See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233160893

Stability Constrained Optimal Power Flow in Deregulated Power

Systems

Article  in  Electric Power Components and Systems · May 2011

DOI: 10.1080/15325008.2010.541409

CITATIONS READS

10 4,979

4 authors, including:

Seyed Hossein Hosseinian Poria Hasanpor Divshali

Amirkabir University of Technology VTT Technical Research Centre of Finland
382 PUBLICATIONS   3,741 CITATIONS    40 PUBLICATIONS   362 CITATIONS   

SEE PROFILE SEE PROFILE

Behrooz Vahidi
Amirkabir University of Technology
513 PUBLICATIONS   4,106 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Designing Secure and Efficient Distributed Smart Grid View project

Optimal Demand Response View project

All content following this page was uploaded by Poria Hasanpor Divshali on 26 March 2015.

The user has requested enhancement of the downloaded file.

This article was downloaded by: [The University of British Columbia]
On: 28 September 2013, At: 13:05
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Electric Power Components and Systems

Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/uemp20

Stability Constrained Optimal Power

Flow in Deregulated Power Systems
a a a
E. Nasr Azadani , S. H. Hosseinian , P. Hasanpor Divshali & B.
a
Vahidi
a
Department of Electrical Engineering, Amirkabir University of
Technology, Tehran, Iran
Published online: 01 May 2011.

To cite this article: E. Nasr Azadani , S. H. Hosseinian , P. Hasanpor Divshali & B. Vahidi (2011)
Stability Constrained Optimal Power Flow in Deregulated Power Systems, Electric Power Components
and Systems, 39:8, 713-732, DOI: 10.1080/15325008.2010.541409

To link to this article: http://dx.doi.org/10.1080/15325008.2010.541409

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
 Electric Power Components and Systems, 39:713–732, 2011
Copyright © Taylor & Francis Group, LLC
ISSN: 1532-5008 print/1532-5016 online
DOI: 10.1080/15325008.2010.541409

Stability Constrained Optimal Power Flow

in Deregulated Power Systems

E. NASR AZADANI,1 S. H. HOSSEINIAN,1

P. HASANPOR DIVSHALI,1 and B. VAHIDI 1
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

1
Department of Electrical Engineering, Amirkabir University of Technology,
Tehran, Iran

Abstract The dynamic security-constrained dispatch of an electric power network is

a difficult task facing an independent system operator mandated to provide equitable
and fair transmission services in an open-market environment. In this article, a novel
technique based on iterative stability-constrained optimum power flow is proposed.
The particle swarm optimization methodology is employed to maximize the social
welfare with consideration of static and dynamic functional operating constraints
and dynamic loading margin requirements with respect to normal condition and
contingencies. Furthermore, since the pattern of load increase is difficult to predict
in the new market environment, a new method for determining the sensitive loading
direction associated with a dynamic loading margin is proposed. An IEEE 14-bus
test system with both supply and demand bidding is used to illustrate and test the
proposed technique.

Keywords dynamic loading margin, optimum power flow, particle swarm optimi-
zation

1. Introduction
In the last decade, power system operation has been remarkably affected by the deregula-
tion process. In the most industrialized countries, it is difficult to build new transmission
lines while demand is constantly increasing. This fact leads market participants and
system operators to look for adequate and practical ways of evaluating, maintaining,
and pricing system security in order to allow secure market transactions. However, in-
creased load demand and the need to operate the system based on economic considerations
have led to many concerns regarding the secure operation of power systems. In this
environment, independent system operators (ISOs) are in charge of guaranteeing stable,
secure, and reliable operation of the power grid.
Pricing security through proper system constraints requires a variety of assumptions
as well as complex models and simulations. In the various market models that have been
proposed, how to properly include system security is still an open question [1, 2].
Oscillatory stability is an inherently non-linear phenomenon that is related to bifurca-
tion from the viewpoint of non-linear dynamic systems. Bifurcations in the power systems
include saddle-node bifurcation (SNB), limit-induced bifurcation (LB), Hopf bifurcation

Received 10 February 2010; accepted 28 October 2010.

Address correspondence to Dr. Seyed H. Hosseinian, Department of Electrical Engineering,
Amirkabir University of Technology, Tehran, Iran. E-mail: [email protected]

713
 714 E. Nasr Azadani et al.

Nomenclature
Aj simple pair of eigenvalues ˙j!0 , !0 ¤ 0
ac1 and ac2 acceleration constants in PSO
CD vectors of demand bids ($/MWh)
CS vectors of supply bids ($/MWh)
d index of iteration in PSO algorithm
d˛=d change rate of ˛ due to load changes
DAE differential-algebraic equations
DLM dynamic loading margin
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

f objective function
ff .0/ ff .0/ W Rn  Rm  Rk  Rn stands for the system non-linear
differential (state) equations
g equality constraints
g.0/ g.0/ W Rn  Rm  Rk  Rn represents the system algebraic
constraints
Gbest index of the best particle among all the particles in the group
HB Hopf bifurcation
Iij current of line between bus i and j
ISO independent system operator
itermax maximum number of iterations in PSO
Iter number of the iterations up to current stage in PSO
k p.u. loading factor
kO maximum loading level in the operating point
LB limit-induced bifurcation
p p 2 Rk are the control variables, such as power demand and
supply bids
Pd demand power bids
PG generation active power
PG0 must-run generation that is not included in the market bidding
Pij active power of line between bus i and j
PL load active power
PL0 base power that represents inelastic loads
PS supply power bids
P0 load condition of current operating point
pp and pv particle coordinates (position) and its corresponding flight speed
(velocity) in a search space
ppd current position of particle at the d th iteration in PSO
pvd velocity of particle at d th iteration
Pbest best previous position of a particle
QL load reactive power
QG generator reactive powers
SNB saddle-node bifurcation
v associated right eigenvector of A./
V bus voltage magnitudes

