Applied Soft Computing 29 (2015) 122–137

Contents lists available at ScienceDirect

Applied Soft Computing

journal homepage: www.elsevier.com/locate/asoc

Large scale economic dispatch of power systems using oppositional

invasive weed optimization
A.K. Barisal ∗ , R.C. Prusty
Department of Electrical Engineering, VSSUT, Burla, Odisha, India

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents an evolutionary hybrid algorithm of invasive weed optimization (IWO) merged with
Received 29 November 2013 oppositional based learning to solve the large scale economic load dispatch (ELD) problems. The oppo-
Received in revised form 17 October 2014 sitional invasive weed optimization (OIWO) is based on the colonizing behavior of weed plants and
Accepted 9 December 2014
empowered by quasi opposite numbers. The proposed OIWO methodology has been developed to min-
Available online 7 January 2015
imize the total generation cost by satisfying several constraints such as generation limits, load demand,
valve point loading effect, multi-fuel options and transmission losses. The proposed algorithm is tested
Keywords:
and validated using five different test systems. The most important merit of the proposed methodology is
Economic load dispatch
Invasive weeds optimization
high accuracy and good convergence characteristics and robustness to solve ELD problems. The simulation
Opposition based learning results of the proposed OIWO algorithm show its applicability and superiority when compared with the
Seeds results of other tested algorithms such as oppositional real coded chemical reaction, shuffled differential
Fitness evolution, biogeography based optimization, improved coordinated aggregation based PSO, quantum-
Valve point loading inspired particle swarm optimization, hybrid quantum mechanics inspired particle swarm optimization,
modified shuffled frog leaping algorithm with genetic algorithm, simulated annealing based optimization
and estimation of distribution and differential evolution algorithm.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction algorithms like genetic algorithm (GA) [4–6], simulated annealing (SA) [7], evo-
lutionary programming (EP) [8] and hierarchical method [9] may prove to be
Economic load dispatch (ELD) is one of the power system optimization problems very efficient in solving complex power system problems but, these heuristic
has high dimensional, high constraints, non-convex, non-smooth and nonlinear methods do not always guarantee the globally optimal solution. In recent years,
characteristics and requires an efficient optimization technique to be solved. The differential evolution (DE) [10,11], Ant colony optimization [12], artificial immune
modern power systems encounter numerous technical and economical difficul- systems (AIS) [13], bacteria foraging optimization [14] modified genetic algorithm
ties under competitive deregulated environment. The ELD problem is usually a sub [15,16], modified particle swarm optimization (PSO) [17–21] and biogeography
problem of unit commitment and also a constrained optimization task. The prime based optimization (BBO) [25] have been successfully applied to ELD problems.
requirement of ELD is to allocate the optimal generation levels of online gener- Quite promising results in terms of fuel cost savings and faster convergence have
ating units so as to accomplish the load demand at the minimum operating cost been obtained by these techniques.
under various system constraints. Over the years, various mathematical program- However, SA algorithm finds the solution trapped by local optimum rather than
ming methods and nature inspired meta-heuristic optimization techniques have at the global optimum. Moreover, tuning of its relevant control parameters is a
been successfully employed to solve the ELD problems. The conventional methods difficult task. The recent research has identified few drawbacks of the stochastic
include classical calculus method [1], base point and participation factor method, methods like GA of its premature convergence causing degradation in performance
gradient search method, linear programming [2], nonlinear programming etc. A and reduction in its search capability and unsuitable when applied to multimodal
dynamic programming (DP) method can solve such problems in different formu- objective functions. The main drawback of SA, GA, EP and AIS, is their slow conver-
lations [3]. However, the drawback of the DP is its huge computational overburden gence toward optimal solution, which is not suitable for real time operation. Though
when applied to practical sized ELD problems in stipulated time zones. These numer- the convergence characteristic of PSO is fast and acceptable when applied to large-
ical methods require the incremental cost curves to be monotonically increasing or scale real time ELD problems still, the generation schedule obtained is not always
piece wise linear. However, these methods have difficulties and are not suitable to global best solution; rather they often achieve a near global optimal solution.
address nonlinear and discontinuous characteristics [4] of actual practical problems Within the last few years, new optimization methods with modifications of
rather complicating the problem solutions. existing methods have been applied in order to obtain the global or nearly global
The great attempts of researchers across the globe to overcome the limitations of optimal solutions to ELD problems. These modified meta-heuristic algorithms
conventional mathematical programming are leaded to introduce meta-heuristics like, GA based ant colony optimization algorithm [5], Nelder–Mead based BFA
(BFA-NM) [14], modified shuffled frog leaping algorithm with genetic algorithm
(MSFLA& GA) [15], quantum-inspired particle swarm optimization (QPSO) [18],
hybrid quantum mechanics inspired particle swarm optimization (HQPSO) [19],
∗ Corresponding author. Tel.: +91 6632430754; fax: +91 6632430204. improved coordinated aggregation based PSO (ICA-PSO) [20], an improved particle
E-mail address: a [email protected] (A.K. Barisal). swarm optimization (IPSO) [21], shuffled differential evolution (SDE) [22], DE with

http://dx.doi.org/10.1016/j.asoc.2014.12.014
1568-4946/© 2014 Elsevier B.V. All rights reserved.
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 123

generators of chaos sequences and sequential quadratic programming (DEC-SQP) (b) The generator limits: The power output of each generator
[23], variable scaling hybrid differential evolution (VSHDE) [24], hybrid differen- should vary within its minimum and maximum limits. That is,
tial evolution with BBO [25] known as (DE/BBO) [26] and oppositional real coded
the following inequality constraint for each generator should
chemical reaction optimization (ORCCRO) [27] have been successfully applied in
constrained ELD problems. The opposition based learning (OBL) [28] have been be defined for each generator:
incorporated in chemical reaction optimization to improve the convergence rate
of the algorithm. However, this ORCCRO algorithm requires a lots of control param-
Pgi min ≤ Pgi ≤ Pgi max (4)
eters tuning which is a difficult task. The advantages of these algorithms that they do
Pi is the power output of ith generator and Pgimin , Pgimax are the
not have any or fewer restrictions on the shape of cost function curves or problem
constraints. However, they are quite sensitive to various parameters tuning, their minimum and maximum real power output of ith generator.
solution is not unique for each trial run and also the problem of large execution time.
The first application of oppositional based learning and back propagation in neu- 2.2. ELD with non-smooth cost functions
ral network was proposed by Ventresca and Tizhoosh [29]. Since then, it has been
applied to many soft computing techniques such as DE [30], PSO by Wang et al.
[31], ant colony optimization [32,33], biogeography based optimization [34], gravi-
Practically, the ELD problems are inherently highly non linear
tational search algorithm [35], harmony search algorithm [36] and teaching learning and discontinuous in nature. Moreover, the cost functions have
based optimization [37]. It has been proved that a quasi opposite number is usually discontinuities corresponding to the change of fuels and also due
closer than an opposite number to the solution. This paper utilizes the improved to valve point effects that make the problem multimodal. There-
computational efficiency of quasi opposition based learning concept in the pro-
fore, most of the techniques fail to obtain global solution instead of
posed invasive weed optimization algorithm. This evolutionary algorithm known as
invasive weed optimization (IWO), is a more robust, stochastic and derivative free quasi-global optimums to power system optimization problems.
optimization tool for the solution of complex real world problems. The algorithm
is based on the invasive habits of growth of weeds in nature and having excel- 2.2.1. Non-smooth cost function with valve point effects
lent exploration and exploitation ability in the search area. It was first developed by
The generators with multiple valve steam turbines possess a
Mehrabian and Lucas [38] and since then, many applications have been found of this
algorithm such as recommender system design [39], antenna system design [40], wide variation in the input–output characteristics due to wire
state estimation of nonlinear systems [41] and unit commitment problem solution drawing effects. The valve point effect introduces ripples in the
[42]. heat rate curves and cannot be represented by the polynomial func-
Furthermore, oppositional based learning empowers the proposed IWO algo- tion as in (2). Therefore, the accurate cost curve is a combination
rithm to obtain best solution in lesser time. The proposed algorithm is tested on five
different test systems breaking down the previous best results in all cases. The sim-
of sinusoidal functions and quadratic functions represented by Eq.
ulation results so obtained show its reliability and superiority in solving constrained (5).
ELD problems.
Fi (Pgi ) = ai + bi Pgi + ci Pgi2 + |ei × sin(fi × (Pgi min − Pgi ))| (5)

2. Problem formulation where ei , fi are the constants of the ith unit with valve point effects.

2.1. ELD with smooth cost function 2.2.2. Cost function with change of fuels
Generally, the dispatching units are practically supplied with
The prime objective of the ELD problem is to determine the most multi-fuel sources, each unit should be represented with several
economic loadings of generators to minimize the generation cost piecewise quadratic functions reflecting the effect of fuel type
such that the load demands PD in the scheduling horizon can be met changes, and the generator must identify the most economic fuel
and simultaneously, the power balance constraint and generating to burn. The generator with multiple fuel options [9] has different
limit constraints are satisfied. Here, this constrained optimization input–output curve. Therefore, it is more appropriate to represent
problem can be written as: the cost functions with piecewise quadratic functions described in
(6).

d
⎧
Minimize FTotal = Fi (Pgi ) (1) ⎪ ai1 + bi1 Pi + ci1 Pi2 if Pi min ≤ Pi ≤ Pi1 , fuel − 1
⎪
⎪
i=1 ⎪ a + b P + c P2
⎪
⎨ i2 i2 i i2 i if Pi1 ≤ Pi ≤ Pi2 , fuel − 2
In general, the cost function of ith unit Fi (Pgi ) is a quadratic Fi (Pi ) = (6)
polynomial and is expressed as: ⎪
⎪
...
⎪
⎪
⎪
⎩
. ..
Fi (Pi ) = ai + bi Pi + ci Pi2 (2) aim + bim Pi + cim Pi2 if Pim−1 ≤ Pi ≤ Pi max , fuel − m
where ai , bi and ci are fuel cost coefficients of ith unit, and d is the where aij , bij , cij are cost coefficients of unit i for the jth fuel type
total number of committed units. and Pi = Pgi , m = (number of generators).

