ARTICLE IN PRESS Computers & Industrial Engineering xxx (2009) xxx–xxx
Contents lists available at ScienceDirect
Computers & Industrial Engineering journal homepage: www.elsevier.com/locate/caie
Centroid defuzzification and the maximizing set and minimizing set ranking based on alpha level sets q Ying-Ming Wang * School of Public Administration, Fuzhou University, Fuzhou 350002, PR China
a r t i c l e
i n f o
Article history: Received 27 April 2007 Accepted 17 November 2008 Available online xxxx Keywords: Centroid defuzzification Maximizing set and minimizing set Alpha level sets Risk assessment
a b s t r a c t Centroid defuzzification and the maximizing set and minimizing set methods are two commonly used approaches to ranking fuzzy numbers and often require membership functions to be known. In this paper, the two methods are reinvestigated when explicit membership functions are not known but alpha level sets are available. Two analytical formulas are derived under the assumption that the exact membership functions can be approximated by using piecewise linear functions based on alpha level sets. The derived analytical formulas are of significant importance and provide very useful decision supports for a wide variety of applications of the two methods in industrial engineering and other areas. Numerical examples are offered to test the derived formulas and illustrate their computational processes and applications in risk assessment of a software development. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction In many industrial engineering applications such as fuzzy weighted average (FWA) (Detyniecki & Yager, 2000; Dong & Wong, 1987; Guh, Hon, & Lee, 2001; Guh, Hon, Wang, & Lee, 1996; Guu, 2002; Kao & Liu, 1999, 2001; Lee & Park, 1997; Vanegas & Labib, 2001), fuzzy data envelopment analysis (FDEA) (Kao & Liu, 2000, 2003; León, Liern, Ruiz, & Sirvent, 2003; Wang, Greatbanks, & Yang, 2005), fuzzy multiple criteria decision making (FMCDM) (Chen & Klein, 1997a; Hon, Guh, Wang, & Lee, 1996; Tseng & Klein, 1992), and so on, explicit membership functions are often not available. The only information available is that their a-level sets (or called acuts). In order to compare or rank the fuzzy numbers without explicit membership functions but with a-level sets available, ranking approaches based on a-level sets are necessary for the fuzzy numbers. Several approaches have been suggested in the literature for ranking fuzzy numbers based on their a-level sets. Chen and Klein (1997a, 1997b) suggested an index of difference based on a-level sets, fuzzy subtraction operation and area measurement. Dubios and Parade suggested a defuzzification procedure by averaging the a-cuts, which is called averaging level cuts (ALC) (Oussalah, 2002). Yager (1981) suggested a valuation method, which ranks fuzzy numbers using valuations (see also Yager and Filev (1999) and Detyniecki and Yager (2000). Filev and Yager (1993) also suggested a generalized level set defuzzification (LSD) method. The
q
This work described in this paper was supported by the Natural Science Foundation of Fujian Province of China (No. A0710005) and the National Natural Science Foundation of China (NSFC) (No. 70771027). * Tel.: +86 591 87893307; fax: +86 591 22866677. E-mail address:
[email protected].
purpose of this paper is not to suggest a new approach for defuzzification, but to derive analytical formulas for the most widely used centroid defuzzification method and the maximizing set and minimizing set ranking approach when only a-level sets are available and provide decision supports for their applications in industrial engineering and other areas. This is of significant importance because the two methods are often utilized to convert a FMCDM problem into a crisp MCDM. However, FMCDM problems can also be solved using a-level sets. If a comparison analysis needs to be conducted to find the difference between the two solution processes, the defuzzification methods employed by the two processes have to be the same; otherwise, their results will be incomparable. The rest of the paper is organized as follows. In Section 2, we briefly review the centroid defuzzification method and the maximizing set and minimizing set approach and give their formulas for known fuzzy membership functions. In Section 3, we assume that exact membership functions can be approximated by using piecewise linear functions based on a-level sets and derive analytical formulas for the two methods. Numerical examples are provided and examined in Section 4. The paper is concluded in Section 5. 2. Centroid and utilities for known membership functions ~ is a normal fuzzy number, whose membership function l ~ Let A A is defined by
8 fA~L ðxÞ; > > > > < 1; lA~ ðxÞ ¼ R > fA~ ðxÞ; > > > : 0;
a x b; b x c; c 6 x 6 d;
ð1Þ
otherwise;
0360-8352/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.cie.2008.11.014
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 2
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
where fA~L : ½a; b ! ½0; 1 and fA~R : ½c; d ! ½0; 1 are two continuous mappings from the real line R to the closed interval [0, 1]. The former is a strictly increasing function called left membership function and the latter is a monotonically decreasing function called right ~ is referred to as a fuzzy interval membership function. If b – c, A (Dubois & Prade, 1983), or a flat fuzzy number (Matarazzo & Munda, ~ is referred to as a trapezoi2001). If fA~L and fA~R are both linear, then A ~ ¼ ða; b; c; dÞ, which is dal fuzzy number and is usually denoted by A plotted in Fig. 1. In particular, when b = c, the trapezoidal fuzzy number is reduced to a triangular fuzzy number, denoted by ~ ¼ ða; b; dÞ. So, triangular fuzzy numbers are special cases of trapA ezoidal fuzzy numbers. The centroid defuzzification method defines the centroid coor~ in horizontal axis as its defuzzified value, which can dinate of A be expressed as (Uehara & Hirota, 1998; Wang & Luoh, 2000)
