1560

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 4, NOVEMBER 2008

Optimal Rescheduling of Generators for Congestion

Management Based on Particle Swarm Optimization
Sudipta Dutta and S. P. Singh, Senior Member, IEEE

AbstractPower system congestion is a major problem that

the system operator (SO) would face in the post-deregulated era.
Therefore, investigation of techniques for congestion-free wheeling
of power is of paramount interest. One of the most practiced and
an obvious technique of congestion management is rescheduling
the power outputs of generators in the system. However, all generators in the system need not take part in congestion management.
Development of sound formulation and appropriate solution technique for this problem is aimed in this paper. Contributions made
in the present paper are twofold. Firstly a technique for optimum
selection of participating generators has been introduced using
generator sensitivities to the power flow on congested lines. Secondly this paper proposes an algorithm based on particle swarm
optimization (PSO) which minimizes the deviations of rescheduled
values of generator power outputs from scheduled levels. The PSO
algorithm, reported in this paper, handles the binding constraints
by a technique different from the traditional penalty function
method. The effectiveness of the proposed methodology has been
analyzed on IEEE 30-bus and 118-bus systems and the 39-bus
New England system.
Index TermsConstraints, generator sensitivities, gradient
methods, optimal rescheduling, optimization techniques, particle
swarm optimization, transmission congestion management.

I. INTRODUCTION

ITH increasing demand for electric power all around

the globe, electric utilities have been forced to meet
the same by increasing their generation. However, the electric
power that can be transmitted between two locations on a
transmission network is limited by several transfer limits such
as thermal limits, voltage limits and stability limits with the
most restrictive applying at a given time. When such a limit
is reached, the system is said to be congested. Ensuring that
the power system operates within its limits is vital to maintain
power system security, failing which can result in widespread
blackouts with potentially severe social and economic consequences. Congestion management, that is, controlling the
transmission system so that transfer limits are observed, is
perhaps the fundamental transmission management problem
[1]. The methods generally adopted to manage congestion include rescheduling generator outputs, supplying reactive power
support or physically curtail transactions. System operators
generally use the first option as much as possible and the last
one as the last resort.

Manuscript received July 16, 2007; revised February 04, 2008. First published August 29, 2008; current version published October 22, 2008. Paper no.
TPWRS-00504-2007.
The authors are with the Department of Electrical Engineering, Institute of Technology, Banaras Hindu University, Varanasi, India (e-mail:
sps_ee@bhu.ac.in; sps5957@indiatimes.com).
Digital Object Identifier 10.1109/TPWRS.2008.922647

Several techniques of congestion management have been reported in literature [2]. The form of deregulated electric power
industry differs from country to country as well as between different regions of a country. Different models to deal the different
transactions, interactions between properties and limitations of
the transmission system and the economic efficiency of the energy market have been mentioned in [3]. Congestion management techniques applied to various kinds of electricity markets
are presented in [4]. Prioritization of electricity transactions and
related curtailment strategies in a system where pool and bilateral/multilateral dispatches coexist is proposed in [5]. In [6],
congestion management ensuring voltage stability is addressed.
An optimal topological configuration of a power system as a
tool of congestion management is presented in [7]. A corrective switching operations of transmission lines is used instead
of generation rescheduling to alleviate congestion in this paper.
Literature on optimal power flow (OPF)-based congestion
management schemes for multiple transaction systems are available. In [8], an OPF-based approach that minimizes cost of congestion and service costs has been proposed. A coordination
mechanism between generating companies and system operator
for congestion management using Benders cuts has been discussed in [9]. In [10], a technique has been proposed for alleviating congestions due to voltage instability and thermal overloads. This also uses OPF which is solved by standard solvers. In
[11], a congestion clusters based method has been presented that
groups the system users having similar effects on the transmission constraints of interest. Here, clusters of type 1, 2 and higher
based on congestion distribution factors have been demarcated,
with type 1 cluster consisting of those with strongest and non
uniform effects on the transmission constraints of interest. The
clusters based on dc load flow form an effective congestion management market where readjustments of transactions in the type
1 cluster help to eliminate congestion. A zonal model based on
ac load flow was proposed in [12] and [13]. Zones have been
identified based on sensitivity values in these works also. However, in both [11] and [12], it is necessary to compute the sensitivity values for all the buses in the system which, given a practical power system, calls for a large amount of computational
effort. Sensitivities of line flow to changes in generation have
been used in [14] to alleviate congestion but no effort has been
made to reduce the number of participating generators. In [15], a
technique has been proposed for selection of participating generators based on sensitivity to current flow on congested line as
well as the generation bids. However an optimal selection of the
design variables is essential for regulating the number of participating generators in this work. A method of overload alleviation
by real power generation rescheduling based on relative electrical distance (RED) concept has been introduced in [16]. This