(a) Active power balance constraint or demand constraint: The 2.2.3. Cost function with valve point effects and change of fuels
d
total generation i=1
(Pgi ) should be equal to the total system In reality, the objective function of the practical economic dis-
demand PD and total transmission loss PLoss . That is represented patch problem has non-differentiable points according to valve
as point loadings and multiple fuels. Therefore, the objective function
should be composed of a set of non-smooth functions to obtain

d
an accurate and practical economic dispatch solution. The cost
(Pgi ) = PD + PLoss (3)
function is framed by combining both valve point loadings and
i=1
multi-fuel options which can be realistically represented as shown
below in (7).
⎧
⎪ ai1 + bi1 Pi + ci1 Pi2 + |ei1 × sin(fi1 × (Pi1 min − Pi1 ))|, for fuel1, Pi1 min ≤ Pi ≤ Pi1
⎪
⎪
⎪
⎨ ai2 + bi2 Pi + ci2 Pi2 + |ei2 × sin(fi2 × (Pi2 min − Pi2 ))|, for fuel2, Pi2 min ≤ Pi ≤ Pi2
Fi (Pi ) = (7)
⎪
⎪ .. .. ..
⎪
⎪ . . .
⎩
aim + bim Pi + cim Pi2 + |eim × sin(fim × (Pim min − Pim ))|, for fuel.m, Pim min ≤ Pi ≤ Pim
124 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

3. Invasive weed optimization to the global optimum solution than a random candidate solution.
They have successfully implemented this learning process in to
Invasive weed optimization is a novel population based numeri- different soft computing techniques [34–37].
cal stochastic, derivative free optimization algorithm inspired from Here, the opposite number and opposite point used in OBL have
the biological growth of weed plants. It was first developed and a straight forward definition as follows:
designed by Mehrabian and Lucas [38]. This technique based on
the colonizing behavior of weed plants [39]. Some of the interesting 4.1. Opposite number
characteristics of weed plants that are invasive, fast reproduction
and distribution, robustness and self adaptation to the changes Opposite number may be defined as the mirror point of the solu-
in climate conditions. The various applications of the proposed tion from the center of the search space and it is mathematically
algorithm include DNA computing, antenna system design [40], expressed as:
state estimation of nonlinear systems [41], optimal arrangement If x be a real number between [a, b], its opposite number x0 is
of piezoelectric actuators on smart meters and unit commitment defined as
problem solution [42].
x0 = a + b − x (8)
Some important properties of IWO algorithm are described
below for which it is considered as a sophisticated tool to solve
4.2. Opposite point
complex optimization problems. One important property of the
IWO algorithm is that it allows all of the plants to participate in the
The opposition point in d dimensional space is given as:
reproduction process. Fitter plants produce more seeds than less fit
If P(x1 , x2 , . . .. . . xi , . . .. . . xd ) is a point in d-dimensional search
plants, which tends to improve the convergence of the algorithm.
space, its opposite point OP(x10 , x20 , . . .. . .xi0 , . . .. . .xd0 ) may be
Furthermore, it is possible that some of the plants with the lower
defined as follows
fitness value carry more useful information in iteration process as
compared to the fitter plants. This IWO algorithm gives a chance to xi0 = ai + bi − xi ; xi ∈ [a, b]; i = 1, 2, . . . d (9)
the less fit plants to reproduce and if the seeds produced by them
The same logic can be extended for Quasi-opposite number and
have good finesses in the colony, they can survive [38]. The other
quasi-opposite point used in QOBL may be defined as follows:
important feature of IWO algorithm is that weeds reproduce with-
out mating. Each weed can produce new seeds, independently. This
4.3. Quasi-opposite number
property adds a new attribute to the algorithm that each agent may
have different number of variables during the optimization process.
It may be defined as the number between the center c of the
Thus, the number of variables can be chosen as one of the optimiza-
search space and the opposite number, quasi-opposite number, xq0
tion parameters in this algorithm. This IWO algorithm has the more
may mathematically be expressed as:
chance to avoid local minima points compared to GA and PSO due
a + b
to its continuous and normally distributed dispersal structure over
xq0 = rand ,a + b − x = rand(c, x0 ) (10)
search space which has a decreasing variance parameter centered 2
on each parent plant [39].
Each individual or agent, a set containing a value of each opti- 4.4. Quasi-opposite point
mization variable, is called a seed set. Each seed in seed set grows
q0 q0 q0 q0
to a flowering plant in the colony (matrix of size Popmax × d). The The quasi opposite point QOP(x1 , x2 , . . .. . .xi , . . .. . .xd ) for
meaning of a plant is one individual or agent after evaluating its fit- d-dimensional search space is given by:
ness. Therefore, growing a seed to a plant corresponds to evaluating a + b
q0 i i
an agent’s fitness. Depending upon the fitness value of individual, xi = rand , ai + bi − xi = rand(ci , xi0 ); i = 1, 2, . . . d
2
new seeds are generated by each plant (row) with respect to a (11)
normalized standard deviation . The standard deviation is very
q0
important in calculating the optimal solution. It can be seen that where xi is a random number uniformly distributed between ci
the SD is reduced from the initial SD to the final SD with different and xi0 .
velocities. The algorithm starts with such a high initial SD that the
optimizer can explore through the whole solution space. The new 4.5. Oppositional invasive weed optimization algorithm
produced weeds are randomly distributed in the field and grow to
flowering weeds for fitness calculation after satisfying all the con- In this section, quasi-oppositional based invasive weed opti-
straints within their limits. Keeping in view that fitter plants have mization is implemented to solve ELD problems. The OIWO
been reproduced more than undesirable plants. In order to limit algorithm employs quasi-oppositional based population initializa-
the maximum number of plants in a colony, the weaker and inap- tion and generation jumping to accelerate the convergence rate of
propriate plants are eliminated. This iterative process continues till IWO algorithm.
the maximum number of iterations is reached. The plant with the The steps of the proposed approach are mentioned below:
best fitness is selected as the optimal solution. Step 1: The seeds are initialized depending upon the number of
selected variables (d) of the given problem, distributed uniformly
over the entire solution space.
4. Opposition based learning Step 2: Create each seed set, after generating all the selected vari-
ables of the given problem randomly within their effective lower
Opposition-based learning (OBL) initially developed by and upper limits and satisfying different constraints. Thus each seed
Tizhoosh [28], is a new concept in computational intelligence contains random values for each variable in the solution space. Each
and is used to accelerate the convergence rate of different opti- seed set represents a potential solution of the given problem. Gen-
mization techniques. OBL considers current population as well erate several seed set to create a Seed matrix (S) of size (Popmax × d).
as its opposite population at the same time in order to get better The total number of plants in the population is Popmax . In similar
candidate solution. Many researchers across the globe proved that way, create a quasi-opposite seed matrix (QOS) using Eq. (11) after
an opposition candidate solution has a better chance to be closer satisfying all the constraints within their limits.
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 125

Step 3: The fitness value of all individuals of the current seed set Step 3: Let, Pj = [ P1 P2 . . .Pd ] be the trial vector denoting the
(S) and quasi opposite seed set ((QOS), (each row (plant) of S and current seed set of population (Popmax × d) to be evolved. The trial
QOS) is calculated according to the objective function considered random current seed matrix (S) is as follows:
in the optimization problem. These seeds then evolve into weed ⎡ ⎤
P11 P12 ... ... P1d
plants capable of producing new units.
⎢ ⎥
Step 4: Each individual (plant) is ranked based on its fitness with ⎢ P21 P22 ... ... P2d ⎥
respect to other weeds. Subsequently, each weed produces new ⎢ ⎥
S=⎢
⎢ ... ... ... ... .
⎥
⎥ (14)
seeds depending on its rank in the whole population including the
⎢ ⎥
quasi opposite seed set. Starting with the maximum number of ⎣ ... ... Pij ... Pid ⎦
seeds (Nmax ) produced by the best fit plant. All plants are partic-
PPopmax ,1 ... PPopmax j ... PPopmax ,d
ipating in reproduction process which adds a new attribute to the
optimization providing chances to contribute useful information The elements of Pi are the real power outputs of the commit-
(good result) by less fit plants during iterative process. ted d generating units subjected to economic fuel types to burn
Step 5: The number of seeds to be created by each weed alters lin- and their respective capacity constraints in (4). To meet exactly the
early from Nmin to Nmax which can be computed using the equation load demand, a dependent unit is randomly selected from among
below. the committed d units. Let, Pdd be the power output of the depend-
Fi − Fworst ent unit to satisfy equality constraint (slack generator), then Pdd is
Number of seeds = (Nmax − Nmin ) + Nmin (12) calculated by
Fbest − Fworst
where Fi is the fitness of ith weed. Fworst , and Fbest denote the 
d

worst and best fitness in weed population. The generated seeds Pdd = PD + PLoss − Pi i = 1, 2 . . . d (15)
are normally distributed over the field with zero mean and varying i=1
standard deviation of  iter described by
i=
/ dd
 iter − iter
n
max
iter = (0 − f ) + f (13) Step 4: Go to step 3 for few iterations and then, go to step 5.
itermax
Step 5: Each seed set (each row) must be a feasible candidate
where itermax and iter are the maximum number of iteration cycles solution that satisfies the given constraints. Similarly, quasi oppo-
assigned by the user and current iteration number, respectively.  0 , site seed set of population (QOS) is created as per Eq. (14). The
 f are the predefined initial and final standard deviations and n is fitness value (fuel cost) of all individuals (each weed) of the cur-
the nonlinear modulation index. rent seed set matrix (S) and quasi opposite seed set matrix (QOS),
Step 6: After that all seeds have found their positions over the is calculated according to the objective function (1). Each individual
search area, the new seeds grow to the flowering plants after satis- (plant) is ranked based on its fitness with respect to other weeds.
fying all the constraints within their limits and then, they are ranked Subsequently, each weed produces new seeds depending on its
together with their parents in both seed set and quasi opposite seed rank in the whole population including the quasi opposite seed set.
set matrices. Plants with lower ranking in the colony are eliminated Starting with the maximum number of seeds (Nmax ) produced by
to reach the maximum number of plants in the colony (Popmax ). the best fit plant.
Step 7: Based on jumping rate ‘jr ’ (i.e., jumping probability), Step 6: Update the iteration as iter = iter + 1.
after generating the new seed set, the opposite seed set is calcu- Step 7: All of the plants are participating in the reproduction
lated based on standard deviation and ranking of its weeds. The process. The numbers of new seeds are generated using Eqs. (12)
quasi opposite set generation using jumping rate may be described and (13). Each individual (plant) should satisfy both equality (3)
below: and inequality constraint (4). The infeasible solutions are replaced
if rand < jr by randomly generated new seed set. The current seed set matrix
for i = 1:Popmax
(S) and quasi opposite seed set matrix (QOS) are updated with
for j = 1:d
q0

aj +bj  increasing individuals which are feasible candidate solutions.
xij = rand 2
, aj + bj − xij ;
end
Step 8: The fitness value of each individual is calculated in the
end entire population. They are ranked in descending order according
end their fitness. The best plant is identified with highest fitness value
Step 8: Survived plants can produce new seeds based on their in the colony (the best fuel cost) and corresponding seed set is the
ranking in the colony. The fittest individual (plant) is selected from optimal solution (generator outputs) stored in a separate memory
the seed-parent combination of current seed set and quasi opposite location for comparison and updated in every iteration.
seed set. If the stopping criterion is satisfied, the iterative process Step 9: Plants with lower ranking are eliminated in order to limit
is terminated and the results are displayed, otherwise go to Step 3 the maximum number of plants in a colony to Popmax .
for continuation. Step 10: Select a new parameter jumping rate ‘jr ’ within [0,1] and
create quasi opposite seed set matrix (QOS) from the newly devel-
4.6. OIWO algorithm for economic load dispatch problems oped seed set generated in the previous steps as per the following
procedure:
The flow chart of the OIWO for solving ELD problem is shown in if rand < jr
for i = 1:Popmax
Fig. 1. The computational procedure for proposed OIWO technique
to solve dispatch problems can be described in the following steps.
for j = 1:d
q0

aj +bj 
xij = rand 2
, aj + bj − xij ;
Step 1: Input parameters of the system and specify the upper
end
and lower boundaries of each variable. Read the value of Popmax , d, end
Pgmin = a, Pgmax = b, cost coefficients, B-coefficient matrix, PD , Nmin , end
Nmax ,  0 ,  f , n, jr and itermax . The individuals of newly generated QOS matrix must satisfy all
Step 2: Initialize randomly the seeds of the population. These the feasible constraints. The fitness (fuel cost) of each individual
seed set must be feasible candidate solutions that satisfy the given is calculated and ranked among all the individuals present in the
constraints. colony.
126 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