di xMi xMi xmin ¼ ; di ci xmax xmin xGi ai xmax xGi ¼ ; bi ai xmax xmin di xMi di xmin uMi ¼ ¼ ; di ci ðdi ci Þ þ ðxmax xmin Þ xG ai xmax ai ¼ uGi ¼ i ; bi ai ðbi ai Þ þ ðxmax xmin Þ
Rc Rd R Rd Rb L ~ Þdx þ b ðxÞdx þ c ðxf A ~ Þdx A ~ ¼ Ra xlA~ ðxÞdx ¼ a ðxf x0 ðAÞ : Rc Rd R Rb L d l ðxÞdx ðf Þdx þ dx þ ðf Þdx ~ ~ ~ A a a A b c A
uT ðiÞ ¼ ½uMi þ 1 uGi =2;
ð2Þ
~ ¼ ða; b; c; dÞ, its centroid turns out For a trapezoidal fuzzy number A to be
dc ab ~ ¼1 aþbþcþd x0 ðAÞ : 3 ðd þ cÞ ða þ bÞ
ð3Þ
Especially when b = c, the above formula is simplified as
~ ¼ x0 ðAÞ
aþbþd ; 3
ð4Þ
~ ¼ ða; b; dÞ . which is the centroid of triangular fuzzy number A The maximizing set and minimizing set method ranks fuzzy ~ N be N fuzzy ~1; . . . ; A numbers according to their total utilities. Let A numbers to be compared or ranked, whose membership functions are denoted by lA~ i ðxÞ; i ¼ 1; . . . ; N. The method first defines a max~ whose membership func~ and a minimizing set G, imizing set M tions are, respectively, defined by (Chen, 1985)
(
lM~ ðxÞ ¼ (
½ðx xmin Þ=ðxmax xmin Þk ; xmin x xmax ; 0;
otherwise;
ð5Þ
ð7Þ ð8Þ ð9Þ ð10Þ
where uMi ¼ supx ðlM~ ðxÞ ^ lA~ i ðxÞÞ is referred to as the right utility ~ i and uG ¼ sup ðl ~ ðxÞ ^ l ~ ðxÞÞ as its left utility value. It value of A x G Ai i ~ i away from the minimizing set G, ~ is obvious that the farther the A ~ i to the maximizing set M, ~ the smaller the uGi and the closer the A ~ i is defined the larger the uMi . So, the final total utility value of A as (Chen, 1985)
i ¼ 1; . . . ; N:
ð11Þ
~ i and the higher its order. The N The greater the uT(i), the bigger the A ~ ~ fuzzy numbers A1 ; . . . ; AN can all be ranked according to their total utilities. 3. Centroid and utilities by a-level sets In the situations that only a-level sets are available, exact membership functions are usually not known. In this study, we assume that exact membership functions can be approximated by using piecewise linear functions based on a-level sets. ~ be a fuzzy number. Its a-level sets Aa or a-cuts Definition 1. Let A are defined as
Aa ¼ fx 2 XjlA~ ðxÞ ag ¼ ½minfx 2 XjlA~ ðxÞ ag; maxfx 2 XjlA~ ðxÞ ag ¼ ½ðxÞLa ; ðxÞUa ;
0 < a 1:
ð12Þ
According to Zadeh’s extension principle (Dubois & Prade, 1980), ~ can also be expressed as the fuzzy number A
~ ¼ [ a a Aa ; A
0 < a 1:
ð13Þ
k
lG~ ðxÞ ¼ ½ðxmax xÞ=ðxmax xmin Þ ; xmin x xmax ; 0;
otherwise;
ð6Þ
S where xmin = inf X, xmax = sup X, X ¼ Ni¼1 X i ; X i ¼ fxjlA~ i ðxÞ > 0g, and k is a constant reflecting decision maker (DM)’s attitude towards risk with k > 1 representing risk-seeking, k < 1 corresponding to risk-averse and k = 1 standing for risk-neutral (Raj & Kumar, 1999; Raja & Kumar, 1998). Usually, k is set as one. Fig. 2 shows the graphical representations of the maximizing set and minimizing set. ~ inter~ and the minimizing set G In Fig. 2, the maximizing set M sect the right and left membership functions of trapezoidal fuzzy ~ i ¼ ðai ; bi ; ci ; di Þ, respectively, at the points Mi and Gi, number A whose coordinates are determined by the following equations:
~ ¼ [a a Aa ¼ [a a ½ðxÞL ; ðxÞU ð0 < a 1Þ be a Definition 2. Let A a a fuzzy number represented by a-level sets. Its membership function fA~ ðxÞ is approximately defined as
lA~ ðxÞ ¼
8 0; > > > > > > > > < ai þ > > 1; > > > > > > : ai þ
x < ðxÞLa0 or x > ðxÞUan ; Dai ðxðxÞLa Þ ðxÞLa
iþ1
i
ðxÞLa
i
; ðxÞLai x ðxÞLaiþ1 ; i ¼ 0; 1; . . . ; n 1; ðxÞLan x ðxÞUan ;
Dai ððxÞU a xÞ i
U ðxÞU a ðxÞa i
iþ1
; ðxÞUaiþ1 x ðxÞUai ; i ¼ 0; 1; . . . ; n 1; ð14Þ
μ ~A( x)
where Dai = ai+1 ai, i = 0, 1,. . . , n 1 and 0 = a0 < a1 < < an 1 < an = 1. Fig. 3 shows the graphical representation of the above piecewise linear membership function. Under the assumption of piecewise linearity, we have the following theorem.
~
A
1
f ~AL
0
a
~ ¼ [a a Aa ¼ [a a ½ðxÞL ; ðxÞU ð0 < a 1Þ be a Theorem 1. Let A a a fuzzy number represented by a-level sets, whose membership function ~ can be is defined by Eq. (14). Then the defuzzified centroid of A determined by
f A~R
b
c
d
x
Fig. 1. Membership functions of trapezoidal fuzzy numbers.
Rd ~ ¼ Ra xlA~ ðxÞdx ; x0 ðAÞ d lA~ ðxÞdx a
ð15Þ
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 3
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
μ (x)
~
1
μ M~ (x)
Ai
μ G~ (x)
Gi
uG i
Mi
u Mi
0 xmin
ai
xG i bi
ci
di
xMi
xmax
x
Fig. 2. Graphical representations of maximizing set and minimizing set.
μ A~ (x) αn =1 α i+1
f
L ~ A
f
R ~ A
αi
x 0
(x) αL0 (x) αLi
(x) αLi+1
(x) αLn
(x) U αn
(x) U αi+1
(x) U αi
(x) U α0
Fig. 3. Piecewise linear membership function represented by a-level sets.
where
Z
lA~ ðxÞdx ¼
a
Remark 1. Let n = 1. Then Eqs. (18), (19), and (15) become
"
d
1 ðxÞUan ðxÞLan 2 þ
n1 X
n1 X i¼1
ai ððxÞUaiþ1 ðxÞLaiþ1 Þ
Z
#
aiþ1 ððxÞUai ðxÞLai Þ ;
ð16Þ
i¼0
Z
"
d
xlA~ ðxÞdx ¼
a
ð18Þ d
a
" # n1 X 1 2U 2L 2U 2L 2U 2L ððxÞa0 ðxÞa0 Þ þ ððxÞan ðxÞan Þ þ 2 xlA~ ðxÞdx ¼ ððxÞai ðxÞai Þ 6n i¼1 þ
1 6n
n1 X
ðxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 :
d
xlA~ ðxÞdx ¼
1h 2L 2U 2L ððxÞ2U a0 ðxÞa0 Þ þ ððxÞan ðxÞan Þ 6 i þððxÞUa0 ðxÞUan ðxÞLa0 ðxÞLan Þ ;
ð19Þ
ð20Þ
ð21Þ
" # ðxÞUa0 ðxÞUan ðxÞLa0 ðxÞLan 1 ; ðxÞLa0 þ ðxÞLan þ ðxÞUan þ ðxÞUa0 3 ððxÞUa0 ðxÞLa0 Þ þ ððxÞUan ðxÞLan Þ ð22Þ
which is exactly the centroid defuzzication formula of trapezoidal fuzzy numbers (see Eq. (3)). Remark 2. If ðxÞLan ¼ ðxÞUan , then Eqs. (18), (19), and (15) can be simplified as
Z
d
lA~ ðxÞdx ¼
a
Z a
i¼0
The proof of the theorem is provided in Appendix A.