(continued)
 Stability Constrained Optimal Power Flow 715

Nomenclature (continued )
x x 2 Rn are the dependent variables, such as bus voltage phasors
y y 2 Rm is a vector of steady-state algebraic variables, such as bus
voltage
Phopf
T hypersurface of Hopf bifurcation
˛ real part of eigenvalue
inertia weight factor in PSO
ı bus voltage phases
 step of load increasing
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

 load-increasing factor
 eigenvalue of A./
 coefficient that maximum allowable velocity set with respect to
variable range
ˆ constant power factor angle

(HB), and others [3]. According to mathematical concepts, SNB occurs when an eigen-
value of state matrix equals zero, and HB occurs when a pair of complex eigenvalues of
state matrix reaches the imaginary axis due to changes in system parameters [4].
The probability of a bifurcation problem occurring depends on the loading level
of the system. For heavily loaded systems, when the operating point approaches the
maximum loading point on the P-V curve, the region of attraction is very small [5];
consequently, perturbations cannot be withstood by the system.
For considering HB, several methods have been proposed. These methods recognize
and predict proximity to bifurcation point in power systems. One popular way is using
the bifurcation index. In [6–8] some indices for prediction of HB point and SNB are
proposed.
In this article, a new bifurcation detection algorithm with a variable load-increasing
step is used. In this method for detection of the dynamic loading margin (DLM), load-
increasing step is determined based on the eigenvalue and its velocity based on the index
in [9]. This bifurcation detection algorithm determines bifurcation occurrence in five or
six iterations.
In addition, determining the sensitive load direction of the network is a vital subject in
system management. Extensive research works have been conducted on loadability margin
determination, considering static voltage stability [10–14]. The continuation power flow
method may be the most reliable technique to calculate loadability margins along a given
loading direction by tracing the system P-V curves, which may be associated with an
SNB or LB [10, 11].
In [12], a mathematical method was proposed for computing a closet SNB and worst-
case load power margin for voltage collapse. This method determines the worst-loading
direction considering SNB. In [13], an iterative and direct method of power flow was
presented to estimate the maximum loading condition. In [14], the authors proposed
a modified continuation power flow for tracing power system stationary behavior due
to parameters variation. This method has used a predictor-corrector continuation method
to trace the solution curve. All the mentioned methods only consider SNB in static models.
A dynamic model of a system can lead to a different loading direction compared
to previous states. In this study, HB has been considered for determining stability and
 716 E. Nasr Azadani et al.

security issues. In [15], a neural network is employed to predict the loadability margins.
Continuous power flow and eigenvalue analysis are used for the security boundary.
Various algorithms have been proposed to include stability constraints in the optimum
power flow (OPF). In [1], a different strategy is proposed based on the use of a multi-
objective OPF technique to maximize both social benefit and the distance to a voltage
instability point. In [16], a new technique based on continuous power flow (CPF) and OPF
is proposed in which fuel cost and voltage stability margin (loadability) are considered to
be optimized. The problem with the inclusion of voltage security constraints alone is that,
in some power systems, oscillatory instabilities are one of the limiting factors in maintain-
ing system security [17, 18] (e.g., WSCC, now WECC, August ’96 blackout). In [18],
the effects of the tuning process of system controllers on the operation of electricity
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

markets and their associated power systems using the system stability constrained-OPF
(SSC-OPF) algorithm is discussed. In [18], the SSC-OPF algorithm uses the stability
index proposed in [6]. In these mentioned researches, the direction of load increase has
not been considered in SSC-OPF, and pattern of load increase is usually defined with
respect to initial load.
A wide variety of optimization techniques has been applied to solve the OPF
problem, such as non-linear programming, quadric programming, linear programming,
Newton-based techniques, sequential minimization technique, and interior method point.
However, most of these methods have their own flaws. A new category of classical
optimization tools has emerged to cope with some of the traditional optimization algo-
rithms’ shortcomings. The main modern optimization techniques are the genetic algorithm
(GA) [19], evolutionary programming (EP) [20], simulated annealing (SA) [21], ant
colony optimization (ACO), and particle swarm optimization (PSO) [22, 23].
The present article proposes an iterative SSC-OPF with a sensitive load direction.
Furthermore, a new approach has been described for applying the sensitive load direction
to SSC-OPF. PSO is employed for the optimization procedure. The proposed method
provides the market solution as a function of security margin with respect to the N 1
contingency criterion. The objective is to maximize the social welfare while maintaining
the adequate distance to a maximum loading condition associated with bus voltage limits
or system stability limits.
This article is organized as follows. Section 2 discusses the basic concepts of a stan-
dard OPF-based market. Section 3 presents system modeling (Section 3.1), determining
of the bifurcation (Section 3.2), sensitivity analysis for load direction (Section 3.3),
and proposed SSC-OPF with the sensitive load direction (SSC-OPF-SLD) (Section 3.4).
Optimization tools and simulation process are described in Section 4. In Section 5, the ap-
plication of SSC-OPF-SLD has been illustrated using an IEEE 14-bus benchmark system.

2. Standard OPF-based Market

The standard OPF-based market is a non-linear constrained optimization problem and con-
sists of an objective function and a set of equality and inequality constraints, as follows:
Min. f .x; p; k/;

s.t. g.x; p/ D 0;
(1)
hmin  h.x; p; k/  hmax

pmin  p  pmax
 Stability Constrained Optimal Power Flow 717

The functions f , g, and h are defined as follows.