Start

Initialize all variables i.e., Popmax , d , pg min , pg max , cost coefficients, B-coefficient
matrix, PD , N min , N max , σ 0 , σ f , n , jr and itermax

Generate initial random seed set (matrix, S). Each Generate initial random quasi
element of the given seed matrix should lie opposite seed set (QOS, matrix)
between the real power generation limits and using equation (11) in similar way
satisfy the load balance constraint (3). of current seed matrix.

The dispersed seeds grow into weed plants in both matrices. Evaluate the fitness of each
individual plant of current seed matrix and in QOS matrix. Cost of power generation of each plant
in the colony.

Sort the fitness in descending order and rank all the plants

Depending upon the rank of the plant generate the new seeds using spatial distribution by equations
(12) & (13).Each element of the both seed matrix should lie between the real power generation limits
and each individual should satisfy the load balance constraint (3)

Determine the new fitness of the seed-plant combination in the colony.Sort the fitness
(cost of power generation) in descending order and rank all the plants.

Elimination of low ranking plants to reach maximum

number of plants in the colony Popmax .
Create new Quasi oppositional
seed set matrix (QOS) as above.
Evaluate fitness of each plant and
Sort the fitness in descending
Yes order and rank all the plants of
Is rand < j r QOS and current seed matrix.
Select Popmax plants as per
ranking.

No

The best plant is identified with highest fitness value in the colony (the best fuel
cost) and corresponding seed set is the optimal solution (generator outputs)

No Is Iteration
= itermax ? Yes

Yes

Print the result with the best fuel cost and corresponding generation levels

Stop

Fig. 1. Flow chart of OIWO algorithm.

Step 11: If the stopping criterion as maximum iterations itermax 5.1. Description of test systems
is satisfied then go to step 12. Otherwise, go to step 6.
Step 12: The best plant is identified with highest fitness value in 5.1.1. Test system-1
the colony (the best fuel cost) and the corresponding seed set is the This system comprising of 13 thermal units with valve point
optimal solution (generator outputs). effect is included in the cost model. The load demands of this sys-
tem are considered as 1800 MW for no transmission losses and
2520 MW for transmission losses. The system input data have been
5. Simulation results and analysis provided in Appendix A and also available in [8,23].
The results without losses are compared with modified shuf-
The proposed IOWO algorithm has been applied to solve ELD fled frog leaping algorithm with GA crossover (MSFLA and GA)
problems of five different benchmark test systems to demonstrate [15], quantum-inspired particle swarm optimization (QPSO) [18],
its performance in comparison to several established optimization improved coordinated aggregation-based PSO (ICA PSO) [20],
techniques reported in literature. The bio-inspired OIWO algo- chaotic differential evolution and sequential quadratic program-
rithm has been implemented using Matlab-7.10.0.499 (R2010a) ming (DEC SQP) [23] and provided in Table 1a. The reported fuel
environment on a 3.06 GHz, Pentium-IV; with 1 GB RAM personal cost and the exact fuel cost should be same or nearly same. Large
computer. difference in fuel cost may be considered as serious error. It is found
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 127

Table 1a
Comparison of results for13-generator system without losses (PD = 1800 MW).

Units Power outputs in MW

DEC-SQP [23] QPSO [18] ICA PSO [20] MSFLA & GA [15] OIWO (proposed)

1 628.3173 538.56 628.32 540.5290 628.3185

2 149.2407 224.70 149.96 225.0747 222.7554
3 223.1685 150.09 299.75 207.9648 149.5936
4 109.8540 109.87 109.86 69.0974 109.8665
5 109.8657 109.87 109.86 84.9624 109.8665
6 109.8666 109.87 60.00 94.7620 109.8665
7 109.8211 109.87 109.87 106.9725 109.8665
8 109.8664 109.87 109.87 109.0098 60.0000
9 60.00 109.87 109.87 108.3280 109.8665
10 40.00 77.41 40.00 79.6389 40.0000
11 40.00 40.00 40.00 63.7670 40.0000
12 55.00 55.01 55.00 58.0643 55.0000
13 55.00 55.01 55.00 72.9603 55.0000

Total power 1800 1800 1821.13 1800

Loss 0.0 0.0 0.0 21.0903 0.0
Fuel cost (reported) 17,963.9401 17,969.01 17,960.37 17,915.64 17,963.83
Exact fuel cost 17,964.13 17,969.02 17,964.06 18,954.27 17,963.83

that the exact fuel cost obtained by MSFLA & GA [15] method is 4
Convergence characteristics of 13 -units system
x 10
much higher in comparison to all other methods reported in this 2.58
table. The reported fuel cost and exact fuel cost is almost same
in case of DEC-SQP [23], QPSO [18] and proposed OIWO methods. 2.56

Total fuel cost generations ($/hr)

However, there is little difference in both costs in case of ICA-
PSO [20] and large difference in MSFLA & GA [15] method as in 2.54
Table 1a. It can be found that the proposed OIWO method provides
cheapest generation schedule as compared to other meta-heuristic 2.52
IWO
algorithms.
The obtained results with losses by proposed OIWO algorithm 2.5
are compared with the best solutions of BBO [25], DE/BBO [26],
2.48 OIWO
SDE [22], different version of PSO [20] and ORCCRO [27] methods
in Table 1b. There is power mismatch of −0.93 MW by ORCCRO [27]
method and +0.052 MW by SDE [26] method reported in Table 1b. 2.46
The negative and positive signs in power mismatch represent the
deficit and surplus of power, respectively. So, the two above meth- 2.44
0 20 40 60 80 10 0
ods do not satisfy the power balance equation.
Number of iteration s
The convergence characteristics of 13-generators system with
transmission losses by IWO and OIWO methods are shown in Fig. 2. Fig. 2. Convergence characteristic of 13-generators system with losses.
In order to verify the robustness of the proposed algorithm, the
statistical performance over 50 runs are presented in Table 2.
demonstrate the proposed algorithm outperforms the other meth-
The minimum, average and maximum costs obtained by differ-
ods in terms of better and consistency in getting optimal solution.
ent methods over 50 trials are depicted in Table 2. The results

5.1.2. Test system-2

Table 1b A system with 40 thermal units with valve point loading effect
Best power output for 13-generator system with losses (PD = 2520 MW). has been considered. The system load demand is 10,500 MW and
Unit OIWO ORCCRO [27] SDE [22]
its input data are provided in Appendix B. The transmission loss has
been neglected in case-1 and then considered in case-2. The B loss
1 628.3185 628.32 628.32
coefficients of this system are multiplication of rows and column
2 299.1989 299.20 299.20
3 299.1991 299.20 299.20 of 6 generating units up to 40 units and also provided in Appendix
4 159.7331 159.73 159.73 B. The results obtained by proposed OIWO algorithm are provided
5 159.7331 159.73 159.73 in Table 3a for case-1, whereas case-2 in Table 3b.
6 159.7331 159.73 159.73
The results without losses are compared with modified shuf-
7 159.7330 159.73 159.73
8 159.7331 159.73 159.73
fled frog leaping algorithm with GA crossover (MSFLA and GA) [15],
9 159.7330 159.73 159.73 quantum-inspired particle swarm optimization (QPSO) [18], hybrid
10 77.3953 77.40 77.40 quantum mechanics inspired particle swarm optimization (HQPSO)
11 113.1079 112.14 113.12 [19] improved coordinated aggregation-based PSO (ICA PSO) [20],
12 92.3594 92.40 92.40
chaotic differential evolution and sequential quadratic program-
13 92.3911 92.40 92.40
ming (DEC SQP) [23] and BBO [25] are provided in Table 3a. In this
Total power output 2560.3686 2559.44 2560.44 table, the reported fuel cost and exact fuel cost are almost same in
Loss MW (reported) 40.3686 39.43 40.43
Exact loss MW 40.3686 40.37 40.367
case of QPSO [18], BBO [25] and proposed OIWO methods. How-
Mismatch of power 0.00 −0.93 +0.052 ever, there is a little difference in both costs in case of DEC-SQP
Load demand PD 2520 2519.07 2520.052 [23] and ICA-PSO [20] and large difference in case of HQPSO [19]
Fuel cost (reported) 24,514.83 24,513.91 24,514.88 and MSFLA & GA [15] methods. Similarly, the result obtained by
Exact fuel cost 24,514.83 24,513.99 24,514.95
proposed OIWO algorithm with losses has been compared with
128 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

Table 2
Comparison between different methods taken after 50 trials (13-generator system with losses).

Methods Generation costs Time (s) No. of hits to minimum solution

Max. Min. Average

OIWO 24,514.83 24,514.83 24,514.83 5.3 47

ORCCRO 24,513.91 24,513.91 24,513.91 8 50
SDE NA 24,514.88 24,516.31 NA NA
ICA-PSO 24,589.45 24,540.06 24,561.46 10.4 NA
STHDE NA 24,560.08 NA NA NA
BBO 24,516.09 24,515.21 24,515.32 15 44
DE/BBO 24,515.98 24,514.97 24,515.05 11 46

GAAPI [5], BBO [25], DE/BBO [26], SDE [22] and ORCCRO [27]. The and can also be taken from [6]. The load demand is considered as
best generation cost and corresponding generation level of gen- 15,000 MW.
erators are provided in Table 3b. The convergence characteristic The minimum fuel cost of 197,989.14 $/h and their correspond-
of 40 generator system using IWO and OIWO are shown in Fig. 3. ing generation levels obtained by proposed OIWO method are pro-
The minimum, average and maximum cost obtained by proposed vided in Table 5. The minimum fuel cost obtained so far for this sys-
algorithm over 50 trials are presented in Table 4. tem by ORCCRO [27] method is found to be 198,016.29 $/h provided
in Table 6. From Tables 5 and 6, it is clear that the proposed method
provides cheapest generation schedule of power generation.
5.1.3. Test system-3
This system comprising of 110 generators with quadratic cost The minimum, average and maximum fuel cost obtained using
characteristic. The input system data are provided in Appendix C OIWO, ORCCRO [27], SAB [7], SAF [7], SA [7], BBO [27], DE/BBO

Table 3a
Best power output for 40-generator system without losses (PD = 10,500 MW).

