Penalty Function Method For Solving Fuzzy Nonlinear Programming Problem

Penalty Function Method For Solving Fuzzy Nonlinear Programming Problem

• 1.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Abstract: In this work, the fuzzy nonlinear programming problem (FNLPP) has been developed and their result have also discussed. The numerical solutions of crisp problems and have been compared and the fuzzy solution and its effectiveness have also been presented and discussed. The penalty function method has been developed and mixed with Nelder and Mend’s algorithm of direct optimization problem solutionhave been used together to solve this FNLPP. Keyword: Fuzzy set theory, fuzzy numbers, decision making, nonlinear programming, Nelder and Mend’s algorithm, penalty function method. Fuzzy nonlinear programming problem (FNLPP) is useful in solving problems which are difficult, impossible to solve due to the imprecise, subjective nature of the problem formulation or have an accurate solution. In this paper, we will discuss the concepts of fuzzy decision making introduced by [2] and the maximum decision [20] that is used in NLPP to find the optimal decision (solution). This decision making was used in fuzzy linear and nonlinear programming problems [1],[8], [9] and [15]. Furthermore, these problemshave fuzzy objective function and fuzzy variables in the constraints [5], [10], [11] and[17] where the fuzzy left and right hand side coefficients on constraints [18]. In addition, the fuzzy NLPP is used in quadratic programming [6], [12] and [16] which hasa fuzzy multi objective function and fuzzy parameters on constraints so in our NLPP that have fuzzy properties on.However, the fuzzy nonlinear programming problem is not just an alternative or even a superior way of analyzing a given problem, it's useful in solving problems in which difficult or impossible to use due to the inherent qualitative imprecise or subjective nature of the problem formulation or to have an accurate solution.The outline of this study is as follows: In section two we introduce some important definitions that are useful in our problem. Section3 we state the general nonlinear programming problem in fuzzy environment by transforming the crisp problem into the fuzzy problem. Section 4 we present and develop the regular penalty function method and mixed it with Nelder and Mend’s algorithm in order to solve FNLPP. Finally, in section 5, we show the efficiency of our study by present numerical example involving FNLPP. Page | 1 Penalty Function Method For Solving Fuzzy Nonlinear Programming Problem 1A.F.Jameel and 2Radhi.A.Z 1School of Mathematical Sciences, 11800 USM, University Science Malaysia, Penang, Malaysia 2Department of Mathematics, College of Science, the University of AlMustansiriyah Iraq Baghdad I. Introduction II. Primetimes Paper Publications 1. Fuzzy Set [20]: If x is a collection of objects denoted generally by X, then a fuzzy set A ~ in X is a set of order pairs:A ~  {(x, (x) A ~  ) | x X},where (x) A ~  : x [0, 1] is called the membership function or grade of membership (also degree of compatibility
