Electr Eng (2011) 93:103–116
DOI 10.1007/s00202-011-0196-4
ORIGINAL PAPER
Optimal reactive power dispatch using ant colony
optimization algorithm
A. A. Abou El-Ela · A. M. Kinawy ·
R. A. El-Sehiemy · M. T. Mouwafi
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
optimal reactive power dispatch (ORPD) problem using ant
colony optimization (ACO) algorithm. The objective of the
ORPD problem is to minimize the transmission power losses
under control and dependent variable constraints. Proposed
sensitivity parameters of reactive power at generation and
switchable sources are derived based on a modified model of
fast decoupled power flow. The proposed ACO-based algo-
rithm is applied to the IEEE standard 14-bus, 30-bus sys-
tems, and a real power system at West Delta Network as a
part of the Unified Egyptian Network. The obtained simula-
tion results are compared with those of conventional linear
programming, genetic algorithm, and particle swarm optimi-
zation technique. Simulation results show the capability of
the proposed ACO-based algorithm for solving the ORPD
problem, especially with increasing the system size.
Keywords Ant colony optimization · Genetic algorithm ·
Particle swarm optimization · Control variables ·
Dependent variables
1 Introduction
optimal reactive power dispatch (ORPD) is one of the opti-
mization problems in power systems that optimized the set-
tings of control variables using conventional, intelligence,
and modern optimization techniques. The objective function
A. A. Abou El-Ela · A. M. Kinawy · M. T. Mouwafi
Electrical Engineering Department, Faculty of Engineering,
Minoufiya University, Shebin El-Kom, Egypt
R. A. El-Sehiemy (B)
Electrical Engineering Department, Faculty of Engineering,
Kafrelsheikh University, Kafr El-Sheikh, Egypt
e-mail: elsehiemy@eng.kfs.edu.eg
of ORPD is to minimize the transmission line losses by opti-
mizing the settings of control variables such as generation
voltages, switchable reactive power, and on-load tap chang-
ers (OLTC) under control and dependent variable constraints.
In recent years, many optimization techniques were used for
solving the reactive power dispatch problem.
Li et al. [1] proposed an integer-coded multi-objective
genetic algorithm (IGA) to reactive power compensation
planning (RCP) problem considering both intact and con-
tingent operating states. The IGA solved both the sitting
and installing of reactive power compensation devices and
the operational characteristics as multi-objective optimiza-
tion problem.
Subbaraj and Rajnarayanan [2] developed a self-adap-
tive real coded genetic algorithm (SARGA) as one of the
techniques to solve the ORPD problem. The problem for-
mulation involves continuous (generator voltages), discrete
(transformer tap ratios), and binary (Var sources) decision
variables.
Zhang and Ren [3] presented a model of ORPD for power
systems based on the principle of income maximization,
which aimed at minimizing active power loss of the whole
networks. On the other hand, the operations would augment
the fault probability of power systems. Therefore, the costs
of adjusting the control devices (CACD) were investigated,
and a mathematical model of ORPD was implemented to
minimize the energy loss at the current time interval and the
CACD.
Yan et al. [4] proposed an improved evolutionary
programming (IEP) and its hybrid version combined with
the nonlinear interior point (IP) technique to solve the ORPD
problem. In an IEP method, the common practices in regulat-
ing reactive power followed in adjusting the mutation direc-
tion of control variables in order to increase the possibility
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

solved using an improved hybrid evolutionary programming PL = Pi (3)

i=1
technique. The seeker optimization algorithm for ORPD was
proposed in [7]. The transmission power losses equal to the sum of the
Recently, a new method based on the ant colony optimi- injected powers (Pi ) at all system buses. The injected power
zation (ACO) is investigated by Dorigo [8]. The ACO was at bus i can be computed in terms of load flow equations:
employed to solve different combinatorial optimization prob-
lems [8,9] as presented for travelling salesman problem [10]. Si = Pi + j Q i = Vi Ii∗ (4)
Currently, most published works are carried out in the direc-
tion of develop ACO algorithm to solve for most of these But the injected current at bus I is computed from:
applications; the results show that the ACO can outperform
other heuristic methods. 
NB
Through few recent years, ACO algorithms were Ii = Yi j V j (5)
employed to solve different optimization problems with more j=1
accurate and efficiently solution compared with conven-
tional and other modern optimization algorithms. In the field The total injected apparent power in Eq. (4) can be derived
of power systems, the ACO has been applied to solve the as shown in (6) as:
optimum generation scheduling problems [11], economic

