Vous êtes sur la page 1sur 6

2010 3rd International Conforence on Advanced Computer Theory and Engineering(ICACTE)

Application of An Improved Particle Swarm Optimization to


Economic Load Dispatch in Power Plant
Ting-Fang1
YU PENG Chun-Hua2
Senior Engineer Associate Professor
School of Mechanical and electronic Engineering , School of Electrical & Electronics Engineering, East China
Nanchang University, Jiaotong University,
Nanchang , Jiangxi Province, China; 330031, Nanchang, Jiangxi Province, China;330013
wtu_tingfy@163.com chinapch@163.com

Abstract: In the application of Particle Swarm Optimization have no special demands for mathematical model and coal
(PSO) for the problem of economic load dispatch (ELD) in consumption curve, so they are more suitable for the ELD.
power plant There are some problems such as premature These methods have the advantage of searching the solution
convergence of PSO, how to deal with constraints of the
space more thoroughly. The main difficulty is their
problem are need to be addressed. An improved PSO with the
sensitivity to the choice of parameters. Among intelligent
constraints partially solved combined with penalty function
were introduced in this paper, the improved PSO refined in methods, PSO is simple and promising. which has the
constraints management, swarm initialization method of PSO advantage of few parameters to adjust, less computation
with respect to the features of ELD in power plant. Based on time, higher accuracy. But there are still some problems
above improvements, the improved PSO solved the problem of such as premature convergence, how to deal with
premature convergence in PSO. In the application to a three constraints of ELD are need to be addressed as PSO is
and a five units power plant respectively, the simulation results employed to approach the ELD problem. An improved PSO
shows that the optimization success rate can reach 100% with with the constraints partially solved combined with penalty
the improved PSO, which has the advantage of few parameters
function were introduced in this paper, the improved PSO
to adjust, easy to implement, with high computational
refined in constraints management, initialization of PSO
efficiency and high accuracy.
swarm with respect to the features of economic load
Keywords:Particle Swarm Optimization(PSO);Economic dispatch in power plant. Based on above improvements, the
Load Dispatch(ELD); Premature convergence improved PSO solved the problem of premature
convergence in PSO.
I INTRODUCTION
II M ATHERM ATICAL MODEL
Economic load dispatch (ELD) in power plants is a
problem of high dimension, non-linear, multiple constraints A. Objective function
and real time specialty[I]. There are various numerical The objective of the economic load dispatch is to
optimization techniques have been employed to approach minimize the generating cost based on the premise that
the ELD problem. ELD is solved traditionally using constraints are satisfied. Coal consumption(standard coal) is
mathematical programming based on optimization selected as the optimization objective in order to emphasize
techniques such as equal increment method, mathematical the main aspects, simplify the mathematical model and
linear programming(LP), homogenous LP, quadratic make the problem comparable . Then the mathematical
programming(QP), non-linear programming (NLP) and description of ELD's objective function is:
dynamic programming(DP) and so on [2-5].Equal increment
method belongs to traditional methods with the advantages i=N
of simple mathematical model and high search speed, but it minF = min LUiF';(Pi) (1)
demands the coal consumption is a convex function to i=1
ensure its convergence, which is not factual. LP and
homogenous LP have high speed , but could bring errors where N is the quantity of units, F is the total coal
while linearizing the model; QP and NLP can calculate consumption" FlPJ is the coal consumption of No.i unit
non-linear model, but still demand the coal consumption under the load of Pi' Pi is the load of No. i unit (MW).
curve is continuous and differentiable , which limits its B .Fuel cost function
application; DP has precise results and less demands for the
Main generation cost is fuel cost. Usually, fuel cost of
coal consumption curve but it brings 'dimension disaster'
a power plant unit is expressed as a second order
when there are more units.
approximate function of each unit output as below.
In recently years, many scholars apply intelligent
methods to ELD, such as neural network(NN)[6,7], genetic
algorithm(GA)[8,9],particle swarm optimization (PSO) (2)
[10-15], et al . Compared with the traditional methods, they

