International Journal of Emerging

Electric Power Systems

Volume 9, Issue 4 2008 Article 5

Effect of Control Parameters on Differential

Evolution based Combined Economic
Emission Dispatch with Valve-Point Loading
and Transmission Loss
Kamal K. Mandal∗ N. Chakraborty†

∗
Jadavpur University, kkm567@yahoo.co.in
†
Jadavpur University, chakraborty niladri2004@yahoo.co.in

Copyright 2008
c The Berkeley Electronic Press. All rights reserved.
 Effect of Control Parameters on Differential
Evolution based Combined Economic
Emission Dispatch with Valve-Point Loading
and Transmission Loss∗
Kamal K. Mandal and N. Chakraborty

Abstract

Differential evolution (DE) has been proved to be a powerful evolutionary algorithm for global
optimization in many engineering problems. The performance of this type of evolutionary algo-
rithms is heavily dependent on the setting of control parameters. Proper selection of the control
parameters is very important for the success of the algorithm. Optimal settings of control param-
eters of differential evolution depend on the specific problem under consideration. In this paper, a
study of control parameters on differential evolution based combined economic emission dispatch
with valve-point loading and transmission loss is conducted empirically. The problem is formu-
lated considering equality constraints on power balance and inequality constraints on generation
capacity limits as well as the transmission losses and effects of valve point loadings. The feasi-
bility of the proposed method is demonstrated on a fourteen-generator system. The results of the
effect of the variation of different parameters are presented systematically and it is observed that
the search algorithm may fail in finding the optimal value if the parameter selection is not done
with proper attention.

KEYWORDS: control parameters, differential evolution (DE), combined economic emission dis-
patch (CEED), price penalty factor

∗
We would like to acknowledge and thank Jadavpur University, Kolkata, India for providing all
the necessary help to carry out this work.
 Mandal and Chakraborty: Effect of Control Parameters

1 Introduction

The Differential Evolution (DE) as proposed by Storn and Price [1] is a powerful
optimization technique designed for global optimization. Besides its good
convergence properties, main advantage of DE lies with its conceptual simplicity;
ease of use and less number of control parameters. Like any other evolutionary
algorithm, the success of DE is heavily dependent on setting of control
parameters. It has three control parameters: (1) the population size N P (2) the
mutation factor f m , which is a real-valued factor that controls the amplification of
differential variation and (3) the crossover factor C R , which is also a real-valued
factor that controls the crossover operation. One of the main problems in
evolution strategies of DE is to choose the control parameters such that it exhibits
good behavior i.e. it does not prematurely converge to a point that is not globally
optimal or stagnate and has an acceptable rate of convergence toward the global
optimum. Premature convergence may occur under different situations: the
population has converged to local optimum of the objective function or the
population has lost its diversity or the search algorithm proceeds slowly or does
not proceed at all [2]. It is seen that DE may sometimes stops proceeding towards
global optimum and stagnation occurs. Stagnation may occur under various
situations: the population has not converge to a local optimum or any other point
or the population is still remaining diverse and occasionally even if the new
individuals may enter in to the population, the searching algorithm does not
progress towards any better solutions [2].
The effectiveness, efficiency and robustness of the DE algorithm are
sensitive to the setting of control parameters. The best setting for the control
parameters depends on the problem in hand and requirement of computation time
and accuracy [3]. Although, as reported in the literature, control parameters of
DE are not difficult to choose [1], but rules for choosing control parameters are
not general [1]. On the other hand, it is also reported that choosing proper control
parameters for DE is more difficult than expected [4]. It is important to select
optimal parameters for each problem separately and carefully to avoid premature
convergence or even stagnation [2], [5], [6]. Brest et al. [7] assessed the selection
of control parameter and reported that efficiency and robustness of DE algorithm
are much more sensitive to the setting of mutation factor f m and crossover
ratio C R than to the setting of the population size N P . Zaharie [8] discussed the
relationship between control parameters of DE and the evolution of population
variance and reported critical interval for control parameters of DE. Teo proposed
a method of self-adapting population size in addition to self-adapting mutation
factor and crossover ratio [9].

Published by The Berkeley Electronic Press, 2008 1

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

Setting of control parameters in evolutionary algorithms like DE can

be classified into two categories: Parameter tuning and parameter control.
Parameter tuning is the commonly practiced approach that is used in finding
proper values for the parameters before the actual run of the algorithm. Then the
algorithm is tested using these values, which remain fixed during the run.
Generally, electric power plants are operated on the basis of least fuel cost
strategies without considering the pollutants produced. However presently
environmental management has topped the list of utility management concern
[10] and it has become almost compulsory for electric utilities to reduce the
pollution level below certain limit. By proper load allocation among the various
generating units of the plants, the harmful effects of the emission of gaseous
pollutants from power stations, particularly from thermal power stations, can be
reduced. A revised power dispatch program is required in such cases that will
consider both the generation cost and emission cost as the objective functions.
One of the major complications in the above considerations is that the cost and
emission functions are conflicting to each other. In other words, minimizing
pollution increases cost and vice versa. Since the ideas of minimum cost and
minimum emission dispatch are conflicting to each other, therefore the problem of
choosing the least cost generation schedule with environmental objectives
becomes much more complicated. In this paper, tests for various parameter
settings of DE are conducted for the combined economic emission dispatch with
generator constraints, valve-point loading and transmission loss. The influence of
parameter settings of DE is tested on a fourteen-generator system and results are
presented. Based on the results obtained, recommendations are suggested for the
suitable range of control parameters of Differential Evolution for the present
problem.

