Sie sind auf Seite 1von 8

2 nd International Symposium on Computing in Science & Engineering

Proceeding Number: 400/14

Comparison of Metaheuristic Search Techniques in


Finding Solution of Optimization Problems

Serdar Çarbaş1, Erkan Doğan2, Ferhat Erdal 3, Mehmet Polat Saka 4


1
carbas@metu.edu.tr, 2erkan.dogan@atauni.edu.tr, 3eferhat@akdeniz.edu.tr, 4 mpsaka@eng.uob.bh

Abstract.In the present study, performance evaluation of three different metaheuristic techniques is carried
out in finding the solution of optimization problems taken from the literature. Three metaheuristic techniques
selected are the classical harmony search algorithm, improved harmony search algorithm and particle swarm
optimizer. Metaheuristic search techniques of optimization are non-deterministic methods and they rely on
heuristics in finding the better solutions in the search space. They use random or probabilistic parameters
while they search for the optimum solution rather than deterministic quantities. The source of random
variables may be several depending on the nature and the type of problem. The heuristics behind these
innovative techniques is borrowed from the nature or physics. Himmelblau’s nonlinear optimization problem
and the welded beam design problem taken from the literature are solved by using the metaheuristic
techniques mentioned and their performance is evaluated and compared.

Keywords: Metaheuristic search techniques, harmony search algorithm, particle swarm method, optimization
problems

1 Introduction

A vast amount of techniques are now available in the optimization literature to deal with various types of
problems encountered in computational structural mechanics [1]. The long-standing research on computational
efficiencies of these techniques has clearly evinced that performance of any technique is affected by a high
number of factors, such as dimensionality, multi-modality, differentiability, continuity, etc. and that each
technique has some weaknesses apart from its strengths in searching for the optimum solution. For example, the
well-known mathematical programming techniques, which have long been used in engineering optimization
applications, are characterized by their rapid convergence rates. Nevertheless, poor convergence reliability to the
global optimum associated with a gradient based search is the major drawbacks of these techniques [2].
Recently, a group of novel techniques that make use of nature as a source of inspiration to develop numerical
search algorithms have emerged as robust and versatile methods for discrete and global optimization [3]. These
metaheuristic search algorithms come up with an increased promise of finding the global optimum due to their
lack of dependence on gradient information during the search. Heuristic methods are quite suitable and powerful
for obtaining the solution of engineering optimization problems. These methods do not require the derivatives of
the objective function and constraints [4]. Many engineering design problems can be formulated as constrained
optimization problems. The presence of constraints may significantly affect the performances of any
optimization algorithms that may work well for unconstrained problems. With the increase of the research and

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


712
2 nd International Symposium on Computing in Science & Engineering

applications based on stochastic search techniques [5,6], constraint handling used in metaheuristic computation
techniques has been a hot topic in both academic and engineering fields [7-11].

2 Mathematical Formulation of an Optimization Problem

One of the most difficult parts encountered in practical engineering design optimizations is the constraint
handling. Real-world limitations frequently introduce multiple, non-linear and non-trivial constraints on a
design. A general engineering optimization problem can be defined as follows;
Minimize f(x), x={x1, x2,…,xNd} which is subjected to gi(x)≤0, i=1,2,…,p
and hj(x)=0, j=1,2,…,m where Lxk≤x≤ Uxk, k=1,2,…,Nd.
Here, f(x) is the objective function, x denotes the decision solution vector, Nd is the number of decision
variables, Lxk and Uxk, are the lower and the upper bound of each decision variable, respectively. p is the
number of inequality constraints and m is the number of equality constraints.

3 Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO) is a relatively new heuristic approach utilized for optimization problems due
to its simple principle and ease of implementation [12]. The PSO algorithm is initialized with a population
(swarm) of random potential solutions (particles). Each particle iteratively moves across the search space and is
attracted to the position of the best fitness historically achieved by the particle itself (local best) and by the best
among the neighbors of the particle (global best), [13]. The steps of the algorithm are listed in the following [14].

Step 1: Initialization. Initialize an array of particles with random positions and their associated
velocities.
Step 2: Local best updating. Evaluate the fitness function of the particles and update local best
position Pikaccording to the best current value of the fitness function.
Step 3: Global best updating. Determine the current global minimum fitness value among the current
positions and update Pgk, the global best position.

