OKE-5116-103 1

DESIGNING AN INTELLIGENT VOLTAGE INSTABILITY DETECTION

SYSTEM IN POWER DISTRIBUTION SYSTEMS IN THE PHILIPPINES
USING IEEE 14 BUS TEST SYSTEM
CHAN, Ian Benedict, GRAY, Janric Micah F., OCAMPO, Anne Bernadine S., RODRIGUEZ, Pocholo L.
Electrical Engineering Department, Faculty of Engineering, University of Santo Tomas
España, Manila, Philippines

Abstract - The state of an electric power system may be classified as either stable or unstable. The borderline of stability is at any
condition for which a slight change in an unfavourable direction of any pertinent quantity will cause instability. Voltage instability
in power distribution systems could lead to voltage collapse and thus power blackouts. The researchers will present an intelligent
system using back propagation algorithm which detects voltage instability and classifies voltage output of an assumed power
distribution system (PDS) as: stable or unstable.

The researcher’s work is the use of voltage output readings as the input patterns to the neural network for training and
generalizing purposes, thus providing a faster instability detection system that simulates a trained operator controlling and
monitoring the 3-phase voltage output of the assumed power distribution system.

Index Terms— Load Stability, Phase Synchronization, Voltage Collapse, Neural Network, Power Distribution System, 3-phase, Back-
Propagation Algorithm
pronounced, the system integrity is endangered mainly due
to protecting devices that trip generation, transmission, or
I. INTRODUCTION1 load equipment. This degradation process may eventually
lead to a blackout in the form of a voltage collapse. (Glavic,
T he study is focused primarily on researching viable
methods for applying real-time voltage instability
detection algorithms, which may be used to improve the
M. & Cutsem, T.V. 2011).

existing models currently being used by substations here in

the Philippines. voltage instability detection algorithms have
been in development starting as early as the 1930’s, in high
voltage transmission lines. since then, there had been major
discoveries on the different parameters which may lead to
these so-called “instabilities”. the study aims to modify and
configure these algorithms, and hopefully be able to apply it
to the systems here in the country. Power system voltage
instability results from the inability of the combined
transmission and generation system to deliver the power
requested by loads.
In many power systems throughout the world, voltage
instability is considered as a major threat for secure
operation. In a voltage unstable situation, the voltage drops
undergo a dramatic, generally monotonic, decline in the
minutes following a disturbance. When this decrease is too
1 Rodriguez, Pocholo L. was with University of Santo Tomas,
Philippines. He is now with Colegio de San Juan de Letran, College of
Engineering and Information Technology (email:
rodriguezpocholo12@gmail.com)
Ocampo, Anne Bernadine S. was with University of Santo Tomas,
Philippines. (email: Ocampo.annenadine@yahoo.com)
Chan, Ian Benedict was with University of Santo Tomas,
Philippines.
Gray, Janric Micah F. is now with University of Santo Tomas,
Philippines.
II. METHODOLOGY
A. Research Method
The application of an intelligent voltage instability
detection system in distribution lines in the Philippines
requires a research method which focuses on the
quantitative aspect, but at the same time utilizes a certain
degree in qualitative. It is qualitative since the research
requires initial data system from the Institute of Integrated
Electrical Engineers (IIEE) and interview gathered from
engineers working in the Manila Electric Company about
their observations and concerns about the voltage
instability occurring in the grid. A test system for the
IEEE 14 Bus system was used for the data that is needed
for the training of the model that will be used for the
prediction of the voltage instability. These initial
parameters of the test system were altered and varied
using MATLAB. The values obtained will be the
information needed to be trained to the WEKA program in
which the researchers will use for the detection system of
the voltage instability.
From there, the additional information accumulated will
require supporting details. These details are where the
quantitative research method comes into the picture. The
said research method will have a major part in the study
 since the numerical data will be used in order to generate A. Training Phase
facts and discover patterns in the research.