0885-8950/$25.00 2008 IEEE

DUTTA AND SINGH: OPTIMAL RESCHEDULING OF GENERATORS FOR CONGESTION MANAGEMENT

technique claims to minimize the system losses and maintain

a better voltage profile and hence more stability margin. However here, the bids of individual generation units and costs of
rescheduling are not taken in concern in this work. Generators
having same RED but different price bids must reschedule their
outputs in such a way that the total cost of rescheduling is minimum. This problem has not been addressed in [16].
The main intent of the present paper is to propose a technique for reducing the number of participating generators and
optimum rescheduling of their outputs while managing congestion in a pool at minimum rescheduling cost. In a congested
power system, the incremental or decremented change in power
outputs of all the generators do not affect the power transmitted
on the congested line to the same extent. As such, there is no
need to reschedule the outputs of generators whose generations
are less critical to the congested line flow. To optimally select
the generators participating in congestion management, the sensitivities of the generators to the congested line are used. The
generators actually participating in congestion alleviation are
chosen based on a criterion which reduces the number of participating generators.
The second major purpose of the present paper is to explore
the ability of particle swarm optimization technique in solving
the congestion management problem. The congested system is
modeled as an optimization problem. Conventional methods of
solution of OPF are based on search direction determined from
derivative of the function. Therefore it becomes imperative to
express the problem in the form of continual differentiable function; otherwise, these methods become less efficient. In order to
overcome this problem, the present paper solves the optimization problem using particle swarm optimization. Generally, in
optimization algorithms, the value of objective function is regarded as the fitness function and the binding constraints are
handled as penalty functions method. This method has many
disadvantages because the penalty parameters are empirically
assigned and are deeply affected by the problem model. However in this paper, the constraints have been handled using a
novel technique (as described in Section IV).
This paper illustrates the effectiveness of the proposed
method on the congestion management problem considering
the 39-bus New England system and the IEEE 30- and 118-bus
systems.
II. PARTICLE SWARM OPTIMIZATION
The method of particle swarm optimization is an evolutionary
algorithm and was first introduced by Kennedy and Eberhart
in 1995 [17]. The motivation was social behavior of organisms
such as fish schooling and bird flocking. It was observed that a
flock of birds stochastically find food in an area. Not all birds
in a flock know the exact location of food but they know the
position closest to it (the food). The simplest and most effective
way to search for food is to search the area around the present
best position, i.e., the position closest to food.
Similar to seeking food, the solution to an optimization
problem or the best solution is found out from a solution
space with a population based search procedure in which the
particles, like birds, change their positions (states) with time.
Each particle represents a potential solution to a problem

1561

dimensional space (where the number of dimenin an

sions corresponds to the number of variables). The particles
are randomly generated (a particle size between 10 to 100
is usually considered sufficient) initially with two parameters each-position and velocity in the dimensional space
and velocity
such as position
.
Each particle is then flown over the search space in order
to find potential solution regions of the landscape and adjusts
its flying velocity and direction according to its own flying
experience as well as that of its neighbors. Positions of the
particles (tentative solutions) are evaluated at end of every
iteration relative to an objective or fitness value. Particles are
assumed to retain memory of the best positions they have
achieved in course of flying and share this information among
the rest. The collective best positions of all the particles
taken together is termed as the global best position given as
and the best position achieved by
the individual particle is termed as the local best or position
.
best and for the th particle given as
Particles use both of these information to update their positions
and velocities as given in the following equations [18]:

(1)
(2)
where is the inertia weight; and are random values beand
are two positive constants, called
tween 0 and 1;
acceleration constants; generally
represents iteration number.
The particles continue flying and seeking solution and hence
the algorithm continues until a pre-specified number of maximum iterations are exceeded or exit criteria are met. The accuracy and rate of convergence of the algorithm depends on the
appropriate choice of particle size, maximum velocity of particles and the inertia weight. However, no specific guideline is
available to select the particle size. Moreover, it also varies from
problem to problem. As a result, one has to choose it by trial
and error. The maximum velocity of individual particles should
, is
be chosen very judiciously. If the maximum velocity,
too high, the particles may fly past the best solution without
discovering it and if it is too low particles may fail to explore
sufficiently beyond local solutions [19]. The inertia weight parameter is considered important for the convergence of the algorithm. It controls the impact of previous history of velocities on the current velocities of particles and hence regulates
the local and global exploration capabilities of the particles. A
large inertia weight facilitates exploration, i.e., searching newer
areas while a small value tends to facilitate exploitation, i.e., a
finer searching of current search area. The value of the inertia
weight parameter is normally kept between 0.4 and 0.9. Thus,
the choice of inertia weight should be carefully made.
Compared with traditional optimization algorithms, PSO
does not need the information of the derivative of functions in
the process model. The algorithm can work as long as fitness
values for optimization model can be calculated. Besides, the