Z
i 1 h U ðxÞa0 ðxÞLa0 þ ððxÞUan ðxÞLan Þ ; 2
ð17Þ
Especially when Dai 1n and ai ¼ ni ; i ¼ 0; . . . ; n; the equations are simplified as " # Z d n1 X 1 lA~ ðxÞdx ¼ ððxÞUai ðxÞLai Þ ; ððxÞUa0 ðxÞLa0 Þ þ ððxÞUan ðxÞLan Þ þ 2 2n a i¼1
Z
a
~ ¼ x0 ðAÞ
i¼0 n1 1X Dai ððxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 Þ: 6 i¼0
lA~ ðxÞdx ¼
a
n1 X 1 2L 2L ai ððxÞ2U ðxÞ2U an ðxÞan aiþ1 ðxÞaiþ1 Þ 6 i¼1 # n1 X 2U 2L þ aiþ1 ððxÞai ðxÞai Þ
þ
d
d
" # n1 X 1 ððxÞUai ðxÞLai Þ ; ððxÞUa0 ðxÞLa0 Þ þ 2 2n i¼1
" n1 X 1 2L 2L ððxÞ2U xlA~ ðxÞdx ¼ ððxÞ2U a0 ðxÞa0 Þ þ 2 ai ðxÞai Þ 6n i¼1 # n1 X U U L L þ ððxÞai ðxÞaiþ1 ðxÞai ðxÞaiþ1 Þ ;
ð23Þ
ð24Þ
i¼0
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 4
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
2L ððxÞ2U a0 ðxÞa0 Þ þ 2
~ ¼1 x0 ðAÞ 3
n1 P i¼1
2L ððxÞ2U ai ðxÞai Þ þ
ððxÞUa0 ðxÞLa0 Þ þ 2
nP 1
i¼0 n1 P i¼1
ððxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 Þ ð25Þ
:
ððxÞUai ðxÞLai Þ
Remark 3. Let n ¼ 1: Then Eqs. (23)–(25) become
uM ¼
aj0 þ1 ððxÞUaj xmin Þ aj0 ððxÞUaj þ1 xmin Þ 0
0 þ1
0
Z
d
a
1 2
lA~ ðxÞdx ¼ ððxÞUa0 ðxÞLa0 Þ;
ð26Þ
uG ¼
d
a
xlA~ ðxÞdx ¼
1 2L U U L ½ððxÞ2U a0 ðxÞa0 Þ þ ððxÞa0 ðxÞan ðxÞa0 6 ðxÞLan Þ;
0
ðxÞLai
2U 2L U U L L ~ ¼ 1 ððxÞa0 ðxÞa0 Þ þ ððxÞa0 ðxÞan ðxÞa0 ðxÞan Þ x0 ðAÞ 3 ððxÞUa ðxÞLa Þ
¼
ð35Þ
:
ð36Þ
0
Remark 5. Let n = 1. Then Eqs. (35) and (36) become
uM ¼
0
1 ððxÞUa0 þ ðxÞUan þ ðxÞLa0 Þ; 3
0
ðxÞLai þ ðai0 þ1 ai0 Þðxmax xmin Þ
;
The proof of the theorem is provided in Appendix B.
ð27Þ
0
þ ðaj0 þ1 aj0 Þðxmax xmin Þ
ai0 þ1 ðxmax ðxÞLai Þ ai0 ðxmax ðxÞLai þ1 Þ 0 þ1
Z
0
ðxÞUaj ðxÞUaj
ð28Þ
which is exactly the centroid defuzzication formula of triangular fuzzy numbers (see Eq. (4)). Remark 4. The equations below developed in Oussalah (2002), Uehara and Hirota (1998), Yager (1981), Yager and Filev (1999), respectively, are not the centroid defuzzification formulas:
uG ¼
ðxÞUa0 xmin U
;
ð37Þ
;
ð38Þ
ðxÞa0 ðxÞUan þ ðxmax xmin Þ xmax ðxÞLa0 ðxÞLan ðxÞLa0 þ ðxmax xmin Þ
which are exactly the same as Eqs. (9) and (10). In order to determine which intervals the intersection points G and M lie in, we introduce the following sign functions:
S1 ðaÞ ¼ lG~ ððxÞLa Þ a;
ð39Þ
U
L
U
S2 ðaÞ ¼ lM~ ððxÞa Þ a:
!
n ðxÞai þ ðxÞai 1X ; n i¼1 2 P ½ððxÞUai ðxÞLai ÞððxÞUai þ ðxÞLai Þ
xALC ¼
x ¼
i
~ ¼ ValðAÞ
2 Z
P i
1
Av eðAa Þda ¼
0
0
1
L
U
S1 ðaÞ > 0; a ai0
and
ðxÞa þ ðxÞa da: 2
0
ð31Þ
ð32Þ
~ ¼ ða; b; cÞ, (31) is simplified as For triangular fuzzy numbers A
S2 ðaÞ > 0; a aj0 S2 ðaÞ < 0; a aj0 þ1
0 þ1
and ½ðxÞUaj
0 þ1
; ðxÞUaj can be readily 0
determined. Tables 2 and 3 show an illustrative example. 4. Numerical examples In this section, we examine two numerical examples. One is a test example, in which exact membership function is known. The purpose of the test is to verify whether the results based on a-level sets are identical with those obtained from known membership function. The other is a risk assessment example, which is taken from Chen (2001).
ð33Þ
The computation of defuzzified centroid can be easily implemented on a table. Table 1 shows an illustrative example.
Table 1 Alpha-level sets and the computation of defuzzified centroid.
~ ¼ [a a ðAÞ ¼ [a a ½ðxÞL ; ðxÞU ð0 < a 1Þ be a Theorem 2. Let A a a a fuzzy number represented by a-level sets. Its membership function is ~ defined by (5) with defined by (14). Suppose the maximizing set M at k = 1 intersects the right membership function fA~R ~ defined by (6) with xM 2 ½ðxÞUaj þ1 ; ðxÞUaj and the minimizing set G 0 0 k = 1 intersects the left membership function fA~L at xG 2 ½ðxÞLai ; ðxÞLai þ1 , 0 0 ~ can be determined as shown in Fig. 4. Then the total utility value of A by
ai
ðxÞLai
ðxÞUai
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Sum
34.78 38.18 41.64 45.15 48.72 52.33 55.98 59.66 63.36 67.08 70.81
99.11 96.38 93.62 90.86 88.08 85.28 82.46 79.61 76.72 73.79 70.81
uT ¼ ðuM þ 1 uG Þ=2; where
:
So, by checking the signs of the above two functions and observing their changes, ½ðxÞLai ; ðxÞLai
!
