This article was downloaded by:[University of Southampton]

On: 22 February 2008

Access Details: [subscription number 773565843]
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.informaworld.com/smpp/title~content=t713399721
Solution of Different Types of Economic Load Dispatch
Problems Using a Pattern Search Method
J. S. Al-Sumait a; J. K. Sykulski a; A. K. Al-Othman b
a
Electrical Power Engineering Group, University of Southampton, Highfield,
Southampton, United Kingdom
b
Electrical Engineering Department, College of Technological Studies, Alrawda,
Kuwait

Online Publication Date: 01 March 2008

To cite this Article: Al-Sumait, J. S., Sykulski, J. K. and Al-Othman, A. K. (2008)
'Solution of Different Types of Economic Load Dispatch Problems Using a Pattern
Search Method', Electric Power Components and Systems, 36:3, 250 - 265
To link to this article: DOI: 10.1080/15325000701603892
URL: http://dx.doi.org/10.1080/15325000701603892

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,
re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be
complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be
independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or
arising out of the use of this material.
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Electric Power Components and Systems, 36:250–265, 2008

Copyright © Taylor & Francis Group, LLC
ISSN: 1532-5008 print/1532-5016 online
DOI: 10.1080/15325000701603892

Solution of Different Types of Economic Load

Dispatch Problems Using a Pattern Search Method

J. S. AL-SUMAIT,1 J. K. SYKULSKI,1 and

A. K. AL-OTHMAN2
1
Electrical Power Engineering Group, University of Southampton,
Highfield Southampton, United Kingdom
2
Electrical Engineering Department, College of Technological Studies,
Alrawda, Kuwait

Abstract Direct search (DS) methods are evolutionary algorithms used to solve
constrained optimization problems. DS methods do not require information about the
gradient of the objective function when searching for an optimum solution. One such
method is a pattern search (PS) algorithm. This study presents a new approach based
on a constrained PS algorithm to solve various types of power system economic load
dispatch (ELD) problems. These problems include economic dispatch with valve point
(EDVP) effects, multi-area economic load dispatch (MAED), companied economic-
environmental dispatch (CEED), and cubic cost function economic dispatch (QCFED).
For illustrative purposes, the proposed PS technique has been applied to each of
the above dispatch problems to validate its effectiveness. Furthermore, convergence
characteristics and robustness of the proposed method has been assessed and inves-
tigated through comparison with results reported in literature. The outcome is very
encouraging and suggests that PS methods may be very efficient when solving power
system ELD problems.

Keywords ELD, DS method, PS method, evolutionary algorithms, optimization

1. Introduction
Scarcity of energy resources, increasing power generation costs, and ever-growing de-
mand for electric energy necessitates optimal economic dispatch (ED) in today’s power
systems. The main objective of ED is to reduce the total power generation cost while
satisfying various equality and inequality constraints. Traditionally, in ED problems, the
cost function for generating units has been approximated as a quadratic function.
A variety of optimization techniques has been applied in solving ELD problems.
Some of these techniques are based on classical optimization formulations, while others
are based on artificial intelligence methods or heuristic algorithms. Many references
present the application of classical optimization methods, such as linear or quadratic
programming, to solve ELD problems [1, 2]. Classical optimization methods are reported
to be highly sensitive to the selection of the starting points and sometimes converge to a

Received 12 October 2006; accepted 13 July 2007.

Address correspondence to Dr. Abdrulrahman Al-Othman, Electrical Engineering Department,
College of Technological Studies, P.O. Box 33198, Alrawda, 73452, Kuwait. E-mail: ak.alothman@
paaet.edu.kw

250
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 251

local optimum or diverge altogether. Linear programming methods are fast and reliable,
but have disadvantages associated with the piecewise linear cost approximation. Non-
linear programming methods are known to have problems of convergence and algorithmic
complexity. Newton-based algorithms struggle with handling a large number of inequality
constraints [3]. Methods based on artificial intelligence techniques, such as artificial
neural networks, are also presented [4, 5]. Heuristic search techniques, such as particle
swarm optimization [3] and genetic algorithms [6], have also been successfully applied
to ELD problems. Finally, hybrid methods have been developed, where the conventional
Lagrangian relaxation approach, a first-order gradient method, and multi-pass dynamic
programming (DP) are combined in [7].
Recently, a particular family of global optimization methods, originally introduced
and developed by researchers in the 1960s [8], has attracted special attention. Known
as DS methods, they are simply designed to explore a set of points around the current
position, looking for a location that has a better objective value than the existing one.
This family includes PS algorithms, simplex methods (different from the simplex used
in linear programming), Powell optimization, and others [9].
DS methods, in contrast to more standard optimization methods, are often called
derivative-free as they do not require information about the gradient or higher derivatives
of the objective function to search for an optimal solution. Thus, DS methods are a popular
choice to solve non-continuous, non-differentiable, and multimodal (i.e., multiple local
optima), optimization problems. Since the ED is one such problem, the proposed method
appears to be a good candidate to achieve efficient solutions [10].
The main objective of this study is to introduce the use of a PS optimization technique
in application to the power system ELD. The performance, effectiveness, and robustness
of the proposed method are assessed via intensive testing and comparison with results
reported in literature. The article is organized as follows: Section 2 introduces the problem
formulation, Section 3 presents a description of the proposed PS algorithm, analysis and
test results are presented in Section 4, followed by concluding remarks.

