Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Electrical Power and Energy Systems 45 (2013) 50–59

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems


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

Dynamic search space squeezing strategy based intelligent algorithm solutions


to economic dispatch with multiple fuels
A.K. Barisal ⇑
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 a new approach to the solution of optimal power generation for economic dispatch
Received 27 March 2009 (ED) using improved particle swarm optimization (IPSO) technique. In this paper an improved PSO tech-
Received in revised form 18 August 2012 nique is suggested that deals with equality and inequality constraints in ED problems. A constraint treat-
Accepted 19 August 2012
ment mechanism called dynamic search space squeezing strategy is devised to accelerate the
optimization process and simultaneously the dynamic process inherent in the conventional PSO algo-
rithm is preserved. The application and statistical performance of various intelligent algorithms such
Keywords:
as differential evolution (DE), particle swarm optimization (PSO) and improved particle swarm optimiza-
Economic dispatch
Differential evolution
tion (IPSO) are considered on economic dispatch problems with non-smooth cost functions considering
Particle swarm optimization valve point effects and multiple fuel options. To determine the efficiency and effectiveness of various
Dynamic search space squeezing strategy intelligent algorithms, three experiments are conducted considering only multiple fuel options, consider-
ing both valve-point and multiple fuel options and also taking into account the valve point loadings, ramp
rate limits and prohibited operating zones. The simulation results reveal that the proposed IPSO has pro-
vided the better solution with a very high probability to demonstrate its robustness over other intelligent
techniques such as DE, PSO and improved genetic algorithm with multiplier updating (IGA_MU), ant col-
ony optimization (ACO), artificial bee colony algorithm (ABC), hybrid swarm intelligent based harmony
search algorithm (HHS) and fuzzy adaptive chaotic ant swarm optimization (FCASO). The proposed IPSO
ensures convergence within least execution time and provides quality solutions as compared to earlier
reported best results.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction Some methods like the hierarchical method (HM) [8], the evolu-
tionary programming [9], the improved evolutionary programming
Economic load dispatch is an important power system optimi- (IEP) [10], the modified Hopfield neural network (MHNN) [11], the
zation task and one of the fundamental issues of power system Adaptive Hopfield neural network (AHNN) [12], consider only fuel
operation for scheduling generation among the committed gener- options. However, Park solves the ED problem by considering both
ators while satisfying system constraints and minimizing the cost valve point loading and multiple fuel options separately using
of energy requirements. For solving ED problems, previously clas- modified particle swarm optimization (MPSO) [19]. Very few at-
sical methods have been successfully employed with some approx- tempts have been made that incorporated both valve point effects
imations due to nonlinear characteristics of practical systems [1]. and multiple fuel options simultaneously, in the dispatch problem.
However, such approximations may cause to huge revenue loss Recently, Chiang [23] has considered both valve point effect and
over the passage of time. A dynamic programming method (DP) multiple fuel options simultaneously, to obtain accurate and prac-
can solve such problems in different formulations [1]. However, tical economic solution for real power system by using improved
this method suffers from massive computational burden for its genetic algorithm with multiplier updating (IGA-MU) [23]
large dimensionality when applied to practical size ELD problems. technique.
In past decades, many salient methods have been proposed The particle swarm optimization (PSO) [13–20], differential
such as Tabu search [2], genetic algorithm [3,4], evolutionary pro- evolution (DE) [21,22] techniques are recent additions to modern
gramming EP [5], particle swarm optimization with sequential heuristic algorithms, termed as global optimization techniques
quadratic programming PSO-SQP [6] and differential evolution and found to be highly flexible and robust in solving complex
(DE) [7] consider only valve point effects to solve ED problems. power system problems. The recent derivative free meta-heuristic
optimization algorithms, such as ant colony optimization (ACO)
⇑ Tel.: +91 6632430754; fax: +91 6632430204. [24], artificial bee colony algorithm (ABC) [25], hybrid swarm intel-
E-mail address: a_barisal@rediffmail.com ligent based harmony search algorithm (HHS) [26] and fuzzy

0142-0615/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2012.08.049
A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59 51

adaptive chaotic ant swarm optimization (FCASO) [27] have been 2.2. ELD with non-smooth cost functions
successfully employed to solve the economic dispatch of complex
real world power system problems in obtaining very promising Practically, the ELD problems are inherently highly nonlinear
and interesting results. and discontinuous in nature. Moreover, the cost functions have dis-
In this paper, a new approach to the inequality constraint (i.e., continuities corresponding to the change of fuels and also due to
generator outputs) treatment mechanism as dynamic search-space valve point effects that make the problem multimodal. Therefore,
squeezing strategy is devised to optimize the non-smooth ELD most of the techniques fail to obtain global solution instead of qua-
problems. Moreover, dynamic search-space squeezing strategy dif- si-global optimums to power system optimization problems.
fers from the dynamic search-space reduction strategy [19] in two
aspects. Firstly, the limits of position are determined from the rel- 2.2.1. Non-smooth cost function with valve point effects
ative distance measured from the best position of the group and The generator cost curve is obtained from the data point taken
position boundaries, but not from the individual distance as in case during ‘‘heat run’’ tests, when the input and output data are mea-
of reduction strategy. Secondly, the margins of reduction in case of sured as the unit slowly varies through its operating region. The
proposed squeezing strategy are considered from both the sides of generators with multiple valve steam turbines possess a wide var-
position simultaneously unlike one side consideration at a time iation in the input–output characteristics due to wire drawing ef-
[19] and also it varies adaptively instead of remaining constant. fects as illustrated in Fig. 1. The valve point effect introduces
The position of any individual or particle is adjusted between their ripples in the heat rate curves and cannot be represented by the
limits which are calculated based on the relative distance between polynomial function as in (2). Therefore, the accurate cost curve
the ever best position of the group and the inequality boundaries is a combination of sinusoidal functions and quadratic functions
to enhance the global exploration abilities. represented by the following equation:
The dynamic search space strategy is adopted to have faster
convergence in case of proposed Improved PSO. The intelligent F i ðPi Þ ¼ ai þ bi Pi þ ci P2i þ jei  sinðfi  ðPi min  Pi ÞÞj ð5Þ
algorithms like PSO and DE approaches are tested on first two cases where ei, fi are the constants of the ith unit with valve point effects.
on the test system-1. But, the proposed IPSO has been employed to
various test systems consisting of 10-, 15- and 40-generator sys- 2.2.2. Cost function with change of fuels
tems. The simulation results, so obtained are compared to that of Generally, the dispatching units are practically supplied with
recent approaches reported in the literature. The proposed meth- multi-fuel sources, each unit should be represented with several
odology gives the cheapest generation schedule and outperforms piecewise quadratic functions reflecting the effect of fuel type
previously reported other methods. changes, and the generator must identify the most economic fuel
to burn. The generator with multiple fuel options [8] has different
input–output curve as illustrated in Fig. 2. Therefore, it is more
2. Problem formulation
appropriate to represent the cost functions with piecewise qua-
dratic functions described in the following equation:
2.1. ELD with smooth cost function 8
>
> ai1 þ bi1 Pi þ ci1 P2i if Pi min 6 P i 6 Pi1 ; fuel  1
>
>
The prime objective of the ELD problem is to determine the >
> 2
< ai2 þ bi2 Pi þ ci2 Pi if Pi1 6 Pi 6 P i2 ; fuel  2
most economic loadings of generators to minimize the generation
F i ðPi Þ ¼ ...
cost such that the load demands PD in the scheduling horizon can >
>
>
> ...
be met and simultaneously and the power balance constraint and >
>
:
generating limit constraints are satisfied. Here, this constrained aim þ bim Pi þ cim P2i if Pim1 6 Pi 6 Pi max ; fuel  m
optimization problem can be written as: ð6Þ
X
m where aij, bij, cij are cost coefficients of unit i for the jth fuel type.
Minimize F Total ¼ F i ðPi Þ ð1Þ
i¼1
2.2.3. Cost function with valve point effects and change of fuels
In general, the cost function of ith unit Fi(Pi) is a quadratic poly- In reality, the objective function of the practical economic dis-
nomial and is expressed as: patch problem has nondifferentiable points according to valve
point loadings and multiple fuels. Therefore, the objective function
F i ðP i Þ ¼ ai þ bi Pi þ ci P2i ð2Þ should be composed of a set of non-smooth functions to obtain an