• 2.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 2 α +   x (x) 1 0.5 0  }.More generally, suppose that we have n goals n 2 1 G ~  ,  , …,  }}  }  Min {  , Paper Publications or degree of truth) of x in A ~ which maps x to the membership range M (when M contains only the two points 0 and 1), A ~ is a nonfuzzy and (x) A ~  is identical to the characteristic function of crisp set. The range of membership function is a subset of the non-negative real numbers whose supremum is finite. Elements with a zero degree of membership are normally not listed. 2. Fuzzy Numbers with Linear Member Ship Function [3]: The function L: X [0, 1] is a function with two parameters defined as:  1,              ( ; , ) , 0, L x if x x if x if x           Where L is called the trapezoidal linear membership function. Fig. 1 L-Function. 3. Fuzzy Decision Making [2]: Assume that we are given a fuzzy goal (fuzzy objective function) G ~ and fuzzy constraints C ~ in a space of alternatives X. The G ~ and C ~ combine to form a decision,D ~ , which is a fuzzy set resulting from intersection of G ~ andC ~ . In symbols, D ~ C ~ is, correspondingly, the membership function of D ~ can be defined as:   Min { G ~ D ~  , C ~ ,..., G ~ , G ~ and m constraints ,..., C ~ , C ~ m 2 1 C ~ . The, the resultant decision is defined as: D ~  1 G ~  2 G ~  …  n G ~  1 C ~  2 C ~  …  m C ~ and correspondingly:   Min {min { D ~  , 1 G ~  , …, 2 G ~  }, min { n G ~ 1 C ~ 2 C ~ m C ~  Min {  , 1 G ~  , …, 2 G ~  , n G ~  , 1 C ~  , …, 2 C ~ m C ~ j G ~  } j C ~ for j  1, 2,…, n and i  1, 2, …, m. 4. Maximum Decision Maker [20]: If the decision-maker wants to have “crisp” decision proposal, it seems appropriate to suggest to him the divided which has the highest degree of membership in the fuzzy set “decision”. Let us call this the maximizing decision, defined by:
• 3.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 3 XmaxMaxM (x) MaxMin (x), (x) i j C ~ Paper Publications D ~ x D ~ x    , where j D  and i C  are in the definition (3) for i  1, 2, …, m; j 1, 2, …, n. III. Penalty Function Method There are survival methods used in fuzzy reliability problems, such as a novel approach for solving unconstrained model mechanical structure [7], also the numerical integration algorithm in the fuzzy general strength model [13]. The penalty method [19] belongs to the first attempts to solve constrained optimization problems satisfactorily. The basic idea is to construct a sequence of unconstrained optimization problems and solve them by standard minimize or maximize of the unconstrained problems converge to the solution of the constrained one. To simplify the notation, we consider the following two NLPP: (1) (2) i    To construct the unconstrained problems, so-called penalty terms are added to the objective function which penalizes f(x) whenever the feasibility region is left. A factor k controls the degree of penalizing f(x). Proceeding from a sequence {k} with k for k  0, 1… penalty function can be defined by [4]: 1  1. Min/Max (x,)  f(x) + k (3) for problem (1). 1  2. Min/Max (x, )  f(x) + k (4) for problem (2) The unconstrained nonlinear programming problems are solved by any standard technique, e.g., Nelder and Mead [14] method combined with a line search. However, the line search must be performed quite accurately due to this step, narrow valleys created by the penalty terms, respectively. The technique of solving a sequence of minimization (maximization) problems of by using a penalty function method is as follows: 1. Choose a sequence {k}. 2. For each finding a local minimizer (maximizer) x(k) say, .By any steeple optimization method. n (A)Min /Maxf (x) Subject to: g (x) ( )0, i  1,2,...,m andx  0,  x  R i n (B)Min /Maxf (x) Subject to: g (x) 0, i  1,2,...,m andx  0,  x  R 2 m i 1 2 (min(0,gi (x))) 2 m i 1 2 (gi (x)) k min(max) (x, k ) x x  