2 Problem formulation
2.1 Economy

The pure Economic Load Dispatch (ELD) problem is one of the major problems
in power system operation and planning. The classical ELD problem may be
described by minimizing the total fuel cost of the generating units under several
operating constraints. The fuel cost curve for any unit is assumed to be
approximated by segments of quadratic functions of the active power output of
the generator. For a given power system network, the problem may be described
as optimization (minimization) of total fuel cost as defined by (1) under a set of
operating constraints;

http://www.bepress.com/ijeeps/vol9/iss4/art5 2
 Mandal and Chakraborty: Effect of Control Parameters

n
FC(Pg ) = ∑(ai Pi2 + bi Pi + ci ) (1)
i=1

where FC( Pg ) is the total fuel cost of generation in the system ($/hr), ai ,bi ,ci are
the fuel cost coefficients of the i th generating unit, Pi is power generated by the
i th unit and n is the number of thermal units. The coefficients a i , bi and ci are
generally obtained by curve fitting [11], [12].
However, for more practical and accurate modeling of fuel cost
function, the above expression is to be modified suitably. Modern thermal power
plants consist of generating units have multi-valve steam turbines in order to
incorporate flexible operational facilities. The generating units with multi-valve
turbines have very different cost curve compared with that defined by (1) and
exhibit a greater variation in the fuel cost curves. Typically, ripples are introduced
in to the fuel cost curve as each steam valve starts to operate. The valve-point
effect may be considered by adding a sinusoidal function [13], [14] to the
quadratic cost function described above. Hence, the problem described by (1) is
revised as follows:

( )
n
FCv (Pg ) = ∑ ai Pi2 + bi Pi + ci + | ei ×sin( f i × (Pi,min − Pi )) | (2)
i=1
where FCv ( Pg ) is the total fuel cost of generation ($/hr) including valve point
loading, ei , f i are the fuel cost coefficients of the i th generating unit reflecting
the valve-point effect.
The cost is minimized with the following generator capacities and
active power balance constraints as;

Pi, min ≤ Pi ≤ Pi, max (3)

n
∑ Pi = PD + PL (4)
i=1

where, Pi , min and Pi , max are the minimum and maximum power generation by i th
unit respectively, PD is the total power demand and PL is the total transmission
loss.
The transmission loss PL can be calculated by using B matrix
technique and is defined by (5) as

Published by The Berkeley Electronic Press, 2008 3

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

n n
PL = ∑∑ Pi Bij Pj (5)
i =1 j =1

where Bij ’s are the elements of loss coefficient matrix B.

Emission

The solution of ELD problem will give the amount of active power to be
generated by different units at a minimum fuel cost for a particular demand. But
the amount of emission or emission cost is not considered in pure ELD problem.
The amount of emission from a fossil-based thermal generator unit depends on the
amount of power generated by the unit. Total emission generated also can be
approximated as a quadratic function of the active power output of the generators.
The emission dispatch problem can be described as the optimization
(minimization) of total amount of emission release defined by (6) as

( )
n
EC(Pg ) = ∑ αi Pi2 + βi Pi + γ i (6)
i=1

where EC( Pg ) is total amount of emission (lb/hr), α i , β i , γ i are the emission

coefficients of the i th unit subjected to the generator capacity and power balance
constraints described by (3) and (4). α i , β i and γ i are generally obtained by
curve fitting [11], [12].

Combined Economic and Emission Dispatch

The economic dispatch and emission dispatch are two different problems.
Emission dispatch can be included in conventional economic load dispatch
problems by the addition of emission cost to the normal dispatch cost. In this
method different types of emissions are modeled as a cost in addition to the fuel
cost. The bi-objective problem of combined economic emission dispatch (CEED)
can be converted into single objective optimization problem by introducing a
price penalty factor h [15] as follows:

Minimize
φt = FCv (Pg ) + h ∗ EC(Pg ) (7)
where, φ t is the total operational cost of the system subject to the constraints

http://www.bepress.com/ijeeps/vol9/iss4/art5 4
 Mandal and Chakraborty: Effect of Control Parameters

defined by (3) and (4).

Now, for a trade off between fuel cost and emission cost, (7) can be revised as (8)
as

Minimize
φt = w1 ∗ FCv ( Pg ) + w2 * h ∗ EC(Pg ) (8)

subjected to the constraints defined by (3) and (4). The weight factors w1 and w2
have many implications. For w1 = 1 and w2 = 0 the solution will yield results for
pure economic dispatch. For w1 = 0 and w2 = 1 results for pure emission dispatch
and for w1 = w2 = 1 results for combined economic emission dispatch can be
obtained.
The price penalty factor h can be found out by a practical method as
discussed in [15]. The following steps can be used to find out the price penalty
factor for a particular load.
1) Find out the average cost of each generator at maximum power output.
2) Find out the average emission of each generator at its maximum output
3) Divide the average cost of each generator by its average emission and thus
hi is given as

FCv (Pi, max ) (Pi, max )

= hi i =1,........., n $ / lb (9)
EC(Pi, max ) (Pi, max )

4) Arrange the values of price penalty factor in ascending order.

