.OOp i . Forapower angle of 19.4". compute the acti\c
and reacti\'e power supplied to the bus.
Problem4.12
A \alient pole machine wpplies a load of I . 20pi' at i i ni t i \ PFto an infinite bu\.
The direct axi\ and q11adrat Ure axi\synchronous re;Kt;inctj.\ are:
t , / =0.9283 \,, =0.4384
The pourer angle 6is 25".Evaluate the excitation and terminal Yoltages.
Problem4.13
Consider the ca\eof an electric machine connected to an infinite bu\ through ;i
reacti\e electric net\horh wch that the magnitude of the po\ser anglc ci i ne i \
uni t y. A change i n the networh results i n a new power angle ciine~ i t h niagni-
tude t . Suppow the machine is delivering a powerp before the chiingc oc'c~ir\.
then \how that the maximum value of p such that the \ystem remain\ \table
satisfies:
Verify that for.v =0.5,the inaxirnum value of prefiiult po\srerp is approximatel>
0.4245.
Problem4.14
A generator is delivering 0.60of P,,,',,to an infinite busthrough a transmission
line. A fault occurs such that the reactance between the generator and the bus
is increased to three times its prefault value. When the fault is cleared, the
maximum power that can be delivered is 0.80of the original maximum valul:.
Determine the critical clearing angle using the equal-area criterion.
Problem4.15
A generator isdelivering 0.50of P,,,,,to infinite bus through atransmission 1in;t.
A fault occurs such that the new maximum power is0.30of the original. Whcn
the fault is cleared, the maximum power that can be delivered is 0.80 of the
original maximum value.
a. Determine the critical clearing angle.
b. I f the fault iscleared at6=75". find themaximum value of 6forwhicsh
the machine swings around its new equilibrium position.
Problem4.16
The power angle curk'es for a single machine against an infinite bus system is
P =2.8sin 6.Under fault conditions, the curve is described by:
P = 1.2sin 6
Assume that the system is delivering a power of 1.0p i i prior to the fault atid
that fault clearing results in the system returning to the prefault conditions. I f
the fault is cleared at 6,=60,would the system be stable'?Find the maximum
angle of swing 6,if the system is stable.
Problem4.17
A generator is delivering 0.55 of P,,,,,to an infinite bus through a transmissicm
line. A fault occurs such that the reactance between the generator and the bus
is increased to three times its prefault value. When the fault is cleared. t i e
maximum power that can be delivered is 0.75of the original maximum valutb.
Determine the critical clearing angle using the equal-area criterion.
Problem4.18
The 60Hzsynchronous machine shown in Fig. I isgenerating 235MW and 3 0
MVAr of power. The voltage at the infinite bus y is 1.O +jO.Op i , and the line
Busp Bus q
Figure1. Impediance diagram of samplepower system.
reactance is 0.065 1 x 1 on a 100MVA base. The machine transient reactance is
0.221711 and the inertia constant is 3.78per unit on a 100MVA base.
a. Solve the initial power flow of the system.
b. Using the Euler integration technique, calculate the changes i n phase
angle and speed of the generatorforathree-phase fault at bus11. which
is cleared after 0.06second. Use a time step size of 0.02second and a
total time of 0.I8seconds.
P,: generator real terminal power
Q,: generator reactive terminal power
f,,y:real power flow
Ql,ll: reactive power flow
E': voltage behind the transient reactance
V/,: voltage at busp
V(/: voltage at bus q
sI,': generator transient reactance
.I-/,(/: transmission line reactance
Problem4.19
a. Draw adetailed flow chart of the Modified Euler technique, asapplied
to power system stability studies.
b. For the problem given in No. 1, using the Modified Euler integration
technique.calculatethechangesinphase angleandspeed of thegenera-
tor for a three-phase fault at busp , which iscleared after 0.06second.
Use a time stepsizeof 0.02second and a total time of 0.18seconds.
Problem4.20
The 100MVA, 13.8kV, 60Hz synchronous machine has a transient reactance
of 0.02s p i r and an inertia constant of 4.00 per uni t on its own base. I t is
supplying ii pouferof 95 MVAat ;I pourer factor o f 0.895lagging to an infinite
bus.The kroltageat the infinite bus is I .O +jO.O 1711 and the net reactance of tlie
I parallel lines i \ 0.30~ I on a I00 MVAbase.
;I. Drau ;I \ingle line diagram of the \y\terii and \olie i t \ initial po~zci
tlou.
b. U\ing the RK- 3 (Runge-Kutta) integration technique to calculate tlie
change\ in phase angle and speed of the generator f or a three-phaw
1';iuIt at the generator bu\, which i \ cleared after 0.06 \econd. U\c ;I
tinit' step \izeot0.02 \econd and ii total time of 0.18second\.
Problem4.21
For the system given in Problem 4.3, consider the loss of one of the parallel
branche\ Mhow impedance is 0.06p i . If thi\ fault occurred at r =0.0\econd
md M;i\ cleared through breaher reclosurc at r =0.50 second, calculate the
chungc\ i n phase angle and speed of the generator from t =0.00 to t=0.I 2
\econds. U\e the Euler integration technique uith a time \tep \i7e of 0. 02
\e cond.
Problem4.22
where T, =electrical torque,
r,, =mechanical torque,
6 =angular speed of the machine. anti
K,.K2=constants that are functions of the network.
a. D e ~ ~ l o p irnplementation algorithms for solving the swing equation by
any two numerical methods.
b. Comment on the computations efficiency of the two algorithms i n ;i.
especially for large multi-machine power systems.
PROBLEMSFORCHAPTER 5
Problem5.1
Define and briefly discuss the following terminologies, as applied to the topic
of dynamic stability assessment (DSA)in electric polver sjtstenis.
a. Angle stability
b. Center of inertia, CO1
c. Stableequilibrium point, SEP
d. Unstable equilibrium point, UEP
e. Critical energy V, ,
f. Energy margin
Problem5.2
Consider a simple power system as shown in Fig. 2. consisting of a generator
delilrering power to a large system represented by an infinite bus through tlvo
transmission circuits. The single generator represents a thermal generating plant
consisting of four 600 MVA, 24.5 kV. 60 Hz units supplying pouw to the
infinite bus. Bus B is the infinite bus, which can be represented by a \ dt age
sourceof constant \,ohageinagnitiude and constant frequency.The initial opera-
ting conditions of the system. with quantities expressed i n pi( on ;i 2200 MVA.
24.5kV base. are asfollows:
P =0.953p i r (2=0.532p i
E, =I .o/"I at 15.50" E,,=0.935p1r ill O.OO@
The classical model of the generators has unitized parameters lumped to
that of a single equivalent generator given as: X,:=0.325p i and the machine
constant, H =3.06MW.s/MVA.
BusA BusB
j 0. 45
j0.83
EB
-
Figure2. Single line diagramof the powersystem for problem 5.2.
If a solid three-phase fault occurs at point F as shown in Fig. 2, and i!,
subsequently cleared by i . s ol dqqthe faulted circuit, then:
a. Write the dynamic equations for the postfault system referred to tht.
center of inertia. COI.
b. Write the expression for the \ystem energy function.
c. Calculate the postfault \ystem stable equilibrium point (SEP).unstablt:
equilibrium point (UEP),and the critical energy V, , .
d. Calculate the energy at f aul t clearing with t , =0. 08, 0.12. and 0.10
\econd.
e. Determine the system stability for each of the three faul t durations.
In this problem. the netuork reactances shown are on ii 2200 MVA base and
the resistances are assumed tobe negligible.
Problem5.3
Re-visit the data given i n Problem 5.2.Consider a lossof line contingency th: t
u.as created due to an open circuit fault on the branch of the parallel transmi>-
sion \ystem whow reactance is 0.83pi(. If the friult was subsequently cleareJ
by controlled breaker reclosure, then:
a. Calculate the postfiiult system stable equilibrium point (SEP),unstable
equilibrium point (UEP),and the critical energy V, , .
b. Calculate the energy at fault clearing wi t h t , =0.10. 0.20, and 0. 30
second.
c. Hence orotherwise. determine the system stability foreach of the thrte
fiiuIt diirations.
I n this problem, the netuork reactances shown are on a 2200 MVA base ard
the resistances iire assumed to be negligible.
Problem5.4
a. Discuss Lyapunov's stability criteria, as applied to . st di l i rj 9m.ses.snio' it
studies performed on electrical power systems.
b. Explain how thegeneralized expressionof the transient energy function
forii multi-machine power system can be usedtodetermine thestability
of the system.
c. What are the challenges faced by engineers and researchers in attempt-
ing to obtain fast and iiseful solution to the transient energy function ,)f
part b?Also cornrnent on the desired accuracy of the solution.
327
PROBLEMS FOR CHAPTER6
Problem6.1
Consider the 550 kV, 370 km (230miles) line transmission system shown i n
Fig. 3a below supplying power to a radial load from a 'strong' power system
represented by an infinite bus. The line parameters, as shown in Fig. 3b, are
expressed in their respective per uni t values on a common system base of 100
MVA and 55 kV.
(a) InfiniteBUS LoadBus
Bus I Bus2
Transmission Line
Shunt Load
Qsh
(b) InfiniteBus LoadBus
Bus I Bus2
Figure3. The550 kV. 370km (230miles)linetransmi\sion system \upplying aradial
load. ( a ) The whematic diagram. (b)The equivalent Uye circuit reprewntation of the
tranmission line.
6.1.1 Computethe f ul l admittance matrix of the two-bus system andwrite the
pou'er tlou'equations from the sending end to the receiving end in the torm:
6.1.2 Hence orotherwise, write down the expressions for the four ( 3 )\ubm;i-
trices of the Jacobian i n the l i near kd load tlow equation\ ;I\ defined by:
6.1.3 When P1=1600 MW. calculate the eigen\alue\ of the reduced Q-\' J I-
cobian niatrix and the \/-@ sen\iti\itie\ Mith the tolloning different reactilc
p o ~ er iii-jcctionsfor each ofthe corre\pondingt L j o L()Itage\on the Q-I'cur\c.
;I. QI=510 MVAR
b. (II=305 MVAR
c. Values of Q, close to the bottom of the Y-Qcurve.
6.1.-IDetermine the coltage stability of the systenibycomputingthe eigenLiiI-
lie\ ofthe reduced \"-QJacobian matrix for the follouiiig ca\es:
;I. P = 1600MW. Q, =360MVAR
b. P =1890MW. @, =965 MVAR
Assumc that the rerictikre pourer Q,is supplied by ;Ishiint capacitor
Problem 6.2
Thi\ problem demonstrate\ the concept of parameterization. Conde l - ;I \ i niplc
2-bus power \ y\ t em ;I\ \ houn in Fig.4,Mhere re are intere\ted i n the beh:t\ ior
on the bus \oltage,V to the amount 01' reactiLc' p o w r demand at the bu\ unclcr
prc\pecified conditions.
Busp Busq
TotalLoad
3.6-ti s
-
GiLren that s is the parameter of interest, which represents reacti\-epo\jer.
and the fo11owing physicaI1y-ineani ngfuI parame terization equations:
0=3.6 - V COS 8+10V sin 8 ....( i )
o=s +10 V' - vsin e- 10 vcos8 ....( i i )
6.2.1 Solvethe parameterization equations in Sec.6.2.I using \ Aws of.s from
0.00to .s,,,',, i n increments of 0.1001x1 MVAr.
6.2.2 What is the value of .s,,,,,, at which point no real solutionsareobtainable?
Plot the resulting lV(vs. s and 0vs. s traces on the same grid.
6.2.3 How safe is i t to deduce the maximum reacti\Te loridabilitj l i mi t of the
given 2-bus power system'?
Problem6.3
This problem demonstrates the basic concepts of Modril nnalysiis. a s upplicd to
voltage stability studies i n power systems.A 4 Bu s p v e r system is shown i n
Fig. 5 and the operating conditions are summarized i n Table 1 .
6.3.1 Compute the full and reduced Jacobian matrices of the system.
6.3.2 Calculate the eigenvalues and eigenvectors of the reduced Jacobian ma-
trix.
Bus I Bus3
212 =O.OOO+j0.095
5
234 =O.OOO+jO.3 15
Bus2 IBus4
~2~ =O.OOO +j0.055 L
Figure5. A -+-Buspm' er \ pt em for Modal anaiy\i\.
Table1 OperatingConditionsfor the +Bus
System
Complexbus voltages, V,
Voltage Voltage
magnitude angle
Bus no. i ' V,I ( p) 8,(deg)
I .ooo +o.ooo
I .000 -25.890
0.655 -38.980
0.795 -38.980
6.3.3 Determine the modal reactive power variation and modal voltage varii-
tion.
6.3.4 Determine the bus, branch, and generator participants.
Problem6.4
Using the algorithm for static assessment, solve the following system of alge-
braic equations represented by:
I . I . ~ ' - ] . I ? = 3.2 ..,.( i )
2. 2s+0 . 9 ~ =- k ....( i i )
where k in aparameter varying from k =0to k, , , , , .
