www.ietdl.

org

Published in IET Generation, Transmission & Distribution

Received on 24th June 2007
Revised on 31st January 2008
doi: 10.1049/iet-gtd:20070515

ISSN 1751-8687

Distribution system reliability enhancement

using optimal capacitor placement
A.H. Etemadi M. Fotuhi-Firuzabad
Department of Electrical Engineering, Sharif University of Technology, Tehran, Iran
E-mail: [email protected]

Abstract: Failure statistics of most utilities indicate that distribution systems make the greatest individual
contribution to the unavailability of supply to customers. Optimal capacitor placement in distribution
systems has a number of advantages such as reducing losses, improving voltage profile, improving power
factor and so on. The conventional objective function of the optimal capacitor placement consists of the
total cost of losses and investments. Since capacitors supply reactive loads locally, they improve the
load-carrying capability of the lines and therefore play the same role as redundant lines. Thus, optimal
capacitor placement can also improve the reliability indices of a distribution system. Therefore two new
objective functions are defined here. The first one is defined as the sum of reliability cost and
investment cost. The second is defined by adding the reliability cost, cost of losses and investment cost.
The latter is a comprehensive objective function which is the resultant of the conventional and
reliability-based objective functions. This problem is solved using a particle swarm optimisation-based
algorithm. The effectiveness and applicability of the proposed approach are examined using a distribution
network.

1 Introduction system. Some of the reliability improvement strategies are

as follows [2]: adding devices for protection, reclosing and
The main function of a power system is to feed loads with switching, automation, faster crew response to speed up the
electrical energy as economically as possible and with a repair process, fewer equipment failures to prevent
reasonable level of reliability and quality. Distribution contingencies and system reconfiguration. Shunt capacitor
system is one of the main three parts of a power system, placement in power systems, particularly in the distribution
responsible for transferring electrical energy to the end part, is one of the most used schemes for reinforcement.
users. A distribution system is relatively cheap and outages Capacitor placement in distribution system has several
in this part of the system have a very localised effect benefits such as reducing lagging component of the circuit
compared with generation and transmission parts. current, improving voltage regulation, reducing active and
However, analysis of the customer failure statistics of most reactive power loss because of reduction in current,
utilities indicates that the distribution system makes the increasing power factor of the system and so on, one or
greatest individual contribution to the unavailability of more of which may be the reason for this application.
supply to a customer [1].
Optimal capacitor placement needs to determine the
Therefore these statistics reinforce the need to be capacitor sizes and locations considering some of the above
concerned with the reliability of the distribution system, to aims. The objective function is usually formulated as the
evaluate quantitatively the merits of various reinforcement total cost of losses and investments. This is a hard, large-
schemes available to the planner and to ensure that the scale combinatorial problem in which the number of local
limited capital resources are used to achieve the greatest minimum solution points and the number of options to
possible incremental reliability and improvement in the be analysed increases exponentially with the size of the

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 621
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008
www.ietdl.org

distribution system. Since capacitor banks are added in placed on distribution feeders in order to reach the
discrete steps, the objective function is non-differentiable optimum level of reliability at minimum investment. The
and the capacitor placement problem is a mixed integer problem of optimally locating and sizing capacitor banks,
nonlinear programme. However, most of the conventional so that the total cost is minimised while technical
optimisation algorithms used so far are unable to generate constraints (e.g. voltage levels and power flow limits) are
optimal solutions for this type of problem. Optimal enforced, is of a combinatorial nature as the consequence of
capacitor placement problem has been the topic of capacitor sizes and locations being discrete variables. PSO
many research works in the past five decades. References is one of the most recent and efficient methods in
[3 – 8] are some of the most recent published papers on this the optimisation literature and can successfully cope with
topic that have solved the problem using ant colony the nonlinear mixed-integer programming nature of the
direction, graph search algorithm, genetic algorithm, optimal capacitor placement problem. In this paper, PSO is
particle swarm optimisation (PSO) and fuzzy evolutionary used to find the optimum point in the reliability cost/
programming. worth curve shown in Fig. 1. Numerical examples are used
and the results are presented to illustrate the applicability
For the first time, Sallam et al. [9] used shunt capacitors in and effectiveness of the proposed method. Comparative
distribution systems to improve the reliability indices. Using studies are conducted in order to compare the existing
the state-space method and Markov process, they techniques with the proposed method.
demonstrated that the reliability of a distribution system
can be enhanced using shunt capacitors due to the
improvement in load-carrying capability of the feeders by
controlling the reactive power flow. Shunt capacitors
2 Problem description
therefore play the same role as redundant lines, viz, both of 2.1 Reliability evaluation
them enhance the system security and reliability.
Most distribution systems are operated as radial networks,
and as such the principles of series systems are applied to
In this paper, a heuristic method is proposed for optimal them [10]. Three basic reliability indices of the system,
capacitor placement problem in distribution systems. average failure rate, lS , average outage time, rS , and annual
However, unlike the conventional research works on this outage time US are given by
topic, the objective function is not only the total cost of
X
losses and investments. Rather, two new objective functions lS ¼ li (1)
are defined incorporating the cost of reliability, cost of i
losses and investments. The first considers only the cost of X
reliability (ECOST) and investment cost. But, the second US ¼ li ri (2)
i
is a comprehensive one which includes ECOST, cost of P
losses and investment cost. The concept of reliability cost/ US lr
rS ¼ ¼ Pi i i (3)
reliability worth is illustrated in Fig. 1. This figure indicates lS i li
that the investment cost of capacitor placement generally
increases when higher reliability level is required. On the where li , ri and liri are, respectively, the average failure rate,
other hand, the customer cost associated with the failures average outage time and annual outage time of the ith
decreases as the reliability increases. The total cost exhibits component.
a minimum, and as such an optimum or target level of
reliability is achieved. The optimal capacitor placement In this paper, expected interruption cost (ECOST) is
solution determines the location and size of capacitors to be calculated for the distribution system. Evaluating ECOST
enables the system planners to determine the acceptable
level of reliability for customers, provide economic
justifications for determining network reinforcement and
redundancy allocation, identify weak points in a system,
determine appropriate maintenance scheduling and develop
appropriate operation policies. ECOST is therefore a
powerful tool for system planning. ECOST at bus i is
calculated as follows [11]