978-1-4244-6542-2/$26.00© 2010 IEEE V2-619


2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

where a;,b;, and c; are positive fuel cost coefficients. position of each particle are calculated using current
velocity and the distance from pbest to gbest
C Constraints as(4)and(S).In(4),the first term indicates the current velocity
The constraints that must be satisfied with the ELD of the particle(inertia), second term presents the cognitive
problem are as follows. part of PSO where the particle changes its velocity based on
its own thinking and memory, and the third term is the
i=N social part of PSO where the particle changes its velocity
L U;P; -PtotaJ = 0 based on the social-psychological adaptation of knowledge.
;=1
u p;min � p; � u p;masx All the terms are multiplied by appropriate parameters.
; ;
i=N B Constraints management
L Uip;ffiaX � P;otaJ (3) There are two kinds of constraints in the problem of
;=1 ELD: inequality constraint and equality constraint. For an
;=N
L UiP min � P;otaJ
,
inequality constraint, Error is the amount of violation from
;=1 the nearest upper or lower limit. Which can be changed into
non-constraints conditions by designing special operator .
where P i is the load of No. i unit,
max and min For an equality constraint, Error is the amount of violation
p; p;
from the system power balance. The paper using penalty
are the maximal and the minimal load of No.i unit, Ptt o a! is function to handle the constraints as follows:
the total load of the whole power plant.. U; is the state of No.
i unit , when No. i unit is runing U;=I, otherwise U;=O. i=N
2
III IMPLEMENTATION OF STANDARD PSO FchJ =K(P;otaJ - L UiP; ) (6)
;=1
COMBINED WITH PURE PENALTY FUNCTION
A Standard PSO Where K is penalty coefficient
As ELD is a minimization problem, a positive penalty
Standard particle swarm optimization is based on a
is introduced into the objective of ELD as follows:
simplified social model that is closely tied to swarming i=N
theory and was first introduced by Kennedy and Eberhart in minF = min LUiFi(P;) + FchJ =
1995.This PSO concept consists of, at each generation, ;=1 (7)
changing the velocity and location of each particle toward i=N i=N
its pbest location depending on memory and gbest location min L U i F; (P i ) + K(P;otaJ - L U;p; )2
depending on knowledge according to(4)and(S),
;=1 ;=1
respectively: N APPLICATON EXAMPLE OF STANDARD PSO
Vi;+! = x Vi; +C1 xrandOx(p;;est -x� )+
w
COMBINED WITH PURE PENALTY FUNCTION AND
(4) ITS PROBLEM ANALYSIS
C2 x randO x (g /est -x:J (i =1,2,"', N)
A Description of study object
(S) We selected a power plant system with five units as
optimization example, fuel consumption of the power plant
Here w. Cl .C2 are inertia weight and 2 trust parameters, is expressed as a second order approximate function for
respectively. The index of d represent the dth particle. D is each unit as below.
swarm size, k represents the kth generations. 11;;+1 the
1 unit no.1: .t; = 0.0007 �2 + 0.3� + 4 (t/h)
(k+1)th generations particle velocity. Xi�+ the location of
unit no.2: f2 = 0.0004P22 + 0.32P2 + 3 (t/h)
d particle at (k+1)th generations.
PSO is similar to the other evolutionary algorithms in unit no.3: f3 = 0.0004SP / + 0.3P3 + 3.S (t/h) (8)
that the system is initialized with a swarm of random
unit no.4: f4 = 0.000179P42 + 0.26P4 + 4.SS (t/h)
solutions. Each potential solution ,call particle ,flies in the
N-dimensional problem space with a velocity , which is unit no.S: f s = 0.000182P / + 0.2SPs + S.O (t/h)
dynamically adjusted according to the flying experiences of
its own and its colleagues .The location of the ith particle is upper and lower limit for each unit as below:
represented as X;=(Pil.P;2 PiN).The best previous position
•...• 100MWsP/:;200MW , 120MW:;P2:;2S0MW
of the ith particle is recorded as pbesti.The index of the best IS0MW:SP3:S300MW , 60MW:SP4:S12SMW
pbest among all the particles is represented by the symbol g. 60MWsP5:S12SMW
The location pbestg is also called gbest. V elocity of the ith
particle is represented as �=(Vil. V;2, ViN). Next velocity and
...•
In order to analysis the influence of the number of units

V2-620
2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

Tab.2 Optimal combination under the typical total loads in a five units power plant

Total coal
Ptot.,(MW) P,(MW) P2(MW) P3(MW) P4(MW) Ps(MW)
consumption(tIh)

510 0 120 150 115 125 181696

600 100 120 150.0 105 125 219.70

686.31 107.14 162.5 166.67 125 125 254.68

778.57 128.57 200 200 125 125 297.582

90158 157.14 250 244.44 125 125 359.087

992.86 192.86 250 300 125 125 408.835

calls the five units system. its optimal combination under the
Tab.l Optimal combination under the typical total loads in a three units
typical total loads is listed on Tab.2. The optimized object
power plant
has many local extreme point near the global optimum point,
Total coal under the whole system load in the 600MW, Optimization
Ptoto,(MW) PtCMW) P2(MW) P,(MW)
consumption(tlh) process can easily fall into local extreme point 219.97 tIh,
260 101.7 0 158.3 104.0165
the 5IOMW total load easily into the local I82.957tlh
extreme point.
350 0 173.5 176.5 1410294

