A HYBRID COA/ε-CONSTRAINT METHOD FOR SOLVING MULTI-OBJECTIVE PROBLEMS

International Journal in Foundations of Computer Science & Technology (IJFCST) Vol.5, No.5, September 2015
DOI:10.5121/ijfcst.2015.5503 27
A HYBRID COA/ε-CONSTRAINT METHOD FOR
SOLVING MULTI-OBJECTIVE PROBLEMS
Mahdi parvizi, Elham Shadkam and Niloofar jahani
Department of Industrial Engineering, Faculty of Eng.; Khayyam University, Mashhad,
Iran
ABSTRACT
In this paper, a hybrid method for solving multi-objective problem has been provided. The proposed
method is combining the ε-Constraint and the Cuckoo algorithm. First the multi objective problem
transfers into a single-objective problem using ε-Constraint, then the Cuckoo optimization algorithm will
optimize the problem in each task. At last the optimized Pareto frontier will be drawn. The advantage of
this method is the high accuracy and the dispersion of its Pareto frontier. In order to testing the efficiency
of the suggested method, a lot of test problems have been solved using this method. Comparing the results
of this method with the results of other similar methods shows that the Cuckoo algorithm is more suitable
for solving the multi-objective problems.
KEYWORDS
Cuckoo optimization algorithm (COA), ε-Constraint, Pareto frontier, MODM (Multi-objective decision
making), Optimization.
1. INTRODUCTION
In the single-objective optimization it is assumed that the decision makers connect to a single
purpose Such as maximizing the profit, minimizing the costs, minimizing the waste, maximizing
the market share etc. But in the real world, the decision maker checks more than a single
objective. For example in order to study the production level in a company, if only the profit
would be examined and all other objectives such as customer satisfaction, staff satisfaction, the
production diversity, market share etc would be rejected, the results won't be reliable. So using
the multi-objective decision making (MODM) is necessary. Finding an optimized answer that
covers all of the restrictions together is impossible in multi-objective problems. So using the
Pareto frontier, reliable answers for a multi-objective problem will be obtained. There are many
different ways for solving multi-objective problems. These ways divide in two groups. Combined
methods (all of the objectives acts as a single one) and the limited methods (one of the objective
function will be kept and other ones would be act as the restriction).
Ehrgott and Gandibleux studied on the approximate and the accurate problems related to the
combination method of multi-objective problems [1]. Hannan and Klein submitted an algorithm
for solving multi-objective integer linear programming. This algorithm use to eliminate the extra
known dominant solutions [2]. Leumanns et al. submitted a meta-heuristic algorithm in order to
find approximate effective solutions of multi-objective integer programming, using the ε-
Constraint [3]. Sylva and Crema submitted a solution for finding the set of non-dominant vectors
in multi-objective integer linear programming [4]. Arakaw et al. combined the GDEA and the GA
methods to generate the efficient frontier in multi-objective optimization problems. [5] Deb used
the evolutionary algorithms for solving the multi-objective algorithms [6]. Nakayama drew the

