Artificial Bee Colony Algorithm for Optimal Placement and Sizing of

Distributed Generation

Ayse Aybike Seker1, Mehmet Hakan Hocaoglu2

1-2
Gebze Institute of Technology, Department of Electronics Engineering, 41400, Gebze, Kocaeli, Turkey
[email protected], [email protected]

Abstract and automation in the deregulated power system environment

[1]. As the penetration of DG units in distribution system
In electric power networks, to satisfy the increasing increases allocating them in an optimal way becomes necessary
demand, various methods have been improved and to achieve the maximum benefits [4].
implemented. Distributed generation (DG) is one of these The DG-unit application is a mixed integer nonlinear
methods which has many advantages as providing loss optimization problem which is the determination of the optimal
reduction, improving voltage profile, and increasing location and size for DG-unit to be installed on a network [5]. It
reliability. Besides renewable energy sources like solar cells, includes maximizing system voltages or minimizing power loss
wind turbines, fuel cells, and small hydro turbines can be and cost [6].
used as a DG-unit. In this paper, DG-unit placement and DG-unit placement has been solved by using many different
sizing process is performed with Artificial Bee Colony approaches. As a classical approach second order algorithm
(ABC) algorithm. ABC algorithm is a meta-heuristic method used in [7] to allocate DG-units for minimum power
approach inspired by the intelligent foraging behavior of loss. In analytical approaches as [8-9] optimal place of DGs are
honey bee swarms. 33, 69, 229 bus radial test systems are determined for the various load profiles in radial systems. As
used in order to show the performance of ABC algorithm in another analytical method in [10] optimal placement is
solving nonlinear optimization problems. The results show determined with unity power factor in radial as well as
that ABC algorithm can successfully determines optimal networked systems. In [11] an analytical method is introduced
place and size of DG-unit. based on the equivalent current injection techniques without use
of admittance matrix, inverse of admittance matrix or Jacobian
I. Introduction matrix which are proved to be problematic for the radial systems
[12]. A methodology, which is based on exact loss formula used
The ever increasing environmental concerns and demand of in [13] and tested in three typical distribution systems. Meta-
energy has brought many challenges on power system heuristic approaches are used as well as analytical ones for DG
companies. Distributed generation (DG) is an alternative to setting and sizing. Tabu search application is used in [14],
solve these problems [1]. DG means integrating small generators Fuzzy-GA method is used in [15], Hereford Ranch Algorithm
in a distribution system in order to meet required level of load used in [16] to reduce power loss. In [2] Craziness based
demand thereby improving the voltage profile, increasing life of Particle Swarm Optimization (CRPSO) used to consider the
system equipment, providing reliability and economic benefit optimal location for DG which in turn improves the voltage
such as minimum power loss and energy efficiency [2]. There is profile of overall system. A method called harmony search
many reasons for increasingly use of DG technologies. It is algorithm with differential operator is studied both for
more economic than running a power line to remote locations minimizing active power loss and improving voltage profile. [6]
and DG-unit usually requires shorter installation times and the presents a new optimization approach called Artificial Bee
investment risk is not so high. It provides backup power during Colony algorithm as another meta-heuristic method and applied
system outages and for system black-start it makes starting it in 69-bus test system. In this study ABC algorithm is also
generation and restoring a portion of the utility system possible applied to 229 bus real system to optimize the size and place of
without outside support after system collapse [3]. DG-unit. In [17] a new method which employs a Modified
The capacity of DG-unit is ranging from 15 kW to 50 MW Teaching-Learning Based Optimization algorithm used. These
and it gains more importance as it make use of renewable energy studies show that DG placement is important for efficiency of
sources such as Solar Cells, Fuel Cells, Small Wind Turbines system. However DG-units still need time to integrate decently
and Small Hydro Turbines as well as non renewable energy in real systems. Due to various technical challenges and DG-
sources [2]. units is one of the solutions for this issue.
Significant impact of embedded distributed generators on In this paper Artificial Bee Colony algorithm, proposed by
voltages, load demand, power loss, system reliability and Karaboga [18] is used for optimization problem of DG-units in
economy make it a key issue for distribution system planning distribution systems. The proposed method is tested on 33, 69