B. Research Design

This study aims to apply an intelligent detection system

for voltage instability in transmission lines going to the
distribution system in the Philippines. It exercises the use
of quasi-experimental research design, particularly the
Single-Sample Repeated Measures. Since IEEE 14 Bus
system is a reliable source is “the world's largest technical
professional organization dedicated to inspire a global
community to innovate for a better tomorrow through its
highly-cited publications, conferences, technology
standards, and professional and educational activities.
IEEE is the trusted “voice” for engineering, computing
and technology information around the globe” (IEEE,
FIGURE 1.2. Testing Flow Chart
2016). Through that, the researchers are going to
manipulate some of the dependent variables to utilize the
For the training phase, this is the phase where the group
effectiveness, performance and efficiency of the said
system. This research focuses on aspects such as “Why” builds the model/system. The training points are fed into
and How” of things. In order to check the criterion ANN (Multilayer Perceptron) classifier and did a validation
mentioned, the researchers will use the programs WEKA testing (model testing). The technique used is 10-fold cross
and MATLAB to simulate the voltage instability in the validation technique. Cross validation partitions the training
lines, and to manipulate the necessary variables. dataset into 10 equal subsets. 9 of the subsets are used for
the training model and the 1 left is used for validating and
After using the said programs, the researchers will now testing the model, repeated for K (10) times (called as
use the analysis gathered in order to develop an intelligent folds), then each of their results are averaged to get a single
system. The system created is based on the results of the estimate. (Note: Each subset is used as validation set).
program which has been manipulated to maximize its
effectiveness, performance and efficiency with respect to
the voltage instability in the distribution system. B. Testing Phase

III. FLOWCHART
Flow chart of WEKA program of a simulation of voltage
instability predictor:

FIGURE 1.3. Testing Flow Chart

FIGURE 1.1: Full Flow Chart
Based on the whole flow chart, on the Data part, the
group separated the number of instances into 80% training
and 20% testing and it is fed to the training phase.
For the testing phase, the researchers used command
prompt to type in the prediction testing script, then the
results are transferred to excel file, then performed
accuracy. The result of this process will then be evaluated
by the researchers in order to see the accuracy of the
prediction of the system before incorporating it to
distribution systems.
OKE-5116-103
The accuracy will depend on the number of data
processed by the system and is then subject to improvement
as stated in the recommendation of the paper.
IV. SIMULATION RESULTS
Designing a system requires training and sample points
that will be made as a basis of the system. There are data
that needs to be properly checked and evaluated whether it
is correct or not. In order to check the efficiency of the
system, with the help of the group’s chosen program, a
legitimate sample of dataset was obtained from WEKA. The
data set that the researchers will use will come from the
result of the simulation of the IEEE 14 Bus Test System in
MATLAB.
The data set will then be used for the training of the
model from WEKA that will be used for the detection and
prediction of the voltage instability. All the buses included
in 24-hour load profile table are load buses. The following
buses included in the load profile will be subjected to the
testing process to be incorporated in the code in WEKA’s
interface (see Appendix A for 24 hour load profile).
The extent of the detection system for voltage instability
covers the behavior of the load buses for 1 day (24 hours).
The initial data for the system was accumulated from a
reliable source which is IEEE 14 Bus System 24-hour load
profile. As observed, the load (P & Q) is varied every hour
for 24 hours. The values where used as the current load data
of respective buses in WEKA, subject to change per hour.
A. MATLAB Load Profile
TABLE 1.1 6AM Load Profile
The sample case of IEEE 14 BUS SYSTEM indicated
above represent a brief representation of the initial stable
values used in the system of WEKA. This is incorporated in
Matlab in wherein the parameters under study are voltage,
load real power, and load reactive power.
As such, the basis for load parameter are from the 24-
hour load profile, and the researchers used MATLAB so
that variation of the load will correspond a reliable voltage
change based on Gauss-Seidel Method.
Also by referring to the table above, the 14 buses have
different types: Slack Bus, Load Bus, and Voltage-
Controlled. Slack Bus is implied as the reference bus and
must be capable of supplying total losses in the system. It
3
has a voltage value of 1.0 + j 0 P.U. The voltage-controlled
bus however, has voltage control capabilities. Instead of a
generator, it has tap-adjustable transformer and/or a static
VAR compensator.
The load buses are the ones subjected to study in this
paper since real power and reactive power are known
variables. This is because the researchers intend to vary the
load and locate and study what instance of load does it
cause a voltage instability.
These load variables are obtained from IEEE Bus 14
System, and since voltage magnitude and voltage angle are
unknown, this instance prove to be a good scenario for this
thesis so that voltage behavior can be observed.
The researchers have then varied the IEEE 14 Bus Test
System 24-hour load profile with respect to a certain
percentage in its load and fed the data to MATLAB in order
to generate the data needed and to determine on what
percentage of load variation the load instability occurred.
TABLE 1.2. 6AM Varied (10%) Load Profile
Table 1.2 indicates a case of load profile in which the
researchers have taken into consideration in the detection
system. Since an increase in load is a factor for
susceptibility in voltage instability of the system, the load is
then varied. As stated in “Voltage Stability Assessment in
Power Network Using Artificial Neural Network,” an
increase in 5% in the voltage can already be considered
unstable. Keeping that in mind, the researchers modified the
initial load parameters (P & Q) from the 24-hour load
profile in IEEE by an increase in 10%, 30%, and 50%. This
is to further analyze what specific increase in load will have
a counterpart effect to make the voltage unstable. Looking
at all the tables, it is observed that Bus 14 proved to be the
weakest link among all the bus since it experiences constant
voltage violation for every increase in load for all cases.
The red highlight indicates that the specific bus exceeds
the nominal voltage by a minimum of 5%. Bus 14 has been
consistent for having a red highlight which means it is
sensitive to voltage violation with just a mere increase in
load. Buses 1, 2, 3 6, 7, 8 are the buses that least likely for
voltage violation to occur. In all the cases, even an increase
of 50% in the load, the said buses still don’t experience
voltage instability. The rest of the buses that wasn’t mention
have at least once experienced voltage instability.
 B. WEKA Data Attributes For next value of load reactive power in bus 2, the value
is 0.05
And so on and so forth for the next buses.