ECOSTi ¼ La(i) Ci li (4)

where La(i) is the average load connected to load point i in

kW and Ci is the cost of interruption (in $/kW) for the ith
bus which is evaluated using composite customer damage
Figure 1 Capacitor placement cost/worth curves function (CCDF). CCDF shows the cost of interruption as

622 IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621– 631
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-gtd:20070515
 www.ietdl.org

a function of interruption duration. A typical CCDF [11] is Branch current has two components: active (IA) and
illustrated in Fig. 2. reactive (IR). Capacitor placement can supply part of the
reactive power demands. Consequently, the reactive
The total ECOST of the distribution feeder is calculated component is reduced and due to the reduction of the
as follows magnitude of current, the resistive losses decrease. As a
result, destructive effects of temperature on the reliability of
X
NB X
NB
overhead lines and underground cables are moderated.
ECOST ¼ ECOSTi ¼ La(i) Ci li (5) Capacitor placement also improves the reliability in another
i¼1 i¼1
way. It increases the loadability of feeders and therefore
where NB is the number of load points in the feeder. plays the similar role as redundant lines and as such can
increase the reliability of the system.
Another common index is energy not served (ENS) which
can be calculated for each load bus i using the following The positive impact of capacitor placement on reliability
equation can be considered as a failure rate reduction of distribution
feeder components [9]. Before capacitor placement, any
uncomp
ENSi ¼ La(i) Ui (6) feeder i has an uncompensated failure rate of li . If the
reactive component of a feeder branch is fully compensated,
comp
its failure rate reduces to li . If the reactive component
2.2 Reliability improvement via capacitor of current is not completely compensated, a failure rate is
placement defined with linear relationship to the percentage of
Feeders in distribution systems deliver power from compensation. Thus, the compensation coefficient of the
distribution substations to distribution transformers. A ith branch is defined as
considerable portion of customer interruptions are caused
i
by equipment failures in distribution systems consisting of IR-old
ai ¼ i (7)
underground cables and overhead lines. Resistive losses IR-new
raise the temperature of feeders which is proportional to
the square of the current magnitude flowing through the i
where IR
new i
and IR
old are the reactive components of the
feeder. For underground cables, there is a maximum ith branch current before and after compensation,
operating temperature which if exceeded would cause the respectively. The new failure rate of the ith branch is
insulation problem and an increase in component failure computed as follows
rates. The life expectancy of the insulation material
decreases exponentially as the operating temperature uncomp comp comp
li-new ¼ ai (li
li ) þ li (8)
increases [12]. Also moisture absorption occurs more
rapidly at high temperatures. When moisture invades
extruded dielectrics such as cross-linked polyethylene or In this paper, three objective functions are considered. The
ethylene – propylene rubber, the voltage withstand capability first one is the conventional objective function which
of the cable is reduced and the probability of dielectric consists of cost of losses and cost of capacitor investment.
breakdown increases, and consequently, the failure rate of The two other newly defined objective functions consider
the cable is increased. Temperature also has impacts on the cost of reliability. The second objective function considers
reliability of overhead lines. High currents will cause lines only the cost of reliability and investment. The third one is
to sag, reducing the ground clearance and increasing the a comprehensive objective function which consists of three
probability of occurring an electric break [2]. terms: cost of reliability (ECOST), cost of losses and
capacitor investment cost.

