A Differential Evolution Algorithm Based On Multi-Population For Economic Dispatch Problems With Valve-Point Effects
A Differential Evolution Algorithm Based On Multi-Population For Economic Dispatch Problems With Valve-Point Effects
A Differential Evolution Algorithm Based On Multi-Population For Economic Dispatch Problems With Valve-Point Effects
2, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2927574
ABSTRACT Economic dispatch (ED) problem with valve-point effects of generation units is a non-smooth
and non-convex problem. This paper presents a differential evolution algorithm based on multi-population
(MPDE) to solve ED problems with valve-point effects. The proposed MPDE algorithm employs the
evolutional methodology of multiple populations to overcome the limitations of the traditional differential
evolution algorithm. In multiple populations, each population has its mutation strategy and parameters to
enhance the searching ability. Moreover, information exchange among multiple populations to increase
the diversity of individuals in a single population. In addition, the algorithm also introduces the normal
distribution function to dynamically adjust the scaling factor and crossover rate to accelerate convergence
speed. The proposed algorithm is tested on the 13-, 40-, 80-, and 140-unit test systems. The simulation
results show that the MPDE algorithm can achieve much smaller variances compared with other intelligent
algorithms. The proposed algorithm has significant performance in accuracy for solving economic dispatch
problems with valve-point effects.
INDEX TERMS Differential evolution, multi-population, economic dispatch problems, valve-point effects.
VOLUME 7, 2019 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ 95585
X. Li et al.: MPDE for ED Problems With Valve-Point Effects
size and a neighbor strategy to form the MCSA algorithm. have improved on the basis of the traditional DE algorithm
Cai et al. [17] proposed a chaotic particle swarm optimiza- and used the improved algorithm to solve the ED problem.
tion algorithm (CPSO), which combines particle swarm opti- Amjady and Sharifzadeh [32] presented a modified differ-
mization algorithm (PSO), adaptive weighting factor and ential evolution algorithm (MDE), where the mutation factor
fusion chaotic local search optimization method to solve ED and the crossover rate are dynamically adjusted, and the
problem. Adarsh et al. [18] proposed a bat algorithm that mutant vector is produced by two mutation operators and a
uses chaotic sequences to enhance its performance, called modified repair operation regulating variables. Bhattacharya
the chaotic bat algorithm (CBA). Park et al. [19] applied and Chattopadhyay [33] introduced a hybrid technique com-
the dynamic search space reduction strategy to the PSO bining differential evolution with biogeography-based opti-
to form the MPSO algorithm that speeds up the optimiza- mization (DE/BBO), which is a combination of DE and BBO
tion process. Jadoun et al. [20] proposed a new Modulated to improve the quality of solution and convergence speed.
Particle Swarm Optimization Algorithm (SMPSO). Unlike Roy et al. [34] proposed a new hybrid difference algorithm
MPSO, SMPSO introduces a truncated sinusoidal contraction (HCR-DE), which combines DE with CRO and utilizes the
function to change the particle velocity for a larger search advantages of HCR and DE to improve the convergence
range. Chaturvedi et al. [21] introduced a time varying speed. Niknam et al. [35] proposed a hybrid algorithm which
acceleration coefficient to PSO algorithm (PSO_TVAC) to combines variable DE with fuzzy adaptive particle swarm
improve the local and global search efficiency and stabil- optimization (FAPSO-VDE) to solve the ED problem with
ity of PSO. He et al. [22] applied the differential evolu- valve-point effects. Zou et al. [36] presented an improved
tion algorithm (DE) and sequential quadratic programming differential evolution algorithm (IDE), where randomiza-
(SQP) techniques to the GA algorithm to form the HGA. tion is incorporated into IDE to overcome the premature
In the HGA algorithm, DE and SQP adjust the GA opti- convergence.
mized solution. Xiong et al. [23] proposed a multi-strategy Srinivasa Reddy and Vaisakh [37] proposed a hybrid shuf-
ensemble biogeography-based optimization (MsEBBO) to fled differential evolution algorithm (SDE), which combines
effectively balance the global exploration and the local the advantages of hybrid hopping algorithm and differential
exploitation. evolution algorithm, and designed a new mutation strat-
The differential evolution algorithm (DE) is an intelli- egy: DE/Memeplexbest/2. The algorithm divides the pop-
gent optimization algorithm that simulates the evolution of ulation into several subpopulations, each subpopulation is
natural organisms, which was proposed by Storn and Price updated by DE/Memeplexbest/2, and after a certain number
in 1995 when studying the Chebyshev polynomial fitting of predefined memetic evolutions, all individuals are col-
problem [24]. The DE algorithm has a strong search abil- lected from each subpopulation and shuffled. Wu et al. [38]
ity, so it has been improved by many some researchers. proposed a multi-population ensemble differential evolu-
Draa et al. [25] proposed a sinusoidal differential evolu- tion algorithm (MPEDE). The population is divided into
tion (SinDE), which uses the new sinusoidal formulas to three smaller indicator subpopulations of the same size
automatically adjust the scaling factor and the crossover rate and a larger reward subpopulation. The three indicator
to balanced global search and local search. Mohamed [26] subpopulations use different mutation strategies, and the
presented an improved differential evolution (IDE) algorithm. reward subpopulation selects the best mutation strategy
The algorithm introduces a new triangular mutation rule and among the three populations for mutation operation. This
adopts a restart mechanism to overcome premature conver- paper proposes a multi-population differential evolution
gence. Li et al. [27] designed a new framework that named algorithm (MPDE), and each population adopts different
HMJCDE by combining MJADE and MCoDE. They are mutation strategies. In the process of evolution, informa-
operated alternatively according to the improvement rate of tion exchanges among the three populations. SDE, MPEDE
the fitness value. Mohamed and Mohamed [28] presented and MPDE algorithms all introduce the concept of multi-
adaptive guided differential evolution algorithm (AGDE) that population, but there are some differences in some respects.
introduces a new mutation scheme to maintain effectively For example, the three algorithms are different in the selec-
the balance between the global search and local search abil- tion of mutation strategies. MPDE chooses three different
ities. Cui et al. [29] presented an adaptive multiple-elites- mutation strategies: DE/rand/1, DE/rand/2 and DE/rand-
guided composite differential evolution algorithm with a to-best/1. The three indicator populations in MPEDE
shift mechanism (AMECoDEs), the purpose of this shift select DE/rand/1, DE/current-to-best/1 and DE/current-to-
mechanism is to solve the problems of search stagnation rand/1 respectively. All subpopulations in the SDE algorithm
and premature convergence. Wu et al. [30] proposed a use the DE/Memeplexbest/2 mutation strategy to update indi-
multi-population based on framework (MPF) to realize the viduals. In addition, in order to improve the diversity of
ensemble of multiple DE variants to derive a new algorithm individuals in the single population, MPDE algorithm uses
named EDEV. Mohamed et al. [31] presented a new semi- a multi-population learning strategy, in which multiple pop-
parameter adaptation approach to effectively adapt the value ulations can learn and exchange information with each other.
of the scaling factor of the differential evolution algorithm to Das et al. [39] used a linearly decreasing strategy to adjust
enhance search capability. In recent years, some researchers the scaling factor. Qin et al. [40] introduced the distribution
C. CROSSOVER OPERATION
xmt generates vtm through mutation operation, and the DE
algorithm performs crossover operation based on xmt and vtm
to generate utm (the trial vector). Compare the random number
and the crossover rate to generate each dimension of the trial
vector. The updating equation is given by:
(
vt If rand(0, 1) ≤ CRtm or n = nrand
t
um,n = m,n t
(9)
xm,n Otherwise
where rand(0, 1) is a uniformly distributed random number
in the interval (0, 1); nrand is a random integer in the interval
[1, D]; CRtm is the crossover rate of individual m at the t − th
generation.