Problem6.5
The synchronous machine shown in Fig. 6isgenerating 250MW and 85MVAr
of power.The machine transient reactance is0.200p i . and the line reactance i \
Bus I Bus 2
Figure6. The impedance diagram of sample powet system.
0.045p i r , both on a 100MVA base. The value of the voltage at the infinite bus
(node2)is I .OOO +j O. OOO pi !
6.5.1 Solvethe power flowof thesystem forboth low and high voltage power
solutions.
6.5.2 Construct the necessary vectors required by the VIPI Method and com-
pute this proximity index.
6.5.3 Consider a bolted three-phase fault occurring i n the middle of the trans-
mission line. Calculate the multiple power flow solutions (asin part 6.5.1).and
the voltage instability proximity index.
6.5.4 Comment on the VIP/ values obtained in 6.5.2and 63.3.
Bibliography
Abed, E.H. and Varaiya P. NonlinearOscillationsin PoLver Systems.I r i r . .I. Elect. Po~. or .
Ellc.r;L!\~ $\Yr., 6. I983,3 7 4 3 .
Adkins, B., 7 h c ) Grric.r-trlTlrrot-vof' Elet~ti. ic. rrlMmhi i i os . Chapman and Hall Ltd.. L - c i n -
don. 1962.
Aggoune. M..Atlas, L. E..Cohn, D. A.,El-Sharkawi, M. A.,and Marks, R. J . Artificial
Neural Networks for Power System StaticSecurityAssessment. I EEEI r i t . S?.rri/u)-
si rrri i or1 C' i ru~i t strrrtl Sys/ ei t i s. Portland,OR (May9-1 I 1989b)pp. 390494.
Aggoune. M., El-Sharkawi, M. A.. Park, D. C., Damborg, M. J.. and Marks 11, R. J .
PreliminaryResultson Using Artificial Neural Net\c,orksfor Securit),Assessnittit.
IEEEPr01..( ! f ' I YKY PICA.Seattle,WA. ( May 1989)pp. 252-258.
Alxac.0.and Stott,B.Optitiial Load Flow with Steady-StateSecurity,I EEETi.trri.v. /',\S.
Vol. 94 ( 1974)pp.715-75 I .
Al\,arado,F. L.Parallel Solutionof TransientProblemsbq'TrapemidalIntegriition,ILl i E
7rrrri.s.Pori.orApprr. S\>.vt..PAS-98,No. 3(Maj,/June 1989)p. 1080.
Atiderson. P. M.and Fouod. A. A. Po\t.t>r S!..storrr Sttrhilit!- crrrtl Cotrtr-ol. 4i ri t i s: loua Slate
University Press, 1977.
Antsaklis, P. J . Neural Networks in Control Systems. I EEE C ' o r i r r d Sj . . vf cvi i s Mt r gt r : i r i c
(April 1989)pp. 3-5.
Athay, T., Podmore. R. , ancl Virmani. S . A Practical Method for Direct Analyxi5 o f
Transient Stability,I EEE Trtrris.Pouvr- Apprr. S\,.vf.. PAS-98.2. (Mar. /Apr. 1979)
pp. 573-583.
Aylett, P. D.The Energy-Integral Criterion of Transient Stability Limits o f Power $04-
tems, Proc. of tho 1n.stitirtioii of Elrctricul Etigiiicv.r.s ( Loi i t l or i ) . IOSC, 8. (Sept.
1958)pp. 257-536.
Ajjarapu, V. and Christy, C. The Continuation Poww Fl o ~, : A Tool for Steady State
Voltage Stability Analysis, IEEE PICA Cor!fii.oriwPi.oc.c.c.rliri,s.s.(May 199I ) pp.
304-3 I I .
Barisilal. D. Ilukaram. and Partliasaradiy, K. An Expert System for Po\+?er System Volt-
age Stability Improvement, Electricul Po\rvr &. Eticr-gj.Sj:sterii.s, Vol. 19.no. 6.
1997.pp. 385-392.
Bartle. R. G. Tlici Elrrtirrits of R e d Aricilysis. John Wiley &Sons. Inc.,New York. 1076.
Bego\,ic,M.and Phadke, A.G.Dynamic Simulation of Voltage Collapse. IEEEPr wc ~wl -
irigs of PICA. Seattle. WA (May 1-5. 1989).
Bellman. R. and Giertz, M. On the Analytic Formalism of the Theory of Fuz1.y Sets.
lr$wtucitioti Scicwce, Vol. 5 , 1973.pp. 149-156.
Bergen, A. R. and Hill, D. J. Structure Preserving Model for Poww System Stability
Analysis. IEEE Trtiiis. P o i l ~ r A p p r . Syst. . PAS- 100. Vol. I . (Jan. I981 ) pp.
25-35.
Bergen, A, R. Analytical Methods for the Problem of Dynamic Stability, Pro(. . Iiitctmi-
tioriril Sjwiposiiini o r 1 Circuitsci i i d Sys t mi s , I 977.
Bergen. A. R. Poit'erSTsterii Aiicilysis,Prentice-Hall, Inc.,Englewood Cliffs. NJ, 1986.
Bergen, A. R., Hill, D. J.. and DeMarco, C.L.A Lyapunov Function for Multi-hlachine
Power Systems with Generator Flux Decay and Voltage Dependent Loads. lilt. J .
of Electric Poi c ~r . ciricl Eriergj,S!,st..8. I , (Jan. 1986)pp. 2- 10.
Berggren, B. Transient Behavior in a Stressed Power Sqrstem, MS Thesis. l o m ~ Statc
University. Ames. 1988.
Bhatia Neelu, G. Incorporation of Nonlinear Load Models and Identification of Inter-
area Mode PhenomenonintheTransient Energy Function Method. Ph.D. Dissertn-
tion. loura State UniLtersity.Ames, 1989.
Boratynska-Stadnicka. D. J., et al. Converting an Existing Computer Codc to a Hyper-
cubeComputer, Proceedings of the 199-9PICA, (May 1089)pp. 39-4-399.
Braae. M.and Rutherford. D. A. Selection of Parameters for a Fuzzy Logic Controller.
Fl i ~, \ 'Set.s( i t i d S!,.stet??.s,Vol. 2, NO.3,(1979)pp. 185-199.
Brasch, F. M.. Jr., et al. Design of Multiprocessor Structures for Simulation of Pourer
System Dynamics, Report No.EL-1756-1, Palo Alto. CA.: EPRI, March 19x1.
Brow'n. J. E., Kovacs, K. P., and Vas, P. A Method of Including the Effects of hkiin
Flux Path Saturation in the Generalized Equations of AC Machines. IEEE Ti.crri.v.
Pouv r Appcir. Syst.. PAS- 102, No. 1. (Jan. 1983)pp. 96- 102.
Canaqp. I . M. Causes of Discrepancies on Calculation of Rotor Quantities and Exact
Equivalent Diagrams of the Synchronous Machine, IEEE Tr(rii.v. Poi\.c.r- Appi r .
Sj'.st.,PAS- 88. NO.7, (July 1969).
Canay. I. M. Determination of Model Parameters of SynchronousMachines. IEE Pro(*..
130,Part B. No. 2, (March 1983)pp. 86-93.
Cardozo. E. and Talukdar,S.N. ADistributed Expert System for Fault Diagnosis. IEEE
Tr.tiri.scic*tiorisor1 Po\iver S y s t e m, (May 1988)pp. 641-6-46.
Carvalho. V. F. et al. Direct Analysis of Transient Stability for Large Power Systems.
EPRI Rrport No. EL-4980.December 1986.
Chandrashekhar.K. S.and Hill, D.J . CutsetStability Criterion for Power SystemsUsinp
a Structure Preserving Model, Iritrrrititioritrl Joirrritil cf Electric. Porrvr trrirl s\-.\ .
terrZ.s 8, NO.3 ( J ul y 1986)pp. 146-157.
Chang, C.S..Sutanto, D.,and Lachs. W. Automatic Control of Voltage Instability by ail
Expert System Utilizing Pattern Recognition Techniques, Proc.of'the TerithPor1.u1.
Sjsterii Couipiittrtiori Cor$ereric*e,Graz, Austria, (August 1990)pp. 1057-1064.
Chen, G. P. and Malik, 0.P. Tracking Constrained Adaptive Power System Stabilizer.
IEEE Proc*twlirig.so r i Geriertitiori. Trtirisrnissioiitiritl Distrilxitiori.Vol. 142.Nc .
2. (1995)pp. 149-156.
Chen. G. P.. Malik, 0.P.,Hope, G.S. , Qin. Y. H.. and Xu. G. Y. An Adaptive Power
S4'stein StiibiIizer Based on Se1f-0ptiinizing PO le-Shift ing ControI Strateg),,IEEl ;
Trrirr.stic~tiori.so t i Errrrgjv Cori\wsiorr, Vol. 8, No. 4. (1993)pp. 639-645.
Chen, S.T.. Yu, D. C.,and Moghaddainjo, A. R. Weather Sensiti\,eShort-Term LoaJ
Forecasting Using an Non-Fully Connected Artificial Neural Network, IEEi:
Trtiris.OI I Poltvr Sw..Vol. 7. No. 3. (Aug. 1992).
Cheng. S.-J., Chow, Y. S.,Malik, 0.P., and Hope. G. S. An Adaptive Synchronouh
Machine Stabilizer,IEEE Trmistictioiisor1 Porr~rSystems, Vol . PWRS- I . No. :#,
(1986)pp. 101-109.
Chiang. H. D. Analytical Results o nDirect Methodsfor Power System Transient Stabil-
ity Assessment, Corztrolt i r i d Dj wt i rrzi c Sj..sterri.s( C. T.Leoritlc~s, etl. ), 43,Academic
Press, San Diego, CA (199I ) 275-334.
Chiang,H.D.,Dobson. I., Thomas. R.J..Thorp,J. S. , and Fekih-Ahmed, L.On Voltage
Collapse i n Electric Power Syst em. Pr- ot w~t 1i ri g. sof'PICA,(May 1-5 1989)Sea-
tle, WA.
Chiang, H. D.. Wu. F. F.,and Varaiya. P. P. Foundations of Direct Methods for Poutbr
System Transient Stability Analysis. IEEE Trtiris. Circirits Sj,.st., CAS-24 (Feb.
1987)pp. 160-173.
Chow, J. H., ( 4 . ) . Time-Scale Modeling of Dynamic Networks uith Applications 10
Power Systems, Vol. 46 of' lectirre notes i r i Control trritl 1nti)rrmitiori Sc~i er i c~~\ .
Springer-Verlag, New York, 1982.
Chomr, M. and Thomas, R. J . Neural Networks Synchronous Machine Modeling, Pr o-
c.rrtlirig.sof tlir lY89ISCAS, Vol. I , Portland. OR (May 1989)pp.495498.
Christie. R. D. and Talukdar. S. N. Expert Systems for On-Line Security Assessment:
A Preliminary Design, IEEE Ti-ciristic.tiorrsor1 Port ~rLlelii~crj~, ( J ul y 1986)pp.
736-743.
Christie. S.,Talukdar,N.,and Nixon, J . C. CQR: A Hybrid Expert System for Securi.y
Assessment, Proc. 1989PICA,pp. 267-273.
Chua, L. 0.and Lin, P. M. Computer-Aided Analysis of Electronic Circuits, Prentic:-
Hall, Inc., Engl ebmdCliffs. NJ , ( 1975).
ClGRE Task Force 38-02- 10, Motli~lirigof' Volttige Colltrpse lridirtliri~q Djvitiriiic Plio-
1101l1e11t1, ( I993).
Concordia. C.,Synchronous Machines. John Wiley &Sons,New York. 1951,
Coultes, M. E. and Watson, W. Synchronous Machine Models by Standstill Frequent:y
Response Tests, IEEE 7'rtrri.s.Porivr Apptir. Sj'st.,PAS- 100. Vol. 4, (Apr. I981)
pp. 14x0- 1489.
Bibliography 33.5
Cutsem. T.V. AMethod toCompute Reactive Power Marginswith Respect to Voltage
Collapse, IEEE Trans.ott PohierSystems,Vol. 6,No.2 (Feb. 1991)pp. 145-156.
Cutsem,T.Van,Wchenkel,L.,Pavella,M.,Heilbronn, B.,Goubin, M.,and Gilliard. M.
DecisionTreesfor PreventiveVoltageStabilityAssessment,Proc.. 2 i d Int. Work-
shop 011Bulk,PokiverSystem VoltagePhetioriteit(r-V~)Itci,geSttrbilitj*( i d Srwrritjv,
McHenry, USA, (August 1991)pp. 217-228.
Dandeno,P.L. et al.Current Usageand Suggested Practices i n Power System Stability
Simulations for Synchronous Machines, IEEE Traits. Energy Coinwsioit, EC- I .