1. Conventional objective function: The conventional objective

function is expressed as follows

min COST ¼ KP LP þ KE LE þ CC (9)

where COST is the total cost of the system ($/year), LP the

peak power losses (kW), LE, the energy losses (kW h), CC
cost of the capacitor ($), KP the factor to convert peak
power losses to dollar ($/kW), and KE the factor to
convert energy losses to dollar ($/kWh).

2. Reliability-based objective function: Using (4) and (5), the

second objective function associated with optimal capacitor
Figure 2 Typical CCDF placement is defined as the sum of the total ECOST of the

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 623
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008
www.ietdl.org

system and capacitor investment which is given by For the first backward sweep, the voltage of each bus is
assumed to be 1 pu and the branch currents are computed
minimise TCOST1 ¼ ECOST þ CC (10) using these voltages in (12) and the Kirchhoff’s current
law. When the current flowing in all branches are
where TCOST1 (in $/year) is the total cost of the system and computed, the bus voltages are updated using Kirchhoff\’s
CC is the total cost of the capacitors. It is assumed that the voltage law in the forward sweep. The voltage at each bus
capacitor placement cannot affect the voltage profile is calculated with voltage drop calculation, starting at the
negatively. Also it is assumed that the capacitors only source bus and traversing out to the end buses using the
change the magnitude of the reactive component of the currents calculated in the previous backward sweep. For
feeder current and real power flow will remain constant. All example, voltage at Bus 2 is calculated as follows
loads are assumed to be three-phase loads. Capacitor banks
are available in discrete sizes and only fixed capacitors are used. V2 ¼ V1
Z1 I1 (13)

3. Comprehensive objective function: The third objective where Z1 is the impedance of Branch 1. The backward sweep
function considers the cost of reliability (total ECOST), will be repeated with the updated bus voltages and the
cost of active losses and cost of investment. Therefore by process is continued iteratively until the algorithm is
solving this problem, both conventional aims and reliability terminated based on the convergence criterion as shown
improvement purposes are obtained and the comprehensive below.
optimal solution is determined. The comprehensive
objective function is given by max jVik
Vikþ1 j , 1 i ¼ 1, . . . , NB (14)
minimise TCOST2 ¼ ECOST þ KP LP þ KE LE
where Vik and Vikþ1 are the voltage of bus i in two successive
þ CC (11) iterations. NB is the number of buses and 1 is the specified
tolerance for voltage mismatch.

2.3 Distribution power flow 3 Solution technique

In capacitor placement problem, distribution power flow is 3.1 Particle swarm
used as a subroutine in each iteration and plays an important
PSO is a robust stochastic evolutionary computation
role in the solution process. Thus, the power flow algorithm
technique based on the movement and intelligence of
should be computationally efficient and numerically robust.
swarms [17]. PSO has been shown to be effective in
Forward– backward sweep or ladder iteration method is used
optimising difficult multidimensional discontinuous
in this paper which is a simple, easy to implement and
problems in a variety of fields [18].
reliable method for radial feeders [13]. In addition to the
distribution load flow methods based on forward– backward
In the literature of PSO, there are some important
sweep, another group of methods inspired from transmission
keywords that are described below [19]:
system power flow calculations, with modifications for
overcoming the difficulty of high R/X ratio of distribution
Particle refers to each individual in the swarm. All particles
systems, have been proposed such as fast decoupled power
should accelerate their movements towards the personal and
flow [14, 15] and the implicit Zbus Gauss method [16].
overall best locations. Position is the place of a particle in a
N-dimensional space of solutions. Fitness is a function that
Consider the simple distribution feeder shown in Fig. 3.
takes the position of a particle and returns a scalar number
The load current for each node is computed by
that shows the goodness of that position and is the


 objective function; like other evolutionary algorithms. pbest
Sn Pn þ jQn 
In ¼ ¼ (12) is the best position that a particle encountered in the search
Vn Vn
process and had the best fitness value. gbest is the global
best position of all particles in the swarm which has the
highest fitness and each particle knows it. Each particle
should change its position with a specified velocity towards
a better position.

The PSO algorithm can be described as follows.