NB
 ∗
dispatch of power systems [12,13], unit commitment [14], Si = Vi Yi j V j
the constrained load flow [15], for solving the reliability opti- j=1
mization problem under performance and cost constraints
[16], adaptive and dynamic ant colony search algorithm for 
N
 
= Vi G i j − j Bi j V j∗
optimal distribution systems reinforcement strategy [17], for j=1
distribution system minimum loss reconfiguration [18],
for daily Volt/Var control in distribution networks consider- 
NB
    
= G i j − j Bi j (|Vi |  δi ) V j   − δ j
ing distributed generators [19] for solving the reactive power j=1
dispatch model [20] and for solving the reactive power pric-
ing problem in the deregulated open access markets [21]. 
N
  
= |Vi | G i j − j Bi j V j   δi j
It is rather difficult to find a single search space, config-
j=1
uration, and parameter set of ACO algorithm that can sat-
isfy every optimization problem. The ACO proposed in this 
N
   
= |Vi | G i j − j Bi j cos δi j + j sin δi j V j 
paper formulates the reactive power dispatch problem as a
j=1
combinatorial optimization problem. The minimization of
the transmission line losses are considered as an objective 
N
 
= |Vi | G i j cos δi j + Bi j sin δi j
function with control and dependent variable constraints.
j=1
   
+ j G i j sin δi j − Bi j cos δi j V j  (6)
2 Problem formulation
Then the active power at bus i can be formulated as:
The ORPD problem can be expressed as an optimization
problem as: 
N
  
Pi = |Vi | G i j cos δi j + Bi j sin δi j V j  (7)
Min f (x) (1) j=1

g (x) = 0
Subject to : (2) At i = j, Eq. (6) can be rewritten as;
h (x) ≤ 0
where f (x) is the objective function such as transmission 
N
  
line losses, generator fuel costs, etc., g(x), h(x) represent Pi = Vi2 G ii + |Vi | G i j cos δi j + Bi j sin δi j V j  (8)
the equality and inequality constraints, respectively, and x is j=1, j ∈i
/
the vector of the control variables. 
NB
The objective function f (x) is aimed at minimizing the PL = Pi
i=1
total transmission line losses.

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:

∂ PL /∂ Q SW = (∂ PL /∂ VSW ) (∂ VSW /∂ Q SW ) (13) VLmin ≤ VL ≤ VLmax

VLmin = VLmin − VLinit (16)
VLmax = VLmax − VLinit
where ∂ VSW /∂ Q SW refers to the change of switchable volt-
age buses with respect to change in switchable reactive power
at that bus and is determined based on sensitivity parameters where VLinit is the initial value of the voltage at each load bus,
that will be illustrated in the next section. VLmin and VLmax are the minimum and maximum of load bus
The objective function (3) is subjected to the following voltages, respectively, VLmain and VLmax are the minimum
constraints for control and dependent variables: and maximum changes in load bus voltages, respectively.

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

VL = − [B L L ]−1 [B LG ] VG

 
where Q init
G is the initial value of reactive power at generation VL = SGL VG (20)
buses, Q min max
G and Q G are the minimum and maximum values
of reactive power at generation buses, respectively, Q main
G where SGL = − (B L L )−1 B LG .
and Q max
G are the minimum and maximum changes in reac-
tive power at generation buses, respectively. 3.1.2 No changes in generation voltages (VG = 0)

The sensitivity parameters relating to the changes in load-

bus voltages due to the changes in switchable reactive power
3 Proposed sensitivity parameters sources can be given as:
 
