Optimal Capacitor Placement on Interconnected

Distribution Systems in Presence of Nonlinear

Loads Using Genetic Algorithms
Ahmad Galal Sayed (1), Hosam K.M. Youssef (2), Member, IEEE
1
Engineering Consultants Group, Cairo, Egypt
2
Electric power and machine Dept., Faculty of Engineering, Cairo University, Egypt

Abstract- Due to simplicity of analysis of radial distribution radially (with one open point in each loop) or with closed
systems, all previous work studied the effect of nonlinear loads on loops, the basic equipment capacity requirements of the loop
the optimal solution of the Capacitor Placement Problem (CPP) feeder design do not change. In terms of complexity, a loop
on only radial distribution systems. The study of the optimal feeder system is only slightly more complicated than a radial
capacitor placement on interconnected distribution systems in
system. Power usually flows out from both sides toward the
presence of nonlinear loads using Genetic Algorithms (GA) is
presented in this paper. Results (power losses, operating middle, and in all cases can take only one of two routes.
conditions and annual benefits) are compared with that obtained Voltage drop, sizing, and protection engineering are only
from radial and loop distribution systems. Computational results slightly more complicated than for radial systems.
obtained showed that the harmonic component affects the optimal Interconnected distribution systems are the most complicated
capacitor placement in all system configurations. When all loads and costly but most reliable method of distributing electric
were assumed to be linear, interconnected and Loop system power. An interconnected distribution system involves multiple
configurations offer lowest power losses and best operating paths between all points in the network. Interconnected
conditions rather than the radial system configuration while systems provide continuity of service (reliability) far beyond
radial system configuration offer best annual benefits due to
that of radial and loop designs: if a failure occurs in one line,
capacitor placement. In distorted networks, interconnected
systems configuration offer lower power losses, best operating power instantly and automatically re-routes itself through other
conditions and best annual benefits due to capacitor placement. pathways.
Interconnected distribution systems are more expensive than
radial distribution systems, but not greatly so in dense urban
I. INTRODUCTION
applications, where the load density is very high, where the
Electric power is supplied to final users by means of distribution must be placed underground, and where repairs
Medium Voltage (MV) or Low Voltage (LV) distribution and maintenance are difficult because of traffic and congestion,
systems, their structures and schemes can differ significantly interconnected systems may cost little more than loop systems.
according to loads location. Overhead lines with short Interconnected systems require little more conductor capacity
interconnection capabilities are mostly employed in rural areas, than a loop system. The loop configuration required "double
whilst cables with a great number of lateral connections for capacity" everywhere to provide increased reliability.
alternative supplies are widespread used in urban areas. Interconnected systems is generally no worse and often needs
Most power distribution systems are designed to be radial, to considerably less capacity and cost, if it is built to a clever
have only one path between each customer and the substation. design and its required capacity margins are minimized.
The power flows exclusively away from the substation and out The solution procedures of the Capacitor Placement Problem
to the customer along a single path, which, if interrupted, (CPP) start with performing a load flow analysis to analyze the
results in complete loss of power to the customer. Its steady-state performance of the power system prior to capacitor
predominance is due to two overwhelming advantages: it is placement and after capacitor placement and to study the
much less costly than the other two alternatives (loop and effects of changes in capacitor sizes and locations.
interconnected systems) and it is much simpler in planning, The biggest advantages of the radial system configuration, in
design, and operation. addition to its lower cost, are the simplicity of analysis and
An alternative to purely radial feeder design is a loop predictability of performance. Because there is only one path
system, which has two paths between the power sources between each customer and the substation, the direction of
(substations, service transformers) and each customer. power flow is absolutely certain. Equally important, the load
Equipment is sized and each loop is designed so that service on any element of the system can be determined in the most
can be maintained regardless of where an open point might be straightforward manner by simply adding up all the customer
on the loop. Because of this requirement, whether operated loads "downstream" from that piece of equipment. Because
load and power flow direction are easy to establish in radial the capacitor placement and the second is the cost of the total
distribution system, voltage profiles can be determined with a energy losses.
good degree of accuracy without resorting to exotic calculation The cost associated with capacitor placement is composed of
methods; equipment capacity requirements can be ascertained a fixed installation cost, a purchase cost and operational cost
exactly; capacitors can be sized, located, and set using (maintenance and depreciation). The cost function described in
relatively simple procedures (simple compared to those this way is a step-like function rather than a continuously
required for similar applications to non-radial (loop and differentiable function since capacitors in practice are grouped
interconnected) system designs, in which the power flow in banks of standard discrete capacities with cost not linear
direction is not a given). proportional to the capacitor bank size.