D. SELECTION OPERATION
The DE algorithm uses the greedy selection method to per- FIGURE 1. The framework of population division.
Qt and Qmin = 5; t is the current number of iterations; tmax If the individual m belongs to the population pop2,
is the maximum iteration number. the updating equation is given by:
For the purpose of increasing the individual diversity of
the single population, individuals can select not only pater- vtm
nal individual vectors in their own population, but also the (
x t + Fmt × (xrt2 − xrt3 ) + Fmt × (xrt4 − xrt5 ) If qm ≤ Qt
paternal individual vectors in other populations to perform = rt1
the mutation operation. When qm ≤ Qt , all paternal indi- xj1 + Fmt × (xjt2 − xjt3 ) + Fmt × (xjt4 − xjt5 ) Otherwise
vidual vectors of individual m have to be selected from (14)
its own population. When qm > Qt , the target vector is
selected in the population to which individual m belong, the where r1 , r2 , r3 , r4 , r5 are five random integers in the range
other individual vectors are selected randomly in the three [NP + 1, 2NP ] and r1 6 = r2 6 = r3 6 = r4 6 = r5 ; j1 is
populations. one random integer in the range [NP + 1, 2NP ]; j2 , j3 , j4 , j5
If the individual m belongs to the population pop1, are four different random integers in the range [1, 3NP ] and
the donor vector vtm can be described as follows: j1 6 = j2 6 = j3 6 = j4 6 = j5 .
( If the individual m belongs to the population pop3, the for-
x t + Fmt × (xrt2 − xrt3 ) If qm ≤ Qt mula is expressed as follows:
vm = rt1
t
(13)
xj1 + Fmt × (xjt2 − xjt3 ) Otherwise
vtm
where r1 , r2 , r3 are three random integers in the range [1, NP ], (
t −x t )+F t × (x t −x t ) If q ≤ Qt
x t +Fmt ×(xbest m
and r1 6 = r2 6 = r3 ; j1 is one random integer in the range = rt1 r1 m r2 r3
t −x t )+F t × (x t −x t ) Otherwise
[1, NP ]; j2 , j3 are two random integers in the range [1, 3NP ], xj1 +Fmt × (xbest j1 m j2 j3
and j1 6 = j2 6 = j3 . (15)
TABLE 3. Output power of generators in the best result for the case 1.
TABLE 1. The parameters of the MPDE algorithm for solving the unit ED
problems.
TABLE 4. (Continued) The results obtained by different intelligent algorithms in solving case 1.
population to have too much influence on the other two popu- individuals in the single population and facilitates the indi-
lations, resulting in rapid convergence of three populations to vidual to escape from the local optimum.
local optimum. On the other hand, the greater the number of
consecutive failures of individual continuous update, which B. DYNAMICALLY ADJUST ALGORITHM PARAMETERS
indicating that the individual may fall into local optimum In the DE algorithm, the scaling factor and crossover rate
or that the paternal individuals in the population to which have an important influence on the global search ability and
the individual belongs cannot meet the requirement of the convergence speed of the algorithm. The control parameters
individual to perform the mutation operation, and from other in the traditional DE algorithm are selected from fixed values,
populations, it is possible to find paternal individuals suitable so the diversity of parameters is relatively poor, for some
for the individual to perform the mutation operation. The special problems, the effect of fixed parameters is not very
information exchange mechanism enhances the diversity of good. This paper proposes a method of dynamically adjusting
TABLE 6. Output power of generators in the best result for the case 2.
parameters, in which the control parameters are converted where N (αt , βt ) represents the normal distribution function
from fixed to dynamic. with a mean of αt and a standard deviation of βt ; αmax is the
maximum value of αt ; αmin is the minimum value of αt .
1) DYNAMICALLY ADJUST THE SCALING FACTOR
If the value of the scaling factor is larger, the global search
ability of the algorithm is stronger, if the value of the scaling
factor is smaller, the convergence speed of the algorithm is 2) DYNAMICALLY ADJUST THE CROSSOVER RATE
faster. In the early stages of evolution, a larger scaling factors The crossover rate has a great influence on the diversity of
enhance global search ability of the algorithm, in the later the population. When the value of the crossover rate is large,
stages of evolution, a smaller scaling factor speeds up the it can be seen from the formula (9) that most of the elements
convergence of the algorithm. This paper uses a nonlinear of the trial vector uti come from the donor vector vti , the
decrement method to dynamically adjust the scaling factor. probability of individual changes in the population is greatly
The scaling factor is given by: improved, the algorithm is easy to search for the optimal
solution. If the crossover rate is small, most of the elements
Fmt = N (αt , βt ) (16) of the trial vector uti come from the xit , which is beneficial for
t × (αmax − αmin ) the algorithm to perform a stable search in space. This paper
αt = αmax − βt = 0.1 (17)
tmax uses a nonlinear incremental method to dynamically adjust
FIGURE 6. The convergence characteristics for case 3. FIGURE 7. The convergence characteristics for case 4.
TABLE 10. Output power of generators in the best result for the case 3. B. CONSTRAINTS
1) UNIT POWER INEQUALITY CONSTRAINTS
The output power of each generator unit should meet the
inequality constraint (22), it shall not be greater than the
maximum of power output of the generator unit and not less
than the minimum of power output of the generator unit.
Pmax
j ≥ Pj ≥ Pmin
j (22)
where Pmax
j is the maximum power of the generator
unit j.
2) RAMP-RATE CONSTRAINTS
The actual operating range of all the online units is restricted
by their corresponding ramp rate limits. The ramp-up and
ramp-down constraints can be written as follows:
where P0j is the previous generation of unit j; DRj and URj are
down and up ramp rate limits.
where gj is the number of the POZs for unit j; Plj,k and Puj,k
applying the MPDE algorithm to the ED problem are are the lower and upper bounds of the k-th POZs of unit j.
introduced.
4) SYSTEM POWER EQUALITY CONSTRAINTS
A. OBJECTIVE FUNCTION The total of all power of the generator unit is equal to the
The purpose of the ED is to optimize the power output of sum of load demand and the transmission loss. The equality
generator unit while satisfying the constraints of the power constraints can be formulated as follows:
system, thereby minimizing the total power generation cost.
