随机数

ELSEVIER An International Journal Available online at www.sciencedirect.com computers & .¢,=.=. ~o , . .=T . mathematics with applications Computers and Mathematics with Applications 51 (2006) 1127-1136 www.elsevier.com/locate/camwa Ranking Fuzzy Numbers with an Area Method using Radius of Gyrat ion YONG DENG* School of Electronics and Information Technology Shanghai J iao Tong University Shanghai 200030, P.R. China dengyong©sj tu. edu. cn ZHU ZHENFU National Defence Key Laboratory of Target and Env i ronment Feature Beijing 100854, P.R. China LIU QI Shanghai Inst i tutes for Biological Sciences Chinese Academy of Sciences Shanghai 200031, P.R. China (Received October 2003; revised and accepted November 2004) Abst ract - -Rank ing fuzzy numbers plays an very important role in linguistic decision making and some other fuzzy application systems. Many methods have been proposed to deal with ranking fuzzy numbers. Chu pointed out some shortcomings of the existing distance method and proposed to rank the fuzzy numbers with the area between the centroid point and original point. However, drawbacks are also found in the area method. For example, it cannot rank fuzzy numbers when some fuzzy numbers have the same centroid point. In this paper, we propose a modified area method to rank fuzzy numbers. The modified method can effectively rank various fuzzy numbers and their images. We also used some comparative examples to illustrate the advantage of the proposed method. @ 2006 Elsevier Ltd. All rights reserved. Keywords - -Rank ing , Fuzzy numbers, Area, Distance index, Radius of gyration. 1. INTRODUCTION The rank ing of fuzzy numbers has been a concern in fuzzy mul t ip le a t t r ibute dec is ion-mak ing (MADM) since its incept ion. More than 20 fuzzy rank ing indices have been proposed since 1976 [1,2]. Var ious techn iques are appl ied in the l i te ra ture that compare fuzzy numbers [3-18]. Some of *Author to whom all correspondence should be addressed. The authors would like to express their great thanks for the anonymous referees for their careful reading and valuable comments, which help us to make much improvements in the present version of the manuscript. This work was supported by National Natural Science Foundation of China Grant No. 30400067, Shanghai Nature Foundation Grant No. 03ZR14065, China Post-doctor Foundation Grant No. 200436273, National Defence Key Laboratory Grant No. 51476040105JW0301,51476040103JW13, Key Laboratory of ducation Ministry for Image Processing and Intelligent Control Grant No. TKLJ0410, and Aerospace Basic Support Foundation Grant No. 2004-JD-4. 0898-1221/06/$ - see front matter (~) 2006 Elsevier Ltd. All rights reserved. Typeset by .A.A/~-TF~ doi: 10.1016/j.camwa.2004.11.022 1128 Y. DENC et al. these ranking methods have been compared and reviewed by Bortolan and Degani [19]. Chen and Hwang [20] thoroughly reviewed the existing approaches, pointed out some illogical conditions that arise among them. In 1988, Lee and Li [21] ranked fuzzy numbers based on two different criteria, namely, the fuzzy mean and the fuzzy spread of fuzzy numbers. They pointed that human intuition would favor a fuzzy number with the following characteristics: higher mean value and at the same time lower spread. However, when higher mean value and at the same time higher spread or lower mean value and at the same time lower spread exists, it is not easy to compare its ordering clearly. Therefore, Cheng [22] proposed the coefficient of variance (CV index), i.e., CV = a (standard error)/l#l,/z ¢ 0, a > 0, to improve Lee and Li's ranking method [21]. Chu [23] pointed out the shortcomings of Cheng's method and suggested to rank fuzzy numbers with the area between the centroid point and the point of origin. It seems that the area-based method can effectively rank various fuzzy numbers with various