5) Add the maximum capacity of each unit ( Pi , max ) one at a time starting
from the smallest hi unit until ∑ Pi , max ≥ PD is realized.
6) At this stage, hi associated with last unit in the process is the price penalty
factor h for the given load.
From the above description, it is clear that the value of the price
penalty factor h is dependent on the total power demand and hence it will have
different values for different power demand. It is also important to note that the
value of the price penalty factor h will be the same for ELD, EED and CEED as
long as the power demand is the same.
With all these functions, the main objective of this study is to find the
optimal values of control parameters for Differential Evolution (DE) based
combined economic emission dispatch with above said constraints including the
valve-point loading and transmission loss.

Published by The Berkeley Electronic Press, 2008 5

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

3 Differential Evolution (DE)

DE or Differential Evolution belongs to the class of evolutionary algorithms [1],

[16] that include Evolution Strategies (ES) and conventional Genetic Algorithms
(GA). DE differs from the conventional genetic algorithms in its use of perturbing
vectors, which are the difference between two randomly chosen vectors. DE is a
scheme by which it generates the trial vectors from a set of initial populations. In
each step, DE mutates vectors by adding weighted random vector differentials to
them. If the fitness of the trial vector is better than that of the target vector, the
target vector is replaced by the trial vector in the next generation.
DE offers several strategies for optimization. They are classified
according to the following notation such as DE/x/y/z, where x refers to the method
used for generating parent vector that will form the base for mutated vector, y
indicates the number of difference vector used in mutation process and z is the
crossover scheme used in the cross over operation to create the offspring
population [16], [17]. The symbol x can be ‘rand’ (randomly chosen vector) or
‘best’ (the best vector found so far). The symbol y, number of difference vector,
is normally set to be 1 or 2. For cross over operation, a binomial (notation: ‘bin’)
or exponential (notation: ‘exp’) operation is used.
The version used here is the DE/rand/1/bin, which is described by the
following steps.

3.1 Initialization

The optimization process in DE is carried with four basic operations:

initialization, mutation, crossover and selection. The algorithm starts by creating a
population vector P of size NP composed of individuals that evolve over G
generation. Each individual Xi is a vector that contains as many as elements as the
problem decision variable. The population size NP is an algorithm control
parameter selected by the user. Thus,

[
P (G) = X i(G) ,................................................., X N(G)
P
] (10)

[
X i(G) = X1(,Gi) ,........................X D(G, i) ]
T
, i = 1,..............N P (11)

The initial population is chosen randomly in order to cover the entire searching
region uniformly. A uniform probability distribution for all random variables in
assumed in the following as;

http://www.bepress.com/ijeeps/vol9/iss4/art5 6
 Mandal and Chakraborty: Effect of Control Parameters

X (j0,i) = X min (
max
j +σ j X j − X min
j ) (12)

where i = 1,....................N P and j = 1,....................D ;

D is the number of decision or control variables, X min
j and X max
j are the lower
and upper limits of the j the decision variables and σ j ∈ [ 0 ,1 ] is a uniformly
distributed random number generated anew for each value of j. X (j 0, i ) is the j th
parameter of the i th individual of the initial population.

3.2 Mutation Operation

Several strategies of mutation have been introduced in the literature of DE [16].

The essential ingredient in the mutation operation is the vector difference. The
mutation operator creates mutant vectors ( Vi ) by perturbing a randomly selected
vector ( X k ) with the difference of two other randomly selected vectors
( X l and X m ) according to;

(
Vi(G) = X (Gk) + f m X l(G) − X m(G) ) (13)

where X k , X l and X m are randomly chosen vectors ∈ [ 1 ,......................, N P ]

and k ≠ l ≠ m ≠ i . In other words, the indices are mutually different and also from
the running index i. The mutation factor f m that lies within [0, 2] is a user
chosen parameter used to control the perturbation size in the mutation operator
and to avoid search stagnation.

3.3 Crossover Operation

In order to extend further diversity in the searching process, crossover operation is

performed. The crossover operation generates trial vectors ( U i ) by mixing the
parameter of the mutant vectors with the target vectors. For each mutant vector,
an index q ∈ [ 1,.................N P ] is chosen randomly using a uniform distribution
and trial vectors are generated according to;

Published by The Berkeley Electronic Press, 2008 7

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

V (G) , if η ≤ C or j = q
 j, i j R
(G) 
U j, i =  (14)
 (G) , otherwise
 X j, i

where i = 1,.................., N P and j = 1,................., D ; η j is a uniformly

distributed random number within [0, 1] generated anew for each value of j. The
crossover factor C R ∈ [ 0, 1] is a user chosen parameter that controls the diversity
of the population. X (j G,i ) , V j(,Gi ) and U (j G,i ) are the j th parameter of the i th target
vector, mutant vector and trial vector at G generation respectively.