28
Pareto frontier of the multi-objective optimization algorithms using DEA (Data Envelopment
Analysis) [7]. Agarwal drew the Pareto frontier of the multi-objective optimization algorithms
using GA (Genetic Algorithm) [8]. Vincova used the DEA in order to find the Pareto frontier [9].
Reyes-Sierra investigated the solution of multi-objective optimization algorithm using the particle
swarm algorithm [10]. Seiford and Tone helped the multi-objective optimization algorithm using
DEA and publishing related software [11]. Pham solved the multi-objective optimization
algorithm using the Bee Algorithm [12]. Durillo and Garc'ıa-Nieto investigated a new solution for
multi-objective optimization algorithm based on the particle swarm algorithm [13]. Yun studied
the solution of multi-objective optimization algorithm using the GA and DEA. Also he found the
Pareto frontiers of efficient points using this method [14]. Yang used the Cuckoo optimization
algorithm in order to find the Pareto frontiers [15]. Gorjestani et al. proposed a COA multi
objective algorithm using DEA method [16].
This article submits a hybrid algorithm that uses the advantages of both the Cuckoo algorithm and
the ε-Constraint method simultaneously. The submitted algorithm solves the multi-objective
problems for allowable εs using the Cuckoo algorithm and the Matlab software. At last for each
iteration, it finds a Pareto answer and linking these answers draw the Pareto frontier. This method
draws a better Pareto frontier than other similar methods. In the second section, the Cuckoo
algorithm will be introduced. The third section explains the multi-objective algorithm and the ε-
Constraint method. In the fourth section, the suggested hybrid algorithm of this article will be
investigated in details. Test problems and their solution with similar algorithms compares in the
fifth section then in the last section, the conclusion and the future offers will be submitted.
2. THE CUCKOO ALGORITHM INTRODUCTION
The cuckoo search was expanded by Xin-She Yang and Suash Deb in 2009. After that the
Cuckoo optimization algorithm was submitted by Ramin Rajabioun in 2011 [17]. This algorithm
applied in several researches such as production planning problem [18](Akbarzadeh and
Shadkam, 2015), portfolio selection problem [19](Shadkam et al., 2015), evaluation of
organization efficiency [20](Shadkam and Bijari, 2015), evaluation of COA [21](Shadkam and
Bijari, 2014) and so on
Flowchart of the Cuckoo algorithm is given in the figure 1

Figure1: the Cuckoo algorithm flowchart
For more information refer to [17
3. THE MULTI-OBJECTIVE ALGORITHM
General form of a multi-objective opt
Max (Min)=
Max (Min)=
⋮
Max(Min)=
s.t.
, 1, 2 …
0, 1,2, … . ,
In the multi-objective problems, we face some objectives in contrast of single
algorithms that has just one objective. In this model, k is the number of objective functions that
can be either max or min and m is the number of restrictions and n is the number problem's
variables.
In the multi-objective optimization problems there is
objective functions simultaneously. For this reason, the Pareto optimal
Figure1: the Cuckoo algorithm flowchart
7].
OBJECTIVE ALGORITHM AND THE ε-CONSTRAINT METHOD
objective optimization problem is as (1):

objective problems, we face some objectives in contrast of single
min and m is the number of restrictions and n is the number problem's
objective optimization problems there is not a certain answer that optimizes
objective functions simultaneously. For this reason, the Pareto optimal concept is introduced.
29
ONSTRAINT METHOD:
(1)
objective problems, we face some objectives in contrast of single-objective
min and m is the number of restrictions and n is the number problem's
not a certain answer that optimizes all of the
concept is introduced.

30
The Pareto optimal concept explanation is that x∗ x ... ,x , x! is an optimal pareto. If for
each allowablex"and i={1,2,..k}, we have (for minimizing problem): ∀ %& ̅∗ ≤ ̅
Then x∗ will be the optimal Pareto that n is the number of decision making variables and k is the
number of objective functions. In other words, x∗ is an optimal Pareto if there is no other x"vector
that doesn’t make at least one objective function worse in order to improve some of the objective
functions.
4. ε-CONSTRAINT METHOD
In this method, one of the different objective functions will be selected and other objective
functions will act as the restrictions considering a specific constraint and the problem changes
into a single-objective problem. Using different εs results optimal pareto answers.
General form of this method is given as (2).
Min F(X)={ , … , )(x)}
*. +.
( ) <=>b
≥ 0
Min F(X)= ( )
s.t.
( ) <=>b
( ) ≤ ε , ≠ , = 1,..,n
≥ 0
(2)
If the objective function is max, the constraint is f/(x) ≥ ε/. Selecting the ε is the most important
thing in this method because the answers are so sensitive to this parameter. So the selected ε must
be in range of f/
01!
≤ ε/ ≤ f/
023
for each objective function.
5. THE COA/ ε-CONSTRAINT HYBRID ALGORITHM
Step 1: first according to the objective function of main problem, the mathematical model will be
written based on the ε-Constraint method and the problem is converted from multi-objective to
single-objective problem.
Step 2: the obtained function from the ε-Constraint method will be described as the meta-heuristic
Cuckoo algorithm function.
Step 3: the iterations including the εs for solving the main problem is formed for the Cuckoo
algorithm, and this loop will be iterate until the Cuckoo algorithm ends.
Step 4: according to the first laying time and the initial number of cuckoos, a matrix will be
formed from the habitats in the beginning of implementing the Cuckoo algorithm.
Step 5: the obtained function from step 2 gets the habitats matrix as its input data and finds the
objective problem magnitude according to the new restrictions for each habitat.
Step 6: the Cuckoo algorithm sorts the habitats according to their quantity and objective functions
as usual and the rest of the tasks will be the same as it is described in references number [17].
Step 7: in each iteration of the loop, the habitat that earns the most quantity of the objective
function called best cuckoo and will be saved in a different matrix.
Step 8: after exiting the formed loop, the functionsf and f calculate for each saved points.