The interconnected distribution systems are much more It should be pointed that since the objective function is non-
complicated than other forms of distribution, and thus much differentiable, all nonlinear optimization techniques become
more difficult to analyze and operate. There is no awkward to apply.
"downstream" side to each unit of equipment. This complicates The second term in the objective function represents the total
load estimation, power flow analysis, and protection planning. cost of energy losses. This term is obtained by summing up the
Due to simplicity of analysis of radial distribution systems, annual real power losses for the system.
all previous work studied the effect of nonlinear loads on D. Operational Constraints
optimal solution of CPP on only radial distribution systems. Voltages along the feeder are required to remain within
The study of the optimal placement and sizing of fixed upper and lower limits after the addition of capacitors on the
capacitor banks placed on distorted interconnected distribution feeder. Voltage constraints can be taken into account by
systems using Genetic Algorithms (GA) is presented in this specifying the upper and lower bounds of the magnitude of the
paper. Results (power losses, operating conditions and annual voltages. The distortion of voltage is considering by specifying
benefits) are compared with that obtained from radial and loop for maximum total harmonic distortion (THD) of voltages and
distribution systems. The radial, loop and interconnected the maximum number of banks to be installed in one location is
distribution systems models are obtained by suitably taken into account.
simplification of a typical Italian grid. Commercial package E. Mathematical Representation
ETAP PowerStation 4.0 program is used for harmonic load The capacitor placement problem is expressed
flow analysis, the proposed solution methodology is mathematically as shown below:
implemented in Microsoft Visual Basic 6 programming J
language. Min F = K p Ploss + ∑ K cj Q cj (1)
j =1
II. PROBLEM FORMULATION
Subject to:
A. Assumptions
The optimal capacitor placement problem has many Vmin ≤ V j ≤ Vmax
variables including the capacitor size, location, capacitor cost, (2)
voltage and harmonic constraints on the system. There are THD j ≤ THD max
(3)
switchable capacitors and fixed type capacitors in practice.
However, considering all variables in a nonlinear fashion will Q ≤Q
c
j
c
max (4)
make the placement problem very complicated. In order to
simplify the analysis, only fixed type capacitors are considered Where
with the following assumptions:
1) The system is balanced.
c
Qmax = L Q0c (5)
2) All the loads are constant F The total annual cost function.
B. Capacitor size and cost Kp
Annual cost per unit of power losses.
Only the smallest standard size of capacitors and multiples
Ploss The total power losses.
of this standard size are allowed to be placed at the buses to
have more realistic optimal solution. The capacitor sizes are (Result from ETAP PowerStation Harmonic Load
treated as discrete variable and the cost of the capacitor is not Flow Program).
linearity proportional to the capacitor size, this makes the J Number of buses.
formulated problem a combinatorial one. K cj
C. Objective Function The capacitor annual cost/kvar.
The objective of the capacitor placement problem is to c
Q j
reduce the total energy losses of the system while striving to The shunt capacitor size placed at bus j.
minimize the cost of capacitors installed in the system. The Vj
The rms voltage at bus j.
objective function consists of two terms. The first is the cost of
(Result from ETAP PowerStation Harmonic Load
 Flow Program). with another similarly partitioned solution. The random
Vmin position is referred to as "the crossover point". In other words,
Minimum permissible rms voltage. crossover defines the outcome as gene exchange. Crossover
Vmax Maximum permissible rms voltage. operator proved very powerful in genetic algorithms. Mutation
is the process of random modification of a particular value of a
THD j solution with a small probability. Mutation is applied to alter
The total harmonic distortion at bus j.