where ai, bi and ci are fuel cost coefficients of ith unit, and m is the
total number of committed units. $/ MW
(a) Power balance constraint or demand constraint: The total gen-
P
eration m i¼1 ðP i Þ should be equal to the total system demand
PD and total transmission loss PLoss. That is represented as
X
m
ðPi Þ ¼ PD þ PLoss ð3Þ
i¼1
C
B
(a) The generator limits: The generation output of each unit
should be between its minimum and maximum limits. That A
is, the following inequality constraint for each generator
should be satisfied.
Pi min 6 P i 6 Pi max ð4Þ
MW
Pi is the power output of ith unit and Pimin, Pimax are the minimum
and maximum real power output of ith unit. Fig. 1. Cost function with three valves from A to C.
52 A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59

techniques by many aspects that are population of solutions, not


Fuel A Fuel B Fuel C a single point solution, no limitation of size of the problem, inde-
$/MWh pendent of initial guess of the variables, simple concept and easy
implementation, very close to hundred percent success rate, signif-
icantly fast and robust due to competition and selection and also
incredibly well in solving the real-world practical ED problems.
In the more than 10 years since the intelligent algorithms were
first described, researcher have devised literally hundreds of mod-
ifications to improve it. Almost every paper reports that its new
version performs better than previous ones. Very few of the mod-
$/h ifications are quite helpful and adopted by the wider community
MW keeping in mind that a simpler algorithm that performs as well
Min P1 P2 Max
as more complicated one will be preferred. This paper presents
Fig. 2. Piecewise quadratic and incremental cost curve with three fuels. an improved version of particle swarm optimization in which a
constraint treatment mechanism called dynamic search space
squeezing strategy is devised to accelerate the optimization pro-
accurate and practical economic dispatch solution. The cost func- cess and simultaneously the dynamic process inherent in the con-
tion is framed by combining both valve point loadings and multi- ventional PSO algorithm is preserved. Here, the intelligent
fuel options which can be realistically represented as shown in algorithms such as differential evolution (DE), PSO and improved
the following equation: PSO are considered.
8
>
> a þ bi1 P i þ ci1 P2i þ jei1  sinðfi1  ðPi1min  Pi1 ÞÞj; for:fuel1; Pi1min 6 Pi 6 Pi1
> i1
>
>
< ai2 þ bi2 P i þ ci2 P2i þ jei2  sinðfi2  ðPi2min  Pi2 ÞÞj; for:fuel2; Pi2min 6 Pi 6 Pi2
F i ðPi Þ ¼ . .. 3.1. Differential evolution
>
>
> .. . ...
>
>
:
aim þ bim Pi þ cim P2i þ jeim  sinðfim  ðPimmin  Pim ÞÞj; for:fuel:m; Pimmin 6 Pi 6 Pim
The DE is a type of evolutionary algorithm originally proposed
ð7Þ
by Price and Storn [21] for optimization over a continuous space.
DE is a simple parallel search stochastic algorithm used for global
2.2.4. Cost function with ramp rate limits optimization. The initial population is chosen randomly and should
In ELD problems, the generator output is usually assumed to be cover the entire search space [22]. The optimization process in DE
adjusted smoothly and instantaneously. However, under practical is carried out with three basic operations: mutation, crossover and
circumstances ramp rate limit restricts the operating range of all selection. DE employs both mutation and crossover to produce one
the online units for adjusting the generation operation between trial vector Ui,j+1 for each target vector Xi,j.
two operating periods. The inequality constraint due to the ramp Then, a selection phase takes place, where each trial vector is
rate limits [20] of ith unit due to the change in generation are given compared with the corresponding target vector (one-to-one); the
by the following constraint. better one will enter into the population of the next generation.
For each target vector X i;j a mutant vector is generated using
MaxðPi min ; P0i  Dri Þ 6 Pi 6 MinðPi max ; P 0i þ U ri Þ ð8Þ rand/rand strategy as

if generation increases; Pi  P 0i 6 U ri ð9Þ U i;jþ1 ¼ X r3;j þ FðX r1;j  X r2;j Þ ð13Þ