~ ¼ a þ b þ c þ d: ValðAÞ 4
a þ 2b þ c : 4
ð30Þ
~ ¼ ða; b; c; dÞ, (31) turns Especially for trapezoidal fuzzy numbers A out to be
~ ¼ ValðAÞ
It can be seen very clearly from Fig. 4 that
S1 ðaÞ < 0; a ai0 þ1 ;
ððxÞUai ðxÞLai Þ Z
ð29Þ
ð40Þ
ð34Þ
ðxÞUai ðxÞLai 64.33 58.20 51.98 45.71 39.36 32.95 26.48 19.95 13.36 6.71 359.03
2L ðxÞ2U ai ðxÞai
ðxÞLai ðxÞLaiþ1
ðxÞUai ðxÞUaiþ1
8613.144 7831.392 7030.815 6217.017 5384.448 4534.250 3665.891 2778.437 1871.469 945.238
1327.900 1589.815 1880.046 2199.708 2549.518 2929.433 3339.767 3780.058 4250.189 4749.935
9552.222 9023.096 8506.313 8002.949 7511.462 7032.189 6564.641 6107.679 5661.169 5225.070
48872.1
28596.37
73186.79
~ ¼ 1 8613:144þ2 ð48872:18613:144Þþð73186:7928596:37Þ ¼ 68:18 x0 ðAÞ 64:33þ2 ð359:0364:33Þ 3
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 5
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
μ M~
~ α n = 1 μG α i +1
0
α i0
uG
αj
0+1
uM
αj
0
x
(x) αLi
xmin
0
(x) αLi
(x) Uα j (x) Uα
0 +1
0+1
xmax
j 0
Fig. 4. Maximizing and minimizing sets for fuzzy numbers represented by a-level sets.
Table 2 Alpha-level sets for the overall risk scores of three new products development (NPD) and the changes of their sign functions.
a
Alpha-level sets
Changes of sign functions
NPD1 ðxÞLa 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 xmin ¼ 5:19;
NPD2 ðxÞUa
ðxÞLa
NPD3 ðxÞUa
NPD1
ðxÞLa
34.78 99.11 25.71 89.21 5.19 38.18 96.38 28.58 86.04 6.29 41.64 93.62 31.55 82.89 7.52 45.15 90.86 34.61 79.77 8.91 48.72 88.08 37.77 76.67 10.46 52.33 85.28 41.01 73.59 12.2 55.98 82.46 44.33 70.53 14.16 59.66 79.61 47.74 67.49 16.36 63.36 76.72 51.23 64.46 18.84 67.08 73.79 54.78 61.43 21.63 70.81 70.81 58.41 58.41 24.79 L U xmax ¼ 99:11; S1 ¼ ðxmax ðxÞa Þ=ðxmax xmin Þ a and S2 ¼ ððxÞa xmax Þ=ðxmax
Table 3 Utilities of the overall risk scores of the three new products development and their rankings. New product development
uM
uG
uT
Rank
NPD1 NPD2 NPD3
0.7706 0.6723 0.4345
0.4986 0.5876 0.8421
0.6360 0.5423 0.2962
1 2 3
~ ¼ ð3; 5; 7; 10Þ Example 1. Given the trapezoidal fuzzy number A and its a-level sets at a = 0, 0.1,. . . , 1.0, respectively, together with ~ ¼ ð0; 15; 15Þ and the minimizing set the maximizing set M ~ ¼ ð0; 0; 15Þ. G By Eqs. (3), and (9), (10), (11), we have its defuzzified centroid and utilities based on known membership function as follows:
dc ab ~ ¼1 aþbþcþd x0 ðAÞ 3 ðd þ cÞ ða þ bÞ 1 10 7 3 5 ¼ 3 þ 5 þ 7 þ 10 ¼ 6:2963; 3 ð10 þ 7Þ ð3 þ 5Þ uM ¼
d xmin 10 0 ¼ 0:5556; ¼ ðd cÞ þ ðxmax xmin Þ ð10 7Þ þ ð15 0Þ
uG ¼
xmax a 15 3 ¼ 0:7059; ¼ ðb aÞ þ ðxmax xmin Þ ð5 3Þ þ ð15 0Þ
NPD2
NPD3
ðxÞUa
S1
S2
S1
S2
S1
S2
61.85 58.23 54.6 50.94 47.27 43.58 39.87 36.13 32.37 28.59 24.79 xmin Þ a
+ + + + +
+ + + + + + + +
+ + + + + +
+ + + + + + +
+ + + + + + + + +
+ + + + +
uT ¼
uM þ 1 uG 0:5556 þ 1 0:7059 ¼ 0:4248: ¼ 2 2
The defuzzified centroid and utilities based on a-level sets are computed in Tables 4 and 5, form which it can be seen quite clearly that the results based on a-level sets are identical with those obtained from known membership function. This verifies the validity of the two analytical formulas developed in this study.
Table 4 ~ ¼ ð3; 5; 7; 10Þ and the computation Alpha-level sets of the trapezoidal fuzzy number A of its defuzzified centroid.
ai
ðxÞLai
ðxÞUai
ðxÞUai ðxÞLai
2L ðxÞ2U ai ðxÞai
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Sum
3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0
10.0 9.7 9.4 9.1 8.8 8.5 8.2 7.9 7.6 7.3 7.0
7.0 6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 49.5
91.00 83.85 76.80 69.85 63.00 56.25 49.60 43.05 36.60 30.25 24.00 624.25
ðxÞLai ðxÞLaiþ1
ðxÞUai ðxÞUaiþ1
9.60 10.88 12.24 13.68 15.20 16.80 18.48 20.24 22.08 24.00
97.00 91.18 85.54 80.08 74.80 69.70 64.78 60.04 55.48 51.10
163.20
729.70
~ ¼ 1 91þ24þ2 ð624:259124Þþð729:7163:2Þ ¼ 6:2963 x0 ðAÞ 7þ2þ2 ð4972Þ 3
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 6
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
Despite the fact that the problem has been investigated many times, all the investigations focus on transforming the problem into a crisp multiple attribute decision analysis. There has been no attempt to solve the problem using a-level sets and the extension principle, which are supposed to provide exact solution to the problem. So, in this study, the problem is regarded as a fuzzy weighted average (FWA) and solved of using a-level sets and the extension principle. Assessments of risk items under each attribute are first aggregated using FWA to generate an aggregated assessment for the attribute and the aggregated assessments of the six attributes are then aggregated again to generate an overall risk assessment for each software project. Eleven a-levels, a = 0, 0.1, . . . , 1.0, are set for computation. The overall risk assessments for the three projects are presented in Table 7, together with their defuzzified centroids, utilities and rankings. The rankings turn out to be the same as those obtained by Chen (2001), but our approach provides much more information on the overall risk of each project and also offers a full picture rather than only a point estimate for the overall risk assessment of each project, as shown in Fig. 6.