1, (Mar. 1986)pp. 77-93.
Dandeno, P. L., Kundur, P., and Schulz, R. P. Recent Trends and Progress i n Synchro-
nousMachineModeling in the Electric Utility Industry,IEEE Proc. . Vol.62,No.
7,(July 1974)pp. 941-950.
Debs,A.ModernPowerSystemControland Operation, KluwerAcademic, Boston,MA
(1988).
DeMarco,C.L. andOverbye,T.J. An EnergyBasedMeasureforAssessing Vulnerabil-
ity to Voltage Collapse, IEEE Traits. Po,zyer Syst., Vol. 5, No. 2. (May 1990)pp.
582-59 1 .
DeMarco, C.L. and Bergen, A. R. A Security Measure for Random Load Disturbances
i n Non-LinearPowerSystemModels,IEEE Trms.oi l CAS,Vol.34.No. 12,(Dec.
1987).
DeMarco, C. L. and Overbye, T. J. An Energy Based Security Measure for Assessing
Vulnerability to VoltageCollapse,IEEE Trcliis.oi i Porc~irSj:stcm.s, Vol.5, No. 2,
(May 1990).
DeMarco, C.L. and Overbye, T.J. Low Voltage Power Flow Solutions and Their Role
i n ExitTime Based SecurityMeasures forVoltageCollapse. Proc. rlte 27th IEEE
Cot$ on Decision arid Control,Austin,TX,(Dec.1988).
Dembart, B., et al. Dynamic Stability Calculations Using Vector and Array Processors,
Report no. EL-3335,Pal0Alto,CA:EPRI,(January 1984).
DeMello, F. P. and Concordia, C. Concepts of Synchronous Machine Stability as Af-
fectedby Excitation Control, IEEE Traits.,PAS-88. (Apr.1969)pp. 316-329.
DeMello. F. P. and Hannett, L. N. Determination of Synchronous Machine Stability
StudyConstants, EPRI Report EL-1424, 3, ElectricPowerResearchInstitute.Palo
Alto. CA,(June 1980).
DeMello, F. P. and Hannett, L. N. Representation of Saturation i n Synchronous Ma-
chines,IEEE Traits.Power Syst.,PWRS-I,4, (Nov. 1986)pp. 8-18.
DeMello,F.P.andLaskowski,T.F.ConceptsofPowerSystemDynamicStability.IEEE
Trms. ot i Pow-crAppar. and Syst., PAS-94 (3),(May/June 1975)pp. 827-833.
DeMello, F. P., Nolan, P. J., Laskowski, T.F.,and Undrill,J. M. Coordinated Applica-
tion of Stabilizers in Multimachine Power Systems, IEEE Tram. Poit-erAppur.
S j ~ t . , PAS-99, (May-June 1980)pp. 892-90 1 .
Dobson, 1. and Chiang, H. Towards a Theory of Voltage Collapse in Electric Pouw
Systems,Systems & Coittrol Letters,(1989).
Donimel. H. W. Digital Computer Solutionof ElectromagneticTransients i n Singleand
Multiphase Networks, IEEE Tram. PAS, Vol. 88, No. 4,(April 1969)pp. 388-
399.
El-Shnrkaui M. A.. Marks, R. J. , Aggoune. M. E., Park, D. C.. Damborg. hl . J.. and
Atlas. L. E. Dynamic Security Assessment of Power SJfstemsusing Bach Error
Propagation Artificial Neural Networks. Scwml Sj ql o. y i i i i r i o ~ i E.ipc~r.tS\..vtcwi .4p-
plic-ciriorit oPoItvr Sj:vtcw.s.Seattle. WA (1989).
El-Sharkawi. M.A..et al.Neural Networks and TheirApplication to PourerEngineering.
Coritrolt i ~i e lI l ~wt i mi cSj'.strni.s 4I , Academic Press, ( I99I ).
El-Sharkawi. M. A., Marks. R. J., and Weerasooriya. S . Neural Networks and Their
Application to PouwEngineering. in Colitr-01ciritl I?\ ~ri t rni i c~Sj,.slm.s.A d ~ ~ i i i t ~ ~ . s i i i
Tliror-j*~ i i i r lApplitwtioiis (C. T. Leondes, ed.).Vol. 41 .Part 1/4, Aciidemic Press.
San Diego, CA (1991), 2373-3 7 (March 1990).
EI-Sharkan4, M. A. and Atteri. R. Static Security Assessment of Power Systcm Using
Kohonen Neural Network. Pr.oc. Secwrit1 Irit. For-rtrii ori Ne~ri r- t i l Net\\~)i.rC.Al ~p l i c * t r -
tioris toP o ~ v r - ,Yj~.stom, ANNPS ' 93. Yokohama,Japan (April I993).
Endrenyi. J. Ke~litihilitj.MotlcjliriQqi r i E 1 c ~ ~ r - i ~ Sj~stc~r~is. Po~*c>r - John Wi ky & Sons. Ncw
York, ( 1978).
EPRI Report. Extended Transient-Midterm Stability Package (ETMSP),EPRl EL-4610.
Electric PonwResearch Institute, Palo Alto,CA (Jan. 1987).
EPRI Report. Arti fici 21 I In teIIigelice Technologies tor Pots, er S4'st em 0perati ons.EPRI
Project 1999-7,(1986).
EPRI Report. Dei.elopment of Expert Systems as On-Line Po\vcr Systeni Operational
Aids. No. EL-56-35,(1987).
EPRI Report. Pourer System Dynamic Analysis-Phase I , EPRl Report EL-484. Electric
Pourer Research Institute. (July 1977).
EPRI Project. PouwSystem Dynarnic Security Analysis Using Artificial Intelligent S j \-
terns. Phase Feasibility ELAuation, EPRI Project RP 3103-2(1993).
Feng. Z. , Ajarapu, V.. and Long, B. Identification of Voltage Collapse Through Direct
Eqiiilibrium Tracing,IEEE Tr amon Port*cJrSystcwi.~, (submitted).
Feng, Z.. AjjaraPu. V.. and Maratukulam, D.J. A Approach for Preventive and Corrcctiw
Control to Mitigate Voltage Collapse,IEEE Tr-ciris. o r i P o , \ ~ rSj ~si erri s.(suhniitted).
Findlay,J . A.. Maria, G.A.,and Wong. V. R. Stability Limits for Bruce LGR Scheme.
Proc. 18thPo,l~.rS\xteviis Cornptiitiori Coufewlicc, Helsinki. Finland. ( I9831.
Fischi. R., Kam, M., Chow,J. C.,and Ricciardi, S. Screening Poww Systeiii Contingen-
cies Using a Back-Propagation Trained Mu1tiperceptron, PI-oc.of'tlic I YKY I S C' AS .
Vol. 1. Portland, OR (May 1989)pp. 486-489.
Fitzgerald. A. E.. Kingsley, C.. and Umans, S. D. Electric Machinery, McGrau-Hill
Book Co.. New York. (1983).
Fleming. P. J . Computer Aided Design of Regulations Using Multi Ob-jecti\,e.In Pro-
cwtliugs qf IFACColitrolApp1iccitioii.v of'No~ilirietirpro^rci1ii11i iiig t I r i t l Opt ilr rizri-
tiori,Capri. Italy (1985)pp. 47-52.
Forgq'.C. OPSHZ Usc~r'sMr i ~i i r t i l .ComputerScienceDepartment. Carnegie-Mellon Cl i i i -
\wsity.(1983).
Fouad. A. A.and Stanton,S . E.Transient Stability of Multimachinc Po\verS l w m . Parts
I d111. IEEE TIU/ I. S. Po, t ~j rAp p r , ci ri rl S\.st., (July I981) pp. 3408-3424.
Fouad, A. A. and Vittal, V. Power System Transient Stability Assessment Using the
Transient Energy Function Method,Controle r r i d L?\~riti111ic~ ST. s i m, \ (C.7.. L c o r i r l c ~ . ~ .
c d . ), 43, Academic Press, San Diego. CA ( 1991) 1 15-1 84.
Fouad, A. A.et al.Calculationof Generation-Shedding Requirements ofthe B.C.Hydro
System Using Transient Energy Functions. IEEE TrLttis.Po, i vrAppt r . ~ i i d S?-.st.,
Vol. I , NO.2 (May 1986)pp. 17-24.
Fouad, A. A.,et al.Transient Stability Margin as aTool for Dynamic Security Assess-
ment, EPRI Report No. EL-1755. March (I981).
Fouad, A. A., et al. Transient Stability Program Output Analysis, Report no. EL-3192.
Palo Alto, CA: EPRI. August (1985).
Fouad. A. A., S. Venkataraman,and Davis. J. A. An Expert System for Security Trend
Analysis of aStability-Limited Power System,Paper no. 91WM216-2 PWRSpre-
sented at the 1991Power Engineering Society Winter Meeting, New York. (Febri-
ary I9911.
Found. A. A., Vittal, V.. and Oh, T. Critical Energy for Transient Stability Assessmerrt
of a Multimachine Power System,lEEE Treiiis.oi l Porivr Appcir. mid Sys t . . Vol.
PAS-103,(1984)pp. 2199-2206.
Fukunaga, K. and KoontL,W. L.G.Application of Karhunen-Loeve Expansion to Futuie
Selection and Ordering,IEEE Trtiris. Co mp t e r s , Vol. c-19,No.3(April 1970).
Fukunaga. K. Introduction to statistical pattern recognition, Academic Press. New,Yoik
(1972).
Galiana. F. D. and Lee. K. On the Steady State Stability of Power Systems. Pro(..# f
P/ CA. Toronto, (May 1977).
Gate. E. C., Henimaplardh,K., Manke, J. W.. and Gelopulos, D. P. Time Frame Noticm
and Time Response of the Models i n Transient, Mid-Term and Long-Term Stabil-
ity Programs, IEEE Treiirs.P ~ H Y ~ Appi r . Syst.. PAS-10.3, I , (Jan. 1984)pp, 133-
150.
Geeves,S.ClGRETF38.02.09,(con\!.) A,sses.siiierito f f rticti(u1Ftist Trtirisieiit Sttrhilrt\,
Methods; Stcrte oJ'tliv Art. ELECTRA. No. 153(April 1994).
Gelopulos. D. P. Midterrn Simulation of Electric Power Systems,,EPRI Report EL-SC;6.
Electric Power Research Institute, Palo Alto. CA,(June 1979).
Glandy, M. G.and Pai. M. A. Analysis of Voltage Collapse in PoawSystems. Pro(. . ( f
the 2 / s t NAPS, (Oct.1989).
Grainger,J.J . andStevenson.W. D.Po\t-rrSj*.stri?i Aiitilysis.McGrawHill,New York(IW1).
Greene,S. , Dobson. L.. and Alvarado. F.L.Sensitivity ofthe Loading Margin to Voltage
Collapse with Respect to Arbitrary Parameters. IEEE Trms. Po,i.ur.Yj*.st..PWh'S'I
I . (May 1996)pp.835-850.
Gross. C.A. Po,\.ri. Sy.ster,i Ai i dy s i s , John Wiley &Sons,Inc., New York, (1986).
Guckenheinier,J. and Holmes,P.Nonlinear Oscillations, Dynamic Systems,and Bifurca-
tion of Vector Fields, Springer-Verlag,New York, ( 1983).
Hahn, W. Stcthilitj. of Motion,Springer-Verlag,Berlin. (1967).
Hakimmashhadi and Heydt, G.T.Fast Transient Security Assessment, l EEEPWRS,\0 1 .
I02 (December 1983)pp. 38I 6-3823.
Hanimons. T. J . and Winning. D. J. Comparisons of Synchronous Machine Models in
the Study of the Transient Behavior of Electrical Power Systems. Pro(. . IEICE.
Vol. 118, NO.10.(Oct. 1971)pp. 1432-1458.
Harris, M. R.. Lawrenson, P. J. , and Stephenson,J . M. Per-unit Systems with Special
Reference to Electrical Machines, Cambridge Unilwsity Press, Cambridge, Eng-
land. (1970).
Bihliogrciphy 339
Hartana. R. K. and Richards,G. G. Harmonic SourceMonitoring and Identification U s -
ing Neural Networks, IEEE PES 1990WinterMectiiig,Paper No. 90 WM 238-6
PWRS, Atlanta, CA,(Feb. 1990).
Hasan. A. R., Martis, T.S.,and Sadral Ula. A. H. M. Design and Implementation of a
Fuzzy Controller Based Automatic Voltage Regulator for a SynchronousGenera-
tor, IEEE Tiuiis.oii Eiiei-gyConif., Vol. 9,No. 3, (1993)pp.550-557.
Hassan, M. A. M., Malik,0.P., and Hope, G.S. A Fuzzy Logic Based Stabilizer for a
Synchronous Machine, IEEE T r a moii Eiiergy Co~i\,., Vol. 6. No. 3, (19911 pp.
407-413.