if generation decreases; P0i  Pi 6 Dri ð10Þ where Ui,j+1 is the ith mutated individual for the next genera-
tion, X the population set, F the mutation constant, [0, 2], j the cur-
where Pi and P0i are the current and previous power output of ith rent iteration, j + 1 the next iteration and Xr1,j, . . ., Xr3,j the
generator, respectively. Uri and Dri are the up ramp and down ramp randomly selected individuals from the population of current
limits of the ith generator. iteration.
To increase the diversity of the perturbed parameter vectors
2.2.5. Cost function with prohibited operating zones crossover is performed after mutation. The crossover constant CR
The input–output characteristics of modern units are inherently controls the diversity of population and aids the algorithm to es-
nonlinear because of the steam valve point loadings [2]. The oper- cape from local optima.
ating zones due to valve point loading or vibration due to shaft
bearing is generally avoided in order to achieve best economy, V i;jþ1 ¼ X i;j ð1  CRÞ þ U i;jþ1 CR ð14Þ
called prohibited operating zones of a unit, which make the cost
curve discontinuous in nature. The feasible operating zones of ith
where CR is the crossover probability constant from interval [0, 1].
unit having k number of prohibited operating zones are repre-
The parents for the next iteration are selected as follows.
sented by
h i 
Pi R Ppzk pzk
iL ; P iU k ¼ 1; 2; . . . ð11Þ V i;jþ1 if f ðV i;jþ1 Þ > f ðX i;jþ1 Þ
X i;jþ1 ¼ ð15Þ
X i;jþ1 if f ðX i;jþ1 Þ > f ðV i;jþ1 Þ
Pi 6 Ppzk
iL and P i P Ppzk
iU ð12Þ
where f(Vi,j+1) is the fitness function value of the ith individual of
where Ppzk
and
iL P pzk
are the lower and upper limits of kth prohibited
iU the population to which the mutation and crossover operators are
zone for ith generating unit. applied and f(Xi,j+1) is the fitness function value of the ith individual
in the original population. The loss of the best individuals in the
3. Intelligent algorithms following iteration is avoided by this selection mechanism, as the
worst individuals are replaced by the best individuals. This process
The intelligent algorithms are population based stochastic continues until the maximum function evaluation is reached or no
parallel search algorithms differ from traditional optimization improvement is seen in the best individual after many generations.
A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59 53

3.2. Particle swarm optimization DkLi þ DkHi ¼ 1 ð19Þ


At iteration k + 1, the adjusted limits of output of generator i are
The PSO idea was originally introduced by Kennedy and Eber-
determined to satisfy inequality constraints as follows:
hart in [14] as an optimization technique inspired by swarm intel-
 
ligence and theory in general such as bird flocking, fish schooling kþ1
Pi;min
k
¼ Pi;min þ gbest i  Pi;min  DkLi ð20Þ
and even human social behavior. Thereafter, PSO became a novel
optimization tool providing a population based search procedure  
kþ1 k
in which individuals, called particles fly in an m-dimensional Pi;max ¼ P i;max  Pi;max  gbesti  DkHi ð21Þ
search space, where each dimension corresponds to a parameter
in a function being optimized. The position mechanism of the par- The limits of output of generators are varying in iterations but
ticle in the search space is updated by adding the velocity vector always dependent on the location of gbest in the boundary. The up-
(16) to its position vector (17), and has been illustrated in Fig. 3. dated maximum and minimum limits are described in (20) and
Let X and V be a particle position and its corresponding velocity (21), and always satisfied by (4). The activation of dynamic space
in a search space respectively. The best position achieved by a par- squeezing process is illustrated in Fig. 4.
ticle is recorded and denoted by Pbest. The best particle among all
particles in the population is represented as gbest. 3.3.2. Proposed algorithm for ED problem
The updated velocity and position of a particle can be calculated The computational procedure for proposed IPSO technique can
as shown in the following formulae: be described in the following steps:
  
~
V kþ1 ¼ w~ V kji þ c1  randðÞ  ~ Pbestji  ~
X kji Step-1: Input parameters of the system and specify the upper
ji
  and lower boundaries of each variable.
þc2  randðÞ  ~ g besti  ~
X kji ð16Þ Step-2: Initialize randomly the particles of the population.
These particles must be feasible candidate solutions that satisfy
~
X kþ1 ¼~
X kji þ ~
V kþ1 j ¼ 1; 2; . . . ; n; i ¼ 1; 2; . . . ; m ð17Þ the given constraints.
ji ji ;
Step-3: Let, Pj ¼ ½ P1 P2 ; . . . ; P m  be the trial vector denoting the
where k is the pointer of iterations (generations); w the inertia particles of population to be evolved. The elements of Pi are the
weight factor; n the number of particles in a group; m the number real power outputs of the committed m generating units sub-
of generators in a particle; c1, c2 the acceleration constant, in gen- jected to economic fuel types to burn and their respective
eral and rand () is the random number in the range [0,1]; capacity constraints in (4). To meet exactly the load demand,
a dependent unit is randomly selected from among the commit-
3.3. Improved PSO for ELD problems ted m units. Let, Pd be the power output of the dependent unit to
satisfy equality constraint (slack generator), then Pd is calcu-
3.3.1. Dynamic search-space squeezing strategy lated by
The committed generators are considered as the particles in
population and its capacity limits as boundaries of search space. X
m

Judicious choice of search space of the particle not only improves Pd ¼ P D þ PLoss  Pi i ¼ 1; 2; . . . ; m ð22Þ
the speed of convergence but also ensures the algorithm to be less i¼1
susceptible to getting trapped on local minima, i.e. the solution of i–d
IPSO is better than others. When there are no significant improve-
ments in the performance of solution achieved, the dynamic Step-4: Go to step-3 for few iterations and then, go to step-5.
search-space squeezing strategy is activated to accelerate the con- Step-5: Compare each particle evaluation value with its Pbest.
vergence speed. In this case, the search space is dynamically read- The best evaluation value among Pbest’s is denoted as gbest.
justed (i.e., squeezed) based on the relative distance between gbest Step-6: Update the iteration as iter = iter + 1.
and lower and upper limits of ith generator denoted by DLi and DHi Step-7: Update inertia weight.
respectively. Both the relative distances are variables, not always Step-8: Update velocity and position of each particle within
equal and constant, which are represented as follows: their assigned limits using (16), (17), and (4).
k
gbest  Pi;min Pi;max  gbest
k Step-9: Each particle is evaluated according to its updated posi-
DkLi ¼ ; DkHi ¼ ð18Þ tion. If the evaluation value of each particle is better than the
Pi;max  P i;min Pi;max  Pi;min
previous Pbest, the current value is set to be Pbest. If the best
Pbest is better than gbest, the value is set to be gbest.
Step-10: If the stopping criterion (maximum iterations set) is
satisfied then go to step-11. Otherwise, go to step-2.

j th Particle Pjk1 Pjk2 Pjik Pjmk

k −1 Pjik ,max
Pjik ,min gbest i

Iteration k

gbestik

Iteration k+1 Pjik ,+min


1 • Pjik,+max
1

Fig. 3. The search mechanism of PSO by jth particle. Fig. 4. Search-space squeezing mechanism of the jth particle during activation.
54 A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59

Step-11: The space squeezing strategy is activated to adjust the Table 1


upper and lower bounds of the particles relative to latest gbest Comparison of optimization methods for case I of system-1.