M
The cost function when considering the valve-point effects X
can be formulated as follows: Pj = PD + PLoss (26)
j=1
M
M X
M M
X
min F = min Fj (Pj ) (20) PLoss =
X
Pj Bji Pi +
X
B0j Pj + B00 (27)
j=1
j=1 i=1 j=1
TABLE 11. The results obtained by different intelligent algorithms in solving case 3.
TABLE 11. (Continued). The results obtained by different intelligent algorithms in solving case 3.
[18]
TABLE 11. (Continued). The results obtained by different intelligent algorithms in solving case 3.
TABLE 13. Output power of generators in the best result for the case 4. 3) EQUALITY CONSTRAINTS HANDLING
This paper takes a new method to deal with equality con-
straints. The difference is distributed to each generation unit
by calculating the difference between the current total output
and the demand output. In this way, not only the genera-
tor unit can be corrected to satisfy the equality constraints,
but also the output power of the generator unit can be
changed less, and the influence due to the variation of the
output power can be reduced. The specific steps are analyzed
below:
Step 1: If Pj violates formula (24) or satisfies formula (25),
set the transition variable Tj = 0, else Tj = Pj .
Step 2: Calculate the difference 1 between the current total
output and the demand output.
M
X
1= Pj − PD − PLoss (30)
j=1
information exchange among populations. The Fmt is obtained DHS and HIS, IDE algorithm in terms of mean cost, max-
by using (16) and (17). imum cost, standard deviation. The maximum, minimum,
Step 5: Obtain CRtm according to formulas (18) and (19). mean, and standard deviation of the MPDE algorithm are
Perform crossover operation and generate the trial vector. the best. Although the running time of the DEC-SQP algo-
Modify the trial vector that violates equality constraints and rithm is faster than the running time of the MPED algorithm,
inequality constraints. Evaluate the fitness value of the indi- the MPDE algorithm is far superior to the DEC-SQP algo-
vidual according to formulas (20) and (21). rithm in terms of accuracy. For DataSet2(13−unit) ), MABC,
Step6: Perform selection operation and update qm by FAPSO-VDE, MsEBBO and MPDE can obtain the lowest
using (11). cost. The maximum, minimum and mean values of MPDE
Step 7: If the maximum number of iterations is reached, are equal. The stability of MPDE is stronger than MABC,
the algorithm terminates. Output the global optimal fitness FAPSO-VDE and MsEBBO. In addition, the running time
value, else go to step 4. of MPDE is shorter than that of MABC and FAPSO-VDE.
Although ST-HDE, NSEO and CBA are superior in time to
MPDE, MPDE is superior to them in other respects. By com-
V. SIMULATION RESULTS AND ANALYSIS paring the results of the MPDE algorithm with other intelli-
In order to test the comprehensive performance of the gent algorithms, the proposed algorithm has great advantages
improved algorithm, we tested six cases of 13-, 40-, 80- in solving case 1.
and 140- unit test systems with this algorithm. All cases are Table 5 depicts the ranks computed through the Friedman
coded in C++ and implemented in Visual Studio 2013, which test [65], [66] and the Friedman test is implemented by
are tested on a PC with Intel i5 2.3GHz processor, 4GB of the software SPSS. Calculate the fuel consumption cost of
RAM and Windows 10 Professional, each case runs 50 times each generator unit for each algorithm (The cost is accurate
independently and we compare them with the results of other to 0.0001$/h.). Rank the algorithms for each generator unit,
intelligent algorithms. and average the ranks obtained in all generator units for
each algorithm. The algorithm with the smaller mean ranking
A. SETTING ALGORITHM PARAMETERS performs better. As can be seen from Table 5, the MPDE ranks
The MPDE algorithm is sensitive to parameters. The parame- the first. MPDE is superior to all algorithms.
ters of the MPDE algorithm for solving the ED problems are
shown in Table 1 and Table 2. The brief introduction to the
test cases is also shown in Table 2, including the valve-point C. CASE 2
effects (VPE), transmission losses (TL), prohibited operating Case 2 is one of the 13- unit test system, which considers
zones (POZs) and ramp rate limits (RRL). the valve-point effects and the transmission losses. Its load
demand is 2520 MW There are four sets of data are used
for the case 2. The first data set (DataSet1(13−unit) ) uses for
B. CASE 1 the loss coefficients B are given in Table A-1 of Appendix A
Case 1 is one case in the 13- unit test system, which considers and the cost coefficients are refer to [4]. The second data set
the valve-point effects, and its load demand is 1800 MW. (DataSet2(13−unit) ) uses for the loss coefficients B are given
there are two sets of data are used for the fuel consumption in Table A-I of Appendix A with correction B0,11 = 0.0017
cost coefficients. The first data set (DataSet1(13−unit) ) uses for and the cost coefficients are refer to [4]. The third data
the fuel consumption cost coefficients and generation limits set (DataSet3(13−unit) ) uses for the loss coefficients B are
are refer to [44]. The second data set (DataSet2(13−unit) ) given in Table A-I of Appendix A with correction B1,10 =
uses for the fuel consumption cost coefficients and generation 0.0005 B00 = 0.0055 and the fuel consumption cost coeffi-
limits are refer to [4]. The difference between them is the e cients are refer to [4]. The fourth data set (DataSet4(13−unit) )
fuel consumption cost coefficient of 3-th unit. uses for the loss coefficients B are given in Table A-I of
Case 1 was run independently for 50 times with the MPDE Appendix A with correction B1,10 = 0.0005 B00 = 0.0055
algorithm. Figure 4 shows the convergence characteristics of and the cost coefficients are refer to [44].
the MPDE algorithm when solving case 1. Table 3 shows Case 2 was run independently for 50 times with the MPDE
the output of each generator unit at the lowest total algorithm. Figure 5 shows the convergence characteristics of
generation cost with the different fuel consumption cost the MPDE algorithm when solving case 2. Table 6 shows the
coefficients. output of each generator unit.
Table 4 shows the results of the MPDE algorithm and other Table 7 shows the results of the MPDE algorithm and
intelligent algorithms to test case 1, including the minimum other intelligent algorithms to test case 2, including the min-
cost, mean cost, maximum cost, standard deviation value, imum cost, mean cost, maximum cost, standard deviation
time and NFE (NFE= NP ∗ tmax ). As can be seen from value, time and NFE. As can be seen from Table 7, for
Table 6, for DataSet1(13−unit) ), HIS, CSOMA, DHS, IDE and DataSet1(13−unit) ), MABC, MCSA, MSOS and MPDE can
MPDE can get the lowest cost and DHS takes less time, get the lowest cost and their maximum, minimum and mean
but the MPDE algorithm is far superior to the CSOMA, values are equal. In terms of time, MPDE is better than
TABLE 14. The results obtained by different intelligent algorithms in solving case 4.
