Upload
omar-m-saad
View
216
Download
0
Embed Size (px)
Citation preview
Applied Mathematics and Computation 170 (2005) 216–225
www.elsevier.com/locate/amc
An iterative goal programming approachfor solving fuzzy multiobjective integer
linear programming problems
Omar M. Saad
Department of Mathematics, College of Science, Qatar University, P.O. Box 2713, Doha, Qatar
Abstract
This paper presents an iterative goal programming approach for solving fuzzy mul-
tiobjective integer linear programming problems. These problems involve fuzzy param-
eters in the right-hand side of the constraints. The concept of a-level set of these fuzzyparameters is introduced with the definition of their membership function. A solution
algorithm is described in sequential steps to solve the formulated model. The suggested
approach in this paper is mainly based on the iterative goal programming technique
together with Gomory cuts. An illustrative numerical example is given to clarify the the-
ory and the solution algorithm.
� 2005 Elsevier Inc. All rights reserved.
Keywords: Goal programming; Multiobjective integer linear programming; Fuzzy parameters;
Gomory cuts
1. Introduction
In literature there are several approaches available to deal with multiobjec-
tive integer linear programming problems. Goal programming is one powerful
0096-3003/$ - see front matter � 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2004.11.026
E-mail address: [email protected]
O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225 217
approach that has been proposed for the modeling, analysis and solution of
multiobjective optimization problems. Previous results on the use of a wide
variety of goal programming models to solve these problems are reported in
[2,6–8,13–15,18].
Effective integer goal programming methods for all-integer, mixed-integer
and zero–one multiobjective problems have been introduced by Lee andMorris [10]. These methods were based on the cutting-plane algorithm,
branch-and-bound method, and implicit enumeration technique.
In Ignizio [6], a hybrid approach combining generalized goal programming
and generalized networks has been presented for the modeling of certain large-
scale multiobjective integer programming problems.
Hughes and Saad [5] discussed stability of efficient solutions in the decision
space for parametric multiobjective integer linear programming problems via
goal programming tools. The approach presented in [5] depends on Balinskialgorithm [1] together with the iterative approach suggested by Dauer and
Krueger [3] with the help of Gomory cuts [9,16]. Later on, Osman et al. [11]
investigated stability of efficient solutions for zero–one multiobjective linear
programming problems using goal programming.
More recently, Saad and Sharif [12] presented an iterative goal program-
ming approach for solving multiobjective integer linear programming prob-
lems. The definition of, and an algorithm to determine the stability set of the
first kind for these problems have been characterized. The present paper isan extension of the study introduced in [12] to cover the problem investigated
previously, but with fuzzy parameters in the right-hand side of the constraints.
To our knowledge, this problem has not yet studied before in the integer case
and under fuzzy environment.
The roots of the present paper lie in the following sections: Section 2
presents the formulation of multiobjective integer linear programming pro-
blem involving fuzzy parameters in the right-hand side of the constraints
with the associated fuzzy integer linear goal programming model. Section 3contains some basic definitions in the fuzzy theory together with the state-
ment of the nonfuzzy version of the formulated goal programming model. In
Section 4, a solution algorithm is suggested and described in sequential steps
to solve such programs. Section 5 provides a numerical example to illustrate
the theory and the solution algorithm. Finally, Section 6 contains the
conclusions.
2. Problem formulation
In this section, we begin by introducing the following multiobjective integer
linear programming problem with fuzzy parameters in the right-hand side of
the constraints:
218 O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225
ðMOILPPÞ~m : max ZðxÞ; ð1:aÞsubject to
x 2 X ð~mÞ; ð1:bÞ
where
X ð~mÞ ¼ x 2 Rn=grðxÞ 6 ~mr; r ¼ 1; 2; . . . ;m; x P 0 and integerf g;
and Z:Rn ! Rk, Z(x) = (z1(x),z2(x), . . .,zk(x)) is a vector-valued criterionwith zi(x), i = 1,2, . . .,k, are real-valued linear objective functions, ~m ¼ð~m1;~m2; . . . ;~mmÞT is a vector of fuzzy parameters and Rn is the set of all ordered
n-tuples of real numbers. Furthermore, the constraints functions gr(x),
r = 1,2, . . .,m, are assumed to be linear.