2. Problem Formulation
Generally, depending on the type of the ELD considered, the traditional formulation is
basically a minimization of summation of the fuel costs of the individual dispatchable
generators, subject to the real power balanced with the total load demand as well as with
the limits on generators’ outputs. The mathematical form of each type of ED problem
considered in this study may be described as in the following subsections.

2.1. Mathematical Form of EDVP

The inclusion of the valve-point effects is advantageous and makes the modeling of the
fuel-cost function more realistic [11–13]. However, the valve-point effects, which appear
as a sinusoidal term added to the fuel-cost functions, introduce ripples to the heat-rate
curve, and, therefore, create more local minima in the search space.
The objective function of an EDVP is to minimize F , where F is given by

N
X
F D Fi .Pi /: (1)
i D1
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

252 J. S. Al-Sumait et al.

The incremental fuel-cost function of the generation units with valve-point loadings are
represented as follows [11]:
2
Fi .Pi / D ai Pgi C bi Pgi C ci C jei  sin.fi  .Pgi.min/ Pgi //j (2)

subject to
N
X
Pgi D PD C PL (3)
i D1

Pgi.min/ < Pgi < Pgi.max/ ; i 2 Ns ; (4)

where
F is the system overall cost function;
N is the number of generators in the system;
ai , bi , and ci are the constants of fuel function of generator number i ;
ei , fi are the constants of the valve-point effect of generator number i ;
Pgi is the active power generation of generator number i ;
PD is the total power system demand;
PL is the total system transmission losses;
Pgi.min/ is the minimum limit on active power generation of generator i ;
Pgi.max/ is the maximum limit on active power generation of generator i ; and
Ns is the set of generators in the system.

2.2. Mathematical Form of MAED

In a MAED problem, the objective is to determine the most economical generation level
in a given area and the power interchange between the areas that minimize the overall
operation cost while satisfying a set of constraints as follows [14]:
hX X i
min F .Pgi / C G.tj k / (5)

subject to
X X X
Pgi C tkj tj k Dm D 0 (6)
i 2˛m j 2ˇm j 2ˇm

Pgi.min/  Pgi  Pgi.max/ (7)

tj k.min/  tj k  tj k.max/ ; (8)

where
F .Pgi / is the fuel-cost function of generator Pgi ,
G.tj k / is the cost of transmission of line tj k ,
tj k is the economic flow on the tie line from area j to k,
˛m is the set of generating units in area m,
ˇm is the set of tie lines in area m,
Pgi is the active power generation of generator number i ,
Dm is the total load demand of area m,
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 253

PL is the total system transmission losses,

Pgi.min/ is the minimum limit on active power generation of generator i ,
Pgi.max/ is the maximum limit on active power generation of generator i ,
tj k.min/ is the minimum limit on active power generation of generator i , and
tj k.max/ is the maximum limit on active power generation of generator i .

2.3. Mathematical Form of CEED

The primary objective of ED can be extended to take into consideration the environmental
impact of power generation due to the emission of different pollutants that could inflict
harm to the environment [15, 16]. The emission-cost function, just like the fuel-cost
function in traditional ED, is modeled as a second-order polynomial

N
X
2
ED di Pgi C hi Pgi C ui (kg/h) (9)
i D1

where di , hi , and ui are emission coefficients of generator i . Thus, the overall objective
function for the combined emission–economic dispatch is
 
.1 w/
min F C E (10)
w

where w is a weighting factor, w 2 Œ0; : : : ; 1, which represents the relative importance
or the trade-off between the fuel-cost function F and the emission-cost function E. The
objective function of Eq. (10) is minimized, subject to the constraints given by Eqs. (3)
and (4), where the transmission losses are given by

N X
X N
PL D Pgn Bn;k Pgk (11)
nD1 kD1

where B is a matrix of loss coefficients and Pgn and Pgk are the active power generation
of generators n and k, respectively.