1. Define the fitness function and solution space: Fitness

function is the interface between the physical world and
optimisation method. The fitness function should be
Figure 3 Sample distribution feeder defined to show the goodness of a solution by a single

624 IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621– 631
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-gtd:20070515
 www.ietdl.org

number. The parameters to be optimised are limited in a size have lower unit prices. The available capacitor size is
predefined reasonable range that specifies the solution space. usually limited to

2. Initialise swarm location and position randomly: Each QC C

max ¼ LQ0 (18)
particle should start with an initial location and velocity
(magnitude and direction) at the beginning of the search where L is an integer. Therefore for each installation location,
process. These initial values are generated randomly and there are L capacitor sizes available {QC C C
0 , 2Q0 , . . . , LQ0 }.
the initial location of each particle is its pbest and the best Thus the parameters to be optimised are in the range of 0
initial location of the swarm is the initial gbest. to L. The solution procedure is described below.

3. Flying towards the optimum: Each particle compares its 1. Collect the distribution system data such as load of buses,
current location with its pbest or gbest using the fitness line data, compensated and uncompensated failure rates, base
function, and if its current fitness is worse, flies to an voltage and so on.
appropriate location. The velocity of the movement
depends on pbest and gbest and is calculated using the 2. Determine the parameters of PSO such as size of particle,
following equation weighting factors, weighting function, maximum number of
iterations and other necessary parameters. The solution is
vkþ1
i ¼ vvki þ c1  rand1  (pbest
xki ) þ c2  rand2 the vector of positions which corresponds to the number of
 (gbest
xki ) (15) capacitor banks [in the range of 0 to L in (18)] installed on
each bus.
where vki is the velocity of particle i at iteration k, v the
weighting function, c1 and c2 the weighting factors which 3. Initialise the algorithm, that is initial particle positions
determine the convergence behaviour of PSO, c1 a factor and velocities are randomly chosen.
determining how much the particle is influenced by the
memory of its best location, c2 a factor determining how 4. Evaluate the values of pbesti related to each particle and
much the particle is influenced by the rest of the swarm gbest using the objective function shown by (10) or (11).
and both are typically set equal to 2, rand1 and rand2 the In order to evaluate the objective function, the following
random numbers between 0 and 1, xki the current position steps must be followed.
of particle i at iteration k, pbesti the pbest of particle i and
gbest the best value so far in the group among the pbest of † Run distribution power flow using (12) and (13) and
all particles. The following weighting function is usually backward/forward sweep and determine the active loss of
used: the feeder (LP).
vmax
vmin † Using the results of power flow and (7) and (8), compute
v¼  iter (16)
itermax new failure rates of branches.
where vmax is the initial weight, vmin the final weight, itermax † Using (1) – (3), calculate the reliability indices of each bus.
the maximum iteration number and iter the current iteration
number. Once the velocity is determined, particles move † Using (4) and the CCDF curve (Fig. 2), calculate ECOST
according to the following equation [14] of each bus.
xkþ1
i ¼ xki þ Dt  vkþ1
i (17)
† Calculate total ECOST using (5).
where Dt is the time step usually chosen to be 1.
† Using the obtained values of total ECOST, LP and the
4. Repeat: The previous step is repeated for each time step capacitors cost, the two new objective functions, (10) and
and the pbest related to each particle and the gbest are (11) can be calculated.
updated and the process is finished when the termination
criterion is met. The maximum number of iterations is one 5. For the next iteration, the values of new velocities and
of the most common criteria for the algorithm to stop. positions are calculated using (15), (16) and the round
value of the right-hand side of (17). Evaluate the objective
function for these new values and if the fitness function is
3.2 Proposed method improved, gbest and pbesti which is related to the particle
The objective function is shown by (10). The number of whose fitness value has improved, are updated.
parameters to be optimised is equal to the number of load
buses. Considering the real-world capacitors, there exists a 6. The aforementioned loop is continued.
finite number of standard sizes which are integer multiples
of the smallest size QC
0 . Besides, the cost per kVAR varies 7. The algorithm is terminated if the iteration number is
from one size to another. In general, capacitors of larger reached to the prespecified maximum number of iterations

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 625
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008
www.ietdl.org

or the objective function is not improved after a specified computational time. Moreover, for very large distribution
number of iterations. systems, the following measure can be considered to reduce
the computational time. At first, in order to find the sensitive
For more clarification, the algorithm associated with the buses to capacitor placement, the problem is solved with a
proposed method is illustrated in Fig. 4 in which i refers to low population size and a small value for maximum number
the bus number. of iterations while the solution vector includes all buses of the
system. Sensitive buses are referred to those buses of the
Since the capacitor placement is not an online action for distribution feeder for which placing a constant size of
distribution systems, its computational solution time is not a capacitor results in more reduction in the value of overall
significant issue. However, in order to avoid excessive defined cost function. Then, in the main solution of the
computational times for large feeders, some parameters of the problem, the solution vector is limited to the sensitive buses.
PSO method such as population size and maximum Now, the population size and the maximum iteration
iterations can be justified in order to obtain a trade-off numbers are increased to find a near-optimal solution with a
between an acceptable solution and a reasonable reasonable computational time.