127
bus test systems and 229 bus real system. Results are compared by Karaboga [18] as a new heuristic method for solving
with grid search method. multidimensional optimization problems. At first it was
The paper is organized as follows: Section II provides the proposed for unconstrained problems, and then another version
formulation of problem; Section III presents the ABC algorithm; is published for constrained optimization problems [20].
Section IV describes the use of ABC algorithm in DG-unit Additionally the performance of ABC algorithm was compared
allocation; Section V includes results and Section VI outlines with differential evolution (DE), particle swarm optimization
conclusions and further studies. (PSO) and evolutionary algorithm (EA) for a set of multi-
dimensional numeric problems [21]. A modified version of ABC
II. Problem Formulation algorithm was published for constrained problems [22].
Honey bee swarm consists of three kinds of bees: employed
DG-unit optimization problem can be explained as finding bees, onlooker bees, and scout bees [22].
the optimal size and location of that DG-unit to minimize the x Employed bees: Randomly search for food source
system total power loss subject to inequality constraints. positions (solutions) and then share the information that is nectar
x The objective function is to minimize the total system amounts by dancing with the bees waiting in the hive. Duration
real power loss.  is number of buses.  and  are the real and of dance depends on the nectar amount (fitness value) of the
reactive power flows from bus  to bus  + 1. Bus voltage at bus food source.
 is  . +1 is the resistance of line connecting buses  and  + 1. x Onlooker bees: Watch dances of various employed bees

2 + 2
and chose the good food source position according to quality of
 =     +1
| |2
(1) that food source.
=1 x Scout bees: An employed bee of the source which is
x The inequality constraints are voltage limits of system abandoned becomes a scout and starts to search a new food
|  |  | |  |  | (2) source randomly [21].
where min-max voltage is ±5 % of the nominal voltage that Onlookers are placed on food sources by using probability
satisfy operating constraints. based selection, as the nectar amount increases on source
x Other inequality constraint is size of DG-unit onlooker bees prefer that source too. Scout bees are the
       (3) explorers and have no guidance while looking for food. That
where DG-units sizes are preselected discretely between 10% creates the possibility of both low or rich quality sources may be
and %80 of the total system demands. discovered. In ABC one of employed bee is selected as the scout
bee and this selection is controlled by a parameter “limit”. If the
x Power factor of DG-unit is the last inequality constraint
food source (solution) is not improved by a predetermined
which is set to operate practical values that is 0.95, 0.90 and
number of trials, its employed bee abandons the source and it
0.85 towards the optimal result [6]
. .   . .  . . 
convert to a scout. So “limit” is the number of trials before
(4)
leaving a food source [18].
x Power balance is satisfied with power flow constraints.
Pseudo-code of main body of ABC algorithm [22]:
 =     =   
1: Initialization
 2: Evaluation
 = | | | |[ cos(   ) +  sin
(   )] 3: cycle = 1
=1 (5)
4: repeat
 5: Employed Bees Phase
 = | | | |[ sin(   )   cos(   )] 6: Calculate Probabilities for Onlookers
=1
where  is the voltage magnitude of bus .  is the voltage
7: Onlooker Bees Phase
8: Scout Bees Phase
angle of bus .  ,  are active and reactive power of DG in 9: Memorize the best solution achieved so far
bus .  ,  are active and reactive power of load in bus .  , 10: cycle = cycle + 1
 are net active and reactive power injected in bus .  ,  11: until cycle =Maximum Cycle Number
are the real and imaginary part of admittance between bus  and In ABC algorithm employed bees number is equal to
. onlooker bees number. At initialization step a randomly
The values P, Q and V in the objective function is calculated distributed population of solutions is generated. Each solution
with Thukaram’s power flow algorithm for radial distribution  ( = 1,2, … ,  ) is a D-dimensional vector. Here  is the
networks that uses forward backward sweeps to calculate branch number of employed bees and D is the number of optimization
currents and node voltages [19]. parameter. After initialization the population of the positions
(solutions) is subjected to repeated cycles (cycle= 1,2,…,MCN)
III. Artificial Bee Colony (ABC) Algorithm of the search process of the employed, onlooker and scout bees.
Then, an employed bee produces a modification on the position
The artificial bee colony algorithm approach is established (solution) in her memory depending on the local information
from foraging behavior of real honey bees. ABC is introduced (visual information) and tests the nectar amount (fitness value)

