Sunteți pe pagina 1din 10

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO.

3, MAY 2015

1455

A Decomposition-Based Practical Approach to


Transient Stability-Constrained Unit Commitment
Yan Xu, Member, IEEE, Zhao Yang Dong, Senior Member, IEEE, Rui Zhang, Student Member, IEEE,
Yusheng Xue, Member, IEEE, and David J. Hill, Fellow, IEEE

AbstractTraditional security-constrained unit commitment


(SCUC) considers only static security criteria, which may however
not ensure the ability of the system to survive dynamic transition
before reaching a viable operating equilibrium following a large
disturbance, such as transient stability. This paper proposes a
tractable mathematical model for transient stability-constrained
unit commitment (TSCUC) and a practical solution approach.
The problem is modeled without explicit differential-algebraic
equations, reducing the problem size to one very similar to a conventional SCUC. The whole problem is decomposed into a master
problem for UC and a range of subproblems for steady-state
security evaluation and transient stability assessment (TSA).
Additional constraints including Benders cut and so-named stabilization cut are generated for eliminating the security/stability
violations. The extended equal-area criterion (EEAC) is used for
fast TSA and analytically deriving the stabilization cut, wherein
multiple contingencies having common instability mode can be
simultaneously stabilized by one cut. The proposed approach is
demonstrated on the New England 10-machine system and the
IEEE 50-machine system, reporting very high computational
efciency and high-quality solutions.

System load demand in period .


Load demand of bus

Machine inertia of unit .


Number of units and dispatching time periods.
System spinning reserve requirement in period .
Ramping down and up limit of unit .
Shut-down and start-up ramping limit of unit .
Minimum up and down time of unit .
Desired transient stability margin.
Steady-state variables:
Real ow on line in period t for contingency .
Binary variable for on/off status of unit in
period .
Active power output of unit in period .

Index TermsBenders decomposition, extended equal-area


criterion, mixed-integer programming, transient stability-constrained unit commitment.

NOMENCLATURE

in period .

Spinning reserve of unit in period .


ON and OFF time of unit in period .
Dynamic variables:
Transient stability margin for contingency
period .

Constants:

in

Generation cost coefcients of unit .

Rotor angle of unit at transient time .

Generation cost function of unit .

Angle speed of unit at transient time .

Start-up and shut-down cost of unit in period .

Mechanical power of unit at transient time .


Electrical power of unit at transient time .

Manuscript received February 12, 2014; revised May 13, 2014 and July
08, 2014; accepted August 14, 2014. Date of publication August 29, 2014;
date of current version April 16, 2015. This work was supported in part by
the Australian Research Council (ARC) through a Linkage Project (Grant
no. LP120100302), in part by the University of Newcastle through a Faculty
Strategic Pilot Grant, and in part by the State Key Laboratory of China for
Alternate Electrical Power Systems with Renewable Energy Sources through
an Open Grant. Paper no. TPWRS-00214-2014.
Y. Xu and R. Zhang are with the Centre for Intelligent Electricity Networks, University of Newcastle, Newcastle, NSW 2308, Australia (e-mail:
eeyanxu@gmail.com; rui.zhang@newcastle.edu.au).
Z. Y. Dong is with the School of Electrical and Information Engineering, University of Sydney, Sydney, NSW 2006, Australia (e-mail: zydong@ieee.org).
Y. Xue is with State Grid Electric Power Research Institute, Nanjing, China
(e-mail: xueyusheng@sgepri.sgcc.com.cn).
D. J. Hill is with the Department of Electrical and Electronic Engineering,
University of Hong Kong, Pokfulam, Hong Kong, and also with the School of
Electrical and Information Engineering, University of Sydney, Sydney, NSW
2006, Australia (e-mail: dhill@eee.hku.hk).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TPWRS.2014.2350476

I. INTRODUCTION
A. Background and Motivation

S an effective tool to clear day-ahead electricity markets,


an unit commitment (UC) program aims to determine the
generators' operating states including their turn on/off statuses
and power outputs with the objective of minimizing the total
production cost while meeting prevailing operational limits [1].
Conventionally, the prevailing constraints include power balance, unit minimum on/off time limits, ramping up/down limits
and system spinning reserve requirements, etc. State-of-the-art
methods for solving a UC are Lagrangian relaxation (LR) [1]
and mixed-integer programming (MIP) [2].
In recent years, due to rapid load growth and unmatched
infrastructure investments, power systems are being pushed
to operate near their security limits, as a consequence, there
is a strong need to consider the security constraints in the

0885-8950 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

1456

UC, yielding security-constrained UC (SCUC) problems [3].