Units Power outputs in MW

HQPSO [19] DEC-SQP [23] QPSO [18] BBO [25] ICA-PSO [20] MSFLA & GA [15] OIWO (proposed)

1 114.00 111.76 111.20 111.0645 110.8 113.1389 110.7998

2 114.00 111.56 111.70 111.5915 110.8 102.2241 110.7999
3 104.88 97.40 97.40 97.6008 97.41 115.3426 97.3999
4 179.80 179.73 179.73 179.7095 179.94 178.1223 179.7331
5 93.04 91.66 90.14 88.30605 88.52 96.235 87.7999
6 139.84 140.00 140.00 139.9992 140.00 138.543 140.0000
7 260.06 300.00 259.60 259.6313 259.60 267.8913 259.5997
8 287.04 300.00 284.80 284.7366 284.60 274.6712 284.5997
9 276.10 284.60 284.84 284.7801 284.60 298.2346 284.5996
10 130.00 130.00 130.00 130.2484 130.00 132.1237 130.0000
11 170.42 168.80 168.80 168.8461 168.80 171.1123 94.0000
12 164.77 94.00 168.80 168.8239 94.00 161.0985 94.0000
13 223.00 214.76 214.76 214.7038 214.76 197.0112 214.7599
14 306.58 394.28 304.53 304.5894 394.28 308.4523 394.2793
15 395.04 304.52 394.28 394.2761 394.28 390.1342 394.2794
16 390.29 304.52 394.28 394.2409 304.52 385.3210 394.2794
17 479.94 489.28 489.28 489.2919 489.28 487.2341 489.2794
18 489.67 489.28 489.28 489.4188 489.28 497.0912 489.2794
19 509.01 511.28 511.28 511.2997 511.28 507.6211 511.2794
20 512.07 511.28 511.28 511.3073 511.28 534.2314 511.2794
21 521.22 523.28 523.28 523.4170 523.28 521.1223 523.2794
22 523.60 523.28 523.28 523.2795 523.28 514.1553 523.2794
23 526.50 523.28 523.29 523.3793 523.29 527.0123 523.2794
24 519.19 523.28 523.28 523.3225 523.28 548.0786 523.2794
25 523.27 523.28 523.29 523.3661 523.29 521.2314 523.2794
26 519.23 523.28 523.28 523.4362 523.28 518.5436 523.2794
27 10.00 10.00 10.01 10.05316 10.00 13.1213 10.0000
28 10.00 10.00 10.01 10.01135 10.00 18.1220 10.0000
29 10.00 10.00 10.00 10.00302 10.00 13.1563 10.0000
30 89.01 90.33 88.47 88.47754 96.39 91.1543 87.7999
31 186.03 190.00 190.00 189.9983 190 179.1452 190.0000
32 188.50 190.00 190.00 189.9881 190 182.1231 190.0000
33 190.00 190.00 190.00 189.9663 190 186.1211 190.0000
34 174.80 200.00 164.91 164.8054 164.82 185.1221 164.7998
35 169.07 200.00 165.36 165.1267 200.00 164.0132 194.3976
36 163.56 200.00 167.19 165.7695 200.00 168.0673 200.0000
37 100.88 110.00 110.00 109.9059 110.00 99.1430 110.0000
38 110.00 110.00 107.01 109.9971 110.00 97.1232 110.0000
39 110.00 110.00 110.00 109.9695 110.00 107.5623 109.9999
40 515.59 511.28 511.36 511.2794 511.28 512.2314 511.2794

Total power 10,500 10,500 10,500 10,500 10,500 10,521.95 10,500

Loss MW 0.0 0.0 0.0 0.0 0.0 21.95 0.0
Fuel cost (reported) 121,318.60 121,741.97 121,448.21 121,426.95 121,413.20 121,263.48 121,412.54
Actual fuel cost 122,318.60 121,532.14 121,447.75 121,426.95 121,422.16 123,818.06 121,412.54
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 129

Table 3b
Best power output for 40-generator system with losses (PD = 10,500 MW).

Unit Power outputs in MW

OIWO ORCCRO [27] SDE [22] GAAPI BBO [25] DE/BBO [26]

1 113.9908 111.68 110.06 114 112.54 111.04

2 114.0000 112.16 112.41 114 113.22 113.71
3 119.9977 119.98 120.0 120 119.51 118.64
4 182.5131 182.18 188.72 190 188.37 189.49
5 88.4227 87.28 85.91 97 90.41 86.32
6 140.0000 139.85 140.0 140 139.05 139.88
7 299.9999 298.15 250.19 300 294.97 299.86
8 292.0654 286.89 290.68 300 299.18 285.42
9 299.8817 293.38 300 300 296.46 296.29
10 279.7073 279.34 282.01 205.25 279.89 285.07
11 168.8149 162.35 180.82 226.3 160.15 164.69
12 94.0000 94.12 168.74 204.72 96.74 94
13 484.0758 486.44 469.96 346.48 484.04 486.30
14 484.0477 487.02 484.17 434.32 483.32 480.70
15 484.0396 483.39 487.73 431.34 483.77 480.66
16 484.0886 484.51 482.30 440.22 483.30 485.05
17 489.2813 494.22 499.64 500 490.83 487.94
18 489.2966 489.48 411.32 500 492.19 491.09
19 511.3219 512.20 510.47 550 511.28 511.79
20 511.3350 513.13 542.04 550 521.55 544.89
21 549.9412 543.85 544.81 550 526.42 528.92
22 549.9999 548.00 550 550 538.30 540.58
23 523.2804 521.21 550 550 534.74 524.98
24 523.3213 525.01 528.16 550 521.20 524.12
25 523.5804 529.84 524.16 550 526.14 534.49
26 523.5847 540.04 539.10 550 544.43 529.15
27 10.0086 12.59 10 11.44 11.51 10.51
28 10.0068 10.06 10.37 11.56 10.21 10
29 10.0123 10.79 10 11.42 10.71 10
30 87.8664 89.70 96.10 97 88.28 90.06
31 190.0000 189.59 185.33 190 189.84 189.82
32 189.9983 189.96 189.54 190 189.94 187.69
33 190.0000 187.61 189.96 190 189.13 189.97
34 199.9940 198.91 199.90 200 198.07 199.83
35 200.0000 199.98 196.25 200 199.92 199.93
36 164.8283 165.68 185.85 200 194.35 163.03
37 110.0000 109.98 109.72 110 109.43 109.85
38 109.9940 109.82 110 110 109.56 109.26
39 110.0000 109.88 95.71 110 109.62 109.60
40 550.0000 548.50 532.43 550 527.82 543.23

Total power 11,457.2965 11,458.75 11,474.43 11,545.06 11,470 11,457.83

Loss MW 957.2965 958.75 974.43 1045.06 970.37 957.83
Fuel cost 136,452.677 136,855.19 138,157.46 139,864.96 137,026.82 136,950.77

[27] are shown in Table 6. The convergence characteristic curve The best generation schedule obtained using OIWO algorithm
obtained using IWO and OIWO are depicted in Fig. 4. is reported in Table 7. The convergence characteristics of 140 gen-
erators system obtained by IWO and OIWO methods are shown in
Fig. 5. The minimum, average and maximum fuel costs obtained by
5.1.4. Test system-4 OIWO and SDE [22] over 50 trials are presented in Table 8.
A power system of Korea having 140 generating units with valve
point loading effects is taken from the literature [21] as test system 5.1.5. Test system-5
4. The system comprising of 140 thermal generating units where A complex system with 160 thermal units with multiple fuel
the hydro and pump storage plants are not considered. 12 genera- options and valve point loading effect is considered here. The sys-
tors have the cost function with valve point loading effects. In this tem load demand is 43,200 MW. The input data are provided in
case, ramp rate limits and prohibited operating zones are not con- Appendix D. The input data of 10 units [9,16] are replicated up to
sidered. The transmission losses are neglected for this test system. 160 units and also given in Appendix D. The transmission loss is
The input data of fuel cost is available in [21]. The total load demand not included in the cost function. The cheapest generation sched-
is set to 49,342 MW. ule obtained using OIWO algorithm is presented in Table 9. The

Table 4
Comparison maximum, average and minimum value taken after 50 trials (40-generators system with losses).

Methods Generation costs Time (s) No. of hits to minimum solution

Max. Average Min.

OIWO 136,452.68 136,452.68 136,452.68 10.7 46

ORCCRO 136,855.19 136,855.19 136,855.19 14 50
BBO 137,587.82 137,116.58 137,026.82 40 41
DE/BBO 137,150.77 136,966.77 136,950.77 32 45
130 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

5 4 Convergence Characteristics of 160-units system

x 10 Con vergence characteristics of 40-units system x 10
1.65 1.04

Total Fuel Cost generations ($/hr)

Total Fuel Cost generations ($/hr)

1.6 1.03

1.55
1.02
OIWO
IWO
1.5
OIWO 1.01
IWO
1.45
1
1.4
0.99
1.35 0 100 200 300 400 500 600
0 50 10 0 15 0 Number of iterations
Number of iteration s
Fig. 6. Convergence characteristic of 160-generators system.
Fig. 3. Convergence characteristic of 40-generators system with losses.

is due to the inherent randomness of the fast seed reproduction by

5 Convergence Characteristics of 110-units system
x 10 weeds without mating, independently, allowing all weeds to partic-
2.4 ipate in reproduction process and quasi opposition based learning.
The algorithm is based on the invasive habits of growth of weeds
Total Fuel Cost generations ($/hr)

2.3 in nature and having excellent exploration and exploitation ability

in the search area to provide cheapest generation schedule.
2.2
OIWO 5.2. Tuning of control parameters
IWO
2.1
The three important parameters affect the convergence of the
algorithm that are the initial SD ( 0 ), the final SD ( f ), and the non-
2 linear modulation index (n). The mostly preferred method used to
select the control parameters is parameter tuning after a series of
1.9 experiments done. The modulation index should be tuned care-
0 50 100 150 200 250 fully in order to achieve the proper value of the standard deviation
Number of iterations in each iteration. A high initial standard deviation should be chosen
to allow the algorithm to explore the whole search space, aggres-
Fig. 4. Convergence characteristic of 110-generators system.
sively. It seems that the IWO works well if the  0 is set around a few
percent (1–5 percent) of the dynamic range of each variable. The
6 Con vergence Characteristics of 140-units system final SD should be selected carefully to allow the optimizer to find
x 10
1.9 the optimal solution as accurate as possible. A finer local optimum
solution can be achieved by decreasing this parameter. However, it
Total Fuel Cost generations (S/hr)