3.4 Selection Operation

Selection is the operation through which better offspring are generated. The
evaluation (fitness) function of an offspring is compared to that of its parent. The
parent is replaced by its offspring if the fitness of the offspring is better than that
of its parent, while the parent is retained in the next generation if the fitness of the
offspring is worse than that of its parent. Thus, if f denotes the cost (fitness)
function under optimization (minimization), then

( ) ( )
Ui(G) , if f Ui(G) ≤ f X i(G)

X i(G+1) = (15)
 X (G) , otherwise
 i

The optimization process is repeated for several generations. This allows

individuals to improve their fitness while exploring the solution space for optimal
values. The iterative process of mutation, crossover and selection on the
population will continue until a user-specified stopping criterion, normally, the
maximum number of generations allowed, is met. The other type of stopping
criterion, convergence to the global optimum, is possible if the global optimum of
the problem is available.

3.5 Constraint Handling

The success of any optimization method, such as DE, lies in the fact that how
does it handle the constraints relating to the problem. Most of the evolutionary
algorithms were originally developed to solve unconstrained optimization

http://www.bepress.com/ijeeps/vol9/iss4/art5 8
 Mandal and Chakraborty: Effect of Control Parameters

problems. However, over the last few decades, several methods have been
proposed to handle constraints in evolutionary algorithms. Michalewicz et al
presented a complete review of constrained optimization problems in evolutionary
algorithms [18] and these methods of constraints handling can be grouped into
four categories: methods based on preserving feasibility of solutions, methods
based on penalty functions, methods which make a clear distinction between
feasible and unfeasible solution and hybrid methods.
The present work is based on strategy to generate and keep control
variable in the feasible region [19]. This can be achieved as follows:


 min
 X j, i if X (jG,i ) ≤ X min
j, i


X (jG,i ) =  (16)
 X j, i
max
if X (jG )
, i ≥ X j,i
max


 (G)
 X j,i otherwise

where, i = 1,...................., N P and j = 1,..................., D .

In other words, when the generated value is less than or equal to the minimum
generating limit, it is set at the minimum. On the other hand, if the generated
value is greater than or equal to the maximum generating limit, it is set at the
maximum; otherwise the generated value is retained.

4 Results of Parameter Study

Combined economic emission dispatch (CEED) problems considering generator

constraints, valve point effects and transmission loss have been used here for
studying the effect of different parameter settings of differential evolution (DE).
The test system considered in this paper consists of fourteen generators. The cost
coefficients, generation limits and emission coefficients, valve point coefficients
are derived from [20], [21] with modifications to incorporate valve-point loading
and transmission losses. Those are shown in Table 1. The problem is solved as a
combined economic emission dispatch one with w1 = 1 and w2 = 1 . Transmission
loss is calculated using B-loss coefficient matrix according to (5). The main
control parameters considered here are the mutation factor f m , the crossover ratio
C R and the population size N P .

Published by The Berkeley Electronic Press, 2008 9

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

Table 1
Unit Data for Fourteen-Generator System with Valve Point Loading; a, b, c, e and
f are Fuel cost coefficients. α, β and γ are Emission coefficients
Unit Pmin Pmax a b c e f α β γ
1 150 455 0.0050 1.89 150 300 0.035 0.016 -1.500 23.333
2 150 455 0.0055 2.00 115 200 0.042 0.031 -1.820 21.022
3 20 130 0.0060 3.50 40 200 0.042 0.013 -1.249 22.050
4 20 130 0.0050 3.15 122 150 0.063 0.012 -1.355 22.983
5 150 470 0.0050 3.05 125 150 0.063 0.020 -1.900 21.313
6 135 460 0.0070 2.75 120 150 0..063 0.007 0.805 21.900
7 135 465 0.0070 3.45 70 150 0.063 0.015 -1.400 23.001
8 60 300 0.0070 3.45 70 150 0.063 0.018 -1.800 24.003
9 25 162 0.0050 2.45 130 150 0.063 0.019 -2.000 25.121
10 25 160 0.0050 2.45 130 100 0.084 0.012 -1.360 22.990
11 20 80 0.0055 2.35 135 100 0.084 0.033 -2.100 27.010
12 20 80 0.0045 1.60 200 100 0.084 0.018 -1.800 25.101
13 25 85 0.0070 3.45 70 100 0.084 0.018 -1.810 24.313
14 15 55 0.0060 3.89 45 100 0.084 0.030 -1.921 27.119

The study on the effect of different parameter settings of differential

evolution (DE) is performed for a demand of 2000 MW.