with new parameters using (20) and (21) for few iterations. If Demand 2400 MW 2500 MW 2600 MW 2700 MW
the performance is improving then, go to step-1. Methods Minimum cost ($)
Step-12: The particle that generates the latest gbest is the opti- HM [8] 488.5 526.7 574.030 625.180
mal solution. EP [9] 481.740 526.350 574.750 626.260
IEP [10] 481.779 526.304 574.473 623.851
MHNN [11] 487.87 526.13 574.26 626.12
3.3.3. IPSO implementation AHNN [12] 481.7 526.23 574.37 626.240
The present work has been implemented in command line of MPSO [19] 481.723 526.239 574.381 623.809
Matlab-7.0 for the solution of economic load dispatch with non- CGA-MU [23] – – – 623.8095
smooth cost functions. The program was run on a 3.06 GHz, Pen- IGA-MU [23] – – – 623.8093
DE 481.723 526.239 574.381 623.8098
tium-IV, with 256 MB RAM PC. Optimal parameter combinations PSO 481.723 526.240 574.381 623.8094
for different methods are experimentally determined by conduct- IPSO 481.722 526.238 574.380 623.8089
ing a series of experiments with different parameter settings be-
fore conducting actual runs to collect the results. In PSO, there
are three parameters namely inertia constant w, velocity vmax
system load demand is varied from 2400 MW to 2700 MW with
and acceleration constants c1, c2 to be adjusted for optimum per-
100 MW increments and the system data and related constraints
formance beside swarm size. The different values of PSO parame-
are adopted from [8] and [23].
ters such as inertia weight, number of particles, maximum
By satisfying one equality as power balance and 20 inequality
allowable velocity, the following selected key parameters as
constraints as generator capacity limits of the system, the intelli-
wmax = 0.7, wmin = 0.1, v max ¼ ðpmax10pmin Þ
, vmin = vmax, c1 = 1.2 gent algorithms such as IPSO, PSO and DE’s best result out of 20
c2 = 1.8, n = 30 give the better performance compared to all other independent trials is compared to that of HM [8], EP [9], IEP [10],
settings in terms best and mean values.When dynamic search MHNN [11], AHNN [12], and MPSO [19] and given in Table 1. From
squeezing strategy is activated in case of IPSO, then, the new se- this Table it is evident that the proposed IPSO provides the better
lected parameters are wmax = 0.01, wmin = 0.0, v max ¼ ðpmax20
pmin Þ
, solution quality than others methods such as HM [8] (except
vmin = vmax, c1 = 1.2, c2 = 1.8, n = 30. 2600 MW case), IEP [10], MHNN [11], and MPSO [19] by satisfying
Similarly, the differential evolution method with the control equality and inequality constraints. However, the HM [8] and
parameters such as F and CR varied from 0.1 to 1 in steps of 0.1. MHNN [11] do not satisfy the equality constraint. AHNN [12] pro-
Out of ten combinations, CR = 0.6 and F = 0.6 yielded the best re- vides slightly better solution than IPSO except 2700 MW load de-
sults in terms of mean and best solutions obtained. Hence, this mand. Moreover, the generation configurations of IPSO are
parameter setting is chosen as the optimal setting for DE. almost similar or better to MPSO [19] and AHNN [12]. Fig. 5 shows
the convergence natures of Improved PSO, PSO and DE algorithms.
4. Systems and results The proposed improved PSO algorithm also yields better solution
quality and consistency over PSO and DE intelligent algorithms.
The proposed dynamic search space strategy based Improved The best power generation schedule and fuel selection out of 20
PSO (IPSO), PSO and DE approaches are tested on the three test sys- independent trials for case I of system-1 by the proposed IPSO for
tems. To demonstrate the robustness of the proposed approach various load demands 2400 MW, 2500 MW, 2600 MW and
over PSO and DE and also the obtained results are compared with 2700 MW is provided in Table 2. The optimal power generation
those obtained from other methods such as hierarchical method schedule and fuel selection of three intelligent algorithms like Im-
(HM) [8], the evolutionary programming [9], the improved evolu- proved PSO, PSO and DE for a load demand of 2700 MW corre-
tionary programming (IEP) [10], the modified Hopfield neural net- sponding to best trial is given in Table 3. The comparisons in
work (MHNN) [11], the Adaptive Hopfield neural network (AHNN) Tables 1 and 3, clearly reveal that the proposed IPSO is more effec-
[12], modified particle swarm optimization (MPSO) [19], the im- tive to PSO and DE methods and also to other methods reported in
proved genetic algorithm with multiplier updating (IGA_MU) literature. Table 4 provides the statistical performance of three
[23], ant colony optimization (ACO) [24], artificial bee colony algo- intelligent algorithms in terms of best, mean, worst cost, standard
rithm (ABC) [25], hybrid swarm intelligent based harmony search deviation and mean computation time over 20 independent trials.
algorithm (HHS) [26] and fuzzy adaptive chaotic ant swarm opti- The performance PSO is better than DE in all aspects except to the
mization (FCASO) [27].
The applicability and validity of the aforementioned technique
for practical applications has been tested on various test systems
Convergence characteristics of intelligent algorithms
consisting of 10-, 15- and 40-generator systems. The first system
632
having three cases such as the case 1 considers only multiple fuel Load Demand of 2700 MW for System 1
options, case 2 and case 3 consider both multiple fuel option and 630
valve point loading effects with different combination of fuel op-
tions for generating units. The second system deals with prohibited 628 IPSO
Cost ($)

operating zones and ramp rate constraint with transmission losses


and the third system considers the famous benchmark large scale DE
626
economic dispatch with valve point effects. PSO
624
4.1. Test system-1
622
0 20 40 60 80 100
4.1.1. Case I
Number of iterations
Here, the non-smooth cost function of 10 unit systems only
with multiple fuel options has been considered and the objective Fig. 5. Convergence characteristics of the intelligent algorithms for case I of
function is represented as piecewise quadratic cost function. The system-1.
A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59 55