31
Step 9: a plot of the f and f functions quantity will be drawn that is the Pareto frontier of the
main multi-objective optimization problem.
6. SOLVING TEST PROBLEMS
A number of test functions have been provided that can help to validate the suggested approach in
table 4.
Table 4. Test problems
According to high importance of the input parameters of meta-heuristic algorithm and its effect
on the final answer, the parameters of the Cuckoo algorithm for solving any problems are given
below:
Number of initial population=5, minimum number of eggs for each=2, maximum number of eggs
for each cuckoo= 4, number of clusters=1.
constraintsObjective function
, ≥ 0
4≤ ( − 2) + ( − 2) =
=
1
( − 1)5
+ ≤ 0
, ≥ 0
= 2 −
= −
2
− −3 5
≥ 0
≥ −1, ≤ 2
=
=
3
, ≥ 0
( ) + ( − 5) ≥ 25 = 4 + 4
= ( − 5) + ( − 5)
4
−5 ≤ ≤5, i=1,2,3
= ( : −10exp (
=
− 0.2> + ? )
= ∑ [| |C.D5
= + 5 Sin( 5
)]
5
Cos(16arctan(
GH
GI
)) ≥ 0-1- 0.1 +
( − 0.5) + ( − 0.5) ≥ −0.5
≥ 0, ≥ π
=
=
6
−4 ≤ x1 ≤4, i = 1,2
min f = 1 − exp(− :(x1 −
1
√n
)
!
1=
)
min f = 1 − exp(−:(x1 +
1
√n
)
!
1=
)
7
∈ [0.1,1N, ∈ [0,5N
9 + ≥ 6
9 + ≥ 1
=
=
(1 + )8
3 − ≥ −10
∈ [−20,20N, ∈ [−20,20N
( ) + ( ) ≥ 225
= ( − 2) +( − 1) + 2
( − 1) - = 9
9

32
Test problem 1: [22,17]
For solving the example using the ε-Constraint method, one of the objective functions will be
kept and the other one will be added to the constraints as it mentioned before. For this test
function, we keep f in the objective function and add f to the constraint and the problem will be
as the equation (3):
=
s.t.
( − 2) + ( − 2) ≥ 4
≥ ε
0
(3)
In order to find the allowable range of ε, f will be solved once with min function and once with
max function. The allowable range will be the 0 ≤ε≤ 4.
For finding allowable εs with the pace of 0.01, the problem will be solved using the Cuckoo
algorithm and the Matlab software for 400 iterations. The Pareto frontier is shown in figure 2.
Also the results of finding the Pareto frontier using the similar methods are shown in this figure
too.
Figure 2. Comparing the suggested method with other methods
Ranking MethodDEA Method
COA/ε–Constraint Method

33
The converted problem is as the equation (4)
= 2 −
s.t.
+( − 1)5 ≤ 0
− ≥ ε
0
(4)
The allowable range of ε will be the −1 ≤ε≤ 0and the pace is 0.0025. The Pareto frontier after
400 iterations is shown in figure 3.
The allowable range of ε will be the −2 ≤ε≤ 2 and the pace is 0.01and the Pareto frontier after