128
of the new source (modified solution). If the nectar amount of
the modified solution is higher than the previous one the bee
memorizes the new position and forgets old one. Otherwise she
keeps the position of previous one in her memory. After all
employed bees complete the search process; they share the
nectar information of the food source and their positions
information with the onlooker bees on the dance area. An
onlooker bee evaluates the nectar information taken from all
employed bees and chooses a food source with a probability
related to its nectar amount. Same procedure is applied as in the
case of employed bee.
Following expression is to calculate probability value related
with that food source:
!"##
 = 
$=0 !"##
(6)

where !"## is the fitness value of the solution  and  is the

number of food source positions (solution) in other words, half
of the colony which is equal to number of employed bees.
Onlooker bees select their food source among the high value
of  . They produce a neighbor food source position  + 1 to the
selected one , then compare nectar amounts (fitness values) of
neighbour and previous position. And the same selection
criterion used by employed bees is applied to onlooker bees.
This is repeated until all onlookers are distributed. If a solution 
does not improve for a predetermined number of time (limit), its
employed bee abandon this solution and she becomes a scout
and search for a new random food source position. Once the
new position is determined, next ABC algorithm cycle (MCN)
starts
Neighbor food source position is determined with the
following expression:
"& = '*- + ('*-   ) (7)
where  /  and both 3 {1,2, … ,  }. is a random number
between [-1,1] and 3 {1,2, … , }. When the food source
position is abandoned, its employed bee becomes a scout. Scout
produces a new food source position as follows:
("& )
 =  + 4    6 (8)
where is a number between [-1,1].
ABC algorithm has three control parameters: 1) the colony
size CS ( " *'7"- ""# +  '*''" ""#). 2) limit
value 3) maximum cycle number MCN.

IV. ABC Algorithm for DG-unit Allocation

Flowchart of ABC algorithm is shown in Fig.1. The steps of

proposed ABC algorithm for DG-unit allocation are described as
follows.
1) Set control parameters of ABC: MCN (maximum cycle
number), D (parameter number to optimize) and limit which
depends on colony size and D.
2) Read system data which contains bus and line data of the
test system.
Fig. 1. Flowchart of ABC algorithm for DG-unit allocation
3) Run power flow based on Thukaram’s technique [19].
4) Initialize the food source positions  (solutions
population), here  = 1,2, … ,  .

