Академический Документы
Профессиональный Документы
Культура Документы
com
Abstract
An ecient hybrid genetic algorithm (HGA) approach for solving the economic dispatch problem (EDP) with valve-point eect is
presented in this paper. The proposed method combines the GA algorithm with the dierential evolution (DE) and sequential quadratic
programming (SQP) technique to improve the performance of the algorithm. GA is the main optimizer, while the DE and SQP are used
to ne tune in the solution of the GA run. To improve the performance of the SQP, the cost function of EDP is approximated by using a
smooth and dierentiable function based on the maximum entropy principle. An initial population obtained by using uniform design
exerts optimal performance of the proposed hybrid algorithm. The combined algorithm is validated for two test systems consisting of
13 and 40 thermal units whose incremental fuel cost function takes into account the valve-point loading eects. The proposed combined
method outperforms other algorithms reported in literatures (EP, EPSQP, PSO, PSOSQP) for EDP considering valve-point eects.
2007 Elsevier Ltd. All rights reserved.
Keywords: Hybrid genetic algorithm (HGA); Dierential evolution (DE); Maximum entropy principle; Uniform design; Economic dispatch problem
1. Introduction
Scarcity of energy resources, increasing power generation cost and ever-growing demand for electric energy
necessitates optimal economic dispatch in todays power
systems [1]. Economic dispatch is one of the most important problems to be solved in the operation of a power system. The primary objective of the economic dispatch
problem (EDP) of electric power generation is to schedule
the committed generating unit outputs so as to meet the
required load demand at minimum operating cost while
satisfying all unit and system equality and inequality constraints [2].
Improvements in scheduling the unit outputs can lead to
signicant cost savings. The input output characteristics of
large units are highly nonlinear because of valve-point
loadings, generating unit ramp rate limits, etc. Furthermore, they may generate multiple local minimum points
*
Corresponding author.
E-mail address: hedakuo@mail.neu.edu.cn (D. He).
0142-0615/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2007.06.023
32
the potential optimal regions within a reasonably short period of time, while DE and SQP technique subsequently
search these regions rapidly. Therefore, the proposed algorithms the possibility of nding global optimal point and
the convergence speed are improved.
In general, the hybrid method oers an exact solution
only when the function is smooth and its gradient information is known [7]. The method is proposed to approximate
EDP by using a smooth and dierentiable function based
on the maximum entropy principle in SQP operator. In this
way, the performance of the hybrid method is improved.
At the same time, to improve rationality of the distribution
of initial population, the hybrid technique integrating the
uniform design with the genetic algorithm is proposed.
Two ED problems with 13 and 40 thermal units with a
nonsmooth fuel cost function are employed to prove the
eectiveness of the proposed hybrid algorithm. The results
obtained through the combined approach are compared
with those reported in the literatures [12,13].
2. EDP formulation
The classic EDP minimizes the following incremental
fuel cost function associated to dispatchable units:
(
)
NP
X
min F min
F i P i ;
1
i1
where Fi(Pi) is the fuel cost function of the unit and ith is
the power generated by the ith unit, Pi subject to power
balance constraints:
NP
X
P i P D P loss ;
i1
where PD is the system load demand and Ploss is the transmission loss, and generating capacity constraints:
P imin 6 P i 6 P imax
for i 1; 2; . . . ; N P ;
where, ai, bi and ci, are the fuel-cost coecients of the unit,
ei and fi are the fuel cost-coecients of the unit with valvepoint eects.
The economic dispatch of generation of real power of
the generating units is to be done to the required load
demand by satisfying the above constraints. The incremental fuel cost function can be modeled in a more practical
fashion by including the valve-point eects [15]. The generating units with multi-valve steam turbines exhibit a greater
variation in the fuel cost functions. The valve-point eects
introduce ripples in the heat-rate curves, thereby the number of local optima is increased. Hence, a technique that
overcomes these complexities has to be evolved.