TABLE 15. Output power of generators in the best result for the case 5.
D. CASE 3
Case 3 is one case in the 40- unit test system, which considers
the valve-point effects, and its load demand is 10500 MW.
There are two sets of data are used for the fuel consumption
cost coefficients. The first data set (DataSet1(40−unit) ) uses for
the fuel consumption cost coefficients and generation limits
are refer to [4]. The second data set (DataSet2(40−unit) ) uses
for the fuel consumption cost coefficients and generation
limits are refer to [48]. The difference between them is the e
fuel consumption cost coefficient of 7-th unit.
Case 3 was run independently for 50 times with the MPDE
algorithm. Figure 6 shows the convergence characteristics of
the MPDE algorithm when solving case 3. Table 10 shows
the output of each generator unit at the lowest total generation
cost.
Table 11 shows the results of the MPDE algorithm and
other intelligent algorithms to test case 3, including the min-
MABC and MSOS, but it is a bit worse than MCSA. Although imum cost, mean cost, maximum cost, standard deviation
HDE, STHDE, ICA-PSO, value, time and NFE. It can be seen from Table 11, for
BBO, SOS, CS and DE/BBO have strong competitiveness DataSet1(40−unit) ), only EMA, MCSA, NSEO and MPDE
in time, they are worse in accuracy. For DataSet2(13−unit) ), can get the lowest cost, MCSA takes less time than MPDE,
MSOS and MPDE can get the lowest cost and their but MPDE has good performance in terms of the mean
maximum, minimum and mean values are equal, and cost, the maximum cost and the standard deviation. Rela-
MPDE takes less time than MSOS. For DataSet3(13−unit) tively speaking, MPDE is relatively stable. HDE, DE/BBO,
and DataSet4(13−unit) , the MPDE has the best perfor- FAPSO–VDE and CBA are better than MPDE in time,
mance in terms of the maximum, minimum, mean val- but in other respects, MPDE is superior to them. For
ues and time, the MPDE algorithm is significantly better DataSet2(40−unit) ), DHS and MPDE can get the lowest cost
than other algorithms. Table 8 depicts the ranks computed and DHS is better than MPDE in terms of time, but the MPDE
through the Friedman test. As can be seen from Table 8, has the best performance in terms of the maximum, minimum
for DataSet1(13−unit) , DataSet3(13−unit) and DataSet4(13−unit) and mean value. By comparing the results of the MPDE algo-
the MPDE has the smallest mean ranking so it ranks rithm with other intelligent algorithms, the proposed algo-
the first. Table 9 depicts the Friedman ranks test for the rithm outperforms other intelligent algorithms in solving case
DataSet2(13−unit). As can be seen from Table 9, RMPDE10 > 3. Table 12 depicts the ranks computed through the Friedman
RFPSOGSA , RMPDE > RFPSOGSA and RMPDE > RFPSOGSA ,
10 12 12 13 13 test. As can be seen from Table 12, for DataSet1(40−unit) and
only RMPDE
11 < RFPSOGSA
11 , so the mean rank of MPDE is DataSet2(40−unit) the MPDE has the smallest mean ranking so
bigger than that of FPSOGSA. But T10 +T11 + T12 + T13 = it ranks the first.
TABLE 16. The results obtained by different intelligent algorithms in solving case 5.
E. CASE 4 Case 4 was run independently for 50 times with the MPDE
Case 4 is one case in the 40- unit test system, which considers algorithm. Figure 7 shows the convergence characteristics of
the valve-point effects and the transmission losses. Its load the MPDE algorithm when solving case 4. Table 13 shows
demand is 10500 MW. there are two sets of data are used the output of each generator unit at the lowest total generation
for the fuel consumption cost coefficients in the 40- unit test cost.
system. The first data set (DataSet1(40−unit) ) uses for the cost Table 14 shows the results of the MPDE algorithm and
coefficients and generation limits are refer to [5]. The second other intelligent algorithms to test case 4, including the
data set (DataSet2(40−unit) ) uses for the fuel consumption minimum cost, mean cost, maximum cost, standard devia-
cost coefficients and generation limits are refer to [48]. The tion value, time, and NFE. As can be seen from Table 14,
B-loss coefficients are refer to [72]. for DataSet1(40−unit) ), the minimum cost, mean cost and
TABLE 17. Output power of generators in the best result for the case 6.
F. CASE 5
Case 5 is one case in the 80- unit test system, which considers
the valve-point effects and the prohibited operating zones.
Its load demand is 21000 MW. This large-scale system is
created by expanding the 40 generating unit system. The fuel
consumption cost coefficients and generation limits are refer
to [4] and The prohibited operating zones are refer to [90].
Case 5 was run independently for 50 times with the MPDE
algorithm. The lowest generating cost is 242794.729463$/h.
Figure 8 shows the convergence characteristic of the MPDE
algorithm when solving case 5. Table 15 shows the output of
each generator unit at the lowest total generation cost.
Table 16 shows the results of the MPDE algorithm and
other intelligent algorithms to test cases 5, including the
minimum cost, mean cost, maximum cost, standard deviation
value, time, and NFE. As can be seen from Table 16, the min-
imum cost, mean cost and the maximum cost of MPDE are is 49342 WM. All the data of valve point cost coefficients and
superior to the results calculated by other intelligent algo- generation limits for each unit are refer to [92].
rithms. The stability and accuracy of MPDE are stronger Case 6 was run independently for 50 times with the MPDE
than all intelligent algorithms in the table. Some intelligent algorithm. The lowest generating cost is 1559708.807042$/h.
algorithms take less time than MPDE, but MPDE is more Figure 9 shows the convergence characteristic of the MPDE
accurate than them in terms of minimum cost, mean cost and algorithm when solving the case 6. 1559708.807042$/h is the
maximum cost. the statistical results obtained by MPDE are minimum value that can be found so far. It can be seen that
highly competitive compared to these by the other intelligent the MPDE algorithm has a good performance in solving the
algorithms. case 6. Table 17 shows the output of each generator unit at
the lowest total generation cost.
G. CASE 6 Table 18 shows the results of the MPDE algorithm and
Case 6 is one of the 140- unit test system, which considers the other intelligent algorithms to test cases 6, including the
valve-point effects and the ramp rate limits. Its load demand minimum cost, mean cost, maximum cost, standard devia-
TABLE 18. The results obtained by different intelligent algorithms in solving case 6.
tion value, time and NFE. As can be seen from Table 18, OWGO, CCPSO, KGMO, and GWO have an advantage in
MPDE can get the lowest cost. The maximum, minimum, terms of time and OGWO is better than MPDE in the standard
and mean of the MPDE algorithm are the best. Although deviation, they are not as good as MPDE in terms of accuracy,
MPDE has a strong competition in accuracy. the proposed [13] G. P. Singh and A. Singh, ‘‘Comparative study of krill herd, firefly and
algorithm has great advantages in solving case 6. cuckoo search algorithms for unimodal and multimodal optimization,’’ Int.
