Institute for Advanced Management Systems Research Department of Information Technologies Åbo Akademi University An Introduction to Fuzzy Linear Programs - Tutorial Robert Fullér Directory • Table of Contents • Begin Article c 2010 October 2, 2010 rfuller@abo.fi Table of Contents 1. Fuzzy programming versus goal programming 2. Fuzzy numbers 3. Measures of possibility 4. Zimmermann’s approach 5. Fuzzy linear programming with fuzzy number coefficients 3 1. Fuzzy programming versus goal programming Consider the following simple linear program x → min, subject to x ≥ 1, x ∈ R What if the decision maker’s aspiration level (or goal) is b0 = 0.5? The goal is set outside of the conceivable values of the objective function under given constraint. The linear inequality system x ≤ 0.5 does not have any solution. Toc JJ II x≥1 J I Back J Doc Doc I Section 1: Fuzzy programming versus goal programming 4 In goal programming we are searching for a solution from the decision set, which minimizes the distance between the goal and the decision set. That is, |x − 0.5| → min, subject to x ≥ 1, x ∈ R The unique solution is x∗ = 1. In fuzzy programming we are searching for a solution that might not even belong to the decision set, and which simultaneously minimizes the (fuzzy) distance between the decision set and the goal. We want to be as close as possible to the goal and to the constraints. Depending on the definition of closeness, the fuzzy version can turn into the following singleobjective problem max min{|x − 0.5|, |x − 1|}, subject to 1/2 ≤ x ≤ 1. The unique solution is x∗ = 0.75. Toc JJ II J I Back J Doc Doc I Section 1: Fuzzy programming versus goal programming 5 Figure 1: Illustration of the optimal solution. By using the minimum operator to aggregate the fuzzy statements ’x is close to 0.5’ and ’x is close to 1’ we get that the optimal solution is x∗ = 0.75. The fuzzy problem is to find an x ∈ R such that, ’x is as close as pos. to 0.5’ and ’x is as close as pos. to 1’ If we use the minimum operator to aggregate the objective functions then we have the following single-objective problem (FigToc JJ II J I Back J Doc Doc I Section 1: Fuzzy programming versus goal programming 6 ure 1) |x − 0.5| |x − 1| , subject to 1/2 ≤ x ≤ 1. max min 1− , 1− 1/2 1/2 Which turns into, max min{|x − 0.5|, |x − 1|}, subject to 1/2 ≤ x ≤ 1 which has a unique optimal solution x∗ = 0.75. We will consider the following two-variable linear program x1 + x2 → max subject to x1 , x2 ≤ 1 x1 , x2 ∈ R, What if the decision maker’s aspiration level is b0 = 3? Toc JJ II J I Back J Doc Doc I Section 1: Fuzzy programming versus goal programming 7 The aspiration level can not be reached since the maximal value of the objective function is equal to two? Figure 2: A simple LP. The unique optimal solution is (1, 1). Toc JJ II J I Back J Doc Doc I Section 1: Fuzzy programming versus goal programming 8 Figure 3: The desired value of the objective function is set to 3. Toc JJ II J I Back J Doc Doc I 9 2. Fuzzy numbers A fuzzy set A of the real line R is defined by its membership function (denoted also by A) A : R → [0, 1]. If x ∈ R then A(x) is interpreted as the degree of membership of x in A. A fuzzy set in R is called normal if there exists an x ∈ R such that A(x) = 1. A fuzzy set in R is said to be convex if A is unimodal (as a function). A fuzzy number A is a fuzzy set of the real line with a normal, (fuzzy) convex and continuous membership function of bounded support. Toc JJ II J I Back J Doc Doc I Section 2: Fuzzy numbers 10 Definition 2.1. A fuzzy set A is called triangular fuzzy number with peak (or center) a, left width α > 0 and right width β > 0 if its membership function has the following form a−t 1− if a − α ≤ t ≤ a α t−a A(t) = 1 − if a ≤ t ≤ a + β β 0 otherwise and we use the notation A = (a, α, β). The support of A is (a − α, b + β). A triangular fuzzy number with center a may be seen as a fuzzy quantity ”x is close to a”or”x is approximately equal to a”. Toc JJ II J I Back J Doc Doc I If A is not a fuzzy number then there exists an γ ∈ [0, 1] such that [A] is not a convex Section 2: Fuzzysubset numbersof R. 11 1 a-! a a+" Fig. 1.2. Triangular fuzzy number. Figure 4: A triangular fuzzy number. Definition 1.1.42.2. A fuzzy set A called fuzzy number peak Definition A fuzzy setis of the triangular real line given by the with mem(or center) a, left width α > 0 and right width β > 0 if its membership bership function function has the following form |a − t| a−t 1 − if |a − t| ≤ α, A(t) = (1) 1 − αα if a − α ≤ t ≤ a otherwise, t−a A(t) = 0 1− if a ≤ t ≤ a + β β a symmetrical triangular fuzzy Where α > 0 will be called otherwise JJ II 0 J I J Doc Doc I Toc Back Section 2: Fuzzy numbers 12 number with center a ∈ R and width 2α and we shall refer to it by the pair (a, α). Let A = (a, α) and B = (b, β) be two fuzzy numbers of the form (1), λ ∈ R. Then we have A + B = (a + b, α + β), λA = (λa, |λ|α). (2) Which can be interpreted as ”x is approximately equal to a” + ”y is approximately equal to b” = ”x + y is approximately equal to a + b” Toc JJ II J I Back J Doc Doc I Section 2: Fuzzy numbers 13 Figure 5: A = (a, 1), B = (b, 2), A + B = (a + b, 3) Definition 2.3. A fuzzy set A is called trapezoidal fuzzy number with tolerance interval [a, b], left width α and right width β if Toc JJ II J I Back J Doc Doc I Section 2: Fuzzy numbers 14 its membership function has the following form a−t 1− if a − α ≤ t ≤ a α 1 if a ≤ t ≤ b A(t) = t−b 1− if a ≤ t ≤ b + β β 0 otherwise and we use the notation A = (a, b, α, β). (3) The support of A is (a − α, b + β). A trapezoidal fuzzy number may be seen as a fuzzy quantity ”x is approximately in the interval [a, b]”. Toc JJ II J I Back J Doc Doc I The support of A is (a − α, b + β). 15 1 a-! a b b+" Fig. 1.3. Trapezoidal fuzzy number. Figure 6: Trapezoidal fuzzy number. A trapezoidal fuzzy number may be seen as a fuzzy quantity 3. Measures of possibility ”x is approximately in the interval [a, b]”. Definition 1.1.6 Any fuzzy number A ∈ F can be as distribuFuzzy numbers can also be considered asdescribed possibility % number & tions. If A ∈ F is afuzzy and x ∈ R a real number a−t if tdegree ∈ [a − α,ofa]possibility of the L then A(x) can be interpreted as the α statement ”x is A”. 1 if t ∈ [a, b] % & A(t) = tJ− b) JJ II I J Doc Doc I Toc Back Section 3: Measures of possibility 16 Let A, B ∈ F be fuzzy numbers. The degree of possibility that the proposition ”A is less than or equal to B” is true denoted by Pos[A ≤ B] and defined by the extension principle as Pos[A ≤ B] = sup min{A(x), B(y)}, (4) x≤y In a similar way, the degree of possibility that the proposition ”A is greater than or equal to B” is true, denoted by Pos[A ≥ B], is defined by Toc JJ II J I Back J Doc Doc I Finally, the degree of possibility that the proposition is true ”A is equal to Section 3: Measures of possibility A 17 Pos[A!B]=1 a B b Fig. 1.13. Pos[A ≤ B] = 1, because a ≤ b. Figure 7: Pos[A ≤ B] = 1, because a ≤ b. B” and denoted by Pos[A = B], is defined by Pos[A = B] = sup