The FDPF method is one of the load flow methods that VL = SSW L
Q SW (21)
assumed the changes in active power flows due to changes
in voltage phase angles which are higher than the changes L = − (B )−1 V −1 , V
where SSW LL SW SW is the bus voltage that
due to voltage magnitudes. On the other hand, the changes in connected to VAR sources.
reactive power flows due to changes in voltage magnitudes
are higher than the changes due to voltage phase angles. So, 3.2 Generation reactive power related to control variables
it can be written as:
The sensitivity parameters corresponding to the changes in
[P/V ] = [B  ][δ] generated reactive power due to the changes in generation-
(18) bus voltages can be written as:
[Q/V ] = [B  ][V ]
Q G /VG = [BGG ] VG + [BG L ] VL (22)
where [P] and [Q] are the vectors of active and reac- By substituting from Eq. (20) into (22), we obtain:
tive power mismatches, respectively, [B  ] and [B  ] are the  
susceptance matrices. These matrices can be computed as: Q G /VG = [BGG ] VG + [BG L ] SGL VG
 

N −1 Q G = SGG VG (23)
Bii = 1/ X i j , i = 1, 2, . . . , N − 1  
j=1, j ∈i
/ where SGG = VG BGG + BG L SGL .
The sensitivity parameters relating to the changes in gener-
Bij = 
B ji = −1/ X i j , i, j = 1, 2, . . . , N − 1
ated reactive power due to the changes in switchable reactive
power sources can be written as:
where X i j is the line reactance between buses i and j, [δ]
 −1
and [V ] are the vectors of changes in voltage angles and Q G /VG = [BGG ] SGL VL + [BG L ] VL (24)
magnitudes, respectively.
The ORPD problem is based on the sensitivity parameters By substituting from Eq. (21) into (24), we get:
derived based on the second equation of the FDPF model   −1  
(18) which can be rewritten as: Q G /VG = BGG SG L
+ BG L SSW L
Q SW
    
Q G /VG BGG BG L VG Q G /VG = SSW G
Q SW (25)
= (19)
Q L /VL B LG B L L VL  G    −1  
where SSW = VG BGG SGL + BG L SSW L .