C. Artificial Neural Network

FIGURE 1.3. WEKA Data Attributes

The figure above shows the code inputted by the

researches needed for training the model in WEKA. The
code shows the prediction training of the values which
determines whether the given set of values is stable or
unstable. They indicate the condition or the classification
and declare the type of parameter of a given set of inputs.
These attributes are arranged vertically and shows the
parameters such as voltage, real power and reactive power
of the load per bus. The buses included in the detection
system are Bus 2, Bus 3, Bus 4, Bus 5, Bus 6, Bus 9, Bus
10, Bus 11, Bus 12, Bus 13 and Bus 14. Theses buses are
under the category of load bus that’s why they are subject to
observation and analysis in the study. Bus 1, Bus 7, and Bus
8 are not included in the code because they are slack bus or
voltage-controlled bus wherein their voltages are constant
and can be manipulated. Looking back at the figure, the
“@attribute cur_” of voltage, voltage angle, load real power
and load reactive power means that it is the parameter’s
current and initial values. It is their data when t=0. The
“@attribute next_” code however indicates the parameter’s
next values after being subjected to a change in time. For
easier understanding of data, these represents the values of
voltage, and real and reactive power when t=1. As such,
each data of load bus is taken into consideration in the code.
FIGURE 1.4. Generated Artificial Neural Network
The Neural Network of the dataset is obtained from
WEKA. For this paper’s study, a neural network is obtained
using the features/attributes stated in this study. After
successfully doing the folds of the system, the group has
built the model and the model log together with its
summary. This model is reliable since the information fed to
WEKA has undergone series of learning and pattern-
recognition to derive a stable prediction system. From this
model log, it will show different visualizations like margin
curve, threshold curve, cost/benefit analysis, and cost curve.
This generated model from WEKA will be the one used for
prediction and in line with that, it is best to test the accuracy
of the model by testing it with a test data.