• 4.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org 1 1 m 3. Stop when the penalty terms ok, or k is zero and the constraints satisfies the solution at once Page | 4 Paper Publications when the penalty terms is zero. 4. The convergence of the solution of (NLPP) is using the penalty function method and its properties can be found in [4]. 5. Use this formula (3-4) in Nelderand Mead’s [14] algorithm for the direct solution of the optimization problem. IV. Fuzzy Nonlinear Programming Problem Consider the crisp NLPP below: Min/Max f(x) Subject to: n  j 1 aijx j  () bi, i  1, 2… m; j  1, 2… n (5) For all xj 0 and xj Rn. Now, the fuzzy version for problem (1) is as follows: x a ~ M / n i ~ M f(xj) Subject to: n  j 1 j ijx a ~  () i b ~ , i  1, 2, …, m; j  1, 2, …, n (6) For all xj 0 and xj Rn. Note that, the i-th constraint of problem (2) is called fuzzy technological constraint with fuzzy technological coefficients and fuzzy right hand side numbers, where ij a ~ is the fuzzy technological coefficient and i b ~ is the fuzzy right hand side numbers, where i1, 2, …, m and j1, 2, …, n.This problem can be solved using fuzzy decision making properties by using definition (3) and (4) in section II as follows: 1. To fuzzify the objective function, calculate the lower and the upper bounds z and zu receptively of the optimal values are obtained by solving the crisp NLPP as follows: z1 Min/Max f(xj) Subject to: n  j 1 aijx j  () bi (7)  xj 0, xj Rn and i  1, 2, …, m; j 1, 2, …, n. z2 Min/Max f(xj) Subject to: 2 2  i 1 2 (gi (x))
• 5.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 5 Paper Publications n  j 1 aijx j  () bi + pi (8)  xj 0, xj Rn and i  1, 2, …, m; j 1, 2, …, n. z3 Min/Max f(xj) Subject to:   n j 1 (aij dij )x j  () bi+ pi (9)  xj 0, xj Rn and i  1, 2, …, m; j 1, 2, …, n. and z4 Min/Max f(xj) Subject to:   n j 1 (aij dij )x j  () bi (10)  xj 0, xj Rn and i  1, 2, …, m; j 1, 2, …, n. Where, pi , dij>0 are any chosen constants represented the limits of bi and aij .Whenever the objective function takes the value between z1, z2, z3 and z4 by letting z min (z1, z2, z3, z4) and zu max (z1, z2, z3, z4). Let M ~ be the fuzzy set representing the objective function f(xj), such that: M ~  {(x, (x) M ~  ) | x  Rn} Where: if zu f x j 1, ( ) ( ) f x j z M ( x ) , if z f ( x j ) zu zu z if f x j z 0, ( )                     This represents the satisfaction of the aspiration level of the objective.Note that, pi and dij are any chosen constant represented the limits of bi and dij and can be found as follows:  ) x ( a ~ ij {(x, (x)  ) | x  R} ij a ~ Where: 1, ( ) , 0, if x aij aij dij x a x if aij x aij dij ij dij if x aij dij                   See Fig.1
• 6.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 6 (the degree of fyziness) a  d  x ij ij Fig. 2Membership function of aij μ (x) . b  p  x i i Paper Publications Also:  ) x ( b ~ i {(x, (x)  ) | x  R} i b ~ where: 1, ( ) , 0, if x bi bj pi x b x if bi x bi pi i pi if x bi pi                   See Fig.4. bi μ (x) . Fig.3The membership functionof  Note that if any ij a ~ and i b ~ is negative, say ij a ~ , then:  ij a ~  {(x, (x)  ) | x  R} ij a ~ where 1, ( ) ( ) ( ) , ( ) 0, if x aij dij x aij dij a x if aij dij x aij ij dij if x aij                        1 0 x μ (x) ij a ~ ij d aij dij aij+bij 1 0 bi pi pi+bi x μ (x) i b ~ i p (the degree of fyziness)