436.31 107.1 162.5 166.71 175.741286


B Success criterion ofPSG searching
To compare search ability between the standard PSO
Tab.3 Optimization results of the standard PSO combined with pure penalty and the improved PSO, Analysis criteria were introduced
function in a three units power plant as follow: compared with the global optimal solution, if the
solution's state for all units match the global optimum point
Average total coal consumption Success times in 20
Ptota,(MW) and the deviation of load for each unit is under 5MW, the
in 20 times simulation (tIh) times simulation
optimization result is successful, otherwise failure.
260 104.178125 18 In simulation calculation, PSO parameters take C1 =
350 14120364 16
C2 = 1.8, inertia coefficient w in (1, 0.5) linear decrease,
penalty coefficient K = 10, swarm size D= 50, the most
436.31 176.883923 18 number of iterations maxgen = 300 Generation.
C Simulation results of standard PSG and its problem
Tab.4 Optimization results of the standard PSO combined with pure penalty
function in a five units power plant Analysis
With the standard PSO combined with pure penalty
Average total coal consumption Success times in 20
Ptoto,(MW)
in 20 times simulation (tIh) times simulation
function, optimized simulation were performed to the above
three units power plant system and five units power plant
510 182.8368 1 system respectively, 20 times optimized calculation results
under each typical total load of power plant were shown on
600 221254 0
Tab.3 and Tab.4.
686.31 26169314 8 Shown as Tab.3, in the application of standard PSO
778.57 303.291968 11
combined with pure penalty function to the three units
power plant system, the success rate can reach 80-90%, but
90158 636.264551' 13 there are still some search process fall into local extreme
992.86 8763.182812' 11
point and become premature.
From the Tab.4, when the number of power plant units
in power plant on the PSO optimization results, define the reach five , the simulation results is poor, by the above
power plant system owns the units from no.1 to no.3 as success criteria of optimization, at the typical total load of
fewer units optimization object, calls the three units system. power plant of 600MW and 5IOMW, which has many local
which represent the fewer units system ,its optimal extreme point near the global optimum point, the 20 times
combination under the typical total loads is listed on Tab.I. search process were all failure, at other typical total load of
The power plant system owns the units from no.I to power plant, the success rate can reach only around 50%.
no.5 above represent the more units optimization object, From the simulation results of Tab.3 and Tab.4, the

V2-621
2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

min min•
standard PSO combined with pure penalty function is very
1 r pr
ijP< then Pr =pr
'
difficult in searching the global optimal point as the power max max
plant has more units. The reason as follows: a) Disadvantage else ifPr > Pr then Pr = Pr
of the standardPSO[13, IS] , from the formula of (4) and (5), max m
Where Pr , Pr in is the rth unit upper and lower
velocity of standard PSO consists of three parts: the
load limit (MW) 0
momentum part, the cognitive part and the social part. As
b) After the location of a particle is updated, the
evolution goes on, the swarm might undergo an undesired
operation (a) above is duplicated
process of diversity loss. Some particles become inactively
After the method of constraints partially solved were
while they lose both global and local searching capabilities
introduced, a large number of particles of ELD's infeasible
in the next generations. For a particle, the loss of global
solutions were changed to feasible solutions, there are still
searching capability means that it will be only flying within
some of particles are infeasible solutions, but their location
a quite small space, which will be occurred when its location
become nearer to its constraints border. The simulation
and pbest are close to gbest (if the gbest has no significant
results shows the operation of constraints partially solved
change)and its velocity is close to zero(for all
may reduce the diversity loss of PSO and exclusion of
dimensions)according to Eq.(4);the loss of local searching
infeasible solutions caused by the penalty function, thus
capability means that the possible flying cannot lead
improved the searching ability of PSO.
perceptible effect on its fitness, the evolution process will be
stagnated. If gbest is located in a local optimum, then the B Improved aspect on the initialization ofPSG swarm
swarm convergence becomes premature, as all the particles
During the initialization of PSO swarm, the paper
become inactive. b)Constraints management is unreasonable, improved on two aspect: first, during the initialization of the
with the pure penalty function, the penalty coefficient K is units state U; for each particle, instead of generate a length
very difficult to chose. a lager part of particles is infeasible of N-dimensional integer series to particle unit state purely,
solution and constraints can not be satisfied after initialized the paper introduced a reasonable criteria: On the basis of
at random, when penalty coefficient K is unreasonable , it the N-dimensional integer series of U;(i=l to N), if
is difficult to reflect the advantages and disadvantages of a
particle location under the influence of penalty function N