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
equations in a compact matrix form as:
   obstacle as shown in Fig. 1c.
VL SGL SSW
L VG
= (26)
Q G SGG SSW
G Q SW
4 ACO algorithm
ACO algorithms were first proposed by Dorigo [8–10].
4.1 Description of real ants
ACO algorithms are based on the behavior of real ants that
are members of a family of social insects. However, a group
of explorer ants leave the colony for finding the food source
in a randomly directions where they marked their routes by
laying a chemical substance on the ground. Other ants are
attracted to the route that has the largest amount of phero-
mone that decays with time. So that, a shorter route will be
found that has a largest amount of pheromone than a longer
route. So that, they are found the shortest route between the
nest and food source by indirect communication media called
pheromone that laid on the ground as a guide for another
ants. Figure 1 shows how the real ants find the shortest path
between nest or colony and food source [22].
In Fig. 1a, there are no obstacles between nest and food
source. However, the shortest route is the straight line. If an
obstacle is located on the route of ants to become two routes
around the obstacle, some of ants choose the left side around
the obstacle and the other will choose the right side as shown
in Fig. 1b; the pheromone laid on the left side will be concen-
trated than right side of obstacle because ants in the shortest
path takes minimum time in leaving and returning for nest
where they moves in the same speed. So, they will lay a larg-
est amount of pheromone than other ants on the other route,
while the other ants are attractive to the shortest route. Hence,
Fig. 1 Illustration of real ant behavior [22]
107
all ants in the colony will take the shortest route around the
4.2 Mathematical model of ACO algorithm
A random amount of pheromone is deposited in each route
after each ant completes its tour; anther antes are attracted
to the shortest route according to the probabilistic transition
rule that depends on the amount of pheromone deposited
and a heuristic guide function as equal to the inverse of the
distance between beginning and ending of each route. The
probabilistic transition rule of ant k to go from city i to city
j can be expressed as in Traveling Salesman Problem (TSP)
[10] as:
 α  β
τi j (t) ηi j (t)
Pikj (t) =  α  β , j, q ∈ Nik (27)
q τiq (t) ηiq (t)
where τi j is the pheromone trail deposited between city i
and j by ant k, ηi j is the visibility or sight and equal to the
inverse
 of the distance
 or the transition cost between city i
and j ηi j = 1/di j . The two parameters α and β have shown
the relative weight of pheromone trail and heuristic guide
function, respectively. If α = 0, the closest cities are more
likely to be selected that corresponding to a classical greedy
algorithm. On the contrary, if β = 0, the probability will be
depend on the pheromone trial only. These two parameters
should be tuned with each other; Dorigo [10] found experi-
mentally the good values of α and β are 1 and 5, respectively;
q is the cities that will be visited after city i while Nrk is a
tabu list in memory of ant that recodes the cities which will
be visited to avoid stagnations. After each tour is completed,
a local pheromone update is determined by each ant depend-
ing on the route of each ant as in Eq. (28); after all ants are
attracted to the shortest route, a global pheromone update is
considered to show the influence of the new addition deposits
by the other ants that are attracted to the best tour as shown
in Eq. (29):
τi j (t + 1) = (1 − ρ) τi j (t) + ρτ◦ (28)
τi j (t + 1) = (1 − ρ) τi j (t) + ετi j (t) (29)
where, τi j (t + 1) is the pheromone after one tour or iteration,
ρ is the pheromone evaporation constant equals to 0.5 as a
good value by Dorigo in [10], ε is the elite path weighting
constant, τo = 1/di j is the incremental value of pheromone
of each ant. τi j is the amount of pheromone for elite path as:
τi j (t) = 1/dbest (30)
where dbest is the shortest tour distance.
123
108
5 ACO algorithm for ORPD problem
The ACO algorithm was used to solve many power system
problems [23]. In this paper, the ACO algorithm is applied
to solve the ORPD problem as an optimization technique
with control and dependent variable constraints where artifi-
cial ants travel in search space to find the shortest route that
has the strongest pheromone trail and a minimum objective
function. Our objective in this paper was to minimize the
transmission power losses as described in (3) with control
variable constraints and dependent variable constraints.
The heuristic guide function is the inverse of the individual
change in losses of each ant that positioned in the reasonable
limit of the control variable to the visibility of each ant, while
heuristic guide function of the problem is the inverse of the
total change in losses at iteration t + 1 as:
 NG
 
η(t + 1) = 1 (∂ PL /∂ VG ) VG
G=1
 Is Max Iteration is reached

NSW
+ (∂ PL /∂ Q SW ) Q SW (31)
SW=1
In the genetic algorithm (GA), a chromosome is subdivided
into genes; each gene represents a variable, which consists of
a binary string with length that depends on the boundary of
the corresponding variable. In particle swarm optimization
(PSO), a swarm consists of particles or populations that cor-
responding to the control variables. The parameters of the GA
and PSO employed in this paper are reported in Appendix A.
In ACO algorithm, a search space creates with dimen-
sions of stages on number of control variables and states or
the randomly distributed values of control variables with in
a reasonable threshold. Artificial ants leave colony to search
randomly in the search space based on the probability in (27)
to complete a tour matrix that consists of the positions of
ants with the same dimension of the search space. Then tour
matrix is applied on the objective function to find a heuristic
guide function to find the best solution and update local and
global pheromone to begin a next iteration. System param-
eters are adjusted by trail and error to find the best values
of theses parameters. The ACO algorithm can be applied to
solve the ORPD problem using the following steps which
can be indexed in a flowchart for ORPD problem as shown
in Fig. 2.
Step 1: Initialization
In this step, it is needed to insert the lower and upper
boundaries of each control variable, system parameters,
and create a search space with a dimensions of number of
control variables and the length of randomly distributed
values with the same dimension of the initial pheromone
123
Electr Eng (2011) 93:103–116
Start
Initialization ; Insert parameters, control limits and
initial pheromone
Generate the initial position randomly of each ant
Apply state transition rule
Apply local pheromone updating rule
Fitness function evaluation (power losses)
Apply global pheromone updating rule
Checking Constraints
No
Yes
Stop
Fig. 2 Flow chart of ACO algorithm
that contains elements with very small equal values to give
all ants with the same chance of searching.
Step 2: Provide first position
In the second step, each ant is positioned on the initial
state randomly within the reasonable range of each con-
trol variable in a search space with one ant in each control
variable in the length of randomly distributed values.
Step 3: Transition rule
Each ant decides to visit a next position in the range of
other control variables according to the probability transi-
tion rule in Eq. (28) that depends on the amount of pher-
omone deposited and the visibility that is the inverse of
objective function (30), where the effect of pheromone and
visibility on each other depends on the two parameters α
and β.
Step 4: Local pheromone updating
Local updating pheromone is different from ant to other
because each ant takes a different route. The initial pher-
omone of each ant is locally updated as in Eq. (28).
Step 5: Fitness function
After all ants are attracted to the shortest path that has
a strongest pheromone, the best solution of the objective
function is obtained.
Step 6: Global pheromone updating
Amount of pheromone on the best tour becomes the
Electr Eng (2011) 93:103–116 109