4.1 Effect of varying Mutation factor - f m

To check the effect of the mutation factor f m , it is varied from 0.95 to 0.05 in
steps of 0.05. The crossover factor C R and the population size are set at 0.90 and
100 respectively following the recommendation from the literature [1], [4], [22],
[23]. C R is set to a relatively higher value in order to have higher diversity in the
population. It means that on an average 90% of the elements of the trial vectors
are identical to the mutant vector that implies a high diversity for the present
setting of C R . One hundred (100) independent runs are performed for every
parameter combinations. Maximum number of iteration is set at 300. Table 2
shows the minimum cost, maximum cost, average cost, minimum emission,
maximum emission, average emission and the average computation time. It is
clearly seen that from Table 2 that both cost and emission increases for lower
values as well as higher values of f m . For example, with f m = 0.1, the minimum
cost and the minimum emission are found to be $10037.00 and 3738.30 lbs
respectively, while the corresponding values with f m = 0.90 are $9740.00 and
3311.20 lbs respectively. Several tests were performed with higher number of
iterations (up to 1000 in step of 100) for lower values of f m (≤ 0.15), but no
improvement in results were observed. It seems that the vector could not reach the

http://www.bepress.com/ijeeps/vol9/iss4/art5 10
 Mandal and Chakraborty: Effect of Control Parameters

minimum value and got stuck somewhere on their way to the minimum with
lower values and higher values of f m . This may be due to the fact that during the
searching process the difference vector for the perturbation has decreased for too
low values of f m . The values in the range of 0.45 ≤ f m ≤ 0.75 yields better results
in terms of cost, emission and computation time. Also, it is observed that for
higher values of f m (≥ 0.75), computation time increases to some extent.
Optimum result is obtained with f m = 0.50, when minimum cost, minimum
emission and computation time are found to be $ 9592.40, 3232.20 lbs and
16.2781 seconds respectively.

Table 2
Effect of mutation factor f m on Combined Economic Emission Dispatch (CEED)
with N P = 100 and C R = 0.90 for a demand of 2000 MW.
Value Fuel Cost ($) Emission (lb) Average
of fm CPU Time
(Sec.)
Minimum Maximum Average Minimum Maximum Average
0.95 9701.20 10917.00 10007.00 3310.10 4117.90 3436.20 37.2875
0.90 9740.00 10215.00 9903.10 3311.20 3566.20 3405.30 35.2766
0.85 9693.60 1.0066.00 9869.10 3399.00 3872.40 3401.80 34.6266
0.80 9697.10 9994.20 9817.60 3380.10 3556.80 3416.60 31.7796
0.75 9648.70 9840.20 9756.30 3385.10 3439.90 3394.10 25.7937
0.70 9628.70 9872.90 9720.30 3287.90 3341.00 3295.40 23.3687
0.65 9598.10 9817.80 9714.20 3285.70 3338.60 3219.30 20.0781
0.60 9589.70 9856.40 9691.30 3285.50 3693.10 3398.20 18.9016
0.55 9596.50 9995.50 9727.70 3284.60 3534.90 3249.60 17.0265
0.50 9592.40 9845.90 9644.80 3232.20 3430.80 3351.50 16.2781
0.45 9615.50 10142.00 9840.10 3252.20 3700.50 3376.20 15.3891
0.40 9684.80 10090.00 9860.30 3425.70 3839.50 3618.60 14.6078
0.35 9764.50 10153.00 9998.10 3432.50 4632.10 3801.70 14.7187
0.30 9877.10 10659.00 10208.00 3405.70 4039.10 3804.50 14.4813
0.25 9935.50 10604.00 10218.00 3644.30 4422.80 4040.30 14.6094
0.20 9954.00 10452.00 10187.00 3777.90 4430.20 4159.60 14.4594
0.15 10137.00 10905.00 10562.00 3686.20 4692.50 4156.30 14.0500
0.10 10037.00 10962.00 10550.00 3738.30 4859.80 4213.00 14.1531
0.05 10159.00 11122.00 10603.00 3955.90 4584.90 4339.80 14.3125

Thus, it is seen that for the present problem, the value of f m should not be
smaller than a certain value (0.45) in order to find the minimum value. It is
observed that smaller value of f m increases the chance of not finding the
minimum at all.

Published by The Berkeley Electronic Press, 2008 11

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

4.2 Effect of varying Crossover factor - C R

To find the influence of the crossover factor C R , it is also varied from 0.95 to 0.05
in steps of 0.05. The mutation factor f m and the population size are set at 0.50
and 100 respectively following the result obtained above and recommendation
from the literature. In this case also, one hundred (100) independent runs are
performed for every parameter combinations and maximum number of iteration is
set at 300. The results for different combination of parameter are shown in Table
3 in terms of the minimum cost, maximum cost, average cost and the average
computation time.