Table 2 computation time. The IPSO takes minimum computational time


Optimal generations and fuel types for case I of system-1 by IPSO. 0.922 s to reach the best solution. The minimum standard devia-
Unit 2400 MW 2500 MW 2600 MW 2700 MW tion in case of IPSO suggests the consistency of quality solution
F P (MW) F P (MW) F P (MW) F P (MW) in comparison to other intelligent methods.
Fig. 6 illustrates the convergence characteristics for case I of
1 1 189.7403 2 206.5536 2 216.5772 2 218.2345
2 1 202.3379 1 206.4974 1 210.8794 1 211.7061
system-1 for different load demands 2400 MW, 2500 MW,
3 1 253.8928 1 265.7000 1 278.5072 1 280.7208 2600 MW and 2700 MW. The simulation results it is found that
4 3 233.0418 3 235.9342 3 239.0669 3 239.6091 the costs, fuel types and dispatch levels from IPSO are quite similar
5 1 241.8399 1 257.9430 1 275.5305 1 278.4580 to that of MPSO [19], CGA_MU [23] and IGA_MU [23] methods and
6 3 233.0439 3 235.9673 3 239.1177 3 239.6704
different from other methods. The overall performance of IPSO is
7 1 253.2670 1 268.8038 1 285.6955 1 288.6079
8 3 233.0452 3 235.9434 3 239.1346 3 239.6341 always better than other existing heuristic methods in applying
9 1 320.3856 1 331.5537 1 343.5195 3 428.4794 the practical ED problems.
10 1 239.4056 1 255.1036 1 271.9715 1 274.8798
Total 2400.00 2500.00 2600.00 2700.00
Cost ($) 481.722 526.238 574.380 623.808
4.1.2. Case II
F: Fuel. The case II of test system-1 contained ten units addressing both
valve point effects and multiple fuel options simultaneously. The
key parameters selected here are same as the parameters in case I.
Fig. 7 presents the convergence nature of three intelligent
Table 3 algorithms such as DE, PSO and IPSO. Table 5 provides the compar-
Optimal output and fuel selection for case I of system-1 with PD = 2700 MW. isons of the results obtained by various intelligent algorithms for
Unit no. DE PSO IPSO
case II of test system-1 with recently applied IGA_MU, CGA_MU
[23] and ACO [24] techniques. The proposed IPSO gives better cost
P (MW) F P (MW) F P (MW) F
for all power demands than other intelligent methods. The fuel
1 218.2620 2 218.3258 2 218.2345 2 costs for all demands are increased due to the addition of valve
2 211.6445 1 211.6818 1 211.7060 1
point loading effects.
3 280.7456 1 280.6169 1 280.7208 1
4 239.7753 3 239.5636 3 239.6091 3 Table 6 shows the optimal power generation schedule and eco-
5 278.0133 1 278.6062 1 278.4580 1 nomical fuel selection for case II and corresponding to the best trial
6 239.5186 3 239.5004 3 239.6704 3 out of 20 independent trials by DE, PSO and IPSO techniques for a
7 288.5972 1 288.4925 1 288.6079 1 load demand of 2700 MW for fair comparison with IGA_MU,
8 239.7293 3 239.6469 3 239.6341 3
9 428.9363 3 428.8116 3 428.4787 3
CGA_MU and ACO methods. Among the six algorithms, IPSO gives
10 274.7780 1 274.7543 1 274.8798 1 the cheapest generation schedule.
Total 2700 2700 2700 Fig. 8 illustrates the convergence characteristics of case II for
Cost ($) 623.8098 623.8094 623.8089 different load demands 2400 MW, 2500 MW, 2600 MW and
2700 MW. For all power demands, the optimal generation schedule
and economic fuel selection by IPSO are provided in Table 7.

Table 4 Convergence characteristics of intelligent algorithms


Comparison of simulation results with PD = 2700 MW. 632
Evolution Best cost Worst Mean Standard Time
Load Demand of 2700 MW for System 2
methods ($) cost ($) cost ($) deviation (s) 630
DE 623.8098 623.8193 623.8133 3.2705e4 1.152
PSO 623.8094 623.8102 623.8098 2.7419e4 1.273 628
Cost ($)

IPSO 623.8089 623.8097 623.8092 1.3516e5 0.922 IPSO


PSO DE
626

624

Convergence plot by IPSO at different load demands 622


0 20 40 60 80 100
650
Number of iterations

2700MW Fig. 7. Convergence characteristics of the intelligent algorithms for case II of


600
system-1.
Cost ($)

550 2600MW
Table 5
Comparison of optimization methods for case II of system-1 with all demands.

2500MW Methods Minimum cost ($)


500
2400 MW 2500 MW 2600 MW 2700 MW
CGA-MU – – – 624.7193
2400MW
450 IGA-MU – – – 624.5178
0 20 40 60 80 100 DE 482.1245 527.0024 574.9744 624.4606
No. of Generations PSO 482.0014 526.9336 574.7146 624.2449
IPSO 481.8044 526.2929 574.4326 623.8730
Fig. 6. Convergence characteristics of the IPSO for case I with various load ACO 482.5267 – – 624.193
demands.
56 A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59

Table 6
Comparison of optimization methods for case II of system-1 with PD = 2700 MW.

Unit CGA-MU IGA-MU DE PSO IPSO ACO [24]


P (MW) F P (MW) P (MW) P (MW) P (MW) P (MW) F
1 222.0108 2 219.1261 219.5864 220.0873 217.5692 221.45 2
2 211.6352 1 211.1645 212.2079 210.4593 211.2166 211.53 1
3 283.9455 1 280.6572 278.7762 280.3836 279.6488 281.62 1
4 237.8052 3 238.4770 238.5645 241.3854 240.1769 239.90 3
5 280.4480 1 276.4179 276.7550 276.3311 276.5743 276.99 1
6 236.0330 3 240.4672 240.0430 239.5128 239.9082 239.11 3
7 292.0499 1 287.7399 281.3123 293.2521 285.3796 284.76 1
8 241.9708 3 240.7614 241.1176 241.9236 240.4456 240.70 3
9 424.2011 3 429.3370 417.2435 414.3107 430.0665 429.61 3
10 269.9005 1 275.8518 294.3936 282.3540 279.0143 274.31 1
Total 2700 2700 2700 2700 2700 2699.98
Cost ($/h) 624.7193 624.5178 624.4606 624.2449 623.8730 624.193
Reported 623.7 $/h

The significance of bold value are the actual calculated values but different from the reported values.

Convergence plot by IPSO at different load demands Table 8 provides the performance of various algorithms statisti-
650 cal measures such as best, mean, worst, standard deviation and
mean computation time over 20 independent trials. The proposed
IPSO has better economic cost, lower mean time and lower stan-
600 dard deviation than other methods reported in Table 8. The fuel
2700MW cost reported by ACO method is 623.7 $/h for the load demand of
2700 MW. However, the actual cost by ACO is found to be
Cost ($)

550 624.193 $/h, when it is tested as per the input data of the fuel com-
2600MW
binations and the outputs of generators provided by the ACO meth-
2500MW od [24]. Similarly, the tested cost of ACO method for 2400 MW is
500
682.5267 $/h, but it is reported as 681.60 $/h.
From the compared results, the proposed IPSO exhibits not only
2400MW better solution quality and consistency but also less computational
450
0 20 40 60 80 100 time than other reported techniques.
No. of Generations To show the robustness of the proposed IPSO for the realistic ED
problem with a load demand of 2700 MW, the relative frequency of
Fig. 8. Convergence characteristics of the IPSO for case II with various load
convergence by IGA_MU, CGA_MU, DE, PSO and IPSO techniques
demands.