through the fitness of a particle. Thus the PSO present an L Vi X Pimin < Ptotal < L Vi X p;max (10)
i=1
adverse situation of random search, and can not convergence
to global optimum. c) Problems caused by its own
characteristics of ELD, ELD is a very complex problem Then the generated Vi is acceptable, otherwise generate
which has many has many local extreme point near the the N-dimensional integer series to particle unit state again,
global optimum point, especially when there are more units until formula (10) is satisfied.
in power plant. Second, according to the coal consumption of thermal
power characteristics that the unit has a lower coal
V IMPROV ED PSO consumption rate in the vicinity of the higher load, while has
a higher coal consumption rate in the vicinity of the lower
As the exist problems above with standard PSO
load, the paper improved the initialization of thePSO swarm
employed to approach the ELD problem, an improved PSO
as follow: select a certain percentage of particles, initialize
is introduced in this paper, which improved in the
constraints management, initialization of PSO swarm with its unit load with the range of ( 80%� 100%) p; "laX at
respect to the features of ELD problem as follows: random.

A Constraints partially solved VI SIMULATON RESULTS OF IMPROV ED PSO


On the basis the management of using penalty function
With the improved PSO proposed above, Simulation
to handle the constraints with formula(6) , a method with the
optimization calculation were performed at the power plant
constraints partially solved combined with penalty function
system with three units and five units respectively,
were introduced in this paper, its realization process is as
simulation results are shown in Tab.S and Tab.6 , Tab.S is
follows:
a) During the initialization ofPSO swarm, distribute the
the results of the three units power plant, Tab.6 is the results
of the five units power plant. From the simulation results ,
load with the range of upper and lower limit to each units at
random first, then select r unit within 1 to N at random , if
the improved PSO shows a strong searching ability in the
application of ELD, at the 20 times optimal calculations
the rth unit's state is running( Vr=I), assign the rth unit the
load as follow:
under each typical load of power plant system, the
optimization success rate can reach 100% with the analysis
i= N ,i#: r criteria , and accuracy is greatly improved.
Pr= P total L U iP i (9)
i= 1
Fig.l and Fig.2 shows the movement of the global best
particle visually using the proposed improved PSO under the

V2-622
2010 3rd International Conforence on Advanced Computer Theory and Engineering(ICACTE)

189
total load of 600MW and 510MW of the five power plant.
""
Horizontal axis represents iterations, vertical axis is the 5'
(()
188
fitness of the global best particle. Initially, they are �
0
...,
frequently changed; however, the change is relatively 5'
(() 187 -
smaller near to the final generations, which indicates a fine fJQ
g \
tuning of the searching space. Fig.l and Fig.2 shows the �
186 -
"0
searching efficiency of improved PSO is very high, the '"
;::.
(i.
global best particle reaches the global optimum solution at ii" 185 -
the 18th generation under the total load of 600 MW , and ?
reaches the global optimum solution at the 12th generation 2:
184 -
the total load of 51OMW .
'---t
183 -
Tab.5 Optimization results of the improved PSO in a three units power plant

182 - The global optimum point(181.696tJh)


Average total coal consumption Success times in 20
Pto",(MW)
in 20 times simulation (tJh) times simulation
I terations
181 -1---,--,--
I ,--,---r---.
I ---,-'
.,.- I----r-,--,
I '''I---.-I
.,.- ,...,.:..:c:r=
1 ;.::.=-
1.:c.; --"-----1
260 104 016518 20 10 20 30 40 50 60 70 80 90 100

