Professional Documents
Culture Documents
Optimal Reactive Power Dispatch Using Ant Colony Optimization Algorithm
Optimal Reactive Power Dispatch Using Ant Colony Optimization Algorithm
DOI 10.1007/s00202-011-0196-4
ORIGINAL PAPER
Received: 5 March 2010 / Accepted: 3 January 2011 / Published online: 28 January 2011
© Springer-Verlag 2011
Abstract This paper proposed a procedure to solve the of ORPD is to minimize the transmission line losses by opti-
optimal reactive power dispatch (ORPD) problem using ant mizing the settings of control variables such as generation
colony optimization (ACO) algorithm. The objective of the voltages, switchable reactive power, and on-load tap chang-
ORPD problem is to minimize the transmission power losses ers (OLTC) under control and dependent variable constraints.
under control and dependent variable constraints. Proposed In recent years, many optimization techniques were used for
sensitivity parameters of reactive power at generation and solving the reactive power dispatch problem.
switchable sources are derived based on a modified model of Li et al. [1] proposed an integer-coded multi-objective
fast decoupled power flow. The proposed ACO-based algo- genetic algorithm (IGA) to reactive power compensation
rithm is applied to the IEEE standard 14-bus, 30-bus sys- planning (RCP) problem considering both intact and con-
tems, and a real power system at West Delta Network as a tingent operating states. The IGA solved both the sitting
part of the Unified Egyptian Network. The obtained simula- and installing of reactive power compensation devices and
tion results are compared with those of conventional linear the operational characteristics as multi-objective optimiza-
programming, genetic algorithm, and particle swarm optimi- tion problem.
zation technique. Simulation results show the capability of Subbaraj and Rajnarayanan [2] developed a self-adap-
the proposed ACO-based algorithm for solving the ORPD tive real coded genetic algorithm (SARGA) as one of the
problem, especially with increasing the system size. techniques to solve the ORPD problem. The problem for-
mulation involves continuous (generator voltages), discrete
Keywords Ant colony optimization · Genetic algorithm · (transformer tap ratios), and binary (Var sources) decision
Particle swarm optimization · Control variables · variables.
Dependent variables Zhang and Ren [3] presented a model of ORPD for power
systems based on the principle of income maximization,
1 Introduction which aimed at minimizing active power loss of the whole
networks. On the other hand, the operations would augment
optimal reactive power dispatch (ORPD) is one of the opti- the fault probability of power systems. Therefore, the costs
mization problems in power systems that optimized the set- of adjusting the control devices (CACD) were investigated,
tings of control variables using conventional, intelligence, and a mathematical model of ORPD was implemented to
and modern optimization techniques. The objective function minimize the energy loss at the current time interval and the
CACD.
Yan et al. [4] proposed an improved evolutionary
A. A. Abou El-Ela · A. M. Kinawy · M. T. Mouwafi
Electrical Engineering Department, Faculty of Engineering, programming (IEP) and its hybrid version combined with
Minoufiya University, Shebin El-Kom, Egypt the nonlinear interior point (IP) technique to solve the ORPD
problem. In an IEP method, the common practices in regulat-
R. A. El-Sehiemy (B)
ing reactive power followed in adjusting the mutation direc-
Electrical Engineering Department, Faculty of Engineering,
Kafrelsheikh University, Kafr El-Sheikh, Egypt tion of control variables in order to increase the possibility
e-mail: elsehiemy@eng.kfs.edu.eg of keeping state variables within bounds.