4 Case study
4.1 Nine section feeder example
A radial distribution feeder with nine sections and nine load
points shown in Fig. 5 is considered [4]. The line data and
load data are given in Table 1. It is assumed that the
section with the highest impedance has the biggest failure
rate of 0.5 f/year and the section with the smallest
impedance has the least failure rate of 0.1 f/year. The
failure rates of other sections are calculated linearly
proportional to these two values according to their
impedances. The reason for this is that the failure rate is
proportional to the length of the line and more impedance
corresponds to longer lines.

The available capacitor sizes and their associated cost are

given in Table 2. Larger capacitor banks can be obtained
by installing the combination of appropriate number of
available sizes. The following assumptions are considered to
calculate the reliability indices and optimal capacitor
placement. If the reactive component of a section current is
fully compensated, its failure rate reduces to 85% of its
uncompensated failure rate [8] and for partial
compensation, the failure rate is calculated using (8). The
feeder has a main breaker and there is a sectionaliser at the
beginning of each section. The repair time is 8 h and
switching time is 0.5 h. Since capacitor placement does not
have any impact on the reliability of other components
except feeder sections, all other components such as
transformers, busbars, breakers, disconnects and so on, are
assumed to be fully reliable.

The capacitor placement problem is solved using the

algorithm shown in Fig. 4. The following parameters are

Figure 4 Solution procedure flowchart Figure 5 9 section test feeder

626 IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621– 631
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-gtd:20070515
 www.ietdl.org

Table 1 Nine-section feeder line and load data

Bus no. Load Feeder section R þ jX, V Failure rate, (f/year)

kW kVAR
1 1840 460 0–1 0.123 þ j0.413 0.106
2 980 340 1–2 0.014 þ j0.605 0.100
3 1790 446 2–3 0.746 þ j1.205 0.153
4 1598 1840 3–4 0.698 þ j0.608 0.150
5 1610 600 4–5 1.983 þ j1.728 0.247
6 780 110 5–6 0.905 þ j0.798 0.165
7 1150 60 6–7 2.055 þ j1.164 0.252
8 980 130 7–8 4.795 þ j2.716 0.459
9 1640 200 8–9 5.343 þ j3.026 0.500

set for PSO: weighting factors, c1 and c2 are both set to 2, terminated. As noted earlier, among many advantages,
population size is set to 90, the maximum number of capacitor placement also improves the voltage profile of the
iterations is set to 2000 and further, the algorithm is feeder as shown in Fig. 7. The least voltage of the feeder
terminated if the objective function is not improved after has been improved from 0.83 to 0.88 pu.
250 iterations, inertia weights vmax and vmin are set to 0.9
and 0.4, respectively. 4.3 Comprehensive objective function
The proposed algorithm is now carried out using the
4.2 Reliability-based objective function comprehensive objective function (11). The location vector
of capacitor banks to be installed is [1050, 0, 0, 2700, 0,
In this section, the proposed algorithm is carried out using 600, 450, 0, 450] kVAR on load points 1– 9, respectively.
the reliability-based objective function (10). The algorithm The cost of customers interruptions is evaluated using the
is terminated at the 751th iteration because the gbest has CCDF function shown in Fig. 2. ECOST of the
not changed for at least 250 iterations. Elapsed time was distribution feeder before capacitor placement is 466 120 $/
552 s. The location vector of capacitor banks to be installed year and after capacitor placement it reduces to 407 460 $/
is [0, 0, 3000, 0, 0, 450, 0, 450, 450] kVAR on load points year and the total cost of capacitors is 12 760 $. Thus, the
1 – 9, respectively. The cost of customer interruptions is overall savings is 45 900 $/year. The new failure rates of
evaluated using the CCDF function shown in Fig. 2. the feeder sections after capacitor placement are 0.0934,
ECOST of the distribution feeder before capacitor 0.0863, 0.1302, 0.1305, 0.2157, 0.1517, 0.2162, 0.4143
placement is 466 120 $/year and after capacitor placement, and 0.4299 f/year for feeder sections, respectively.
it reduces to 410 860$/year and the total cost of capacitors
is 8 820 $. Thus, the overall savings is 46 440 $/year. The 4.4 Comparison of different solutions
new failure rates of the feeder sections after capacitor
placement are 0.0923, 0.0858, 0.1306, 0.1416, 0.2184, The nine-section test feeder has been investigated in a
0.1485, 0.2168, 0.4147 and 0.4297 f/year for sections 1 – 9, number of research works using the conventional objective
respectively. Complete reliability calculation of the function (9). Different case studies are conducted to
distribution feeder for the optimal solution is given in determine the possible capacitor arrangements when various
Table 3. objective functions are used. The results are used to
compare the existing techniques with the proposed ones as
The best values of the objective function in each iteration shown in Table 4. They include the location and sizes of
(gbest) are shown in Fig. 6. It can be seen from this figure capacitors, total active power loss, the values of
that the values of gbest are constant for 250 iterations at conventional and new objective functions and the least
the end of the procedure and thus the algorithm is voltage magnitude of the feeder which is the magnitude of
RMS voltage at Bus 9. It can be seen that different
Table 2 Available capacitor sizes and costs capacitor arrangements are obtained with different objective
functions.
Size, kVAR 150 300 450 600 900 1200
The base case is the main feeder before installing any
Cost, ($) 750 975 1140 1320 1650 2040
capacitor banks.