Fig.2 Convergence of the global best particle under the load of


350 141.029404 20
510MW in five units power plant

436.31 175.741287 20
VII CONCLUSION

Tab.6 Optimization results of the improved PSO a five units power plant An improved PSO with the constraints partially
solved combined with penalty function were introduced in
Average total coal consumption Success times in 20 this paper, the improved PSO refined in constraints
Pto",(MW)
in 20 times simulation (tJh) times simulation
management , initialization of PSO swarm with respect to
510 181.696033 20 the features of ELD in power plant. Based on above
600 219.702173 20 improvements, the improved PSO solved the problem of
premature convergence in PSO. In the application to a
686.31 254.681934 20
three and a five units power plant respectively , the
778.57 297.582861 20
simulation results shows that the optimization success rate
901.58 359.0875 20 can reach 100% with the improved PSO, which has the
992.86 408.835 20 advantage of few parameters to adjust, easy to implement,
higher computational efficiency and a higher accuracy.
226
"" ACKNOWLEDGMENT
5'
(()
� This work was supported by Jiangxi Provincial Office of
0 225 -
...,
5' Education Research Fund (NO:GJJ10293) and the Natural
(()
00
0- 224 -
Science Foundation of Jiangxi Province of China
� (NO:2009GZSOOI6) .
"0
'"
;::.
(i.
ii"
223
I REFERENCES
?
[1] Kang Song, Zhang Xiangdong, Sun Xingping. Optimization of
2: 222 - load dispatching among multiple generating units of a thermal power
plant[J]. Electric Power, 1997, 30(3): 25-28(in Chinese).
[2] Padhy N P. Unit Commitment-a Bibliographical Survey[J].IEEE
221 -
Trans on Power Systems, 2004, 19(2): 1196-1205.
[3] Snyder W L, Powell H D Jr, Rayburn J C. Dynamic Programming
The global optimum point(219. 70tJh) Approach to Unit Commitment[J]. IEEE Trans on Power Systems,
220 -
1987, 2(2): 339-350.
Iterations
[4] Han Xueshan, Liu Zhuo. Optimal unit commitment considering
219 -t---'-'I -'--"
I -'--
1 "'-'" 1 "--
1 -'--' 1 - 1 '--',,----1
-' "1 -'-- 1 unit's ramp-rate limits[J] . Power System Technology , 1994 ,
10 20 30 40 50 60 70 80 90 100 18(6): 11-16 (in Chinese).
Fig.1 Convergence of the global best particle under the load of [5] J.A. Momoh, M.E.El-Hawary, R.Adapa, A review of selected optimal
600MW in five units power plant power flow literature to 1993, Part I:Nonlinear and quadratic
programming approaches ,IEEE Trans. Power
Syst.14(l X1999)96-104.

V2-623
2010 3rd International Conference on Advanced Computer Theory and Engineering(1CACTE)

[6] WAN Wenjun ZHOU Keyi XU Jianqun, et al. Dynamic system on [12] HOU Yunhe , LU Lijuan , XIONG Xinyin , et al . Enhanced
economic dispatch among thermal power units[J]. Proceedings of the particle swarm optimal algorithm and its application on one economic
CSEE, 2005, 25(2): 125-129. dispatch of power systems [J]. Proceedings of the CSEE, 2004,
[7] Yalcinoz T, Cory B J, Short M 1. Hopfield neural network a: roaches 24(7):95-100.
to economic dispatch Problems [1]. Electrical Power & Energy [13] Ahmed Yousuf Saber , Shantanu Chakraborty , S.M.Abdur
System, 2001, 23(6):435-442 Razzak, et al. Optimization of economic load dispatch of higher
[8] CHEN Yanqiao, NI Min, LIU Jizhen, et al. Application of real order general cost polynomials and its sensitivity using modified
coded genetic algorithm to economic load dispatch in power particle swarm optimization [J]. Electric Power Systems Research,
plants[J]. Proceedings of the CSEE, 2007, 27(20): 107-112. 2009, 79(1): 98-106.

[9] 1. Tippayachai, W.Ongsakul, LNgamroo, Parallel micro genetic [14] Ting T 0, Rao M V C, Loo C K. A novel approach for unit
algorithm for constrained economic dispatch, IEEE Trans. Power commitment problem via an effective hybrid particle swarm
Syst.I 7(August(3)X2003):790-797. optimization[J]. IEEE Trans on Power Systems, 2006, 21(1):
411-418.
[10] ZHANG Zhen-yu, GE Shao-yun, LIU Zi-fa . Particle swarm
optimization algorithm and its application in unit commitment [15] Zeng Jianchao, Jie Jing, Cui ZhiHua . Particle Swarm
[J]. Eiectric Power Automation Equipment, 2006, 26(5): 28-31. Optimization[M]. Beijing China:: Science Press, 2004.

[11] CHEN Gong-gui . Power station short-term generation optimal


dispatch based on local random search PSO algorithm [J]. Electric
Power Automation Equipment, 2008, 28(5): 52-55.

V2-624

Vous aimerez peut-être aussi