min{A(x), B(x)} = (A − B)(0), Pos[A ≥ B]x = sup min{A(x), B(y)}. (1.21) (5) x≥y Let A = (a, α) and B = (b, β) fuzzy numbers of symmetric triangular form. It is easy to compute that, Let A = (a, α) and B = (b, β) fuzzy numbers of symmetric if athat, ≤b to1 compute triangular form. It is easy a−b Pos[A ≤ B] = 1 − (1.22) otherwise α+β 0 if a ≥ b + α + β II I ”A Back J Doc Doc I The Toc degree ofJJ necessity that theJproposition is less than or equal to B” The degree of necessity that the proposition ”A is less than or equal to B” Section 3: Measures of possibility 18 B A Pos[A!B] b a Fig. 1.14. Pos[A ≤ B] < 1, because a > b. Figure 8: Pos[A ≤ B] < 1, because a > b. is true, denoted by Pos[A ≤ B], is defined as Nes[A ≤ B] = 1 − Pos[A ≥ B]. 1 if a ≤ b If A = (a, α) and B = (b, β) are fuzzyanumbers − b of symmetric triangular form (6) Pos[A ≤ B] = 1− otherwise α + β 0 if a ≥ b + α + β Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 19 Pos[A = (a, α) ≤ B = (b, β)] = 1 ⇐⇒ a ≤ b. Figure 9: Pos[A = (a, α) ≤ B = (b, β)] = 1 ⇐⇒ a ≤ b Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 20 Figure 10: The left-hand side of B does not play any role in computing Pos[A ≤ B]. Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 21 Figure 11: The right-hand side of B does matter as to the value of Pos[A ≤ B] if a > b. Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 22 1 if a ≥ b b−a Pos[A ≥ B] = 1− otherwise α + β 0 if a ≤ b − (α + β) (7) The degree of possibility that the proposition ”A is equal to B” is true, denoted by Pos[A = B], is defined by Pos[A = B] = sup min{A(x), B(x)}, x Let x ∈ R and let A be a fuzzy number. The degree of possibility of the statement that ’A takes value x’ is defined by A(x). Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 23 Figure 12: Pos[A = (a, α) ≥ B = (b, β)] = 1 ⇐⇒ a ≥ b Let A = (a, α) be a fuzzy number of symmetric triangular form Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 24 Figure 13: The right-hand side of B does not matter as to the value of Pos[A ≥ B]. Toc JJ II J I Back J Doc Doc I Section 3: Measures of possibility 25 and let x ∈ R. Then, Pos[A ≤ x] = Pos[A ≥ x] = Toc JJ 1 if a ≤ x A(x) = 1 − 0 1 A(x) = 1 − 0 II J I a−x otherwise α if a ≥ x + α if a ≥ x x−a otherwise α if x ≤ a − α Back J Doc Doc I Section 3: Measures of possibility 26 Figure 14: Pos[A ≥ x]. Toc JJ II J I Back J Doc Doc I 27 4. Zimmermann’s approach The conventional model of linear programming (LP) can be stated as ha0 , xi → min; subject to Ax ≤ b. In many real-world problems instead of minimization of the objective function ha0 , xi it may be sufficient to determine an x such that a01 x1 + · · · + a0n xn ≤ b0 ; subject to Ax ≤ b. (8) where b0 is a predetermined aspiration level. Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 28 Suppose that the decision maker’s aspiration level is set below the optimal value of the objective function. In this case linear inequality system (8) does not have a solution. Then we use flexible constraints. We state the fuzzy linear programming problem as follows (Zimmermann [1]) Find an x∗ ∈ Rn such that it satisfies the following inequalities as much as possible (in the sense we defined above) a01 x1 + · · · + a0n xn ≤ (b0 , d0 ) a11 x1 + · · · + a1n xn ≤ (b1 , d1 ) .. . am1 x1 + · · · + amn xn ≤ (bm , dm ) Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 29 Figure 15: Illustration of constraint satisfacion hai , xi ≤ (bi , di ). Let us µi (x) denote the degree of satsifaction of the i-th con- Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 30 straint by x ∈ Rn . Then we have, 1 if hai , xi ≤ bi , hai , xi − bi µi (x) = 1− if bi < hai , xi ≤ bi + di , di 0 if hai , xi > bi + di , for i = 0, 1, . . . , m. • if for an x ∈ Rn the value of hai , xi is less or equal than bi then x satisfies the i-th constraint with the maximal conceivable degree one; • if bi < hai , xi < bi + di then x is not feasible in classical sense, but the decision maker can still tolerate the violation of the crisp constraint, and accept x as a solution with a positive degree, however, the bigger the violation Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 31 the less is the degree of acceptance; • if hai , xi > bi + di then the violation of the i-th costraint is untolerable by the decision maker, that is, µi (x) = 0. Then the (fuzzy) solution of the FLP problem is defined as a fuzzy set on Rn whose membership function is given by µ(x) = min{µ0 (x), µ1 (x), . . . , µm (x)}, In this setup µ(x) denotes the degree to which all inequalities are satisfied at point x ∈ Rn . The maximizing solution x∗ of the FLP problem satisfies the equation Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 32 λ∗ = µ(x∗ ) = max µ(x). x That is, λ∗ denotes the maximal degree to which both the aspiration level and the constraints can be satisfied. max λ µ0 (x) ≥ λ µ1 (x) ≥ λ ... µm (x) ≥ λ 0 ≤ λ ≤ 1, x ∈ Rn . In this case we get the following LP problem Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 33 max λ λd0 + ha0 , xi ≤ b0 + d0 , λd1 + ha1 , xi ≤ b1 + d1 , ······ λdm + ham , xi ≤ bm + dm , 0 ≤ λ ≤ 1, x ∈ Rn . For example, let the optimal value of LP problem ha0 , xi → min; subject to Ax ≤ b be equal to 2, that is, ha0 , x∗ i = 2, and let the aspiration level be 1 and let all the tolerance leveles be equal to one. Then our FLP problem turns into the following LP Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 34 max λ λ + ha0 , xi ≤ 1 + 1, λ + ha1 , xi ≤ b1 + 1, ······ λ + ham , xi ≤ bm + 1 0 ≤ λ ≤ 1, x ∈ Rn . It is clear that the solution of this LP is different from the solution of the original LP since for x = x∗ we get λ = 0. Really, from λ + 2 ≤ 2 and 0 ≤ λ ≤ 1 it follows that λ = 0. Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 35 Let us consider the problem x1 + x2 → min subject to x1 ≥ 1 x2 ≥ 1 x1 , x2 ∈ R, Let b0 = 1 and let d0 = d1 = d2 = 1. The aspiration level is less than 2, the value we can reach under given constraints. Then we get the following fuzzy linear programming problem: Find an x∗ ∈ R such that Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 36 Figure 16: The optimization problem: unreachable goal (we are in the red). Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 37 x1 + x2 ≤ (1, 1) subject to x1 ≥ (1, 1) x2 ≥ (1, 1) x1 , x2 ∈ R Then we have µ0 (x) = Toc JJ 1 1− 0 II if x1 + x2 ≤ 1, x1 + x2 − 1 otherwise 1 if x1 + x2 > 2, J I Back J Doc Doc I Section 4: Zimmermann’s approach 38 Figure 17: Degrees of satisfactions of the goal x1 + x2 ≤ (1, 1) by the values of the objective function x1 + x2 . if x1 + x2 ≤ 1, 1 2 − (x1 + x2 ) otherwise µ0 (x) = 0 if x1 + x2 > 2, Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 39 Figure 18: Degrees of satisfactions of the first constraint x1 ≥ (1, 1) by the values of x1 . µ1 (x1 , x2 ) = Toc JJ II 1 1− 0 J if x1 ≥ 1, 1 − x1 otherwise 1 if x1 < 0, I Back J Doc Doc I Section 4: Zimmermann’s approach 40 1 if x1 ≥ 1, x1 otherwise µ1 (x1 , x2 ) = 0 if x1 < 0, 1 if x2 ≥ 1, x2 otherwise µ2 (x1 , x2 ) = 0 if x2 < 0, The fuzzy solution is defined by µ(x1 , x2 ) = min{µ0 (x1 , x2 ), µ1 (x1 , x2 ), µ2 (x1 , x2 )} and the maximizing solution is obtained from Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 41 λ∗ = max min{µ0 (x1 , x2 ), µ1 (x1 , x2 ), µ2 (x1 , x2 )}. x1 ,x2 Figure 19: Degrees of satisfactions of the goal x1 + x2 ≤ (1, 1) by the values of the objective function x1 + x2 . Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 42 Figure 20: Degrees of satisfactions of the first constraint x1 ≥ (1, 1) by the values of x1 . That is, Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 43 max λ 2 − (x1 + x2 ) ≥ λ x1 ≥ λ x2 ≥ λ 0 ≤ λ ≤ 1, x1 , x2 ∈ R. The optimal solutions are x∗1 = x∗2 = λ∗ = 2/3. Toc JJ II J I Back J Doc Doc I Section 4: Zimmermann’s approach 44 Figure 21: Illustration of the solution. Toc JJ II J I Back J Doc Doc I 45 5. Fuzzy linear programming with fuzzy number coefficients Assume that all parameters in (8) are fuzzy numbers of symmetric triangular form. Then the following flexible (or fuzzy) linear programming (FLP) problem can be obtained by replacing crisp parameters aij , bi with symmetric triangular fuzzy numbers ãij = (aij , α) and b̃i = (bi , di ) respectively, (a01 , α)x1 + · · · + (a0n , α)xn ≤ (b0 , d0 ) (a11 , α)x1 + · · · + (a1n , α)xn ≤ (b1 , d1 ) .. . (am1 , α)x1 + · · · + (amn , α)xn ≤ (bm , dm ) (9) Here d0 and di are interpreted as the tolerance levels for the Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 46 objective function and the i-th constraint, respectively. We denote by µi (x) the degree of satisfaction of the i-th restriction at the point x ∈ Rn in (9), i.e. µi (x) = Pos(ãi1 x1 + · · · + ãin xn ≤ b̃i ). Especially, µ0 (x) denotes the degree of satisfaction of the aspiration level of the decision maker at point x ∈ Rn . µ0 (x) = Pos(ã01 x1 + · · · + ã0n xn ≤ b̃0 ). Then the (fuzzy) solution of the FLP problem (9) is defined as a fuzzy set on Rn whose membership function is given by Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 47 µ(x) = min{µ0 (x), µ1 (x), . . . , µm (x)}, In this setup µ(x) denotes the degree to which all inequalities are satisfied at point x ∈ Rn . The maximizing solution x∗ of the FLP problem (9) satisfies the equation λ∗ = µ(x∗ ) = max µ(x). x That is, λ∗ denotes the maximal degree to which both the aspiration level and the constraints can be satisfied. From (6) it follows that the degree of satisfaction of the i-th restriction at x in (9) is the following: Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 48 1 if hai , xi ≤ bi , hai , xi − bi µi (x) = 1− otherwise, α|x|1 + di 0 if hai , xi > bi + α|x|1 + di , where |x|1 = |x1 | + · · · + |xn | and hai , xi = ai1 x1 + · · · + ain xn , for i = 0, 1, . . . , m. To find a maximizing solution to FLP problem (9) we have to solve the following nonlinear programming problem Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 49 λ + h0, xi → max ha0 , xi − b0 1− ≥λ α|x|1 + d0 ha1 , xi − b1 ≥λ α|x|1 + d1 ······ ham , xi − bm 1− ≥λ α|x|1 + dm 1− 0 ≤ λ ≤ 1, x ∈ Rn . That is, Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 50 max λ λ(α|x|1 + d0 ) − α|x|1 + ha0 , xi ≤ b0 + d0 , λ(α|x|1 + d1 ) − α|x|1 + ha1 , xi ≤ b1 + d1 , ······ λ(α|x|1 + dm ) − α|x|1 + ham , xi ≤ bm + dm , 0 ≤ λ ≤ 1, x ∈ Rn . Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 51 Example 5.1. As a very simple example consider the following LP x → min, subject to x ≥ 1, x ∈ R that is, 1 × x → min subject to 1 × x ≥ 1, x ∈ R, with a unique solution x∗ = 1. (10) Let us set the aspiration level to 0.5 and derive from it the following FLP: Find an x∗ ∈ R from the possibilistic inequality system Toc JJ II (1, α)x ≤ (0.5, d0 ) (1, α)x ≥ (1, d1 ) J I Back (11) J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 52 which satisfies both inequalities as much as possible. If α = 0.5, d0 = 0.5 and d1 = 0.5 we get the following statement: Find an x∗ ∈ R from the possibilistic inequality system (1, 0.5)x ≤ (0.5, 0.5) (1, 0.5)x ≥ (1, 0.5) which satisfies both inequalities as much as possible. After the multiplication by scalar we get the problem: Find an x∗ ∈ R from the possibilistic inequality system Inequality 1. (x, 0.5 × |x|) ≤ (0.5, 0.5) Inequality 2. (x, 0.5 × |x|) ≥ (1, 0.5) which satisfies both inequalities as much as possible. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 53 Figure 22: Explanation of the fuzzified aspiration level ’the value of the objective should be less then 0.5’ Let us consider the case when x = 1. Then we get Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 54 Figure 23: Explanation of the fuzzified constraint, Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 55 Figure 24: The goal and the constraint are in a trade-off relation. Inequality 1. (1, 0.5) ≤ (0.5, 0.5) Inequality 2. (1, 0.5) ≥ (1, 0.5) Inequality 1. is satisfied with degree 0.5 since µ0 (1) = 0.5 and Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 56 the second one is satisfied with degree one since µ1 (2) = 1. That is, x = 1 can be considered as a solution to the FLP problem with degree min{0.5, 1} = 0.5. Figure 25: The situation for x = 1. Let us consider the case when x = 2. Then we get Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 57 Figure 26: The situation for x = 2. Inequality 1. (2, 1) ≤ (0.5, 0.5) Inequality 2. (2, 1) ≥ (1, 0.5) Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 58 Inequality 1. is satisfied with degree 0 since µ0 (2) = 0 and the second one is satisfied with degree one since µ1 (1) = 1. Figure 27: The situation for x = 0.75. Using (6) we get that the degree of satisfaction of the first inToc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 59 equality at point x is equal to 1 if x ≤ 0.5 x − 0.5 µ0 (x) = 1− if 0.5 ≤ x ≤ 1 0.5|x| + 0.5 0 if x > 1 Using (6) we get that the degree of satisfaction of the second inequality at point x is equal to Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 60 1 if −x ≤ −1 −x+1 µ1 (x) = 1− if −1 < −x ≤ −0.5 0.5|x| + 0.5 0 if −x > −0.5 The fuzzy solution is µ(x) = min{µ0 (x), µ1 (x)} and the maximizing solution is obtained from λ∗ = max µ(x) = max min{µ0 (x), µ1 (x)} x Toc JJ x II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 61 That is, λ → max x − 0.5 1− ≥λ 0.5|x| + 0.5 −x+1 1− ≥λ 0.5|x| + 0.5 0 ≤ λ ≤ 1, x ∈ R. That is, the uniqe maximizing solution of (11) is then obtained from the equation 1− −x+1 x − 0.5 =1− 0.5|x| + 0.5 0.5|x| + 0.5 where x∗ = 0.75 and the degree of consistency is Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 62 Figure 28: µ0 and µ1 . λ∗ = 1 − Toc JJ II 0.75 − 0.5 = 5/7 0.5 × 0.75 + 0.5 J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 63 The degree of consistency is smaller than one, because the aspiration level, b0 = 0.5, is set below 1, the minimal value of the crisp goal function, 1 × x under the crisp constraint x ≥ 1. Figure 29: Illustration of the optimal solution. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 64 Example 5.2. Consider the following LP x1 + x2 → max subject to x1 ≤ 1 x2 ≤ 1 x1 , x2 ∈ R, that is, x1 + x2 → max subject to x1 + 0 × x2 ≤ 1 0 × x1 + x2 ≤ 1 x1 , x2 ∈ R, with a unique solution x∗1 = x∗2 = 1 and x∗1 + x∗2 = 2. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 65 Figure 30: An illustration of example 2. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 66 Suppose that the decision maker sets the desired value of the objective function to 3. Then we get the following fuzzy linear program (1, α)x1 + (1, α)x2 ≥ (3, d0 ) (1, α)x1 + (0, α)x2 ≤ (1, d1 ) (0, α)x1 + (1, α)x2 ≤ (1, d2 ) x1 , x2 ∈ R, with α = 0.5, d0 = d1 = d2 = 1. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 67 That is, (x1 , 0.5|x1 |) + (x2 , 0.5|x2 |) ≥ (3, 1) (x1 , 0.5|x1 |) + (0, 0.5|x2 |) ≤ (1, 1) (0, 0.5|x1 |) + (x2 , 0.5|x2 |) ≤ (1, 1) x1 , x2 ∈ R, After the addtions and multiplications by scalar we get (x1 + x2 , 0.5(|x1 | + |x2 |)) ≥ (3, 1) (x1 , 0.5(|x1 | + |x2 |)) ≤ (1, 1) (x2 , 0.5(|x1 | + |x2 |)) ≤ (1, 1) x1 , x2 ∈ R, Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 68 Figure 31: The fuzzified goal: if x1 + x2 ≥ 3 then the goal is satisfied with degree 1. Let us consider the situation when x1 = x2 = 1. (2, 1) ≥ (3, 1) (1, 1) ≤ (1, 1) Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 69 Figure 32: The fuzzified constraint: if x1 ≤ 1 then the first constraint is satisfied with degree 1. Using (6-7) we get Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 70 Figure 33: The value of the objective function for x1 = x2 = 1. µ0 (x1 , x2 ) = 1 if x1 + x2 ≥ 3 3 − (x1 + x2 ) otherwise 1− 1 + 0.5(|x1 | + |x2 |) 0 if x1 + x2 ≤ 3 − (1 + 0.5(|x1 | + |x2 |)) Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 71 µ1 (x1 , x2 ) = 1 if x1 ≤ 1 x1 − 1 otherwise 1− 1 + 0.5(|x1 | + |x2 |) 0 if x1 > 1 + (1 + 0.5(|x1 | + |x2 |)) µ2 (x1 , x2 ) = 1 if x2 ≤ 1 x2 − 1 otherwise 1− 1 + 0.5(|x 1 | + |x2 |) 0 if x2 > 1 + (1 + 0.5(|x1 | + |x2 |)) The fuzzy solution is defined by Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 72 µ(x1 , x2 ) = min{µ0 (x1 , x2 ), µ1 (x1 , x2 ), µ2 (x1 , x2 )} and the maximizing solution is obtained from λ∗ = max min{µ0 (x1 , x2 ), µ1 (x1 , x2 ), µ2 (x1 , x2 )}. x1 ,x2 That is, max λ µ0 (x) ≥ λ µ1 (x) ≥ λ µ2 (x) ≥ λ 0 ≤ λ ≤ 1, x ∈ R2 . Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 73 To find a maximizing solution to FLP problem we have to solve the following nonlinear programming problem max λ 3 − (x1 + x2 ) 1− ≥λ 1 + 0.5(|x1 | + |x2 |) 1− x1 − 1 ≥λ 1 + 0.5(|x1 | + |x2 |) 1− x2 − 1 ≥λ 1 + 0.5(|x1 | + |x2 |) 0 ≤ λ ≤ 1, x1 , x2 ∈ R. The unique solution is x∗ = (4/3, 4/3) and λ∗ = 6/7. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 74 Figure 34: The optimal value of the fuzzy objective function. (x1 + x2 , 0.5(|x1 | + |x2 |)) ≥ (3, 1) (x1 , 0.5(|x1 | + |x2 |)) ≤ (1, 1) (x2 , 0.5(|x1 | + |x2 |)) ≤ (1, 1) Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 75 Figure 35: The optimal value of fuzzy objective and its relation to the fuzzified goal. (4/3 + 4/3, 0.5(4/3 + 4/3)) ≥ (3, 1) (4/3, 0.5(4/3 + 4/3)) ≤ (1, 1) Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 76 Figure 36: Constraint satisfaction. References [1] H.-J. Zimmermann, Fuzzy programming and linear programming with several objective functions, Fuzzy Sets and Systems, 1(1978) 45-55. 28 Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 77 Figure 37: An illustration of example 2. Toc JJ II J I Back J Doc Doc I Section 5: Fuzzy linear programming with fuzzy number coefficients 78 Figure 38: An illustration of the functions µ0 , µ1 and µ2 . Toc JJ II J I Back J Doc Doc I