33
i1
Moreover,
Wx 6 Wx; l 6 Wx l ln m
8x 2 Rn
and l > 0:
8
Based on the maximum entropy principle the incremental fuel cost function can be approximated by entropic
smoothing approximation function:
WP i ; l ai P 2i bi P i ci
l ln eei sinfi P imin P i =l eei sinfi P imin P i =l :
10
The function W(Pi, l) provides a good approximation to
the function Fi(Pi) in the sense that
0 6 WP i ; l F i P i 6 l ln 2 and l > 0:
11
It is clear that this function uniformly approximates the
maximum function Fi(Pi) when a parameter l tends to
innitesimal and is greater than Fi(Pi) with an error bound
less than lln 2.
i 6 j;
hi 6 n 1 hj
12
i 6 j;
13
where u(n + 1) is Euler function of n + 1; hi and hj are different generating elements in generating vector.
Theorem 2. If (12) and (13) are satised, the optimal generating vector of uniform design with n dimensionalities
and 12 un 1 1 factors is composed of natural numbers
which are prime numbers with n + 1 in [1, n + 1/2) with any
one taken out.
Based on Theorem 1 and Theorem 2, for the uniform
designs with 12 un 1 or 12 un 1 1 factors, the
usable tables can be obtained immediately without any
34
14
i
P ti Dt; P ti P imin if r 1;
where r is a random bit, and function D(t, y) returns a value
in the range [0, y] such that the probability of the value returned is close to 0 increases with t
b
Dt; y y 1 n1t=gen ;
around the local area of the parameter resembling a hillclimbing operator. For our experiments, b was chosen
equal to 2.
15
SQP method seems to be the best nonlinear programming methods for constrained optimization. It outperforms every other nonlinear programming method in
terms of eciency, accuracy and percentage of successful
solutions over a large number of test problems. The
method resembles closely to Newtons method for constrained optimization just as is done for unconstrained
optimization. At each iteration an approximation is made
by the Hessian of the Lagrangian function using a BroydenFoldfarbShanno (BFGS) quasi-Newton updating
method. This is then used to generate a quadratic programming (QP) sub-problem whose solution is used to
form a search direction for a line search procedure. In this
combined method, SQP is used as a local optimizer to
ne-tune the region explored by GA in its run. The GA
is applied to the optimization problem, and the randomly
chosen solution obtained by GA is applied as the initial
starting point for the SQP and generate SQP ospring
by the nal solution obtained using the SQP.
5.5. Dierential evolution operator
The DE operator used in this paper can be described as
following steps and procedures.
Step 0. Set counter k = 1.
0 0
Step 1. Generate the initial individuals (points) x1 ; x2 ;
0
. . . ; xD randomly from the current GA population.
Step 2. While (stopping criterion is not met, namely
k < SSmax).
Generate D new individuals (points) as follows:
k
Step 2.1. Find the best individuals xb in the current
population;
k
k k
k
For each preset vector xi xi;1 ; xi;2 ; . . . ; xi;n ;
i 1; 2; . . . ; D.
k k
Step 2.2. (First selection). Chose two vectors xr1 ; xr2
randomly from the current population, where r1, r2 2
{1, 2, . . ., D}.
Step 2.3. (Mutation). Generate a vector v = (vi,1, vi,2,
. . ., vi,n) according to
k
k
k
k
k
fm xr1 xr2 :
16
v xi f m xb xi
Step 2.4. (Crossover). Generate a new vector u = (ui,1,
k
xi;2 ; . . . ; xi;n and the vector v = (vi,1, vi,2, . . ., vi,n) generated in Step 2.3 as follows:
(
uij
j 1;2;...;n;
17
6. Solution methodology
The proposed hybrid approach for EDP with valvepoint eects can be summarized as follows:
Step 1: Get the data for the system.
Step 2: Initialize parameter of algorithm and count t.
Step 3: Generate initial population by the usable tables
of uniform design.
Step 4: Evaluate the objective function and update count t.
Step 5: Identify the Fitbest(t) of the current run t.
Step 6: If Fitbest(t) < Fitbest(t 1) replace Fitbest(t 1)
with Fitbest(t), otherwise goto Step 7.
Step 7: Generate selection ospring using selection
operation.
Step 8: Generate crossover ospring using simplex crossover operation.
Step 9: Generate mutation ospring using nonuniform
mutation operation.
Step 10: Generate DE ospring using DE operation.