2.4. Mathematical Form of QCFED

Traditionally, the cost function of the generating units is approximated as a quadratic
function. A crucial issue in ED studies is to determine the order and approximate the
coefficients of the polynomial used to model the fuel-cost function [17]. This issue is
particularly important in terms of reducing the error between the approximated polyno-
mial, along with its coefficients, and the actual operating cost. According to [17, 18], a
third-order (cubic) polynomial is realistic to model the operating cost. For a generating
unit with non-monotonically increasing cost curves, a cubic polynomial is usually used
to obtain accurate dispatch results [19]. The third-order polynomial function is expressed
as
3 2
Fi .Pgi / D ˛i;1 Pgi C ˛i;2 Pgi C ˛i;3 Pgi C ˛i;4 (12)
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

254 J. S. Al-Sumait et al.

subject to

N
X
Pgi D PD C PL (13)
i D1

Pgi.min/ < Pgi < Pgi.max/ ; i 2 Ns ; (14)

where ˛i;1 , ˛i;2 , ˛i;3 , and ˛i;4 are the coefficients of the cubic fuel-cost function.

System Losses Representation. It important to mention at this stage that the B-coefficients,
or loss coefficients, have been adopted for the modeling of the system losses in the
previous formulations. The representation of the system losses by B-coefficients is indeed
constant and suitable for actual interpretation of the real power system losses under
certain conditions. If the actual operating conditions are close to the base case where
B-constants were computed, then the B-coefficients method should compute the system
losses with reasonably high accuracy [20, 21]. In other words, the use of constant values
for the loss coefficients in the transmission losses equation yields practically precise
results when the coefficients are calculated for some average operating condition and if
extremely wide shifts of load between plants or in total load do not occur. In practice,
large systems are economically loaded by calculations based on just one set of loss
coefficients that are sufficiently accurate throughout the daily variations of load on the
system [21].

3. Overview of the PS Method

The PS optimization routine is an evolutionary technique that is suitable to solve a
variety of optimization problems that lie outside the scope of the standard optimization
methods. Generally, PS has the advantage of being very simple in concept, being easy to
implement, and resulting in a computationally efficient algorithm. Unlike other heuristic
algorithms, such as genetic algorithms (GA) [22, 23], PS possesses a flexible and well-
balanced operator to enhance and adapt the global and fine tune the local search. A
historic discussion of DS methods for unconstrained optimization is presented in [9],
where a modern perspective on the classical family of derivative-free algorithms is given
with focus on the development of DS methods.
The PS algorithm proceeds by computing a sequence of points that may or may not
approach the optimal point. The algorithm starts by establishing a set of points called a
mesh around the given point. This current point could be the initial starting point supplied
by the user, or it could be computed from the previous step of the algorithm. The mesh
is formed by adding the current point to a scalar multiple of a set of vectors called a
pattern. If a point in the mesh is found to improve the objective function at the current
point, the new point becomes the current point at the next iteration.
This may be better explained by the following:

First: The PS begins at the initial point X0 that is given as a starting point by the user.
At the first iteration, with a scalar D 1 called the mesh size, the pattern vectors are
constructed as Œ0 1, Œ1 0, Œ 1 0, and Œ0 1; they may be called direction vectors.
Then the PS algorithm adds the direction vectors to the initial point X0 to compute the
following mesh points:
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 255

X0 C Œ1 0I X0 C Œ0 1

X0 C Œ 1 0I X0 C Œ0 1

Figure 1 illustrates the formation of the mesh and pattern vectors. The algorithm computes
the objective function at the mesh points in the order shown.
The algorithm polls the mesh points by computing their objective function values
until it finds one whose value is smaller than the objective function value of X0 . If
there is such a point, then the poll is successful and the algorithm sets this point equal
to X1 .
After a successful poll, the algorithm steps to iteration 2 and multiplies the current
mesh size by 2. (This is called the expansion factor and has a default value of 2.)
The mesh at iteration 2 contains the following points: 2  Œ1 0 C X1 , 2  Œ0 1 C X1 ,
2  Œ 1 0 C X1 , and 2  Œ0 1 C X1 . The algorithm polls the mesh points until it finds
one whose value is smaller the objective function value of X1 . The first such point it
finds is called X2 , and the poll is successful. Because the poll is successful, the algorithm
multiplies the current mesh size by 2 to get a mesh size of 4 at the third iteration because
the expansion factor D 2.