Table 7
Optimal solutions for cases II & III of system-1 with different fuel options and loads by IPSO.

Unit 2400 MW 2400 MW [24] 2500 MW 2600 MW 2700 MW (case III only)
F P (MW) F P (MW) F P (MW) F P (MW) F IPSO [25]
1 1 189.3089 1 190.3 2 205.2267 2 218.4885 2 218.6744 203.2661
2 1 202.5518 1 203.1 1 207.7520 1 211.7119 1 212.9650 197.3535
3 1 255.4520 1 253.3 1 263.5269 1 276.6181 3 332.6432 347.8936
4 3 231.4428 3 233.1 3 235.3398 3 239.3707 3 241.3894 234.4000
5 1 242.4392 1 241.5 1 258.4485 1 276.0437 1 287.9013 294.3970
6 3 234.3990 3 232.5 3 236.2800 3 240.1769 3 238.5691 235.0708
7 1 250.2464 1 252.8 1 271.0193 1 285.3564 1 296.7754 296.6126
8 3 232.5178 3 233.1 3 235.6078 3 238.1613 3 243.4030 248.3842
9 1 321.5403 1 320.1 1 331.4669 1 342.0228 1 348.1188 351.3578
10 1 240.0479 1 240.2 1 255.3320 1 272.0497 1 279.5604 291.2644
Cost ($/h) 481.8044 482.5267 526.2929 574.4326 629.2984 632.8052
Reported [24] 481.6 $/h P means Output of Generator Reported [25] 609.2250

The significance of bold value are the actual calculated values but different from the reported values.

Table 8
Comparison of statistical results of case II of system-1 with PD = 2700 MW.

Methods Best cost ($) Worst cost ($) Mean cost ($) Standard deviation Mean time (s)
CGA_MU 624.7193 633.8652 627.6087 – –
IGA_MU 624.5178 630.8705 625.8692 – –
DE 624.4606 624.4918 624.4724 0.0076 1.220
PSO 624.2449 624.2744 624.2543 0.0028 1.453
IPSO 623.8730 623.8900 623.8887 8.5615  104 1.032
ACO 623.7/624.193 624.09 623.9 0.11 5.16
A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59 57

Table 9
Comparisons of different methods on relative frequency of convergence in the ranges of cost.

Methods Range of cost ($)


633.5– 632.5– 631.5– 630.5– 629.5– 628.5– 627.5– 626.5– 625.5– 624.5– 624.0– 623.5–
634.5 633.5 632.5 631.5 630.5 629.5 628.5 627.5 626.5 625.5 624.5 624.0
CGA_MU 1 2 3 7 10 21 31 20 5
IGA_MU 1 2 2 11 45 39
DE 100
PSO 100
IPSO 100

Table 10
Optimal generations and fuels of case III of system-1.

Unit 2400 MW 2500 MW 2600 MW 2700 MW


F IPSO [25] F IPSO [25] F IPSO [25] F IPSO [25]
1 1 181.9673 183.4274 2 198.4447 182.4037 2 210.8717 225.9520 2 221.8579 233.2278
2 1 198.1485 197.6764 1 203.4430 182.4032 1 207.2779 215.0370 1 213.3369 210.6630
3 3 332.0000 342.6159 3 332.0000 343.6259 3 332.0068 342.4434 3 332.0000 335.9482
4 3 230.1110 229.8365 3 233.8224 224.9392 3 236.9980 234.1474 3 240.8200 234.7204
5 1 226.2664 226.3649 1 246.0063 246.2421 1 263.7957 253.6373 1 284.6940 278.7617
6 3 230.2063 230.0831 3 233.0153 233.6105 3 239.0008 227.4553 3 240.7574 232.7875
7 1 237.8645 236.8642 1 257.5676 296.6863 1 275.7265 276.9986 1 294.6405 283.9141
8 3 230.2220 223.4053 3 233.3000 229.3996 3 236.4008 233.6542 3 240.7403 241.4519
9 1 309.4490 305.9907 1 321.4824 324.5771 1 335.2384 327.7284 1 350.0180 352.3731
10 1 223.7650 223.7356 1 240.9183 236.1124 1 262.6834 262.9464 1 281.1350 296.1523
Cost ($/h) 491.9448 494.1713 533.5923 539.9589 579.1937 581.774 629.0219 630.5187
Reported in [25] 470.9506 Reported in [25] 516.2793 Reported in [25] 558.5632 Reported in [25] 607.7481
Time (s) 0.9016 1.8103 0.9022 1.3653 0.9174 1.6619 0.9233 2.0361

The significance of bold value are the actual calculated values but different from the reported values.

are listed in Table 9 for each cost range among 100 randomly ini- When both multiple fuel options and valve point effects are
tiated trials. The minimum generation cost obtained by IPSO is considered, the actual costs of IPSO and ABC method for
623.8862 $/h which lies within the cost range of 623.5 $/h to 2700 MW are 629.2984 $/h and 632.8052 $/h, respectively.
624.0 $/h. Table 9 reveals that the proposed IPSO has provided However, it is reported as 609.2250 $/h by ABC method [25]. From
the global solution with a very high probability to demonstrate the Tables 7–10, it is clear that IPSO provides better high quality
its effectiveness and efficiency over other intelligent techniques. solution than other methods.