123
104 Electr Eng (2011) 93:103–116
Zhao et al. [5] suggested the use of an integrated The transmission losses can be computed as:
multi-agent system (MAS) and PSO algorithm for solving
the reactive power dispatch problem. In [6], the ORPD was
NB
123
Electr Eng (2011) 93:103–116 105
⎛ ⎞
NB
N
2.1 Control variable constraints
PL = ⎝V 2 G ii +|Vi |
i G i j cos δi j + Bi j sin δi j V j ⎠
i=1 j=1, j ∈i
/
2.1.1 Generation voltage constraints
(9)
The change in generator voltage must be within their permis-
sible limits as:
To compute the change, transmission line losses with respect
to changes in generator voltage can be derived based on
the load flow equations and in terms of the change in the VGmin ≤ VG ≤ VGmax
generator voltages and switchable reactive power, which are VGmin = VGmin − VGinit (14)
defined as: VGmax = VGmax − VGinit
NG
where VGinit is the initial value of the generator voltage at
MinPL = (∂ PL /∂ VG )VG each generator bus, VGmin and VGmax are the minimum and
G=1
maximum of generation voltages, respectively, VGmain and
NSW
VGmax are the minimum and maximum changes in genera-
+ (∂ PL /∂ Q SW ) Q SW (10)
tion voltages, respectively.
SW=1
where PL is the objective function of the change in trans- 2.1.2 Switchable reactive power constraints
mission line losses, VG and Q SW are the changes in the
control variables of generation voltages and switchable reac- The change in switchable reactive power must be within their
tive power, respectively, NG is the number of generation permissible limits as:
buses, NSW is the number of buses which the switchable
reactive sources are located, NB is the number of system
Q min
SW ≤ Q SW ≤ Q SW
max
buses.
Now, the values of ∂ Pi /∂ Vi and ∂ P j /∂ Vi can be calcu- Q min
SW = Q SW − Q SW
min init
(15)
lated by differentiating Eq. (7) as: Q max
SW = Q max
SW − Q init
SW
N
where Q init
SW is the initial value of the switchable reactive
∂ Pi /∂ Vi |i∈NG = 2Vi G ii + G i j cos δi j power, Q min max
SW and Q SW are the minimum and maximum of
j=1, j ∈i
/
SW and Q SW
switchable reactive power, respectively, Q main max
+Bi j sin δi j V j (11) are the minimum and maximum changes in switchable reac-
∂ P j /∂ Vi j ∈NG
/
= V j G ji cos δi j − B ji sin δi j tive power, respectively.
(12)
2.2 Dependent variable constraints
where Yi j is the admittance element of admittance matrix
between buses i and j; δi and δ j are the angles of the volt- 2.2.1 Load voltage constraints
ages at buses i and j, respectively.
Moreover, the change in transmission line losses due to The voltage at each load bus must be within their permissible
changes in switchable reactive power can be written as: limits as:
123
106 Electr Eng (2011) 93:103–116
2.2.2 Generation reactive power constraints 3.1 The load bus voltages related to control variables
The reactive power of each generator must be within their 3.1.1 No changes in the reactive power at load-bus
permissible limits as: (Q L = 0)
G ≤ Q G ≤ Q G
Q min max The sensitivity parameters relating the changes in load-bus
voltages due to the changes in generation-bus voltages are
Q min
G = G − QG
Q min init
(17)
given as:
Q max
G = G − QG
Q max init
The sensitivity parameters between control and dependent Now, the proposed sensitivity between the control and the
variables can be derived as follows. dependent variables can be formulated based on the previous
123
Electr Eng (2011) 93:103–116 107
equations in a compact matrix form as: all ants in the colony will take the shortest route around the
obstacle as shown in Fig. 1c.
VL SGL SSW
L VG
= (26)
Q G SGG SSW
G Q SW
4.2 Mathematical model of ACO algorithm
Fig. 1 Illustration of real ant behavior [22] where dbest is the shortest tour distance.
123
108 Electr Eng (2011) 93:103–116
123
Electr Eng (2011) 93:103–116 109
1.09
1.08
VLmax.
1.03
1.02
strongest due to attraction of ants to this path. Moreover,
1.01
the pheromone on the other paths is evaporated in time.
Vsp.
Step 7: Program termination
0.99
The program will be terminated when the maximum iter- 3 4 5 6 7 8 9 10 11 12 13 14
ation is reached or the best solution is obtained without Load buses
the ants stagnations.