Second: If in iteration 3 (mesh size D 4) none of the mesh points have a smaller objective
function value than the value at X2 , the poll is declared unsuccessful. In this case the
algorithm does not change the current point at the next iteration, which is X3 D X2 . At
the next iteration, the algorithm multiplies the current mesh size by 0.5, a contraction
factor, so that the mesh size at the next iteration is smaller. The algorithm then polls
with a smaller mesh size.
The PS optimization algorithm will repeat the illustrated steps until it finds the
optimal solution for the minimization of the objective function. The algorithm stops
when any of the following conditions occurs:
 the mesh size is less than mesh tolerance,
 the number of iterations performed by the algorithm reaches a predefined maxi-
mum,

Figure 1. PS mesh points and the pattern.

Downloaded By: [University of Southampton] At: 12:17 22 February 2008

256 J. S. Al-Sumait et al.

 the total number of objective function evaluations performed by the algorithm

reaches a predefined maximum,
 the distance between the point found at one successful poll and the point found
at the next successful poll is less than a predefined tolerance, or
 the change in the objective function from one successful poll to the next successful
poll is less than a predefined function tolerance.

Constraint Handling
Many ideas were suggested to insure that the solution will satisfy the constraints [24].
Currently, the only minor weakness in the nonlinear constraints handling procedure
is due to the lack of specific information related to first-order derivatives. Although
many systematic approaches have been employed to handle the nonlinear constraints, the
augmented Lagrangian approach, which has been utilized by PS method, has overcome
such shortcomings in dealing with nonlinear constraints. Lewis and Torczon [25] stated
that, despite the absence of an explicit estimation of any derivatives (a characteristic of PS
methods), PS-augmented Lagrangian approach exhibits all of the first-order convergence
properties of the original algorithm advocated by Conn, Gould, and Toint [26, 27]. Lewis
and Torczon [25] were able to overcome such drawbacks by proceeding with successive,
inexact minimization of the augmented Lagrangian via PS methods, even without knowing
exactly how inexact the minimization is. As a result, the size of the problem will increase
by introducing new parameters.
The augmented Lagrangian pattern search (ALPS) has been used to solve nonlinear
constraint problems in PS algorithm. ALPS proceeds to solve a nonlinear optimization
problem with nonlinear constraints, linear constraints, and bounds [25, 28–30]. The
variables bounds and linear constraints are handled separately from nonlinear constraints,
in which a sub-problem is constructed and solved (having the objective function and
nonlinear constraint function) using the Lagrangian and the penalty factors. Such a sub-
problem is minimized using a PS method, where the linear constraints and bounds are
satisfied. ALPS starts with an initial value for the penalty parameter, where the PS
algorithm minimizes a series of the subproblem, which estimates the original problem. If
the required accuracy and feasibility conditions are met, then the Lagrangian estimates
are updated. If not, a penalty factor is added to the penalty parameter. This, in turns,
leads to a new formation of a subproblem and ultimately results in a new minimization
problem. The above steps are repeated until one of the stopping criteria is reached. For
further explanation on how PS handles constraints, refer to [25–27, 31].

4. Numerical Results
A set of Matlab files, incorporated in the GA and DS toolbox, implementing the proposed
PS method, have been used to solve various ED problems. Thus, cost coefficients of the
fuel cost and the combined objective function for all test cases have been coded in Matlab
environment.
Initially, several runs have been carried out with different values of the key parameters
of PS, such as the initial mesh size and the mesh expansion and contraction factors. In
this study, the mesh size and the mesh expansion and contraction factors are selected as
1, 2, and 0.5, respectively. In addition, a vector of initial points, i.e., X0 , was randomly
generated (each initial point is bounded within the generators limits) to provide an initial
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 257

guess for the PS to proceed. As for the stopping criteria, all tolerances were set to 10 6 ,
and the maximum number of iterations and function evaluations set to 1000.