J. Intell. Syst. Appl., vol. 6, no. 3, pp. 35–49, Feb. 2014.
[14] R. Kumar, A. Sadu, R. Kumar, and S. K. Panda, ‘‘A novel multi-objective
VI. CONCLUSION directed bee colony optimization algorithm for multi-objective emission
A differential evolution algorithm based on multi-population constrained economic power dispatch,’’ Int. J. Elect. Power Energy Syst.,
vol. 43, no. 1, pp. 1241–1250, Dec. 2015.
(MPDE) is proposed to solve the ED problem with [15] P. K. Roy, S. P. Ghoshal, and S. S. Thakur, ‘‘Biogeography-based optimiza-
valve-point effects. The MPDE algorithm applies a tion for economic load dispatch problems,’’ Electr. Power Compon. Syst.,
multi-population strategy to the traditional differential evolu- vol. 38, no. 2, pp. 166–181, Dec. 2009.
[16] J. Zhao, S. Liu, M. Zhou, X. Guo, and L. Qi, ‘‘Modified cuckoo search
tion algorithm. In order to enhance its search ability, the muta- algorithm to solve economic power dispatch optimization problems,’’
tion strategy and algorithm parameter in each population are IEEE/CAA J. Autom. Sinica, vol. 5, no. 4, pp. 794–806, Jul. 2018.
also different. During the evolution process, individuals in [17] C. Jiejin, M. Xiaoqian, L. Lixiang, and P. Haipeng, ‘‘Chaotic particle
swarm optimization for economic dispatch considering the generator con-
different populations can also learn from individuals in other straints,’’ Energy Convers. Manage., vol. 48, no. 2, pp. 645–653, Feb. 2007.
populations, this method of information exchange enhances [18] B. R. Adarsh, T. Raghunathan, T. Jayabarathi, and X.-S. Yang, ‘‘Economic
the diversity of a single population. Moreover, the MPDE dispatch using chaotic bat algorithm,’’ Energy, vol. 96, pp. 666–675,
Feb. 2016.
algorithm adopts the normal distribution function to dynami-
[19] J.-B. Park, K.-S. Lee, J.-R. Shin, and K. Y. Lee, ‘‘A particle swarm
cally adjust the scaling factor and crossover rate. The MPDE optimization for economic dispatch with nonsmooth cost functions,’’ IEEE
algorithm is tested on the 13-, 40-, 80- and 140- unit test Trans. Power Syst., vol. 20, no. 1, pp. 34–42, Feb. 2005.
systems. Statistical results are compared with the reported [20] V. K. Jadoun, N. Gupta, K. R. Niazi, and A. Swarnkar, ‘‘Modulated particle
swarm optimization for economic emission dispatch,’’ Int. J. Elect. Power
results in literature, the MPDE algorithm has better accuracy Energy Syst., vol. 73, pp. 80–88, Dec. 2015.
and robustness than other intelligent algorithms, and can [21] K. T. Chaturvedi, M. Pandit, and L. Srivastava, ‘‘Particle swarm opti-
provide satisfactory global optimization solutions. Therefore, mization with time varying acceleration coefficients for non-convex eco-
nomic power dispatch,’’ Int. J. Electr. Power Energy Syst., vol. 31, no. 6,
the MPDE algorithm is a very suitable tool for solving the ED pp. 249–257, Jul. 2009.
problem with the valve point effects. [22] D. He, F. Wang, and Z. Mao, ‘‘A hybrid genetic algorithm approach based
on differential evolution for economic dispatch with valve-point effect,’’
Int. J. Electr. Power Energy Syst., vol. 30, no. 1, pp. 31–38, Jan. 2008.
APPENDIX A [23] G. Xiong, D. Shi, and X. Duan, ‘‘Multi-strategy ensemble biogeography-
See Table A-1. based optimization for economic dispatch problems,’’ Appl. Energy,
vol. 111, pp. 801–811, Nov. 2013.
[24] R. Storn and K. Price, ‘‘Differential evolution-a simple and efficient heuris-
REFERENCES tic for global optimization over continuous spaces,’’ J. Global Optim.,
[1] D. C. Walters and G. B. Sheble, ‘‘Genetic algorithm solution of economic vol. 11, no. 4, pp. 341–359, Dec. 1997.
dispatch with valve point loading,’’ IEEE Trans. Power Syst., vol. 8, no. 3, [25] A. Draa, S. Bouzoubia, and I. Boukhalfa, ‘‘A sinusoidal differential evo-
pp. 1325–1332, Aug. 1993. lution algorithm for numerical optimisation,’’ Appl. Soft Comput., vol. 27,
[2] R. A. Jabr, A. H. Coonick, and B. J. Cory, ‘‘A homogeneous linear program- pp. 99–126, Feb. 2015.
ming algorithm for the security constrained economic dispatch problem,’’ [26] A. W. Mohamed, ‘‘An improved differential evolution algorithm with
IEEE Trans. Power Syst., vol. 15, no. 3, pp. 930–936, Aug. 2000. triangular mutation for global numerical optimization,’’ Comput. Ind. Eng.,
[3] J.-Y. Fan and L. Zhang, ‘‘Real-time economic dispatch with line flow and vol. 85, pp. 359–375, Jul. 2015.
emission constraints using quadratic programming,’’ IEEE Trans. Power [27] G. Li, Q. Lin, L. Cui, Z. Du, Z. Liang, J. Chen, N. Lu, and Z. Ming,
Syst., vol. 13, no. 2, pp. 320–325, May 1998. ‘‘A novel hybrid differential evolution algorithm with modified CoDE and
[4] N. Sinha, R. Chakrabarti, and P. K. Chattopadhyay, ‘‘Evolutionary pro- JADE,’’ Appl. Soft Comput., vol. 47, pp. 577–599, Oct. 2016.
gramming techniques for economic load dispatch,’’ IEEE Trans. Evol. [28] A. W. Mohamed and A. K. Mohamed, ‘‘Adaptive guided differential
Comput., vol. 7, no. 1, pp. 83–94, Feb. 2003. evolution algorithm with novel mutation for numerical optimization,’’ Int.
[5] N. Noman and H. Iba, ‘‘Differential evolution for economic load dis- J. Mach. Learn. Cybern., vol. 10, pp. 253–277, Feb. 2019.
patch problems,’’ Electr. Power Syst. Res., vol. 78, no. 8, pp. 1322–1331, [29] L. Cui, G. Li, Z. Zhu, Q. Lin, K.-C. Wong, J. Chen, N. Lu, and J. Lu, ‘‘Adap-
Aug. 2008. tive multiple-elites-guided composite differential evolution algorithm with
[6] Z.-L. Gaing, ‘‘Particle swarm optimization to solving the economic dis- a shift mechanism,’’ Inf. Sci., vol. 422, pp. 122–143, Jan. 2018.
patch considering the generator constraints,’’ IEEE Trans. Power Syst., [30] G. Wu, X. Shen, H. Li, H. Chen, A. Lin, and P. N. Suganthan, ‘‘Ensemble of
vol. 18, no. 3, pp. 1187–1195, Aug. 2003. differential evolution variants,’’ Inf. Sci., vol. 423, pp. 172–186, Jan. 2018.