Objective function: the objective function is defined as

f .x; p; k/ D .CDT Pd CST PS /: (2)

The negative of Eq. (2) represents consumer surplus plus the producer surplus, i.e.,
the net social welfare.
Equality constraints: g.x; p/ D 0 represents the standard power flow equations;

g.x; p; k/ D g.ı; V; QG ; Ps ; Pd / D 0: (3)

Downloaded by [The University of British Columbia] at 13:05 28 September 2013

The loads in the system are typically represented in steady state as constant PQ
loads with a constant power factor and are assumed to increase for the purpose of
stressing the system as follows:

PL D PL0 C Pd ;

Pd  kPL0 ;
(4)
QL D PL tan ;

PG D PG0 C PS :

Inequality constraints: inequality constraints consist of the physical and security limits
of the system. The physical and security limits considered transmission line thermal
limits;

Iij .ı; V /  Iij max : (5)

Generator reactive power limits:

QG min  QG .ı; V /  QG max : (6)

Voltage “security” limits:

Vmin  V  Vmax : (7)

Power limits on transmission lines:

jPij .ı; V /j  Pij max ; (8)

which are used to represent the security limits of the system. The limits are repre-
sented as follows:

Ps min  Ps  Ps max ;
(9)
Pd min  Pd  Pd max :
 718 E. Nasr Azadani et al.

3. Small Perturbation Stability-constrained OPF

3.1. System Modeling

In order to precisely analyze bifurcation in any given system, providing accurate dynam-
ical models is of prime importance. Dynamical and load flow equations are required to
assess small signal stability in power grids. The DAE in the power system is obtained as
Eq. (10) [6, 8]:
(
xP D ff .x; y; p/
: (10)
0 D g.x; y; p/
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

In small-perturbation stability analysis, Eq. (10) is linearized around an equilibrium

or operating point. Thus,
" # " #" #
xP J1 J2 x
D ; (11)
0 J3 J4 y
„ ƒ‚ …
J

where J1 D @ff @ff @g @g

@x , J2 D @y , J3 D @x , and J4 D @y . In fact, J4 is the Jacobian matrix of
the load flow. Assuming that J4 is not singular, Eq. (11) can be expressed in the form
of Eq. (12):


xP D J1 J2 J4 1 J3 x D A:x: (12)

Equation (12) is an ordinary differential equation (ODE) in which A is the state

matrix of the system. As an important point, the dynamic order of system elements
should be selected meticulously. Increasing the dynamic order leads to decrease in the
speed of equation solving. On the other hand, decreasing the dynamic order might lead to
low accuracy. For example, in [24], it was shown that when a synchronous machine in a
simple system is modeled by a second dynamical order, bifurcation cannot be monitored.
In contrast, with a higher-order modeling, bifurcation can be observed by changing the
stator resistor.

3.2. Determination of Bifurcation

The occurrence of bifurcation in a power system depends on the position of state matrix
eigenvalues. If a pair of eigenvalue reaches the imaginary axis with no zero imaginary
part, HB will occur. If an eigenvalue with zero imaginary parts reaches the imaginary axis,
SNB will occur.
The basic equation for finding an eigenvalue of a state matrix of a system is given by

Av D v;
(13)
v T v D 1:

Bifurcation in power system depends on two important parameters: the initial value
of loads and the generation of all buses, as load and generation increase direction. The
DLM can be defined as the minimum amount of additional load on a specified pattern of
load increase that would cause the power system to be dynamically unstable. The simplest
 Stability Constrained Optimal Power Flow 719

way to determine the distance of the system to the bifurcation point includes three stages.
First, increasing consummation load and generation power of all buses in certain direction
and performing load flow calculation; second, formation of the state matrix and eigenvalue
calculation of this matrix; and third, studying the eigenvalue position and determining
whether or not bifurcation occurs. If bifurcation does not occur, it will go back to the
first stage and will increase consummation load and generation power. In the simplest
method, increasing the step in all iterations is constant. In order to increase the accuracy
of bifurcation, the step of load increasing must be decreased; therefore, the higher the
accuracy, the higher the calculation time. For a decreasing number of iterations, the step
of load increasing must be varied with a change in eigenvalue position of the state matrix.
In this article, the load-increasing step is determined by the vicinity of the eigenvalues
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

to the imaginary axis and, also, the velocity of their movement to the imaginary axis. By
using this method, the number of iterations for bifurcation detection will reduce to five
or six iterations.
All eigenvalues do not move similarly with increasing consummation load and
generation power of all buses of system. Some eigenvalues are almost constant, and some
of them move quickly in different directions. The increase in load and generation moves
some eigenvalues away from the imaginary axis and other towards the imaginary axis.
Therefore, in order to determine the step of load increasing,  should be calculated for
a certain number of nearest eigenvalues. If eigenvalues move with present speed , the
load-increasing step causes that corresponding eigenvalue to reach the imaginary axis.
 is calculated by Eq. (14) [25]:
˛
 D : (14)
d˛=d

In order to determine the load-increasing step,  is calculated for a definite number

of eigenvalues with the largest real part.  > 0 means that the eigenvalue is moving
toward the imaginary axis. Actually, the smaller the , the more critical the eigenvalue.
In contrast,  < 0 shows that the distance of the eigenvalue to the imaginary axis
increases. Therefore,  should be equal to the smallest positive index, and, if the
system is unstable, the load should be decreased until the largest eigenvalue comes back
to the left half of the complex plane. Therefore,  should be equal to the smallest
negative index.
Numerical differentiation can be used to compute d˛=d because the conventional
algorithm for eigenvalue calculation cannot keep the eigenvalue ordering.
Differentiating Eq. (13) respect to ,
8̂
dA dv dv d
<v
ˆ CA D C v
d d d d
: (15)
ˆ dv T dv
:̂v C vT D0
d d
T
Since v dv
d
D v T ddv , Eq. (15) can be simplified into
2 3
" # dv 2
dA./
3
I A./ v 6 d 7 v
6 7 D 4 d 5 : (16)
vT 0 4 d 5
0
d
 720 E. Nasr Azadani et al.