Case I: EDVP
This test case consists of 13 generating units with a quadratic cost function combined
with the effects of valve-point loading. The unit’s data (upper and lower bounds) along
with the cost coefficients for the fuel cost (a, b, c, e, and f ) for the 13 generators with
valve-point loading are given in [32, 33]. Note that losses are ignored in this case.
The PS algorithm has been executed 50 times with different starting points to study
its performance and effectiveness. The solution of the PS method and the execution times
for the 50 runs were compared with the outcome of other evolutionary methods applied
to the same test system as reported in [33]. This numerical experiment compares the
performance of the PS with the other methods in terms of the dispatching costs and
convergence speed. Table 1 shows the optimal solutions determined by PS for the 13
generators, while the execution time and cost comparison are shown in Table 2.
As can be seen from Table 2, the optimum solution of the PS is better than the solu-
tions of all the other evolutionary methods. Moreover, the execution time is significantly
shorter than for the other methods [33].
The convergence of the PS process is shown in Figure 2, where only 70 iterations
were needed to locate the optimal solution. However, PS may be allowed to continue
the search in the neighborhood of the optimal solution to increase the confidence in the
result. In this case, the PS terminates after 52 further iterations (a total of 122).
Figure 3 illustrates the mesh size throughout the convergence process. It is apparent
that the mesh size decreases until the algorithm terminates, in this case at a mesh size

Table 1
Generator loading and fuel cost determined by
PS with total load demand of 1800 MW

Generator production
Generator (MW)

Pg1 538.5587
Pg2 224.6416
Pg3 149.8468
Pg4 109.8666
Pg5 109.8666
Pg6 109.8666
Pg7 109.8666
Pg8 109.8666
Pg9 109.8666
Pg10 77.4666
Pg11 40.2166
Pg12 55.0347
Pg
P13 55.0347
Pgi D 1800 MW Total cost: $17,969.17
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

258 J. S. Al-Sumait et al.

Table 2
Comparison of PS, GA, and EP

Evolution Mean time Best time Mean cost Maximum cost Minimum cost
method (s) (s) ($) ($) ($)

CEP 294.96 293.41 18,190.32 18,404.04 18,048.21

FEP 168.11 166.43 18,200.79 18,453.82 18,018.00
MFEP 317.12 315.98 18,192.00 18,416.89 18,028.09
IFEP 157.43 156.81 18,127.06 18,267.42 17,994.07
PS 5.88 1.65 18,088.84 18,233.52 17,969.17
CEP: Classical evolutionary programming; FEP: fast evolutionary programming; MFEP: modi-
fied fast evolutionary programming; IFEP: improved fast evolutionary programming.

of 1.5259e–005, which is more than the stopping criteria, indicating that this particular
run did not terminate on the mesh size tolerance. Figure 3 shows that the polling was
initially successful, leading to enlarging the mesh size. At iteration number 8, the mesh
size has decreased, indicating an unsuccessful poll in the previous iteration. This scenario
continues throughout until reaching one of the termination criteria.
The PS has determined the optimal solution for 50 different initial points. The
solutions of each run are shown in Figure 4. The total execution time for the 50 runs

Figure 2. Convergence of PS for the 13 generating units.

Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 259

Figure 3. Convergence of PS mesh size for the 13 generating units.

was 294.06 s. For this test case, looking at the maximum, mean, and minimum costs in
Table 2, it may be clearly seen that PS outperforms GA and evolutionary programming
(EP) by reaching a lower value of the maximum and mean costs for 50 different runs.

Case II: MAED

The MAED problem considered consists of four areas with tie lines connecting these
areas. Each area contains four generation units. Note that quadratic cost functions are
used to model the cost of generation F .Pgi /, but the tie line transmission costs, G.tj k /,
are assumed to be linear functions of the power transfer. The generators’ data and the tie
lines’ coefficients, along with their limits, are all given in [14].
Different heuristic methods, such as particle swarm optimization (PSO) and EP, have
been applied to the same problem, and the results are reported in [34, 35]. The results
using the PS and other methods are shown in Tables 3 and 4. The optimal solution
obtained by PS is obviously better than those obtained by various EP algorithms and is
similar to the result obtained by network flow programming (NFP) [14], which is not
heuristic in nature. In addition, the computation time of PS is less than the execution
times of the variants of EP.
As illustrated in Figure 5, PS has located the optimal solution after only 200
iterations, but it continues computating and refining the result. Figure 6 shows the mesh
size expansion and contraction behavior during the PS search for the global minimum.
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

260 J. S. Al-Sumait et al.

Figure 4. Objective function value for 50 different starting points.