For a given set of attributes, there is a respective and

corresponding value for each voltage, real power and
reactive power for every load bus. This is indicated in the
figure above. The data is listed horizontally respective of
the type of parameter indicted vertically in the attributes
code. The values included are in per-unit basis. To make
things clearer, by looking at the attribute code and input
data in WEKA, it shows that:
For current voltage in bus 2, the value is 1.045
For current voltage angle in bus 2, the value is 0
For current load real power in bus 2, the value is 0.217
For current load reactive power in bus 2, the value is
0.127
For next value of voltage in bus 2, the value is 1.045
For next value of voltage angle in bus 2, the value is
28.138
For next value of load real power in bus 2, the value is FIGURE 1.5. Actual Artificial Neural Network Design
0.217
OKE-5116-103
A neural network usually has 3 basic layers, namely the
input layer, hidden layers, and the output layer. From the
neural network, the attributes declared on the data is
analyzed and processed by WEKA with declaration of
values that result to a specific output, namely stable or
unstable. These attributes serve as the input layer and
parameter of the artificial neural network. When
processed by WEKA, relationships and patterns are to be
generated in the form of weights, which is the lines
connecting the attributes to the neurons
These weights are usually multipliers that can be seen
from the model data generated by WEKA. These weights
are then fed into the hidden layers and then
assessed/studied by the hidden layers with neurons to
further identify relationships and perform back
propagation process to minimize the error and increase the
integrity of the system.
D. Prediction Accuracy
FIGURE 1.5. Accuracy Results
From the figure above, different parameters/results can be
seen. This is the prediction output of the system based from
a given set of codes indicated from previous discussions.
These codes are derived from command prompt and copied
to excel for more clarity of output. Instances, the first
column, represent the number of entries the test data has. In
the 2nd column, the “actual” column shows the actual result
that the test data has. This indicates the ideal voltage
condition the detection system should predict based from its
measured value. The “predicted column” in the 3rd, shows
the predicted results of the detection system using the model
generated from WEKA or in any case, the output. The error
column will determine whether there is an error or none in
the prediction stage. The (+) sign indicates that an error is
detected, whereas no sign indicates that the predictions were
correct. The prediction column in the 4th however,
represents the prediction accuracy of the system. In Figure
A4, 1 from the 2nd row indicates that the system correctly
detected the first entry with a confidence level of 50.2%. In
the first row with the error the system depicted the entry by
1. This means that the system is 100% sure that the output
WEKA has predicted is wrong and inaccurate. As such, the
next column shows the bus location and time (24 hour
setting) where and when the voltage violation had occurred.
The 1’s and 0’s in the last column indicates whether the
system is accurate or not, 1 being accurate and 0 being
5
inaccurate. The accuracy of the prediction was only
80.18%. This is because the code in WEKA doesn’t
indicate a right or wrong in a given set of inputs and thus, it
is the system’s job to study and create computational pattern
from the input to provide a reliable and accurate detection
system. It has to study intensively the given values to
predict a stable value. Because of this, the prediction
obtained a low percentage in accuracy.
V. CONCLUSION
Voltage instability was found to be a significant
impediment on the efficiency and safe operation of power
systems, causing fluctuations in the actual voltage readings
in the power system, which may then lead to the so called
“voltage collapse” or blackouts when left unchecked. The
researchers had proposed an early voltage instability
detection system based from the PMU’s as input
components which then utilize the artificial neural network
to predict the onset of a voltage collapse.
Designing an intelligent detection system for voltage
instability detection requires intensive study in order for the
research study to be effective and efficient. The researchers
gathered data from IEEE 14 Bus System and incorporated it
into the program WEKA. From there, programming
techniques were used in order to create a system that can
detect whether the voltages accumulated are unstable or not.
In totality, the researchers therefore concluded that
detecting voltage instability in the distribution grid is a
possibility. WEKA is a favorable program to be used for
detecting voltage instability since it can easily adapt to new
inputs based from its learning and study derived from
previous computations by taking into consideration real-
time basis. Methods such a Gauss-Seidel for variation of
load parameters and Artificial Neural Network for
generating accurate outputs proved to be a viable option for