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 627
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008
www.ietdl.org

Table 3 Reliability evaluation of the nine-section feeder for the objective function (10)
Feeder Failure Bus no.
section ratea
1 2 3 4 5 6 7 8 9
l r u r u r u r u r u r u r u r u r u
1 0.092 8 0.74 8 0.74 8 0.74 8 0.74 8 0.74 8 0.74 8 0.74 8 0.74 8 0.74
2 0.086 0.5 0.04 8 0.69 8 0.69 8 0.69 8 0.69 8 0.69 8 0.69 8 0.69 8 0.69
3 0.131 0.5 0.07 0.5 0.07 8 1.04 8 1.04 8 1.04 8 1.04 8 1.04 8 1.04 8 1.04
4 0.142 0.5 0.07 0.5 0.07 0.5 0.07 8 1.13 8 1.13 8 1.13 8 1.13 8 1.13 8 1.13
5 0.218 0.5 0.11 0.5 0.11 0.5 0.11 0.5 0.11 8 1.70 8 1.70 8 1.70 8 1.70 8 1.75
6 0.149 0.5 0.07 0.5 0.07 0.5 0.07 0.5 0.07 0.5 0.07 8 1.19 8 1.19 8 1.19 8 1.19
7 0.217 0.5 0.11 0.5 0.11 0.5 0.11 0.5 0.11 0.5 0.11 0.5 0.11 8 1.73 8 1.73 8 1.73
8 0.415 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 8 3.32 8 3.32
9 0.430 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 0.5 0.21 8 3.44
total 1.878 0.9 1.63 1.2 2.28 1.7 3.25 2.3 4.32 3.2 5.95 3.8 7.07 4.6 8.69 6.3 11.80 8 15.03
subtotal ECOST 13 350 10 540 28 780 34 910 49 510 28 710 52 490 61 290 13 1290
subtotal ENS 3002 2230 5826 6898 9587 5513 9999 11569 24645
total ECOST ¼ 410 860 $/year
total ENS ¼ 79269 kWh/yr
a
The failure rate of each bus is the sum of the preceding branches failure rates which is not included because of the lack of
sufficient space.

Cases 1 – 4 are the solutions of the conventional optimal The last case (case 6) is the solution of the problem using
capacitor placement problem found in References [20 – 23] the proposed method with comprehensive objective function
respectively with KP ¼ 168 $/kW. The methods used for (11). This objective function comprehensively considers
solving the problem are analytical [17], fuzzy-reasoning the reliability cost, cost of losses and capacitor investment
approach [18], heuristic combinátorial optimisation [19] cost.
and differential evolution method [20].
It can be seen from the results that case 6 (comprehensive
Case 5 is the solution of the problem using the objective function) results in a better solution compared with
proposed reliability-based objective function algorithm that of case 5 (reliability-based objective function). However,
(10), in which the cost of reliability and investment are comparing the results associated with cases 1 – 5, case 5
considered. (reliability-based objective function) results in the best

Figure 6 Values of gbest in each iteration for the objective

function (10) Figure 7 Voltage improvement after capacitor placement

628 IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621– 631
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-gtd:20070515
 www.ietdl.org