129
 5) Make power flow for each position and check results if Place and size of DG-units calculated with grid search to
the voltage limits are in the given range. verify results of ABC algorithm. In this method all possible
6) If voltages are between the range check if all food combinations examined. DG-unit sizes are determined with a
sources are produced, otherwise turn to step 4. 100 step interval between %10-%80 of total system demands.
7) If all food sources are produced calculate fitness values And power factor (pf) is set to 0.85 (lagging). Results of grid
of  .using search method are shown on Table-2. Total active losses
1
!"## =
decrease in all systems after DG-unit placement. In 33 bus test
1+ 
(9)
system total loss reduces 68%, in 69 bus it reduces 90%, and in
where   is the response of (1) at solution . Otherwise 229 bus system total loss is 88% less than the systems default
turn to step 4. loss.
8) Modify  solutions to produce neighbor with the
equation given in (6). And evaluate them as indicated by step 5. Table-2. Grid Search Method Optimal Results for DG-unit
Follow as step 6 and 7 with turning back to modifying step 8 33-bus 69-bus 229-bus
Optimal placement 6 61 155
instead of step 4
Optimal size (MW) 3.0921 2.2456 0.7293
9) Compare food positions and apply selection process. Total MW loss 0.0682 0.0239 0.0052
10) If all onlooker bees are distributed go to step 13. CPU time (s) 20.06 137.99 4094.96
Otherwise go to next step. Vmax / Vmin 1 / 0.9126 1 / 0.9126 1/0.9212
11) Calculate probabilities by using (5).
12) For the selected onlooker bee, produce neighbor solution ABC algorithm results are recorded in Table-3, 4, 5 for 33,
using (6) and evaluate them as step 5. Follow as step 6 and 7 69 and 229-bus systems. ABC parameters are set to different
with turning back to producing neighbor step 12, instead step 4. values according to systems size and experiment results.
13) Go back step 9. Optimal placement, optimal size, total system real power loss
14) Determine the abandoned solution and replace it with a and calculation time are shown. Simulation has been run 30
new solution using (7). Check voltage limits, if it is in the range times and results of total loss are given as its best, worst and
calculate fitness and go to next step. Otherwise produce a new mean value.
solution using (7).
15) Memorize the next solution  . Table-3. Simulation Results of ABC for 33 bus system
16) If cycle=MCN, stop and print optimal solution  . 33-bus (30 independent runs)
Optimal placement Bus 6
In meta-heuristic optimization algorithms’ parameter setting
Optimal size (MW) 3.0971 (pf=0.85)
has significant impact on the performance. As an advantage ABC parameters CS=20, MCN=30
ABC has few control parameters. Furthermore structure of the Best Worst Mean
ABC algorithm directs the solutions to feasible region. limit Total MW loss 0.0679 0.0692 0.0681
CPU time (s) 2.03
parameter is important in the algorithm and is taken as
0.58.
In 33 bus system 6th and 26th buses are connected. ABC
algorithm finds randomly 6th or 26th bus in each run. In both
V. Results
cases total losses are between the best and worst values given in
Table-3 which shows both buses are acceptable as the optimal
ABC algorithms performance and validity is checked in 33
place.
and 69 radial distribution test systems and also using a real test
33-bus voltage profile
system with 229 nodes [23]. Results are compared with grid 1.05

1.04 without DG
with DG
search method and an optimization package in [24]. In this study 1.03

1.02

ABC algorithm implemented in MATLAB platform and 1.01

executed on an Intel Pentium 4 with 3.20 GHz speed and 2GB

bus voltage

0.99

0.98

RAM. 0.97

33, 69 and 229 bus tests systems data without DG-unit is 0.96

0.95

shown on the Table-1. It shows that all bus voltages are satisfied 0.94

0.93

10% limits however the regulations are restricted to 5%. 0.92

0.91

0.9
1 5 9 13 17 21 25 29 33

bus number
Table-1. Tests Systems Default Case
33-bus 69-bus 229-bus Fig. 2. Voltage profile of 33 bus system
Total MW loss 0.2110 0.2250 0.0429
Total MVar loss 0.1430 0.1021 0.0163 Results of ABC algorithm are identical to those calculated
Vmin (p.u.) 0.9038 0.9092 0.9159 using grid search method. The real power loss reduction is 68%
Vmax (p.u.) 1.0000 1.0000 1.0000 in 33-bus system. Furthermore Vmin is improved by 0.0589 p.u.
Load (MVA) 3.7150+2.3000i 3.8021+2.6945i 0.7900+0.4869i