attributes (normal/nonnormal, triangular/trapezoidal, and general). However, Chu's method still has some drawbacks, i.e., it cannot correctly rank fuzzy numbers in some situations (numerical examples are illustrated in Section 3). In this paper, we propose a modified area-based ranking approach. In Section 2, we introduce some preliminaries. In Section 3, we briefly review the distance-based method presented by Cheng [22] and area method presented by Chu [23]. In Section 4, we present a new area-based ranking method. In Section 5, we use some examples to show the advantage of the proposed method. The paper ends with conclusion in Section 6. 2. BACKGROUD INFORMATION In this section, we first introduce the basic concepts of fuzzy numbers and the centroid point of fuzzy numbers. Then, we describe the radius of gyration (ROG) method of fuzzy numbers, which plays an important role in our proposed ranking method. 2.1. Fuzzy Numbers A fuzzy number A = (a, b, c, d; w) is described as any fuzzy subset of the real line R with membership function fA which has the following properties [24]. (a) fA is a continuous mapping from R to the closed interval [0, w], 0 ~< w ~< 1. (b) :A(X) -~- 0 for all z E (--co, a]. (c) fA is strictly increasing on [a, b]. (d) :A(x) = ~., for all x e [b,e], where w is a constant and 0 < ~. < 1. (e) fa is strictly decreasing on [c, d]. (f) fA(x) = 0 for all x E [d, +co), where 0 ~< w ~< 1, a, b, c, and d are real numbers. If w = 1, then the generalized fuzzy number A is called a normal trapezoidal fuzzy number denote A = (a, b, c, d). If a = b and c = d, then A is called a crisp interval. If b = c, then A is called a generalized triangular fuzzy number. If a = b = c = d, then A is called a real number. The membership function fA of A can be expressed as { f~(x), ay Figure 2. A generalized trapezoidal fuzzy numbers (a, b, c, d; w). In a similar way, we define the radius of gyration of an area A with respect to the y axis is I v = r~A, ry -~ ~. (7) (8) In the following of this paper, when a generalized fuzzy number A is given, the radius of gyration r A A (ROG) points of the generalized fuzzy number A is denoted as ( x,ru ) whose value can be obtained by equations (6) and (8). For an area made up of a number of simple shapes, the moment of inertia of the entire area is the sum of the moments of inertia of each of the individual area about the axis desired. For example, the moment of inertia of the generalized trapezoidal fuzzy number in Figure 2 can be obtained as follows, (I~) = (I~)~ + (I~)2 + (I~)a, (9) (z~) = (~r~) 1 + (z~) 2 + (z~) 3 . ( lO) EXAMPLE 1. Determine the moment of inertia (with respect to x axis and to y axis) and the radius of gyration of the generalized trapezoidal fuzzy number (a, b, c, d; w), shown in Figure 2. First, the trapezoid (a, b ~, d, d) can be divided into three parts, abb ~, bb~c'c, and cdd. The moment of inertia of the area abb ~ with respect to x axis, and the moment of inertia of the area abb' with respect to y axis can be calculated, according to equations (5) and (7), respectively, as (i~) 1 = y2dA= y2. (b -a ) (w-Y) dy_ (b -a ) w 3 bb' W 12 ' (11) ( Iy ) l=~ x2dA - (b -a ) 3w (b -a )a2w 2(b -a )2aw bb' 4 + 2 + 3 The moment of inertia of area bb~dc and cdd, with respect to x axis and y axis, can be obtained, respectively, as follows, (c - b )w 3 ( /~)~ = 3 ' (a - c )w 3 (/z)a - 12 ' (Iy)2 (c - b)3w (12) -- ~ +(c -b ) b2w+(c -b) 2bw, (d - c )aw (d - c) c2w (d - c) 2 cw (Iu)3 -- 12 + 2 + 3 So, the (ROG) point of generalized trapezoidal fuzzy number (a, b, c, d; w) can be calculated as i (I~)1 + (Ix)2 + (Ix)a r~ = ( ( (c - b) + (d - a ) ) .w) /2 ' (13) ry = (((c-- b) + (d - a)). w) /2 ' where the (I~)1, (Ix)S, (Ix)3, (Iv)i, (Iy)2, (Iy)3 can be obtained from equations (11) and (12). Rank ing Fuzzy Numbers / \ EK-- W . . . . . . . . . . . . . . . . . . a A=(a,a,a,a,w) F igure 3. The moment of iner t ia of a c r i sp number . 