Hawing,J. N.,Choi,J.J., SehoOh,andMarks, R.J. I I . Query-Based Learning Applied
ToPartially Trained Multilayer Perceptrons. IEEE Ti-rriis. Neiii-(11Nrt\twi-ks.No.
(January 1991)pp. 131-136.
Heffron. W. G. and Phillips, R. A. Effects of Modern Amplitude Voltage Regulator i n
Under Excited Operation of Large Turbine Generators, AIEE Trrtris.,PAS-71 .
(Aug. 1952)pp. 692-697.
Heydt,G.T.Coiupiitei-Aiiu1y.si.sMethodsforP m v r S!-.stei?i.s, Macmillan PublishingCo. ,
New York, (1986).
Hirsch. M. W. and Smale. S. Dlffereiitial Eqitatiotis, Dymriiiicol Systcwis. ri i i t l Liiicwi-
Algehrcr,Academic Press, New York, (1974).
Hiyama, T. Robustness of Fuzzy Logic Power System Stabilizers Applied to Multi
Machine Power System,IEEE Traits.o i i Eiiergjv, Vol. 9, No. 3. ( 1993)pp. 351-
459.
Hsu, Y. Y. and Cheng, C. H. Design of Fuzzy Power System Stabilizers for Multi-
Machine Power Systems, IEEE Proceediiigs oi t Geiti~r~itioii, Trciiisi,~i.s.sioit t r i i t l
Distrihirtioii. Vol. 137,Part C,No. 3, (May 1990),pp. 233-238.
IEEE Committee Report. Computer Representation of Excitation Systems. IEEE Trtriis.
P o u ~ r A p p r . Swt . , PAS-87,6,(June 1968)pp. 1460-1464.
IEEE Committee Report. Dynamic Models for Steam and Hydro Turbines in Power
System Studies,IEEE Tram.Pou7er Appar. Syst.,PAS- 92, 6,(Nov/Dec 1973)pp.
1904-1915.
IEEE CommitteeReport.Excitation SystemModels forPower System Stability Studies,
IEEE Trciiis.P o ~ v r Appar. Syst.,PAS-100.2, (Feb. I981) pp.494-509.
IEEEPES Engineering Education Committee. IEEE Tutorial Course: Digital Simulation
of Electrical Transient Phenomena, IEEE Tittoricil 8I E H 0 17-3-5-PCVR.IEEE,
New York, (1980).
IEEE Std 115A. IEEE Trial Use Standard Procedures for Obtaining Synchronous Ma-
chine Parameters by Standstill FrequencyResponse Testing,Si ~ppl ei i i ~~i i t t o ANSII-
IEEE Std. 115-1983,IEEE,New York, (1984).
IEEE. Symposium on Synchronous Machine Modeling for Power System Studies,
83THOlOl-6-PWR, Tittorial at IEEE/PES 1983 Wiiitrr Meetiii<q.New York,
(1983).
IEEE. Special Publication 90TH0358-2-PWR, Vol t t i pStdi1it.vof Pon.ei.Sj'.sreni.s: Coii-
cepts,Aiicilyticd Toolsmid Itidustiy E.rperieiice,(1990).
Ignizio, J. P. Generalized Goal Programming-An Overview, Computer and Ops. Re-
search (Pergamon Press) 10,No.4,(1983)pp. 277-289.
Ilk-Spong,M.,Crow.M.L.,and Pai,M.A.Transient Stability Simulationby Waveform
I
Relaxation Methods, IEEE 7'rtrri.s. o i i P o \ c ~ r.Sjs.sr. PWKS-2. No. 3 (No\.embes
1987)pp. 943-953.
Ipnkchi. K. R., Brandwiijan. A. . El-Sharkaui, M. A. arid Caule~r, G. Neural Works f o .
Dynamic Security Assessment o f Large-Scale Pomw Systems: requirement^,
O\w\, iebr. Proi.. I " l r i t i ~r ~i i t r t i oi i i i l Forrri,i o r i NNAPS. Seattle, WA (199I ) pp.
65-7 1 .
lwan S. N.a n d Tan. 0. T.Neural-Net Based Real-Time Control Capacitors Insralled on
Dixtribution Systeiiis.IEEE ~ I W J . L O I I Pori~r.l)e/i,'t~r;\~. Vol. 5 , No. 1, ( 1990)p p .
266-373.
IEEEWorkingGrouponDynamicSystemPerformanceo f thc SysteniPlanningSubcom-
mittee, IEEE Power Engineering Society Power System Engineering Cornmittet.
Sjcmposiuin on Adequacy and Philosophy of Modeling: Dynamic Sqsstem Pertoi-
iiiiince, IEEEPirblii~irtioii75C'HO 97O-J-PWK I EEE New York, ( I975).
Jackson. W.B. und Winchester. R. L. Direct and Quadrature Axis Eqiii\,alent Circuits
for Solid-RotorTurbine Generators, IEEE 7'rtrri.s. P o ~ t ~ r &pi t . . .Sj,.st., PXi-A'8.-'.
(JliI>r 1969)pp. 1121-1 136.
Jaris, J . and Galiana. F. D. Quantitati\re Analysis of Steady-State Stability i n P o w ~
Nctu.orks,IEEE 7~rirrr.s. o r 1 PAS, Vol. 100,(Jan. I98I 1.
r Jevasurjma, B. P o u ~Syste111 Loading Miirgi n Est iiiiat ion Using Arti t'i cial Neural Net-
w x k . 2rZtli ~Vort hAt rrt ~ri cwrPort~r5' ~wi po. s i r r t r i .M.I.T., MA. (November 1996).
Kakimoto, N., Ohsaua, Y., and Hayashi. M. Transient Stability Analysis of Electr~c
Po\vcr Sq,sternsvia Lure Type Lyiipiiiiov Functions, Parts I I . and I . I EEE 7 r ~ r r i \.
c!f'Jtrptrri.YK, Y6.(M:iy/Jlltie 1978).
Kam. F. R.,Chow, M. J . C..and Ricciardi. S. An Improved Hopfield Model t or POM.Y
S j w m Contingency Classification, Pr.oc.. of'tlic 1990/SC',4S. Vol. 3. Ne n OF
leans, LA (May 1990) pp. 2925-3928.
Kcssel, P. and Glavitsck. H. Estimating the Voltage Stability of a F'o~wSqrstem. i /i l 'L
Tr~trri. stri~tiori. s, / Y M . PWK1)- I ( 31, pp. 346-354.
Kundur. P. Po\i.rrSwrri i Srtrhility trriil Corrtrol. McGraw Hill, New York. ( I9931.
Ku,antjc,H.G.,et al.Static Bifurcation i n ElectricP o w r Networks:Lossof SteacljrSt; tc
Stability and Voltage Collapse. IEEE Trtriis.o i i CAS. Vol. CAS-33. No. 10.( 0 ~ .
1986)pp. 981-991.
Khorasani. K., Pai. M. A.. and Saiier, P. W. Modal-Based Stability Analysis of PouCI'
Systems Using Energy Functions, I t i t . J . of'l < I c ~ c ~ t r i c ~ Po\t,c)t.t i t i d Errei;q>..S\~.srcr,i\.
Vol . 8.NO.I , (JiinLILirj' 1986)pp. I 1-16.
Kimbiirk. E. W. Po, i ~) r.Sj~.sttwi Stirhilitj.: S\ , i zc. l rr. ori orr. s M d z i i r i ~ s ,Doiw Publicatio14.
Inc., New York, ( 1956).
Kohonen. T. Self-OrganiLed Formation of Topologically Correct Feature Maps, Ni c d .
~ ' \ ~ / ~ f ~ l . l l J ~ ~ f l ~ ' , S .( 1982)pp. 59-69. NO. 33
Kokotovic.P. V. andSauer.P.W. Integral Manifold;is;i ToolforRcdircedOrder hlodel-
ingof Nonlinear Systeiiix:A Sjmchronous MachineCase Study,IEEE Trtrii.\. ( ' / I . -
c,rrir.sSj..sr.. Vol. 36,No. 3, (Mar. 1989)pp.303410.
Kokotovic. P. V.. Khalil, H. K, andO' Reilly, J.Si ri grr/ t ri . Prrtirr-lxrtioiiMet hods i r i ( . t ) r i -
t rol : ,-\ritrlysi.s ti ri i / Dt ) . si , gi i . Academic Press Inc., London. ( 1986).
Komai. K. , Sakaguchi.T.. andTakeda,S.Power S j W m Fault Diagnosis ui t h an E\ l icrt
System Enhanced by the General Problem Solving Method. IASTEI) C' oi ?f i ~r - c wc ~c ~.
Bozeman, Montana (Aug. 1986).
Kosko. B. Fi i q ' Thirrkiiig. Prentice-Hall, Englewood Cliffs,NJ. ( I993).
Kosko. B. Neural Netwrorks and FuzzySystems:A llj*ritiriii(,Aplwotic*lit oMr i c ~ l i i r i c ~ Iri t ol -
ligtvice,Prentice-Hall,Englewood Cliffs, NJ (1992).
Krause, P. C.Ariti1j~si.sof Elrc*tric. Mt di i i i or - j ~,McGraw-Hill,New York, ( 1986).
Kreyszig, E.Aeliwticwl Eri,qiric.c)ririg Mtitlwiii[itics. John Wile),Br Sons. Inc.. New York.
(1972).
Kuffel, E. and Abdullah, M. Hi gh- Voltcigc. Eri,qiricvririg, Pergamon Press, Oxford. Eng-
land,(1970).
Kundur, P. Portx)r- Sjxturii Sttihility ciricl Cor1rt-d.McGra\s,-Hill.New York.( 1994).
Kundur, P.. Klein, M., Rogers, G. J.. and Zywno. M. S. Application of Power S>rstem
Stabilizer for Enhancement of Overall System Stability, IEEE 7-1uri.\,P\\'KS-J.
( Mt i j ?1989)pp. 61-1-626.
Lt i Sctrlcr. M.,et til. A High!\%Peirtillc4 Method,fi)r*Ti-tirisiuit Sttihilitj.Ari t i 1j xi . s. 111 Pro-
c . r r d i i i ~ . S(!f' tlitj 1989PICA, (May 1989)p ~ . 380-386.
Larsen. E. V. and Swann, D. A. Applying Power System Stabilim-s.Part I to I l l . IEEE
Trciris.o r 1 Po1ix.r Aplxrr. e r r i d Sj?.st.,Vol. PAS-100.No.6. ( I981 pp. 30I 7-3046.
Lee, B. H. and Lee, K. Y. A Study on Voltage Collapse Mechanism i n Electric Pou,er
Systems. IEEE Trtiris. o r 1 Po\ivr- Apperr. Sj ' st . , Vol. 6, No. 3. (Aug. I99I ) pp.
966-974.
Lee, B. H. and Lee, K. Y. Dynamic and Static Voltage Stability Enhancement of Po\ver
Systems,IEEE Trciri.scic*tioriso r 1 P o , i ~ r . Vol. 8,No. I , (Feb. 1993)pp. 231-238.
Lee,C.C.Fuzzy Logic i n Control Systems:Fuzzy Logic Controller-Parts I and I I . IEEE
Trtiris.o r 1 Sjvst. Mtiri ridCylwr.rietic*.s.Vol. 20, No. 2, ( 1990)pp.404435.
Lee. S. Y..etal.Parallel PowerSystemTransientStability AnalqpsisonHqpxcubcMulti-
processors, I n Procwciiiigs of t he I 989PICA, (May 1989)pp.400406.
Lesieutre, B. C.. Sauer, P. W., and Pai, M. A. Sufficient Conditions oii Sratic Load
Models forNetwork Solvability. I n Proc. 24t hAri ri i rri l Nortli An i c i r . . Po\ t ~r - s\wip..
Reno. NV (Oct. 1992)pp. 262-27 I .
Liu, C. C. and Dillon. T. S. State-of-the-Art of Expert Sj'stem Applications to Po\+.cr
a r i d Ei i c ~r - g~ Systems,I r i t . J. Elec-triculP OHY J ~ Sj:stetii.s,( 1989)in press.
Liu,C.C.DynamicSecurityAssessment:TheRoleof ExpertSystems,minutesof theJ ul y
27. 1988meeting of the Dynamic Security Assessment Working Group mccling.
PowerSystemEngineeringCommitteeof the PowerEngineeringSociety.IEEE.
Liu,C.C.,Lee,S.J.. and Venkata,S.S.An Expert System Operational Aid for Restora-
tion and Loss Reduction of Distributed Systems.I EEETrtiris. o i i Po i t ~r . I ) c ~ l i \ ~ ~ i - \ .
(May 1988)pp. 619-626.
Liu. C.-W.,Chang,J.-S.,Su,M.-C. Neuro-Fuzzy Networks for Voltage Securit.8 hloni-
toringBasedon SynchronizedPhaserMeasurements.IEEESitiiiiiic.r. Meoririg. Berlin.