Table 3
Effect of crossover ratio C R on Combined Economic Emission Dispatch (CEED)
with N P =100 and f m = 0.50 for a demand of 2000 MW.
Value Fuel Cost ($) Emission (lb) Average
of C R CPU
Time
(Sec.)
Minimum Maximum Average Minimum Maximum Average
0.95 9618.40 9783.80 9684.90 3262.90 3533.90 3310.30 16.2984
0.90 9592.40 9845.90 9644.80 3232.20 3430.80 3351.50 16.2781
0.85 9598.30 9821.80 9690.80 3287.10 3446.30 3392.50 16.2781
0.80 9596.90 10071.00 9770.10 3255.00 3911.90 3365.10 16.0765
0.75 9684.90 10188.00 9736.20 3261.05 3572.50 3358.40 16.1297
0.70 9614.80 10503.00 9792.40 3283.90 3664.50 3313.30 16.6781
0.65 9603.40 98710.00 9723.50 3289.90 3340.20 3366.80 16.4156
0.60 9639.10 10059.00 9791.00 3286.50 3887.70 3381.80 16.4890
0.55 9672.50 10058.00 9725.80 3291.50 3642.40 3315.40 16.5687
0.50 9677.70 10233.00 9844.20 3325.90 3816.80 3393.10 17.9094
0.45 9690.00 10235.00 9769.10 3382.90 3518.30 3318.70 16.9797
0.40 9615.70 10153.00 9812.60 3305.80 3580.60 3307.30 17.6750
0.35 9694.40 10136.00 9793.50 3293.90 3815.10 3341.50 17.6640
0.30 9660.50 10135.00 9833.60 3268.00 3768.90 3356.60 20.4797
0.25 9803.10 10320.00 9976.80 3288.60 3900.70 3361.30 18.7765
0.20 9740.40 10268.00 9952.80 3370.20 3452.40 3381.90 19.0890
0.15 9888.40 10528.00 10132.00 3350.00 3773.60 3414.10 23.6359
0.10 10028.00 10936.00 10217.00 3389.40 3898.30 3561.00 28.0437
0.05 10139.00 10492.00 10339.00 3442.40 3940.30 3642.90 28.4640

From the Table 3, it is seen that for lower values of C R , the minimum cost,
minimum emission and the computation time increase. For example, with C R =
0.1, the minimum cost, the minimum emission and the average computation time
are found to be $10028.00, 3389.40 lbs and 28.0437 seconds respectively. In this

http://www.bepress.com/ijeeps/vol9/iss4/art5 12
 Mandal and Chakraborty: Effect of Control Parameters

case also, several tests were performed with higher number of iterations (up to
1000 in step of 100) for lower values of C R (≤ 0.10), but no improvement in
results were observed. The values in the range of 0.30 ≤ C R ≤ 0.95 yields better
results in terms of cost, emission and computation time. The optimum result is
obtained with C R = 0.90, when the minimum cost, minimum emission and
computation time are found to be $9592.40, 3232.20 lbs and 16.2781 seconds
respectively.

4.3 Effect of varying Population size - N P

In order to obtain the effect of the population size N P , it is varied from 30 to 160
in steps of 10 considering the higher dimension (for the present case it is 14,
explain earlier) of the problem under consideration. The mutation factor f m and
the crossover factor C R are set at 0.50 and 0.90 respectively based on the process
described in section 4.1 and 4.2. It this case also, one hundred (100) independent
runs are performed for every parameter combinations and maximum number of
iteration is set at 300.

Table 4
Effect of Population size N P on Combined Economic Emission Dispatch (CEED)
with C R = 0.90 and f m = 0.50 for a demand of 2000 MW.
Value Fuel Cost ($) Emission (lb) Average
of CPU
NP Time
(Sec.)
Minimum Maximum Average Minimum Maximum Average
30 10030.00 10945.00 10341.00 3342.00 4750.20 4178.70 4.6703
40 9884.10 10279.00 10082.00 3571.40 4499.40 3945.50 6.5609
50 9810.30 10284.00 10073.00 3486.40 4041.80 3743.80 8.1410
60 9864.70 10421.00 10014.00 3339.70 3962.10 3593.30 10.6720
70 9764.00 9943.00 9814.50 3257.70 3845.80 3447.20 11.5310
80 9698.50 10040.00 9741.30 3237.70 3649.80 3385.30 13.1828
90 9594.40 10052.00 9742.60 3275.90 3466.40 3298.30 14.4984
100 9592.40 9945.90 9644.80 3232.20 3430.80 3351.50 16.2781
110 9591.40 9970.00 9688.40 3238.20 3724.60 3306.30 17.9687
120 9592.40 9905.70 9695.50 3237.30 3691.10 3300.90 19.7750
130 9588.00 9929.40 9727.60 3233.60 3561.40 3248.30 22.2343
140 9593.70 9998.20 9762.30 3231.00 3717.30 3357.20 27.7953
150 9593.30 10009.00 9702.20 3231.20 3338.60 3253.20 24.9750
160 9592.70 9811.00 9659.10 3232.10 3360.10 3242.70 26.6390

Published by The Berkeley Electronic Press, 2008 13

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

The results for different combinations are presented in Table 4. It is seen

that convergence rate is 100% above a value of N P = 80. However, computation
time increases gradually along with population size. The optimum result is
obtained with N P = 100, when the minimum cost, emission and computation time
are found to be $9592.40, 3232.20 lbs and 16.2781 seconds respectively.

4.4. Results of Combined Economic Emission Dispatch (CEED)

Control parameters of DE were selected through the process as above and the
optimal parameters as obtained are N P =100, f m = 0.50 and C R = 0.9. Maximum
iteration was set at 300. Table 5 shows the results for optimized cost, generation
schedule, penalty factors, losses, emission output and computation time for
combined economic emission dispatch (CEED) with a demand of 1500 MW,
2000 MW and 2500 MW.