Therefore, d=d is obtained via the calculation of dA./=d from numerical differen-
tiation. The real part of this value is equal to d˛=d. Therefore, DLM detection is used
to determine the distance of the operating point to bifurcation with the initial load and
generation power along a given loading direction. Moreover, DLM detection has been
employed to find the sensitive direction to the Hopf hypersurface.

3.3. SSC-OPF Market-clearing Model

In this article, the following optimization problem is used to represent an OPF market-
clearing model with inclusion of small perturbation stability constraints:
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

Min. f .x; p; k/ D .CDT Pd CST PS /;

s.t. g.ı; V; QG ; Ps ; Pd ; k/ D 0;

O
k D k; (17)

hmin  h.x; p; k/  hmax ;

pmin  p  pmax ;

where kO is the maximum dynamic loading level in the operating point and kO > DLM req ,
where DLM req is the DLM requirement under normal state and, hence, is assumed to be
an input. Furthermore, k  DLM max , where DLM max stands for the maximum loading
margin. SSC-OPF (Eq. (17)) is basically a non-linear optimization problem with an
implicit constraint; hence, an optimization technique must be employed to be able to
solve this problem. In this article, the PSO method is used as the base to solve the
proposed optimization problem.

3.4. Sensitivity Analysis for Load Direction

As mentioned in Section 3.2, the direction of load increase is one of the major parameters
that affects the DLM. However, the increase of load in the power market depends on the
electricity price, so it is difficult to predict the pattern of load increase and how the system
network is stressed by load. When loads increase in a different pattern or the network is
stressed in different directions, different DLMs will be obtained. This section addresses
the new method for determining the sensitive direction by calculating the normal vector
to HB. Phopf Phopf
Write the hypersurface of HB T for a set of  2 T , where  is the vector
of loading level of the system for which Eq. (12) has an HB with Aj having a simple
pair of eigenvalues ˙j!0 , !0 ¤ 0 and all of the eigenvalues with non-zero real parts
and satisfying the transversality condition. Since Aj is invertible, the implicit function
theorem implies that there is a smooth function u defined as a neighborhood of  with
u. / D x  , u./ D x  and f .u./; / D 0. There is also a smooth function  defined
in a neighborhood of  with . / D j!0 and . /, an eigenvalue of A.u./; /.
Denoting v and w for right and left complex eigenvectors of A.u./; / corresponding
to ./, these eigenvectors are normalized according to the jvj D 1, vw D 1. Then v
and w are smooth functions of  in a neighborhood of . /.
 Stability Constrained Optimal Power Flow 721
Phopf
The transversersality conditions satisfied by  2 T are

c ¤ 0;

N (18)
d Ref./g
¤ 0;
d
where c is a coefficient of cubic terms in the flow reduced to the center manifold and is
Phopf
a complicated function of triple derivatives of f .  2 T states that there is an open
Phopf
set in a smooth hypersurface given by Ref./g D 0. It follows that T has a normal
Phopf Phopf
vector N. / at  2 T . In [26], it has been shown that the normal vector to T
d Ref./ d˛
at . / is , .
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

d
i.e.,d
d Ref./
For calculating ,
the Hessian of f with respect to x and  needs to be
d
computed. In order to get rid of calculating the Hessian, d˛
d
is calculated with respect to
dA d˛
d
. According to Section 3.2, d  can be obtained when  is a scalar variable.
Here,  is a vector of the loading level of the system. For each dimension of vector
, d˛=d  is obtained with respect to d˛=d  by Eq. (16). In a similar way, d˛=di for
the i th load can be calculated. In this case, the changes in A, i.e., A, is obtained when
there are small changes in  for an individual bus.
Assume that the hypersurface is continuous and convex. Therefore, the iterative
method can be used to compute the sensitive direction of load increase.
The procedure of the proposed method for the detection of the sensitive direction is
as follows.
1. Set the iteration i D 0 and assume the initial direction of load increase N 0. Here,
N 0 D . PmP 1 P i ; PmP 2 P i ; PmP 3 P i ; : : : ; PmP mP i /, where m is the total number of
iD1 iD1 iD1 iD1
the load and P i is the active power of the i th load. It is assumed that the power
factor of each load keeps constant along the load increase.
2. Set i C 1 and increase the load along direction Ni 1 until HB occurs. The
increasing of load in the first iteration is accomplished according to initial load.
3. Determine the DLM and the corresponding active load power vector by Pi D
P0 C Ni 1 DLM i .
4. In the neighborhood of  in which bifurcation hass occurred, the load of the j th
bus increases and ddA d˛
j , and as a consequence d j , can be computed. This process
d˛
continues until d
is computed. Set
0 1
d˛1 d˛2 d˛m
B C
B d1 d2 dm C
Ni D B m
B ; m ;:::; m C: (19)
@ X d˛ X d˛ X d˛ C A
i D1
d i D1
d i D1
d

5. Repeat Steps 2–4 until the changes of Ni is within the specified tolerance.
The final Ni corresponding P  is obtained as the sensitive direction. Note that the
hopf
iteration converges in one step if T is a hyperplane. At each iteration, Ni indicates
Phopf
the direction of the point closest to P 0 on the tangent hyperplane T .
If the iteration converges exponentially to the fixed point, then the parameter PS D
P0 C NS DLM S specifies a locally closest bifurcation. If  is the parameter of the
 722 E. Nasr Azadani et al.
Phopf
locally closest bifurcation and T is not too concave at  , then the direction N s is
an exponentially stable fixed point of the iteration. Note that when the iteration converges
to the locally closest bifurcation, it is not necessarily a globally closest bifurcation.