4.2. Test system-2: (15 unit system with ramp rate and prohibited
4.1.3. Case III: a new set of fuels operating zones)
To verify the feasibility and applicability of the proposed IPSO
method is compared with the results obtained by ABC algorithm This system consists of 15-unit system to verify the feasibility
considering the multi fuel options and valve point loading effects. and applicability of the proposed approach in solving non-smooth
Firstly, only multiple fuels and secondly, both multiple fuel options functions exhibiting prohibited operating zones, transmission
and valve point effects simultaneously are considered. Each unit losses and ramp rate constraints. The unit input data and B-loss
has three options for fuel, and the optimal generation is obtained coefficients are adapted from [20], and the load demand of the sys-
based on the most economical fuel to be burned. However, Hem- tem is 2630 MW. The key parameters for IPSO of this system are k:
amalini and Simon [25] have considered a new set of fuel options 100, m:15, wmax = 0.01, wmin = 0.0, v max ¼ ðpmax10
pmin Þ
, vmin = vmax, c1 = 2,
in order to achieve the best economy in obtaining the total cost of c2 = 2 and n = 30. The optimal solutions obtained by the proposed
generation of ten dispatching units. The input data of this system is IPSO method along with other methods such as ABC [25] and
adopted mainly from [25]. The key parameters selected here are HHS [26] are provided in Table 11. The global optimum solution
same as the parameters in case II of test system-1. for 15-generators system is yet to be discovered. It was reported
Table 10, provides the optimal generation, total generation cost, that, the optimal solution for 15 generator system was
execution time and also the fuel combinations as per the reference 32692.8361 $/h by the HHS method [26]. The transmission loss is
[25] with varying load demands from 2400 MW to 2700 MW with recalculated by using the optimal solution given by [26] and with
100 MW increments by neglecting the valve point effects. The fuel the input data [20] is found to be 30.83945 MW. However, it is re-
costs reported by ABC methods are 470.9506, 516.2793, 558.5632 ported by HHS method [26] that the loss as 29.66314 MW different
and 607.7481 $/h for the load demands 2400 MW, 2500 MW, from the tested results. This method fails to satisfy the power bal-
2600 MW and 2700 MW, respectively. However, the actual cost ance equation i.e., the load demand is 2628.81995 MW instead of
by ABC method are found to be 494.1731, 539.9581, 581.774 and 2630 MW. The optimal solution among 50 trials by the IPSO meth-
630.5187 $/h, when it is tested according to the input data of the od is found as 32706.6580 $/h, the loss 30.85745 MW, average
fuel combinations and the outputs of generators provided by the computational time 2.356 s with the standard deviation 21.882
ABC method [25], are different to the reported fuel costs for various by satisfying all the constraints, such as power balance, ramp rate
load demands. The Proposed IPSO provides the better fuel cost as limits, prohibited operating zones, generation limits and transmis-
per the fuel combination [25] is shown in Table 10. sion loss thereby validating the heuristics applicability. Moreover,
58 A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59

Table 11 Table 13
Best solution for system-2 with a demand of 2630 MW. Comparisons of results for system-3 with demand of 10500 MW.

Unit power output (MW) IPSO (proposed) ABC [25] HHS [26] Methods Best cost Avg. cost ($/ Worst cost Time Standard
($/h) h) ($/h) in (s) deviation
P1 455.0000 455.0000 455.0000
P2 380.0000 380.0000 379.9954 ACO [24] 121532.41 121606.45 121679.64 52.45 45.58
P3 130.0000 130.0000 130.0000 ABC [25] 121441.03 121995.82 – 30.02 –
P4 130.0000 130.0000 130.0000 HHS [26] 121415.592 121615.8544 – 16.385 114.56
P5 170.0000 169.9997 169.9572 FCASO [27] 121516.47 122082.59 – 145.18
P6 460.0000 460.0000 460.0000 FAMPSO[26] 121412.57 121413.388 – 37.4 –
P7 430.0000 430.0000 430.0000 IPSO
P8 71.8762 71.9698 81.8563
(proposed)121412.866121509.5223121546.84218.08442.89
P9 58.98125 59.1798 47.8546
P10 160.0000 159.8004 160.0000
P11 80.0000 80.0000 80.0000
P12 80.0000 80.0000 79.9959
P13 25.0000 25.0024 25.0000
comparatively better than other techniques except the FAMPSO
P14 15.0000 15.0056 15.0000 [26] fuzzy adaptive PSO algorithm very close result. The IPSO
P15 15.0000 15.0014 15.0000 method gives the cheapest generation schedule for which huge
Total power output 2660.85745 2660.95910 2659.65940 revenue is saved over a long period of time say, annually. Addition-
Ploss (MW) reported 30.85745 30.86010 29.66314
ally, the IPSO method can obtain a lower mean cost, mean compu-
Ploss (MW) tested 30.85745 30.86010 30.83945
Load demand (MW) 2630.0000 2630.09900 2628.81995 tational time and standard deviation than the other techniques
Total Gen. cost ($/h) 32706.6580 32707.8551 32692.8361 reported in literature, thus signifying in higher quality solution,
stable convergence characteristics and good computational
The significance of bold value as the Load demand of 2630.0000 MW is violated
efficiency.