Now, going back to problem ðMOILPPÞ~m (1.a) and (1.b), we can write an
associated fuzzy integer linear goal programming model ðILGPÞ~m consisting
of k goals and having ~m 2 Rm a vector of fuzzy parameters in the right-hand side
of the constraints. This model may be expressed as:
ðILGPÞ~m : Achieve : z1ðxÞ ¼ h1; ð2:aÞz2ðxÞ ¼ h2;
..
.
zkðxÞ ¼ hk;
and the constraints are given by :
grðxÞ 6 ~mr; r ¼ 1; 2; . . . ;m; ð2:bÞx P 0 and integer; ð2:cÞ
where h1,h2, . . .,hk are scalars and represent the aspiration levels associated
with the objectives z1(x),z2(x), . . .,zk(x), respectively.
3. Fuzzy concepts
The fuzzy theory has been advanced by L.A. Zadeh at the University of Cali-fornia in 1965. This theory proposes a mathematical technique for dealing with
imprecise concepts and problems that have many possible solutions. The con-
cept of fuzzy mathematical programming on a general level was first proposed
by Tanaka et al. (1974) in the framework of the fuzzy decision of Zadeh and
Bellman [17].
For the development that follows, we shall introduce some definitions con-
cerning trapeziodal fuzzy numbers and their membership functions, come from
[4], and that will be used throughout this paper. It should be noted that anequivalent approach can be used in the triangular fuzzy numbers case.
O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225 219
Definition 1. A real fuzzy number ~a is a fuzzy subset from the real line R with
membership function l~aðaÞ that satisfies the following assumptions:
(1) l~aðaÞ is a continuous mapping from R to the closed interval [0,1],
(2) l~aðaÞ ¼ 0 8a 2 ð1; a1�,(3) l~aðaÞ is strictly increasing and continuous on [a1,a2],(4) l~aðaÞ ¼ 1 8a 2 ½a2; a3�,(5) l~aðaÞ is strictly decreasing and continuous on [a3,a4],
(6) l~aðaÞ ¼ 0 8a 2 ½a4;þ1Þ,
where a1, a2, a3, a4 are real numbers and the fuzzy number ~a is denoted by~a ¼ ½a1; a2; a3; a4�.
Definition 2. The fuzzy number ~a is a trapezoidal number, denoted by[a1,a2,a3,a4], and its membership function l~aðaÞ is given by (see Fig. 1)
l~aðaÞ ¼
0; a 6 a1;
1 aa2a1a2
� �2
; a1 6 a 6 a2;
1; a2 6 a 6 a3;
1 aa3a4a3
� �2
; a3 6 a 6 a4;
0; otherwise:
8>>>>>>>><>>>>>>>>:
Definition 3. The a-level set of the fuzzy number ~a is defined as the ordinary
set Lað~aÞ for which the degree of their membership function exceeds the level
a 2 [0, 1]:
∼µ a (a)
1
0 a1 a2 a3 a 4 a
Fig. 1. Membership function of a fuzzy number ~a.
220 O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225
Lað~aÞ ¼ a 2 Rjl~aðaÞ P af g:For a certain degree a� 2 [0,1] with the corresponding a-level set of the fuzzynumbers ~mr, problem ðILGPÞ~m can be understood as the following nonfuzzy lin-ear goal programming model written as:
ðILGPÞm : Achieve : z1ðxÞ ¼ h1; ð3:aÞ
z2ðxÞ ¼ h2;
..
.
zkðxÞ ¼ hk;
subject to
grðxÞ 6 mr; r ¼ 1; 2; . . . ;m; ð3:bÞ
mr 2 Lað~mrÞ; r ¼ 1; 2; . . . ;m; ð3:cÞ
x P 0 and integer; ð3:dÞ
where Lað~mrÞ is the a-level set of the fuzzy parameters ~mr; r ¼ 1; 2; . . . ;m.We now rewrite problem (3.a) and (3.b) above in the following equivalent
form:
ðILGPÞm : Achieve : z1ðxÞ ¼ h1; ð4:aÞz2ðxÞ ¼ h2;
..
.
zkðxÞ ¼ hk;
subject to
grðxÞ 6 mr; r ¼ 1; 2; . . . ;m; ð4:bÞ
nð0Þr 6 mr 6 N ð0Þr ; r ¼ 1; 2; . . . ;m; ð4:cÞ
x P 0 and integer: ð4:dÞ
It should be noted that the constraint mr 6 Lað~mrÞ; r ¼ 1; 2; . . . ;m, has been re-
placed by the equivalent constraint nð0Þr 6 mr 6 N ð0Þr ; r ¼ 1; 2; . . . ;m, where nð0Þr
and N ð0Þr are lower and upper bounds on mr.