4. PSO

4.1. Overview of PSO

PSO is a multi-agent search technique that traces its evolution to the emergent motion of
a flock of birds searching for food. It uses a number of particles that constitute a swarm.
Each particle traverses the search space looking for the global minimum (or maximum).
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

In a PSO system, particles fly within a multi-dimensional search space. During flight,
each particle adjusts its position according to its own experience as well as the experience
of neighboring particles, making use of the best position encountered by its neighbors and
itself. Similarly, the swarm direction of a particle is determined by the history experience
obtained by a set of its neighboring particles and itself.
The modified velocity and position of each particle can be calculated as shown in
Eqs. (20) and (21):

pvd C1 D cf  Π pvd C ac1 rand. /  .Pbest ppd /

C ac2 rand. /  .Gbest ppd /; (20)

ppd C1 D ppd C pvd C1 ; (21)

where rand. / is a uniform random value in the range Œ0; 1, and cf is the constriction
factor, which is a function of ac1 and ac2 according to Eq. (22):

2
cf D p ; (22)
j2 ac ac2 4acj

where ac D ac1 C ac2 and ac  4. The appropriate selection of inertia weight provides
a balance between global and local explorations. In general, is set according to Eq. (23):
max min
D max  iter: (23)
itermax

In the above procedures, the particle velocity is limited by a maximum value pvmax.
This value determines the resolution by regions that are to be searched between the present
position and the target position. This limit enhances the local exploration of the problem
space. It also realistically simulates the incremental changes of human learning. If pvmax
is too high, particles may ignore good solutions. On the other hand, a small pvmax avoids
particles to sufficiently explore beyond local solutions. The latter may lead the PSO to
fall into local minima. In many experiences with PSO, pvmax was often set at 10–20%
of the dynamic range of the variable on each dimension [27].
In [28, 29], a decaying inertia weight is proposed and tested to favor the global search
at the start of the algorithm and then the local search afterward. If the inertia weight is
not reduced with time, it is suggested to select a value 2 Œ0:8; : : : ; 1:2. Several studies
propose different values for these parameters that are considered adequate for some of
the usual benchmark functions in [29].
 Stability Constrained Optimal Power Flow 723

4.2. Simulation Process

Generally, the optimization problem can be defined as Eq. (24), in which the fitness
function includes equality and inequality constraints:

Min. F .X/;

s.t. G.X/ D 0; (24)

H  H.X/  HN ;

where x is the vector of optimization variables, i.e., with lower bounds x and upper
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

bound x; F is the scalar optimization function; G is the equality vector function defined
in Eq. (3); and H is an inequality vector function with lower bound H and upper
bound H .
In this case, the control variables, i.e., power demand and supply bid, are the input
variables of the PSO algorithm. Therefore, the position vector of PSO can be represented
as x D ŒPg ; Pl .
The computation procedure of the PSO technique is described in the following steps.
Step 1. Input the parameters of system and specify the lower and upper boundaries
of each variable and constraint.
Step 2. Initialize the particles of the population randomly. After that, each pv should
be initialized to a value drawn from the uniform random distribution in the
interval Œ pv max; pv max. To ensure uniform velocity through all dimensions,
the maximum velocity in the j th dimension is represented as follows:

pv max D .pp max pp min /  : (25)

The initial Pbest of individual i is a set of initial positions of the individual

i , and the initial Gbest is determined as the position of an individual with
minimum fitness.
Step 3. Update the velocity vector using Eq. (20). The parameters in Eqs. (20)
and (22) are selected as follows:

D 0:7968; ac1 D ac2 D 2:

Step 4. Modify the position considering constraints: The position of each individual
is modified by Eq. (21) based on its updated velocity. If any element of an
individual violates its constraints, then it should be replaced by the particle
calculated in the previous iteration that satisfies the constraints.
Step 5. Update Pbest and Gbest; the Pbest of each particle in iteration d C 1 is
updated as follows:

Pbest id C1 D ppid C1 if Fid C1  Fid ,

(26)
Pbest id C1 D ppid if Fid C1 > Fid ,

where Fi is the fitness function in individual i which is defined in Eq. (2).

Gbest in iteration d C 1 is a set of the best evaluated positions among all
elements of Pbestd C 1.
 724 E. Nasr Azadani et al.

Step 6. End the iterations if the stopping criteria is satisfied; otherwise, go to Step 3.
Clearly, if the maximum number of iterations is reached, the operation shall
be terminated.

5. Mixed SSC-OPF with Sensitive Loading Direction

(SSC-OPF-SLD)
Figure 1 depicts the flowchart of the proposed technique for SSC-OPF with sensitive
loading direction associated with an optimal market solution.
If the procedure for DLM detection and determining the sensitive load direction
accomplished for each particle, the time consumption for running this procedure would be
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

too expensive and impractical for any larger system. In the proposed method, determining
the sensitive direction with the associated DLM is computed after running SSC-OPF.
This technique works as follows.
0) The loading parameter is initialized, which is the base case or pure market clearing
problem. In this case, kO can be set to the maximum loading level for the operating
point.
1) The SSC-OPF problem described in Eq. (17) is solved using the current value
of parameters. In the first iteration, the increase load direction for computing the
DLM is set by the initial generation and load.

Figure 1. Flowchart of proposed SSC-OPF-SLD.

 Stability Constrained Optimal Power Flow 725

2) The current SSC-OPF solution is used to set the initial direction of load increase
N 0 for determining the sensitive loading direction.
3) Given the loading direction Ni , the procedure of DLM detection is used to deter-
mine the DLM considering the SSC-OPF solution as an initial value, operating
point, and sensitive direction as a direction of load increase. Note that determining
the sensitive direction is accomplished after SSC-OPF; therefore, maybe the DLM
associated with this sensitive direction does not guarantee DLM req .
4) If DLM  DLM req , then the algorithm stops; otherwise, the procedure returns to
Step 1 and the sensitive loading direction obtained in Step 2 is used to compute
the DLM in the SSC-OPF.
Notice that contingencies can be taken into account directly so that system security
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

can be properly handled. In addition, the iterative process proposed in this article allows
controlling the value of the loading parameter.