Liu, C.C.and Vu. K. T. Analysis of Tap-Changer Dynamics and Construction ot'Volt-
age Stability Regions,IEEE Trcrris. or 1 CAS,Vol. 36. (April 1989)pp. 575-590.
Liu,C.C.Characterimtion of a Voltage Collapse Mechanism due to the Eft'ectsof On-
Line Tap Changers. Procw~cliiigsof IEEE liitc~riititiori(i1Sj wi pos i i i r i i or1 Ci t +(xi t , \
crriel Sj..strr?is, San Jose,CA. Vol. 3,(May 1986)pp. 1028- 1030.
Lopes,J. A.. Pecas, F. M.. Fernandes, Matos, M. A. Fast E\uluation ofVoltageCollapsta
Risk Using Pattern Recognition Techniques. Ptrper APT 300-20-09r i c ~ ~ p t e r l , f i j , .
if t itiori tit the IEEE/NTUAAflieri s POr t ~ p i - t ~s e r I - TecA Coil/>reri Athens. Greece. i t ( > ,
(Sept. 5-8, 1993).
Long, B.andAjarapu,V.TheSparseFormulationof ISPS and Its Applicationto Voltagc:
Stability Margin Sensitivity and Estimation, IEEE Tvtiris. o i i PortvrApptir. S\,.\-
ttnr.s, (Oct . 1997).
Lyapuno\., A. M. The General Problem of the Stability of Motion (in Rirssian), 7/io
Mtitlirriititic~tilSoc.iet~9of Khcirkoi?. Rir.s.si~r,1892, English translation, Taylor an11
Francis Ltd..London, (1992).
hledanic,M. Illic-SpongandChristensen,J. DiscreteModelso f Slotv Voltage Djmamics
for Under Load Tap-ChangingTransformerCoordinations,IEEE Trtim.o i i Poit.cr.
A1111tir.SJ.S~.. Vol. 2, (NOV. 1987)pp. 873-882.
Morison, G. K.. Gao, B., and Kundur, P. Voltage Stability Using Static and Dynami2
Analysis. IEEEbE.5 Si i mi i er Mrct i i i g, Seattle, WA, (July 12-16 1992) 92 Shl
590-0-PWRS.
Ma, X..El-Keib,A.A.,Smith,R. E.,andMa,H. Estimationof VoltageStability Margins
Using Artificial Neural Networks, Proi.. Ofthe 26th NAPS, (Sept. 1994).
Maeda, M. and Murakami, S.A Self-Organizing Fuzzy Controller, f - i ~ : ~ Sets tiritl S\,: -
terris,Vol . 5I . ( 1992)pp. 29-40.
Magnusson, P. C.Transient Energy Method of Calculating Stability,A/ E Tintis. , Vol.
66,( 1937)pp. 747-755.
Mansour. Y., Vaahedi, E.. Chang. A. Y. , Corns. B. R.. Garrett, B. W.. Demaree. K. .
Atha).T.,Cheung, K. , and Hydro. B. C. On-Line Transient StabilityAssessmciit
(TSA): Model De\relopment. Analysis and Post-Processing. / EKE Trtiiis. Poi i ~*r -
A/l/>tit+. S\..S~.. Vol. 10,No. I . (Feb. 1995) 241-253.
Marathe. H. Y., Liu, C.C. , Tsai, M. S. , Rogers. R. G.. and Maurer. J. M. An On-Li re
Operational Expert System with Data Validation Capabilities, Pi - oi ~./ Y8Y t/C,~I.
pp.56-63.
hlarceau. R.J . , Mailhot, R..andGaliana,F.D.AGeneralized ShellforDynamic Securiiy
Analysis in Operation Planning, I EEE Trtiris. or1 torc.fr SJsterris, Vol. 8, No. 3
( 1993), pp. 1098- I 106.
Maria, G. A. , Tang. C.. and Kiln, J. Hybrid Transient Stability Analysis. Paper 11. ).
89SM684-2-PWRS. IEEE Siuriiiwr Portver Mcwi ri g, Longbeach. CA, (July 1989) .
McCallcy. J . D.. et al. Power System Security Boundary Visualization Using Neural
Networks, Riilk porrvr .sj%sterti c l wt i r i i i c - s t i ri t l Corifro1IV-Ke. st t ~i i c~t i rri rl g. (August
24-28. 1998)Santorini,Greece.
McClelland, E. C.and Van Home. P. R. Fast Voltage Predication Using 21 Knowledge
Based Approach, /ETrtrris. PWRS. (February 1983)pp. 315-3 19.
Michel, A. N., Fouad, A. A., and Vittal, V. PourerSystem Transient Stabilitj, Using
Indi\ridual Machine Energy Functions, IEEE Trtiris. Circirits S\ . . st . , CAS-30.Vol .
5 , (May 1983)pp. 266-276.
Miller,R. K., and Michel, A. N. Ordiiitii:\~tXfc~r-twtidEyircitioris, Academic Press. Ncw
York. (19x3).
Minnich, S . H. et al. Saturation Functions for SynchronousGenerators from Finite Ele-
ments, Pciper87 WM207-4,IEEE/PES I987 WinterMer4iig.New Orleans. LA,
(Feb. 1-6. 1987).
Minnich, S.H. Small Signal. LargeSignalsand Saturation in Generator Modeling. IEEE
7rtin.s.Eiiergy Cori\~er.sioii, EC-I, Vol. I , (Mar. 1986)94102.
Mori, H., Tamaru, Y., and Tsuzuki, S. An Artificial Neural-Net Based Technique for
Power System Dynamic Stability with the Kohonen Model, IEEE Proc. of 1901
PICA,Baltimore, MD, (May 1991)pp. 293-301.
Miranda, V. and Saraiva,J. T. Fuzzy Modeling of Power Syst emOptimal Load Flomt.
IEEE Trciris.oi i Po,i,erAppcir.Syst em, Vol. 7, No. 2,( 1992)pp. 8.13-849.
Mori,H.andTarnura,Y. OnVoltageSecurity On-LineIndexin Electric PowerSystems.
Poiver System Eiigineering Cminiittee of IEEE of Jqxr i i . Paper No. PE85-41,
(June 1985)( i nJapanese).
Mori, H. and Tsuzuki, S. Determination of Power System Topological ObservabilitJc
Using the Boltzmann Machine, IEEE Proc. ofl Y Y 0 ISCAS. New Orleans. LA.
(May 1990)pp. 2938-294 I.
Mori, H. and Tsuzuki, S. Estimation of Critical Points on Static Voltage Stability i n
Electric Power Systems,IFACProc. of I990Synip.on Po\i,er Sjwc-'iii.smid Pou.i.r
Plcirir-Control,Seoul,Korea, (Aug. 1990)pp. 550-554.
Mori, H. and Tsuzuki. S. Power System Topological Observability Analysis Using a
Neural Network Model. Proc. cf Secoiid S wi I ms i i r i i i o ~ i E.vpcJr*tSj~stcwisApplicw
tioii to Pobtver Sj st ems, Seattle. WA (July 19)pp. 385-391.
Mori,H. Decentralized PowerSystemVoltageControl UsingArtificialNeuralNet\t<orks.
Proc. of 1992IEEEISCAS. San Diego. CA, (May 1992)pp. 1701 -1 704.
Mori, H. Monitoring and Control of Power System Voltage Stability Using an Artificial
Neural Networks, Preprints-ofI992IFAC/lFlP/IMACS'IiiteriieitioiitilSj wi pos i r r l i i
o i i Artifcictl
Mori,H.andTama-ru. Y. An ArtificialNeural-NetBased ApproachtoMonitoringPo\ver
SystemVoltageStability,Proc.qf Nulk Poriqer Systcwi Voltage Plieiioiiicwci 11: Volt-
cige Stcihilitjq ciiicl Seciirih,DeepCreekLake,MD. (Aug.1991 ) pp. 347-358.
Mori, H., Itou. K.,Uematsu, H. , andTsuzuki,S.An Artificial Neural Net BasedMethod
for Predicting Power System Voltage Harmonics," IEEE Trrrris. 0 1 1 Po,i*c.i*Dolii,-
et-j*,Vol. 7,No. 1, (Jan. 1982)pp. 402409.
Mori, H., Uematsu, H., Tsuzuki, S.,Sakurai,T., Kojima, Y. , and Suzuki, K. Identifica-
tion of Harmonic Loads in Power Systems Using an Artificial Neural Netwrorks,
Proc. of Secoiid Sjwiposiiaii on Expert Systeni.~Applie.cirioii i i i Po~qer Sj:vteiii.v,
Seattle,WA. (July 1989)pp. 371-378.
Niebur,D. and Germond,A. J. PowerSystemStaticSecurityAssessmentUsing Kohonen
Neural NetworkClassifier,IEEE Trurzs.h v e r Deli\*er-j(May 1992)pp. 865-87 1.
Nims,J.W.,El-Keib.A. A.,andSmitk,R.E.Contingency Ranking forVoltage Stability
Using a Genetic Algorithm, Electric Poitv?rSystemResrarch, (October 1996).
Nuhanovic. A.. Glavic. M., and Prijaca, N. Validation of Clustering Algorithm for Volt-
age Stability Analysis on the Bosnian Electric Power System. IEEE Pi *ocwdi t i gs
o i i Geiiertitor Tr~iiisniissionaridDistribution,Vol. 145, No. 1 , (1998)pp. 21-26.
Obadina,0.0.and Berg, G.J.VAR Planning for Power System Security,I EEE Ti-ri1i.s.
o i i Pouvi- Sjsteiiis.Vol. 4,No. 2,(May 1989).
O'Grandy, M.G. and Pai. M. A. Analysis of Voltage Collapse in PoLtrer Systems. PI.oc..
( ?/ ' i / i e.?/"NAYS. Rolla, MO, (Oct.1989).
Ot wbye, T. J . and Demarco, C. L. Improved Techniques t i w PoLver S j w m Voltage
Stability Assessment Using Energy Methods, / EELTrrrris.o t i Po,c.r~i. .S\.,siotti.\. Vol.
6. No. 4. ( Nov. 19911,
Ojo. J . D. and Lipo, T. A. An Improved Model t or Saturated Salient Pole Syiichronoux
Motor\. Ptrl)t)r88 SM 6 / 4 0 , /EEE/PES 1988 S i ~ t t i t t r o t ~ Mw r i t i ~ .Portland. OR.
( J ul y 24-29, 1988).
Oli\,e.D. W. New Techniques for the Calculation o f Dq'n;i~iiic Stubitit),. / EEE 7 i w i . )
Po\\.cI.A/ >/ X~I . . S\,.Y/.,PAS-&S. Vol. 7, ( Jul y 1966).
O\tojic ancl Heydt, G. T.Transient Stability Assessment by Pattern Recognition i n t ht .
Frequencjr Doniain. //:EL' Tt wi s . Po\ t ~t . Ap p i r . S!,.\t.. Vol. 6.(Februarj,1091 ) pp.
231- 37.
0s) r c~ka.A. Multicriterion Optimi~ation in Engineering, Ellis Hor\wod. Ltd. (Di\ri\ioii
ot'John Wileyl. New York, (1984).
0~crhise. T. J. ancl DeMarco. C. L. ImproFd Techniques for Power S ~ ~ e i i i Volt;igl:
Stabilit,, Assesxnient Using Energy Based Sensitic,ities. IEEI:' Trtrrrs.O I I Po\\.cr.
S\,.st.. Vol. 6. No. 3, (Aug. I991 ).
O\whq' e. T.J . and DeMarco. C.L. Voltage Stabilitj,Enhnncement Lrsing Energj' BascJ
Sensiti\,ities./ EELTrtrr,.\. otr Po\t.c>rS\*st..Vol. 6. No. 3.(Aug.199I 1 .
O\ , crhj e. T.J. Application of 2111 Energy Based Securitj hlethod to Voltage Iristiibilit? i ii
EIectricUI POWer Syste111, lX.\. sert t i tiorI , Uriiversi t 1,of Wisconsin- hl adison. ( 199I I .
O\crh>,e.T. J . . Dobson. I. . and DeMiirco, C. L. Q- V Cur\,e Interpretations of' Energl
hltxsures t or Voltage Seciiritj,, II:'EE/PES lY9.j \ Vi i i rcr Mec. i i i i <g, ( J un 3I -Feh 5.
1993).
Padij~ar,K. K. and Chosh. K. K. Direct Stubitit)! E\'aluation of'Power Systems nI111
Det ai Ied Gcncrator ModeIs Using Struct lire Preser\ring Energy Furictions./ I I 1. 1.
El/r)c.tr. Po\t.ri.( / r i d EilcJt;q\S\,.s/.,Vol. 11 . NO. I , (JLIII.1989)pp. 47-56.
Padi>rar.K. I?. and Sastry. H. S . Y. Topological Encr g~ Function Analyis ot' Stabili
S>! S~~' IIIS. S\..\./.. ot'P OWC~ / i t / . li/cJ('tt.. Po\\'c't. t r t t ~ lI : ' I I o I . ~~\ ~ Vol . 9.I-. (Ji111. 1987)j>,?.