Table 3
Comparison of PS and other heuristic methods (16 generators)

Optimization techniques

Generator IFEP FEP CEP NFP PS

Area 1 P1 (MW) 149.998 149.997 150.000 150.000 150.000
Demand P2 (MW) 099.986 099.968 100.000 100.000 100.000
400 MW P3 (MW) 068.270 067.017 068.826 066.970 066.971
P4 (MW) 099.940 099.774 099.985 100.000 100.000
Area 2 P5 (MW) 056.349 057.181 056.373 056.970 56.9718
Demand P6 (MW) 096.753 095.554 093.519 096.250 96.2518
200 MW P7 (MW) 041.264 041.736 042.546 041.870 41.8718
P8 (MW) 072.586 072.748 072.647 072.520 72.5218
Area 3 P9 (MW) 050.003 050.030 050.000 050.000 050.002
Demand P10 (MW) 035.985 036.552 036.399 036.270 036.272
350 MW P11 (MW) 038.012 038.413 038.323 038.490 038.492
P12 (MW) 037.426 037.001 036.903 037.320 037.322
Area 4 P13 (MW) 149.988 149.986 150.000 150.000 150.000
Demand P14 (MW) 099.964 099.995 100.000 100.000 100.000
300 MW P15 (MW) 057.601 057.568 056.648 057.050 057.051
P16 (MW) 095.874 096.482 095.826 096.270 096.271
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 261

Table 4
Comparison of PS and other heuristic methods (tie lines)

Area Tie lines values (MW)

From To IFEP FEP CEP NFP PS

1 2 00.094 00.062 00.000 00.000 00.000

1 3 18.649 18.241 19.587 18.180 18.181
1 4 00.000 00.000 00.000 00.000 00.000
2 1 00.018 00.000 00.018 00.000 00.000
2 3 69.997 69.790 68.861 69.730 69.730
2 4 00.000 00.000 00.000 00.000 00.000
3 1 00.000 00.000 00.000 00.000 00.000
3 2 00.000 00.000 00.000 00.000 00.000
3 4 00.000 00.000 00.000 00.000 00.000
4 1 00.549 01.548 00.758 01.210 01.210
4 2 02.951 02.509 01.789 02.110 02.111
4 3 99.927 99.974 99.927 100.00 100.00
Total cost ($/h) 7337.51 7337.52 7337.75 7337.0 7336.98
Computation time (s) 23.97 7.47 7.82/11.49 — 5.77
Population size 100 100 100 — —
No. of iterations 585 645 758/920 — 1225

Figure 5. Convergence of PS for MAED.

Downloaded By: [University of Southampton] At: 12:17 22 February 2008

262 J. S. Al-Sumait et al.

Figure 6. Mesh size.

Case III: CEED

In this combined environmental ED case, a six-generator system is considered. Infor-
mation about the generators’ fuel cost, NOx emission functions, the B matrix, loss
coefficients, and the operating limits are detailed in [36]. The total load demand is set to
700 MW, and the weighting factor is 0.5.
The PS results of the line losses, emission, fuel cost, total cost, and computation
time are presented and compared with results of other heuristic methods (GA and EP
from [34, 35]) in Table 5. It can be seen that PS has reached the best total and fuel costs,
and also has produced the best time of computation compared with the other methods. In
addition, PS came third in line losses and emissions. The convergence of the PS needs
only 40 iterations and 2.05 s to reach the optimal solution, which are significantly less
for EP and GA.

Case IV: QCFED

According to [17], it is an industry practice to adopt a cubic polynomial for modeling
fuel costs of generation units. This is particularly important in situations with generation
units having non-monotonically increasing incremental curves [19]. In this case, a three-
generator system is considered with third-order cost functions. Information about the
generators’ fuel-cost coefficients, the B matrix, loss coefficients, and the operating limits
are detailed in [18].
The optimal solution of PS is given in Table 6, along with the results obtained by
conventional DP from [18] for comparison purposes. Clearly, the PS has converged to a
better solution, while the execution time is less than 1 s. Also, the significant reduction
in line losses is obvious.
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 263

Table 5
Comparison of PS and other heuristic methods

Optimization techniques

Generator IFEP FEP CEP FCGA PS

P1 (MW) 077.142 077.358 077.274 080.16 77.4318

P2 (MW) 049.925 049.669 049.639 053.71 048.894
P3 (MW) 048.764 048.316 048.535 040.93 048.516
P4 (MW) 103.486 104.369 103.525 116.23 104.5679
P5 (MW) 259.805 260.663 260.695 251.20 260.8632
P6 (MW) 191.828 190.473 191.233 190.62 190.6723
Line losses (MW) 30.949 30.849 30.901 32.850 30.945
Emission (kg/h) 530.5164 532.5046 524.49 527.46 528.33
Fuel cost ($/h) 38,214.02 38,214.23 38,216.47 38,408.82 38,208.63
Total cost ($/h) 19,369.84 19,369.89 19,369.84 19,468.14 19,368.48
Computation time (s) 3.874 1.598 4.48 — 2.05
No. of iterations 57 65 77 — 40

FCGA: Fuzzy controlled genetic algorithm.