Step 11: Generate SQP ospring using SQP operation.
Step 12: Generate whole ospring with selection ospring, crossover ospring, mutation ospring, DE ospring and SQP ospring.
Step 13: While (termination criterion not met).
35
generating vector of uniform design with 28 dimensionalities and 13 factors are all natural numbers in [1,13]. The
optimal generating vector of uniform design with 82
dimensionalities and 40 factors are all natural numbers in
[1].Then the good lattice point method is employed to
obtain the usable table with optimal generating vector.
Hence the population size was chosen as 28 for test case
1 and 82 for test case 2.
The selection, crossover, DE, SQP and mutation operation proportions were chosen as 3:3:7:1:14 for test case 1
and 6:15:20:1:40 for test case 2. The simulation parameters
of the proposed hybrid approach are xed as follows,
D = 7, SSmax = 30, F = 0.48, CR = 0.88, b = 2, l = 1,
itermax = 30 for test case 1 and D = 20, SSmax = 50,
F = 0.48, CR = 0.88, b = 2, l = 1, itermax = 60 for test case
2. To validate the comparison of results obtained using the
proposed technique with the results obtained using the EP,
EPSQP, PSO and PSOSQP techniques, solution procedure of the HGA technique is terminated when the maximum number of iterations is reached.
The tness function Fit(Pi) is given as
F if
P i is feasible;
FitP i
18
F F gmax otherwise;
where F is the value of objective function, is used to keep
penalty, gmax is given as
gmax maxf0; gi P ; jhj P j; i 1; 2; . . . ; m1 ;
j 1; 2; . . . ; m2 g;
19
This case study consisted of 13 thermal units of generation with the eects of valve-point loading, the complexity
and nonlinearity to the solution procedure is increased.
The expected power demands to be met by the all thirteen
generating units are 1800 MW [12] and 2520 MW [27]. The
system data can be found from [12]. The problem has a
number of local optimum points as there are more possibilities for any method to stick on any one of the local optimum points. The problem is solved for two dierent
power demands in order to show the eectiveness of the
proposed method in producing quality solutions.
The nal fuel costs obtained using the EP, EPSQP,
PSO, PSOSQP and the proposed method for power
demand of 1800 MW were summarized in Table 1, which
shows that the minimum cost and the mean cost value
obtained by the proposed method is comparatively less
compared to all the other methods. The best results
obtained for solution vector, with HGA with the minimum
cost of $17 963.83/h is given in Table 2. Table 3 reports the
36
Table 1
Comparison of fuel costs for PD = 1800 MW for case 1
Table 4
Comparison of fuel costs for case 2
Method
Method
EP [12]
EPSQP [14]
PSO [14]
PSOSQP [14]
HGA
157.43
121.93
77.37
33.97
8.48
17 994.07
17 991.03
18 030.72
17 969.93
17 963.83
18 127.06
18 106.93
18 205.78
18 029.99
17 988.04
EP [12]
EPSQP [14]
PSO [14]
PSOSQP [14]
HGA
1167.35
997.73
933.39
733.97
84.21
122 624.35
122 323.97
123 930.45
122 094.67
121 418.27
123 382.00
122 379.63
124 154.49
122 245.25
121 784.04
Table 2
Best result obtained for PD = 1800 MW for case 1 using HGA
Power
Generation
Power
Generation
1
2
3
4
5
6
7
628.3185
222.7491
149.5996
109.8665
109.8665
109.8665
109.8665
8
9
10
11
12
13
60.0000
109.8665
40.0000
40.0000
55.0000
55.0000
Table 3
Dispatch results for a PD = 2520 MW for case 1
Generator
EPSQP [14]
PSOSQP [14]
628.23
299.22
299.17
159.12
159.95
158.85
157.26
159.93
159.86
110.78
75.00
60.00
92.62
628.3136
299.1715
299.0474
159.6399
159.6560
158.4831
159.6749
159.7265
159.6653
114.0334
75.0000
60.0000
87.5884
628.3205
299.0524
298.9681
159.4680
159.1429
159.2724
159.5371
158.8522
159.7845
110.9618
75.0000
60.0000
91.6401
1
2
3
4
5
6
7
8
9
10
11
12
13
Total cost
($/h)
24 275.71
24 266.44
24 261.05
HGA
628.3184
299.1992
299.1988
159.7330
159.7329
159.7324
159.7330
159.7330
159.7331
77.3994
77.3996
87.6879
92.3992
24 169.92
Table 5