Fig. 4 Initial values of load voltages for 14-bus system
123
110 Electr Eng (2011) 93:103–116
Table 2 Minimum, maximum limits, and initial values for control and Table 3 Minimum, maximum limits, and initial values for control and
dependent variables for 30-bus system dependent variables for WD real power system
Control/dependent Min. Max. Initial Control/dependent Min. Max. Initial
variables variables
VLmax. VLmax.
1.04
1.025
Initial value of load voltages
1.03
Initial value of load voltages
Vsp.
1.02
0.975
1.01
VLmin.
Vsp.
0.99 0.925
0.98 0.9
0.97
0.875
0.96
0.85
10 15 20 25 30 35 40 45 50
VLmin.
3 4 6 7 9 10 12 1415161718192021222324252627282930 Load buses
Load buses
Fig. 6 Initial values of load voltages for WDN system
Fig. 5 Initial values of load voltages for IEEE 30-bus system
123
Electr Eng (2011) 93:103–116 111
1.08
Table 4 ORPD settings of control and dependent variables using dif-
ferent optimization techniques for IEEE 14-bus system
Control/dependent LP GA PSO ACO 1.06
variables VLmax.
1.04
1.08 VLmax.
Load bus voltage (pu.)
1.04
1.06
VLmax.
1.02
Load bus voltage (pu.)
1.04
Vsp.
1.02
Vsp. 0.98
0.98 0.96
VLmin.
3 4 5 6 7 8 9 10 11 12 13 14
0.96
VLmin.
Load buses
3 4 5 6 7 8 9 10 11 12 13 14
Load buses Fig. 9 Load bus voltages using PSO technique for IEEE 14-bus system
Fig. 7 Load bus voltages using LP technique for IEEE 14-bus system
level of reserve power compared to other optimization tech- algorithm for ORPD problem leads to the obtain the high-
niques. est reduction of transmission power losses (0.1631 pu) while
Figures 7, 8, 9 and 10 show the load bus voltages using other optimization techniques lead to the following transmis-
different optimization techniques LP, GA, PSO, and ACO sion power losses 0.2014, 0.1922, and 0.1634 pu for LP, GA,
algorithm, respectively. The voltages at load-buses 6 and 8 and PSO, respectively. The highest reduction of power losses
still violated their limits using the conventional LP only while leads to the lowest payment for power losses of about 19%
other optimization techniques have preserved all load volt- reduction compared to LP optimization technique. Further-
ages within their permissible limits. more, the use of ACO algorithm leads to the lowest voltage
deviation (V.D.)rms of 0.0135 compared to 0.03, 0.0269, and
6.2.2 IEEE 30-bus test system 0.0257 using the other optimization techniques, respectively.
Added to the above performance, ACO algorithm leads to the
Tables 6, 7 and Figs. 11, 12, 13 and 14 show the ORPD using highest has RPR (0.1769) compared with other techniques.
different optimization techniques. In Table 6, the use of ACO Similar to the remark presented above for the IEEE 14-bus
123
112 Electr Eng (2011) 93:103–116
1.08
VLmax.
1.04
1.06
1.04
1.01
Vsp.
1.02
0.99
0.98
Vsp.
0.97
0.96
0.98
VLmin.
3 4 6 7 9 10 12 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
VLmin.
3 4 5 6 7 8 9 10 11 12 13 14 Fig. 11 Load bus voltages using LP technique for IEEE 30-bus system
Load buses
Fig. 10 Load bus voltages using ACO for IEEE 14-bus system
VLmax.
1.04
Table 6 ORPD solution using different optimization techniques for
VLmin.
3 4 6 7 9 10 12 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
123
Electr Eng (2011) 93:103–116 113
0.99
0.98 1.06
VLmax.
0.97 1.04
1.02
0.96
0.9
0.88
system size, it is found that the percentage of power losses
0.86
reduction is increased from 15 to 19%.
10 15 20 25 30 35 40 45 50
Table 7 shows the results of control variables (VG and Load buses
Q SW ) and dependent variables (Q G ).
Figures 11, 12, 13, and 14 show the load-bus voltages (VL ) Fig. 15 Load bus voltages using LP for WDN system
that eliminate the voltage violations using LP, GA, PSO, and
1.06
ACO, respectively. VLmax.