Table 5
Solution for Fourteen Generator system for Combined Economic Emission
Dispatch (CEED) with N P = 100, C R = 0.90 and f m = 0.50
Demand (MW)
Generation of Units (MW) 1500 2000 2500
P1 150.20 239.76 329.53
P2 150.23 150.01 219.54
P3 96.51 95.21 129.99
P4 120.01 119.75 120.06
P5 150.10 199.86 249.75
P6 135.25 284.59 384.25
P7 135.07 234.86 284.43
P8 60.59 159.73 209.56
P9 125.10 124.89 161.74
P10 139.09 137.32 159.89
P11 58.77 66.65 79.95
P12 79.97 79.95 79.93
P13 64.07 84.97 84.84
P14 52.95 52.43 52.65
Total Generation (MW) 1517.88 2029.98 2546.11
Losses (MW) 17.88 29.98 46.11
Penalty Factor (h) 1.2191 1.5299 1.5716
CPU Time (Sec). 36.2812 16.2736 18.3542
Iterations 300 300 300
Fuel Cost ($/hr) 6869.90 9592.40 12871.00
Emission (lb/hr) 1340.20 3232.20 6316.30

http://www.bepress.com/ijeeps/vol9/iss4/art5 14
 Mandal and Chakraborty: Effect of Control Parameters

It is important to note that the process takes longer time when the demand
is near minimum or maximum limits. For example, the CPU time is 36.2812 sec.
and 18.3542 sec. with a demand of 1500 MW and 2500 MW respectively,
whereas it is 16.2736 sec. with a demand of 2000 MW as shown in Table 5.This
is due to the fact that the algorithm takes more time for generating initial set of
feasible solutions satisfying constraints near the lower and upper generation
limits.

5 Conclusion

The success of DE is heavily dependent on setting of control parameters. Optimal

settings of the control parameters of DE depend on the specific optimization
problem in hand. In this paper, a parameter study of control parameters of
differential evolution (DE) is conducted for combined economic emission
dispatch with generator constraints, valve-point loading and transmission loss.
The effect of variation of all the three control parameters of DE i.e. NP, fm and CR
are investigated and the results are presented. It is seen that premature
convergence and even stagnation can occur due to wrong parameter selection.
Based on the results obtained, recommendations are suggested for the range of
control parameters of DE.

Appendix A: List of Symbols

FC( Pg ) : total fuel cost of generation in the system ($/hr)

ai ,bi , ci : fuel cost coefficients of the i th generating unit
Pi : power generated by the i th unit
n : number of thermal units
FCv ( Pg ) : total fuel cost of generation in ($/hr) including valve point loading
ei , f i : fuel cost coefficients of the i th generating unit reflecting valve-point
effect
Pi , min : minimum power generation by i th unit
Pi , max : maximum power generation by i th unit
PD : total power demand
PL : total transmission loss
Bij : elements of loss coefficient matrix B
EC( Pg ) : total amount of emission (lb/hr)
αi ,βi ,γ i : emission coefficients of the i th unit

Published by The Berkeley Electronic Press, 2008 15

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

φt : total operational cost of the system

w1 , w2 : weight factors

References

[1] Storn R.and Price K., “Differential evolution: a simple and efficient
adaptive scheme for global optimization over continuous spaces”, Technical
Report TR-95-012”, Berkeley, USA: International Computer Science Institute,
1995.
[2] Lampinen J. and I. Zelinka (2000), “On Stagnation of the Differential
Evolution Algorithm”, in Osmera P.(ed), Proc. of MENDEL 2000, 6th
International Mendel Conference on Soft Computing, pp. 76-83.
[3] Liu J.and Lampinen J., “On Setting the Control Parameter of the
Differential Evolution Method”, In Proceedings of the 8th International
Conference on Soft Computing (MENDEL 2002), pages 11–18, 2002.
[4] G¨amperle R., M¨uller S. D., and Koumoutsakos P., “A Parameter Study
for Differential Evolution,” in Advances in Intelligent Systems, Fuzzy
Systems, Evolutionary Computation, A. Grmela and N. Mastorakis, Eds.
WSEAS Press, 2002, pp. 293–298.
[5] Zielinski K., Weitkemper P., Laur R.and Kammeyer K., “Parameter study
for differential evolution using power allocation problem including
interference cancellation,” Proceedings of IEEE Congress on Evolutionary
Computation, Vancouver, British Columbia, Canada, 2005, pp. 1857–64.
[6] Ro¨nkko¨nen J, Kukkonen S. and Price KV. “Real-parameter optimization
with differential evolution”, Proceedings of IEEE Congress on Evolutionary
Computation. Edinburgh, UK; 2005, pp. 506–13.
[7] Brest J., Zumer V. and Mauˇcec M. S., “Self-Adaptive Differential
Evolution Algorithm in Constrained Real-Parameter Optimization,” 2006
IEEE Congress on Evolutionary Computation Sheraton Vancouver Wall
Centre Hotel, Vancouver, BC, Canada July 16-21, 2006, pp.910-926.
[8] Zaharie D, “Critical Values for the Control Parameter of the Differential
Evolution Algorithms,” MENDEL 2002, 8th International Conference on Soft
Computing (Matouˇsek R. and Oˇsmera P. eds). University of Technology,
Brno (2002), pp.62 – 67.
[9] Teo J., “Exploring dynamic self-adaptive populations in differential
evolution,” Soft Computing - A Fusion of Foundations, Methodologies and
Applications, 2005. DOI: 10.1007/s00500-005-0537-1.
[10] IEEE Current Operating Problems Working Group, “Potential Impacts of
Clean Air Regulations on System Operations”, IEEE Transactions on Power
Systems, Vol. PWRS-10, (1995), pp. 647–654.