(Result from ETAP PowerStation Harmonic Load some genes in the solutions. When a gene exchange resulting
Flow Program). from application of a crossover operator is not meeting
THDmax Maximum permissible total harmonic distortion. appropriate restriction, mutation might be very helpful in
providing a proper gene exchange amendment. Mutation is
c generally seen as a background operator that provides a small
Qmax Maximum permissible capacitor size.
amount of random search. It increases the population diversity.
It also helps expand the search space by reintroducing
L An integer. information lost due to premature convergence. Therefore, it
Q0c drives the search into unexplored regions.
Smallest capacitor size. In addition to the above components, the stopping criterion
of the algorithm is of great significance. It determines when the
Bounds for (2), (3) are specified by the IEEE-519 standard algorithm shall be stopped or terminated and thus, considering
[1]. the best solution obtained so far as the optimal solution.
C. Design of a successful GA-based solution methodology
III. GENETIC ALGORITHM
In designing a GA-based solution methodology, several
decisions concerning the algorithm parameters shall be
A. Solution Algorithm
properly made in order to obtain high-quality solutions.
Combinatorial optimization problems can be solved either
Premature convergence to local optimum may result if the
by exact or by approximate methods. In exact methods, all the
algorithm parameters are not selected in an appropriate
feasible solutions are evaluated and the best one is selected as
manner.
the optimal solution. However, exact methods are impractical
Population size and the way the initial population is selected
when a real-life problem is to be evaluated. In this paper, GA
will have a significant impact on the results. Initial population
will be used for the solution of CPP.
could be seeded with heuristically chosen solutions or at
B. Genetic Algorithm Framework random. In either case, the initial population should contain a
There are four components in the design of a GA-based wide variety of structures. The population size and the initial
solution methodology. These include the initialization of the population are selected such that the solution domain
algorithm, fitness evaluation, selection and genetic operators. associated with the population is sufficiently covered. The
Algorithm initialization is the process of randomly population size depends on the criteria for selecting the initial
generating a set of initial feasible solutions forming the so- solutions. A constant size population of solutions shall be
called "initial population". The number of these solutions is judged by the algorithm designer. If the population size is too
referred to as the "population size". Each iteration in a genetic small, the solution domain will not be adequately searched and,
algorithm, known as a "generation", results in a new set of thus, resulting in poor performance. Premature convergence to
feasible solutions. local solutions can be prevented by using a large population
Genetic algorithm needs some fitness measure to determine size. However, this may slow down the convergence rate.
the relative 'goodness' of a particular solution. This can be The performance of a genetic algorithm is highly sensitive to
obtained either by direct evaluation of the objective function or the fitness values. The fitness value may be selected as the
by some other indirect means. Fitness evaluation is the objective function to be optimized. Some researchers, however,
criterion guiding the search process of a genetic algorithm. believe that the objective function value is a naive fitness
In genetic algorithms, parents are selected to produce measure. Therefore, using the objective function value
offspring. Selection process can be carried out in different associated with each solution as a fitness measure is rarely a
ways as discussed before in chapter three. good idea to some researchers. When applying a crossover
Genetic operators are the probabilistic transition rules operator, the resulting offspring can be either feasible of
employed by a genetic algorithm. A new and improved infeasible. There are two ways to deal with infeasible solutions.
population is generated from an old one by applying genetic One way is to design heuristic operators transforming
operators. Operators used by genetic algorithms include infeasible solutions to feasible ones. The second is by
crossover and mutation. penalizing this infeasibility in the objective function. In the
Crossover is the process of choosing a random position in latter case, selection of proper penalty factors is another
the solution and swapping the characters around this position decision to be made by the algorithm designer.
 Crossover rate and mutation rate play a very important role mutation rate in order to obtain a high quality solution.
in the performance of genetic algorithm. A higher crossover
D. Application of GA to the CPP in distorted distribution systems
rate introduces new solutions more quickly into the population.