Taking into account restrictions grðxÞ 6 m�r ; r ¼ 1; 2; . . . ;m, and for the pur-pose for solving the integer linear goal program (4.a)–(4.d) at mr ¼m�r ¼ N ð0Þ
r ; r ¼ 1; 2; . . . ;m, for a certain degree a = a� 2 [0, 1], we use the itera-
tive approach developed in [3] together with the Gromory cuts shown in
[9,16]. First, we solve the following integer linear optimization problem associ-ated to the first goal, viz:
O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225 221
P 1ðm�r Þ : minD1 ¼ d1 þ dþ
1 ; ð5:aÞsubject to
z1ðxÞ þ d1 dþ
1 ¼ h1; ð5:bÞgrðxÞ 6 m�r ; r ¼ 1; 2; . . . ;m; ð5:cÞd1 P 0; dþ
1 P 0; x P 0 and integer; ð5:dÞ
where d1 and dþ
1 are the underattainment and the overattainment, respectively,
of the first goal where d1 � dþ
1 ¼ 0.
Suppose this problem has integer optimal value D�1 ¼ d�
1 þ dþ�1 with at least
one value d�1 or dþ�
1 nonzero, i.e., d�1 � dþ�
1 ¼ 0.
Now, the attainment problem for goal 2 is equivalent to the integer optimi-
zation P 2ðm�r Þ, where
P 2ðm�r Þ : minD2 ¼ d2 þ dþ
2 ; ð6:aÞsubject to
z2ðxÞ þ d2 dþ
2 ¼ h2; ð6:bÞz1ðxÞ þ d
1 dþ1 ¼ h1; ð6:cÞ
d1 þ dþ
1 ¼ D�1; ð6:dÞ
grðxÞ 6 m�r ; r ¼ 1; 2; . . . ;m; ð6:eÞdi P 0; dþ
i P 0; x P 0 and integer; i ¼ 1; 2: ð6:fÞ
Letting D�2 ¼ d�
2 þ dþ�2 denotes the integer optimal value of P 2ðm�r Þ, we can pro-
ceed to goal 3.
The general attainment problem Pjðm�r Þ for goal j is written as:
P jðm�r Þ : minDj ¼ dj þ dþ
j ; ð7:aÞsubject to
ziðxÞ þ di dþ
i ¼ hi; 1 6 i 6 j; ð7:bÞdi þ dþ
i ¼ D�i ; 1 6 i 6 j 1; ð7:cÞ
grðxÞ 6 m�r ; r ¼ 1; 2; . . . ;m; ð7:dÞdi P 0; dþ
i P 0; x P 0 and integer; 1 6 i 6 j; ð7:eÞ
where di and dþ
i are the underattainment and the overattainment, respectively,
of the ith goal level and di � dþ
i ¼ 0.The integer objective value of P jðm�r Þ, D�
j , is the maximum degree of attain-
ment for goal j subject to the maximum attainment of goals 1,2, . . ., j 1. No-
tice that D�j ¼ 0 if and only if goal j is attained.
Let x� be the optimal integer solution of the integer attainment problem
Pkðm�r Þ associated with the minimum D�k , then the solution of the integer goal
program (ILGP)m (4.a)–(4.d) is given by x� with a = a� 2 [0, 1].
222 O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225
4. Solution algorithm
In this section we develop a solution algorithm to solve the integer linear
goal program (4.a)–(4.d). The outline of this algorithm is as follows:
Step 0. Set a = a� = 0.Step 1. Determine the points (a1,a2,a3,a4) for each fuzzy parameter
~mr; r ¼ 1; 2; . . . ;m, in program ðILGPÞ~m (2.a)–(2.c) with the corre-
sponding membership function l~að~vÞ P a� for the vector of fuzzy
parameters ~m ¼ ð~m1;~m2; . . . ;~mmÞT.Step 2. Convert program ðILGPÞ~m (2.a)–(2.c) into the nonfuzzy integer linear
goal program ðILGPÞ~m (4.a)–(4.d).Step 3. Choose mr ¼ m�r ¼ N ð0Þ
r ; r ¼ 1; 2; . . . ;m, and solve problem P 1ðm�r Þ (5.a)–(5.d) using Gomory�s cutting-plane method [9,16] and obtain D�
1.Step 4. Set j = 2.