[7] Y. H. Song, C. S. Chou, and T. J. Stonham, ‘‘Combined heat and power [31] A. W. Mohamed, A. A. Hadi, A. M. Fattouh, and K. M. Jambi, ‘‘LSHADE
economic dispatch by improved ant colony search algorithm,’’ Electr. with semi-parameter adaptation hybrid with CMA-ES for solving CEC
Power Syst. Res., vol. 52, no. 2, pp. 115–121, Nov. 1999. 2017 benchmark problems,’’ in Proc. IEEE Congr. Evol. Comput. (CEC),
[8] P.-H. Chen and H.-C. Chang, ‘‘Large-scale economic dispatch by genetic Jun. 2017, pp. 145–152.
algorithm,’’ IEEE Trans. Power Syst., vol. 10, no. 4, pp. 1919–1926, [32] N. Amjady and H. Sharifzadeh, ‘‘Solution of non-convex economic dis-
Nov. 1995. patch problem considering valve loading effect by a new modified differ-
[9] K. P. Wong and C. C. Fung, ‘‘Simulated annealing based economic dispatch ential evolution algorithm,’’ Int. J. Elect. Power Energy Syst., vol. 32, no. 8,
algorithm,’’ IEE Proc. C Gener., Transmiss. Distrib., vol. 140, no. 6, pp. 893–903, Oct. 2010.
pp. 509–515, Nov. 1993. [33] A. Bhattacharya and P. K. Chattopadhyay, ‘‘Hybrid differential evolution
[10] R. Naresh, J. Dubey, and J. Sharma, ‘‘Two-phase neural network based with biogeography-based optimization for solution of economic load dis-
modelling framework of constrained economic load dispatch,’’ IEE Proc. patch,’’ IEEE Trans. Power Syst, vol. 25, no. 4, pp. 1955–1964, Nov. 2010.
Gener., Transmiss. Distrib., vol. 151, no. 3, pp. 373–378, May 2004. [34] P. K. Roy, S. Bhui, and C. Paul, ‘‘Solution of economic load dispatch using
[11] W.-M. Lin, F.-S. Cheng, and M.-T. Tsay, ‘‘An improved tabu search for hybrid chemical reaction optimization approach,’’ Appl. Soft Comput. J.,
economic dispatch with multiple minima,’’ IEEE Power Eng. Rev., vol. 17, vol. 24, pp. 109–125, Nov. 2014.
no. 1, pp. 108–112, Feb. 2002. [35] T. Niknam, H. D. Mojarrad, and H. Z. Meymand, ‘‘A novel hybrid par-
[12] B. K. Panigrahi and V. R. Pandi, ‘‘Bacterial foraging optimisation: ticle swarm optimization for economic dispatch with valve-point load-
Nelder-Mead hybrid algorithm for economic load dispatch,’’ IET Gener., ing effects,’’ Energy Convers. Manage., vol. 52, no. 4, pp. 1800–1809,
Transmiss. Distrib., vol. 2, no. 4, pp. 556–565, Jul. 2008. Apr. 2011.
[36] D. Zou, S. Li, G.-G. Wang, Z. Li, and H. Ouyang, ‘‘An improved dif- [57] C.-L. Chiang, ‘‘Improved genetic algorithm for power economic dispatch
ferential evolution algorithm for the economic load dispatch problems of units with valve-point effects and multiple fuels,’’ IEEE Trans. Power
with or without valve-point effects,’’ Appl. Energy, vol. 181, pp. 375–390, Syst., vol. 20, no. 4, pp. 1690–1699, Nov. 2005.
Nov. 2016. [58] X. He, Y. Rao, and J. Huang, ‘‘A novel algorithm for economic
[37] A. S. Reddy and K. Vaisakh, ‘‘Shuffled differential evolution for large load dispatch of power systems,’’ Neurocomputing, vol. 171, no. 1,
scale economic dispatch,’’ Electr. Power Syst. Res., vol. 96, pp. 237–245, pp. 1454–1461, Jan. 2016.
Mar. 2013. [59] J. Cai, Q. Li, L. Li, H. Peng, and Y. Yang, ‘‘A hybrid FCASO-SQP method
[38] G. Wu, R. Mallipeddi, P. N. Suganthan, R. Wang, and H. Chen, ‘‘Differ- for solving the economic dispatch problems with valve-point effects,’’
ential evolution with multi-population based ensemble of mutation strate- Energy, vol. 38, no. 1, pp. 346–353, Feb. 2012.
gies,’’ Inf. Sci., vol. 329, pp. 329–345, Feb. 2016. [60] M. S. P. Subathra, S. E. Selvan, T. A. A. Victoire, A. H. Christinal, and
[39] S. Das, A. Konar, and U. K. Chakraborty, ‘‘Two improved differential U. Amato, ‘‘A hybrid with cross-entropy method and sequential quadratic
evolution schemes for faster global search,’’ in Proc. 7th Annu. Conf. programming to solve economic load dispatch problem,’’ IEEE Syst. J.,
Genetic Evol. Comput. (GECCO), Washington, DC, USA, Jun. 2005, vol. 9, no. 3, pp. 1031–1044, Sep. 2015.
pp. 991–998. [61] T. Niknam, ‘‘A new fuzzy adaptive hybrid particle swarm optimization
[40] A. K. Qin, V. L. Huang, and P. N. Suganthan, ‘‘Differential evolution algorithm for non-linear, non-smooth and non-convex economic dispatch
algorithm with strategy adaptation for global numerical optimization,’’ problem,’’ Appl. Energy, vol. 87, no. 1, pp. 327–339, Jan. 2010.
IEEE Trans. Evol. Comput., vol. 13, no. 2, pp. 398–417, Apr. 2009. [62] M. N. Alam and T. R. Chelliah, ‘‘A new sensitivity evaluation based
[41] J. Zhang and A. C. Sanderson, ‘‘JADE: Adaptive differential evolution optimization algorithm for economic load dispatch problems,’’ in Proc. Int.
with optional external archive,’’ IEEE Trans. Evol. Comput., vol. 13, no. 5, Conf. Energy Efficient Technol. Sustainability, Apr. 2013, pp. 995–1000.
pp. 945–958, Oct. 2009. [63] H. Da-kuo, W. Fu-li, and M. Zhi-Zhong, ‘‘Hybrid genetic algorithm for
[42] M. M. Ali and A. Törn, ‘‘Population set-based global optimization algo- economic dispatch with valve-point effect,’’ Electr. Power Syst. Res.,
rithms: Some modifications and numerical studies,’’ Comput. Oper. Res., vol. 78, no. 4, pp. 626–633, Apr. 2008.
vol. 31, no. 10, pp. 1703–1725, Sep. 2004. [64] D. C. Secui, ‘‘A new modified artificial bee colony algorithm for the
[43] N. Ghorbani and E. Babaei, ‘‘Exchange market algorithm for economic economic dispatch problem,’’ Energy Convers. Manage., vol. 89, no. 1,
load dispatch,’’ Int. J. Elect. Power Energy Syst., vol. 75, pp. 19–27, pp. 43–62, Jan. 2015.