Traditionally, SCUC aims to satisfy the static security criteria
including network power ow limits and voltage limits at
steady states (before and/or after a contingency) [3]. Presently,
SCUC can be effectively solved by decomposition-based
methods, such as Benders decomposition (BD) [3]. As a necessary extension of UC, SCUC has become an essential tool
to balance the economy and security requirements in day-head
electricity market operations.
While the static security requirements can be soundly satised in SCUC, today's power systems is also facing signicant risk of dynamic insecurity, which means a power system
failing to survive the dynamic transition before reaching a viable steady-state operating equilibrium after a contingency. According to previous investigations [25], [26], lack of adequate
dynamic performance is one key driven force for cascading failures and wide-spread blackouts around the world. As one essential dynamic performance criterion, transient stability refers to
the ability of the power system to maintain synchronism after
being subjected to a large disturbance [4]. The loss of transient
stability can be extremely fast (e.g., 100 200 ms following
the fault clearance) and can trigger generator and load shedding, leading to system-wide failures and/or blackouts. While
conventional SCUC only considers the static security, it may
be unable to meet the transient stability requirement. To further enhance the system security and reduce the risk of blackouts, it is sensible and imperative to include transient stability in
the day-ahead generation dispatch stage, which yields the transient stability-constrained UC (TSCUC) problem studied in this
paper.
B. Literature Study
Although very important, very limited work on TSCUC has
been reported due to the tremendous complexity of the problem.
On one hand, SCUC is a large-scale, multi-stage, mixed nonlinear integer programming problem. On the other hand, transient stability study usually calls for a vast number of differential-algebraic equations (DAEs) which are intractable for current programming algorithms.
To the best of our knowledge, [5] is the rst and sole work
on this topic to date. In [5], the authors propose an augment
LR method combined with variable duplication technique to
solve the TSCUC model. The whole problem is divided into a
basic UC subproblem and a transient stability-constrained optimal power ow (TSCOPF) subproblem. The rst subproblem
is solved by dynamic programming and the second by a reduce-space interior point method.
Although effective, the method reported in [5] can suffer from
several practical limitations. First, the transient stability constraint is formulated as DAEs, and time-domain numerical discretization is applied to convert the differential equations into
the algebraic form. It is clear that this would result in a dramatic
explosion of the problem size proportional to the number of integration time steps times the number of generators. The length
of the whole integration period is usually arbitrarily selected.
Although a reduced-space interior method is used to solve the
TSCOPF model, the overall problem dimension and computational burden remain extensive, making the parallel computing
a necessity for implementing the method. Second, the transient
stability is constrained by a rotor angle limit index, i.e., the maximum rotor angle deviation of each generator against the center
of inertia (COI) during the transient period being bounded by a
pre-dened threshold. However, it has been widely shown that

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 3, MAY 2015

this threshold is usually system dependent and not easy to dene: when it is set to a small value, the operation tends to be
conservative and less economic; while if it is too relaxed, the
transient stability may not be ensured [9], [16]. Meanwhile, it
provides very little information about the system stability degree, which is an important metric for system operators. Furthermore, this method only considers a single contingency case,
whereas in practice, it is usually necessary to stabilize multiple
contingencies [13].
C. Contributions of This Paper
This paper proposes 1) a new TSCUC model which dramatically reduces the problem size and 2) a practical approach to
efciently solve the problem. The model does not explicitly
contain any DAEs, yielding a tractable problem size. Following
the state-of-the-art SCUC solution strategy, the proposed
approach is based on a decomposition framework, where
the master problem consists of solving a basic UC model to
determine the unit status and the generation output, and the
slave subproblems consist of feasibility checks including both
network steady-state security evaluation (NSE) and transient
stability assessment (TSA), and generating additional constraints for the master problem to retrieve the security/stability.
For solving the master problem, the MIP method is used.
Compared with LR, MIP is advantageous in that high-performance commercial solvers can be employed and higher-quality
solutions can be usually obtained [2], [3], [27]. For the slave
subproblems, to deal with the transient stability constraints, a
hybrid TSA method called extended equal-area criteria (EEAC)
[8] is utilized. EEAC, also called single machine equivalent
(SIME) [9], is a quantitative TSA method, which can not only
measure the stability degree (margin) but also quantify the
generation shifting to stabilize the system. Based on EEAC,
the transient stability can be quantitatively constrained and the
stability control can be analytically derived and formulated as
linear constraints (named stabilization cuts in this paper). In
this way, the TSCUC can be solved in a process very similar to
a standard SCUC problem. Meanwhile, contingencies that have
common instability mode can be simultaneously stabilized
(i.e., one stabilization cut to stabilize multiple contingencies),
reducing further the dimension of the problem. Furthermore,
while major computing effort is on the TSA stage, EEAC-based
earlier termination for time-domain simulation (TDS) can be
used to speed-up the solution signicantly.
II. PROPOSED MODEL
A. Objective Function
The objective of TSCUC is to determine a day-ahead UC for
minimizing the total production cost:
(1)
where

is the generation cost function:


(2)

The decision variables are unit on/off status and active power
dispatch.
B. Operational Constraints
The following operational constraints are considered:

XU et al.: A DECOMPOSITION-BASED PRACTICAL APPROACH TO TRANSIENT STABILITY-CONSTRAINED UNIT COMMITMENT

a) Power balance:

(3)
b) Generation limits:
(4)
c) Spinning reserve limits:
(5)
d) Ramping limits:
(6)
e) Minimum up and down time limits:
(7)

C. Steady-State Security Constraints


Since this paper mainly focuses on the transient stability criterion, for the steady-state security, a DC network is used and
the transmission ow limits are considered:

(8)
denotes the base case, and
denotes
where
is the power transfer distribution
a contingency case,
is the
factor of bus to line for contingency at period ,
load demand of bus .
It should be indicated that AC network constraints containing
bus voltage limits [10] can also be used for this method.