Step 4: Solution construction. Change the velocities and move each particle to the new position
considering the related velocity.
Step 5: Terminating criterion controlling. Repeat Steps 2–4 until a terminating criterion is satisfied.

4 Harmony Search Optimization (HSO)


This meta-heuristic method imitates the improvisation process of a musician seeking a pleasing harmony.
Musician can play a note from existing memory or perform variations on an existing piece or create an entirely
new piece. These actions represent the basic three operations of the harmony search method. A note can be
played from pleasing songs stored in memory or a note can be played close in pitch to one that is in the memory
or a note can be played totally randomly from the entire range of the instrument. Harmony search method
follows the same logic. The algorithm consists of five steps that are given in the following. The detailed
explanation of the method can be found in [15]

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


713
2 nd International Symposium on Computing in Science & Engineering

Step 1. Initialization. Initialize the optimization operators of HS algorithm includes the harmony
memory (hm), the harmony memory size (hms), the harmony memory considering rate (hmcr), and the
pitch adjusting rate (par).
Step 2.Harmony memory matrixis initialized. Each row of harmony memory matrix contains the
values of design variables which are randomly selected feasible solutions from the design pool for that
particular design variable.

Step 3.New harmony memory matrix is improvised.Generate a new harmony vector from the hm,
based on memory considerations, pitch adjustments, and randomization.

4.1 Improved Harmony Search Optimization (IHSO)

In classical harmony search method the parameters hmcr and par are selected prior to the application of the
method and they are kept constant until the end of the iterations. The numerical applications have shown that the
selection of values for hmcr and par is problem dependent and the initial values selected affect the performance
of the algorithm. Hence, in order to determine the optimum solution it is necessary to solve the optimization
problem several times with different values of these parameters and select the solution with minimum weight. It
is apparent that such application devaluates the efficiency of the algorithm. In order to overcome this
discrepancy, numbers of improvements are suggested in the literature [16-21]. In [22], different strategies are
proposed for hmcr and par. par is updated using the concept suggested by Coelho and Bernert [23] as follows:

par(i)= par min+( par max- par min).degree(i) (14)

where, par(i) is the pitch adjusting rate for generation i, parmin is the minimum adjusting rate, parmax is the
maximum adjusting rate, and i is the generation number. The degree is updated according to the following
expression:

degree ( i ) =
( HCost ( i ) − HCost
max mean ) (15)
( HCost ( i ) −
max HCost min ( i ) )

where, HCostmax(i) and HCostmin(i) are the maximum and minimum function objective values in generation
i, respectively; HCostmean is the mean of objective function value of the harmony memory.The improvisation of
hmcr is carried out using the following expression;

hmcr (i)= hmcr max-( hmcr max- hmcr min).degree(i) (16)

where, hmcr(i) is the harmony memory considering rate for generation i, hmcrmax is the maximum
considering rate, hmcrmin is the minimum considering rate, and i is the generation number.

Step 4.Harmony Memory matrix is updated. If a new harmony vector is better than the worst harmony
in the hm, judging in terms of the objective function value, the new harmony is included in the hm and
the existing worst harmony is excluded from the hm.
Step 5.Termination.Repeat steps 2 and 3 until the terminating criterion is satisfied.

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


714
2 nd International Symposium on Computing in Science & Engineering

5 Design Examples

5.1 Himmelblau’s Nonlinear Optimization Problem

The first problem, called Himmelblau’s function [24], is a commonly used benchmark function for nonlinear
constrained optimization problems. This problem is adopted to test the performance of proposed metaheuristic
techniques. Problem has five design variables and fifteen constraints. Problem definition and optimum design
results are as in the following;
Minimize;f ( x) = 5.3578547 x32 + 0.8356891x1 x5 + 37.293239 x1 − 40792.141 which is subjected to
0 ≤ g1 ( x ) ≤ 92, 90 ≤ g 2 ( x ) ≤ 110, 20 ≤ g 3 ( x ) ≤ 25 and side constraints are
78 ≤ x1 ≤ 102, 33 ≤ x 2 ≤ 45, 27 ≤ x3 ≤ 45, 27 ≤ x 4 ≤ 45, 27 ≤ x5 ≤ 45

where;
g1 ( x) = 85.334407 + 0.0056858 x2 x5 + 0.0006262 x1 x4 − 0.0022053 x3 x5
g 2 ( x) = 80.51249 + 0.0071317 x2 x5 + 0.0029955 x1 x2 + 0.0021813 x32
g3 ( x ) = 9.300961 + 0.0047026 x3 x5 + 0.0012547 x1 x3 + 0.0019085 x3 x4