6. Numerical Results
The proposed technique is applied to the IEEE 14-bus test system [30] for comparison
purposes. In this section, the results of applying the standard OPF and the SSC-OPF-SLD
to the IEEE 14-bus test system are presented and discussed.
A single-line diagram of the IEEE 14-bus test system is shown in Figure 2. It consists
of five synchronous machines with IEEE type-1 exciters, two of which are synchronous
compensators used only for reactive power support. There are 11 loads in the system
totaling 259 MW and 81.3 Mvar. The market bidding data is illustrated in Table 1 (the
GENCO and ESCO numbers in this table correspond to the bus number in Figure 2).
This test system has enough generation and load to simulate an electricity market
and, thus, produce meaningful results that allow the analysis of the proposed techniques.
In addition, the use of a larger system would not allow the ready analysis and, thus, the

Figure 2. Single-line diagram of the IEEE 14-bus system.

 726 E. Nasr Azadani et al.

Table 1
Market bidding data for IEEE 14-bus system

PL or PG (102 MW) Cs or CD ($/MWh)

GENCO 1 2.3 9.5

GENCO 2 0.4 10.20
GENCO 3 0 11.30
ESCO 1 0.076 13.00
ESCO 2 0.217 13.20
ESCO 3 0.112 12.10
ESCO 4 0.478 12.33
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

ESCO 5 0.149 12.24

ESCO 6 0.061 13.55
ESCO 7 0.09 14.66
ESCO 8 0.295 13.67
ESCO 9 0.035 14.62
ESCO 10 0.135 14.22
ESCO 11 0.942 14.45

illustration of the differences between the SSC-OPF-SLD technique and a standard OPF,
since a large number of data and computations would be required to arrive at conclusions
that can also be attained with this reduced size system.
Table 2 illustrates generation and load powers, which are associated with GENCOs
and ESCOs in standard OPF, as the loading level (k) is increased from its normal value.
The DLM does not take into account the standard OPF. As shown in Table 2, the social
welfare increases when the loading level is higher. In other words, the system tends
to move to a higher loading level, and consequently, it will have a smaller distance to

Table 2
Generation and load powers with respect to loading level in standard OPF

k 0.05 0.1 0.15 0.2 0.25 0.3 0.35

GENCO 1 2.4720 2.6185 2.7553 2.7994 2.8428 2.8856 2.9276

GENCO 2 0.4 0.4 0.4 0.4 0.4 0.4 0.4
GENCO 3 0 0 0 0.0981 0.1888 0.2798 0.3716
ESCO 11 0.0367 0.0385 0.0402 0.042 0.0437 0.0455 0.0472
ESCO 13 0.1417 0.1485 0.1552 0.1620 0.1687 0.1755 0.1822
ESCO 3 0.9891 1.0362 1.0833 1.1304 1.1775 1.2246 1.2717
ESCO 5 0.0798 0.0836 0.0874 0.0912 0.0950 0.0988 0.1026
ESCO 2 0.2278 0.2387 0.2495 0.2604 0.2712 0.2821 0.2930
ESCO 6 0.1176 0.1232 0.1288 0.1344 0.14 0.1456 0.1512
ESCO 4 0.5019 0.5258 0.5497 0.5736 0.5975 0.6214 0.6452
ESCO 14 0.15645 0.1639 0.1713 0.1788 0.1862 0.1937 0.2011
ESCO 12 0.0640 0.0671 0.0701 0.0732 0.0762 0.0793 0.0823
ESCO 10 0.0945 0.099 0.1035 0.108 0.1125 0.117 0.1214
ESCO 9 0.3097 0.3245 0.3392 0.354 0.36875 0.3835 0.3982
Social 26.05753 57.43323 87.58106 115.2853 142.6497 169.6674 196.2777
welfare ($)
 Stability Constrained Optimal Power Flow 727

Table 3
Social welfare and DLM with line 1-5 outage

k 0.15 0.2 0.25 0.3 0.35

Social welfare($) 0 24.06454 47.75707 70.97609 93.7102

DLM 0.358922 0.337742 0.308688 0.272857 0.233039

bifurcation. In addition, when the loading level is increased to 0.2 p.u., GENCO 2 and
GENCO 3 inject “base” powers to the system. Power injection of GENCO 2 is constant
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

regarding increased load level.

The outage of each line affects social welfare and the DLM when different contin-
gencies are applied to the system. In this article, a line 1-5 outage is determined to be
the most sensitive contingency from an oscillatory stability point of view. This outage
is the main concern in this article, and thus, it is used in standard OPF and SSC-OPF-SLD
under the N 1 security criterion.
The social welfare and DLM as the loading level with line 1-5 outage are illustrated
in Table 3. As shown in Table 3, if the loading level (k) is 0.15, the social welfare will
be zero. In addition, the DLM of the system will be lower than the normal state because
of line outage. Note that the increased load direction, which is used in DLM in standard
OPF, is regarding the initial load.
The generation powers by GENCOs with and without contingency obtained by
solving Eq. (1) with respect to the ESCO loading factor (k) are illustrated in Figure 3.
Notice that GENCO 2 supplies all the demand needs when a line outage occurs, and
it supplies the “base” power to the system in a normal state. Furthermore, the power
ejection of GENCO 1 and GENCO 3 increase as the ESCO loading factor.