1457

E. Model Features
It is important to note that, instead of modeling the transient
stability constraints as a large set of DAEs, which can result in
an enormous problem dimension (i.e., for each contingency, the
number of stability constraints equals the number of generators
times the number of integration time steps), we strive to simplify
the stability constraints in the TSCUC model while retaining the
inherent instability mechanism and essential accuracy. This can
reduce drastically the dimension of the programming problem.
Even more importantly, it provides the opportunity for decomposing the TSCUC problem in a BD manner, wherein high solution efciency can be gained.
III. EXTENDED EQUAL-AREA CRITERION
A. Basics
The original EEAC was rstly proposed by Xue et al. in
[8]. Its improved versionIEEAC [17] (also known as SIME
[9]) is a hybrid TSA approach combining a full TDS process
and the well-known equal-area criterion (EAC). The principle
of IEEAC is to transform the multi-machine trajectories to
an equivalent one-machine-innite-bus (OMIB) trajectory,
and apply the EAC to the equivalent OMIB. In such a way, it
provides a good engineering approximation of investigating the
stability characteristics of the original multi-machine system.
Specically, IEEAC or SIME drives a full TDS engine to
obtain the multi-machine trajectories (wherein complex system
model is not a limitation), and then separate them into two exclusive clusters: one composed of critical machines (CMs) which
are responsible for the loss of synchronism, and the other composed of non-critical machines (NMs) which correspond to the
remaining machines. The two clusters of CMs and NMs are represented as two corresponding equivalent machine trajectories
[8], [9]:
(10)

(11)
denote the CMs and NMs, respecwhere subscripts and
and
are respectively the inertia coefcient of
tively;
CMs and NMs, calculated as
(12)

D. Transient Stability Constraints


Historically, transient stability constraints have been modeled
as a large set of DAEs [4], [5], [11], [12]. Including them in a
programming model can result in a prohibitive problem size and
computational difculties as discussed in Section I.
To alleviate the computational complexity, we express the
transient stability constraints by enforcing a positive stability
margin for each contingency:

The multi-machine equivalent OMIB trajectory is then constructed as follows [8],[9]:


(13)
(14)

(9)
is calculated through a rigorous time-domain simulationbased TSA procedure which will be introduced later. Basically,
a larger stability margin would result in a more conservative
operating condition and therefore higher operating costs. Hence,
it is an engineering practice to limit the stability margin with a
small threshold according to practical needs.

(15)

(16)
(17)

1458

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 3, MAY 2015

Based on IEEAC, TSA can be realized much more efciently


with abundant system stability information while maintaining
necessary accuracy. Currently, there are mature commercial
IEEAC software tools available, such as one called FASTEST
[17]. FASTEST incorporates a powerful full TDS program and
robust IEEAC algorithm to accurately and reliably calculate
the stability margin and other TSA parameters. It has been
widely implemented in China and many other countries for
years reporting satisfactory engineering performance [17].
Fig. 1. Multi-machine rotor angle trajectories (left window) and the correrepresentation (right window)an unstable case for
sponding OMIB
the New England 10-machine system.

C. Transient Stability Control Based on EEAC


IEEAC provides signicant information for transient stability
control (TSC). In the literature, it has been used for efciently
solving preventive TSC and TSCOPF problems [12][16].
For preventive actions, stabilizing an unstable system consists of modifying the pre-contingency conditions until the stability margin becomes zero (or positive). This can be achieved
by increasing the decelerating area
and/or decreasing the
of the OMIB
representation. In
accelerating area
practice, this can be realized by decreasing the OMIB mechan, that is
ical power
(21)

Fig. 2. Multi-machine rotor angle trajectories (left window) and the correrepresentation (right window)a stable case for the
sponding OMIB
New England 10-machine system.

where and here denote the rotor angle and angular speed of
and
denote the mechanical and
the OMIB, respectively;
electrical power of the OMIB, respectively.
plane for quanThen, the EAC is applied to the OMIB
tifying the transient stability degree and extracting stability information of the original multi-machine system. Illustrations of
multi-machine rotor angle trajectories and corresponding OMIB
representations are shown in Figs. 1 and 2. The gures
are obtained from the simulation results in Section V.
B. Transient Stability Assessment Based on EEAC
IEEAC or SIME can offer the following TSA results [8],[9]:
1) CMs and NMs, which determines an instability mode.
2) The time to instability , which indicates the time that
system loses synchronism. At this time, the curve of
crosses
see Fig. 1, that is
(18)
3) The time to first-swing stability , which indicates the
time that system can be declared as rst-swing stable. At
stops its excursion and return back
this time, the curve
before crossing see Fig. 2, that is

and
are
where denotes the pre-contingency state,
respectively the changes in the total power of CMs and NMs:
(22)
To maintain the power balance, the following condition
should be satised:
(23)
Substituting (23) into (21), we have
(24)
Equations (21)(24) reveal that by shifting real power output
of CMs to NMs, the transient stability can be restored [9],
[13][15].
Numerous examples have reported a quasi-linear relationship between changes of stability margin and OMIB mechanical
power at pre-contingency state [9], [13][15], [16], that is
(25)
where is the approximate linear sensitivity of the stability
margin with respect to generation change.
In practice, the sensitivity value around the operating point
can be numerically estimated via two successive IEEAC runs:

(19)

(26)

Note that and can be used to terminate the TDS much


earlier, saving a great deal of computation time.
4) Transient stability margin , which quanties the degree
of system stability. It is calculated by the decelerating area
minus the accelerating area
of the OMIB
plane:
(20)

With , the required generation shifting for TSC can be analytically calculated. Specically, to control an unstable case,
, if the desired stability
whose stability margin is
, the required increment in stability margin is
margin is
. Combining (24)(26), the required generation
shifting between CMs and NMs can be calculated as

means the system is unstable; otherwise stable.