Because of its simplicity, generality and ability to cope with
If the crossover rate is too high, high-performance solutions are
practical constraints, a genetic algorithm has been designed to
eliminated faster than selection can produce improvements. A
solve the general CPP in a distribution system. The following
low crossover rate may cause stagnation due to the lower
remarks shed some light on the design aspects of the algorithm
exploration rate. Mutation rate shall not be too high in order
as applied to the CPP:
not to prevent crossover from doing its work properly. Some
• The population size is a fixed value and the same is
researchers reported that a variable mutation rate rather than a
determined empirically by trial and error process.
fixed one is more beneficial to be utilized. In the initial stage of
• The objective function itself is used to provide fitness
the GA, it is recommended to start with a high crossover rate
values of the newly generated solutions. Once a new
and a low mutation rate since the crossover is mainly
solution is generated, its associated feasibility is checked.
responsible for the search at this stage. As the algorithm
If the solution is infeasible, the penalty factor is applied.
progresses, the crossover operator becomes less productive and
therefore, the mutation rate shall be increased. • Selection, crossover and mutation are applied as genetic
Simple GA involving just mutation and crossover has proved operators in the algorithm design. No additional operators
to be quite powerful enough. Some designers proposed to add such as the inversion operator are considered in the design
an inversion operator where a section of a solution is 'cut out' process.
and then re-inserted reversely. However, it has not been found • The algorithm is designed based on a fixed, rather than a
significantly useful. It may be used instead of mutation variable, mutation rate throughout the search.
operator to explore new regions of the searching domain. • Steady-state replacement approach is selected in the design
As pointed out earlier, newly generated solutions replace of the algorithm.
existing solutions in subsequent generations of GA. Two • The algorithm is designed to stop after predetermined
replacement approaches are available to the algorithm designer number of generation.
to select among which. These are the incremental or 'steady-
state' approach and the generational approach. In incremental Based on the above remarks, a GA-based solution
or steady state replacement, once a new feasible solution has methodology applied to the CPP has been implemented. The
been generated, it will replace an above-average fitness value algorithm implementation can be summarized as follows:
member chosen randomly. Members that fitness values are In this paper the representation by means of strings of
better than the calculated average are transferred to the next integers was chosen. Each gene(i) (represent buses) of the
generation with no change. In generational replacement a new chromosome (its length is equal to the total number of the
population of children is generated to replace the whole parent system buses (m)) can store a 0, which indicates absence of
population. The steady-state replacement approach has the capacitors on the corresponding bus or an integer different
following advantages. from 0 that indicates the number of added capacitor sizes that
1. If steady-state replacement approach is not employed, is added in the bus(i). Therefore a chromosome can be
there will be no guarantee that the best solution in the represented as Figure (1)
current population will survive into the next generation. Bus number 1 2 3 ... ... m-2 m-1 m
With this approach, best solutions are always kept in the gene 0 3 4 ... ... 2 1 3
Figure 1: A genetic algorithm chromosome of the capacitor placement problem
population and the newly generated solution is
immediately available for selection and reproduction.
The algorithm procedure can be summarized as follow:
2. It is more efficient when compared to the generational
1. Input GA parameters, i.e. population size, crossover
approach. Faster convergence is usually expected with this
rate, mutation rate, selection type, crossover type,
approach.
mutation type and termination mode.
3. It prevents the occurrence of duplicates. Duplication is
2. Generate a set of initial solutions, forming the initial
unhelpful since it wastes resources on evaluating, the same
population, randomly.
fitness function and it distorts the selection process by
3. Run ETAP Power Station 4.0 program to calculate
giving extra chances to the duplicate solutions to
system power losses, bus voltages and total harmonic
reproduce.
distortion at each bus.
Researchers, however, reported some success with the
4. Calculate the associated fitness value (objective
generational replacement approach. The genetic algorithm can
function) of each solution.
be designed to stop if a pre-specified number of iterations are
5. Check the constraints and applying the penalty factor
completed or if no improvement is encountered in the optimal
(by adding 1E10 to the fitness value) if the constraints is
solution during a given number of consecutive iterations. The
not satisfied.
stopping criterion shall be properly tuned with the other
6. Create a new population by performing selection,
parameters like the population size, crossover rate and
 crossover and mutation on the individuals. source –typical IEEE- XFMR Magnet.