1562

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 4, NOVEMBER 2008

algorithm of PSO is simple enough for people to understand it

easily and has a profound intellectual background at the same
time.
The technique of PSO has already been applied in several
problems of optimization in the power system. In [20], PSO
has been applied to solve the economic dispatch of generators
in a power system. A technique has been proposed in [21] to
control reactive power and voltage to maintain power system
security from voltage stability point of view. Sensitivity based
congestion management using PSO has been discussed in [22].
However, it does not reveal about procedure of handling of constraints. The PSO algorithm proposed in the present paper has
been modified to handle the constraints in a way different from
the traditional penalty function method. Moreover the algorithm
in [22] has only been tested on the 39-bus New England system.
The robustness of the proposed method however, has been tested
on two other systems namely the IEEE 30-bus and 118-bus
systems.

The active power injected at a bus-s can be represented as

(9)
where

is the active load at bus-s.

can be expressed as

(10)
where is the number of buses in the system.
Differentiating (10) w.r.t.
and , the following relations
can be obtained:

III. CONGESTION MANAGEMENT PROBLEM FORMULATION

The generators in the system under consideration have different sensitivities to the power flow on the congested line. A
change in real power flow in a transmission line k connected
between bus i and bus j due to change in power generation by
generator g can be termed as generator sensitivity to congested
line (GS). Mathematically, GS for line k can be written as
(3)
where
is the real power flow on congested line-k;
is the
real power generated by the th generator.
The basic power flow equation on congested line can be
written as

(11)

(12)
Neglecting P-V coupling, the relation between incremental
change in active power at system buses and the phase angles of
voltages can be written in matrix form as
(13)
where

..
.

(4)
where and are the voltage magnitude and phase angle reand
represent, respectively,
spectively at the th bus;
the conductance and susceptance of the line connected between
buses i and j; neglecting P-V coupling, (3) can be expressed as

..
.

(14)

Thus
(15)
(16)

(5)
where
The first terms of the two products in (5) are obtained by
differentiating (4) as follows:

(6)

(7)
(8)

(17)
and
in (5),
To find the values of
needs to be found out. However,
is a singular matrix of
rank one deficiency. So it is not directly invertible. The slack bus
in the present work has been considered as the reference node
and assigned as bus number 1. The elements of first row and first
can be eliminated to obtain a matrix
which
column of
, where
represents a
can be inverted to obtain matrix
matrix with its first row and column deleted or a vector with the

DUTTA AND SINGH: OPTIMAL RESCHEDULING OF GENERATORS FOR CONGESTION MANAGEMENT

first element deleted. Using these relations the following equation can be obtained:
(18)
The actual vector
can be found by simply adding the eleto (18) as shown by the following relation:
ment

..
.

(19)

, being the
The second term of the sum in (19) vanishes as
change in phase angle of slack bus is zero. Accordingly, (19)
reduces to
(20)

Thus required elements of

and
are
found out from (20).
It is to be noted that the generator sensitivity values thus obtained are with respect to the slack bus as the reference. So the
sensitivity of the slack bus generator to any congested line in the
system is always zero.
denotes how much active power flow over a transmission line connecting bus-i and bus-j would change due to active
power injection by generator g. The system operator selects the
generators having non uniform and large magnitudes of sensitivity values as the ones most sensitive to the power flow on the
congested line and to participate in congestion management by
rescheduling their power outputs.
Based on the bids received from the participant generators,
the amount of rescheduling required is computed by solving the
following optimization problem:

1563

the line flow limit of the line connecting bus-i and bus-j.
is the number of participating generators,
is the number of
and
denote respectransmission lines in the system,
tively the minimum and maximum limits of generator outputs.
It can be seen that the power flow solutions are not required
during the process of optimization. The power balance, except
accounting for losses, is taken care by (24). However, final generation allocation at slack bus is obtained at the end of optimization process which takes care of system losses; although (24)
includes change in active power at slack bus. This is the added
advantage of the proposed formulation.
IV. SOLUTION BY PSO
In this paper each particle has N variables where N is the total
number of generators taking part in congestion management.
Each variable represents output of participating generators submitting the bidding curves. Particle evolution based on fitness
of particles and selection operation of GB (global best) and
(local best), is used to meet the constraints.
Fitness is an index used to evaluate the superiority of the
particle. Traditionally, the objective function is regarded as the
fitness function and the inequality constraints are converted to
penalty functions and added to the objective function. The drawback of this method is that an excellent particle can be misjudged as inappropriate for the penalty factors. Besides, penalty
parameters are usually assigned by empirical approach and are
deeply affected by the problem model. For the sake of avoiding
this, a binary fitness has been used: one for optimal objective and
the other for the binding constraints. Optimal objective fitness
is equal to the value of the expression (21) which represents the
cost of active power rescheduling and hence the cost acquired
to curb congestion or congestion cost. Binding constraints fitness value is adopted to scale the level of violation, calculated
as follows:
(25)

Minimize

(21)

subject to

(22)

(23)
(24)
is the real power adjustment at bus-g and
are
where
the incremental and decremented price bids submitted by generators. These are the prices at which the generators are willing
is the power flow caused
to adjust their real power outputs.
is
by all contracts requesting the transmission service.

where z is the value of the inequality constraint,

and
are the lower and higher limits of the inequality constraints.
The fitnesses to binding constraints of the particles are considered first and if a particle does not satisfy the binding constraints, it is regenerated. This way feasible particles are generated that guarantee the fulfillment of binding constraints superior to infeasible particles that violate the binding constraints.
Thus entering into feasible region is considered before obtaining
global optimal solution. There is no need to set up the penalty
parameter.
The PSO algorithm works as follows.
Step 1) Particles are generated and initialized with values of
position and velocity. Each particle has N dimensions where N denotes the number of participating
generators, and the values of these N variables are
the amount of rescheduling required by generators
to manage congestion.
Step 2) Equation (24) is tested based on the system states
represented by an individual particle. If that particle

1564

Step 3)

Step 4)

Step 5)
Step 6)

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 4, NOVEMBER 2008

does not satisfy the equality constraints, it is regenerated.

The binding constraints fitness values for the particles are determined. If a particle does not satisfy the
fitness requirement, it is regenerated.
The optimal objective fitness values are calculated
for all the particles. Then the values of position best
and global best are determined.
Position and velocities of particles are updated.
If the maximum number of iterations is exceeded
or some pre-specified exit criteria is satisfied, the
program is stopped. Else returned to Step 2.

Fig. 1. Comparison of results of 39-bus New England system.

TABLE I
GSS FOR 39-BUS NEW ENGLAND SYSTEM (CONGESTED LINE 15-16)

V. NUMERICAL EXAMPLES
The PSO algorithm for congestion management, delineated in
the previous sections, has been implemented using Visual C++
programming language. The performance of the algorithm has
been studied on 39-bus New England system [23], IEEE 30-bus
[24], 118-bus [25] systems. The 39-bus system was intentionally
chosen in order to compare the performance of the proposed
method with four reported methods in [11][13] and [16]. The
approach presented in [12] and [13] being similar, results of [12]
only have been considered for comparison. The performance of
the proposed method on IEEE 30-bus and 118-bus has also been
compared with [15].
While using PSO to solve the congestion problem, algorithm
validity and influence of different PSO parameters have also
been studied.

TABLE II
COMPARISON OF GENERATOR RESCHEDULING FOR CONGESTION
MANAGEMENT IN LINE (16-17) OF 39-BUS NEW ENGLAND SYSTEM

A. The 39-Bus New England System

The 39-bus New England system has been considered for
bringing out the effectiveness of the proposed technique. The
39-bus system consists of ten generator buses and 29 load
buses. The congested line is connected between buses numbered 14 and 34. Generators selected for rescheduling are those
connected to buses 3, 8 and 10. The PSO parameters selected
for the algorithm are as follows:
Particle size
Acceleration constants
Inertia weight,
The algorithm is found to converge within 100 iterations. The
value of the objective function, i.e., the cost of active power
rescheduling is dependent on the incremental and decremental
price bids submitted by the participating generators. The amount
of rescheduling required for congestion management has been
compared with three different methods for this case.
Method 1This refers to the approach discussed in [11].
Method 2This refers to the approach discussed in [12] and
[13].
Method 3This refers to the proposed method as discussed
in the present paper.
The comparison of the amounts of active power rescheduled
by the three methods has been shown in Fig. 1. As can be seen
Method 3 (proposed) generates better results as it suggests lower
amounts of rescheduled generation of generators.
For comparing the performance of proposed method with
method reported in [16], an outage has been created on the
line joining buses 14 and 34 resulting in congestion of the line