1.85
should be noticed that tuning the final SD much smaller than the
1.8 precision criteria of the optimization variables, does not improve
the final error level and may deteriorate the convergence rate of the
1.75 OIWO optimization. Therefore, the final SD in each dimension should be
IWO selected based on the precision effect of that variable on the objec-
1.7
tive function. It was shown that the value of nonlinear modulation
1.65 index has a considerable effect on the performance of OIWO. It was
suggested [40] that the best choice for modulation index is 3. It has
1.6 been noted that control parameters have great influences on the
1.55 performance of heuristic optimization technique.
0 100 200 300 400 500 The tuning of the variables like initial standard deviation is var-
Number of iteration s ied from 0.1 to 0.9 with an increment of 0.1, the maximum number
of seeds are also varied from 8 to 1 by a step of 1, the jumping
Fig. 5. Convergence characteristic of 140-generators system. rate is varied from 0.1 to 0.6 with a step of 0.1 and also the final
standard deviation, modulation index, population size are varied
convergence characteristic of 160 generators system obtained by with different values possible combinations. For different combi-
IWO and OIWO are shown in Fig. 6. The minimum, average and nation of parameters the minimum fuel cost of 160 generator test
maximum fuel costs obtained by OIWO, ORCCRO [27], ED-DE [10], system is evaluated. At a time one single parameter is varied to get
and different GA [10] methods, BBO [25], DE/BBO [26] over 50 trials the minimum cost by keeping other parameters fixed within their
are presented in Table 10. prescribed limits. To avoid the local minima and to enhance the
This IWO algorithm gives a chance to the less fit plants to efficiency of convergence property a perturbation such as OBL is
reproduce and if the seeds produced by them have good finesses given in search space to improve the candidate solution. Owing to
in the colony, they can survive. The OIWO algorithm employs space limitation, it is not possible to present all the tuning results
quasi-oppositional based population initialization and generation of possible combinations of parameters in tabular form. However,
jumping to accelerate the convergence rate of IWO algorithm. This the tuning of population size on the performance on 160 units
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 131

Table 5
Best power output for 110-generator system (PD = 15,000 MW).

Unit Power Unit Power Unit Power Unit Power Unit Power
output MW output MW output MW output MW output MW

1 2.4000 24 350.0000 47 5.4004 70 360.0000 93 440.0000

2 2.4004 25 400.0000 48 5.4000 71 400.0000 94 499.9670
3 2.4026 26 400.0000 49 8.4015 72 400.0000 95 600.0000
4 2.4000 27 499.9994 50 8.4000 73 107.8321 96 469.2687
5 2.4000 28 500.0000 51 8.4000 74 188.8113 97 3.6000
6 4.0011 29 199.9905 52 12.0000 75 89.9993 98 3.6000
7 4.0000 30 99.9997 53 12.0000 76 49.9990 99 4.4000
8 4.0000 31 10.0009 54 12.0011 77 160.0146 100 4.4005
9 4.0000 32 19.9898 55 12.0003 78 291.3563 101 10.0077
10 63.0551 33 79.4851 56 25.2000 79 176.9965 102 10.0010
11 59.2747 34 250.0000 57 25.2000 80 97.7532 103 20.0046
12 35.6585 35 359.9995 58 35.0000 81 10.0005 104 20.0053
13 57.4385 36 399.9920 59 35.0000 82 12.3053 105 40.0000
14 25.0000 37 39.9995 60 45.0009 83 20.0423 106 40.0017
15 25.0000 38 69.9814 61 45.0009 84 199.9858 107 50.0000
16 25.0005 39 99.9942 62 45.0000 85 324.5127 108 30.0000
17 155.0000 40 120.0000 63 184.9958 86 439.9894 109 40.0000
18 155.0000 41 156.8006 64 184.9933 87 18.8673 110 20.0001
19 155.0000 42 219.9998 65 184.9989 88 23.3342 Fuel cost ($/h) = 197,989.1358
20 154.9959 43 440.0000 66 185.0000 89 84.4027
21 68.9000 44 560.0000 67 70.0000 90 91.9002
22 68.9000 45 660.0000 68 70.0000 91 58.2896
23 68.9000 46 619.5259 69 70.0015 92 98.0707

Table 6
Comparison between different methods taken after 50 trials (110-generators system).

Methods Generation costs Time (s) No. of hits to minimum solution

Max. Average Min.

OIWO 197,989.93 197,989.41 197,989.14 31 46

ORCCRO [27] 198,016.89 198,016.32 198,016.29 45 48
SAB [7] NA 207,764.73 206,912.9057 NA NA
SAF [7] NA 207,813.37 207,380.5164 NA NA
SA [7] NA 201,595.19 198,352.6413 NA NA
BBO [27] 199,102.59 198,413.45 198,241.166 115 41
DE/BBO [27] 198,828.57 198,326.66 198,231.06 132 43

Table 7
Best power output for 140-generator system (PD = 49,342 MW).

Unit Power output MW Unit Power output MW Unit Power output MW Unit Power output MW Unit Power output MW

1 113.3601 29 500.9999 57 103.0000 85 115.000 113 94.0000

2 189.0000 30 501.0000 58 198.0000 86 207.0000 114 94.0000
3 189.9999 31 506.0000 59 312.0000 87 207.0000 115 244.0005
4 190.0000 32 505.9998 60 281.3455 88 175.0018 116 244.0003
5 168.5486 33 505.9993 61 163.0000 89 175.0000 117 244.0000
6 190.0000 34 506.0000 62 95.0000 90 175.1042 118 95.0002
7 490.0000 35 500.0000 63 160.0000 91 175.0000 119 95.0000
8 489.9995 36 500.0000 64 160.0005 92 580.0000 120 116.0003
9 495.9965 37 241.000 65 490.0000 93 645.0000 121 175.0000
10 496.0000 38 241.0000 66 196.5687 94 983.9999 122 2.0000
11 496.0000 39 774.0000 67 489.2261 95 978.0000 123 4.0000
12 496.0000 40 769.0000 68 490.0000 96 681.9990 124 15.0000
13 506.0000 41 3.0000 69 130.0000 97 719.9998 125 9.0000
14 509.0000 42 3.0000 70 234.7119 98 717.9999 126 12.0000
15 506.0000 43 249.9996 71 137.0009 99 720.0000 127 10.0000
16 505.0000 44 248.5722 72 325.5000 100 964.0000 128 112.0001
17 506.0000 45 249.9987 73 195.1150 101 958.0000 129 4.0000
18 506.0000 46 250.0000 74 175.0377 102 1007.000 130 5.0000
19 505.0000 47 240.3950 75 175.0000 103 1005.9991 131 5.0000
20 505.0000 48 250.0000 76 175.0000 104 1012.9995 132 50.0000
21 505.0000 49 250.0000 77 175.0000 105 1020.0000 133 5.0000
22 504.9999 50 250.000 78 330.1887 106 954.0000 134 42.0000
23 505.0000 51 165.0000 79 530.9990 107 952.0000 135 42.0000
24 505.0000 52 165.0000 80 531.0000 108 1006.0000 136 41.0000
25 536.9999 53 165.0001 81 397.3314 109 1013.0000 137 17.0004
26 537.0000 54 165.0000 82 56.0000 110 1021.0000 138 7.0000
27 549.0000 55 180.0000 83 115.0001 111 1015.0000 139 7.0000
28 548.9997 56 180.0000 84 115.0004 112 94.0003 140 26.0000
Fuel cost ($/h) = 1,559,405.4669 Load demand = 49,342 MW
132 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

Table 8
Comparison between different methods taken after 50 trials (140-generators system).

Methods Generation costs Time (s) No. of hits to minimum solution

Max. Min. Average

OIWO 1,559,405.88 1,559,405.45 1,559,405.61 46.8 46

SDE [22] NA 1,560,236.85 NA NA NA

Table 9
Best power output for 160-generator system (PD = 43,200 MW).

Unit Power Unit Power Unit Power Unit Power Unit Power Unit Power
output MW output MW output MW output MW output MW output MW

1 217.5691 28 240.4456 55 276.5746 82 211.2162 109 430.0674 136 239.9082

2 211.2162 29 430.0674 56 239.9082 83 279.6483 110 279.0141 137 285.3795
3 279.6483 30 279.0140 57 285.3795 84 240.1770 111 217.5692 138 240.4457
4 240.1770 31 217.5693 58 240.4457 85 276.5745 112 211.2162 139 430.0674
5 276.5746 32 211.2163 59 430.0674 86 239.9082 113 279.6482 140 279.0141
6 239.9082 33 279.6484 60 279.0140 87 285.3795 114 240.1770 141 217.5692
7 285.3794 34 240.1770 61 217.5693 88 240.4457 115 276.5745 142 211.2162
8 240.4457 35 276.5746 62 211.2162 89 430.0674 116 239.9082 143 279.6482
9 430.0674 36 239.9082 63 279.6483 90 279.0140 117 285.3795 144 240.1770
10 279.0139 37 285.3795 64 240.1770 91 217.5692 118 240.4457 145 276.5744
11 217.5692 38 240.4457 65 276.5744 92 211.2162 119 430.0674 146 239.9082
12 211.2162 39 430.0674 66 239.9082 93 279.6484 120 279.0140 147 285.3795
13 279.6484 40 279.0139 67 285.3796 94 240.1770 121 217.5692 148 240.4457
14 240.1770 41 217.5692 68 240.4457 95 276.5745 122 211.2162 149 430.0675
15 276.5754 42 211.2163 69 430.0674 96 239.9082 123 279.6483 150 279.0139
16 239.9083 43 279.6482 70 279.0140 97 285.3795 124 240.1769 151 217.5692
17 285.3795 44 240.1770 71 217.5692 98 240.4457 125 276.5746 152 211.2162
18 240.4457 45 276.5745 72 211.2161 99 430.0674 126 239.9082 153 279.6482
19 430.0674 46 239.9082 73 279.6482 100 279.0141 127 285.3794 154 240.1770
20 279.0140 47 285.3796 74 240.1770 101 217.5692 128 240.4457 155 276.5746
21 217.5692 48 240.4457 75 276.5743 102 211.2162 129 430.0674 156 239.9082
22 211.2162 49 430.0674 76 239.9082 103 279.6484 130 279.0140 157 285.3796
23 279.6484 50 279.0141 77 285.3795 104 240.1770 131 217.5692 158 240.4457
24 240.1770 51 217.5692 78 240.4457 105 276.5744 132 211.2162 159 430.0669
25 276.5745 52 211.2162 79 430.0674 106 239.9082 133 279.6482 160 279.0141
26 239.9082 53 279.6484 80 279.0140 107 285.3795 134 240.1770 Fuel cost = 9981.9834 ($/h)
27 285.3796 54 240.1770 81 217.5691 108 240.4457 135 276.5745

system is shown in Table 11. When the population size increases, changing the parameters one at a time. It is found that the nonlinear
there is no significant improvement of fuel cost of the system rather modulation index value of 3 gives better execution time. The initial
the execution time has been increased. The appropriate population standard deviation value has been varied to get the variation of exe-
size of this system is found to be 20. To validate the convergence cution time for the 160 units test system. Similarly, final standard
effect of the tuning of control parameters such as the value of deviation value varies from lower to higher, but an intermediate
initial SD, final SD and nonlinear modulation index are being var- chosen value gives better performance. Moreover, a few selected
ied. The different combinations of parameters are experimented by combinations of parameters are only shown in Tables 12a and 12b.