(27)

XU et al.: A DECOMPOSITION-BASED PRACTICAL APPROACH TO TRANSIENT STABILITY-CONSTRAINED UNIT COMMITMENT

Fig. 3. Proposed decomposition strategy for TSCUC.

Meanwhile, IEEAC reveals that multiple contingencies


having common instability mode, i.e., common CMs, can be simultaneously stabilized. This is to apply the most constraining
power shifting, imposed by the severest contingency, to these
common CMs only [9], [13][15]. Hence, the computational
efforts for multi-contingency can be remarkably reduced.
IV. PROPOSED APPROACH
A. Decomposition Strategy
Decomposition-based methods, such as BD, have become a
mature approach to SCUC problems. Typically, it decomposes
the whole problem into a master problem and a range of slave
subproblems. The master problem is essentially a UC problem,
which determines the commitment and dispatch of the units
without the network security constraints. Given the UC results
from the master problem, the slave subproblems check the security constraints. If violation exists, Benders cuts (or similarities) are generated and added to the master problem. The whole
problem is iteratively solved between the master and slave problems until no violation exits. The major advantage of the BD
method is that it can decompose the large problem into a series
of smaller and tractable problems, diminishing signicantly the
complexity of the whole problem and enabling parallel computing to reduce the total execution time.
This paper adopts a similar decomposition strategy to efciently solve the TSCUC problem. The description of the proposed decomposition strategy is shown in Fig. 3.
Let denote the UC status I and generation dispatch P, and y
denote the system state variables. The TSCUC problem can be
rewritten as the following standard BD form:
(28)
(29)
(30)
(31)
where (28) corresponds to the cost function (1), (29) corresponds to the operational constraints (3)(7) as well as the additional constraints generated from the subproblem, (30) corresponds to the network steady-state security constraints (8), and
(31) corresponds to the transient stability constraints (9).

1459

At the initial iteration, (29) only comprises the operational


constraints (3)(7), where the network steady-state security constraints and transient stability constraints are disregarded. After
the subproblems are solved, additional constraints are generated
(if necessary) and added to (29) in subsequent iterations to mitigate the security and stability violations.
As already mentioned, there are generally two mature
methods for solving a UC problem: LR and MIP. A comprehensive comparison and discussion on the two methods for UC can
be found in [3]. In this paper, we apply the MIP method since
high-performance commercial MIP packages are available
today, such as CPLEX [6] and GUROBI [7]. Actually, most
ISOs in the US are switching to MIP method for UC calculation
[27]. However, it should be indicated that the LR method can
also be adopted for the proposed approach here.
To apply the MIP, the nonlinear UC model should be reformulated into a solvable form [i.e., mixed integer linear programming (MILP)] for standard MIP solvers. In this paper, a computationally efcient MILP formulation reported in [2] is adopted.
It has been shown that this MILP formulation requires fewer binary variables and constraints than other reported models and
can signicantly saving the computational efforts.
C. Subproblem
The subproblem evaluates the hourly network steady-state security and the transient stability of a master UC solution and
generates additional constraints (if necessary): Benders cut and
stabilization cut.
1) Network Steady-State Security Evaluation (NSE): The
NSE involves both the base case and contingency cases. For
each case, a linear programming (LP) model is built [3]:
(32)
(33)
where is the vector of ones, is the slack vector used to check
the violation of line ow constraints, and is the Lagrangian
multiplier vector of inequality constraints in (33).
means the violation occurs, and the Benders cut is generated as
(34)
mathematically represents the marginal decrement or
increment of the objective function (32) when is adjusted.
In the next iteration, (34) will be added to (29) of the master
problem to eliminate the steady-state security violation.
2) Transient Stability Assessment (TSA): For each contingency case, the EEAC-based TSA is performed. The TSA results include those presented in Section III-B. The left side of
constraint (31) is checked: if the stability margin is negative, the
stabilization cut, i.e., required generation shifting between CMs
and NMs, is generated.
In order to simultaneously stabilize multiple contingencies
with least number of additional constraints, the contingencies
can be grouped according to their resulting instability modes.
The contingencies having a common instability mode are divided into one group. Each contingency group is then represented by the severest contingency which is the one with the
smallest stability margin in that group. For each representative
contingency, the stabilization cut is generated as

B. Master Problem
The master problem is to solve the UC model, i.e., (28), (29),
determining the commitment and generation dispatch .

(35)

1460

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 3, MAY 2015

Fig. 5. Implementation of the proposed approach.