Best result obtained for case 2 using HGA
Power
Generation
Power
Generation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
111.3793
110.9278
97.4104
179.7331
89.2188
140.0000
259.6198
284.6570
284.6588
130.0000
168.8214
168.8496
214.7524
394.2848
304.5361
394.2987
489.2877
489.2869
511.2752
511.2857
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
523.2961
523.3202
523.2916
523.3014
523.2675
523.2787
10.0000
10.0000
10.0000
88.6376
190.0000
190.0000
190.0000
164.9795
165.9970
165.0464
110.0000
110.0000
110.0000
511.3005
PSO, PSOSQP and the proposed method were summarized in Table 4. The best result obtained for solution vector, with HGA with the minimum cost of $121 418.27/h is
given in Table 5. It is clear from Table 4, HGA has the
best probability of attaining better solutions, the mean
cost value, and the minimum cost amongst the ve methods in this test case. It is noticeable that the mean cost
value obtained using HGA is less than the minimum cost
obtained using other methods. Hence, for power system
EDP of greater size with more nonlinearities, the proposed method is proved to be the best algorithm amongst
the ve methods.
8. Discussion and conclusion
In the past, to solve economic dispatch problem
eectively, most algorithms require convex cost function
and the incremental cost curves to be of monotonically
smooth increasing nature and strict continuity of the search
space. For the generating units, the incremental fuel cost
curves are highly nonlinear and noncontinuous and
nonmonotonically.
GA is an ecient tool for solving complex optimization
problems. It is applied to solve various problems in various
diverse elds. It is also eectively used in solving complex
problems in the power system eld such as EDP. GA is faster in nding the high performance region but displays difculties in performing local search for complex functions.
It leads to a poor ne tuning of the nal solution. The
DE is a population-based direct search algorithm. In general, the DE provided fast convergence. However, as it is
a greedy scheme, with an explicit bias towards the best ospring, DE sometimes runs into the problem of losing the
diversity of the population, which, in turn, leads to a premature convergence. It shows that the DEs main advantages are its simple structure, speed and local searching
property. The SQP as a best nonlinear programming
method can explore the search space quickly with a gradient direction but this local gradient search technique is sensitive to the initial starting points and always gets stuck at a
local optimum solution. The GA, DE and SQP have
advantages that complement one another.
To overcome these drawbacks and exert these advantages, this paper proposed new hybrid approach combining GA, DE and SQP to solve the EDP of electric energy
with the valve-point eect. Basically, the proposed algorithm consists of two parts. The rst part employs the
property of GA, which can provide a near global search
region. The second part employs DE to explore the
search region and SQP to ne tune solutions of the
GA run in order to obtain a local optimum solution.
The cost function of EDP is approximated by using a
smooth and dierentiable function based on the maximum entropy principle. This is advantaged to improve
the performance of the SQP. An initial population
obtained by using uniform design reects the information
of solution space scientically and exerts optimal performance of the proposed hybrid algorithm.
The eectiveness of the HGA is illustrated for two test
cases of EDP with valve-point eects and compared with
the results obtained using the EP, EPSQP, PSO and
PSOSQP methods. The numerical results reveal the superiority and feasibility of the proposed hybrid approach for
solving the EDP of units with value-point eects.
Acknowledgement
This work is supported by the National Nature Science
Foundation of China under Grant 60374063.