Table 6
Comparison of PS and DP with total demand
1400 MW

Optimization technique

Generator DP PS

P1 (MW) 360.2 372.29

P2 (MW) 406.4 356.0
P3 (MW) 676.8 712.0
Line losses (MW) 43.4 40.29
Fuel cost ($/h) 6642.26 6639.01
Computation time (s) — 0.619
No. of iterations 5 6

5. Conclusions
This paper introduces a new approach based on PS optimization to solve various prob-
lems of power system ELD. The proposed method has been applied to four types of
problems, including the EDVP effects, MAED, CEED, and QCFED. Through extensive
comparisons, it has been demonstrated that the PS approach outperforms other heuristic
methods in terms of reaching a better optimal solution; at the same time, the simplicity of
the PS process makes the algorithm computationally more efficient. However, the PS is
more sensitive than the GA or EP to the initial guess (starting point) and appears to rely
more heavily on how close the given initial point is to the expected optimum. This, in
turn, makes the PS method more susceptible to getting trapped in local minima. Overall,
the PS approach has been found to be a very efficient technique to study a wide range
of optimization problem in the area of power system.
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

264 J. S. Al-Sumait et al.

Acknowledgment
This work has been supported by the government of The State of Kuwait, Public Authority
of Applied Science & Training, Kuwait (Project # TS-06-02).

References
1. Adler, R. B., and Fischl, R., “Security constrained economic dispatch with participation factors
based on worst case bus load variations,” IEEE Transactions on: Power Apparatus and Systems,
Vol. 96, pp. 347–356, 1977.
2. Bui, R. T., and Ghaderpanah, S., “Real power rescheduling and security assessment,” IEEE
Trans. Power Apparatus Syst., Vol. PAS-101, pp. 2906–2915, 1982.
3. Pancholi, R. K., and Swarup, K. S., “Particle swarm optimization for security constrained eco-
nomic dispatch,” International Conference on Intelligent Sensing and Information Processing,
Chennai, India, pp. 7–12, 2004.
4. El-Sharkawy, M., and Nieebur, D., “Artificial neural networks with application to power
systems,” IEEE Power Engineering Society, A Tutorial Course, 1996.
5. Yalcinoz, T., and Short, M. J., “Neural networks approach for solving economic dispatch
problem with transmission capacity constraints,” IEEE Trans. Power Syst., Vol. 13, pp. 307–
313, 1998.
6. Youssef, H. K., and El-Naggar, K. M., “Genetic based algorithm for security constrained power
system economic dispatch,” Elec. Power Syst. Res., Vol. 53, pp. 47–51, 2000.
7. Chen, C. L., and Chen, N., “Direct search method for solving economic dispatch problem
considering transmission capacity constraints,” IEEE Trans. Power Syst., Vol. 16, pp. 764–
769, 2001.
8. Hooke, R., and Jeeves, T. A., “Direct search solution of numerical and statistical problems,”
J. Assoc. Computing Mach., Vol. 8, pp. 212–229, 1961.
9. Lewis, R. M., Torczon, V., and Trosset, M. W., “Direct search methods: Then and now,” J.
Computat. Appl. Math., Vol. 124, pp. 191–207, 2000.
10. Deb, K., Optimization for Engineering Design: Algorithms and Examples, New Delhi: Prentice
Hall, 1995.
11. Walters, D. C., and Sheble, G. B., “Genetic algorithm solution of economic dispatch with
valve point loading,” IEEE Trans. Power Syst., Vol. 8, pp. 1325–1332, 1993.
12. Victoire, T. A. A., and Jeyakumar, A. E., “Hybrid PSO-SQP for economic dispatch with
valve-point effect,” Elec. Power Syst. Res., Vol. 71, pp. 51–59, 2004.
13. Victoire, T. A. A., and Jeyakumar, A. E., “A modified hybrid EP-SQP approach for dynamic
dispatch with valve-point effect,” Int. J. Elec. Power Energy Syst., Vol. 27, pp. 594–601, 2005.
14. Streiffert, D., “Multi-area economic dispatch with tie line constraints,” IEEE Trans. Power
Syst., Vol. 10, pp. 1946–1951, 1995.
15. Farag, A., Al-Baiyat, S., and Cheng, T. C., “Economic load dispatch multiobjective optimiza-
tion procedures using linear programming techniques,” IEEE Trans. Power Syst., Vol. 10,
pp. 731–738, 1995.
16. Song, Y. H., Wang, G. S., Wang, P. Y., and Johns, A. T., “Environmental/economic dispatch
using fuzzy logic controlled genetic algorithms,” IEE Proc. Generat. Transm. Distribut.,
Vol. 144, pp. 377–382, 1997.
17. Liang, Z. X., and Glover, J. D., “Improved cost functions for economic dispatch compensa-
tions,” IEEE Trans. Power Syst., Vol. 6, pp. 821–829, 1991.
18. Liang, Z. X., and Glover, J. D., “A zoom feature for a dynamic programming solution to
economic dispatch including transmission losses,” IEEE Trans. Power Syst., Vol. 7, pp. 544–
550, 1992.
19. Jiang, A., and Ertem, S., “Economic dispatch with non-monotonically increasing incremental
cost units and transmission system losses,” IEEE Trans. Power Syst., Vol. 10, pp. 891–897,
1995.
Downloaded By: [University of Southampton] At: 12:17 22 February 2008