TABLE III
SOME SYSTEM PARAMETERS BEFORE AND AFTER RESCHEDULING

joining buses 15 and 16. According to [16], all of the ten generators take part in congestion management. However, based
on sensitivity analysis proposed in the present paper (given in
Table I), it is apparent that only six of them are sufficient to
manage congestion successfully without exceeding the generation limits of generators. The comparative results are tabulated
in Tables IIIV. From these results, it can be clearly seen that
the system losses are lower, voltage profile obtained is better
and congestion is managed better as indicated by lower overload factor, by the proposed method.
B. Modified IEEE 30-Bus System
The IEEE 30-bus system consists of six generator buses and
24 load buses. The numbering of buses has been done in a way
that the generator buses are numbered first followed by the load
buses. Slack node has been assigned bus number 1. Here, two

DUTTA AND SINGH: OPTIMAL RESCHEDULING OF GENERATORS FOR CONGESTION MANAGEMENT

TABLE IV
FLOW CHANGE THROUGH THE CRITICAL LINE
BEFORE AND AFTER RESCHEDULING

1565

TABLE VIII
EFFECT OF VARIATION IN PSO PARAMETERS (RESCHEDULING
COST IN $/DAY AFTER 1000 ITERATIONS), 30-BUS SYSTEM

TABLE V
CONGESTED LINE DETAILS OF 30-BUS SYSTEM

TABLE VI
GSS FOR IEEE 30-BUS SYSTEM (CONGESTED LINE 2-1)

TABLE VII
GENERATOR PRICE BIDS FOR IEEE 30-BUS SYSTEM ($/MW -DAY)

lines have been found to be congested, that are between buses 2

and 1 and that between buses 9 and 2.
Congested line flow of the 30-bus system has been presented
in Table V.
The values of generator sensitivities computed for the congested line 2-1 are presented in Table VI.
Close values of sensitivities point out that the 30-bus system
is practically a very small system compared to a realistic power
network. All the generators show strong influence on the congested line flow. This is because a small system is generally
very tightly connected electrically. Thus, all the generators are
chosen to participate in congestion management and the next
part of the algorithm, i.e., solving the congestion management
problem using PSO has been proceeded with. It is also indicative that significant simplifications can be visible only on large
networks such as the 118-bus system described in the following
section.
The generator cost curves have been assumed to be quadratic
such that cost of rescheduling is proportional to the square of the
change in active power output as represented in Table VII. Generators selected for participation in congestion management are
asked to reschedule their outputs optimally on the basis of their
bids so that the cost of rescheduling gets minimized. However
the algorithm does not take into account the change of nodal
prices or generator bids at pre or post congestion management
situation.
The influence of the PSO parametersinertia weight and
population size on the convergence of the algorithm has been
studied, as tabulated in Table VIII. The size of particles has been
increased from 10 to 100 in steps of 10 and the inertia weight has
also been varied between 0.4 and 0.9 in steps of 0.1 for determining the optimal parameters. The parameters finally selected
for the algorithm for which consistent and superior results were
found are as follows:

Fig. 2. Plot of generator sensitivities and active power rescheduled for generators of 30-bus system.

Particle size
Acceleration constants
Inertia weight,
The algorithm is found to converge between 400500
iterations.
Total system losses before congestion management were
found to be 21 MW, while the system losses after congestion
management decreased to 15 MW.
The relationship of generator sensitivities with the change
in power outputs of generators for congestion management has
been graphically represented in Fig. 2 and rescheduled power at
various generators in Table IX.
For comparison purpose, the methodology for generator selection proposed in [15] has been made use of using two different values of MF, which are design variables that regulate the
number of generators taking part to be chosen by the operator
(as defined in [15]). Generators having positive values of sensitivities are multiplied by their respective bidding values and arranged in a decreasing order according to the resultant product.
Generators having negative sensitivities are divided by their bidding values and arranged in increasing order according to the result. The top entry in each category (positive and negative) multiplied by MF determines the cut-off criteria for generators to be
selected. MF varies between 0 to 1, the higher its value, lesser
is the number of selected generators. Following this procedure,
the generators selected to participate in congestion management
and their rescheduled powers are presented in Table IX.