1.04
Vsp.
Load bus voltage (pu.)
Tables 8, 9 and Figs. 15, 16, 17, and 18 show the ORPD using 0.98
different optimization techniques.
0.96
VLmin.
0.94
Table 8 Settings of control/dependent variables using different opti-
mization techniques for WDN System 0.92
Vg 4 1.0019 1.0374 1.0241 0.9835 Fig. 16 Load bus voltages using GA for WDN system
Vg 5 1.0130 0.9638 1.0113 1.0465
Vg 6 1.0084 1.0223 0.9794 0.9544
Vg 7 1.0076 1.0125 1.0387 1.0182 Table 8 shows the results of control variables (VG ) and
Vg 8 1.0066 0.9872 0.9686 0.9673 dependent variables (Q G ). In Table 9, the use of ACO algo-
Qg 1 −0.9314 −0.4238 −0.106 −0.682 rithm for ORPD problem leads to obtain the highest reduc-
Qg 2 0.7907 0.4649 0.530 0.131 tion of transmission power losses (PL = 0.1698 pu) while
Qg 3 −0.7342 0.6398 0.940 −0.898 other optimization techniques lead to the following transmis-
Qg 4 −1.7452 −1.2347 −1.470 −1.237 sion power losses 0.2786, 0.2116, and 0.1873 pu for LP, GA,
Qg 5 1.1691 0.9674 0.630 0.862 and PSO, respectively. The highest reduction of power losses
Qg 6 −1.4325 −1.3832 −1.001 −0.799 leads to the lowest payment for power losses of about 61%
Qg 7 1.4185 1.4924 1.362 1.037
of the power losses obtained using LP optimization tech-
Qg 8 −2.0135 −0.1865 −0.500 −0.217
nique. Moreover, the use of ACO algorithm leads to the low-
est voltage deviation (V.D.) rms of 0.024 compared to 0.044,
123
114 Electr Eng (2011) 93:103–116
1.06 0.005
Vsp. -0.005
0.98
-0.01
0.96
VLmin. -0.015
0.94
0.92 -0.02
0.9
-0.025
0.88
0.86 -0.03
0 10 20 30 40 50 60 70 80 90 100
10 15 20 25 30 35 40 45 50
Load buses Iteration
Fig. 17 Load bus voltages using PSO technique for WDN system Fig. 19 Convergence curve of different optimization techniques for
14-bus test system
1.06
VLmax.
1.04
1.02 0.25
Vsp. PSO
ACO
0.2
0.98
0.96 0.15
VLmin.
0.94
0.1
0.92
0.9 0.05
0.88
0
0.86
10 15 20 25 30 35 40 45 50
-0.05
Load buses 0 10 20 30 40 50 60 70 80 90 100
Iteration
Fig. 18 Load bus voltages using ACO algorithm for WDN system
Fig. 20 Convergence curves of different optimization techniques for
IEEE 30-bus test system
0.034, and 0.0317 using the other optimization techniques,
respectively. Similar to the remarks presented above for the
IEEE 14-bus test system and for the IEEE 30-bus test sys-
0.12
tem, the use of ACO for solving the ORPD increases the GA
Incremental Change in Line Losses (pu.)
reductions of power losses and voltage deviation in the case 0.1 PSO
ACO
of the real power system at WDN. Then the proposed ACO 0.08
can be considered as an efficient tool to face the emergency
0.06
conditions the power system face according to the highest
level of reserve power compared to other optimization tech- 0.04
niques. Figures15, 16, 17, and 18 show that the load voltages 0.02
based on ACO algorithm are the nearest to the flat voltage
0
compared to the other techniques.