Feb. 2016. [65] J. Derrac, S. García, D. Molina, and F. Herrera, ‘‘A practical tutorial on
the use of nonparametric statistical tests as a methodology for comparing
[44] L. dos Santos Coelho and V. C. Mariani, ‘‘Particle swarm approach based
evolutionary and swarm intelligence algorithms,’’ Swarm Evol. Comput.,
on quantum mechanics and harmonic oscillator potential well for economic
vol. 1, no. 1, pp. 3–18, Mar. 2011.
load dispatch with valve-point effects,’’ Energy Convers. Manage., vol. 49,
no. 11, pp. 3080–3085, Nov. 2008. [66] A. W. Mohamed, A. A. Hadi, and K. M. Jambi, ‘‘Novel mutation
strategy for enhancing SHADE and LSHADE algorithms for global
[45] L. dos Santos Coelho, R. C. T. Souza, and V. C. Mariani, ‘‘Improved
numerical optimization,’’ Swarm Evol. Comput., to be published. doi:
differential evolution approach based on cultural algorithm and diversity
10.1016/j.swevo.2018.10.006.
measure applied to solve economic load dispatch problems,’’ Math. Com-
[67] S. K. Wang, J. P. Chiou, and C. W. Liu, ‘‘Non-smooth/non-convex
put. Simul., vol. 79, no. 10, pp. 3136–3147, Jun. 2009.
economic dispatch by a novel hybrid differential evolution algorithm,’’
[46] L. dos Santos Coelho and V. C. Mariani, ‘‘An efficient cultural
IET Gener., Transmiss. Distrib., vol. 1, no. 5, pp. 793–803, Sep. 2007.
self-organizing migrating strategy for economic dispatch optimization
[68] K. Bhattacharjee, A. Bhattacharya, and S. H. N. Dey, ‘‘Oppositional
with valve-point effect,’’ Energy Convers. Manage., vol. 51, no. 12,
Real Coded Chemical Reaction Optimization for different economic dis-
pp. 2580–2587, Dec. 2010.
patch problems,’’ Int. J. Elect. Power Energy Syst., vol. 55, pp. 378–391,
[47] L. dos Santos Coelho and V. C. Mariani, ‘‘An improved harmony search Feb. 2014.
algorithm for power economic load dispatch,’’ Energy Convers. Manage.,
[69] D. C. Secui, ‘‘A modified Symbiotic Organisms Search algorithm for
vol. 50, no. 10, pp. 2522–2526, Oct. 2009.
large scale economic dispatch problem with valve-point effects,’’ Energy,
[48] L. dos Santos Coelho and V. C. Mariani, ‘‘Combining of chaotic differential vol. 113, no. 15, pp. 366–384, Oct. 2016.
evolution and quadratic programming for economic dispatch optimiza- [70] X.-S. Yang, Nature-Inspired Optimization Algorithms. Amsterdam,
tion with valve-point effect,’’ IEEE Trans. Power Syst., vol. 21, no. 2, The Netherlands: Elsevier, 2014.
pp. 989–996, May 2006.
[71] S. Duman, N. Yorukeren, and I. H. Altas, ‘‘A novel modified hybrid
[49] Y. Labbi, D. Ben Attous, and B. Mahdad, ‘‘Artificial bee colony optimiza- PSOGSA based on fuzzy logic for non-convex economic dispatch prob-
tion for economic dispatch with valve point effect,’’ Front Energy, vol. 38, lem with valve-point effect,’’ Int. J. Elect. Power Energy Syst., vol. 64,
no. 7, pp. 786–803, May 2010. pp. 121–135, Jan. 2015.
[50] B. Mohammadi-Ivatloo, A. Rabiee, A. Soroudi, and M. Ehsan, ‘‘Iteration [72] A. K. Barisal and R. C. Prusty, ‘‘Large scale economic dispatch of power
PSO with time varying acceleration coefficients for solving non-convex systems using oppositional invasive weed optimization,’’ Appl. Soft Com-
economic dispatch problems,’’ Int. J. Electr. Power Energy Syst , vol. 42, put., vol. 29, pp. 122–137, Apr. 2015.
no. 1, pp. 508–516, Nov. 2012. [73] M. Pradhan, P. K. Roy, and T. Pal, ‘‘Oppositional based grey wolf opti-
[51] J. G. Vlachogiannis and K. Y. Lee, ‘‘Economic load dispatch—A com- mization algorithm for economic dispatch problem of power system,’’ Ain
parative study on heuristic optimization techniques with an improved Shams Eng. J., vol. 9, no. 4, pp. 2015–2025, Dec. 2018.
coordinated aggregation-based PSO,’’ IEEE Trans. Power Syst., vol. 25, [74] A. I. Selvakumar and K. Thanushkodi, ‘‘A new particle swarm optimization
no. 1, pp. 591–592, Feb. 2010. solution to nonconvex economic dispatch problems,’’ IEEE Trans. Power
[52] M. A. Al-Betar, M. A. Awadallah, A. T. Khader, and A. L. Bolaji, Syst., vol. 22, no. 1, pp. 42–51, Feb. 2007.
‘‘Tournament-based harmony search algorithm for non-convex economic [75] M. Kumar and J. S. Dhillon, ‘‘Hybrid artificial algae algorithm for eco-
load dispatch problem,’’ Appl. Soft Comput., vol. 47, pp. 449–459, nomic load dispatch,’’ Appl. Soft Comput., vol. 71, pp. 89–109, Oct. 2018.
Oct. 2016. [76] N. Amjady and H. Nasiri-Rad, ‘‘Economic dispatch using an efficient real-
[53] L. Wang and L.-P. Li, ‘‘An effective differential harmony search algorithm coded genetic algorithm,’’ IET Gener., Transmiss. Distrib., vol. 3, no. 3,
for the solving non-convex economic load dispatch problems,’’ Int. J. Elect. pp. 266–278, Mar. 2009.
Power Energy Syst., vol. 44, no. 1, pp. 832–843, Jan. 2013. [77] N. Amjady and H. Nasiri-Rad, ‘‘Solution of nonconvex and nonsmooth
[54] V. Hosseinnezhad, M. Rafiee, M. Ahmadian, and M. T. Ameli, ‘‘Species- economic dispatch by a new Adaptive Real Coded Genetic Algorithm,’’
based quantum particle swarm optimization for economic load dispatch,’’ Expert Syst. Appl., vol. 37, no. 7, pp. 5239–5245, Jul. 2010.