http://www.bepress.com/ijeeps/vol9/iss4/art5 16
 Mandal and Chakraborty: Effect of Control Parameters

[11] Wong, K.P and Yuryevich, J.,“ Evolutionary Programming Based

Algorithm for Environmentally Constrained Economic Dispatch,” IEEE
Transaction on Power Systems, Vol. 13.No.2, May (1998), pp 301.
[12] Ramanathan, R., “Emission constraints economic dispatch,” IEEE
Transaction on Power Systems, vol. 9, No. 4, November (1994), pp. 1994–
2000.
[13] Park, J. B., Lee, K.S., Shin, J.R. and Lee, K.Y., “A Particle swarm
optimization for solving the economic dispatch with nonsmooth cost
functions,” IEEE Transaction on Power Systems, Vol. 20.No.1, February
(2005), pp34-42.
[14] Walters, D. C. and Sheble, G. B., “Genetic algorithm solution of economic
dispatch with the valve point loading,” IEEE Trans. Power Systems, Vol. 8,
August (1993), pp. 1325–1332.
[15] Kulkarni, P. S., Kothari, A. G., and Kothari, D. P., “Combined economic
and emission dispatch using improved backpropagation neural network,”
Electric Power Components and System, vol. 28, (2000), pp. 31–44.
[16] K. Price, “Differential Evolution: A Fast and Simple Numerical
Optimizer,” Biennial Conference of the North American Fuzzy Information
Processing Society. NAFIPS. 19-22 Jun 1996, pp. 524-527.
[17] Storm R.and Price K., “Differential Evolution - A simple and efficient
adaptive scheme for global optimization over continuous spaces”, Journal of
Global Optimization, 1997, vol. 11, Dordrecht, pp. 341- 359.
[18] Michalewicz Z. and Schoenauer M., “Evolutionary Algorithms for
Constrained Optimization Problems,” Evolutionary Computation, Vol. 4, No.
1, 1996, pp. 1-32.
[19] Lampinen J., “A constraint Handling Approach for the Differential
Evolution Algorithm,” Proceedings of the 2002 Congress on Evolutionary
Computation, CEC '02, Volume 2, 2002, pp. 1468-73.
[20] P. Venkatesh, Ganandass R. and Padhy N.P., “Comparison and application
of evolutionary programming techniques to combined economic emission
dispatch with line flow constraints,” IEEE Transaction on Power Systems,
Vol. 18, No.2, May (2003), pp. 688–697.
[21] Gaing Zwe-Lee, “ Particle swarm optimization to solving the economic
dispatch considering the generator constraints,” IEEE Transaction on Power
Systems, Vol. 18, No.3, August (2003), pp. 1187–1195.
[22] Storn, Rainer (1996), “On the usage of differential evolution for function
optimization,” In the proceedings of Biennial Conference of the North
American Fuzzy Information Processing Society (NAFIPS 1996), Berkeley,
pp. 519–523.

Published by The Berkeley Electronic Press, 2008 17

 International Journal of Emerging Electric Power Systems, Vol. 9 [2008], Iss. 4, Art. 5

[23] Lampinen J. and Storn R., “Differential Evolution,” In New Optimization

Techniques in Engineering, G. C. Onwubolu and B. Babu, Eds. Berlin
Heidelberg: Springer-Verlag, 2004, pp. 123–166.

Biographies

K. K. Mandal was born in Calcutta, India on 25th December, 1964. He received

the B.E Degree in Electrical Engineering from Jadavpur University, Kolkata,
India in 1986 and M.E Degree from Allahabad University, Allahabad, India in
1998. His employment experience includes Indian Telephone Industries, National
Institute of Technology, Durgapur, India. He is presently working as a Reader in
the Department of Power Engineering, Jadavpur University, Kolkata, India. His
present research interest includes Power Economics, Deregulated Electricity
Industry and Power Electronics.

N. Chakraborty was born in Kolkata, India on 27th August, 1964. He received his
Bachelor of Electrical Engineering in 1986 and Masters of Electrical Engineering
in 1989 from Jadavpur University, Kolkata. He was awarded with the D.I.C from
Imperial College, London, U.K. and Ph.D Degree from University of London in
1999. At present he is a professor in the Department of Power Engineering,
Jadavpur University, Kolkata. His fields of research interest include Power and
Energy Economics, Applied Superconductivity and Environmental Measurements
and Analysis.

http://www.bepress.com/ijeeps/vol9/iss4/art5 18