Figure 3. Generation powers by GENCOs with and without contingency. (color figure available
online)
 728 E. Nasr Azadani et al.
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

Figure 4. DLM with respect to ESCO loading factor in standard OPF. (color figure available
online)

Figure 4 shows the DLM of a system in standard OPF as the ESCO loading factor.
The generation and load of GENCOs and ESCOs and their social welfare in SSC-OPF-
SLD are shown in Table 4. In this case, the required DLM is set to 0.3 p.u. and 0.4 p.u.,
and the proposed sensitive direction applied in the OPF.
The computed effort for the proposed method is acceptable, and the proposed SSC-
OPF-SLD converges in three or four iterations; i.e., it is needed to compute OPF three
or four times for having a DLM required with respect to sensitive loading direction.

Table 4
Generation and load of GENCOs and ESCOs and their social welfare in SSC-OPF-SLD

Normal state Line outage 1-5

DLM req D 0:3 DLM req D 0:4 DLM req D 0:3 DLM req D 0:4
GENCO 1 2.1920 1.8265 1.920 1.6812
GENCO 2 0.8980 1 1 1
GENCO 3 0.4084 0.4944 0.6 0.6
ESCO 11 0.0439 0.0396 0.0455 0.035
ESCO 13 0.1755 0.1755 0.1755 0.1755
ESCO 3 1.2246 1.2246 1.2246 1.2246
ESCO 5 0.0989 0.076 0.0808 0.0970
ESCO 2 0.2818 0.2821 0.2821 0.2821
ESCO 6 0.1293 0.1120 0.1305 0.112
ESCO 4 0.6214 0.5182 0.6150 0.478
ESCO 14 0.1937 0.1908 0.1937 0.1935
ESCO 12 0.0793 0.0793 0.0793 0.0793
ESCO 10 0.117 0.117 0.117 0.117
ESCO 9 0.3835 0.3835 0.3835 0.3312
Iteration 3 4 3 4
Social welfare ($) 152.659 118.217 64.83044 37.53603
 Stability Constrained Optimal Power Flow 729
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

Figure 5. PSO convergence in procedure of SSC-OPF-SLD. (color figure available online)

The advantage of the proposed method is that the DLM requires it as an input and
it can be set. Furthermore, N 1 contingencies criterion can be considered in DLM
detection. The optimization method, i.e., PSO, used for OPF abolishes the complexity of
Lagrangian multi-player computation.
The PSO convergence in the procedure of SSC-OPF-SLD is depicted in Figure 5.
The number of particles in this study is set at 30. As shown in [16], faster convergence
is achieved when the number of particles increased from 30 to 50.
The sensitive load direction when the required DLM is set at 0.3 and 0.4 is illus-
trated in Figure 6. The proposed method for determining the sensitive loading direction
converges in about seven iterations. When the required DLM is changed, the operating
point as the output of OPF will be changed, leading to different sensitive load directions.
Figure 7 depicts the differences of sensitive load increase direction when the system
is in normal state and when the contingency is considered. As shown in Figure 6, load

Figure 6. Sensitive load direction with different required DLM.

 730 E. Nasr Azadani et al.
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

Figure 7. Sensitive load increase direction with and without contingency.

increased direction of buses 3 and 4 is more sensitive than others when the contingency
occurs (line 1-5 outage). In this case, the required DLM is set at 0.3 p.u.
To evaluate the computational burden of the proposed SSC-OPF-SLD technique, the
CPU times were obtained from a Pentium 4 with 2.4 GHz and 1 GB RAM (Intel, China)
and took less than 10 min for the test system. The algorithm is easy to implement and
is capable of finding optimal solutions to the problem, giving more flexibility.

7. Conclusion
In this article, a technique based on iterative SSC-OPF-SLD, which includes a stability
constraint along with sensitive loading direction in the power system, is proposed.
Determining the sensitive loading direction is based on normal vector computing. The
results demonstrate the advantages of the proposed technique, which provides a set of
optimal market solutions as a function of system security. Therefore, this method allows
system operators to investigate the effect of system security on the market clearing
mechanism. Furthermore, the PSO methodology is employed to optimize this problem.
The proposed technique based on PSO is easy to implement, and it is capable of finding
optimal solutions for the non-linear constrained problem.

References
1. Milano, F., Canizares, C. A., and Invernizzi, M., “Multi-objective optimization for pricing
system security in electricity markets,” IEEE Trans. Power Syst., Vol. 18, No. 2, pp. 596–604,
May 2003.
2. Milano, F., Canizares, C. A., and Conejo, A. J., “Sensitivity-based security-constrained OPF
market clearing model,” IEEE Trans. Power Syst., Vol. 20, No. 4, pp. 2051–2060, November
2005.
3. Marszalek, W., and Trzaska, Z. W., “Singularity-induced bifurcations in electrical power
systems,” IEEE Trans. Power Syst., Vol. 20, No. 1, pp 312–320, February 2005.
4. Yue, M., Brookhaven National Lab, “Bifurcation subsystem and its application in power system
analysis,” IEEE Trans. Power Syst., Vol. 19, No. 4, pp. 1885–1893, 2004.
5. Canizares, C. A., “On bifurcation voltage collapse and load modeling,” IEEE Trans. Power
Syst., Vol. 10, No. 1, pp. 512–522, February 1995.
 Stability Constrained Optimal Power Flow 731