1566

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 4, NOVEMBER 2008

TABLE IX
COMPARISON OF RESULTS OF 30-BUS SYSTEM

TABLE X
CONGESTED LINE DETAILS OF 118-BUS SYSTEM

Fig. 3. Plot of generator sensitivities versus generators of 118-bus system.

It is found that although the total rescheduled power is less in

technique used in [15], the cost of rescheduling comes out more
than the technique proposed in the present paper.
C. Modified IEEE 118-Bus System
The IEEE 118-bus system has been considered as an example
of a large power network. It consists of 54 generator buses and
64 load buses. Once again the numbering of buses is started
with the generator buses followed by the load buses with slack
node designated as bus number 1. Here the congested line is
connected between buses 13 and 16.
Congested line flow of the 118-bus system has been presented
in Table X.
The values of generator sensitivities computed for congested
line 1316 are presented in Table XI. The plot of generator sensitivities for the generators in the 118-bus system is depicted in
Fig. 3. It can be seen from the table as well as the figure that
generators numbered 13 through 18 have non uniform values

TABLE XI
GSS FOR IEEE 118-BUS SYSTEM (CONGESTED LINE 1316)

of sensitivities to the congested line power flow. The magnitudes of the sensitivity values are also much larger. Generator
numbered 13 has the maximum negative sensitivity whereas the
generator numbered 16 has the maximum positive sensitivity to
the congested line power flow. It is interesting to note that these
represent the same buses to which the congested line is connected. Accordingly generators numbered 13 through 18 are selected for participation in congestion management. Bus 1 is a
slack bus and is also chosen to participate in the rescheduling
process to manage system losses. Thus, while the number of
generators in the system is 54, the number of generators participating in congestion management is only 7. It is evident that
there is a drastic reduction in number of generators whose power
outputs need to be rescheduled to manage congestion. The participating generators can be divided into two groups based on
the sign of sensitivity values. A positive sensitivity value indicates that an increase in generation for that generator increases
the power flow on the line under consideration whereas a negative sensitivity value indicates that an increase in generation
decreases the power flow on the congested line.
The generator cost curves have been assumed to be quadratic
in this case as well, such that cost of rescheduling is proportional
to the square of the change in active power output as represented
in Table XII.
The influence of the PSO parametersinertia weight and
population size on the convergence of the algorithm has been
studied, as tabulated in Table XIII. In this case, PSO parameters
selected are as follows:
Particle size
Acceleration constants
Inertia weight,
The algorithm is found to converge between 400500 iterations in this case also.
Total system losses before congestion management is found
to be 140 MW. However, post congestion management the
system losses declined to 137 MW.
The rescheduled power at various generators obtained using
the proposed method has been tabulated in Table XIV. It is to be

DUTTA AND SINGH: OPTIMAL RESCHEDULING OF GENERATORS FOR CONGESTION MANAGEMENT

TABLE XII
GENERATOR PRICE BIDS FOR IEEE 118-BUS SYSTEM ($/MW -DAY)

1567

TABLE XIV
COMPARISON OF RESULTS OF 118-BUS SYSTEM

TABLE XIII
EFFECT OF PSO PARAMETERS (RESCHEDULING COST IN $/DAY
AFTER 1000 ITERATIONS), 118-BUS SYSTEM
TABLE XV
EFFECT OF CONGESTION ON GENERATOR BIDS (STUDIED FOR GENERATOR 13)

noted that although generator numbered 14 has a positive sensitivity, yet its power generation has been increased for congestion management. It can also be seen that among the generators
having positive sensitivities, generator 14 has the minimum positive sensitivity value. An increase of generation is required in
this case for power balance and generator 14 having the least
positive sensitivity is selected to increase its output. It should
also be noted here that the price bids submitted by the different
generators are extremely influential in determining the rescheduled generation levels.
For comparison purpose, once again the methodology for
generator selection proposed in [15] has been made use of using
two different values of MF. Here also it is found that although
the total rescheduled power is less in technique used in [15], the
cost of rescheduling comes out lesser in the technique proposed
in the present paper. These results have also been presented
,
in Table XIV. It is also to be noted here that for
the number of generators selected is only three, excluding the
slack bus. However, congestion is not fully managed (line limit
being 200 MW) because there is a maximum limit of increasing
or decreasing the generation on any generator which in this
MW on all generators. This
system has been taken as or
indicates that the design parameter MF must be chosen very
judiciously.