Table 4 Results of various capacitor arrangement on the nine-section feeder

compensated capacitor sizes, PLoss , investment ECOST, Objective functions Least
buses kVAR kW cost, $ $ voltage, pu
(9)($) (10)($) (11)($)
Base — — 783.78 — 466 120 131 670 466 120 597 800 0.8375
case
Case 1 4, 7, 9 4050, 900, 900 704.71 10 560 428 040 128 950 438 600 556 990 0.9001
Case 2 3, 4, 5, 9 3300, 2100, 689.11 13 920 424 180 129 690 438 100 553 870 0.9001
1650, 600
Case 3 4, 5, 9 2700, 2850, 900 719.45 11 920 432 790 132 790 444 710 565 580 0.9070
Case 4 4, 5, 9 4050, 1650, 750 698.14 12 510 423 080 129 800 435 590 552 880 0.9000
Case 5 3, 6, 8, 9 3000, 450, 450, 695.87 8 820 410 860 125 720 419 680 536 580 0.8798
450
Case 6 1, 4, 6, 7, 9 1050, 2700,600, 683.31 12 760 407 460 127 560 420 220 535 010 0.8830
450, 450

solution. In all cases, similar to the aforementioned reliability cost/reliability worth curve. It can be seen from
references, the value of KV is assumed to be 0 which means Fig. 8 that the total ECOST reduces as the capacitor banks
that the cost of energy loss is not considered. As the and as such their cost are increased. This will continue up
dominant term in the two new objective functions is the to the optimal point. Beyond the optimum, two capacitor
ECOST (about four times the cost of loss), the values banks are randomly added to Bus 1 of the system with
associated with reliability-based and comprehensive sizes 600 and 750 kVAR. It is clear that this extra capacitor
objective functions (cases 5 and 6) are not very different. If installation reduces the ECOST of the system but it
the cost of energy loss is considered, the solutions and the deviates the solution from the optimal point. In this
values of objective functions (10) and (11) would be more section, objective function (10) was considered.
different.
4.6 Sensitivity analysis on PSO
4.5 Reliability cost/worth curve parameters (c1 and c2)
In the previous section, it was shown that adding capacitor As noted earlier, PSO algorithm has some parameters
banks improves the reliability of the feeder. As expected, including weighting factors c1 and c2 . These parameters are
the cost of capacitors increases as the capacitor banks are typically set to 2 because, according to the literature,
increased. Thus, there is an optimal capacitor placement extensive investigations have shown that this selection leads
solution for which the total cost of the system is to the best answer. In this section, a sensitivity analysis has
minimised. Fig. 8 shows this optimum point and the been carried out in order to determine the impacts of these
parameters on the final solution of objective function (10).

Figure 8 Reliability cost/worth curve

A, ECOST; þ, capacitor cost, w, total cost; w, optimal point Figure 9 Sensitivity analysis of c1 and c2 parameters of PSO

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 629
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008
www.ietdl.org

The two parameters are chosen from this area: (c1 , c2) [ [1.5, [5] GALLEGO R.A., MONTICELLI A., ROMERO R.: ‘Optimal capacitor
2.5]  [1.5, 2.5] in definite steps and the corresponding total placement in radial distribution networks’, IEEE Trans.
cost of the system for the optimal solution is determined. The Power Syst., 2001, 16, (4), pp. 630 – 637
results are illustrated in Fig. 9. As expected, the best results
occurred near the point for which c1 ¼ c2 ¼ 2. However, [6] DE SOUZA B.A. , ALVES H. , FERREIRA H.A. :
since the procedure is heuristic and not analytic and ‘Microgenetic algorithms and fuzzy logic applied to the
random to some extent, the results do not obey this rule optimal placement of capacitor banks in distribution
entirely. networks’, IEEE Trans. Power Syst., 2004, 19, (2),
pp. 942– 947