• 7.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 7 μ (x) x  (a  d ) ij ij d Fig. 4 Membership function of -aij μ (x). b , i  1, 2, …, m if b a x    0 (11) Paper Publications Similarly, use the same way in i b ~ . n  2.Now, to fuzzify the i-th constraints j 1 j ijx a ~  () i ~ Let i C ~ be the fuzzy set for i-th constraints, such that: i C ~ {(x, (x)  ) | x  Rn} i C ~  can be defined by (x) i C ~ 1 1 1 1 1 1 0, ( ) , ( ) 1, ( ) i n i ij j j n i ij j n n j C n ij j i ij ij j i j j ij j i j n i ij ij j i j b a x x if a x b a d x p d x p if b a d x p                                    Then we have the following crisp NLPP problem: Max  Subject to: g1 :  (x) M ~   0 g2 :  (x) 1 C ~  gm1 :  (x)   0 m C ~ Where x  0, 0  1 and x  Rn, which is equivalent to the problem (12): Min () Subject to: 1 0 x ij a ~  ij -(aij+dij) dij -aij
• 8.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 8 1 equal to zero, for any given points x11.5, x2 2, we have: Paper Publications g1 : f (x ) z    z z u j   0 g2 : b1 [(a11 + d11)x1 + … + (a1n + d1n)xn] p1 0 (12)  gm1 : bm [(am1 + dm1)x1 + … + (amn + dmn)xn] pm 0  xj (x1, x2, …, xn)  0, xj Rn and 0  1. The penalty function for problem (8) is given by: m  1 1 Min (xj, , )  +     i 1 2 Min(0,gi ) 2 , j 1, 2, …, n. V. Numerical Example Consider the following crisp nonlinear programming problem: Min z  x1 + x2 x1x2 Subject to: g1: 8x13x26 (13) g2: 3x1 + 6x2 4 x1, x2 0 The solution of the crisp problem by using the penalty function method: 1  Min (x1, x2, )  x1 + x2 x1x2 +    2 i 1 2 Min(0,gi ) 2 in algorithm (2.7.1), we get the following optimal results:  At 75104 the penalty term    2 i 1 2 Min(0,gi ) 2 * x1  0.16437888, and * x2  0.58449170, Therefore z* 0.6279249 which satisfy the constraints: g1 * *  2.931 and g2  8.684105 The fuzzy version of the above problems: n i ~ M x1 + x2 x1x2 z Subject to: g1: 8 ~ x13 ~ x26 ~ (14) g2: 3 ~ x1 + 6 ~ x24 ~ x1, x2.0
• 9.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 9 if x x x x z x x x x z M x if z x x x x zu zu z if x x x x zu    7 9 Paper Publications 1. Let M ~ be the fuzzy set of the objective function z, such that M ~ {(x, (x) M ~  ) | x  R}, where (x) M ~  can be defined as: 0, 1 2 1 2 ( ) 1 2 1 2 , 1 2 1 2 1, 1 2 1 2                             First, we find the chosen coefficients dij and pi then we obtain :    3 6 aij   8 3 1 2  4 3  , dij   , then aij + dij   7 1   4 bi   6   4 , pi   1   8 , then bi + pi   5 , where i  1, 2 and j  1, 2. and the lower and upper bounds z and zu for the optimal values can be found by solving four crisp NLPP as follows: z1 Min x1 + x2 x1x2 Subject to: g1: 8x13x26 (15) g2: 3x1 + 6x2 4 x1, x2 0. z1 have the same solution of z in (13). z2 Min x1 + x2 x1x2 Subject to: g1: 8x13x25 g2: 3x1 + 6x2 8 (16) x1, x2 0. Using algorithm in section IV, we get the following optimal results: 1  At  1107 the penalty term    2 i 1 2 Min(0,gi ) 2 equal to zero, we get the results: * x1  0.15383577 and * x2  1.25643698, Therefore z2 1.2169878 , * g1  2.9106 and * g2  1.291104 z3 Min x1 + x2 x1x2 Subject to: g1: 7x1 x25 (17) g2: 7x1 + 9x2 8