Fig. 4. Computation owchart.

where
denotes the generation output of unit obtained
from the master problem.
The stabilization cut (35) conveys information about how the
generation dispatch should be modied to preventively retain
the transient stability.
It is important to note that the derived stabilization cut is in
the same mathematical form as the Benders cut, namely, both of
them are the linear constraints. Hence, they can be seamlessly
integrated together to be added to the master problem. In such a
way, the programming problem size can be limited to as small
as a traditional SCUC.
D. Solution Procedure
The general computation owchart of the TSCUC is presented in Fig. 4 and the detailed steps are as follows.
Step 1) Solve the master problem (28), (29) using MILP.
Step 2) Given the master UC solution, perform the hourly
NSE and hourly TSA for each considered contingency.
Step 3) If all the contingencies are both steady-state secure
and transient stable, stop; otherwise, go the next
step.
Step 4) For the steady-state security block, for each insecure contingency, generate the Benders cut (34) by
solving (32), (33); for the transient stability block,
for each representative unstable contingency, generate the stabilization cut (35) following the procedure presented in Section III-C.
Step 5) Add the generated Benders cut and stabilization cut
(if any) to the master problem, and go back to Step
1).
V. NUMERICAL RESULTS
A. Implementation of the Proposed Approach
The simulation is conducted on an ordinary 64-bit PC with
3.10-GHz CPU and 4.0 GB of RAM. Tim-domain simulation
is performed using the commercial power system simulation
package PSS/E [20], and the IEEAC algorithm is realized
in the MATLAB platform (note that, commercial packages
for IEEAC combined with TDS can also be used here, e.g.,
FASTEST [17] which has been put into practice in many
countries). An interface developed in our previous work [21] is
used to connect PSS/E and MATLAB. The optimization model

Fig. 6. One-line diagram of the New England 10-machine system.

is written in MATLAB language and solved by invoking the


commercial solver GUROBI [7] which has a built-in interface
with MATLAB. The implementation of the proposed approach
is schematically described in Fig. 5.
B. New England 10-Machine System
The proposed approach is rst tested on the New England
10-machine system. This test system consists of 10 machines,
39 buses, and 46 branches (see Fig. 6). The system network data
and machine dynamic parameters are obtained from [18], and a
10-unit UC data is obtained from [19]. The UC data is modied
to accommodate the base loading level of this system. Specifically, the total load demand, generation limits, and ramping
limits are increased by 3 times. Then, the total load is proportionally distributed to the load buses according to the base
loading level.
1) SCUC Results: For comparison purposes, the SCUC
model is rst solved without including the transient stability
constraints. The total generation cost is $1 590 910.6 and
the detailed unit status and generation dispatch are given in
Table I and Fig. 7, respectively.
2) Single Contingency Case: A single contingency (called
C1) is considered rst. The contingency is a three-phase shortcircuit at bus 21 with the duration time of 0.12 s. Under this
contingency, it is found that the SCUC solution, although viable

XU et al.: A DECOMPOSITION-BASED PRACTICAL APPROACH TO TRANSIENT STABILITY-CONSTRAINED UNIT COMMITMENT

1461

TABLE I
SCUC/TSCUC RESULTS (NEW ENGLAND 10-MACHINE SYSTEM)

Fig. 8. TSCUC generation dispatchsingle contingency (C1).

Fig. 9. TSCUC generation dispatchmultiple contingencies (C1 and C2).

Fig. 7. SCUC generation dispatch.

for steady-state security, fails to maintain the transient stability


in 17 out of 24 h. The transient stability margin and CMs are
also given in Table I.
To illustrate, the system multi-machine rotor angle trajectorepresentations of the
ries and the corresponding OMIB
5th and 1st dispatch hour are respectively shown in Fig. 1 and
Fig. 2see Section III-A. In Fig. 1, it can be seen that, there is
one CM which is unit G31 and the system losses synchronism
at 0.63 s. Consequently, the TDS can be early terminated at that
time. In Fig. 2, the time to rst-swing stability is 0.4 s, which
means the TDS can be early terminated at that time.
The TSCUC model considering C1 is then solved using the
proposed approach. The desired stability margin is set to a
small positive value 1.0 to avoid over-stabilization. Note however, that this threshold can also be set larger depending on practical needs. After the 1st iteration, hour 7, 10, and 11 remain unstable (but their stability margin have been increased to 8.8,
8.1, and 9.8, respectively) and the other unstable hours become stable with positive stability margin close to the margin
target. The 2nd iteration is then executed and after this iteration, all hours have become transient stable with respect to C1.
For the obtained TSCUC solution, the total generation cost is
$1 615 187.1, which increases only 1.53% over the SCUC solution. The increased cost accounts for preventively stabilizing

the system over the 24 h for C1. The difference in unit status between TSCUC and SCUC solutions is highlighted in Table I, and
the single-contingency TSCUC generation dispatch is shown in
Fig. 8.
Comparing the results of TSCUC and the SCUC, it can be
seen that the generation output of unit G31 has been remarkably
decreased, and to compensate its generation shifting, the on/off
status and generation dispatch of other units are modied.
3) Multi-Contingency Case: The proposed approach is tested
for multi-contingency stabilization, which is not treated in previous work [5]. Another three-phase short-circuit applied at bus
30 with the duration time of 0.09 s is considered (called C2). For
C2, 19 out of 24 hours are unstable, which are the hour 523.
For common unstable hour 512 with C1, C2 has the same composition of CMs as C1, but C2 is more severe, i.e., has smaller
stability margin value, in the hour 5, 79, and 12.
To stabilize C1 and C2 simultaneously, a representative contingency is selected (given the same instability mode) as the one
with smaller stability margin for each unstable hour. Then, the
multi-contingency stability constraint is imposed for the representative contingency only. In this way, the computational efforts can be effectively reduced.
After 2 iterations, a viable TSCUC solution has been obtained, which can ensure the transient stability for both contingencies for each hour. The total generation cost of the multi-contingency TSCUC solution is $1 616 903.5, which increases only
0.11% and 1.64% over the single-contingency TSCUC and the
SCUC solution, respectively. The difference in unit status between the multi- and single-contingency TSCUC solutions is
on the unit G34 at hour 8. The multi-contingency TSCUC generation dispatch is shown in Fig. 9.
4) Computation Efficiency Analysis: Compared with the
existing method for TSCUC [5], it is manifest that the computation efciency of the proposed approach is substantially
higher: it does not solve a time-consuming high-dimensional
programming problem such as TSCOPF; rather, it decomposes