-0.02
-0.04
6.3 Results evaluation
-0.06
Figures 19, 20, and 21 evaluate the previous results for the 0 10 20 30 40 50 60 70 80 90 100
Iteration
three studied systems. It is cleared that the ACO algorithm
leads to the lowest power losses, the highest reactive power Fig. 21 Convergence curves of different optimization techniques for
reserve and minimum deviation from the flat voltage. The IEEE 30-bus test system
123
Electr Eng (2011) 93:103–116 115
Table 10 Statistical evaluation of different optimization tools for the Table A.1 Genetic algorithm (GA) parameters
IEEE 14-bus, IEEE 30-bus, and WDN systems
Number of populations (iterations) Npop 100
Tool Test system Best Ploss Worst Ploss Computation Population size Pop 80
time (s)
Probability of crossover Pc 50%
GA 14-bus −0.02373 0.00362 1.33 Probability of mutation Pm 5%
30-bus 0.01713 0.24113 2.11 Number of crossover points Cop 2
WDN −0.01651 0.11105 2.45 Precision Pr 1,000
PSO 14-bus −0.02661 0.00435 1.15 Termination of precision Pter 0.001
30-bus −0.01192 0.16893 1.83 Termination number of iteration Nter 100
WDN −0.04091 0.08105 2.05
ACO 14-bus −0.02491 0.00311 0.99
30-bus −0.01215 0.15341 1.69 Table A.2 PSO parameters
WDN −0.05851 0.07321 1.97 Number of populations (iterations), (npop) 100
Number of particles, (pop) 100
characteristics of the proposed algorithm have been com- Inertia Constant Range (w) 0.4–0.9
pared to other optimization technique. Particle velocity limits ±0.8 × range
In terms of the convergence evaluation, fast convergence Learning coefficients c1 = c2 2
with minimum computation time as shown in Table 10 at the
best solution for different tested systems is achieved using
ACO. Table 10 shows the computation time needed for solv-
ing the ORPD using different optimization techniques for Appendix A
three power systems. It is cleared that the ACO leads to the
highest power losses reduction and leads to minimum com- The ORPD problem is solved using the proposed ACO algo-
putation time compared to GA and PSO optimization algo- rithm. The suggested ACO-based procedure is compared
rithms. with two search-based optimization algorithms, namely
genetic algorithm and particle swarm optimization tech-
nique, which are used in this paper. The parameters of these
7 Conclusions optimization techniques are genetic algorithm parameters as
shown in Table A.1 and particle swarm optimization algo-
This paper has been proposed as an optimal procedure based rithm parameters as shown in Table A.2.
on ACO algorithm to solve the ORPD problem. In the sug-
gested procedure for ORPD, the objective is to minimize
the transmission line losses under control and dependent References
variable constraints. The control and dependent variables
are computed using proposed sensitivity parameters of reac- 1. Li F, Pilgrim JD, Dabeedin C, Chebbo A, Aggarwal RK
tive power sources at generation and switchable buses. The (2005) Genetic algorithms for optimal reactive power compensa-
proposed sensitivity parameters have been computed based tion on the national grid system. IEEE Trans Power Syst 20:493–
500
on fast decoupled power flow model. Three test systems
2. Subbaraj P, Rajnarayanan PN (2009) Optimal reactive power dis-
have been used to show the capability of the proposed algo- patch using self-adaptive real coded genetic algorithm. Electr
rithm compared with other optimization techniques as LP, Power Syst Res 79:374–381
GA, and PSO. The simulation results show that the ACO 3. Zhang Y, Ren Z (2005) Optimal reactive power dispatch consider-
ing costs of adjusting the control devices. IEEE Trans Power Syst
algorithm leads to highest reduction of power losses with 20:1349–1356
the lowest level of voltage deviation for the tested systems. 4. Yan W, Lu S, Yu DC (2004) A novel optimal reactive power dis-
Moreover, the ACO algorithm leads to more reduction in the patch method based on an improved hybrid evolutionary program-
cost of switchable reactive power while efficiently increases ming technique. IEEE Trans Power Syst 19:913–918
5. Zhao B, Guo CX, Cao YJ (2005) A multiagent-based particle
the reactive power reserve compared to the other proposed swarm optimization approach for optimal reactive power dispatch.