Int. J. Elect. Power Energy Syst., vol. 63, pp. 311–322, Dec. 2014. [78] M. Ghasemi, M. Taghizadeh, S. Ghavidel, and A. Abbasian, ‘‘Colonial
[55] T. A. A. Victoire and A. E. Jeyakumar, ‘‘Hybrid PSO–SQP for economic competitive differential evolution: An experimental study for optimal eco-
dispatch with valve-point effect,’’ Electr. Power Syst. Res., vol. 71, no. 1, nomic load dispatch,’’ Appl. Soft Comput., vol. 40, pp. 342–363, Mar. 2016.
pp. 51–59, Sep. 2004. [79] M. Moradi-Dalvand, B. Mohammadi-Ivatloo, A. Najafi, and A. Rabiee,
[56] W.-M. Lin, H.-J. Gow, and M.-T. Tsai, ‘‘Combining of direct search and ‘‘Continuous quick group search optimizer for solving non-convex eco-
signal-to-noise ratio for economic dispatch optimization,’’ Energy Con- nomic dispatch problems,’’ Electr. Power Syst. Res., vol. 93, pp. 93–105,
vers. Manage., vol. 52, no. 1, pp. 487–493, Jan. 2011. Dec. 2012.
[80] M. Gholamghasemi, E. Akbari, M. B. Asadpoor, and M. Ghasemi, ‘‘A new XUEPING LI received the M.Eng. degree in elec-
solution to the non-convex economic load dispatch problems using phasor trical engineering and the Ph.D. degree in power
particle swarm optimization,’’ Appl. Soft Comput., vol. 79, pp. 111–124, electronics and power transmission from Yan-
Jun. 2019. shan University, Qinhuangdao, China, in 2009
[81] J.-B. Park, Y.-W. Jeong, J.-R. Shin, and K. Y. Lee, ‘‘An improved particle and 2013, respectively. He is currently with the
swarm optimization for nonconvex economic dispatch problems,’’ IEEE School of Electrical Engineering, Yanshan Uni-
Trans. Power Syst., vol. 25, no. 1, pp. 156–166, Feb. 2010. versity. His research interests include economic
[82] D. Aydin and S. Özyön, ‘‘Solution to non-convex economic dispatch
operation and control of power system, power grid
problem with valve point effects by incremental artificial bee colony with
state estimation, and inspired optimization algo-
local search,’’ Appl. Soft Comput., vol. 13, no. 5, pp. 2456–2466, May 2013.
[83] R. P. Parouha and K. N. Das, ‘‘DPD: An intelligent parallel hybrid rithms and their application on various optimiza-
algorithm for economic load dispatch problems with various practical tion problems.
constraints,’’ Expert Syst. Appl., vol. 63, pp. 295–309, Nov. 2016.
[84] H. Shayeghi and A. Ghasemi, ‘‘A modified artificial bee colony based on
chaos theory for solving non-convex emission/economic dispatch,’’ Energy
Convers. Manage., vol. 79, pp. 344–354, Mar. 2014.
[85] A. Ghasemi, ‘‘A fuzzified multi objective interactive honey bee mating
optimization for environmental/economic power dispatch with valve point HONGJIE ZHANG received the B.Eng. degree in
effect,’’ Int. J. Elect. Power Energy. Syst., vol. 49, pp. 308–321, Jul. 2013. electric power systems and automation from the
[86] D. Singh and J. S. Dhillon, ‘‘Ameliorated grey wolf optimization for eco-
Hebei Normal University of Science and Technol-
nomic load dispatch problem,’’ Energy, vol. 169, pp. 398–419, Feb. 2019.
ogy, Qinhuangdao, China, in 2017. He is currently
[87] M. Singh and J. S. Dhillon, ‘‘Multiobjective thermal power dispatch using
opposition-based greedy heuristic search,’’ Int. J. Elect. Power Energy pursuing the M.Eng. degree with the Department
Syst., vol. 82, pp. 339–353, Nov. 2016. of Electric Power Systems and Automation, Yan-
[88] I. Ciornei and E. Kyriakides, ‘‘A GA-API solution for the economic shan University. His research interests include
dispatch of generation in power system operation,’’ IEEE Trans. Power optimization algorithm and economic operation,
Syst., vol. 27, no. 1, pp. 233–242, Feb. 2012. and control of power systems.
[89] P. K. Roy and S. Bhui, ‘‘Multi-objective quasi-oppositional teaching learn-
ing based optimization for economic emission load dispatch problem,’’ Int.
J. Electr. Power Energy Syst., vol. 53, pp. 937–948, Dec. 2013.
[90] B. Mandal, P. K. Roy, and S. Mandal, ‘‘Economic load dispatch using
krill herd algorithm,’’ Int. J. Elect. Power Energy Syst., vol. 57, pp. 1–10,
May 2014.
[91] T. Jayabarathi, T. Raghunathan, B. Adarsh, and P. N. Suganthan, ‘‘Eco-
nomic dispatch using hybrid grey wolf optimizer,’’ Energy, vol. 111,
pp. 630–641, Sep. 2016. ZHIGANG LU received the B.Sc. degree in
[92] S. M. A. Bulbul, M. Pradhan, P. K. Roy, and T. Pal, ‘‘Opposition-based krill automatic control and the M.S. degree in sys-
herd algorithm applied to economic load dispatch problem,’’ Ain Shams tems engineering from Xi’an Jiaotong University,
Eng. J., vol. 9, no. 3, pp. 423–440, Sep. 2018. Xi’an, China, in 1985 and 1988, respectively, and
[93] M. Pradhan, P. K. Roy, and T. Pal, ‘‘Grey wolf optimization applied to the Ph.D. degree in electric power systems and
economic load dispatch problems,’’ Int. J. Elect. Power Energy Syst., automation from North China Electric Power Uni-
vol. 83, pp. 325–334, Dec. 2016. versity, Beijing, China, in 1998. He is currently the
[94] A. I. Selvakumar and K. Thanushkodi, ‘‘Optimization using civilized Dean of the School of Electrical Engineering, Yan-
swarm: Solution to economic dispatch with multiple minima,’’ Electr.
shan University, Qinhuangdao, China. His current
Power Syst. Res., vol. 79, no. 1, pp. 8–16, Jan. 2009.
research is engaged in electric power systems and
[95] T. Niknam, H. D. Mojarrad, and H. Z. Meymand, ‘‘Non-smooth economic
dispatch computation by fuzzy and self adaptive particle swarm optimiza- automation. His current research is mainly engaged in multi-fault repair in
tion,’’ Appl. Soft Comput., vol. 11, no. 2, pp. 2805–2817, Mar. 2011. power systems, economic operation and control of power systems, power
[96] M. Basu, ‘‘Kinetic gas molecule optimization for nonconvex economic grid state estimation, power system planning, and inspired optimization
dispatch problem,’’ Int. J. Elect. Power Energy Syst., vol. 80, pp. 325–332, algorithms and their application on various optimization problems.
Sep. 2016.