Table 10
Comparison between different methods taken after 50 trials (160-generators system).

Methods Generation costs Time (s) No. of hits to minimum solution

Max. Min. Average

OIWO 9983.998 9981.9834 9982.991 17.3 46

ORCCRO [27] 10,004.45 10,004.20 10,004.21 19 48
ED-DE [10] NA 10,012.68 NA NA NA
CGA-MU [10] NA 10,143.73 NA NA NA
IGA-MU [10] NA 10,042.47 NA NA NA
BBO [27] 10,010.59 10,008.71 10,009.16 44 40
DE/BBO [27] 10,010.25 10,007.05 10,007.56 35 42

Table 11
Effect of population size on 160 generator system.

Popmax No of hits to best solution Simulation time Maximum cost Minimum cost Average cost

5 31 10.43 10,003.53 10,001.33 10,002.61

10 38 12.07 10,002.05 10,000.03 10,001.42
15 40 14.48 9984.32 9982.81 9983.25
20 44 17.32 9983.45 9981.98 9982.64
25 41 18.15 9983.17 9982.87 9982.39
30 38 21.68 9983.66 9982.87 9982.02
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 133

Table 12a
Effect of various parameters on performance of OIWO (minimum fuel cost $/h obtained for test case-5).

Smax Smin n jr f 0

0.1 0.2 0.4 0.5 0.6 0.7 0.8 0.9

8 0 3 0.1 0.01 9984.82 9984.27 9983.80 9983.03 9983.68 9983.67 9983.67 9983.73
7 0 3 0.2 0.05 9983.90 9983.54 9982.14 9982.87 9982.42 9982.668 9983.55 9982.96
6 0 3 0.2 0.0015 9983.12 9983.95 9982.42 9982.84 9982.65 9982.191 9982.27 9982.31
5 0 3 0.3 0.001 9982.81 9982.04 9982.78 9982.64 9982.11 9881.98 9981.99 9982.95
4 0 2 0.3 0.002 9983.63 9983.15 9983.91 9983.93 9982.17 9982.347 9982.04 9982.03
3 0 2 0.4 0.10 9984.09 9984.48 9983.79 9983.67 9982.70 9982.884 9982.82 9982.43
2 0 2 0.5 0.15 9985.22 9984.92 9984.88 9984.69 9983.76 9983.651 9983.71 9983.76
1 0 2 0.6 0.2 9984.57 9984.48 9984.65 9984.39 9984.03 9983.842 9984.69 9985.238

Table 12b
Effect of various parameters on performance of OIWO (execution time in second obtained for test case-5).

n f 0

0.1 0.2 0.4 0.5 0.6 0.7 0.8 0.9

3 0.0005 17.70 17.82 18.00 17.57 17.83 17.77 18.03 18.15

3 0.0003 17.34 17.64 17.81 17.56 17.94 17.71 17.83 18.29
3 0.0015 17.72 17.62 17.53 18.10 17.77 17.64 18.19 17.92
3 0.001 17.53 17.48 17.69 17.76 17.33 17.31 17.87 17.72
2 0.001 17.58 17.69 18.01 17.81 17.91 17.62 18.47 17.85
3 0.01 17.78 17.95 17.73 17.98 18.40 17.69 17.99 17.75
2 0.01 17.88 18.04 17.96 17.85 17.83 17.95 17.97 17.87
2 0.10 18.18 18.37 18.06 18.19 18.48 18.42 18.38 17.98
3 0.15 18.71 18.53 18.26 18.59 18.68 18.82 18.35 18.77
2 0.2 18.19 18.41 18.74 18.51 18.68 18.22 18.59 18.68

Table 13
Optimal parameter setting of different test cases.

Test systems Parameters

Smax Smin n jr 0 f Popmax

13 unit 3 0 3 0.2 0.6 0.0015 10

40 unit 4 0 3 0.2 0.6 0.0015 10
110 unit 5 0 3 0.3 0.7 0.001 16
140 unit 5 0 3 0.3 0.7 0.001 20
160 unit 5 0 3 0.3 0.7 0.001 20

Brief summarized results of all test system are provided in shown in Tables 2, 4, 6, 8 and 10. The OIWO method takes mini-
Table 13. mum time to obtain minimum fuel costs which are presented in
Tables 2, 4, 6, 8 and 10. From the above tables, out of 50 num-
5.3. Comparative study bers of trials for five different test systems, OIWO reaches to the
minimum costs 47, 46, 46, 46 and 46, respectively. Moreover, the
5.3.1. Solution quality and robustness computational efficiencies of OIWO are 94% for first system and 92%
Tables 1, 3, 5, 7 and 9 present the minimum fuel cost produced for remaining four large systems. The convergence characteristics
by OIWO algorithm for 5 different test systems. The proposed OIWO OIWO are very smooth due to the inherent randomness involved
provides the cheapest generation schedule in comparison to other in the optimization process and also the use of oppositional based
existing techniques reported in literature. In order to show the learning into proposed algorithm.
robustness of OIWO, the minimum, average and maximum values
6. Conclusion
of fuel cost for all cases are compared with different methods. These
are reported in Tables 2, 4, 6, 8 and 10 and emphasizing consistency In this paper, a newly developed OIWO algorithm has been pro-
in achieving the good result and better in terms of quality of solu- posed for non-convex, large scale economic load dispatch problems
tions obtained by OIWO algorithm. It can be seen from Figs. 2–6 solution. The feasibility and robustness of the proposed algorithm
that the OIWO has faster convergence than IWO method because of have been investigated on five test systems having 13, 40, 110
opposition based learning. This is due to the inherent randomness 140 and 160-units. The analysis of simulation results reveal that
of the fast seed reproduction by weeds without mating, indepen- the proposed method succeeded in achieving the cheapest genera-
dently, allowing all weeds to participate in reproduction process tion schedule in all cases. The opposition based learning with IWO
and quasi opposition based learning. makes it faster in convergence characteristics. The performance of
the OIWO algorithm is better in comparison to other established
5.3.2. Computational efficiency methods reported in literature in terms of convergence rate, qual-
The execution time generally increases for large scale systems ity and consistency in the solution. In future, attempts can be made
using any algorithm. The average CPU time by the proposed OIWO to implement this proposed methodology in solving complex real
method is smaller in comparison to other established methods are world power system problems.
134 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

Appendix A.

Table A.1
Cost coefficient and generation limits of 13 unit system.

Unit Pgmin MW Pgmax MW ai bi ci ei fi

1 0 680 0.00028 8.10 550 300 0.035

2 0 360 0.00056 8.10 309 200 0.042
3 0 360 0.00056 8.10 307 150 0.042
4 60 180 0.00324 7.74 240 150 0.063
5 60 180 0.00324 7.74 240 150 0.063
6 60 180 0.00324 7.74 240 150 0.063
7 60 180 0.00324 7.74 240 150 0.063
8 60 180 0.00324 7.74 240 150 0.063
9 60 180 0.00324 7.74 240 150 0.063
10 40 120 0.00284 8.6 126 100 0.084
11 40 120 0.00284 8.6 126 100 0.084
12 55 120 0.00284 8.6 126 100 0.084
13 55 120 0.00284 8.6 126 100 0.084

Table A.2
The B loss coefficients matrix of a 13 unit system with a base capacity of 100 MVA is as follows.
⎡ 0.0014 0.0012 0.0007 −0.0001 −0.0003 −0.0001 −0.0001 −0.0001 −0.0003 0.0005 −0.0003 −0.0002 0.0004
⎤
0.0012 0.0015 0.0013 0 −0.0005 −0.0002 0 0.0001 −0.0002 −0.0004 −0.0004 0 0.0004
⎢ 0.0007 0.0013 0.0076 −0.0001 −0.0013 −0.0009 −0.0001 ⎥
⎢ 0 −0.0008 −0.0012 −0.0017 0 −0.0026 ⎥
⎢ −0.0001 0 −0.0001 0.0034 −0.0007 −0.0004 0.0011 0.0050 0.0029 0.0032 −0.0011 0 0.0001 ⎥
⎢ −0.0003 −0.0005 −0.0013 −0.0007 0.0090 0.0014 −0.0003 −0.0012 −0.0010 −0.0013 0.0007 −0.0002 −0.0002 ⎥
⎢ ⎥
⎢ −0.0001 −0.0002 −0.0009 −0.0004 0.0014 0.0016 0 −0.0006 −0.0005 −0.0008 0.0011 −0.0001 −0.0002 ⎥
= ⎢ −0.0001 −0.0001 0.0011 −0.0003 −0.0005 0.0007 ⎥
B13×13
⎢ −0.0001 0.0001
0 0 0.0015 0.0017 0.0015 0.0009 0
⎥
⎢ 0 0.0050 −0.0012 −0.0006 0.0017 0.0168 0.0082 0.0079 −0.0023 −0.0036 0.0001 ⎥
⎢ −0.0003 −0.0002 −0.0008 0.0029 −0.0010 −0.0005 0.0015 0.0082 0.0129 0.0116 −0.0021 −0.0025 0.0007 ⎥
⎢ −0.0005 −0.0004 −0.0012 0.0032 −0.0013 −0.0008 0.0009 0.0079 0.0116 ⎥
⎢ 0.0200 −0.0027 −0.0034 0.0009 ⎥
⎣ −0.0003 −0.0004 −0.0017 −0.0011 0.0007 0.0011 −0.0005 −0.0023 −0.0021 −0.0027 0.0140 0.0001 0.0004 ⎦
−0.0002 0 0 0 −0.0002 −0.0001 0.0007 −0.0036 −0.0025 −0.0034 0.0001 0.0054 −0.0001
 0.0004 0.0004 −0.0026 0.0001 −0.0002 −0.0002 0 0.0001 0.0007 0.0009 
0.0004 −0.0001 0.0103
B013 = −0.0001 −0.0002 0.0028 −0.0001 0.0001 −0.0003 −0.0002 −0.0002 0.0006 0.0039 −0.0017 0 −0.0032
B0013 = 0.0055

Appendix B.

Table B.1
Cost coefficient and generation limits of 40 unit system.