Solution of Different Types of Economic Load Dispatch 265

20. Saadat, H., Power System Analysis, Boston, MA/London: WCB/McGraw-Hill, 1999.
21. Grainger, J. J., and Stevenson, W. D., Power System Analysis, New York: McGraw-Hill, 1994.
22. Goldberg, D. E., Genetic Algorithms in Search, Optimization, and Machine Learning, Reading,
MA: Addison-Wesley, 1989.
23. Michalewicz, Z., Genetic Algorithms C Data Structures D Evolution Programs, 3rd rev. and
extended ed., New York: Springer-Verlag, 1996.
24. Youssef, H. K., El-Shibini, M., and Hazaa, G. A., “Some new aspects in power system dynamic
security using pattern recognition,” Second Middle East Power Conference (MEPCON-92),
pp. 308–313, Egypt, January 1992.
25. Lewis, R. M., and Torczon, V., “A globally convergent augmented Lagrangian pattern search
algorithm for optimization with general constraints and simple bounds,” SIAM J. Optim.,
Vol. 12, pp. 1075–1089, 2001.
26. Conn, A. R., Gould, N. I. M., and Toint, P. L., “A globally convergent augmented Lagrangian
algorithm for optimization with general constraints and simple bounds,” SIAM J. Numer. Anal.,
Vol. 28, pp. 545–572, April 1991.
27. Conn, A. R., Gould, N. I. M., and Toint, P. L., “A globally convergent augmented Lagrangian
pattern search algorithm for optimization with general constraints and simple bounds,” Math.
Comput., Vol. 66, pp. 261–288, 1997.
28. Lewis, R. M., and Torczon, V., “Pattern search algorithms for linearly constrained minimiza-
tion,” SIAM J. Optim., Vol. 10, pp. 917–941, 1999.
29. Lewis, R. M., and Torczon, V., “Pattern search algorithms for bound constrained minimization,”
SIAM J. Optim., Vol. 9, pp. 1082–1099, 1999.
30. Torczon, V., “On the convergence of pattern search algorithms,” SIAM J. Optim., Vol. 7,
pp. 1–25, 1997.
31. The Math Works Inc., Genetic Algorithm and Direct Search Toolbox for Use with Matlab
User’s Guide, 2nd ed.
32. Wong, K. P., and Wong, Y. W., “Thermal generator scheduling using hybrid genetic simulated-
annealing,” IEE Proc. Generat. Transm. Distrib., Vol. 142, pp. 372–380, July 1995.
33. Sinha, N., Chakrabarti, R., and Chattopadhyay, P. K., “Evolutionary programming techniques
for economic load dispatch,” IEEE Trans. Evolut. Comput., Vol. 7, pp. 83–94, 2003.
34. Jayabarathi, T., Jayaprakash, K., Jeyakumar, D. N., and Raghunathan, T., “Evolutionary pro-
gramming techniques for different kinds of economic dispatch problems,” Elec. Power Syst.
Res., Vol. 73, pp. 169–176, 2005.
35. Jeyakumar, D. N., Jayabarathi, T., and Raghunathan, T., “Particle swarm optimization for
various types of economic dispatch problems,” Int. J. Elec. Power Energy Syst., Vol. 28,
pp. 36–42, 2006.
36. Dhillon, J. S., Parti, S. C., and Kothari, D. P., “Stochastic economic emission load dispatch,”
Elec. Power Syst. Res., Vol. 26, pp. 179–186, 1993.