130
 Table-4. Simulation Results of ABC for 69 bus system In further studies, different cases can be tested as multiple
69 bus (30 independent runs) DG-unit allocation, or different combinations of active and
Optimal placement Bus 61 reactive power injection which depends on the DG-unit’s
Optimal size (MW) 2.2389 (pf=0.85)
ABC parameters CS=30, MCN=40
variety.
Best Worst Mean
Total MW loss 0.0238 0.0238 0.0238 VII. References
CPU time (s) 26.84
[1] P. N. Hrisheekesha and J. Sharma. “AI Applications to Distribution
69-bus voltage profile
System with Distributed Generation”, International Journal of Computer
1.05
Applications 1(14):38–42, February 2010.
1.04 without DG
1.03
[2] K. Mistry; R. Roy, “CRPSO based optimal placement of multi-
1.02 distributed generation in radial distribution system”, Power and Energy (PECon),
1.01 2012 IEEE International Conference on , vol., no., pp.852,857, 2-5 Dec. 2012.
1
[3] M. Gandomkar, M. Vakilian, M. Ehsan, “Optimal distributed generation
bus voltage

0.99

0.98
allocation in distribution network using Hereford Ranch algorithm”, Proceedings
0.97 of the Eighth International Conference on Electrical Machines and Systems,
0.96 2005.Volume 2, Issue , 27-29 Sept. 2005, pp. 916 – 918.
0.95
[4] A. Pigazo, M. Liserre, R. A. Mastromauro, V. M. Moreno, and A.
0.94

0.93
Dell’Aquila, “Wavelet-based islanding detection in grid-connected PV systems,”
0.92 IEEE Trans. Ind. Electron., vol. 56, no. 11, pp. 4445–4455, Nov. 2009.
0.91 [5] M. R. Aghaebrahimi, M. Amiri, “Distributed generator placement
0.9
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 techniques using artificial intelligence,” Sustainable Power Generation and Supply,
bus number
SUPERGEN '09. International Conference on , vol., no., pp.1,6, 6-7 April 2009.
Fig. 3. Voltage profile of 69 bus system [6] F. S. Abu-Mouti, M. E. El-Hawary, “Optimal Distributed Generation
Allocation and Sizing in Distribution Systems via Artificial Bee Colony
In 69 bus system ABC algorithm results shows that total Algorithm”,Power Delivery, IEEE Transactions on, vol.26, no.4, pp.2090,2101,
power loss is reduced 90% and Vmin improved by 0.0634 p.u. Oct. 2011.
[7] S. Rau and Y.H. Wan, “Optimum Location of Resources in Distributed
In addition it is 5 times faster than grid search method. Planning”, IEEE Trans. Power Syst., Vol. 9, Nov. 1994, pp.2014–2020.
[8] H. L. Willis, “Analytical Methods and Rules of Thumb for Modelling
DG-Distribution Interaction”, IEEE PES Summer Meeting, vol. 3, Seattle, WA,
Table-5. Simulation Results of ABC for 229 bus system July 2000, pp. 1643–1644.
229-bus (30 independent runs) [9] T. Griffin, K. Tomsovic, D. Secrest, and A. Law, “Placement of
Optimal placement Bus 155 Dispersed Generation Systems for Reduced Losses”, 33rd Annu. Hawaii Int. Conf.
Optimal size (MW) 0.7320 (pf=0.85) Systems Sciences, Maui, HI, 2000, pp.1-9.
[10] C. Wang, M. H. Nehrir, “Analytical Approaches for Optimal Placement
ABC parameters CS=40, MCN=40
of DG Sources in Power Systems”, IEEE Trans. On Power Syst., Vol. 19, No. 4,
Best Worst Mean November 2004; pp. 2068–2076.
Total MW loss 0.0052 0.0052 0.0052 [11] T. Gozel, M. H. Hocaoglu, “An analytical method for the sizing and
CPU time (s) 273.78 siting of distributed generators in radial systems”, Electric Power Systems
Research, Volume 79, Issue 6, June 2009, Pages 912-918, ISSN 0378-7796
[12] T. Gozel, “Optimal Allocation of Distributed Generators in Distribution
In 229-bus system total power loss is reduced 88% and Systems”, PhD Thesis, Gebze Institute of Technology, 2009
system Vmin improved 0.0395 p.u. In addition as the size of the [13] N. Acharya, P. Mahat and N. Mithulananthan, “An analytical approach
for DG allocation in primary distribution network”, Int. J. Electr. Power Energy
system increase importance of time factor is seen from the Syst., 2006, 28, (10), pp. 669–746.
results. As in the 229-bus system ABC algorithm is 15 times [14] K. Nara, Y. Hayashi, K. Ikeda, and T. Ashizawa, “Application of Tabu
Search to Optimal Placement of Distributed Generators”, IEEE PES Winter
faster than the grid search method. Meeting, 2001, Vol. 2, pp.918 – 923.
[15] K. H. Kim, Y. J. Lee, S. B. Rhee, S. K. Lee, and S.K. You, “Dispersed
1.05