Table 1 Control/dependent variables limits and initial values for IEEE

12 13 14
14-bus system
Control/dependent Min. Max. Initial
variables
6 11 10 9
Vg 1 1.00 1.10 1.060
Vg 2 1.00 1.10 1.045
8 7 Q sw 9 0.00 0.20 0.000
Q sw 14 0.00 0.15 0.000
Qg 1 −0.50 1.50 −0.169
1 5 4 Qg 2 −0.45 1.50 −0.424

1.09

1.08

Initial value of load voltages

1.07
2 3
1.06

VLmax.

Fig. 3 Single line diagram for 14-bus test system 1.04

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

The combination between the ACO-based procedures and

the modified FDLF method appears on the following two
main points:
• Using the proposed sensitivity parameters for evaluating
the dependent variables in terms of control variables as
discussed in Sect. 3.
• At each iteration is checked the system constraints using
the FDLF method.
6 Applications
6.1 Test systems
Two standard test systems and a real power system are used
to show the capability of the proposed technique to find the
ORPD using ACO algorithm. The two standard systems are
the standard IEEE 14-bus as shown in Fig. 3 and 30-bus test
systems [24].
In addition, a real power system at West Delta Network
as a part of the Unified Egyptian Network (UEN) [25,26] is
used as a test power system. The MVA base is taken as 100,
and the cost of power losses is assumed to be 0.07 E.P./kW
while the cost of reactive power is assumed 15 E.P./kVar.
The results obtained are compared with those obtained using
a conventional linear programming (LP), GA, and PSO tech-
nique. The parameters of the GA and PSO are presented in
Appendix A.
As default starting values, the parameters from [10,22]
were used: α = 0.5, β = 3ρ = 0.2, and ε = 0.2. The best
values of ACO algorithm parameters for the ORPD problem
are α = 1, β = 5ρ = 0.5, and ε = 5. The load flow is
done using the Newton–Raphson load flow to get the val-
ues of power loss that are 13.593, 17.528, and 22.811 MW
for IEEE 14-bus, 30-bus systems, and West Delta Network
(WDN), respectively. Table 1 and Fig. 4 show the minimum,
maximum limits, and initial values of control and dependent
variables for IEEE 14-bus test system, while Table 2 and
Fig. 5 show the minimum, maximum limits, and initial val-
ues of control and dependent variables for IEEE 30-bus test
system. In addition, Table 3 and Fig. 6 show the minimum,
maximum limits, and initial values for control and dependent
variables for WDN.
6.2 Results and comments
The results of the ACO algorithm are obtained using the
MATLAB code version 7.1 that was setup on a Pentium 4,
3.0 GHz PC, 0.99 GB of RAM.