Table 5 ~ ¼ ð3; 5; 7; 10Þ and the computation Alpha-level sets of the trapezoidal fuzzy number A of its utilities.
ai
ðxÞLai
ðxÞUai
S1(ai)
S2(ai)
Utilities
0.0 0.1 0.2 0.3
3.0 3.2 3.4 3.6
10.0 9.7 9.4 9.1
+ + + +
+ + +
¼ 0:7059 uG ¼ 1:0 ð154:8Þ0:9 ð155:0Þ ð5:04:8Þþ0:1 ð150Þ
0.4 0.5
3.8 4.0
8.8 8.5
+ +
uM ¼ 0:3 ð9:40Þ0:2 ð9:10Þ ð9:49:1Þþ0:1 ð150Þ ¼ 0:5556
0.6 0.7 0.8 0.9 1.0
4.2 4.4 4.6 4.8 5.0
8.2 7.9 7.6 7.3 7.0
+ + + +
G ¼ 0:4248 uT ¼ uM þ1u 2
Example 2. Consider a software development risk assessment problem which was investigated by Chen (2001), Lee (1996a, 1996b, 1999) and Lee, Lee, Lee, and Chen (2003, 2004). The hierarchical structure for the assessment problem is shown in Fig. 5 and consists of six attributes and 14 risk items (or called risk factors). The risk of each item is defined as the product of grade of risk and grade of importance that are described by linguistic variables and characterized by triangular fuzzy numbers. Table 6 shows the assessments of each risk item and the weights of the attributes for three software projects (A1, A2 and A3) provided by two experts (decision makers). For brevity, the original assessments and weights presented in Chen (2001) by two experts have been averaged as a whole in Table 6.
5. Concluding remarks Many applications of fuzzy set theory require defuzzification and ranking approaches based on alpha level sets because exact membership functions may not always be available. In this paper, we have assumed that exact membership functions can be approximated using piecewise linear functions based on alpha level sets and derived two analytical formulas to meet such a requirement. The two formulas are, respectively, extensions of the most widely used centroid defuzzification approach and the maximizing set and minimizing set method to alpha level sets. One is to capture the
Attribute W1
X1: Personnel
Risk item W11
X11: Personnel shortfalls, key person(s) quit W21 W22
W2
X2: System requirement
W23 W24
X21: Requirement ambiguity X22: Developing the wrong software function X23: Developing the wrong user interface X24: Continuing stream requirement changes
W31 W3
X3: Schedules and budgets
X31: Schedule not accurate W32
X32: Budget not sufficient
W41
Aggregative risk assessment
X41: Gold-piating W42 W4
W5
W6
X4: Developing technology
X5: External resource
X6: Performance
W43
X42: Skill levels inadequate X43: Straining hardware
W44
X44: Straining software
W51
X51: Shortfalls in externally furished components
W52
X52: Shortfalls in externally performed tasks
W61
X61: Real-time performance shortfalls
Fig. 5. Hierarchical structure for software development risk assessment (Chen, 2001).
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 7
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx Table 6 Weights and risk assessment data for three software development projects. Risk item
X1 X11
X2 X21
X22
Weight
Grade of risk
(0.125, 0.275, 0.425) (0.75, 0.875, 1) A1: (0.45, 0.55, 0.65) A2: (0.65, 0.75, 0.85) A3: (0, 0.05, 0.15) (0.275, 0.45, 0.6) (0.2, 0.325, A1: (0.2, 0.3, 0.4) 0.425) A2: (0.7, 0.8, 0.9) A3: (0.15, 0.25, 0.35) (0.2, 0.35, 0.5)
Grade of importance (0.5, 0.675, 0.8) (0.7, 0.8, 0.9) (0, 0, 0.1) (0.4, 0.5, 0.6) (0.7, 0.8, 0.9) (0.1, 0.2, 0.3)
A1: (0.25, 0.4, 0.55) A2: (0.5, 0.6, 0.7) A3: (0, 0.05, 0.15)
(0.3, 0.4, 0.5) (0.6, 0.7, 0.8) (0, 0.1, 0.2)
X23
(0.175, 0.3, 0.4)
A1: (0.25, 0.45, 0.55) A2: (0.7, 0.8, 0.9) A3: (0.05, 0.15, 0.25)
(0.25, 0.35, 0.45) (0.6, 0.7, 0.8) (0.05, 0.15, 0.25)
X24
(0.15, 0.325, 0.5)
A1: (0.3, 0.4, 0.5)
(0.25, 0.375, 0.5)
A2: (0.7, 0.8, 0.9) A3: (0.15, 0.25, 0.35)
(0.7, 0.8, 0.9) (0.1, 0.2, 0.3)
X3 X31
X32
X4 X41
(0.15, 0.25, 0.35) (0.225, 0.35, A1: (0.3, 0.4, 0.5) 0.55) A2: (0.7, 0.8, 0.9) A3: (0.1, 0.2, 0.3) (0.3, 0.525, 0.625)
A1: (0.3, 0.5, 0.7)
(0.2, 0.3, 0.4)
A2: (0.5, 0.6, 0.7) A3: (0.15, 0.25, 0.35)
(0.6, 0.7, 0.8) (0.15, 0.25, 0.35)
(0.2, 0.325, 0.45) (0.25, 0.35, A1: (0.225, 0.35, 0.45) 0.475) A2: (0.65, 0.75, 0.85) A3: (0.2, 0.3, 0.4)
ðR2 ÞUa
ðR3 ÞLa
ðR3 ÞUa
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0.077 0.086 0.096 0.106 0.116 0.127 0.138 0.150 0.161 0.174 0.186
0.345 0.325 0.306 0.288 0.271 0.255 0.240 0.226 0.212 0.199 0.186
0.327 0.345 0.363 0.380 0.398 0.417 0.435 0.454 0.473 0.492 0.511
0.719 0.696 0.674 0.653 0.632 0.611 0.591 0.571 0.551 0.531 0.511
0.009 0.011 0.014 0.017 0.020 0.024 0.028 0.032 0.037 0.041 0.046
0.116 0.108 0.100 0.092 0.085 0.077 0.071 0.064 0.058 0.052 0.046
Centroid Total utility Rank
0.198 0.294 2
0.517 0.665 1
Project 1
0.055 0.092 3
Project 2
Project 3
1 0.8 0.6
(0.2, 0.3, 0.4)
A1: (0.3, 0.4, 0.5) A2: (0.7, 0.8, 0.9) A3: (0.15, 0.25, 0.35)
(0.2, 0.3, 0.4) (0.65, 0.75, 0.85) (0.2, 0.3, 0.4)
X44
(0.2, 0.3, 0.4)
A1: (0.2, 0.3, 0.4) A2: (0.6, 0.7, 0.8) A3: (0.15, 0.25, 0.35)
(0.2, 0.3, 0.4) (0.65, 0.75, 0.85) (0.15, 0.25, 0.35
X51
(0.1, 0.275, 0.4) (0.45, 0.55, 0.65)
A1: (0.2, 0.3, 0.4)
(0.45, 0.55, 0.65)
(0.5, 0.6, 0.7)
A2: (0.55, 0.65, 0.75) A3: (0.15, 0.25, 0.35) A1: (0.3, 0.4, 0.5) A2: (0.7, 0.8,0.9) A3: (0.15, 0.25, 0.35) A1: (0.4, 0.5, 0.6) A2: (0.5, 0.6, 0.7) A3: (0.15, 0.25, 0.35)
0.2 0
0
0.1
0.2
0.3
(0.55, 0.65, 0.75) (0.1, 0.2, 0.3)
X43
X61
ðR2 ÞLa
(0.275, 0.4, 0.525)
(0.1, 0.25, 0.4) (0.65, 0.75, 0.85) (0.1, 0.2, 0.3)
(0.125, 0.3, 0.5) (0.85, 0.95, 1)
Project 3
ðR1 ÞUa
0.4
A1: (0.15, 0.25, 0.35) A2: (0.65, 0.75, 0.85) A3: (0.05, 0.15, 0.25)
X6
Project 2
ðR1 ÞLa
(0.6, 0.7, 0.8) (0.05, 0.15, 0.25)
(0.1, 0.2, 0.3)
X52
Project 1
(0.225, 0.35, 0.45)
X42
X5
a
Alpha
Attribute
Table 7 Alpha-level sets for the overall risks of the three software projects at different alevels.