1462

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 3, MAY 2015

TABLE II
CPU TIME (NEW ENGLAND 10-MACHINE SYSTEM)

the whole large-scale problem into small-scale, tractable subproblems which are efciently solved using commercial solvers
and fast TDS tools.
Given the clear computational structure, the total CPU time
required for a single contingency calculation can be roughly estimated as follows:

(36)
Fig. 10. Portion of one-line diagram of the IEEE 50-machine system.

respectively denote the CPU time for


where
denotes the
UC solution, NSE and TDS for a contingency,
denote the number of unstable hour
total iteration number,
denotes all the other elapsed time
for the th iteration, and
during the whole computation process, including the time for
reading and exporting data les, interfacing between different
software tools, etc. In particular, the summation item in (36)
represents that an unstable contingency requires an additional
TDS to calculate the sensitivity value [see (26)] for deriving the
stabilization cut. Table II lists the CPU time for each task of the
proposed approach.
For the master problem, the MILP solver GUROBI only costs
0.24 s. For each contingency, the NSE and generating BD cut requires only 0.01 s. The major computational burden lies in the
TSA phase, which has been signicantly alleviated thanks to
EEAC-based early termination: for a contingency, it only cost
about 0.7 s 0.9 s using PSS/E package. For this test system, the
total CPU times are 62.8 s and 89.5 s for single- and multi-contingency TSCUC calculations, respectively. Given the decomposition structure, the proposed approach is also ideal for parallel implementing, e.g., paralleling 24 h or paralleling contingencies. Such a parallel computing platform tailored for PSS/E
is developed in [21].
C. IEEE 50-Machine System
The IEEE 50-machine system is used to further demonstrate
the high solution speed of the proposed approach. This system is
derived from a representative model of a realistic power system
in North America [23]. It consists of 50 machines, 145 buses,
and 453 branches. Fig. 10 shows a portion of the one-line diagram of the high-voltage lines of this test system.
The system data is obtained from [24], and is extended
for a TSCUC study. The size of this system is similar to the
IEEE 118-bus and 300-bus systems used in [5], which have
54 machines/186 branches and 69 machines/411 branches,
respectively. Therefore it is sensible to use the reported CPU
time in [5] to benchmark the proposed approach.
A 3-phase short-circuit fault at the branch between bus 6
and 10 is studied here. Using the proposed approach, a viable
TSCUC solution is obtained after 3 iterations. Fig. 11 shows
the system trajectories (in different colors to distinguish CMs
and NMs) for a specic dispatch hour from the SCUC solution,

Fig. 11. Multi-machine rotor angle trajectories (upper window) and the correrepresentation (lower window)an unstable case for
sponding OMIB
the IEEE 50-machine system.

where it can be seen that the system losses synchronism after


the fault and its stability margin is 18.6.
Fig. 12 shows the system trajectories for the same dispatch
hour from the TSCUC solution, where it can be seen that the
system is able to maintain synchronism after the fault and its
stability margin becomes to 0.72. Note that the trajectories of
CMs and NMs are respectively plotted in red and blue colors in
the upper windows of these two gures.
Since the major computational burden lies at the TDS stage, it
is important to note that, for the unstable case (Fig. 10), the instability condition is reached at 1.09 s; for the stable case (Fig. 11),
the rst-swing stability condition is met at 1.36 s. Hence, there
is no need to run the TDS for the whole simulation time (e.g., 5

XU et al.: A DECOMPOSITION-BASED PRACTICAL APPROACH TO TRANSIENT STABILITY-CONSTRAINED UNIT COMMITMENT

1463

decomposes the whole problem into a master problem for UC


and a range of subproblems for NSE and TSA. Benders cut and
so-named stabilization cut are generated for eliminating the
security/stability violations. The EEAC is used for fast TSA
and analytically deriving the stabilization cut. Compared with
the existing method [5], the proposed approach in this paper is
advantageous for much higher solution speed, more accurate
and quantitative modeling of stability constraints, and high
efciency in stabilizing multiple contingencies simultaneously.
Case studies on the New England 10-machine system and the
IEEE 50-machine system have validated the approach.
Given the strong and inherent coupling relationship between
the active power and the rotor angle stability, this paper only
optimize the active power quantities with the fully time-domain
simulation-based TSA check. Note that this is consistent with
industry practice, where DC OPF with AC feasibility check is
adopted (see [27]). Future efforts can be devoted to co-optimize
both active and reactive power in the TSCUC problem. This can
be achieved using the AC network-based UC modeling reported
in [10]. In addition, future works can be done to integrate more
dimensions into the problem, such as FACTS devices which can
impact system's dynamic performance.
Fig. 12. Multi-machine rotor angle trajectories (upper window) and the correrepresentation (lower window)an stable case for the
sponding OMIB
IEEE 50-machine system.