Himmelblau [24] first solved this problem by using the generalized reduced gradient (GRG) method. Then it is
studied by Gen and Cheng [25] using genetic algorithms (GAs). Runarsson and Yao [26] proposed an
evolutionary strategies (ES) algorithm with stochastic ranking for the solution of this problem.

Table 2.Optimal results for Himmelblau’s nonlinear optimization problem.

Optimum solutions obtained by different metaheuristic methods

Design
IHSO PSO HSO ES GRG GAs
variables
x1 78.0000 78.0000 78.0000 78.0000 78.6200 81.4900

x2 33.2777 33.0003 33.0000 33.0000 33.4400 34.0900

x3 27.3060 29.9962 29.9950 29.9953 31.0700 31.2400

x4 44.9988 44.9999 45.0000 45.0000 44.1800 42.2000

x5 44.2466 36.7734 36.7760 36.7758 35.2200 34.3700


g1(x) 91.9544 92.0000 91.7147 92.0000 91.7927 91.7819
g2(x) 100.4151 98.8402 98.8405 98.8405 98.8929 99.3188
g3(x) 20.0000 20.0000 19.9999 20.0000 20.1316 20.0604

f(x) -31004.19 -30665.40 -30665.50 -30665.54 -30373.95 -30183.58

The optimal solution is obtained at x = (78.0000, 33.2777, 27.3060, 44.9988, 44.2466) with corresponding
function value equal to f(x) = -31004.19 using IHSO as demonstrated in Table 1. The comparisons of the best
solutions of this example obtained using the proposed metaheuristic search techniques with previous best
solutions reported in [24-26].

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


715
2 nd International Symposium on Computing in Science & Engineering

5.1 Welded Beam Design

A rectangular beam, designed as a cantilever beam, is selected as second example. The geometric view and the
dimensions of the beam are illustrated in Fig 1. The beam is designed to carry a certain load with minimum
overall cost of fabrication. The optimization problem has four design variables; h=x1 : the thickness of the weld,
l=x2 : the length of the welded joints, t=x3 : the width of the beam, b=x4 : the thickness of the beam [27].

Fig. 1. Welded Beam Structure

2
Minimize
f ( x) = 1.10471 x1 x 2 + 0.04811 x3 x 4 (14.0 + x 2 ) which is subjected to g 1 ( x ) = τ ( x ) − τ max ≤ 0 as
g ( x ) = σ ( x) − σ max ≤ 0 g 3 ( x) = x1 − x4 ≤ 0
shear stress, 2 as bending stress in the beam, ,
2
g 4 ( x ) = 0 . 10471 x 1 + 0 .04811 x 3 x 4 (14 . 0 + x 2 ) − 5 ≤ 0 g ( x ) = 0 . 125 − x ≤ 0
, and 5 1 as side constraints,
g 6 ( x) = δ ( x) − δ max ≤ 0
as end deflection of the beam, and g 7 ( x ) = P − Pc ( x ) ≤ 0 as buckling load on the bar. And
also, design variables of the problems are limited as;
0 .1 ≤ x1 ≤ 2 .0, 0 .1 ≤ x 2 ≤ 1 0, 0 .1 ≤ x 3 ≤ 1 0, 0 .1 ≤ x 4 ≤ 2 .0

where;
2
x +x   x 2 x + x  
τ' =
P , τ '' = M R , M = P ( L + x2 ) , R = x2
+ ( 1 3 ) 2 , J = 2  x1 x2 2  2 + ( 1 3 )2   ,
2 x1 x2 J 2 4 2   12 2  

( x32 x46 )
3 4.013E
4 PL 6 PL , 36 (1 − x3 E )
δ ( x) = 3
, σ ( x) = Pc ( x) =
E x3 x4 x4 x32 L2
2L 4 G

P = 6000 lb, L = 14 in., E = 30 × 106 psi, G = 12 ×106 psi


τ max = 13600 psi, σ max = 30000 psi, δ max = 0.25 in.