It is well known that congestion can reflect the nodal pricing

in the transmission network. Congestion often results in a
market condition which allows certain units to maximize
profits. To demonstrate this phenomenon, a study has been conducted on the 118 bus system. It is evident from the sensitivity
values that generator numbered 13 has the maximum negative
sensitivity to power flow on the congested line. It is hence at
an advantageous position among the participating generators.
When congestion has occurred on the said line, this generator
can increase its bidding price and increase its profit. Table XV
tabulates these results. It is clear that as the bids increase, the
amount of power required to be rescheduled by it decreases
(due to optimal distribution by system operator). However, the
amount paid to the generator 13 for rescheduling its output
increases.
D. General Observations
From the results of all the three systems studied, it can be
observed that the PSO algorithm has successfully been used
to manage congestion and reduce system losses. The convergence of the PSO method with parameter values as described
above has been shown in Fig. 4. The graph demonstrates that the
rescheduling cost gradually decreases with number of iterations
and converges to a minimum value. The convergence pattern is
also indicative of the fact that the selection of PSO parameters is
appropriate. Execution time of this algorithm is approximately

1568

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 4, NOVEMBER 2008

Fig. 4. Convergence property of PSO algorithm.

12 s on a Pentium IV, 2.4-GHz personal computer for all the

systems tested.
VI. CONCLUSION
The present paper focuses on demonstrating a technique for
optimum selection of generators for congestion management
and additionally the application of PSO in the solution of the
congestion management problem. Generators from the system
are selected for congestion management based on their sensitivities to the power flow of the congested line followed by corrective rescheduling. The problem of congestion is modeled as an
optimization problem and solved by particle swarm optimization technique. The method has been tested on 39-bus New England system, IEEE 30-bus and 118-bus systems successfully.
Results obtained on the 39-bus New England system has been
compared with the results reported using three other techniques.
The appropriateness of the generator selection methodology has
also been compared with reported techniques on IEEE 30-bus
and 118-bus systems. PSO algorithm has many advantages such
as simple concept and easy understanding; the entire complex
decision making is modeled by two simple (1) and (2). The robustness of the algorithm is demonstrated by solving three different networks of different sizes and complexities with equal
performance. Since the convergence of the PSO algorithm depends on the appropriate selection of particle size, inertia weight
and maximum velocity of particles, improper choice of these parameters may lead to inferior results or nonconvergence. However, test results reveal that the proposed implementation is effective in managing congestion and outperforms.
REFERENCES
[1] R. D. Christie, B. Wollenberg, and I. Wangensteen, Transmission
management in the deregulated environment, Proc. IEEE, vol. 88, no.
2, pp. 170195, Feb. 2000.