1.04
Generator Placement Using Fuzzy-GA in Distribution Systems”, IEEE PES,
YES yok
1.03
YES var
Summer Meeting, Vol. 3, July 2002, pp.1148–1153
1.02 [16] J. O. Kim, S. W. Nam, S. K. Park, and C. Singh, “Dispersed Generation
1.01
Planning Using Improved Hereford Ranch Algorithm”, Electric Power System
1

0.99
Research , Vol. 47, No. 1, Oct. 1998, pp.47–55.
bara gerilimleri

0.98 [17] Juan Andrés Martín García, Antonio José Gil Mena, “Optimal
0.97 distributed generation location and size using a modified teaching–learning based
0.96
optimization algorithm”, International Journal of Electrical Power & Energy
0.95

0.94
Systems, Volume 50, September 2013, Pages 65-75, ISSN 0142-0615
0.93 [18] D. Karaboga, “An idea based on honey bee swarm for numerical
0.92
optimization,” Technical Report TR06, Computer Engineering Department,
0.91
Erciyes University, Turkey, 2005
0.9
1 20 39 58 77 96 115
EDUDQXPDUDVÕ
134 153 172 191 210 229
[19] D. Thukaram, H.M. Wijekoon Banda, Jovitha Jerome, “A robust three
phase power flow algorithm for radial distribution systems”, Electric Power
Fig. 4. Voltage profile of 229 bus system Systems Research, Volume 50, Issue 3, 1 June 1999, Pages 227-236.
[20] D. Karaboga and B. Basturk, “Artificial Bee Colony (ABC)
Optimization Algorithm for Solving Constrained Optimization Problems”. Berlin,
VI. Conclusion Germany: Springer-Verlag, 2007, vol. LNAI 4529, pp. 789–798.
[21] D. Karaboga, B. Basturk, “On the performance of artificial bee colony
(ABC) algorithm”, Applied Soft Computing, Volume 8, Issue 1, January 2008,
A population based ABC algorithm is used for an Pages 687-697, ISSN 1568-4946.
[22] D. Karaboga, B. Akay, “A modified Artificial Bee Colony (ABC)
optimization problem. Simulations were tested on IEEE 33 and algorithm for constrained optimization problems”, Applied Soft Computing,
69 bus test systems and 229-bus real system. Objective function Volume 11, Issue 3, April 2011, Pages 3021-3031, ISSN 1568-4946.
[23] A. S. Safigianni, G. J. Salis, “Optimum voltage regulator placement in a
is chosen to minimize total system real power loss and ABC radial power distribution network”, IEEE Trans. on Power Systems, vol. 15, no. 2,
algorithm successfully finds optimized size and place of DG- pp. 879-886,May 2000.
[24] T. Gözel, U. Eminoglu, M.H. Hocaoglu, “A tool for voltage stability and
units. The ABC results are in close agreement with the grid optimization (VS&OP) in radial distribution systems using matlab graphical
search method. user interface (GUI)”, Simulation Modelling Practice and Theory, Volume 16,
Issue 5, May 2008, Pages 505-518, ISSN 1569-190X.

131