The same problem was also solved by Ragsdell and Philips [28] using geometric programming (GP). Deb [29]
used a simple genetic algorithm (GAs) with traditional penalty function to solve the same problem. Ray and
Liew solved this problem using a society and civilization algorithm (SCA) [30]. Also same problem has solved
by using proposed IHSO, HSO, and PSO algorithms. The optimum solutions and comparison of results for this
problem are tabulated in Table 2.

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


716
2 nd International Symposium on Computing in Science & Engineering

Table 2.Optimum results for welded beam design.

Optimum solutions obtained by different metaheuristic methods

Design
IHSO PSO HSO SCA GP GAs
variables
x1 (h) 0.18461 0.23886 0.25381 0.2444 0.2455 0.2489

x2 (l) 3.95998 2.5296 3.07156 6.2379 6.1960 6.1730

x3 (t) 9.10543 9.1796 7.82838 8.2885 8.2730 8.1789

x4 (b) 0.20626 0.2389 0.27413 0.2445 0.2455 0.2533

f(x) 1.77192 1.90340 1.98118 2.3854 2.3859 2.4331

6 Conclusions

The highly promising outcome of this research suggests that the proposed metaheuristic algorithms, which are
classical harmony search algorithm, improved harmony search algorithm, and particle swarm optimizer are
effective alternative for solving engineering optimization problems. Especially the new optimization approach,
improved harmony search, can also be extended to other real-world optimization problems in manufacturing and
design area. Himmelblau’s nonlinear optimization problem and the welded beam design problem taken from the
literature are solved by using the metaheuristic techniques mentioned and their performance is evaluated and
compared. The design examples considered to demonstrate the effectiveness and robustness of the improvements
suggested in the optimization techniques in this study.

References

[1] Arora, J.S. (2002). Methods for Discrete Variable Structural Optimization. Recent Advances in Optimum Structural
Design, ASCE, Ed. S. A. Burns, 1-40, USA.
[2] Sun, W and Y-X, Yuan (2006). Optimization Theory and Methods; Nonlinear Programming, Springer-Verlag.
[3] Saka, M.P. (2007). Optimum Design of Steel Frames using Stochastic Search Techniques Based on Natural
Phenomena: A Review. Civil Engineering Computations: Tools and Techniques, Ed. B.H.V. Topping, Saxe-Coburgh
Publications, pp: 105-147.
[4] Kaveh, A. and Talatahari, S. (2009). Particle swarm optimizer, ant colony strategy and harmony search scheme
hybridized for optimization of truss structures. Computers andStructures, 5-6, 87, 267-83.
[5] Michalewicz, Z. (1995). A survey of constraint handling techniques in evolutionary computation methods. In:
McDonnell, J.R. et al. (Eds.), Proceedings of the fourth Annual Conference on Evolutionary Programming, MIT
Press, Cambridge, MA, pp. 135–155.
[6] Coello, C.A.C. (2002). Theoretical and numerical constraint handling techniques used with evolutionary algorithms:
a survey of the state of the art. Computer Methods in Applied Mechanics and Engineering, 191 (11/12), 1245–1287.
[7] Saka, M.P. and Erdal, F. (2009).Harmony Search Based Algorithm for The Optimum Design of Grillage Systems To
LRFD-AISC. Structural and Multidisciplinary Optimization, 38, 1, 25-41.
[8] Saka, M.P. (2007). Optimum Geometry Design of Geodesic Domes Using Harmony search Algorithm. Advances in
Structural Engineering, 10, 6, 595-606.
[9] Çarbaş, S. and Saka, M.P. (2009). Optimum Design of Single Layer Network Domes Using Harmony Search Method.
Asian Journal of Civil Engineering, 10, 1, 97-112.
[10] Hasançebi, O., Çarbaş, S., Doğan, E., Erdal, F., and Saka, M.P. (2009). Performance evaluation of metaheuristic
search techniques in the optimum design of real size pin jointed structures. Computers and Structures, 87, 5-6, 284-
302.
[11] Hasançebi, O., Çarbaş, S., Doğan, E., Erdal, F., and Saka, M.P. (2010). Comparison of non-deterministic search
techniques in the optimum design of real size steel frames.Computers and Structures, 88, 17-18, 1033-1048.
[12] Eberhart R.C. and Kennedy, J. (1995). A new optimizer using particle swarm theory. In proceedings of the sixth
international symposium on micro machine and human science, Nagoya,Japan.
[13] Kennedy, J., Eberhart, R.C., and Shi, Y. (2001), Swarm intelligence. San Francisco (CA), Morgan Kaufman
Publishers.