(0.55, 0.65, 0.75) (0.2, 0.3, 0.4) (0.45, 0.55, 0.65) (0.7, 0.8, 0.9) (0.15, 0.25, 0.35)
0.5
0.6
0.7
Fig. 6. The overall risks of the three software development projects.
also been illustrated in detail. A software development risk assessment example has been reinvestigated as a fuzzy weighted average and has demonstrated the applications of alpha level sets and the two formulas. It can be concluded that the two formulas provide very helpful support to the applications of fuzzy logic. Appendix A. The proof of Theorem 1 From Definition 2 and Fig. 3, it is known that
Z
d
lA~ ðxÞdx ¼
a
Z
þ
a
1
lA~ ðxÞdx þ þ
ðxÞLa
þ
Z
ðxÞLa
Z
d
xlA~ ðxÞdx ¼
Z
ðxÞU an
lA~ ðxÞdx þ
ðxÞU a
0
ðxÞLa
1
ðxÞLa
Z
ðxÞU a
n1
ð41Þ
xlA~ ðxÞdx þ þ
0
þ þ
Z
Z
ðxÞLan
ðxÞLa
xlA~ ðxÞdx
n1
ðxÞU an
ðxÞLan
Z
lA~ ðxÞdx þ
ðxÞU an
lA~ ðxÞdx;
ðxÞU a1
Z
lA~ ðxÞdx
n1
ðxÞLan
Z
ðxÞLan
ðxÞLa
0
(0.25, 0.35, 0.45) (0.45, 0.55, 0.65) (0.2, 0.3, 0.4)
centroid of a fuzzy number from its alpha level sets. The other is to calculate the utilities of a fuzzy number using its alpha level sets and a predetermined maximizing set and minimizing set. The validity of the two formulas has been examined and verified through a test example and their computational processes have
0.4
Overall risk
ðxÞU a
ðxÞU a1
0
xlA~ ðxÞdx þ xlA~ ðxÞdx:
Z
ðxÞU a
ðxÞU an
n1
xlA~ ðxÞdx þ ð42Þ
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 8
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
Let
Z
Q iL ¼
iþ1
lA~ ðxÞdx; i ¼ 0; 1; . . . ; n 1;
ðxÞLa
ðxÞU an
lA~ ðxÞdx;
ðxÞLan
Q iU ¼
Z
i
ð44Þ
xlA~ ðxÞdx;
ð46Þ
¼
iþ1 iþ1
i ¼ 0; 1; . . . ; n 1;
Z
xlA~ ðxÞdx;
ðxÞU a
ð47Þ
Z
xlA~ ðxÞdx;
ðxÞU a
# dx
ai
2U ððxÞ2U ai ðxÞaiþ1 Þ Dai 1 U 1 2U 3U 3U þ U ðxÞai ððxÞ2U ai ðxÞaiþ1 Þ ððxÞai ðxÞaiþ1 Þ U 2 3 ðxÞa ðxÞa
2
ai
iþ1
Dai U U ððxÞai ðxÞ2U ððxÞ2U aiþ1 Þ þ ai þ ðxÞai ðxÞaiþ1 2 6 2ðxÞ2U aiþ1 Þ; i ¼ 0; 1; . . . ; n 1: 2U
lA~ ðxÞdx ¼ Q m þ
ð54Þ
i ¼ 0; 1; . . . ; n 1:
ð48Þ
"
iþ1
ai þ
ðxÞLa
# Dai ðx ðxÞLai Þ ðxÞLaiþ1 ðxÞLai
i
dx
¼ ai ððxÞLaiþ1 ðxÞLai Þ Dai 1 2L 2L L L L ðxÞ Þ ðxÞ ððxÞ ðxÞ Þ þ L ððxÞ aiþ1 ai ai aiþ1 ai ðxÞaiþ1 ðxÞLai 2 1 ¼ ðai þ Dai ÞððxÞLaiþ1 ðxÞLai Þ; 2
ðQ iL þ Q iU Þ
¼ ðxÞUan ðxÞLan n1 X 1 þ ðai þ Dai ÞððxÞLaiþ1 ðxÞLai Þ 2 i¼0 1 þðai þ Dai ÞððxÞUai ðxÞUaiþ1 Þ 2 n1 X 1 ¼ ðxÞUan ðxÞLan þ ðai þ Dai Þ½ððxÞUai ðxÞLai Þ 2 i¼0
Then we have ðxÞLa
n1 X i¼0
iþ1
Q iL ¼
ðxÞUai ðxÞUaiþ1
d
a
i
Z
x ai þ
ðxÞU a
Dai ððxÞUai xÞ
Further, we have
i
ðxÞU an
ðxÞLan
RiU ¼
¼
ð45Þ
ðxÞLa
Rm ¼
ð43Þ
lA~ ðxÞdx; i ¼ 0; 1; . . . ; n 1;
ðxÞLa
Z
i
i
ðxÞU a
ðxÞU a
Z
"
ðxÞU a
iþ1
i
Z
Qm ¼
RiL ¼
RiU ¼
ðxÞLa
Z
ððxÞUaiþ1 ðxÞLaiþ1 Þ " n1 X 1 ðxÞUan ðxÞLan ai ððxÞUaiþ1 ðxÞLaiþ1 Þ ¼ 2 i¼1 # n1 X U L þ aiþ1 ððxÞai ðxÞai Þ ;
i ¼ 0; 1; . . . ; n 1; ð49Þ
ð55Þ
i¼0
Z
Qm ¼
ðxÞU an
lA~ ðxÞdx ¼
ðxÞLan
Q iU ¼
Z
"
ðxÞU a
i
ai þ
ðxÞU a
iþ1
Z
ðxÞU an
ðxÞLan
dx ¼ ðxÞUan ðxÞLan ;
Z a
Dai ððxÞUai xÞ U
ð50Þ
#
U
ðxÞai ðxÞaiþ1
d
xlA~ ðxÞdx ¼ Rm þ
1 2L ½ðxÞ2U an ðxÞan 2 n1 X ai Dai 2L þ ððxÞ2L ð2ðxÞ2L aiþ1 ðxÞai Þ þ aiþ1 2 6 i¼0 i L L ðxÞ2L ai ðxÞai ðxÞaiþ1 Þ n1 X ai Dai 2U þ ððxÞ2U ððxÞ2U ai ðxÞaiþ1 Þ þ ai 2 6 i¼0 i þðxÞUai ðxÞUaiþ1 2ðxÞ2U aiþ1 Þ