[2] A. Kumar, S. C. Srivastava, and S. N. Singh, Congestion management

in competitive power market: A bibliographical survey, Elect. Power
Syst. Res., vol. 76, pp. 153164, 2005.
[3] Y. H. Song and I.-F. Wang, Operation of Market Oriented Power Systems. New York: Springer, 2003, ch. 6.
[4] K. L. Lo, Y. S. Yuen, and L. A. Snider, Congestion management in
deregulated electricity markets, in Proc. Int. Conf. Electric Utility
Deregulation and Restructuring and Power Technologies, London,
U.K., 2000, pp. 4752.
[5] R. S. Fang and A. K. David, Transmission congestion management
in an electricity market,, IEEE Trans. Power Syst., vol. 14, no. 3, pp.
877883, Aug. 1999.
[6] A. J. Conejo, F. Milano, and R. G. Bertrand, Congestion management
ensuring voltage stability, IEEE Trans. Power Syst., vol. 21, no. 1, pp.
357364, Feb. 2006.
[7] G. Granelli, M. Montagna, F. Zanellini, P. Bresesti, R. Vailati, and M.
Innorta, Optimal network reconfiguration for congestion management
by deterministic and genetic algorithms, Elect. Power Syst. Res., vol.
76, pp. 549556, 2006.
[8] F. Jian and J. W. Lamont, A combined framework for service identification and congestion management,, IEEE Trans. Power Syst., vol.
16, no. 1, pp. 5661, Feb. 2001.
[9] H. Y. Yamina and S. M. Shahidehpour, Congestion management coordination in the deregulated power market, Elect. Power Syst. Res.,
vol. 65, no. 2, pp. 119127, May 2003.
[10] F. Capitanescu and T. V. Cutsem, A unified management of congestions due to voltage instability and thermal overload,, Elect. Power
Syst. Res., vol. 77, no. 10, pp. 12741283, Aug. 2007.
[11] C. N. Yu and M. Ilic, Congestion clusters based markets for transmission management, in Proc. IEEE Power Eng. Soc. Winter Meeting,
New York, Jan. 1999, pp. 111.
[12] A. Kumar, S. C. Srivastava, and S. N. Singh, A zonal congestion management approach using ac transmission congestion distribution factors,, Elect. Power Syst. Res., vol. 72, pp. 8593, 2004.
[13] A. Kumar, S. C. Srivastava, and S. N. Singh, A zonal congestion management approach using real and reactive power rescheduling,, IEEE
Trans. Power Syst., vol. 19, no. 1, pp. 554562, Feb. 2004.
[14] A. S. Nayak and M. A. Pai, Congestion management in restructured
power systems using an optimal power flow framework, M.S. thesis,
Univ. Illinois, Urbana-Champaign, 2002, pp. 1217.
[15] B. K. Talukdar, A. K. Sinha, S. Mukhopadhyay, and A. Bose, A computationally simple method for cost-efficient generation rescheduling
and load shedding for congestion management, Int. J. Elect. Power
Energy Syst., vol. 27, no. 5, pp. 379388, Jun.Jul. 2005.
[16] G. Yesuratnam and D. Thukaram, Congestion management in open
access based on relative electrical distances using voltage stability criteria, Elect. Power Syst. Res., vol. 77, pp. 16081618, 2007.
[17] A. Kennedy and R. Eberhart, Particle Swarm Optimization, in Proc.
IEEE Int. Conf. Neural Networks, Nov. 29Dec. 1 1995, vol. IV, pp.
19421948.
[18] R. C. Eberhart and J. Kennedy, A new optimizer using particle swarm
theory,, in Proc. 6th Int. Symp. Micro Machine and Human Science,
Nagoya, Japan, 1995, pp. 3943.
[19] H. Y. Fan and Y. Shi, Study on Vmax of particle swarm optimization, in Proc. Workshop Particle Swarm Optimization, Indianapolis,
IN, 2001, Purdue Sch. Eng. Techol..
[20] Z. L. Gaing, Particle swarm optimization to solving the economic dispatch considering the generator constraints, IEEE Trans. Power Syst.,
vol. 18, no. 3, pp. 11871195, Aug. 2003.
[21] H. Yoshida, A particle swarm optimization for reactive power and
voltage control considering voltage security assessment, IEEE Trans.
Power Syst., vol. 15, no. 4, pp. 12321239, Nov. 2000.
[22] T. Meena and K. Selvi, Cluster based congestion management in
deregulated electricity market using PSO, in IEEE Indicon 2005
Conf., Chennai, India, Dec. 1113, 2005, pp. 627630.
[23] K. R. Padiyar, Power System Dynamics: Stability and Control.. New
York: Wiley, 1996, pp. 601.
[24] O. Alsac and B. Stott, Optimal load flow with steady-state security,
IEEE Trans. Power App. Syst., vol. PAS-93, pp. 745751, 1974.
[25] L. L. Freris and A. M. Sasson, Investigation of the load flow problem,
Proc. Inst. Elect. Eng. , vol. 115, no. 10, pp. 14591466, 1968.

DUTTA AND SINGH: OPTIMAL RESCHEDULING OF GENERATORS FOR CONGESTION MANAGEMENT

Sudipta Dutta was born in Durgapur, India, in

1983. She received the B.E. degree in electrical engineering from Jadavpur University, Kolkata, India, in
2005. She is currently pursuing the M.Tech. degree
in power systems in the Department of Electrical
Engineering, Institute of Technology, Banaras Hindu
University, Varanasi, India.
Her research interests are in optimization and AI
techniques in the fields of power system operation
and control.

1569

S. P. Singh (SM94) was born in India in 1957. He

received the B.E. and M.E. degrees in electrical engineering from Allahabad University, Allahabad, India,
in 1978 and 1981, respectively, and the Ph.D. degree
from Banaras Hindu University, Varanasi, India, in
1990.
He was Postdoctoral scholar at the University of
Calgary, Calgary, AB, Canada, during 19931994.
He joined the Electrical Engineering Department of
Banaras Hindu University as a faculty member in
1981, where he is presently serving as a Professor.
His main research interests are in optimal operations of power systems,
deregulation, distribution automation, and AI applications to power systems.
Dr. Singh is a Fellow of the Institution of Engineers (India) and a life member
of the Systems Society of India.