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

Vg 1 1.00 1.10 1.050 Vg 1 1.00 1.10 1.05

Vg 2 0.95 1.05 1.034 Vg 2 0.95 1.05 1.00
Vg 5 0.95 1.05 1.006 Vg 3 0.95 1.05 1.00
Vg 8 0.95 1.05 1.023 Vg 4 0.95 1.05 1.00
Vg 11 1.00 1.10 1.091 Vg 5 0.95 1.05 1.00
Vg 13 1.00 1.10 1.099 Vg 6 0.95 1.05 1.00
Q sw 10 0.00 0.20 0.000 Vg 7 0.95 1.05 1.00
Q sw 17 0.00 0.15 0.000 Vg 8 0.95 1.05 1.00
Q sw 24 0.00 0.10 0.000 Qg 1 −1.50 1.50 0.3595
Qg 1 −0.30 1.00 −0.026 Qg 2 −1.50 1.50 0.0640
Qg 2 −0.40 0.50 −0.103 Qg 3 −1.20 1.20 1.0817
Qg 5 −0.20 0.60 0.036 Qg 4 −2.00 1.50 1.0315
Qg 8 −0.20 0.70 0.048 Qg 5 −1.00 1.20 0.5169
Q g 11 −0.06 0.34 0.308 Vg 6 −1.50 1.50 0.8388
Q g 13 0.06 0.37 0.379∗ Vg 7 −1.50 1.50 0.9836
* Violation of its limit Vg 8 −2.50 1.20 0.3114

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

power losses (0.1094 pu) while other optimization techniques

6.2.1 IEEE 14-bus test system
Tables 4, 5 and Figs. 7, 8, 9 and 10 show the ORPD using
different optimization techniques. Table 4 shows the results
of control variables (VG and Q SW ) and dependent variables
(Q G ).
Table 5 shows the results of power loss, cost of power loss
(CPL), root mean square of load-bus voltage deviation with
respect to the flat voltage (V.D.)rms , reactive power reserve
(RPR) and cost of the connected reactive power (CRP).
In Table 5, the use of ACO algorithm for ORPD prob-
lem leads to the obtain the highest reduction of transmission
lead to the following transmission power losses 0.1273,
0.1124, and 0.1111 pu for LP, GA, and PSO, respectively.
The highest reduction of power losses leads to the lowest
payment for power losses of about 15% reduction compared
to LP optimization technique. Moreover, the use of ACO
algorithm leads to the lowest voltage deviation (V.D.)rms of
0.0137 compared to 0.0414, 0.0258, and 0.022 using the other
optimization techniques, respectively. Added to the above
performance, ACO algorithm leads to the highest has RPR
compared with other techniques. The last result means that
the ACO-based may be an efficient tool to face the emergency
conditions the power system face according to the highest

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

Load bus voltage (pu.)

Vg 1 1.060 1.0059 1.0002 1.0000
Vg 2 1.000 1.0042 1.0033 1.0002
Q sw 9 0.0773 0.0748 0.0847 0.0021 1.02

Q sw 14 0.0582 0.0471 0.0323 0.0015

Qg 1 −0.1919 −0.1825 −0.1745 −0.1593 Vsp.
Qg 2 0.46068 0.4563 0.4185 0.3862
0.98

Table 5 A comparison between different optimization techniques for

14-bus system 0.96

LP GA PSO ACO VLmin.

3 4 5 6 7 8 9 10 11 12 13 14
Load buses
Power loss (pu.) 0.1273 0.1124 0.1111 0.1094
CPL (EP) 891.1 786.69 777.7 765.44 Fig. 8 Load bus voltages using GA for IEEE 14-bus system
(V.D.)rms 0.0414 0.0258 0.022 0.0137
RPR 0.2145 0.2281 0.233 0.253 1.08
CRP (EPx103 ) 203.25 182.85 175.5 145.5
1.06