¼
dx
¼ ai ððxÞai ðxÞUaiþ1 Þ Dai 1 2U ðxÞUai ððxÞUai ðxÞUaiþ1 Þ ððxÞ2U þ U ai ðxÞaiþ1 Þ U 2 ðxÞai ðxÞaiþ1 1 ¼ ai þ Dai ðxÞUai ðxÞUaiþ1 ; i ¼ 0; 1; . . . ; n 1; 2 ð51Þ RiL ¼
ðxÞLa
iþ1
ðxÞLa
" x ai þ
# Dai ðx ðxÞLai Þ ðxÞLaiþ1 ðxÞLai
i
¼
ai
2L
dx
Z
n1 1X 2L Dai ½ððxÞ2U ai ðxÞai Þ 6 i¼0
2L þððxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 Þ 2ððxÞ2U aiþ1 ðxÞaiþ1 Þ
i
Dai 2L L ¼ ððxÞaiþ1 ðxÞai Þ þ ð2ðxÞ2L aiþ1 ðxÞai ðxÞai 2 6 ðxÞLaiþ1 Þ; i ¼ 0; 1; . . . ; n 1; 2L
n1 1 1X 2L 2L ai ½ððxÞ2U ½ðxÞ2U an ðxÞan þ ai ðxÞai Þ 2 2 i¼0 2L ððxÞ2U aiþ1 ðxÞaiþ1 Þ þ
2L
ððxÞaiþ1 ðxÞai Þ Da i 1 1 L 3L 3L 2L 2L ðxÞ Þ ððxÞ ðxÞ Þ ððxÞ ðxÞ þ L aiþ1 ai aiþ1 ai 2 ai ðxÞa ðxÞLa 3 iþ1
Rm ¼
¼
2
ai
ðRiL þ RiU Þ
i¼0
U
Z
n1 X
¼
2L
ð52Þ
n1 X 1 2L 2L ai ððxÞ2U ½ðxÞ2U an ðxÞan aiþ1 ðxÞaiþ1 Þ 6 i¼1
þ
n1 X
2L aiþ1 ððxÞ2U ai ðxÞai Þ
i¼0 ðxÞU an
ðxÞLan
xdx ¼
1 2L ½ðxÞ2U an ðxÞan ; 2
ð53Þ
þ
n1 1X Dai ððxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 Þ; 6 i¼0
ð56Þ
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014
ARTICLE IN PRESS 9
Y.-M. Wang / Computers & Industrial Engineering xxx (2009) xxx–xxx
Rd
2L ðxÞ2U an ðxÞan
xlA~ ðxÞdx
1 ~ ¼ Ra x0 ðAÞ ¼ d lA~ ðxÞdx 3 a
n1 P i¼1
2L ai ððxÞ2U aiþ1 ðxÞaiþ1 Þ þ
ðxÞUan ðxÞLan
n1 P i¼1
n1 P i¼0
2L 1 aiþ1 ððxÞ2U ai ðxÞai Þ þ 6
ai ððxÞUaiþ1 ðxÞLaiþ1 Þ þ
Appendix B. The proof of Theorem 2 Let
Dai0 ðxG ðxÞLai Þ xmax xG 0 ¼ ai0 þ : xmax xmin ðxÞLai þ1 ðxÞLai 0
ð58Þ
0
From Eq. (58), we get
xG ¼
xmax ððxÞLai
0 þ1
ðxÞLai Þ þ ðai0 þ1 ðxÞLai ai0 ðxÞLai 0
L
0
0 þ1
Þðxmax xmin Þ :
L
ðxÞai
0 þ1
ðxÞai þ Dai0 ðxmax xmin Þ 0
ð59Þ Accordingly, we have
xmax Dai0 ai0 þ1 ðxÞLai þ ai0 ðxÞLai þ1 xmax xG 0 0 uG ¼ ¼ xmax xmin ðxÞLai þ1 ðxÞLai þ Dai0 ðxmax xmin Þ 0
¼
0
ai0 þ1 ðxmax ðxÞLai Þ ai0 ðxmax ðxÞLai þ1 Þ 0
ðxÞLai
0 þ1
0
ðxÞLai þ ðai0 þ1 ai0 Þðxmax xmin Þ
ð60Þ
:
0
Let
Daj0 ððxÞUaj xM Þ xM xmin 0 ¼ aj0 þ : xmax xmin ðxÞUaj ðxÞUaj þ1 0
ð61Þ
0
Accordingly, we obtain
xM ¼
xmin ððxÞUaj ðxÞUaj
0 þ1
0
U
Þ þ ðaj0 þ1 ðxÞUaj aj0 ðxÞUaj
ðxÞaj ðxÞaj 0
0 þ1
0
U
0 þ1
Þðxmax xmin Þ
þ Daj0 ðxmax xmin Þ
; ð62Þ
uM ¼
aj0 þ1 ðxÞUaj aj0 ðxÞUaj þ1 xmin Daj0 xM xmin 0 0 ¼ xmax xmin ðxÞUaj ðxÞUaj þ1 þ Daj0 ðxmax xmin Þ 0
¼
0
aj0 þ1 ððxÞUaj xmin Þ aj0 ððxÞUaj þ1 xmin Þ 0
ðxÞUaj ðxÞUaj 0
0
0 þ1
þ ðaj0 þ1 aj0 Þðxmax xmin Þ
:
ð63Þ
References Chen, S. H. (1985). Ranking fuzzy numbers with maximizing set and minimizing set. Fuzzy Sets and Systems, 17, 113–129. Chen, S. M. (2001). Fuzzy group decision making for evaluating the rate of aggregative risk in software development. Fuzzy Sets and Systems, 118, 75–88. Chen, C. B., & Klein, C. M. (1997a). An efficient approach to solving fuzzy MADM problems. Fuzzy Sets and Systems, 88, 51–67. Chen, C. B., & Klein, C. M. (1997b). A simple approach to ranking a group of aggregated fuzzy utilities. IEEE Transactions on Systems, Man, and Cybernetics – Part B: Cybernetics, 27, 26–35. Detyniecki, D., & Yager, R. R. (2000). Ranking fuzzy numbers using a-weighted valuations. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 8, 573–591.