5 Conclusions [7] ESMIN A.A., LAMBERT-TORRES G., ZAMBRONI DE SOUZA A.C.: ‘A

hybrid particle swarm optimization applied to loss power
In addition to the benefits such as resistive loss reduction, minimization’, IEEE Trans. Power Syst., 2005, 20, (2),
voltage profile improvement and releasing feeders capacity, pp. 859– 866
capacitor placement in distribution systems can also
improve the reliability indices by failure rate reduction. [8] VENKATESH B., RANJAN R.: ‘Fuzzy EP algorithm and dynamic
Therefore unlike the conventional objective function of data structure for optimal capacitor allocation in radial
optimal capacitor placement problem which includes the distribution systems’, IEE Proc., Gener. Transm. Distrib.,
cost of resistive losses, two new objective functions are 2006, 153, (1), pp. 80– 88
proposed in this paper. The first one is given by adding the
ECOST of the system and the cost of investment. The [9] SALLAM A.A., DESOUKY M., DESOUK H.: ‘Shunt capacitor effect
second one is a comprehensive objective function which on electrical distribution system reliability’, IEEE Trans.
considers ECOST, cost of power losses and cost of Reliab., 1994, 43, (1), pp. 170– 176
investment. A PSO algorithm was proposed in this paper
for optimal capacitor allocation in distribution feeders in [10] BILLINTON R., ALLAN R.N. : ‘Reliability evaluation of
order to find the solutions associated with the two new engineering systems’ (Plenum, New York, 1992)
proposed objective functions. A nine-section feeder was
studied and the optimal capacitor placement was applied to [11] GOEL L., BILLINTON R.: ‘Evaluation of interrupted energy
it. The results presented indicate that considerable savings assessment rates in distribution systems’, IEEE Trans.
in annual cost of the system was observed due to an Power Deliv., 1991, 6, (4), pp. 1876 – 1882
increase in the reliability. It was shown that capacitor
placement also improves the voltage profile of the system. [12] LEWIN P.L., THEED J.E., DAVIES A.E., LARSEN S.T.: ‘Method for
The reliability cost/reliability worth curve of the system was rating power cables buried in surface troughs’, IEE Proc.,
illustrated and the optimal point was determined. The Gener. Transm. Distrib., 1999, 146, (4), pp. 360– 364
results presented also indicate that by installing additional
capacitors, the ECOST reduces but it pushes the solution [13] KERSTING W.H.: ‘Distribution system modeling and
away from the optimum. A sensitivity analysis for analysis’ (CRC Press LLC, New York, 2002)
weighting factors of PSO was also carried out. The results
showed that the best solutions are found when both c1 and [14] RAJIČIĆ D., BOSE A.: ‘A modification to the fast decoupled
c2 are near 2. power flow for networks with high R/X ratios’, IEEE Trans.
Power Syst., 1988, 3, (2), pp. 743 – 746

6 References [15] ZIMMERMAN R.D., CHIANG H.: ‘Fast decoupled power flow
for unbalanced radial distribution systems’, IEEE Trans.
[1] BILLINTON R., ALLAN R.N.: ‘Reliability evaluation of power Power Syst., 1995, 10, (4), pp. 2045 – 2052
systems’ (Plenum, New York, 1996, 2nd edn.)
[16] CHEN T.H., CHEN M.S., HWANG K.J., KOTAS P. , CHELBI E.:
[2] BROWN R.E. : ‘Electric power distribution reliability’ ‘Distribution system power flow analysis – a rigid
(Marcel Dekker Inc., New York, Basel, 2002) approach’, IEEE Trans. Power Deliv., 1991, 6, (3),
pp. 1146 – 1152
[3] CARPINELLI G., VARILONE P., DI VITO V., ABUR A. : ‘Capacitor
placement in three-phase distribution systems with [17] KENNEDY J.: ‘The particle swarm: social adaptation of
nonlinear and unbalanced loads’, IEE Proc., Gener. Transm. knowledge’. Proc. IEEE Int. Conf. Evol. Comput.,
Distrib., 2005, 152, (1), pp. 47– 52 Indianapolis, IN, 1997, pp. 303 – 308

[4] CHIOU J., CHANG C., SU C.: ‘Ant direction hybrid differential [18] EBERHART R.C., SHI Y.: ‘Evolving artificial neural networks’.
evolution for solving large capacitor placement problems’, Proc. 1998 Int. Conf. Neural Networks and Brain, Beijing,
IEEE Trans. Power Syst., 2004, 19, (4), pp. 1794 – 1800 People’s Republic of China, 1998

630 IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621– 631
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-gtd:20070515
 www.ietdl.org

[19] ROBINSON J., RAHMAT-SAMII Y.: ‘Particle swarm optimization systems’. Proc. IEEE Int. Conf. Industrial Technology, 1996,
in electromagnetics’, IEEE Trans. Antennas Propag., 2004, pp. 237– 241
52, (2), pp. 397– 407
[22] BAGHZOUZ Y., ERTEM S.: ‘Shunt capacitor sizing for radial
[20] ABDEL SALAM T.S., CHIKHANI A.Y., HACKAM R.: ‘A new technique distribution feeders with distorted substation voltages’,
for loss reduction using compensating capacitors applied to IEEE Trans. Power Deliv., 1990, 5, pp. 650– 657
distribution systems with varying load condition’, IEEE
Trans. Power Deliv., 1994, 9, (2), pp. 819 – 827 [23] SU C., LEE C.: ‘Modified differential evolution method for
capacitor placement of distribution systems’. Transmission
[21] SU C., TSAI C.: ‘A new fuzzy-reasoning approach to and Distribution Conf. Exhibition: Asia Pacific. IEEE/PES,
optimum capacitor allocation for primary distribution 6 – 10 October 2002, vol. 1, pp. 208 – 213

IET Gener. Transm. Distrib., 2008, Vol. 2, No. 5, pp. 621 – 631 631
doi: 10.1049/iet-gtd:20070515 & The Institution of Engineering and Technology 2008