TABLE III
CPU TIME (IEEE 50-MACHINE SYSTEM)

TABLE IV
TOTAL CPU TIME FOR TSCUC COMPUTATIONS

or 10 s); rather, the TDS can be early terminated at these times,


saving a great deal of computation time.
The CPU time for each task of the TSCUC calculation is
given in Table III. For this test system, the total CPU time is
around 319 s. To compare, the CPU time of the existing approach on the two similar systems are listed in Table IV. It can
be seen that the proposed approach is around 140 times faster
on the similar test system.
VI. CONCLUSION AND FUTURE WORK
Lack of adequate dynamic performance is one key driven
force for cascading failures and/or widespread blackouts. In
day-ahead generation dispatch stage, TSCUC is a reasonable
and necessary extension of SCUC to satisfy transient stability
requirements for day-ahead generation dispatch. As studied in
the pioneer work [5], its difculty lies in the high dimension,
nonlinearity and DAE-based nature of the problem. This paper
proposes a new, tractable TSCUC model which signicantly
reduces the problem size to one very similar to a conventional
SCUC. A decomposition-based practical approach is then
developed. Similar to the BD strategy, the proposed approach

REFERENCES
[1] J. Wood and B. F. Wollenberg, Power Operation, Generation and Control. New York, NY, USA: Wiley, 1996.
[2] M. Carrion and J. Arroyo, A computationally efcient mixed-integer
linear formulation for the thermal unit commitment problem, IEEE
Trans. Power Syst., vol. 21, no. 3, pp. 13711378, Aug. 2006.
[3] Y. Fu, Z. Li, and L. Wu, Modeling and solution of the large-scale
security-constrained unit commitment, IEEE Trans. Power Syst., vol.
28, no. 4, pp. 35243533, Nov. 2013.
[4] P. Kundur, J. Paserba, V. Ajjarapu, G. Andersson, A. Bose, C.
Canizares, N. Hatziargyriou, D. J. Hill, A. Stankovic, C. Taylor, T. Van
Cutsem, and V. Vittal, Denition and classication of power system
stability, IEEE Trans. Power Syst., vol. 19, no. 2, pp. 13871401,
May 2004.
[5] Q. Jiang, B. Zhou, and M. Zhang, Parallel augment Lagrangian relaxation method for transient stability constrained unit commitment,
IEEE Trans. Power Syst., vol. 28, no. 2, pp. 11401148, May 2013.
[6] CPLEX Optimizer [Online]. Available: http://www.ibm.com/us/en/
[7] GUROBI Optimizer [Online]. Available: http://www.gurobi.com/
[8] Y. Xue, T. Van Cutsem, and M. Pavella, A simple direct method for
fast transient stability assessment of large power systems, IEEE Trans.
Power Syst., vol. 3, no. 2, pp. 400412, May 1988.
[9] M. Pavella, D. Ernst, and D. Ruiz-Vega, Transient Stability of Power
Systems: A Unified Approach to Assessment and Control. Norwell,
MA, USA: Kluwer, 2000.
[10] Y. Fu, M. Shahidehpour, and Z. Li, Security-constrained unit commitment with AC constraints, IEEE Trans. Power Syst., vol. 20, no.
3, pp. 15381550, Aug. 2005.
[11] D. Gan, R. J. Thomas, and R. D. Zimmerman, Stability-constrained
optimal power ow, IEEE Trans. Power Syst., vol. 15, no. 2, pp.
535540, May 2000.
[12] Y. Xu, Z. Y. Dong, K. Meng, J. Zhao, and K. P. Wong, A hybrid
method for transient stability constrained-optimal power ow computation, IEEE Trans. Power Syst., vol. 27, no. 4, pp. 17691777, Nov.
2012.
[13] D. Ruiz-Vega and M. Pavella, A comprehensive approach to transient stability control. I: Near optimal preventive control, IEEE Trans.
Power Syst., vol. 18, no. 4, pp. 14461453, Nov. 2003.
[14] Y. Xue, W. Li, and D. J. Hill, Optimization of transient stability control Part-I: For cases with identical unstable modes, Int. J. Control
Autom. Syst., vol. 3, no. 2, pp. 334340, Jun. 2005.
[15] Y. Xue, W. Li, and D. J. Hill, Optimization of transient stability control Part-II: For cases with different unstable modes, Int. J. Control
Autom. Syst., vol. 3, no. 2, pp. 341345, Jun. 2005.
[16] A. Pizano-Martinez, C. R. Fuerte-Esquivel, and D. Ruiz-Vega, A new
practical approach to transient stability-constrained optimal power
ow, IEEE Trans. Power Syst., vol. 26, no. 3, pp. 16861696, Aug.
2011.