training and generating a model for the system. With this
paper, voltage instability will be detected and thus pave way
for distribution utilities to make their response the more
efficient and minimize further casualties such as brownouts
and blackouts from prolonging longer. Since the detection
system proved to be viable to detect bus location and time
which violation occurs, repair response time will be
shortened and maximize time efficiency. An increase or
decrease of 5% in voltage can be further identified for the
system to be unstable.
VI. RECOMMENDATIONS
In applying this study, the researchers recommend that
the engineer check for the compatibility of the algorithm to
the distribution system for practicality purposes. The system
had only been tested, at present, on the inputs from the data
gathered exclusively from IEEE 14 Bus System. The study
was conducted primarily on the notion that voltage
instability was a problem that the distribution utility faces.
An advancement to this study is also to specify what kind of
load is the bus under observation, whether it is residential,
commercial, or industrial. Likewise, it is also better if the
system can determine what increase in load does the voltage
violation occurred. That way, the nature of the parameters
indicated in the bus as well as the instance of time where
voltage instability occurs can be further comprehended and
observational pattern can be acquired.
APPENDIX A. 24 HOUR LOAD PROFILE
APPENDIX B. TREE VISUALIZATION
ACKNOWLEDGMENT
The researchers would like to thank all those people who
have been very supportive throughout the research and
those who assisted them in finishing the research.
First and foremost, the researchers would like to thank
Engr. Danilo C. Guinto, Jr. their EE 515 thesis adviser, for
his valuable insights in aiding the progress and
conceptualization of the research study.
The researchers would also like to thank Engr. Hyson
Peter R. David, the research project instructor for his
commitment in guiding the researchers in the correct
process flow of the study.
The researchers would also like to thank the UST
Electrical Engineering Department who lent a helping hand
and gave meaningful advice upon working and finishing the
research. Moreover, the researchers would like to thank
their parents for supporting and guiding them in finishing
this study.
And most of all, the researchers would like to thank the
Almighty God who guides them from the start till the end of
the research.
REFERENCES
 Ainsworth, J. D., Gavrilovic, A., &Thanawala, H. L. (1980)
Static and Synchronous Compensators for HVDC
Transmission Convertors Connected to Weak AC Systems. In:
28th session CIGRE, pp 31–41
 Andersson, G., Donalek, P., Farmer, R., Hatziargyriou, N.,
Kamwa, I., &Kundur, P. (2005) Causes of the 2003 major
grids blackouts in North America and Europe, and
Recommendation Means to Improve System Dynamic
Performance. IEEE Trans Power Syst 20(4): 1922–1928.
 Bar-Yam, Y. (1997). Dynamics of Complex Systems. Addison-
Wesley.
 Berizzi, A. (2004). TheItalian2003 Blackout. In IEEE: Power
Engineering Society General Meeting.Denver, CO, pp
1673–1679.
 Cutsem, T. V., &Mailhot, R. (1997) Validation of a Fast Voltage
Stability Analysis Method on the Hydro-Quebec system.
IEEE Trans Power Systems 12(1): 282–292
 Cutsem, T. V., &Vournas, C. D. (1998) Voltage Stability of the
Electric Power Systems. Kluwer Academic,
NorwellDECISION - ERC Case No. 2005-280 MC - In the
Matter of the Application for Authority to Implement
Major Capital Projects for Year 2006, with Prayer for
Provisional Authority - Angeles Electric Corp. (AEC)
(7/6/2006), available on line- www.erc.gov.ph.
 DECISION, ERC Case No. 2005-010 RC, In the Matter of the
Application for Approval of the (A) Renewal of Electricity
Purchase Agreement and (B) Extended Electricity Purchase
Agreement (EPA) entered into and between Angeles Electric
Corporation (AEC) and Angeles Power, Inc. (API), with
Prayer for Provisional Authority (5/3/2006), available online-
www.erc.gov.ph.
 Glavic, M. (2012). A short Survey of Methods for Voltage
Instability Detection.
 Gonen, T., Electric Power Distribution System Engineering,
McGraw-Hill Book Company, New York, 1986
 Guide to WECC/NERC Planning Standards I.D: Voltage Support
and Reactive Power, March 30, 2006, Western Electricity
Coordinating Council (WECC), available on line-
http://www.wecc.biz/modules.php?op=modload&name=Downl
oads&file=index&req=viewsd ownload&sid=100
 Hossain, M. J., Pota, H. R., &Ugrinovski, V. (2008) Short and
Long-term Dynamic Voltage Instability. In: 17th IFAC World
Congress, Seoul, Korea, pp 9392–9397
 IEEE (1990) Special publication 90TH0358-2-PWR: voltage
stability of power systems: concepts, analytical tools, and
industry experience
 Kundur, P., Power System Stability and Control, McGraw-Hill,
1994
 Kundur, P., Power Systems Stability and Control, McGraw-Hill,
New York (1994) pp-959-1024.
 Li, F., Zhang, W., Tolbert, L. M., Kueck, J. D., &Rizy, D.
T.,"Assessment of the Economic Benefits from Reactive
Power Compensation," IEEE Power Systems Conference and
Exposition (PSCE), October 29 - November 1, 2006, Atlanta,
Georgia, pp. 1767-1773
OKE-5116-103 7