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-
niques.
Figures 7, 8, 9 and 10 show the load bus voltages using
different optimization techniques LP, GA, PSO, and ACO
algorithm, respectively. The voltages at load-buses 6 and 8
still violated their limits using the conventional LP only while
other optimization techniques have preserved all load volt-
ages within their permissible limits.
6.2.2 IEEE 30-bus test system
Tables 6, 7 and Figs. 11, 12, 13 and 14 show the ORPD using
different optimization techniques. In Table 6, the use of ACO
algorithm for ORPD problem leads to the obtain the high-
est reduction of transmission power losses (0.1631 pu) while
other optimization techniques lead to the following transmis-
sion power losses 0.2014, 0.1922, and 0.1634 pu for LP, GA,
and PSO, respectively. The highest reduction of power losses
leads to the lowest payment for power losses of about 19%
reduction compared to LP optimization technique. Further-
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
0.0257 using the other optimization techniques, respectively.
Added to the above performance, ACO algorithm leads to the
highest has RPR (0.1769) compared with other techniques.
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

Load bus voltage (pu.)

1.03
VLmax.
1.02
Load bus voltage (pu.)

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

0.96 Load buses

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

Load bus voltage (pu.)

1.03
30-bus system
1.02
LP GA PSO ACO
1.01

Power loss (pu.) 0.2014 0.1922 0.1634 0.1631 Vsp.

CPL (EP) 1,409.8 1,345.4 1,143.8 1,141.7 0.99

(V.D.)rms 0.0300 0.0269 0.0257 0.0135 0.98

RPR 0.0860 0.093 0.1055 0.1769 0.97

CRP (EP.103 ) 546.00 535.5 516.75 409.65 0.96

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

Table 7 Settings of control/dependent variables using different opti- Load buses

mization techniques for IEEE 30-bus system
Fig. 12 Load bus voltages using GA for IEEE 30-bus system
Control/dependent LP GA PSO ACO
variables
VLmax.
Vg 1 1.000 1.003 1.0945 1.0120 1.04
Vg 2 1.000 1.010 1.0107 1.0217
1.03
Load bus voltage (pu.)

Vg 5 1.032 1.050 1.0183 1.0137

1.02
Vg 8 1.000 1.000 1.0469 1.0148
1.01
Vg 11 1.050 1.048 1.0460 1.0430
Vsp.
Vg 13 1.050 1.050 1.0500 1.0380
0.99
Q sw 10 0.114 0.112 0.1323 0.1023
0.98
Q sw 17 0.150 0.145 0.1476 0.0886
0.97
Q sw 24 0.100 0.100 0.0646 0.0822
Qg 1 −0.048 −0.006 −0.097 −0.230 0.96

Qg 2 −0.253 −0.382 −0.250 −0.187 VLmin.

3 4 6 7 9 10 12 14 15 1617 18 19 20 21 22 2324 25 26 27 28 29 30
Qg 5 0.054 0.591 0.578 0.511 Load buses
Qg 8 0.006 0.005 0.503 0.185
Q g 11 0.339 0.314 0.325 0.236 Fig. 13 Load bus voltages using PSO technique for IEEE 30-bus sys-
tem
Q g 13 0.370 0.367 0.352 0.335

power. Then the proposed ACO is an efficient tool to face

test system and here for the IEEE 30-bus test system, the use the emergency conditions the power system face according
of ACO for solving the ORPD increases the power losses to the highest level of reserve power compared to other opti-
and voltage deviation reduction and improves the reserve mization techniques. On the other hand, for the increased

123
Electr Eng (2011) 93:103–116 113

VLmax. Table 9 ORPD results using different optimization techniques for

1.04
WDN system

1.03 LP GA PSO ACO

Load bus voltage (pu.)

1.02 Power losses (pu.) 0.2768 0.2116 0.1873 0.1698

1.01 CPL (EP) 1,937.6 1,481.2 1,311.1 1,188.6
Vsp. (V.D.)rms 0.044 0.034 0.0317 0.024

0.99

0.98 1.06
VLmax.
0.97 1.04

1.02
0.96

Load bus voltage (pu.)