6. Canizares, C. A., Mithulananthan, N., Milano, F., and Reeve, J., “Linear performance indexes
to predict oscillatory stability problems in power system,” IEEE Trans. Power Syst., Vol. 19,
No. 2, pp. 1023–1031, May 2004.
7. Tomim, M. A., Lopes, B. I. L., Leme, R. C., Jovita, R., Zambroni de Souza, A. C., de Carvalho
Mendes, P. P., and Lima, J. W. M., “Modified Hopf bifurcation index for power system stability
assessment,” IEE Proc. Generat. Transm. Distrib., Vol. 152, No. 6, pp. 906–912, November
2005.
8. Hasanpor Divshali, P., Hosseinian, S. H., Nasr Azadani, E., and Vahidi, B., “Modified fast
indices for prediction of Hopf bifurcation by matrix reciprocal condition number,” The Iranian
Conference on Electrical Engineering (ICEE 2008), Tehran, Iran, May 2008.
9. Wen, X., A Novel Approach for Identification and Tracing of Oscillatory Stability and Damping
Ratio Margin Boundaries, Ph.D. Thesis, Iowa State University, Ames, IA, 2005.
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

10. Ajjarapu, V., and Christy, C., “The continuation power flow: A tool for steady state voltage
stability analysis,” IEEE Trans. Power Syst., Vol. 7, No. 7, pp. 416–423, February 1992.
11. Canizares, C. A., and Alvarado, F. L., “Point of collapse and continuation methods for large
AC/DC systems,” IEEE Trans. Power Syst., Vol. 8, No. 1, pp. 1–8, February 1993.
12. Dobson, I., and Lu, L., “New methods for computing a closest saddle node bifurcation and
worst case load power margin for voltage collapse,” IEEE Trans. Power Syst., Vol. 8, No. 3,
pp. 905–913, August 1993.
13. Zeng, Z.-C., Galiana, F. D., Ooi, B. T., and Yorino, N., “A simplified approach to estimate
maximum loading conditions in the load flow problem,” IEEE Trans. Power Syst., Vol. 8,
No. 2, pp. 648–654, May 1993.
14. Chiang, H.-D., Flueck, A. J., Shah, K. S., and Balu, N., “CPFLOW: A practical tool for tracing
power system steady-state stationary behavior due to load and generation variations,” IEEE
Trans. Power Syst., Vol. 10, No. 2, pp. 623–630, May 1995.
15. Gu, X., and Canizares, C. A., “Fast prediction of loadability margins using neural networks
to approximate security boundaries of power systems,” IET Proc. Generat. Transm. Distrib.,
Vol. 1, No. 3, pp. 466–475, May 2007.
16. Nasr Azadani, E., Hosseinian, S. H., and Moghani, J., “Market clearing considering power
system security,” Electric Power Conference (EPEC 2008), pp. 1–6, Vancouver, Canada, 6–7
October 2008.
17. Gan, D., Thomas, R. J., and Zimmerman, R. D., “Stability-constrained optimal power flow,”
IEEE Trans. Power Syst., Vol. 15, No. 2, pp. 535–540, May 2000.
18. Kodsi, S. K. M., and Cañizares, C. A., “Application of a stability-constrained optimal power
flow to tuning of oscillation controls in competitive electricity markets,” IEEE Trans. Power
Syst., Vol. 22, No. 4, pp. 1944–1954, November 2007.
19. Bakirtzis, A. G., Biskas, P. N., Zoumas, C. E., and Petridis, V., “Optimal power flow by
enhanced genetic algorithm,” IEEE Trans. Power Syst., Vol. 17, No. 2, pp. 229–236, 2002.
20. Yuryevich, J., and Wong, K. P., “Evolutionary programming based optimal power flow algo-
rithm,” IEEE Trans. Power Syst., Vol. 14, No. 4, pp. 1245–1250, 1999.
21. Poa-Sepulveda, C. A., and Pavez-Lazo, B. J., “A solution to the optimal power flow using
simulated annealing,” Int. J. Elect. Power Energy Syst., Vol. 25, No. 1, pp. 47–57, 2003.
22. Abido, M. A., “Optimal power flow using particle swarm optimization,” Elect. Power Energy
Syst., Vol. 24, No. 7, pp. 563–571, October 2002.
23. AlRashidi, M. R., and El-Hawary, M. E., “Hybrid particle swarm optimization approach for
solving the discrete OPF problem considering the valve loading effects,” IEEE Trans. Power
Syst., Vol. 22, No. 4; pp. 2030–2038, 2007.
24. Xize, N., and Jiajun, Q., “Investigation of torsional instability, bifurcation, and chaos of a
generator set,” IEEE Trans. Energy Convers., Vol. 17, No. 2, pp. 164–168, June 2002.
25. Wen, X., and Ajjarapu, V., “Application of a novel eigenvalue trajectory tracing method to
identify both oscillatory stability margin and damping margin,” IEEE Trans. Power Syst.,
Vol. 21, No. 2, pp. 817–824, May 2006.
26. Dobson, I., “Computing a closest bifurcation instability in multidimensional parameter space,”
J. Nonlin. Sci., Vol. 3, pp. 307–327, 1993.
 732 E. Nasr Azadani et al.

27. Zhao, B., Guo, C. X., and Cao, Y. J., “A multiagent-based particle swarm optimization approach
for optimal reactive power dispatch,” IEEE Trans. Power Syst., Vol. 20, No. 2, pp. 1070–1078,
May 2005.
28. Shi, Y., and Eberhart, R. C., “Parameter selection in particle swarm optimization,” Proceedings
of the Seventh Annual Conference on Evolutionary Programming, pp. 591–600, Springer-Verlag
London, UK, 1998.
29. Shi, Y., and Eberhart, R. C., “Empirical study of particle swarm optimization,” Proceedings
of the IEEE Congress on Evolutionary Computation (CEC), Vol. 3, pp. 1945–1950, 1999.
30. Milano, F., An Open Source Power System Analysis Toolbox (PSAT software version 2).
Available at: http://www.power.uwaterloo.ca/fmilano/psat.htm
Downloaded by [The University of British Columbia] at 13:05 28 September 2013

View publication stats