n1 P i¼0
n1 P i¼0
Dai ððxÞUai ðxÞUaiþ1 ðxÞLai ðxÞLaiþ1 Þ :
ð57Þ
aiþ1 ððxÞUai ðxÞLai Þ
Dong, W. M., & Wong, F. S. (1987). Fuzzy weighted averages and implementation of the extension principle. Fuzzy Sets and Systems, 21, 183–199. Dubois, D., & Prade, H. (1980). Fuzzy sets and systems: Theory and application. New York: Academic Press. Dubois, D., & Prade, H. (1983). Ranking fuzzy numbers in the setting of possibility theory. Information Sciences, 30, 183–224. Filev, D. P., & Yager, R. R. (1993). An adaptive approach to defuzzification based on level sets. Fuzzy Sets and Systems, 54, 355–360. Guh, Y. Y., Hon, C. C., & Lee, E. S. (2001). Fuzzy weighted average: The linear programming approach via Charnes and Cooper’s rule. Fuzzy Sets and Systems, 117, 157–160. Guh, Y. Y., Hon, C. C., Wang, K. M., & Lee, E. S. (1996). Fuzzy weighted average: A max–min paired elimination method. Computers & Mathematics with Applications, 32, 115–123. Guu, S. M. (2002). Fuzzy weighted averages revisited. Fuzzy Sets and Systems, 126, 411–414. Hon, C. C., Guh, Y. Y., Wang, K. M., & Lee, E. S. (1996). Fuzzy multiple attributes and multiple hierarchical decision making. Computers & Mathematics with Applications, 32, 109–119. Kao, C., & Liu, S. T. (1999). Competitiveness of manufacturing firms: An application of fuzzy-weighted average. IEEE Transactions on Systems, Man and Cybernetics – Part A: Systems and Humans, 29, 661–667. Kao, C., & Liu, S. T. (2000). Fuzzy efficiency measures in data envelopment analysis. Fuzzy Sets and Systems, 113, 427–437. Kao, C., & Liu, S. T. (2001). Fractional programming approach to fuzzy weighted average. Fuzzy Sets and Systems, 120, 435–444. Kao, C., & Liu, S. T. (2003). A mathematical programming approach to fuzzy efficiency ranking. International Journal of Production Economics, 86, 145–154. Lee, H. M. (1996a). Applying fuzzy set theory to evaluate the rate of aggregative risk in software development. Fuzzy Sets and Systems, 79, 323–336. Lee, H. M. (1996b). Group decision making using fuzzy sets theory for evaluating the rate of aggregative risk in software development. Fuzzy Sets and Systems, 80, 261–271. Lee, H. M. (1999). Generalization of the group decision making using fuzzy sets theory for evaluating the rate of aggregative risk in software development. Information Sciences, 113, 301–311. Lee, H. M., Lee, S. Y., Lee, T. Y., & Chen, J. J. (2003). A new algorithm for applying fuzzy set theory to evaluate the rate of aggregative risk in software development. Information Sciences, 153, 177–197. Lee, H. M., Lee, T. Y., Lee, S. Y., & Chen, J. J. (2004). A fuzzy group assessment model for evaluating the rate of aggregative risk in software development. International Journal of Reliability, Quality and Safety Engineering, 11, 17–33. Lee, D. H., & Park, D. (1997). An efficient algorithm for fuzzy weighted average. Fuzzy Sets and Systems, 87, 39–45. León, T., Liern, V., Ruiz, J. L., & Sirvent, I. (2003). A fuzzy mathematical programming approach to the assessment of efficiency with DEA models. Fuzzy Sets and Systems, 139, 407–419. Matarazzo, B., & Munda, G. (2001). New approaches for the comparison of L-R fuzzy numbers: A theoretical and operational analysis. Fuzzy Sets and Systems, 118, 407–418. Oussalah, M. (2002). On the compatibility between defuzzification and fuzzy arithmetic operations. Fuzzy Sets and Systems, 128, 247–260. Raj, P. A., & Kumar, D. N. (1999). Ranking alternatives with fuzzy weights using maximizing set and minimizing set. Fuzzy Sets and Systems, 105, 365–375. Raja, P. A., & Kumar, D. N. (1998). Ranking multi-criterion river basin planning alternatives using fuzzy numbers. Fuzzy Sets and Systems, 100, 89–99. Tseng, T. Y., & Klein, C. M. (1992). A new algorithm for fuzzy multicriteria decision making. International Journal of Approximate Reasoning, 6, 45–66. Uehara, K., & Hirota, K. (1998). Parallel and multistage fuzzy inference based on families of a-level sets. Information Sciences, 106, 159–195. Vanegas, L. V., & Labib, A. W. (2001). Application of new fuzzy-weighted average (NFWA) method to engineering design evaluation. International Journal of Production Research, 39, 1147–1162. Wang, Y. M., Greatbanks, R., & Yang, J. B. (2005). Interval efficiency assessment using data envelopment analysis. Fuzzy Sets and Systems, 153, 347–370. Wang, W. J., & Luoh, L. (2000). Simple computation for the defuzzifications of center of sum and center of gravity. Journal of Intelligent and Fuzzy Systems, 9, 53–59. Yager, R. R. (1981). A procedure for ordering fuzzy subsets of the unit interval. Information Sciences, 24, 143–161. Yager, R. R., & Filev, D. (1999). On ranking fuzzy numbers using valuations. International Journal of Intelligent Systems, 14, 1249–1268.
Please cite this article in press as: Wang, Y.-M. Centroid defuzzification and the maximizing set and minimizing set ranking based ... Computers & Industrial Engineering (2009), doi:10.1016/j.cie.2008.11.014