Vsp.
VLmin.
3 4 6 7 9 10 12 1415161718192021222324252627282930 0.98
Load buses
0.96
VLmin.
Fig. 14 Load bus voltages using ACO algorithm for IEEE 30-bus sys- 0.94
tem 0.92

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

6.2.3 West Delta Network system 1.02

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

Control/dependent LP GA PSO ACO 0.9

variables
0.88

Vg 1 1.0744 1.0753 1.0313 1.0385 0.86

Vg 2 1.0159 1.0435 1.0478 0.9692 10 15 20 25 30 35 40 45 50

Vg 3 0.9950 0.9872 0.9986 0.9695 Load buses

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

Incremental Change in Line Losses (pu.)

VLmax. GA
1.04 PSO
0
ACO
1.02
Load bus voltage (pu.)

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

Incremental Change in Line Losses (pu.)

GA
Load bus voltage (pu.)

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
8. Dorigo M (1992) Optimization, learning and natural algorithms.
PhD dissertation, Department of Electronics, Politecnico di Mi-
lano, Milan
9. Dorigo M, Stützle T (2004) Ant colony optimization. MIT Press,
New York
10. Dorigo M, Gambardella LM (1997) Ant colonies for the traveling
salesman problem. Biosystems 43:73–81
11. Yu I-K, Chou CS, Song YH (1998) Application of the ant colony
search algorithm to short-term generation scheduling problem of
thermal units. In: Proceedings of international conference on power
system technology, vol 1, POWERCON ’98, Aug 1998, Beijing,
China, pp 552–556
12. Song YH, Chou CSV, Min Y (1999) Large-scale economic dis-
patch by artificial ant colony search algorithms. Electr Power Com-
pon Syst 27:679–690
13. Hou YH, Wu YW, Lu LJ, Xiong XY (2002) Generalized ant col-
ony optimization for economic dispatch of power systems. IEEE
Int Conf Power Syst Technol 1:225–229
14. EL-Sharkh MY, Sisworahardjo NS, Rahman A, Alam MS (2006)
An improved ant colony search algorithm for unit commitment
application. In: IEEE power systems conference and exposition,
pp 1741–1746
15. Vlachogiannis JG, Hatziargyriou ND, Lee KY (2005) Ant colony
system-based algorithm for constrained load flow problem. IEEE
Trans Power Syst 20:1241–1249
16. Meziane R, Massim Y, Zeblah A, Ghoraf A, Rahli R (2005) Reli-
ability optimization using ant colony algorithm under performance
and cost constraints. Electr Power Syst Res 76:1–8
123
Electr Eng (2011) 93:103–116
17. Favuzza S, Graditi G, Sanseverino ER (2006) Adaptive and
dynamic ant colony search algorithm for optimal distribution sys-
tems reinforcement strategy. Appl Intell 24:31–42
18. Carpaneto E, Chicco G (2008) Distribution system minimum loss
reconfiguration in the hyper-cube ant colony optimization frame-
work. Electr Power Syst Res 78:2037–2045
19. Niknam T (2008) A new approach based on ant colony optimiza-
tion for daily Volt/Var control in distribution networks considering
distributed generators. Energy Convers Manage 49:3417–3424
20. Lenin K, Mohan MR (2006) Ant colony search algorithm for opti-
mal reactive power optimization. Serbian J Electr Eng 3:77–88
21. Ketabi A, babaee AA, Feuillet R (2010) Application of the ant col-
ony search algorithm to reactive power pricing in an open electricity
market. Electr Power Energy Syst 32:622–628
22. Pechač P (2002) Electromagnetic wave propagation modeling
using the ant colony optimization algorithm. Radio Eng 11:1–6
23. Lee KY, Vlachogiannis JG (2005) Optimization of power systems
based on ant colony system algorithms: an overview. In: 13th inter-
national conference on intelligent systems application to power
systems, 6–10 Nov 2005, pp 22–35
24. Washington University. http://www.ee.washington.edu/research/
pstca/
25. El-Sehiemy RA (2008) Performance of transmission network
under deregulated electrical power system. PhD thesis, Faculty of
Engineering, Minoufiya University, Egypt
26. Abou El-Ela AA, Allam SM, Shatla MM (2010) Maximal optimal
benefits of distributed generation using genetic algorithms. Electr
Power Syst Res 80:869–877