34
Test problem 4:[23,18]
The allowable range of ε will be the 0 ≤ε≤ 50 and the pace is 0. 125. The Pareto frontier after

35
The allowable range of ε will be the −11 ≤ε≤ 20 and the pace is 0. 0775. The Pareto frontier
after 400 iterations is shown in figure 6.
Ranking MethodGDEA Method
SPEA MethodNSGA-II Method

36
The allowable range of ε will be the 0 ≤ε≤ 1.2 and the pace is 0. 008. The Pareto frontier after
The allowable range of ε will be the −25 ≤ε≤ 1 and the pace is 0. 065. The Pareto frontier after
Ray–Tai–Seow’s MethodNSGA-II Method

37
The allowable range of ε will be the 1 ≤ε≤ 9 and the pace is 0.02. The Pareto frontier after 400
iterations is shown in figure 9.
Ranking MethodGDEA Method

38
The allowable range of ε will be the −196 ≤ε≤ 72 and the pace is 2.68. The Pareto frontier after
NSGA-II_MethodRay–Tai–Seow’s Method
Ray–Tai–Seow’s MethodNSGA-II Method

39
According to the Pareto frontier resulted from different functions, it is evident that the suggested
method provides uniform and exact frontiers in fewer iterations than other similar methods.
7. CONCLUSION
In this paper, we presented a hybrid method for solving multi-objective problems using the
Cuckoo algorithm and the ε-Constraint method. According to the obtained results from the
proposed method and comparing the obtained Pareto frontiers with the results of similar methods
such as GDEA/GA, DEA/GA, RANKING, NSGA-II, Ray–Tai–Seow’s and SPEA, we concluded
that not only the Cuckoo algorithm finds better Pareto frontiers but also, it needs shortest time to
give the Pareto frontier. Pareto frontier of proposed method has more dispersion than the other
similar algorithms. So the COA/ε-Constraint method is a suitable and reliable method for solving
multi-objective optimization problems. In the future, solving the problems with more objectives,
multi-objective allocation problem and multi-objective problems of project controlling with
minimizing the time and cost target would be in order.
REFERENCES
[1] Ehrgott, M., Gandibleux, X., Bound Sets for Biobjective Combinatorial Optimization Problems,
Computers & Operations Research, Vol. 34, Issue 9, pp. 2674-2694, 2007.
[2] Klein, D., Hannan, E., An Algorithm for the Multiple Objective Integer Linear Programming
Problem, European Journal of Operational Research, 9, 378–385, 1982.
[3] Laumanns, M., Thiele, L., Zitzler, E., An Efficient Adaptive Parameter Variation Scheme for Meta-
heuristics Based on the Epsilon-Constraint Method. European Journal of Operational Research, 169,
pp. 932–942, 2006.
[4] Sylva, J., Crema, A., A Method for Finding the Set of Non-Dominated Vectors for Multiple Objective
Integer Linear Programs, European Journal of Operational Research, 158, pp. 46–55, 2004.
[5] Arakawa, M., Nakayama, H., Hagiwara, I., Yamakawa, H., Multiobjective Optimization using
adaptive range genetic algorithms with data envelopment analysis, Vol.3, 1998.
[6] Deb, K., Multi-Objective Optimization using Evolutionary Algorithms, John & Wiley Sons, Ltd.,
2001.
[7] Yun, Y.B., Nakayama, H., Tanino, T., Arakawa, M., Generation of efficient frontiers in multi-
objective optimization problems by generalized data envelopment analysis, European Journal of
Operational Research, Vol.129, No.3, pp.586–595, 2001.
[8] Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.: A fast and elitist multiobjective genetic algorithm:
NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197, 2002.
[9] Vincova, K., Using DEA Models to Measure Efficiency, BIATEC, Volume XIII, 8/2005.
[10] Reyes-Sierra M., and Coello Coello CA., multiple objective particle swarm optimizers: A survey of
the state-of-art. International Journal of Computational Intelligence Research 2(3), 287–308, 2006.
[11] Cooper W.W., Seiford, L.M. and Tone, K., Data Envelopment Analysis: A Comprehensive Text with
Models, Applications, References and DEA Solver Software. Springer, New York, 2007.
[12] Pham D.T., Ghanbarzadeh A., multi-objective optimization using the bees algorithm. In: Third
international virtual conference on intelligent production machines and systems (IPROMS 2007):
Whittles, Dunbeath, Scotland, 2007.
[13] Nebro A.J., Durillo J.J., Garc´ıa-Nieto J., Coello Coello C.A., Luna F., and Alba E., SMPSO: A new
PSO-based metaheuristic for multi-objective optimization. 2009 IEEE Symposium on Computational
Intelligence in Multicriteria Decision-Making (MCDM 2009). IEEE Press, New York, pp. 66–73,
2009.
[14] Yun, Y.B., Nakayama, H., Utilizing expected improvement and generalized data envelopment
analysis in multi-objective genetic algorithm, J Glob Optim, 2013.
[15] Yang, X.S., Deb, S., Multiobjective cuckoo search for design optimization, Elsevier- Computers &
Operations Research40 1616-1624, 2013.
[16] Gorjestani, M., Shadkam, E., Parvizi, M., Aminzadegan, S., A HYBRID COA-DEA METHOD FOR
SOLVING MULTI-OBJECTIVE PROBLEMS, International Journal on Computational Science &
Applications, Vol.5, No.4, 2015.