9- 16.
Pal, hl . A. et al. Direct Method of'Stability Analysis i n Dj~niiiiiic Security A\st.sumetit.
t ' t rpci . iV0. /.]/A-!. IFAC World Congress,Bh p e s t . (Jltlj, 1984).
Pai. M. A. Cottr/wtcr7'c. chriqire. s irr Po\t,or S:\~stettrAmr/\%.\i.s.Tata McGrau-Hill. NCW
Delhi, Indiii, ( 1979).
Prabhakara. F.m c l He),clt,G. T.Re\,icwo f Pottcrn Rccogni[ion Methods forRapid AII~II-
ysisof Transient Stability,/LEE-PESTc~chrricu/ H e p o r t o t i Kt r pi t l Aittr/\~.\i,sc!f' Tt ~rt r-
,sit'ili ,sitO)i/iI\'. #87THO169-3-PWR(1987).
Pai, h4. A.Gr c ~r ~gy Fiiti(*tiottArrtr1ysi.sj i ) r Po\tvr.S~, s / or , i . Sttihi/itj, . KliiwerAcademic Pih-
lishers. Boston. ( 1989).
Pai, M. A. Power System Stability Studies by Lyapiino\.,-Popo\.' Approach. /'i-oc. :"/i
IFAC' \.t'or.lti C' o r ~g r ~~s s , Paris, ( 1972).
Pai, M. A. . Po r i ~ r . S:\wcrri S' t d~i l i rj ~. North Holland Publishing Co.. Nett,York. ( 1981 .
Pai. M.A..Laufeiiberg. M.. and Saucr.P.W.Some Clarification i n the Transient Encig!'
Function hlethod, / i l l . ./. of'Elec.. Po r t ~~t . . trtitl Etici.,g!*S\.xiciiis, Vol. I 8, No. I .
(1996).pp. 65-72.
Pang, C.. Kprabhakara. F. S. El-Abiad, A. H., and Koi\u. A. I . Security E\.aluation
Pourer Systems Using Pattern Recognition, IEEE Tr.rrri.s.PAS. Vol. 03, (h4aj./Julq,
1974)pp. 969-976.
Pao, Y. and Sobs-jic, D. J. Combined use of Unsupcr\.ised and Super\,ised Learning
Dynamic Security Assessment. Pro(.. of' PICA. Baltimore. MD, ( May 19911 pp.
278-284.
Park, D.C.,El-Sharkawi, M. A.,Marks, R.J. , Aggoune, M. E.. Atlas. L. E.. and Damb-
org, M. J. Electric Load Forecasting Using an Artificial Neural Nct\j.ork, IEEE
PES I YYOSi i i mwr Mwriii'q, Paper No. 90 SM 377-2 PWRS, Minneapolis. hl N.
( Jul y 1990).
Patrick. E. FirricltiriicJrittil.v of'futtorriRocogiiitjoii. Prentice-Hall Information and S!.stem
Science Series. Englewood Cliffs, NJ, (1972).
Pa\rella, M. and Murthy,P. C. TrurisientStcihi1it.v c!f' Ponjc~r. Systerus: Tl ~c ~or y trritl PI-CCC. -
ri w. John Wiley &Sons, Inc., New York. (1994).
Peng,T.M.. Hubele, N. F.,and Karady, G.G.Ad\~ancemcnti n the Application of Neural
Networks, for Short-Term Load Forecasting, / EELTr-(iris.or1 Pou.c)r-.4ppcrr. .J'\'.vt..
Vol. 1 , No. I , (Feb.1992).
Peterson, T. S. Ccilcw/iisiivitli Ariulytic Georiietr:\,,Harper & Row Publishers. Inc.,N ~ N
York, ( 1960).
Policy i n a Fuzz),-Logic Controller. Electi-oriicxLottc~r-s. Vol. I I . ( 1975)pp. 625-626.
Pottle. C.Array and Parallel Pro~~. s. sor. s in On-LineComputations.Report No. EL-2363.
Palo Alto, CA: EPRI,(April 1982).
Prabhashank w, K.andJani schewsyj,W.Digital SimuIation of MuItimacliine P o u ~ Sjrs-
tenis for Stability Studies, IEEE Trtriis. o r 1 Poii'erAp p t i i - . Syst., PAS-87. Vol. 1.
(Jan. 1968)pp. 73-79.
Proceedings 8th Po\+,erSystemsComputation Conference. Helsinki, Finland. ( 1984)pp.
1063- 1069.
Proccedings of Sympcjsiurn on Expert Systems Application to Pourer S>stc'iiis.Stock-
holm. Helsinki, (August 1988).
Proceedings of the IEEE Third International Conference on fu/,zy Logic. San Francisco.
CA,( 1993).
Proceedings of the Workshopon Power System Security Assessment. lou,oState LI ni \ er-
sity, (1988).Library of CongressCatalog No 88-091308.
R. I . 11. Artificial Neural Networks forPower SystemsStatic Security Assessment. PI-oc. .
of f heISCAS,Vol. 1. Portland. OR (May 1989a).
Rahimi, F. A., Lauby, M. C.. Wrubel, J. N., and Lee, K. W. E\,aluation of the Transient
Energy Function Method for On-Line Dynamic Security Analyhis, IEEE Trt i ri . \ .
0 1 1 P0rt.c.r A/)/?(ir.SJ,.vt., Vol. 8,NO.2, (May 1993)pp,497-506.
Rajagopalan. C. Dynamic of Powrer Systemsat Critical Load Le~vl s, Pt1.D. Thcsis. Uni -
\w-sityof Illinois, Urbana-Champaign, (1989).
Rajagopalan, C.. Lesieutre. B., Sauer. P. W.. and Pai, M. A. Dynamic Aspects of Volt-
age/Po\ver Characteristics, IEEE Tram.Po\tvr Apptir: S\-.vt.,Vol. 7. (Aug.1992)
pp. 990- 1000.
Rajagopalan, C., Sailer. P. W.,and Pai, M. A. An Integrated Approach to Dqmniic and
Static Voltage Stability, Proc. IY89Arwri cwi Coritr-olCoi ~/ i ) r mw. Pittsburgh. PA.
Vol. 3. (June 1989)pp. 1231-1235.
Ribbens-Pavetla,M. And Evans. F. J. Direct Methods forStudyingof the Dynamicsof
Large Scale Electric Power Systems-A Survey, Aut ort i t i t i cu, Vol. 21, No. 1.
(1985)pp. 1-21.
Ramey,D. G.and Skooglurid. J. W. Detailed Hydrogovernor Representation for System
Stability Studies, IEEE Trciris. Ponx)rApptir. Syst., PAS-89. Vol. I . (Jan. 1970)
pp. 106-112.
Ramo. S., Whinnery, J. R., and Van Duzer, T. Fielcls t i ri cl Wtr1v.si n Coninrirriic~citiori
E1ectroiiic.s.John Wiley 62 Sons,Inc., New York. (1965).
Ranjan, R. K.. Pai, M. A., and Sauer, P. W. Analytical Formulation of Small Signal
Stability Analysis o f Power Systems with Nonlinear Load Models, In ScrclIitirrcI.
Proc.. in Eng. Sci.. Indian Acad. Sci., Bangalore, India. Vol. 18, No. 5, (Sept
1993)869-889. Errata. Vol. 20. No. 6,(Dec. 1995)p. 971.
Rankin, A. W. Per-Unit Impedance o f SynchronousMachines, AIEE Trcrris.,Vol. 64
(Aug. 1945).
Rastgoufard. P., Yazdankhah. A.,and Schlueter,R. A. Multi-Machine Equal Area Based
Power SystemTransient Stability Measure, IEEE Trerris. o r i Por \ ~r Appcir.. S!s.st..
Vol . 3, No. I . (February 1988)pp. 188-196.
Russell, B. D. and Watson, K. PowerSubstation Automation Using a Knoujledge Basccl
Syst eni J11stificat i on and PreIim i nary FieId Experiinen ts. IEEE Trtiiis. oi i PONI) 1.
Deliivr.i.,(October 1987)pp. 1090-1097.
Sobajic. D. and Pao, Y. Artificial Neural Net based Dynamic Security Assessment Elec-
tric P o a wSystems,I EEETrtiris.PIVRS, Vol. 4, (February 1989)pp. 22-228.
Sobajic. D., Pao, Y. H., and Dolce, J. On-Line Monitoring and Diagnosis of Po1vt.r
System Operating ConditionsUsing Artificial Neural Networks, Proc..of the) 1 7
ISCAS. Vol. 3, Portland. OR.(May 1989)pp. 2243-22-16.
Stagg. G. W. and El-Ablad,A. H. Corripirter Metliotls iii Pori ~rSysttwr Aritilj~sis,Mc-
Grau-Hill Book Co..New York, (1968).
Stott. B. Power System Dynamic Response Calculations, Pro(*.IEEE, Vol. 67, (Fe\ ).
1979).pp. 219-241.
Sit,C.-C.and Hsu, Y.-Y. F u u y DynamicProgramming: An Application to Unit Conr-
mitment, IEEE Treiri.~. o i i Po,i*erAppcir. Sj,.st., Vol. 6, No. 3, ( 1991) pp. 133I -
1237.
Syst tn is COntroI 1nc.DeveI opmen t of Dyna111ic Equivalents f orTransient St abi1i t y St11c1-
ies. Final KrJportEPKI EL-456,Electric Pobver Research Institute. Palo Alto, Ci\,
(Apr. 1977).
Tamura, Y., Iba, K., and Iwamoto, S. A Method f orFinding Multiple Load Flow Solti-
tions tor General Power Systems,IEEE/PES WirirrrMueririg, NY. (1980)Papx
043-0.
Tamura. Y.. Mori, H., and Iwamoto, S. Relationship Between Voltage Instability arid
Multiple Load Flow Solutions in Electric Power Systems,IEEE Trcrrls. 011 PAS.
Vol. PAS-102, NO.5,(May 1983).
Taniura, Y..Nakanishi, Y. , and I~~amoto, S.Onthe MultipleSolutionStructure,Singulat-
Point and ExistenceCondition o f MultipleLoadFlow Solutions,IEEE/PES Wiri/or
Mret i ri g. NY,(1980)Paper 043-0.
Tamura. Y..Sakamoto. K. , and Tayatiia, Y. Voltage Instability Proximity Index (VIII)
Basedon 111-ConditionedPower Systems,Proc. of tlir Z7tli coi$ o i i Dc~c*i.sioii r r i i t l
Coiitrol. Austin, TX,(Dec. 1988).
Tranuchit,A. andThomas, R. J. APosturingStrategyAgainstVoltageInstabilityi n Elec-
tric PowerSystems.IEEE Trcriis.o i i P o ~ v r Applri-.Sj - st . , (Feb.1988)pp. 87-93.
Tamura, Y. , Mori, H., and Iwamoto, S. Relationship BetM,een Voltage Instability and
Multiple Load Flow Solutions in Electric Pomter Systems,/ E Trriiis.o i i Po\t.c.i*
AI7I>lir.Syst., Vol. PAS-102. NO.5,(May 1983)pp. 1 1 15-1 125.
Tinney, W. F.. Brandwajm, V.,and Chan, S.M. Sparse Vector Methods. IEEE Trr1ii.s.
Vol. PAS-104,No. 2, (Feb.1985).
Tomsovic, K., Liu, C. C.. Ackerman, P., and Pope, S. An Expert System Dispatchers'
Aidforthe Isolationof LineSection Faults. / E Trriiis.o i i Port~i- Dcl i \ x~i : \ ~. ( J ul y
1987)pp. 736-743.
Treinen, R.T., Vittal, V. , and Kliemann, W. An I mpr o\ ~d Technique to Determine the
Controlling Unstable Equilibrium Point in a Power System, / E Trciiis. oi i Cir-
c-irits criid S?.steim-l.Firiichnerital Tlieoi-yt i t i d Applic*citioii.Vol. 43. No. 1,(Apr.
1996)313-323.
Toomey, L.J. IBM Systeiii.sJoitriznl Vol. 27,No. 4 (1988)pp. 416335.
Tou, J. T. and Gonzalez, R. C. Pcltterri Recoguitioii Pri ri ci pl c~s,Addison Weslqr. MA
(1974).
Tranuchit. A. and Thomas R. J. A Posturing Strategy Against Voltage Instabilitj, i n
Electric Power Systems,IEEE Trctiis.oi i Porc~)i- Appcri'. Sj*.st..(Feb. 1988).
Tucker, S.G.The IBM 3090System: An Oveririew,IBM Sj,.stcvii.sJoiri-iiul Vol . 25. No.
1 ( I 986)pp.4-20.
Tyla~tsky, D. J., et al . Frequency Domain Relaxation of Pob\rer System Djwmiicx. I n
Proceedings oftlie I989PICA. (May 1989)pp. 387-393.