it is evident from this table that there is a power mismatch in other 5. Conclusion
methods except the swarm intelligent based IPSO method.
This paper presents a new approach as Improved PSO to non-
4.3. Test system-3: (40 unit system with valve point loading effects) smooth ELD cost functions while preserving the inherent basics
of PSO algorithm. The dynamic search space squeezing strategy is
The developed derivative free meta-heuristic optimization algo- proposed to efficaciously handle inequality constraints and to re-
rithms such as IPSO, ACO [24], ABC [25], HHS [26] and FCASO [27] duce the search space margin judiciously to make convergence
methods have been applied to 40 unit system with valve point speed faster. The intelligent algorithms such as DE, PSO and IPSO
loading effects. A load demand of 10,500 MW is considered in this are applied to solve the non-smooth ED problem considering valve
case. The input data of 40 units is given in [20]. The key parameters point effects and multiple fuel options. Three types of ED problems
for IPSO of this system are k: 500, m: 40, wmax = 0.01, wmin = 0.0, with prohibited operating zones, ramp rate limits, valve-point
v max ¼ ðpmax20pmin Þ, vmin = vmax, c1 = 2, c2 = 2 and n = 30. The genera- loading and multiple fuels effects have been considered to evaluate
tion outputs and the corresponding costs of the best solution by the performance of IPSO. Additionally, the findings of IPSO are
the proposed IPSO method are provided in Table 12. The obtained compared with those obtained by the conventional heuristic ap-
optimal results by the IPSO method are compared with results ob- proaches, i.e. ACO, ABC, HHS and FCASO. Studied results affirm that
tained from other methods [24–27] are given in Table 13. the proposed ACO is much superior to other established methods
The obtained minimum cost reported in [26] is by HHS method, in terms of high-quality solution, better convergence property
which is found to be 121, 415.592 $/h. The best cost obtained by and computational efficiency.
IPSO method is 121412.866 $/h, as shown in Table 12, which is The IPSO has shown very similar results to that of other meth-
ods in case of non-smooth cost function only with multi-fuel op-
tions as in case 1 and outperforms in the case of complicated
Table 12 case 2 of system 1 which incorporates both valve point loading ef-
Generation output of each generator and the corresponding cost of the system-3. fects and multiple fuel options simultaneously. The performance of
DE is better than PSO in terms of computational time, but the pro-
Unit Gen. (MW) Cost ($/h) Unit Gen. (MW) Cost ($/h)
posed IPSO is faster than DE technique in both of the systems. In
1 110.8001 925.101 21 523.2802 5071.306
this paper, I have successfully employed the proposed IPSO method
2 110.7988 925.097 22 523.2791 5071.290
3 97.3996 1190.548 23 523.2787 5057.224 to solve the ED problem with nonlinear characteristics of the gen-
4 179.7339 2143.567 24 523.2792 5057.223 erator such as ramp rate limits, prohibited operating zones and
5 87.7993 706.501 25 523.2775 5275.089 non-smoothness are incorporated in cost function. The proposed
6 140.0000 1596.464 26 523.2807 5275.116 methodology gives the cheapest generation schedule, consistently
7 259.5997 2612.885 27 10.0000 1140.524
8 284.5994 2798.228 28 10.0006 1140.538
quality solution, simple concept and easy implementation and
9 284.5983 2779.836 29 10.0000 1140.524 good convergence characteristics over any other methods consid-
10 130.0000 2502.065 30 87.7999 706.500 ered for solving real world ED problems.
11 94.0000 1893.305 31 190.0000 1643.991
12 94.0005 1908.178 32 189.9987 1643.986
13 214.7577 3792.062 33 189.9994 1643.989 References
14 394.2833 6414.960 34 164.7999 1585.546
15 394.2808 6436.622 35 194.3975 1985.423 [1] Wood AJ, Wollenberg BF. Power generation operation and control. New
16 394.2784 6436.582 36 200.0000 2043.727 York: John Willey and Sons; 1984.
17 489.2771 5296.710 37 109.9997 1220.164 [2] Lin WM, Cheng FS, Tsay MT. An improved Tabu search for economic dispatch
18 489.2799 5288.777 38 110.0000 1220.166 with multiple minima. IEEE Trans Power Syst 2002;17:108–12.
19 511.2792 5540.929 39 110.0000 1220.166 [3] Walters DC, Sheble GB. Genetic algorithm – economic dispatch example. IEEE
20 511.2848 5541.027 40 511.2781 5540.928 Trans Power Syst 1993;8:1325–32.
Total load (MW) and Cost ($/h) 10500.000 121412.866 [4] Wong KP, Wong YW. Thermal generator scheduling using hybrid genetic
simulated annealing. Proc IEE Trans Gener, Transm Distrib 1995;142:372–80.
A.K. Barisal / Electrical Power and Energy Systems 45 (2013) 50–59 59

[5] Sinha N, Chakrabarti R, Chattopadhyay PK. Evolutionary programming [17] Abido MA. Optimal design of power system stabilizers using particle swarm
techniques for economic load dispatch. IEEE Trans Evol Comput optimization. IEEE, Trans Energy Convers 2002;17(3):406–13.
2003;7(1):83–94. [18] Yoshida H, Kawata K, Fukuyama Y, Takayama S, Nakanishi Y. A particle swarm
[6] Victoire TAA, Jeyakumar AE. Hybrid PSO-SQP for economic dispatch with optimization for reactive power and voltage control considering voltage
valve-point effect. Electr Power Syst Res 2004;71(1):51–9. security assessment. IEEE Trans Power Syst 2000;15:1232–9.
[7] Coelho LDS, Mariani VC. Combining of chaotic differential evolution and [19] Park JB, Lee KS, Shin JR, Lee KY. A particle swarm optimization for economic
quadratic programming for economic dispatch optimization with valve point dispatch with non-smooth cost functions. IEEE Trans Power Syst
effect. IEEE Trans Power Syst 2006;21(2):989–96. 2005;20(1):34–42.
[8] Lin CE, Viviani GL. Hierarchical economic dispatch for piecewise quadratic cost [20] Zwe-Lee G. Particle swarm optimization to solving the economic dispatch
functions. IEEE Trans Power Appl Syst 1984;PAS-103(6):1170–5. considering the generator constraints. IEEE Trans Power Syst
[9] Jayabarati T, Sadasivam G. Evolutionary programming based economic 2003;18:1187–95.
dispatch for units with multiple fuel cost functions. Eur Trans Electr Power [21] Price K, Storn R. Differential evolution: numerical optimization made easy. Dr.
2000;10(3):167–70. Dobbs J 1997;22(4):18–24.
[10] Park YM, Won JR, Park JB. A new approach to economic load dispatch based on [22] Storn R, Price KV. Differential evolution-A simple and efficient heuristic for
improved evolutionary programming. Eng Intell Syst Electr Eng Commun June global optimization over continuous spaces. Dr. Dobbs J 1997;11:341–59.
1998;6(2):103–10. [23] Chiang CL. Improved genetic algorithm for power economic dispatch of units
[11] Park JH, Kim YS, Eom IK, Lee KY. Economic load dispatch for piecewise with valve point effects and multiple fuels. IEEE Trans Power Syst
quadratic cost function using Hopfield neural network. IEEE Trans Power Syst 2005;20(4):1690–9.
1993;8:1030–8. [24] Pothiya S, Ngamroo I, Kongprawechnon W. Ant colony optimisation for
[12] Lee KY, Sode-Yome A, Park JH. Adaptive Hopfield neural network for economic economic dispatch problem with non-smooth cost functions. Int J Electr Power
load dispatch. IEEE Trans Power Syst 1998;13(May):519–26. Energy Syst 2010;32:478–87.
[13] Eberhart RC, Shi Y. Comparison between genetic algorithms and [25] Hemamalini S, Simon SP. Artificial bee colony algorithm for economic load
particle swarm optimization. Proc IEEE Int Conf Evol Comput dispatch problem with non-smooth cost functions. Int J Electr Power Compon
1998(May):611–6. Syst 2010;38:786–803.
[14] Kennedy J, Eberhart R. Particle swarm optimization. Proc IEEE Int Conf Neural [26] Pandi VR, Panigrahi BK, Bansal RC, Das S, Mohapatra A. Economic load dispatch
Netw, Australia, Perth 1995;4:1942–8. using hybrid swarm intelligence based harmony search algorithm. Int J Electr
[15] Shi Y, Eberhart RC. A modified particle swarm optimizer. Proc IEEE Int Conf Power Compon Syst 2011;39:751–67.
Evol Comput 1998(May):69–73. [27] Cai JJ, Li Q, Li LX, Peng HP, Yang YX. A fuzzy adaptive chaotic ant swarm
[16] Eberhart RC, Shi Y. Comparing inertia weights and constriction factors in optimization for economic dispatch. Int J Electr Power Energy Syst
particle swarm optimization. Proc Cong Evol Comput 2000:84–8. 2012;34:154–60.

You might also like