1131 EXAMPLE 2. Consider a generalized trapezoidal fuzzy number (a, b, c, d; w), where a = b = c = d and 0 ~< w ~< 1. In this special situation, the generalized fuzzy number has no area. How are we to determine the radius of gyration point of the special fuzzy number? Suppose the fuzzy numbers has the width E, where E --* 0. So the moment of inertia with respect to x axis and to y axis can be calculated as follows, / /0 ( I~ ) = y2 dA = y2 . ~ dy = aw----~3 3 ' / /: (Iv) = x 2 dA = a 2. w dx = a2wz. By applying the limit function, the radius of gyration of the generalized fuzzy number (a, a, a, a; w) with respect to the x axis and with respect to the y axis can be obtained as follows, r x ~ ~ - - - - w , 0 V ~w 3 r A rA)of the fuzzy number (a, a, a, a; w) is (v~/3w, a). Thus, the (ROG) points ( ~, generalized 3. D ISTANCE-BASED RANKING AND AREA-BASED RANKING Consider a fuzzy number A, its centroid point is denoted as (~,zj), where 2 and ~j can be calculated by equations (2) and (3), respectively. Cheng proposed the distance-based method for ranking fuzzy numbers [22]; i.e., R (A) = V/(~) 2 + (~)2. (14) For any two fuzzy numbers Ai and Aj, if R(A i ) < R(A j ) , then A~ < Aj; if R(A i ) =- R(A j ) , then A~ = Aj; if R(A~) > R(A j ) , then A~ > Aj. In addition, Cheng [22] proposed the coefficient of variance (CV index), i.e., CV = a (standard error)/l#l, tt ~t 0, a > 0, to improve Lee and Li's ranking method [21]. In the coefficient of variance approach, the fuzzy number with smaller CV index is ranked higher. Shortcomings are found in the distance-based method and the CV index [23]. Moreover, the distance method contradicts the CV index in ranking some fuzzy numbers. To solve these problems, Chu suggests ranking fuzzy numbers with the area between the centroid and original point [23]. 1132 Y. DENG et al. 4/ . . . . . . . . . . . B 0.1(0.~5-~) 0.40.55 O. ? (o.55+~)> Figure 4. Three fuzzy numbers A, B, C. Then, the area between the centroid point (5,~) and the original point (0,0) of the fuzzy number A is defined as S(A) = x x y, (15) where ~ and ~ is the centroid point of fuzzy number A. For any two fuzzy numbers, A~ and Aj, if S(A.i) < S(Aj), then A~ < Aj; if S(A 0 = S(Aj), then Ai = Aj; if S(Ai) > S(Aj), then Ai > Aj. It seems that the area-based can effectively rank various fuzzy numbers and their images (normal/nonnormal, triangular/trapezoidal, and general). However, Chu's method still has some drawbacks, i.e., it cannot correctly rank the fuzzy numbers in some situations. EXAMPLE 3. In Figure 4, three generalized fuzzy numbers A, B, C are shown, where A -- (0.1, 0.2, 0.3, 0.4; 4 ) , B--(0.4,0.5,0.6,0.7;4), The centroid point (£~, !70, Ri = x/(£i) 2 + (~7~) 2 and S, = 2i x gi, i = A, B, C, can be calculated by equations (14) and (15). The results can be shown in Table 1. Table 1. The rank results of distance method and area method. 2 A 0.25 2 B 0.55 2 C 0.55 /~i = V/(~) 2 + (~)2 0.3345 0.0556 0.5932 0.1222 0.5932 0.1222 As can be seen form Table 1, The rank result of distance method is [22] A S(Aj), then A~ > Aj. 5. COMPARATIVE EXAMPLES In this section, all the numerical examples of Cheng's paper [22], one numerical example from Liou and Wang's paper [5], one numerical example from Chu and Tsao's paper [23] and one self-designed numerical example are displayed to illustrate the validity and the advantage of the proposed ranking method. The two triangular fuzzy numbers, U1 = (0, 1, 2) and U2 = (1/5, 1, 7/4) shown in Figure 5, are ranked by our method. First, we obtain the ROG point of U1 and U2, respectively, as follows, rxl = 0.4082, ry~ = 1.0801, rx2 : 0.4082, ry2 : 1.0330. So, the area between the ROG point and original point of fuzzy numbers U1 and U2 can be obtained, respectively, as follows, 1 1 : 0.4082 × 1.0801 -- 0.4410, S(U l ) = r x x 7"y 2 2 _-- 0.4082 × 1.0330 ---- 0.4217. S(U2) = r~ × ry Obviously, the ranking order is U1 :> U2. The images of these two fuzzy numbers are -U1 = 1 x 1 ( -2 , -1 ,0 ) and -U2 : ( -7 /4 , -1 , -1 /5 ) , respectively. By our method, S( -U I ) : r~ ry : 2 ___ 0.7144 × -1.1284 : -0.4217, producing 0.4082 x (-1.0801) = -0.4410 and S(-U2) = rz 2 x ry the ranking order -Us < -U2. Clearly, similar to Chu's area method [23], the proposed method can also overcome the shortcomings of the inconsistency of Cheng's CV index in ranking fuzzy numbers and their images [22]. 