Appendix A
List of symbols
F
Fi(Pi)
Pi
NP
PD
Ploss
Pimin/Pimax
u(n + 1)
P ti
P mut
i
r
n
t
gen
b
k
SSmax
D
k
xb
k
xi
fm
CR
v
u
rand(0,1)
rnbr(i)
37
Euler function of n + 1
i th units output that is chosen to be
mutated (MW)
ith units new output after mutation (MW)
a random bit
a random oating-point number in the
interval [0, 1]
the current generations
the maximum number of generations of GA
a parameter that determines the degree of
dependence on the number of generations
current iteration number of DE operator
maximum iteration number of DE operator
population scale of DE operator
the best individuals in the k iteration
the i individuals in the k iteration
mutation parameter of DE operator
crossover parameter of DE operator
a vector after mutation in DE operator
a vector after crossover in DE operator
a uniform random number in range (0,1)
a randomly chosen index in the
set {1, 2, . . ., n}
References
[1] Jeyakumar DN, Jayabarathi T, Raghunathan T. Particle swarm
optimization for various types of economic dispatch problems.
Electric Power Energy Syst 2006;28 (1):3642.
[2] Abido MA. A novel multiobjective evolutionary algorithm for
environmental/economical power dispatch. Electric Power Syst Res
2003;65 (1):7181.
[3] Zixiong Liang, Glover JD. A zoom feature for a dynamic programming solution to economic dispatch including transmission losses.
IEEE Trans Power Syst 1992;7 (2):5449.
[4] Park JB, Park YM, Won JR, Lee KY. An improved genetic
algorithm for generation expansion planning. IEEE Trans Power
Syst 1992;15 (3):91622.
[5] Wona Jong-Ryul, Park Young-Moon. Economic dispatch solutions
with piecewise quadratic cost functions using improved genetic
algorithm. Electric Power Energy Syst 2003;25 (5):35561.
[6] Orero SO, Irving MR. A genetic algorithm modeling framework and
solution technique for short term optimal hydrothermal scheduling.
IEEE Trans Power Syst 1998;13 (2):50118.
[7] Baskar S, Subbaraj P, Rao MVC. Hybrid real coded genetic
algorithm solution to economic dispatch problem. Comput Electric
Eng 2003;29 (3):40719.
[8] Storn R, Price K. Dierential evolutionA simple and ecient
heuristic for global optimization over continuous spaces. J Global
Optim 1997;vol. 11 (4):34159.
[9] Coelho Leandro dos Santos, Mariani Viviana Cocco. Combining of
chaotic dierential evolution and quadratic programming for economic dispatch optimization with valve-point eect. IEEE Trans
Power Syst 2006;21 (2):98996.
[10] ChungFu C, JiJen W, JiPyng C, et al. Robust searching hybrid
dierential evolution method for optimal reactive power planning in
large-scale distribution systems. Electric Power Syst Res 2007;77 (56):4307.
[11] Aruldoss Albert Victoire T, Ebenezer Jeyakumar A. Hybrid PSO
SQP for economic dispatch with valve-point eect. Electric Power
Syst Res 2004;71 (1):519.
38
[12] Sinha N, Chakrabarti R, Chattopadhyay PK. Evolutionary programming techniques for economic load dispatch. IEEE Trans Evol
Comput 2003;7 (1):8394.
[13] Bhagwan Das D, Patvardhan C. Solution of economic load dispatch
using real coded hybrid stochastic search. Int J Electr Power Energy
Syst 1999;21 (3):16570.
[14] Walters DC, Sheble GB. Genetic algorithm solution of economic
dispatch with valve-point loadings. IEEE Trans Power Syst 1993;8
(3):132531.
[15] Chowdhury BH, Rahman S. A review of recent advances in economic
dispatch. IEEE Trans Power Syst 1990;5 (4):124859.
[16] Li Xingsi. An aggregate function method for nonlinear programming.
Sci China (Ser A) 1991;34 (12):146773.
[17] Xingsi L, Shucherng F. On the entropic regularization method for
solving minmax problems with applications. Math Methods Operat
Res 1997;46 (1):11930.
[18] Houduo Q, Lizhi L, Zhenghua L. Regularized smoothing approximations to vertical nonlinear complementarity problems. J Math
Anal Applicat 1999;230 (1):26176.
[19] Tseng P, Bertsekas DP. On the convergence of the exponential
multiplier method for convex programming. Math Program 1993;60
(1):119.
[20] Shucherng F, Jiye H, Zhenghai H, et al. On the nite termination of
an entropy function based non-interior continuation method for
[21]
[22]
[23]
[24]
[25]
[26]
[27]