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


717
2 nd International Symposium on Computing in Science & Engineering

[14] Kaveh, A. and Talatahari, S. (2009). Engineering Optimization with Hybrid Particle Swarm and Ant Colony
Optimization, Asian Journal of Civil Engineering, 10, 6, 611-628.
[15] Lee, K.S. and Geem, Z.W. (2004). A New Structural Optimization Method Based on theHarmony Search Algorithm.
Computers and Structures, 87, 5-6, 284-302.
[16] Mahdavi, M., Fesanghary, M., and Damangir, E. (2007). An improved harmony search algorithm for solving
optimization problems. Applied Mathematics and Computation 188(2):1567–1579.
[17] Taherinejad, N. (2009). Highly reliable harmony search algorithm. European Conference on Circuit Theory and
Design (ECCTD’09), Antalya, Turkey, August 23-27, 818-822.
[18] Omran, M.G.H. and Mahdavi, M. (2008). Global-best harmony search. Applied Mathematics and Computation,
198(2):643–656.
[19] Geem, Z.W. (2009). Particle-Swarm Harmony Search for Water Network Design. Engineering Optimization,
41(4):297-311.
[20] Geem, Z.W. and Roper, W.E. (2010). Various continuous harmony search algorithms for web-based hydrologic
parameter optimization. Mathematical Modelling and Numerical Optimisation, 1(3):231-226.
[21] Hasançebi, O., Erdal, F., and Saka, M.P. (2010). An Adaptive Harmony Search Method for Structural Optimization.
Journal of Structural Engineering, ASCE, 136, 4,419-431.
[22] Çarbaş, S. and Saka, M.P. (2011). Optimum Topology Design of Various Geometrically Nonlinear Latticed Domes
Using Improved Harmony Search Method. Structural and Multidisciplinary Optimization, (under review).
[23] Coelho, L.S. and Bernert, D.L.A. (2009). An improved harmony search algorithm for synchronization of discrete-time
chaotic systems Chaos. Solitons and Fractals, 41(5):2526-2532.
[24] Himmelblau, D.M. (1972). Applied Nonlinear Programming. Mc-Graw Hill, Daryaganj, New Delhi, India.
[25] Gen, M., and Cheng, R. (1997). Genetic Algorithms and Engineering Design. John Wiley and Sons, NY, USA.
[26] Runarsson, T.P. and Yao, X. (2000). Stochastic ranking for constrained evolutionary optimization. Institute of
Electrical and Electronics Engineers International Transactions on Evolutionary Computation, 4, 284-294.
[27] Doğan, E. (2010). Optimum Design of Rigid and Semi-Rigid Steel Sway Frames Including Soil-Structure Interation.
PhD Dissertation, Middle East Technical University, Ankara, Turkey.
[28] Ragsdell, K.M. and Phillips, D.T. (1976). Optimal design of a class of welded structures using geometric
programming. American Society of Mechanical Engineering Journal of Engineering for Industries,98, 1021-1025.
[29] Deb, K. (1991). Optimal design of a welded beam via genetic algorithms. American Institute of Aeronautics and
Astronautics Journal, 29, 2013-2015.
[30] Ray, T. and Liew, K.M. (2003). Society and civilization: An optimization algorithm based on the simulation of social
behavior. Institute of Electrical and Electronics Engineers International Transactions on Evolutionary Computation,
7, 386-396.