7. Discard the old population and iterate using the new Kp was selected to be 168 $/kW, and the voltage limits on
population if the stopping criterion is not satisfied. the rms voltages were selected as Vmin= 0.95 pu, and
Vmax=1.05 for case (1) and Vmin= 0.93 pu, and Vmax=1.05
for case (2). The maximum THD was selected as THDmax = 5
IV. NUMERICAL RESULTS
B. Test System results
A. Test System data
The distribution network models are obtained by suitably Table (2) Comparison of results between radial, loop and
simplification of a typical Italian grid [2]. Single line diagram interconnected distribution systems for case (1)
of the network is shown in Fig. (2) and the system data as
Variable Before OCP Case (1)
follows:
Radial Loop Intercon Radial Loop Intercon
Two 132 kV HV networks with the same short circuit power nected nected
MVAsc of 6000 MVA; QC1 (kvar) --- --- --- 4050 3600 3900
Two HV/MV substations, comprising each a 132 kV HV QC2 (kvar) --- --- --- 3900 4050 4050
busbar, a 132/20 kV 40 MVA transformer and a 20 kV MV QC3 (kvar) --- --- --- 4050 3900 3600
QC4 (kvar) --- --- --- 1500 1800 3600
busbar; QC5 (kvar) --- --- --- 3300 2400 3300
A feeder, subdivided in three line sections (L01, L12 and QC6 (kvar) --- --- --- 1650 2700 2100
L23) of 3 km each with % positive sequence impedance (100 QC7 (kvar) --- --- --- 3450 3300 3450
MVA base) R=5.17, X=4.23, Z=6.68 QC8 (kvar) --- --- --- 750 3300 2400
A series of further passive overhead feeders; QC9 (kvar) --- --- --- 3450 3450 2400
Link lines between various feeder (Lm1 and Lm2); QC10 (kvar) --- --- --- 1800 2400 1200
Configuration switches (S1, S2, S3, S4, S5 and S6). QC11 (kvar) --- --- --- 2250 2550 2100
QC12 (kvar) --- --- --- 2700 1650 2250
Table (1) show the system load data
Total --- --- ---
The proposed GA was applied to the test system-2 for three capacitor 32850 35100 34350
different network configurations: (kvar)
Min. 0.91 0.91484 0.92671
1. Radial configuration (S1 open, S2 open, S3 open , S4 voltage(pu)
0.96535 0.97364 0.97521
open , S5 open and S6 open); Max. 0.9551 0.95516 0.94462
0.98778 0.98860 0.98554
2. Loop configuration (S1 open, S2 open, S3 close , S4 voltage(pu) 6
Power losses 1244.1 1211.1 1206.5
close , S5 close and S6 close); 812.8 795.7 796.8
(kW)
3. Interconnected configuration (S1 close, S2 close, S3 Cap. cost --- --- ---
6142.8 6301.65 6156.9
($/year)
open , S4 open , S5 open and S6 open). Total cost 20900 203464 202692
142685 139904 139952
($/year ) 8
Benefits ($/ --- --- ---
Table (1): Test system load data year )
66323 63560 62739
Bus No. Load (80% motor, 20% static)
MW MVAR
1 5.25 5.356
2 5.25 5.356
3 5.25 5.356
4 4.5 2.18
5 4.5 2.18
6 4.5 2.18
7 4.5 2.18
8 4.5 2.18
9 4.5 2.18
10 4.5 2.18
11 4.5 2.18
12 4.5 2.18

Commercially-available capacitor sizes with real costs/kvar

were used in the analysis. It was decided that the largest
capacitor size Qcmax should not exceed the total reactive load,
i.e., 35688 kvar. The yearly costs of capacitor sizes as
described in [3]
Optimum shunt capacitor sizes have been evaluated for the
following cases: Figure 2: Test system - Simplified Typical Italian Grid
• Case (1): All loads are assumed to be linear
• Case (2): Each 40 MVA transformer has harmonic current
 Table (3) Comparison of results between radial, loop and 40000
interconnected distribution systems for case (2) 35000
30000
25000 Total sizes of installed