U.S.EnergyResearchandDevelopment Administration. SystemsEngineering forPourer:
Status and Prospects. Proceedings Etigitieeritig Foiriihtioii Coi!fc.imic*r.CONF-
750867,Henniker, NH. (Aug. 17-22, 1975).
Varaiya, P.P..Wu, F. F..andChen,R.-L. Direct Methods forTransient Stability Analy-
sis of Power Systems: Recent Results. In Prc~cecc1iiig.s IEEE 73(December1985)
pp. 1703-1715.
Venckov, V. A., et al. Estimation of Electrical Power System Steady State Stability i n
Load Flow Calculations, IEEE Traits. Pobtper Appui-. Sysr., Vol. 94, (May/June
1975).
Venokov. V.A.. Stroev,V.A.,Idelchick,V.I . , andTrasov.V. I.Estimationof Electrical
Power System Steady State Stability in Load Flow Calculations, IEEE Ti.rrii.s.
Pouvr Apprr. Syst.,Vol. 94,(May/June 1975).
Venkatasubramanian, V.,Schattler,H. , and Zaborszky,J. Voltage Dynamics: Stud).of ;I
Generator with Voltage Control, Transmission and Matched MW Load. /EEL-
Tl.(iii.\..Al/toiti.Coiitr.,Vol. 37,(NOV.1992)pp. 1717-1 733.
Verghese, G. C., Perez-Arriaga, I . J. , and Scheweppe, F. C. Selective Modal Analysis
with Applications to Electric Power Systems, Part I and 11, IEEE Ti-trri.~. P ~ H Y I .
A p p r . Sj ' xt . , PAS- I0I (Sept.1982)pp. 31 17-3 134.
Vidyasagar, M. Noiiliiicwr Sj:stet?zsAiirilwis, Prentice-Ha11, Englewmd CIiffs, NJ.
(1978).
Vittal. V.. et al. Derivation of Stability Limits Using Anulytical Sensitivityof the Tran-
sient Energy Margin. Ptipor- No. 89IVM207-2PIVRSI EEEPES I989Wi t i t et . Mror-
i t i g , New York.
Vittal, V.. Bhatia, N., and Fouad, A. A. Analysis of the Inter-Area Mode Phenomenon
in Power Systems FolloN+lg Large Disturbances, Ptrprr No. 9/ lVM228-7-PlVR.5
pwsetited Lit the I991 Po\\,rt. Etigiticvritig S o c ~ i c ~ r > ~ \\'ititor Mwt i t i g , New York.
(February 199I 1.
Vittal, V.. Fouad,A.A.,and Kundur,P. Deterniination of'TransientStability-Constrainec
Plant Generation Limits, Pt *ocwdi t i g. sof't l i o IFACS j w p o s i i ~ t ~ i o t i , 41rrot 1i t i ri ot i l i t u '
Iti. sfnrtrierittrtioti $) I . Po,r.rr Pl t i t i t s, Bangalore. India. (Deceinber I 5- I 7. 1986)pp
A-8- I through A-8-5.
Vittal. V.. Prabhu, G. M.,and Lini,S.L. A Parallel Cornputer Implcrnentation of P o u ~ .
System Transient Stability Assessment Using the Transient Energy Functioii
Method,IEEETrcitis.Pou-er-Sjssr.,Vol.6,No. 2, (February I99I ) pp. 167- 173.
Vulnerabilit>'to Voltage Collapse. IEEE 7 i i i t i . v . o t i P o r ~ ~ t . Sj,.sr.,Vol. 5. No. 2. A p 1 ~ t ~ .
(Maq' 1990).
U'nda, N.et al.A Real TimeExpert System forPowerSjrstcm FaultsAnalysis,IAS7f i 1)
Corzfi~tw1c-e. Bo~eni an, Montana (August 1986).
Weerasooric.a,S. and El-Sharkau,i. M. A. Towards Static Security Assessment of Lurg:
Scale Power Systems Using Neural Net wxks. IEEE Pt.oc. C: Goriot-tiriori,Ttwi,(-
~ititll>/.~frihl1ti(>/i, / t i i , ~ . ~ i ( ~ t i Vol. 139.NO.1 (Ja11uar)' 1992)pp. 64-70.
Weerasooriya, S. and El-Sharkau,i. M. A. Dynamic Security Assessrnent of Poicw Sj " -
tenis Using Neural Networks, I t i t . Cot$ o / i E.spet.r .Sj..srctii.sApplic.., f i ) t . t he El cct t . / ( ,
ltitlii. ui:\~.Phoenix. AZ(December8- 10, 1993).
Weerisooriyi, S.and El-Sharkawi, M. A. Useof Karhunen Loek'e Espansion i n Training
Neural Networks for StaticSecuriryAssessment, Pt-oc.. I " I t i t . For - i ut i o t i NNA P,;.
Sciittle,WA (July 1991)pp. 59-64.
Wecrisooriya, S.Application of Neural Networks for Classificotion and System Identif i -
cation i n P o u wSystems. Ph.D.Thesis, Uni \ w-si t ~, of Washington (1992).
Wehenkel. L., Van Cutsem. Th. , and Ribbens-Pawlla. M. An Artif'icial Int el l i genc~
Frame\+*orktor On-Line Transient Stability Assessmcnt of P o w r Systems. I EkE
Tturis. P o \ i ~ r Apptir. Syst eni s, (Ma 1989)pp. 789-800.
Werners. B. Aggregation Models in Mathematical Pt-ograin~ning, i n Mr i t l i cwi t i t i c~r lMO)/-
ols,fi)t. t k . i . s i o t i Sirppot.r, G.Mitra (ed.). Birkhcluser, Berlin, ( I988),pp. 295-3 19.
Willems. J . Direct Methods for Transient Stability Studies in Power System Anal\,s\.
/E&? 7rtrti.s.A ~ m t i . Cotirrol,A C-16.4,(July-Aug. 197I ) pp. 1469- 1-48I .
U'i t h NoseCur\re usingHomo t opy Conti nuatio11Method,1EElElPE.S IY90Si r i t i t ~ i o r illcot -
i t i s , Minneapolis, Minnesota, (July 15-19, 1990).
Wollenberg, B. and Sakaguchi, T. Artificial Intelligence i n Power System Operatioii\.
P t ~ o ~ ~ r ~ ~ l i t i ~ q . ~ of'rlie IEEE, (Dec. 1987)pp. 1678- 1685.
W'ollenberg, B. F. Feasibility Studyt or anEnergy Management System Intelligent Alurw
Processor.IEEE Trtitis. Po\r,c>r Sj xt . , (May 1986) pp.231-247.
Wood and Wollenberg. B. F. Porc.or- Getior-titioti,Ope ~t ut i ot i& Cotitt-01. John Wile!, &
Sons. Inc..New York, ( 1984).
Woodson. H. H. and Melchet-, J. R. El c ~ c ~ t t ~ o t ~ i c ~ ~ ~ l i t i t i i ~ ~ t r l - l l j w t i i i c , v Ptrrr I: Il i . sc. t - t ~roS,,..\
t t ' rt i s, John Wile) &L Sons,Inc., New York. ( 1968).
Xie.G. and Ranishaw. R. S.Nonlinear Model of Synchronous Machines u.ith Saliency.
Plr1Jc.i. 85SM 347-0,IEEE/PES 1985S//iiiiiici*Mrc)riil,q.V~il1cOu\eS, Caliiida. (JuI!
14-19. 1985).
Xue, Y., Gao. Z.. Zhu, J., and Liu, J. A Planning Decision Support Expert S \ W m for
Transient and Dynamic Security of Power S!stenis. Pint, . 1080 E.\-lwi.t.Sj..vtt~iii.\
A1/Jli(.t/tioii t oPO\ I~OI. pp. 21 8-224. S!~,~tlrii.~,
Young. C. C. and Webler, R. M. A New Stability Prograni for Predicting the D! naiiiic
PerformanceofElectricPowerSystems.Pi.ocwtii1i~q.s Po, i x)i .C oi i - c!/ t l i ci A i i i i ~ i . i t ~ i i i
j i ~l . r i i c ~. Vol. 29,Chicago. (1967)pp. 1 126-1 138.
Young, C. C. Equipment and System Modeling for Large Scale Stabilit) Stuclics. I EEE
Ti-t/ri.s.PO\\~OI- A//c/r.Sj . ~t . . PAS-9I ,I , (Jan./Feb. I972) pp. 99- 109.
Young. C. C. The S!mchronous Machine, IEEE Tirroi-itrlC or i i ~. s 1~ o i i h9otliw Coiic.cpt.\(!/
Po,ivr Sj%. striii1Ij wrri i i i c. s. 70 M 62-P\VR, Institute of Electrical and Electronics
Engineers. Ne ~ c York, (1970).
Young, T.and Fu, K. S. Hrriirlhook cf Prrttriw Recogiiirioii l i i i t rLqci Pi ~o c ~. s . v i i i ~q . Acxlemic
Press, New. York, ( 1986).
YU, Y. N. Elrc.ti-icPo \ i ~~i * Sj ~s t e i i iq\ wci ni i t . . s , Academic Press, New Yorh. 1983.
Zadeh, L. A. Fuzzy Setsas a Basis fora Theory of Possibility. F1i::j. Set.\ t i i i t l .Sj.,\rciii.$,
Vol. I . NO. I , ( 1978)pp. 3-28.
Zadeh. L. A. Fuzzy Sets.li!fi)riiiririoiI ciiitl Coiiti-ol.Vol. 8.( 1965)pp. 338-353.
Zadek.Lofti A.KII o\\,/orl,q11 Kcpr.c).ser 1tatioi 1 i t i FiI : : y Losi c .,KIUn. er Acaden i i c PubIi h ers.
(1992).
Zhang, B. S. and Edmunds. J. M. Self-Organizing Fun! Logic Controller. IEEE Pro-
c ~c ~c ~t 1 i i I g . s I ) , Vol. 139.No. 5,(1992)pp.360464.
Zhang, Y., Chen. G. P., Malik, 0.P., arid Hope. G. S. An Artificial Ncuriil Networh.
Based Adi1ptiL.e Poww System Stabilizer,l EEE Ti ~r r i i . ~i r t ~t i o r i . s Elici;q\. C o i i i ~ i - - 0 1 1
. s ~ o I ~ , Vol. 8, NO. I , (1993)pp. 71-77.
Zhou. Q., DaLiison, J., and Fouad. A. A. Application of Artificial Neural Net\\~oi-kxi i i
Pou.er System Securityand VulnerabilityAssessmcnt, lEEE/PES / Y Y j blf.(Jan
3l-Feb 5 , 1993).
Ziminermann. H.-J.arid Zysno, P. Latent Connectives i n Human Decision hlahing. Fii::j,
S1t.S t ri l l 1 S j ~ . ~ t r i i 1 . ~ . Vol. 4, No. 1. (1980)pp. 37-5 I .
This page intentionally left blank
Index
cihc-Phase sequence, IS, I6
Artificial neuron (seeNeuron)
Accelerating power. 72
Automatic Rule Generation. 253
Accelerating torque, 36 Automatic Voltage Regulators ( . s w
ActiLtationfunctions, 182
AVR)
Activation level, 182 Average power, 12
Active load, 5 AVR, 98. 255, 1.35
Actibre power. 12,2I
Axon, 181
AGC, IS8
AI (seeArtificial Intelligence) Back propagation, 183. 188
Alpha cut,(a-cut),211
Balanced 3-pha\e sy\tems, 15
Angular momentum,38 Base value. 26
ANN ( . s m Artificial Neural Network) Belief measure. 2IS
ANN-Based TEF,222
Bellman. 206
algorithm, 223
Benefit-to-cost index, 9
learning process, 222
Bifurcation point\, 148
Antecedents, 250
Bifurcation theory. 9, 141
Apparent power, 13
Branch switching. 87
Approximate reasoning. 214-2 15
Bulk power system,3
ARG (.seeAutomatic Rule Generation)
Armature windings. 22
Carpenter, 184
Artificial Intelligence, 221, 259, 296 CCT(seeCritical clearing time)
Artificial Neural Networks, 9, 177, 178
Cell body, 181
architecture, I80
Centerof inertia, 122
3s1
Certainty lxtors. 20I
Chuos theory. 292
Clustering. I84
Coriiplcx potvcr. I I . I3
CO1( SCT Centcr ol incrtiu)
Condition nunihcr. I43
Confidence scnles (.w,Certainty l ~i ct ors)
Conjugate citrrent. I3
Contiriti:itioii nicthod. I 4s. I47
Control \ alvc position.42
Corrector. 90. I49
Counter propagation. I X4
Critical clearing time. 234
Cylindric;il rotor synchronous ni;ichinc.