 Lu, W., Besanger, Y.,Zamai, E.,&Radu, D. (2006). Blackouts:

Description, Analysis and Classification. In: 6th
WSEAS International Conference on Power Systems, pp
429–434.

 Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer,

Peter Reutemann, Ian H. Witten (2009); The WEKA Data
Mining Software: An Update; SIGKDD Explorations,
Volume 11, Issue 1.

 McCulloch, W., & Pitts, W., (1943). A Logical Calculus of the

Ideas Immanent in Nervous Activity. Bulletin of
Mathematical Biophysics, Vol. 5, pp. 115-133.

 Ohno, T.,&Imai, S. (2006). The1987 Tokyo Blackout. In

IEEEPES: Power Systems Conference and
Exposition. Atlanta, GA, pp 314–318.

 Philippine Grid Code, Energy Regulatory Commission, San

Miguel Avenue, Pasig City, Philippines, available on line-
www.erc.gov.ph.

 Saha, T. K., & Ha, L. T.,“Investigation of Power Loss and

Voltage Stability Limits for Large Wind Farm Connections to
a Subtransmission Network”. IEEE Power Engineering
Society General Meeting, 6-10 June, 2004, Denver, Colorado,
USA.

 Summary of WECC Voltage Stability Assessment Methodology,

Western Electricity Coordinating Council (WECC), available
on line-
www.wecc.biz/modules.php?op=modload&name=Downloads
&file=index&req=g etit&lid=11 03

 Taylor, C. W. (1994) Power System Voltage Stability. McGraw-

Hill, New York

 Taylor, C. W., Power System Voltage stability, McGrawhill,

1993.

 U.S. Canada Power System Outage Task Force(2004).

FinalreportontheAug14, 2003 blackout in the United States
and Canada: causes and recommendations

 Vassell, GS. (1991). Northeast Blackout of 1965. IEEE Power

Eng Rev 11(1): 4–8.

 Voltage Stability Criteria, Undervoltage Load Shedding Strategy

and Reactive Power Reserve Monitoring Methodology, May
1998, Western Electricity Coordinating Council (WECC),
available on line:
http://www.wecc.biz/modules.php?op=modload&name=Dow
nloads&file=index &req=e wsdownload&sid=100

 Vournas, C. D., Sauer, P. W., &Pai, M. A. (1996) Relationships

Between Voltage and Angle Stability of Power Systems. Int J
Electr Power Energy Syst 18(8): 493–500