Step 5. Using D�1;D
�2; . . . ;D
�j1, solve P jðm�r Þ using the same method used in Step
3.
Let D�j denotes the minimum.
Step 6. If j 5 k, set j = j + 1 and go to Step 5. Otherwise, go to Step 7.
Step 7. Let x� denotes the optimal integer solution of problem Pkðm�r Þ associ-ated with the minimum D�
k .
Step 8. Set a = (a� + step) 2 [0,1] and go to Step 1.Step 9. Repeat again the above procedure until the interval [0,1] is fully
exhausted. Then stop.
5. An illustrative example
Consider the following integer linear goal program involving fuzzy param-eters ð~m1;~m2;~m3Þ in the right-hand side of the constraints:
ðILGPÞ~m : goal 1 : Achieve 2x1 þ x2 ¼ h1;
goal 2 : Achieve x1 þ 2x2 ¼ h2;
subject to
x1 þ x2 6 ~m1;
x1 þ x2 6 ~m2;
6x1 þ 2x2 6 ~m3;
x1 P 0; x2 P 0 and integers:
Assume that the membership function corresponding to the fuzzy parame-
ters is in the form:
O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225 223
l~aðaÞ ¼
0; a 6 a1;
1 aa2a1a2
� �2
; a1 6 a 6 a2;
1; a2 6 a 6 a3;
1 aa3a4a3
� �2
; a3 6 a 6 a4;
0; a P a4:
8>>>>>>><>>>>>>>:
where ~a corresponds to each ~mi; i ¼ 1; 2; 3. In addition, we assume also that the
fuzzy numbers are given by the following values:
~m1 ¼ ð2; 4; 6; 8Þ; ~m2 ¼ ð0; 3; 5; 7Þ; ~m3 ¼ ð18; 20; 22; 24Þ:Setting a = a� = 0, then we get:
2 6 ~m1 6 8; 0 6 ~m2 6 7; 18 6 ~m3 6 24:
By choosing m� ¼ ðm�1; m�2; m�3Þ ¼ ð8; 7; 24Þ, then the aspiration levels of the goals
have been found h1 = 10 and h2 = 15, respectively.The integer optimization problem associated with the first goal is:
P 1ðm�r Þ : minD1 ¼ d1 þ dþ
1 ;
subject to
2x1 þ x2 þ d1 dþ
1 ¼ 10;
x1 þ x2 6 8;
x1 þ x2 6 7;
6x1 þ 2x2 6 24;
d1 P 0; dþ
1 P 0; x1 P 0; x2 P 0 and integers:
The maximum degree of attainment of problem P 1ðm�r Þ is D�1 ¼ 0 with the
optimal integer solution
x1 ¼ ð2; 6Þ and d1 ¼ 0; dþ
1 ¼ 0:
The attainment problem for goal 2 is equivalent to the integer optimization
problem P 2ðm�r Þ whereP 2ðm�r Þ : minD2 ¼ d
2 þ dþ2 ;
subject to
x1 þ 2x2 þ d2 dþ
2 ¼ 15;
2x1 þ x2 þ d1 dþ
1 ¼ 10;
d1 þ dþ
1 ¼ 0;
x1 þ x2 6 8;
x1 þ x2 6 7;
6x1 þ 2x2 6 24;
di P 0; dþ
i P 0; x1 P 0; x2 P 0 and integers; i ¼ 1; 2:
224 O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225
The maximum degree of attainment of goal 2 is D�2 ¼ 1 with the optimal
integer solution
x2 ¼ ð2; 6Þ and d2 ¼ 1; dþ
2 ¼ 0:
Therefore, the optimal integer solution of the original integer linear goal pro-
gram is:
x� ¼ ð2; 6Þ;
D�1 ¼ 0 with d�
1 ¼ 0; dþ�1 ¼ 0;
D�2 ¼ 1 with d�
2 ¼ 1; dþ�2 ¼ 0
with the corresponding used Gomory cut: x2 6 7.