55
Dcl'w/.ificurion.2. IS. 219. 249
I)elt~i-conncctioti.I X
1)ciidritc.s. I8I
Direct axis. 56
Direct axis synchronot~srciict;tncc. 54
1)istrihutionsystctii. 4
JqO tnasloriiintion. 24
Dynaiiiicsecurity. 289
I1yi;tinicsecurity ;isscsc~iicnt(DSA). 8.
22 I
Dynamic stability. 164
Dyiuiiiic voltogc stnhility. IS7
nlgorithlii. I hh
cquilihriitnipoint. I hl
I'tist subsystetns. I6I
linc:iri/at ion. I62
lo;d Ilou. I hI
\ IOW sllhsystclns, 160
EdisonElectric. I
EHV (.we Exwct high t~olt~tgc)
Eigcnvnlucs. 114. 156
Elcctric;iltorque. 46
Electricpoucr systeni. 3
l w/ y theory. 209
EI t ~ ~ t r o ~ i i c ~ ~ hi t ~ ~ ~ c ; ~ l Iriiii~icJil.I 09
Energy titnction. 127. I30
Energy Systcnis Network Lnhoratory.
97
Eqi d;irc;i criteria. 73-75
Error signal. I87
tS ( s w Expertsysteni)
ESNL ~ s wEnergy systctiis network lah-
oratory)
Euclidctiiidistiince. I78
Euclidean nortii. I47
Excitationcontrol. 6
Excitationcontrol systcni. 37
Excitationsysteni. 40
nlodrl. 36
Excilcr niodel. 37
lineardifferential cquntions. 39
Expert rcnsoning.2 I4
Expert systeni. 9. 176. 191
characteristics. I92
ilcfinition. I91
DSA. 238
iufcrciiccengine. I9I
kno\vlcdgc acquisition. 273
knowledge base. I I
riinn-niachincintcrl'clce. I91
nieiisiircs. 200
prohuhility. I99
rule-b;iscd structure. 740
subjective proh;ihility. 199
E?ctrn high wltnge. 2. 135. 2.59
FACTS. 1%
FIRE ( . s w Fui i y Inlcrcncc Reverse En
piticering)
FL ( . s w F u z ~ ylogic)
FL,C ( s c o Fuuy LogicControllcr)
FLPSS(sw Fii/./.y Logic Power Systeiii
Stiihilizer)
Flux. 55
Flux linkages.23
Four-rnxhincpower ry~ccrii.239
Frcqueiicy dependent lond. 34
Fuui ficntion. 2-10
F u z ~ yinlcrcncc. 2s I
Fui / y Inlcrcnce Reverse Engineering.
254
Fuzxy logic.9. 177. 204
correlatioti niethod. 2 I8
linguisticviiriahle. 2IS
iiionotonicnicthod. 2I7
IFu//y logicI
paraiiieter t11ning.252
power \y\teni \tabili/er, 248, 255
rule\, 252
Fu//) Logic Controller. 248, 349
Fw/ y propmition, 2I6
Fwz y\et\. 206
definition,2I3
operation\, 2I2
Fw/ y \et theorj. I77
Gate opening, 40
Generator
real poiaer. 1I8
reactance\.54
\hart circuit current, 53
Generator model, 52
Generator voltage,6
Generic electric po\ver \ptriii.4
Global er i wfunction. 189
Gro\\berg Outstar algorithm, I84
Hidden layer. 187
Hidden node. 226
High Voltage DC linh\, 133
Hi gh-VoIt age Di rect Current ,2
HVDC tran\nii\\ion control, 7
HVDC ( $ 0 1 ) High-Voltage Direct Cur-
rent)
Hqdraulic turbine model. 40
IEEE 5 7 - b ~ \ te\t \ Y\ ~CI TI .278
IEEE 39-bujsystem, 227. 229
IEEE \sorhing group, 134
II1ductance
mut ua l , 23
\elf, '3
Induction motor. 43
equi\dent circuit.44
\hp. 44
torque. 43
Inertia,46
Inertia coii\tant, 38,47-50
Inferencechain, 196
Inference engine. 175. 196.276
Inference. rule\, 3I7
Infinite bu\. 63. 68, I 8
In\tantaneou\ current. I I
In\tnntaneou\ PO\\ er. I
II1t egral-\y11are error, 25
Integration method\. 9
Int egratIon techn Iqiic\
Euler method.91
modified Eulrr method. 9I
predictorcorrector. 89
RUnge-Kutta method. 9. 94
4;in1pIIng incthod,(17
trape/oidal method. 9. 94
Ua\eform rt'la\iallon. 9
Intelligent \ j \tern\, 173
appronche\. I79
Jacobian matri\, I4I . I43
reduced, I56
KB ( ~ C OKIN)\\ledge haw)
KBVCDP ($OO Kilo\+Icdgc-Ba\ed Volt-
ageCollap\c, Detectionand Prc-
Lention)
h-fi n i te difference eqLI;I tIon\,89
Kinetic energj,17-48
Kirchoff' \ lau. 16
Knowfledgebnw. 176, 194. 249
Kllouledge-bi\\ed\>\teiIi. 195.338
Knowledge-Ba\ed VoltageCollap\e. De-
tection and Preiention. 373
implemcntatIon, 274. 378
5tructural de\lgn. 375
Kohonen laqer. I84
Learning algorithm\. global error. I83
L Index, 140
Linear prograinming. 9
Linear sy\tem
sen\iti\i t ) . I44
Line \ dt age. 16
Load characteri\tic\. 31
Load model\
con\tant po\+er. I49
dynamic. 43. I58
linearmodel. 150
Load paran1etcrs. 33
3.54
Load representation. 85
Local stability. I 1 1
LP (.weLinear programming)
LTC, 277
LU decomposition, 148
Lya punov
function, 115
theorem. I12
Machine models, 59
Machine rating, 39
Machine switching, 88
Marceau, 290
Maximum norm. 147
Mechanical torque. 46
Membershipfunction, 21 I , 250
Mininiiini singular value. 142
Min-niax inference. 2I7
Mismatch, I45
Modal analysis, 156
Models
exciter, 38
generator,52
gobw-ningsystem,40, 42
hydraulic turbines, 40
PSS, 39
static electricnetwork, 10
static loads. 32
synchronousmachine, 2I
terminal voltage transducer, 39
\ dt ageregulator, 38
Modified Eider method. 91
Mutual inductance, 23
Necessity measure. 215
NERC (,weNorth American Electric Re-
liability Council)
Neural networks, 221
learning algorithms, 183
training, 183
Neuron, I80
New England 39-buspower jystem,
264
Newton-Raphson. 145. I56
New Yorh power pool, 132
n-Generator SystemModel, 1 17
Nodal matrix method, 83
Normalization (seeper uni t representa-
tion)
North American Electric Reliability
Council. 3
OPF( seeOptimal power flow)
Optimal pouwtlow, 9
Parameterization, 151
Participation factor, 157,262
Pendulum, I14
kinetic energy, 115
potential energy, I15
Per unit representation, 25
PF(see power factor)
Phase diagram. 13.56-60
Phase-shift transformers, 7
Phase voltage, 16
PID(.seeProportional-lntegral Deriva-
tive)
Plausibility measure, 215
Post-fault curve,81
Power angle cur\re, 70, 120
wound rotor machine, 66
Power factor, 1 1 . 12. I4
Power failures, 132
Power tlou,, 165
linearired model, 165
Power interchange. 6
Power relationships. 20
Power system
basic elements,5
complex power concepts, 1 1
security assessment, 45
three-phase systems, 14-2 I
Power system stability, 7
Power system stabilixrs,38, 158
Power triangles, I3
Predictor, 149
Predictor-corrector step, 147
Pre-fault curve,81
Prime nioiw,40
Prime mover controls,6
Probability measure, 215
Proportional-Integral Derivati\,e. 254
Proposition, 216
PSS ( s c v Power system stabilizers)
3-55 1iide.r
p.u. (seePer unit representation)
P-V curve, 146, IS1
QP(.seeQuadratic programming)
Quadratic programming,9
Quadrature axis,56
Quality of power, 5
Quantization,210
Q-V curve, 146
Q-V sensitiktity. 1.56, 167
Ray.Jame\ J. , 97
RBS ( WORule-based system)
Reactive capability limits, 31
Reactive load, 5
Reactice power, 6, 13. 21
Reactors, 7
Robu\t ANN. 222
Rotor voltage, 21
Rule-based system, 194
Rule\
divergence. 276
\electing modeh for VCdetection.
277
selecting VC parameters, 276
Runge-Kutta. 9, 95
Saliency effects,65
Salient pole synchronousmachine, 56,
57
Self inductance. 23
Semitivity analysi5. 146
SEP( t e e Stable equilibrium point)
Sigmoidal activation function. 225
Sigmoidal function, 182
SIL. 13s
Speed regulation, 6
Squashing function, 182
S[ability
angle, 8
BCU. 131
boundary, I27
definitions, 1 1 1
equal area method, 72
lemma, 112
Lyapunov's theorem, 112
positice definite function, 112
[Stability]
SEPs,260
simple pendulum, I14
single-machine system, I 18
UEP, 129
voltage, 8
Stability assessment. 63. 71-82
fuzzy expert systems, 205
TEFmethod, 121
voltage, 132
Stable equilibrium point, 69, 130.260
Starconnection ( . see WYEconnection1
Static assessment
algorithm, IS3
Static load model, 32
Static security assessment, 8
Staticstability. I63
Static varcompensators,6
Static voltage compensating deiricea,
133, 158
Static voltage stability assessment
modal analysis. IS8
Stator voltage, 21
Steamturbine. 42,43
Steamturbine model.42
Sub-transient equations 59
Sub-transient reactance, SO. 53
Sub-transmission system,4
SVCs(see Static voltage compensating
devices)
Swingequation,46, 72
Switched capacitors,6
Synapse, 181
Synchronouscondensers. 6
Synchronousmachine. 3
classic model, 27
dynamic model, 29
equivalent circuits, 29-30
rnodeling, 21-30
network representation,84
rotorcircuit, 22
statorcircuit, 22
steady state model. 30
subtransient niode1. 30
transient equations. 58
transient model. 30
voltage equations,22
S j iichronoii\ machine mockling. 21
Sy\tttni haw. 26
S ) \tern t-lllilt\. 86
Tamiira, Y .I 4 1
Tangent \ector. 150
Tap-changing tran\toriner\. 7
Target \alue. I87
Ta! tor eup,in\ion. 93
TEF(wt' Tran\ient encrgj tiinction)
Three-phaw pofier, 20. 2I
Three-phaw \)\tern. 14
Thre\hold tiinction. 182
Time traiiie\
rllodel\. I39
RLC component\, 139
Tiaining algorithin\. 184
Gro\\bcigOut\tar. I84
Kohonen. I84
Training proce\\. 263
output ~ect or . I83
Traii\ient Jroop.4I
Ttaii\ient energy tunetion, I 10. 120,
185. 212
tlouchart. I28
inte~ral-\qiiarederror([SE). I22
w t e d energj. I23
Trim4Ie11t ccii i ation\.58
Trnn\i t ' nt Ie;lct ance,53. SO
Tramient 4tabillt). 68. 109
algorithm. 92-93. 98- I01
concept\. 68
Tran\ient \ t'i biIit j ;i4\e\\iiient. 222
Tran\ient \Uing. I21
Trun\nii\\ton networh,4
Tr,ipe/oid:iI iule,0
Trllth tLinctloti\ ( S O P MeilIber\hip tliiic-
tion1
Turbo generator
high pre\\iire, 49
Ion prt.\\ure. 49
UEP ( , weUnxtableequilibrium point)
LlLTCs. 135
Llncertainty. I97
Unxtablc equilibrium point. 69. 130,260
UnxiiperLi d training. I83
VAK, 265
VC (WO\oltage collapw)
VIPI. I41
Voltage collap\e. 134. I35
cla\\itication\, I 34
deflnl(1on. I33
detection, 1-18. 272
pietliction method\. 138
predictor \tep, I SO
prevention.272
time-frame. I36
Voltiige i i i \ t ~i b~l ~t j , 159(\('(I Voltage \t;i-
bilitj problem\)
cla\\itication, 159
Voltage iii\tabilit) indicator\, 140
Voltage regulator. 37
Voltage \tabilitj ;i\\e\\iiicnt. 132. 165.
259
ANN-ba\ed, 260
technique\. 140
time-traiiie. 139
VoItage \tabiI i t y
Inodellng. I36
pit. \cnti\c control, 168
tiiiie-franie (Cnr\onTaj lor).I37
Volt,ige \tabilitj problem\, 138
cla\\ilication. I38
Voltage tr,in\duccr. 38
V- Q \ Ci i \ i t t \ i t ) , I57
VSTAB, (EPRI). 105. 263
U'aler hammer effect,4()
Water 4tarting tiine. 40
Wi i \ ~ forin reI;ixation 9
Wjv-conncction. I6
Y-connection, IS (WJWye-connectic11)
Y-Bus or admittance matrix. 85.
118
Zadeh. 177
ZIP load iiiodel, 34