Unit Pgmin MW Pgmax MW ai bi ci ei fi

1 36 114 0.00690 6.73 94.705 100 0.084

2 36 114 0.00690 6.73 94.705 100 0.084
3 60 120 0.02028 7.07 309.54 100 0.084
4 80 190 0.00942 8.18 369.03 150 0.063
5 47 97 0.01140 5.35 148.89 120 0.077
6 68 140 0.01142 8.05 222.33 100 0.084
7 110 300 0.00357 8.03 278.71 200 0.042
8 135 300 0.00492 6.99 391.98 200 0.042
9 135 300 0.00573 6.6 455.76 200 0.042
10 130 300 0.00605 12.9 722.82 200 0.042
11 94 375 0.00515 12.9 635.20 200 0.042
12 94 375 0.00569 12.8 654.69 200 0.042
13 125 500 0.00421 12.5 913.40 300 0.035
14 125 500 0.00752 8.84 1760.4 300 0.035
15 125 500 0.00708 9.15 1728.3 300 0.035
16 125 500 0.00708 9.15 1728.3 300 0.035
17 220 500 0.00313 7.97 647.85 300 0.035
18 220 500 0.00313 7.95 649.69 300 0.035
19 242 550 0.00313 7.97 647.83 300 0.035
20 242 550 0.00313 7.97 647.81 300 0.035
21 254 550 0.00298 6.63 785.96 300 0.035
22 254 550 0.00298 6.63 785.96 300 0.035
23 254 550 0.00284 6.66 794.53 300 0.035
24 254 550 0.00284 6.66 794.53 300 0.035
25 254 550 0.00277 7.10 801.32 300 0.035
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 135

Table B.1 (Continued)

Unit Pgmin MW Pgmax MW ai bi ci ei fi

28 10 150 0.52124 3.33 1055.1 120 0.077

29 10 150 0.52124 3.33 1055.1 120 0.077
30 47 97 0.01140 5.35 148.89 120 0.077
31 60 190 0.00160 6.43 222.92 150 0.063
32 60 190 0.00160 6.43 222.92 150 0.063
33 60 190 0.00160 6.43 222.92 150 0.063
34 90 200 0.00010 8.95 107.87 200 0.042
35 90 200 0.00010 8.62 116.58 200 0.042
36 90 200 0.00010 8.62 116.58 200 0.042
37 25 110 0.01610 5.88 307.45 80 0.098
38 25 110 0.01610 5.88 307.45 80 0.098
39 25 110 0.01610 5.88 307.45 80 0.098
40 242 550 0.00313 7.97 647.83 300 0.035

Table B.2
The B loss coefficients matrix of a 40 unit system with a base capacity of 100 MVA is as follows.

A loss coefficient of 40 unit system is obtained by replicating the data of 6 unit system in row and column wise as given below.
⎡ 0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 ⎤
0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001
⎢ 0.0007 0.0009 0.0031 ⎥
B6×6 = ⎢
0 −0.0010 −0.0006 ⎥
⎣ −0.0001 0.0001 0 0.0024 −0.0008 −0.0006 ⎦
−0.0005 −0.0006 −0.0010 −0.0006 0.0129 −0.0002
⎡−0.0002
⎤ ⎡ −0.0001 −0.0006 −0.0008 −0.0002 0.0150
⎤
B1 b1,1 b1,2 . . b1,40
⎢ ⎥ ⎢
B2
B40×40 = ⎣ . ⎦ = ⎣ .
. .
.
. .
. .
.
⎥
. ⎦ /100
. . . . . .
B40 b40,1 . . . b40,40
B1 = [0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017
0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017 0.0012 0.0007 −0.0001 −0.0005 −0.0002 0.0017 0.0012 0.0007
−0.0001];
B2 = [0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014
0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014 0.0009 0.0001 −0.0006 −0.0001 0.0012 0.0014 0.0009 0.0001];
B40 = [−0.0001 0.0001 0.000 0.0024 −0.0006 −0.0008 −0.0001 0.0001 0.000 0.0024 −0.0006 −0.0008 −0.0001 0.0001 0.000 0.0024 −0.0006 −0.0008 −0.0001 0.0001
0.000 0.0024 −0.0006⎡ −0.0008 −0.0001 0.0001 0.000 0.0024 −0.0006 −0.0008 −0.0001 0.0001 0.000 0.0024 −0.0006 −0.0008 −0.0001 ⎤ 0.0001 0.000 0.0024];
−0.3908 −0.1297 0.7047 0.0591 0.2161 −0.6635 −0.3908 −0.1297 0.7047 0.0591
B01,40 = 1.0e − 03 × ⎣
0.2161 −0.6635 −0.3908 −0.1297 0.7047 0.0591 0.2161 −0.6635 −0.3908 −0.1297 ⎦
0.7047 0.0591 0.2161 −0.6635 −0.3908 −0.1297 0.7047 0.0591 0.2161 −0.6635
−0.3908 −0.1297 0.7047 0.0591 0.2161 −0.6635 −0.3908 −0.1297 0.7047 0.0591
B0040 = 0.0056 ;

Appendix C.

Table C.1
Cost coefficient and generation limits of 110 unit system.

Unit Pgmax MW Pgmin MW ai bi ci Unit Pgmax MW Pgmin MW ai bi ci

1 12 2.4 0.0253 25.547 24.389 56 96 25.2 0.0098 14.327 82.136

2 12 2.4 0.0265 25.675 24.411 57 96 25.2 0.0099 14.354 82.298
3 12 2.4 0.0280 25.803 24.638 58 100 35 0.0092 14.380 82.464
4 12 2.4 0.0284 25.932 24.760 59 100 35 0.0094 14.407 82.626
5 12 2.4 0.0286 26.061 24.888 60 120 45 0.0072 19.0 218.895
6 20 4.0 0.0120 37.551 117.755 61 120 45 0.0071 19.1 219.335
7 20 4.0 0.0126 37.664 118.108 62 120 45 0.0070 19.2 219.775
8 20 4.0 0.0136 37.777 118.458 63 185 54.3 0.0066 11.694 143.735
9 20 4.0 0.0143 37.890 118.821 64 185 54.3 0.0057 11.715 144.029
10 76 15.2 0.0088 13.327 81.136 65 185 54.3 0.0058 11.737 144.318
11 76 15.2 0.0089 13.354 81.298 66 185 54.3 0.0059 11.758 144.597
12 76 15.2 0.0091 13.80 81.464 67 197 70 0.0036 24.0 269.131
13 76 15.2 0.0093 13.407 81.626 68 197 70 0.0036 24.1 269.649
14 100 25.0 0.0062 18.0 217.895 69 197 70 0.0036 24.2 270.176
15 100 25.0 0.0061 18.1 218.335 70 360 150 0.0025 11.862 187.057
16 100 25.0 0.0060 18.2 218.775 71 400 160 0.0029 8.492 320.002
17 155 54.3 0.0046 10.694 142.735 72 400 160 0.0030 8.503 321.910
18 155 54.3 0.0047 10.715 143.029 73 300 60 0.0054 13.327 52.136
19 155 54.3 0.0048 10.737 143.318 74 250 50 0.0055 12.354 42.298
20 155 54.3 0.0049 10.758 143.597 75 90 30 0.0099 11.380 32.464
21 197 68.9 0.0026 23.0 259.131 76 50 12 0.0031 9.407 23.626
22 197 68.9 0.0026 23.1 259.649 77 450 160 0.0024 14.0 220.0
23 197 68.9 0.0026 23.2 260.176 78 600 150 0.0023 13.1 190.0
24 350 140 0.0015 10.862 177.057 79 200 50 0.0036 13.2 250.0
136 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137

Table C.1 (Continued)

Unit Pgmax MW Pgmin MW ai bi ci Unit Pgmax MW Pgmin MW ai bi ci

25 400 100 0.0019 7.492 210.002 80 120 20 0.0049 13.5 230.0

26 400 100 0.0019 7.503 211.910 81 55 10 0.0061 24.0 70.0
27 500 140 0.0014 12.0 210.0 82 40 12 0.0070 14.5 60.0
28 500 140 0.0013 12.1 180.0 83 80 20 0.0088 14.2 210.0
29 200 50 0.0026 12.2 240.0 84 200 50 0.0022 13.4 150.0
30 100 25 0.0039 12.5 220.0 85 325 80 0.0048 11.3 130.0
31 50 10 0.0051 23.0 60.0 86 440 120 0.0053 8.9 80.0
32 20 5 0.0050 13.5 50.0 87 35 10 0.0021 14.4 90.0
33 80 20 0.0078 13.2 200.0 88 55 20 0.0033 14.3 80.0
34 250 75 0.0012 12.4 140.0 89 100 20 0.0034 13.9 125.0
35 360 110 0.0038 10.3 120.0 90 220 40 0.0037 13.8 160.0
36 400 130 0.0043 9.9 90.0 91 140 30 0.0066 13.7 50.0
37 40 10 0.0011 13.4 80.0 92 100 40 0.0043 13.6 400.0
38 70 20 0.0023 13.3 70.0 93 440 100 0.0022 8.4 260.0
39 100 25 0.0034 12.9 115.0 94 500 100 0.0055 7.6 110.0
40 120 20 0.0067 12.8 150.0 95 600 100 0.0032 7.5 170.0
41 180 40 0.0056 12.7 40.0 96 700 200 0.0077 7.2 140.0
42 220 50 0.0023 12.6 300.0 97 15 3.6 0.0353 26.547 26.389
43 440 120 0.0012 7.4 250.0 98 15 3.6 0.0365 26.675 25.411
44 560 160 0.0045 6.6 100.0 99 22 4.4 0.0380 26.803 25.638
45 660 150 0.0022 6.5 160.0 100 22 4.4 0.0384 26.932 25.760
46 700 200 0.0067 6.2 130.0 101 60 10 0.0210 15.3 65.0
47 32 5.4 0.0353 26.547 34.389 102 80 10 0.0230 16.0 82.0
48 32 5.4 0.0365 26.675 34.411 103 100 20 0.0240 20.2 86.0
49 52 8.4 0.0380 26.803 34.638 104 120 20 0.0350 20.2 84.0
50 52 8.4 0.0384 26.932 34.761 105 150 40 0.0340 25.6 75.0
51 52 8.4 0.0386 17.061 34.888 106 280 40 0.0370 30.5 56.0
52 60 12 0.0320 38.551 127.755 107 520 50 0.0390 32.5 67.0
53 60 12 0.0326 36.664 128.108 108 150 30 0.0350 26.0 68.0
54 60 12 0.0236 38.777 128.458 109 320 40 0.0280 25.8 69.0
55 60 12 0.0243 38.890 128.821 110 200 20 0.0260 27.0 72.0

Appendix D.

Table D.1
10 units with multi fuel options and valve point loading effect.