0 1/5 1 7/4 2 ~ 1.922.1 3 4 Figure 5. Triangular fuzzy numbers U1 = Figure 6. Triangular fuzzy numbers B1 = (0, 1, 2) and U2 = (1/5, 1, 7/4). (1.9, 2, 2.1) and B2 = (2.1, 3, 4). 1134 Y. DENG et al. / \ 1 Ul U2 U3 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Figure 7. Triangular fuzzy numbers U 1 : (0.2,0.3,0.5), U2 = (0.17,0.32,0.58), and U3 = (0.25,0.4,0.7). > / \ 1 4 & a, 0.1 0.2 0.3 0.4 0.$ 0.6 0.7 0.8 0.9 1.0 Figure 8. Triangular fuzzy numbers A1 = (0.4,0.5, 1), A2 = (0.4,0.7, 1), and A3 = (0.4, 0.9, 1). > The two triangular fuzzy numbers, B1 = (1.9, 2, 2.1) and B2 = (2.1, 3, 4) shown in Figure 6, taken from paper [23], are ranked by our method, hrtuitively, the ranking order is B1 < B2. However, by Cheng's CV index, the ranking order is B1 > B2, which is an unreasonable result. By the proposed method, 1 1 = 0.4082 x 2.0004 = 0.8167, S(B1) = r~ × % 2 2 = 0.4082 x 3.0580 = 1.2484, S(B2) = r~ × r~ the ranking order is B1 < B2. Again, our method also can overcome the shortcomings of Cheng's CV index. The three triangular fuzzy numbers, U1 = (0.2,0.3,0.5), [72 = (0.17,0.32,0.58) and Ua = (0.25, 0.4, 0.7) shown in Figure 7, from paper [22], are ranked by our method. I i = 0.4082 × 0.3391 = 0.1384, S(U1) = r~ × % 2 2 = 0.4082 x 0.3666 = 0.1497, S(U~) = T~ x % 3 a = 0.4082 × 0.4596 = 0.1876, s(ga) = rx x ry the ranking order is U1 < (-72 < Ua. Also, the ranking order of images of the three fuzzy numbers is -U1 > -(-72 > -Ua. However, by Cheng's CV index, the ranking order of the images is -U1 < -U2 < -Ua. Clearly, our method has consistency in ranking fuzzy numbers and their images, which could not be guaranteed by Cheng's CV index. Consider the three triangular fuzzy numbers, A1 = (0.4,0.5, 1), A2 = (0.4, 0.7, 1), and A3 = (0.4, 0.9, 1) shown in Figure 8, from [22]. By Cheng's CV index, the ranking order is A1 < Aa < A2. However, by Cheng's distance method, the ranking order is A1 < A2 < A3. By our method, 1 1 = 0.4082 x 0.6468 = 0.2640, S(A1) = r~ x ry 2 2 = 0 .4082 x 0 .7106 = 0 .2901, S(A2) = r~ x ry 3 a = 0.4082 x 0.7778 = 0.3175, S(A3) = r~ × r9 the ranking order is A1 < A2 < Aa. Consider the two triangular fuzzy numbers A1 = (3, 5, 7; 1), A2 = (3, 5, 7; 0.8), and the three trapezoidal fuzzy numbers B1 = (5, 7, 9, 10; 1), B2 = (6, 7, 9, 10; 0.6), Ba = (7, 8, 9, 10; 0.4) shown in Figure 9, from [22]. By the proposed method, S(A1) = S(&) = S(B1) = S(B=) = S(B3) = 0.4082 x 5.0662 = 2.0683, 0.3266 × 5.0662 = 1.6546, 0.5118 × 7.7935 = 3.9884, 0.3162 x 8.0519 = 2.5462, 0.2000 × 8.5245 = 1.7049, Ranking Fuzzy Numbers / 1 4 g 0.4 . . . . . . . 1 2 3 4 5 6 7 8 9 10 > Figure 9. Triangular fuzzy numbers A1 = (3,5,7;1), A2 = (3, 5,7,0.8), and trapezoidal fuzzy numbers B1 = (5,7,9,10;1), B2 = (6,7,9,10;0.6), and B3 = (7,8,9, 10;0.4). 0 1 2 3 4 5 Figure 10. Triangular fuzzy numbers A = (1,2,5; 1) and the general fuzzy number, B = (1,2,2,4;1). 1135 the ranking order is A2 < Ba < A1 < B2 < B3. Here, we see that the proposed method can also rank normal/nonnormal triangular and trapezoidal fuzzy numbers. Consider the triangular fuzzy numbers, A = (1, 2, 5; 1) and the general fuzzy number, B = (1, 2, 2, 4; 1), shown in Figure 10, from paper [5]. The membership of B is defined as [1 - - (1 - -2 )211/2 1 ~X ~ 2, f s (x )= [1 (x--2)2] 1/2, 2~