• 10.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org 1 equal to zero at  1105, so we get Page | 10 1 1, if 1.216980 x x x x x x x x 0.33426089 ,if 0.33426089 x x x x 1.21698780           0, if 6 8x 3x 6 ( 8x 3x )            (x) , if 8x 3x 6 7x 1x 1        Paper Publications x1, x2 0.  By using the algorithm method in section IV, the penalty term    2 i 1 2 Min(0,gi ) 2 the optimal results: * x1  0.43310236 and * x2  0.55203368. Therefore z30.74604895 that satisfies the constraintsg11.416 andg2 1.963103 Finally, z4 Min x1 + x2 x1x2 Subject to: g1 : 7x1 x26 (18) g2 : 7x1 + 9x2 4 x1, x2 0.  Also, by using the algorithm section IV, the penalty term    2 i 1 2 Min(0,gi ) 2 equal to zero at  1108, with the optimal results: * x1  0.42857293,and * x2  0.14166517, Therefore z4 0.33426089That satisfy the constraintsg1 3.674,and g2 3.999. Hence z4 Max {z1, z2, z3, z4}1.21698780 and z Min {z1, z2, z3, z4}  0.33426089. Therefore:        1.21698780 0.33426089                 0, if x x x x 0.33426089 (x) 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 M ~ 2-also let 1 C  be the fuzzy set for the first constraint g1, such that: 1 2 1 2 C1 1 2 1 2 1 2 1 2 4x 3x 4 1, if 4 7x 9x 4  And let 2 C  be the set for the second constraint g2, such that:
• 11.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 11        0, if 4 3x 6x 4 (3x 6x )         (x) , if 3x 6x 4 7x 9x 4        4x 3x 4 1, if 4 7x 9x 4 Paper Publications 2 1 2 1 2 C 1 2 1 2 1 2 1 2  1.By using fuzzy decision making as in problem (12), we have the following crisp NLPP: Max  Subject to: g1 :  (x) M ~   0 g2 :  (x)   0 1 C ~ g3 :  (x)   0 m C ~ Where x  0, 0  1 and x  R2, which is equivalent to the problem below: Min () Subject to: g1 : x1  x2  x1x2  0.33426098 1.2169878  0.33426098  0 g2 :    6x 8x 3x 1 2 2   x 2x 1 1 2  0 (19) 4  3x  6x 4x  3x  4 g3 : 1 2 1 2  0 x1, x2 0, 0  1 The penalty function of the above problem is: 1  Min (x1, x2, , )  +    3 i 1 2 Min(0,gi ) 2 Table I Result of problem (13)  x1 x2  10 1.5 2 0.5 100 0.84819258 0.06900019 0.34780614 1000 0.80510498 0.00681909 0.23532029 1104 0.80037352 0.00049138 0.22322137 1105 0.80228107 0.00009775 0.22153426 25104 0.79977446 00000580 0.22114719 5105 0.79977446 00000580 0.22114719 75105 0.79975981 0.00000066 0.22115274 1106 0.79975981 0.00000066 0.22115274 25105 0.79975981 0.00000066 0.22115274 5106 0.79975981 0.00000066 0.22115274 75105 0.79975981 0.00000066 0.22115274
• 12.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org Page | 12 1107 0.79975981 0.00000047 0.22115274 25106 0.79975981 0.00000047 0.22115274 5107 0.79975981 0.00000047 0.22115274 75106 0.79974898 0.00001033 0.2114567 1108 0.79974898 0.00001033 0.2114567 1 At  1108 the penalty term    Paper Publications  2 i 1 2 Min(0,gi ) 2 equal to zero, so we have the optimal solutions: * x1  0.79974898, * x2  0.00001033 and * 0.2114567, that satisfiesapproximately the constraints g1 3.061101, g2 6.42106 and g3 1.202103 and zAf 0.79975104, such that z< zAf< zu , wherezAfis the solution after the fuzziness . VI. Conclusion In this work, the a numerical method of fuzzy nonlinear programming problem is presented. Furthermore, it is proposed that the results solutionof fuzzy optimization is a generalization of the solution of the crisp optimization problem. In our work, the penalty function mixed with Nelder and Mend’s algorithm have been successfully employed to solve numerical problems in fuzzy environment with fuzzy objective function and fuzzy technological constraints. The numerical results of our proposed method satisfied the fuzzy set theory properties. REFERENCES [1]. F. J Ali 5and S. Amir, “Solving Nonlinear Programming Problem in