algorithms. So, the proposed ACO algorithm leads to more IEEE Trans Power Syst 20:1070–1078
accurate, efficient and fast convergence solution for ORPD 6. Yan W, Lu S, Yu DC (2004) A novel optimal reactive power dis-
problem compared to other optimization techniques. The pro- patch method based on an improved hybrid evolutionary program-
ming technique. IEEE Trans Power Syst 19:913–918
posed algorithm represents a potential tool to aid the power 7. Dai C, Chen W, Zhu Y, Zhang X (2009) Reactive power dispatch
system operators in the on-line environment, especially for considering voltage stability with seeker optimization algorithm.
the case of large-scale power systems. Electr Power Syst Res 79:1462–1471
123
116 Electr Eng (2011) 93:103–116
8. Dorigo M (1992) Optimization, learning and natural algorithms. 17. Favuzza S, Graditi G, Sanseverino ER (2006) Adaptive and
PhD dissertation, Department of Electronics, Politecnico di Mi- dynamic ant colony search algorithm for optimal distribution sys-
lano, Milan tems reinforcement strategy. Appl Intell 24:31–42
9. Dorigo M, Stützle T (2004) Ant colony optimization. MIT Press, 18. Carpaneto E, Chicco G (2008) Distribution system minimum loss
New York reconfiguration in the hyper-cube ant colony optimization frame-
10. Dorigo M, Gambardella LM (1997) Ant colonies for the traveling work. Electr Power Syst Res 78:2037–2045
salesman problem. Biosystems 43:73–81 19. Niknam T (2008) A new approach based on ant colony optimiza-
11. Yu I-K, Chou CS, Song YH (1998) Application of the ant colony tion for daily Volt/Var control in distribution networks considering
search algorithm to short-term generation scheduling problem of distributed generators. Energy Convers Manage 49:3417–3424
thermal units. In: Proceedings of international conference on power 20. Lenin K, Mohan MR (2006) Ant colony search algorithm for opti-
system technology, vol 1, POWERCON ’98, Aug 1998, Beijing, mal reactive power optimization. Serbian J Electr Eng 3:77–88
China, pp 552–556 21. Ketabi A, babaee AA, Feuillet R (2010) Application of the ant col-
12. Song YH, Chou CSV, Min Y (1999) Large-scale economic dis- ony search algorithm to reactive power pricing in an open electricity
patch by artificial ant colony search algorithms. Electr Power Com- market. Electr Power Energy Syst 32:622–628
pon Syst 27:679–690 22. Pechač P (2002) Electromagnetic wave propagation modeling
13. Hou YH, Wu YW, Lu LJ, Xiong XY (2002) Generalized ant col- using the ant colony optimization algorithm. Radio Eng 11:1–6
ony optimization for economic dispatch of power systems. IEEE 23. Lee KY, Vlachogiannis JG (2005) Optimization of power systems
Int Conf Power Syst Technol 1:225–229 based on ant colony system algorithms: an overview. In: 13th inter-
14. EL-Sharkh MY, Sisworahardjo NS, Rahman A, Alam MS (2006) national conference on intelligent systems application to power
An improved ant colony search algorithm for unit commitment systems, 6–10 Nov 2005, pp 22–35
application. In: IEEE power systems conference and exposition, 24. Washington University. http://www.ee.washington.edu/research/
pp 1741–1746 pstca/
15. Vlachogiannis JG, Hatziargyriou ND, Lee KY (2005) Ant colony 25. El-Sehiemy RA (2008) Performance of transmission network
system-based algorithm for constrained load flow problem. IEEE under deregulated electrical power system. PhD thesis, Faculty of
Trans Power Syst 20:1241–1249 Engineering, Minoufiya University, Egypt
16. Meziane R, Massim Y, Zeblah A, Ghoraf A, Rahli R (2005) Reli- 26. Abou El-Ela AA, Allam SM, Shatla MM (2010) Maximal optimal
ability optimization using ant colony algorithm under performance benefits of distributed generation using genetic algorithms. Electr
and cost constraints. Electr Power Syst Res 76:1–8 Power Syst Res 80:869–877
123