1464

[17] Y. Xue, Fast analysis of stability using EEAC and simulation technologies, in Proc. 1998 Int. Conf. Power Syst. Tech (POWERCON),
Beijing, China, 1998.
[18] M. A. Pai, Energy Function Analysis for Power System Stability.
Norwell, MA, USA: Kluwer, 1989.
[19] S. A. Kazarlis, A. G. Bakirtzis, and V. Petridis, A genetic algorithm
solution to the unit commitment problem, IEEE Trans. Power Syst.,
vol. 11, no. 1, pp. 8392, Feb. 1996.
[20] Siemens PTI, PSS/E 33.0 Program Application Guide, May 2011.
[21] K. Meng, Z. Y. Dong, K. P. Wong, and Y. Xu et al., Speed-up the
computing efciency of PSS/E-based power system transient stability
simulations, IET Gener., Transm., Distrib., vol. 4, no. 5, pp. 652661,
May 2010.
[22] Y. Xue, Integrated extended equal area criterion-theory and application, in Proc. 5th Symp. Specialists in Electric Operational and Expansion Planning, Recife, Brazil, 1996.
[23] IEEE Committee Report, Transient stability test systems for direct
stability methods, IEEE Trans. Power Syst., vol. 7, no. 1, pp. 3744,
Feb. 1992.
[24] Power Systems Test Case Archive [Online]. Available: http://www.ee.
washington.edu/research/pstca/
[25] IEEE PES Power System Dynamic Performance Committee, Causes
of the 2003 major grid blackouts in North America and Europe, recommended means to improve system dynamic performance, IEEE Trans.
Power Syst., vol. 20, no. 4, pp. 19221928, Nov. 2005.
[26] Y. V. Makarov, V. I. Reshetov, A. Stroev, and I. Voropai, Blackout
prevention in the United States, Europe, and Russia, Proc. IEEE, vol.
93, no. 11, pp. 19421955, Nov. 2005.
[27] FERC Report, Recent ISO Software Enhancements and Future Software and Modeling Plans [Online]. Available: http://www.ferc.gov/industries/electric/indus-act/rto/rto-iso-soft-2011.pdf

Yan Xu (S'10M'13) received the B.E. and M.E. degrees from South China
University of Technology, China, in 2008 and 2011, respectively, and the Ph.D.
degree from the University of Newcastle, Australia, in 2013.
He was with the Hong Kong Polytechnic University, Hong Kong, between
2009 and 2011. He is now a Research Fellow at the Center for Intelligent Electricity Networks (CIEN), University of Newcastle, Australia. His research interests include power system stability and control, power system planning, Smart
Grid, and intelligent system applications to power engineering.

Zhao Yang Dong (M'99SM'06) received the Ph.D. degree from the University
of Sydney, Australia, in 1999.
He is now Professor and Head of School of Electrical and Information Engineering, University of Sydney, Australia. He was previously Ausgrid Chair and
Director of the Centre for Intelligent Electricity Networks (CIEN), The University of Newcastle, Australia, and is now a conjoint professor there. He also
held academic and industrial positions with the Hong Kong Polytechnic University and Transend Networks, Tasmania, Australia. His research interest includes Smart Grid, power system planning, power system security, load mod-

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 30, NO. 3, MAY 2015

eling, electricity market, and computational intelligence and its application in


power engineering.
Prof. Dong is an editor of the IEEE TRANSACTIONS ON SMART GRID, the
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, IEEE POWER ENGINEERING
LETTERS, and IET Renewable Power Generation.

Rui Zhang (S'12) received the B.E. degree from the University of Queensland,
Australia, and the Ph.D. degree from the University of Newcastle, Australia, in
2009 and 2014, respectively.
She is now a Research Associate at the Centre for Intelligent Electricity Networks (CIEN), University of Newcastle, Australia. She was with Mawan Electric Power Company, Shenzhen, China, from 2010 to 2011. Her research interests include power system operation, stability, and control.

Yusheng Xue (M'87) received the Ph.D. degree from the University of Liege,
Belgium, in 1987.
He is a member of Chinese Academy of Engineering (CAE), and a Standing
Committee member of Division of Energy and Mining Engineering, CAE. He
was Chief Engineer at the Nanjing Automation Research Institute (NARI),
China during 19932009. He is now the Honorary President of State Grid
Electric Power Research Institute (SGEPRI or NARI), China. His research
interests are power system automation and control, power system dynamics
and stability, and power system computing methods.
Prof. Xue is a member of the PSCC Council, and the Editor-in-Chief of Automation of Electric Power System since 1999, and a member of Editorial Board
of IET Generation, Transmission & Distribution.

David J. Hill (F'93) received the B.E. degree in electrical engineering and the
B.Sc. degree in mathematics from the University of Queensland, Australia, in
1972 and 1974, respectively, and the Ph.D. degree from the University of Newcastle, Australia, in 1976.
He is now Chair of Electrical Engineering at The University of Hong Kong
and part-time professor with the School of Electrical and Information Engineering, The University of Sydney, Australia. He was previously Ausgrid Chair
at the University of Sydney and is a Principal Researcher in National ICT Australia. During 20052010, he was an Australian Research Council Federation
Fellow at the Australian National University. During 20062010, he was also a
Chief Investigator of the ARC Centre of Excellence for Mathematics and Statistics of Complex Systems. Since 1994, he has held various positions at the
University of Sydney, Melbourne, California (Berkeley), Newcastle (Australia),
Lund (Sweden) and City University and The University of Hong Kong. His research interests are in network systems, stability analysis, distributed control
and applications to infrastructure type networks, with work now focused on future electricity networks.
Dr. Hill is a Fellow of Engineers Australia, the Society for Industrial and Applied Mathematics (SIAM), USA, the Australian Academy of Science, and the
Australian Academy of Technological Sciences and Engineering (ATSE). He is
a Foreign Member of the Royal Swedish Academy of Engineering Sciences.

S-ar putea să vă placă și