Fuzzy Environment”, Int. J. Contemp. Math. Sciences, Vol. 7, no. 4, pp. 159 – 170, 2012. [2]. R. E. Bellman and L.A. Zadeh,” Decision-making in a fuzzy environment”, Management Sci, Vol. 17, No. 4, pp. 141- 164, 1970. [3]. D. Drinkov, H. Hellendoorn and M. Roinfrank, “An Introduction to fuzzy Control”, Norosa Publishing House, (1996). [4]. R. Fletcher, “Directional Methods of Optimization”, John Wiley and Sons, 1987. [5]. P. Iyengar, “Non-Linear Programming; Introduction”, IEOR, Handout 19, 16 October 2002. [6]. B. Kheirfamand F. Hasani, “Sensitivity analysis for fuzzy linear Programming problems with Fuzzy variables”, Advanced Model and Optimization, Vol. 12, No. 2, pp. 1878-1888, 2010. [7]. B. Li, Zhu. M and Xu. K,” A Practical Engineering Method for Fuzzy Reliability Analysis of Mechanical Structures”. Reliability Engineering and System Safety, Vol. 67, pp. 311–315, 2000. [8]. S. H. Nasseri, E. Ardil, A. Yazdani and R. Zaefarian, “Simplex method for solving linear programming problems with fuzzy numbers”, Transactions on Engineering, Computing and Technology, Vol. 10, pp. 284-288, 2005. [9]. S. H. Nasseri, “ Fuzzy nonlinear optimization, ” The Journal of Nonlinear Analysis and its Applications, Vol. 1, No. 4, pp. 230-235, 2008. [10]. J. Nayakand K.B. Sanjaya, ” Optimal Solution of Fuzzy Nonlinear Programming Problems with Linear Constraints”, International Journal of Advances in Science and Technology, Vol. 4, No. 4, pp. 43-52, 2012. [11]. V. Pandian, R. Nagarajan, and S. Yaacob, “Fuzzy Linear Programming: A Modern Tool for Decision Making”, JurnalTeknologi, Vol. 37, pp. 31-44, 2004. [12]. Z. Preitl, J. K. Tar and M. Takács, “Use of Multi-parametric Quadratic Programming in Fuzzy Control Systems”, ActaPolytechnicaHungarica, Vol. 3, No. 3, pp. 29-43, 2006. [13]. J. Qimiand C. Chun-Hsien,” A Numerical Algorithm of Fuzzy Reliability”, Reliability Engineering and System Safety, Vol. 80, pp. 299–307, 2003. [14]. S. F. Shu-Kai and Z. Erwie,” Simulation Optimization using An Enhanced Nelder-Mead Simplex Search Algorithm”. Proceedings of the Fifth Asia Pacific Industrial Engineering and Management Systems Conference 2004.
• 13.
  International Journal ofRecent Research in Mathematics Computer Science and Information Technology Vol. 1, Issue 1, pp: (1-13), Month: April - June 2014, Available at: www.paperpublications.org [15]. Y. Song, Y. Chen and X. Wu, “A Method for Solving Nonlinear Programming Models with All Fuzzy Coefficients [16]. H. C. Wu, “Duality Theory In Fuzzy Optimization Problems Formulated By The Wolfe's Primal And Dual Pair”, Fuzzy [17]. H. C. Wu, “Duality Theory in Fuzzy Optimization Problems”, Fuzzy Optimization and Decision Making, Vol. 3, pp. Page | 13 Based on Genetic Algorithm”, Advances in Natural Computation, Vol. 36, No. 11, pp 1101-1104, 2005. Optimization and Decision Making, Vol. 6, pp. 179-198, 2007. Paper Publications 345-365, 2004. [18]. M. Yenilmez, N. Rafail and K. Gasimor, Solving Fuzzy Liner Programming Problems with Linear Membership Function, Turk H. Math, TUBITAK, Vol. 26, pp. 375-369, 2002. [19]. M. Zhiqing, Qiying. H and Chuangyin. D, “A Penalty Function Algorithm with Objective Parameters For Nonlinear Mathematical Programming”, Journal Of Industrial And Management Optimization, Volume 5, Number 3, pp. 585–601, 2009. [20]. H.J. Zimmermann, “ Fuzzy Sets :Decision Making and Expert Systems”, Kluwer-Nijhoff Publishing, Boston (1987).