On the other hand, setting a = a� = 1, we get:
4 6 ~m1 6 6; 3 6 ~m2 6 5; 20 6 ~m3 6 22:
Choosing m� ¼ ðm�1; m�2; m�3Þ ¼ ð6; 5; 22Þ, then the optimal integer solution of the
original program has been found:
x� ¼ ð2; 4Þ;
D�1 ¼ 0 with d�
1 ¼ 0; dþ�1 ¼ 0;
D�2 ¼ 1 with d�
2 ¼ 1; dþ�2 ¼ 0
with the corresponding used Gomory cut: 3x1 + x2 6 10.
Remark. It should be noted that a systematic variation of a 2 [0, 1] will yield anew optimal integer solution to the integer linear goal program ðILGPÞ~m.
6. Conclusions
In this paper, an iterative goal programming approach has been proposed
for solving multiobjective integer linear programming problems. These prob-
lems involve fuzzy parameters in the right-hand side of the constraints. Thealgorithm presented in this paper has the advantage of dealing with many inte-
ger linear goal programs by varying the parameters in the right-hand side of the
constraints. Certainly, there are many other points for future research in the
area of fuzzy integer linear goal programs and should be studied. Some of these
points are:
(i) A study is needed to solve fuzzy integer linear goal programs with different
a-level sets of the fuzzy parameters, i.e., the solutions obtained fordifferent values of a should be compared and saved in any step of the
algorithm.
O.M. Saad / Appl. Math. Comput. 170 (2005) 216–225 225
(ii) A solution algorithm is required to treat integer linear goal programs with
fuzzy aspiration levels of the objectives.
(iii) An algorithm should be carried out to solve integer linear goal programs
involving fuzzy parameters in the left-hand side of the constraints.
References
[1] M. Balinski, An algorithm for finding all vertices of convex polyhedral sets, Journal of the
Society of Industrial and Applied Mathematics 1 (1961) 72–88.
[2] A. Charnes, W.W. Cooper, Management Models and Industrial Applications of Linear
Programming, Wiley, New York, 1961.
[3] J.P. Dauer, R.J. Krueger, An iterative approach to goal programming, Operational Research
Quarterly 28 (1977) 671–681.
[4] D. Dubois, H. Prade, Fuzzy sets and systems, in: Theory and Applications, Academic Press,
New York, 1980.
[5] J.B. Hughes, O.M. Saad, Stability sets for integer linear multiobjective and goal programming,
Paper presented at EuroX, The 10th European Conference on Operational Research, Belgrade,
Yugoslavia, June 27–30, 1989.
[6] J.P. Ignizio, GP-GN: An approach to certain large scale multiobjective integer programming
models, Large Scale Systems 4 (1983) 177–188.
[7] J.P. Ignizio, The determination of a subset of efficient solutions via goal programming,
Computers and Operations Research 8 (1981) 9–16.
[8] J.P. Ignizio, Goal Programming and Extensions, D. C. Health, Lexington Books, Lexington,
MA, 1976.
[9] D. Klein, S. Holm, Integer programming post-optimal analysis with cutting-planes, Manage-
ment Science 25 (1) (1979) 64–72.
[10] S.M. Lee, R.L. Morris, Integer goal programming methods, TIMS Studies in the Management
Sciences 6 (1977) 273–289.
[11] M.S. Osman, M.M. Awad, O.M. Saad, Stability in multiobjective integer programming,
AMSE Review 9 (3) (1989) 13–22.
[12] O.M. Saad, W.H. Sharif, Stability set for integer linear goal programming, Applied
Mathematics and Computation 153 (3) (2004) 743–750.
[13] M. Sasaki, M. Gen, A method for solving fuzzy multiobjective decision making problems by
interactive sequential goal programming, Transactions of the Institute of Electronics,
Information and Communication Engineers J75-A (10) (1992) 1590–1995.
[14] M. Sasaki, M. Gen, K. Ida, Interactive sequential fuzzy goal programming, Computers and
Industrial Engineering 19 (1–4) (1990) 567–571.
[15] R. Steuer, in: S. Zionts (Ed.), Vector-Maximum Gradient Cone Contraction Techniques,
Multiple Criteria Problem Solving, Springer-Verlag, Berlin, 1978.
[16] H.A. Taha, Integer Programming: Theory, Applications, and Computations, Academic Press,
1975.
[17] L. Zadeh, R. Bellman, Decision making in a fuzzy environment, Management Sciences 17
(1970) 141–164.
[18] H.J. Zimmermann, Fuzzy programming and linear programming with several objective
functions, Fuzzy Sets and Systems 1 (1978) 45–55.