Biographies

Serdar Çarbaş- Mr. Çarbaş was born in Erzurum, Turkey. He received his BSc degree in civil engineering department from
Atatürk University, in 2002, Erzurum, Turkey. His MSc degree is from Engineering Sciences Department of Middle East
Technical University in 2008, Ankara, Turkey. Also, he is currently continuing his PhD studies in same department.
He has carried out researches on shape, size and topology optimization of latticed domes such as lamella, network, and
geodesic domes. He has several proceedings of international and national conferences on optimum shape and topology design
of latticed domes as well as optimization of steel structures. Besides, he has five international journal publication related with

structural optimization using recent metaheuristic search techniques. His research interests focuses on the optimum design of
steel structures, structural optimization of geometrically nonlinear structures such as steel domes, size, shape and topology
optimization of pin and rigidly connected steel structure systems, optimization algorithms, optimum design of cold-formed
steel sections. His present studies concern the use of meta-heuristic optimizations techniques in structural optimization
problems.
Mr. Çarbaş is a member of Chamber of Civil Engineers of The Union of Turkish Engineers and Architects.

Erkan Doğan- Dr. Doğan was born in Erzurum, Turkey. He received his BSc degree in civil engineering department from
Atatürk University, in 2002, Erzurum, Turkey. He is awarded his doctoral degree from Engineering Sciences Department of
Middle East Technical University in 2010, Ankara, Turkey.

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


718
2 nd International Symposium on Computing in Science & Engineering

His research covers the topics of optimum design of rigid and semi-rigid steel sway frames, soil-structure interaction, size,
shape, and topology optimization of pin and rigidly connected steel structural systems. He has mainly focused on the
application of metaheuristic search algorithms in structural optimization especially particle swarm optimization during his
studies. Recently, he has started working on the optimum design of grillage systems. He, also, has various publications on
international and national journals and proceedings.
Dr. Doğan has attended two international courses and seminars and he is a member of Chamber of Civil Engineers of The
Union of Turkish Engineers and Architects.

Ferhat Erdal- Dr. Erdal was born in Gaziantep, Turkey. He graduated from Akdeniz University, in 2003, Antalya, Turkey.
He undertook his both MSc and PhD degrees from Engineering Sciences Department of Middle East Technical University
(METU), in Ankara, Turkey.
His early studies involve optimum design of grillage systems and effect of beam spacing on these systems. And then, he
concentrated on cellular beams. Furthermore, his researches also involve with the optimum design of real-size steel trusses
and frames. He has eight international, two national journal paper and six studies printed in national and international
proceedings. He professionally deals with stochastic search techniques and its applications on structural systems. Lately, he
has carried out an experimental research on cellular beams. He has developed an optimum design algorithm for cellular
beams.
Dr. Erdal got involved in some international courses and seminars. He was in charge of mechanics laboratory of Engineering
Sciences Department in METU.

Mehmet Polat Saka- Prof. Saka was born in Trabzon, Turkey. He graduated from Đstanbul Technical University in 1969. He
had his PhD degree from Civil Engineering Department of University of Aston in Birmingham, United Kingdom. He is a
professor of structural engineering in the department of civil engineering at The University of Bahrain, in Bahrain. Prior to
his current post, he worked as a professor in the Department of Engineering Sciences at the Middle East Technical
University, in Ankara, Turkey.
He has been carrying out research on structural optimization since he has attained his PhD degree. Since then he has
published 6 chapters in books printed by international publishers, 52 articles in the international journals and 63 papers in the
proceedings of international conferences on structural optimization. His work covers size, shape and topology optimization of
linear and nonlinear skeletal structures as well as evolutionary structural optimization of 3-D continuous structures. He is one
of the early researchers who applied structural optimization to geometrically nonlinear structures. Furthermore, some of his
early works covers shape and topology optimization of pin and rigidly connected steel structures. In his recent publications
he has carried out shape and topology optimization of various braced domes such as geodesic, lamella and network domes
where the geometrical nonlinear behaviors of these domes are taken into consideration. His current research covers the use of
metaheuristic optimizations techniques in structural optimization.
He has membership of so many societies related with his professions and he is in editorial boards of 6 international journals.

June 1 -4, Kusadasi, Aydin, Turkey http://iscse2011.gediz.edu.tr


719

Das könnte Ihnen auch gefallen