Unit Fuel type Pgmin (MW) Pgmax (MW) Cost coefficients

ai bi ci ei fi

1 2 196 250 0.1861e−2 −0.3059e0 0.2113e2 0.2113e−1 −0.3059e1

2 1 157 230 0.4194e−2 −0.1269e1 0.1184e3 0.1184e0 −0.1269e2
3 1 200 332 0.1457e−2 −0.3116e0 0.3979e2 0.3979e−1 −0.3116e1
4 3 200 265 0.5935e−2 −0.2338e1 0.2668e3 0.2668e0 −0.2338e2
5 1 190 338 0.1066e−2 −0.8733e−1 0.1392e2 0.1392e−1 −0.8733e0
6 3 200 265 0.5935e−2 −0.2338e1 0.2668e3 0.2668e0 −0.2338e2
7 1 200 331 0.1107e−2 −0.1325e0 0.1893e2 0.1893e−1 −0.1325e1
8 3 200 265 0.5935e−2 −0.2338e1 0.2668e3 0.2668e0 −0.2338e2
9 3 370 440 0.6121e−3 −0.1817e−1 0.1423e2 0.1423e−1 −0.1817e0
10 1 200 362 0.1102e−2 −0.9938e−1 0.1397e2 0.1397e−1 −0.9938e0

Table D.2
Input data of 160 units with multiple fuel options and valve point loading effects can be evaluated as follows: (by replicating the 10 unit data given below).

a = zeros(1,160); b = zeros(1,160); c = zeros(1,160); e = zeros(1,160); f = zeros(1,160);

Pgmax = zeros(1,160); Pgmin = zeros(1,160);
load Data 10; (input data of 10 unit system in Table D.1)
for i = 1:1:150
a(i + 10) = a(i);
b(i + 10) = b(i);
c(i + 10) = c(i);
e(i + 10) = e(i);
f(i + 10) = f(i);
Pgmax (i + 10) = Pgmax (i);
Pgmin (i + 10) = Pgmin (i);
end
 A.K. Barisal, R.C. Prusty / Applied Soft Computing 29 (2015) 122–137 137

References [22] A. Srinivasa Reddy, K. Vaisakh, Shuffled differential evolution for large scale
economic dispatch, Electr. Power Syst. Res. 96 (2013) 237–245.
[1] A.A. El-Keib, H. Ma, J.L. Hart, Environmentally constrained economic dispatch [23] L.D.S. Coelho, V.C. Mariani, Combining of chaotic differential evolution and
using the Lagrangian relaxation method, IEEE Trans. Power Syst. 9 (4) (1994) quadratic programming for economic dispatch optimization with valve-point
1723–1729. effect, IEEE Trans. Power Syst. 21 (2) (2006) 989–996.
[2] S. Fanshel, E.S. Lynes, Economic power generation using linear programming, [24] J.-P. Chiou, Variable scaling hybrid differential evolution for large-scale eco-
IEEE Trans. Power Appar. Syst. PAS-83 (4) (1964) 347–356. nomic dispatch problems, Electr. Power Syst. Res. 77 (3–4) (2007) 212–218.
[3] A.J. Wood, B.F. Wollenberg, Power Generation, Operation, and Control, 2nd ed., [25] A. Bhattacharya, P.K. Chattopadhyay, Biogeography-based optimization for dif-
John Wiley and Sons, New York, 1984. ferent economic load dispatch problems, IEEE Trans. Power Syst. 25 (2) (2010)
[4] D.C. Walters, G.B. Sheble, Genetic algorithm solution of economic dispatch with 1064–1077.
valve point loadings, IEEE Trans. Power Syst. 8 (3) (1993) 1325–1331. [26] A. Bhattacharya, P.K. Chattopadhyay, Hybrid differential evolution with
[5] I. Ciornei, E. Kyriakides, A GA-API solution for the economic dispatch of genera- biogeography-based optimization for solution of economic load dispatch, IEEE
tion in power system operation, IEEE Trans. Power Syst. 27 (1) (2012) 233–242. Trans. Power Syst. 25 (4) (2010) 1955–1964.
[6] S.O. Orero, M.R. Irving, Large scale unit commitment using a hybrid genetic [27] K. Bhattacharjee, A. Bhattacharya, S.H. Dey, Oppositional real coded chemical
algorithm, Electr. Power Energy Syst. 19 (1) (1997) 45–55. reaction optimization for different economic dispatch problems, Int. J. Electr.
[7] K.K. Vshwakarma, H.M. Dubey, Simulated annealing based optimization for Power Energy Syst. 55 (2014) 378–391.
solving large scale economic load dispatch problems, Int. J. Eng. Res. Technol. [28] H. Tizhoosh, Opposition-based learning: a new scheme for machine intel-
(IJERT) 1 (3) (2012) 1–8. ligence, in: Proceedings of the International Conference on Computational
[8] N. Sinha, R. Chakrabarti, P.K. Chattopadhyay, Evolutionary programming tech- Intelligence for Modeling Control and Automation, Austria, 28th–30th
niques for economic load dispatch, IEEE Trans. Evol. Comput. 7 (1) (2003) 83–94. November, 2005, pp. 695–701.
[9] C.E. Lin, G.L. Viviani, Hierarchical economic dispatch for piecewise quadratic [29] M. Ventresca, H. Tizhoosh, Improving the convergence of back propagation by
cost functions, IEEE Trans. Power Appar. Syst. PAS-103 (6) (1984) 1170–1175. opposite transfer functions, in: IEEE International Joint Conference on Neural
[10] Y. Wang, L. Bin, T. Weise, Estimation of distribution and differential evolu- Networks, Vancouver, BC, 16–21, July, 2006, pp. 9527–9534.
tion cooperation for large scale economic load dispatch optimization of power [30] S. Rahnamayan, H. Tizhoosh, M. Salama, Opposition-based differential evolu-
systems, Inform. Sci. 180 (12) (2010) 2405–2420. tion, IEEE Trans. Evol. Comput. 12 (1) (2008) 64–79.
[11] N. Nomana, H. Iba, Differential evolution for economic load dispatch problems, [31] H. Wang, Y. Liu, S. Zeng, H. Li, C. Li, Opposition-based particle swarm algorithm
Electr. Power Syst. Res. 78 (8) (2008) 1322–1331. with Cauchy mutation, in: IEEE Congress Evolutionary Computation, Singapore,
[12] Y.H. Hou, Y.W. Wu, L.J. Lu, X.Y. Xiong, Generalized ant colony optimization for 2007, pp. 4750–4756.
economic dispatch of power systems, in: Proceedings of International Confer- [32] A.R. Malisia, Investigating the Application of Opposition-based Ideas to Ant
ence on Power System Technology, Power-con, 13–17 October 2002 (1), 2002, Algorithms (Master’s thesis), University of Waterloo, Waterloo, ON, Canada,
pp. 225–229. 2007.
[13] B.K. Panigrahi, S.R. Yadav, S. Agrawal, M.K. Tiwari, A clonal algorithm to solve [33] M. Haiping, R. Xieyong, J. Baogen, Oppositional ant colony optimization algo-
economic load dispatch, Electr. Power Syst. Res. 77 (10) (2007) 1381–1389. rithm and its application to fault monitoring, in: Proc 29th Chinese Control
[14] B.K. Panigrahi, V.R. Pandi, Bacterial foraging optimization: Nelder–Mead hybrid Conference, 2010, pp. 3895–3898.
algorithm for economic load dispatch, IET Gener. Transm. Distrib. 2 (4) (2008) [34] D. Simon, M. Ergezer, D. Du, Oppositional biogeography-based optimization.
556–565. http://embeddedlab.csuohio.edu/BBO/
[15] P. Roy, P. Roy, A. Chakrabarti, Modified shuffled frog leaping algorithm with [35] B. Shaw, V. Mukherjee, S.P. Ghoshal, A novel opposition-based gravitational
genetic algorithm crossover for solving economic load dispatch problem with search algorithm for combined economic and emission dispatch problems of
valve-point effect, Appl. Soft Comput. 13 (2013) 4244–4252. power systems, Int. J. Electr. Power Energy Syst. 35 (1) (2012) 21–33.
[16] C.L. Chiang, Improved genetic algorithm for power economic dispatch of units [36] A. Chatterjee, S.P. Ghoshal, V. Mukherjee, Solution of combined economic and
with valve-point effects and multiple fuels, IEEE Trans. Power Syst. 20 (4) (2005) emission dispatch problems of power systems by an opposition-based har-
1690–1699. mony search algorithm, Int. J. Electr. Power Energy Syst. 39 (1) (2012) 9–20.
[17] Z.-L. Gaing, Particle swarm optimization to solving the economic dispatch [37] B. Mandal, P.K. Roy, Optimal reactive power dispatch using quasi-oppositional
considering the generator constraints, IEEE Trans. Power Syst. 18 (3) (2003) teaching learning based optimization, Int. J. Electr. Power Energy Syst. 53 (2013)
1187–1195. 123–134.
[18] K. Meng, H.G. Wang, Z.Y. Dong, K.P. Wong, Quantum-inspired particle swarm [38] A.R. Mehrabian, C. Lucas, A novel numerical optimization algorithm inspired
optimization for valve-point economic load dispatch, IEEE Trans. Power Syst. from weed colonization, Ecol. Inform. 1 (4) (2006) 355–366.
25 (1) (2010) 215–222. [39] R.H. Sepehri, C. Lucas, A recommender system based on invasive weed opti-
[19] S. Chakraborty, T. Senjyu, A. Yona, A.Y. Saber, T. Funabashi, Solving eco- mization algorithm, in: IEEE Congress on Evolutionary Computation, CEC, 2007.
nomic load dispatch problem with valve-point effects using a hybrid quantum [40] S. Karimkashi, A.A. Kishk, Invasive weed optimization and its features in elec-
mechanics inspired particle swarm optimization, IET Gener. Transm. Distrib. 5 tromagnetic, IEEE Trans. Antennas Propag. 58 (4) (2010) 1269–1278.
(10) (2011) 1042–1052. [41] M. Ahmadi, H. Mojallali, R. Izadi-Zamanabadi, State estimation of nonlinear
[20] J.G. Vlachogiannis, K.Y. Lee, Economic load dispatch—a comparative study on stochastic systems using a novel meta-heuristic particle filter, Swarm Evol.
heuristic optimization techniques with an improved coordinated aggregation Comput. 4 (6) (2012) 44–53.
based PSO, IEEE Trans. Power Syst. 24 (2) (2009) 991–1001. [42] B. Saravanan, E.R. Vasudevan, D.P. Kothari, Unit commitment problem solution
[21] J.B. Park, Y.W. Jeong, J.R. Shin, K.Y. Lee, An improved particle swarm optimiza- using invasive weed optimization algorithm, Int. J. Electr. Power Energy Syst.
tion for nonconvex economic dispatch problems, IEEE Trans. Power Syst. 25 (1) 55 (2014) 21–28.
(2010) 156–166.