Variable Before OCP Case (2) capacitors (Kvar) - Case(1)
20000
Total sizes of installed
Radial Loop Intercon Radial Loop Intercon 15000 capacitors (Kvar)- Cas e(2)
nected nected 10000
QC1 (kvar) --- --- --- 2700 900 3600 5000
QC2 (kvar) --- --- --- 3000 3150 3450 0
QC3 (kvar) --- --- --- 2550 3450 3150 Radial Loop Interconnected
QC4 (kvar) --- --- --- 1350 2100 2700
QC5 (kvar) --- --- --- 2100 1650 1800 Figure 5: Total sizes of installed capacitors (Kvar) for case
QC6 (kvar) --- --- --- 2100 2850 750 (1) & (2)
QC7 (kvar) --- --- --- 300 2400 1650
QC8 (kvar) --- --- --- 2850 2100 1800 V. CONCLUSION
QC9 (kvar) --- --- --- 2550 2550 1650 The study of the optimal capacitor placement on
QC10 (kvar) --- --- --- 3450 2400 3000 interconnected distribution systems in the presence of
QC11 (kvar) --- --- --- 900 1350 3600
QC12 (kvar) --- --- --- 3000 nonlinear loads using Genetic Algorithms (GA) is presented in
2400 300
Total capacitor --- --- ---
this paper. Results (power losses, operating conditions and
26850 27300 27450
(kvar) annual benefits) are compared with that obtained from radial
Min. 0.9099 0.9148 0.92669
voltage(pu) 0.948 0.9544 0.9649 and loop networks. The radial, loop and interconnected
Max. 0.9551 0.9551 0.94461 distribution systems models are obtained by suitably
0.98657 0.9866 0.9772
voltage(pu) simplification of a typical Italian grid. Computational results
Max. THD (%) 2.68 2.68 2.65 4.95 4.99 4.96
Power losses 1244.2 1211.1 1206.5
obtained showed that the harmonic component affects the
873.7 844.928 828.92
(kW) optimal capacitor placement in all system configurations.
Cap. cost --- --- ---
($/year)
5007 4992 5135 When all loads were assumed to be linear, interconnected and
Total cost 20903 20347 202707
151788 146940 144413
Loop system configurations offer lowest power losses and best
($/year ) 1 4 operating conditions rather than the radial system configuration
Benefits ($/ --- --- ---
57242.9 56534 58293 while radial system configuration offer best annual benefits
year )
due to capacitor placement. In distorted networks,
880
interconnected systems configuration offer lower power losses,
860
best operating conditions and best annual benefits due to
840
Power losses (KW/Year) -
capacitor placement.
820 Case(1)
Power losses (KW/Year) -
800
Case(2)
REFERENCES
780 [1] IEEE Recommended Practices and Requirements for Harmonic Control
760 in Electrical Power Systems, IEEE Std. 519-1992, 1993.
[2] M.Brenna, R.Faranda and E.Trioni, “Non-conventional Distribution
740 Network Schemes Analysis with Distributed Generation”, TEQREP,
Radial Loop Interconnected Bucharest, 2004
[3] Y. Baghzouz, S. Ertem, “Shunt Capacitor Sizing for Radial Distribution
Feeders with Distorted Substation Voltages”, IEEE Trans. on Power
Figure 3: Power losses (Kw/year) for case (1) & (2) Delivery, Vol. 5, No. 2, April 1990, pp. 650-657.

68000 Ahmad Galal Sayed received the B.Sc. degree in electrical power and
66000 machines from the Faculty of Engineering at Alexandria University, Egypt, in
64000
2003. He is currently working toward the M.Sc. degree in the department of
electrical power and machines, Faculty of Engineering, Cairo University. His
62000
research interest includes computer aided analysis of power systems, and
60000 Benefits($/year) - Case(1) power system harmonics.
58000 Benefits($/year) - Case(2)
56000
54000
52000
50000
Radial Loop Interconnected

Figure 4: Benefits ($/year) for case (1) & (2)