40
[17] Rajabioun, R., Cuckoo Optimization Algorithm”, Applied Soft Computing, Vol 1, pp 5508–5518,
2011.
[18] Akbarzadeh, A., Shadkam, E., THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR
PRODUCTION PLANNING PROBLEM, International Journal of Computer-Aided technologies,
Vol.2, No.3, 2015.
[19] Shadkam, E., Delavari, R., Memariani, F., Poursaleh, M., PORTFOLIO SELECTION BY THE
MEANS OF CUCKOO OPTIMIZATION ALGORITHM, International Journal on Computational
Sciences & Applications, Vol.5, No.3, 2015
[20] Shadkam E., Bijari M., The Optimization of Bank Branches Efficiency by Means of Response
Surface Method and Data Envelopment Analysis: A Case of Iran, Journal of Asian Finance,
Economics and Business Vol. 2 No. 2, 13-18, 2015.
[21] Shadkam, E., Bijari, M., Evaluation the Efficiency of Cuckoo Optimization Algorithm, International
Journal on Computational Sciences & Applications, Vol.4, No.2, 2014.
[22] Yun, Y.B., Nakayama H., Tanino T., Arakawa, M., Generation of efficient frontiers in multi
objective optimization problems by generalized data envelopment analysis, European Journal of
Operational Research 129, 586-595, 2001.
[23] Yun, Y.B., Nakayama, H., Arakdwa, M., Fitness Evaluation using Generalized Data Envelopment
Analysis in MOGA, ieee6ecdccf0-70f0-20131005123201.
[24] Deb, K., Associate Member, IEEE, Amrit Pratap, Sameer Agarwal, and T. Meyarivan, A Fast and
Elitist Multiobjective Genetic Algorithm: NSGA-II, IEEE TRANSACTIONS ON EVOLUTIONARY
COMPUTATION, VOL. 6, NO. 2, 2002.

A HYBRID COA/ε-CONSTRAINT METHOD FOR SOLVING MULTI-OBJECTIVE PROBLEMS

More Related Content

What's hot (17)

Similar to A HYBRID COA/ε-CONSTRAINT METHOD FOR SOLVING MULTI-OBJECTIVE PROBLEMS (20)

More from ijfcstjournal (20)

Recently uploaded (20)

A HYBRID COA/ε-CONSTRAINT METHOD FOR SOLVING MULTI-OBJECTIVE PROBLEMS