4

FUZZY NUMBERS

4.1 INTRODUCTION

Fuzzy numbers are fuzzy sets used in connection with applications where an
explicit representation of the ambiguity and uncertainty found in numerical
data is desirable. In an intuitive sense, they are fuzzy sets representing the
meaning of statements such as "about 3" or "nearly flve and a half." In other
words, fuzzy numbers take into account the "about," "almost," and "not
quite" qualities of numericallabels. Fuzzy set operations such as union and
intersection, as well as the notions of a-cuts, resolution, and the extension
principie (Chapter 2), are all applicable to fuzzy numbers. In addition, a set of
operations very similar to the familiar operations of arithmetic, addition,
subtraction, multiplication, and division can be defined for fuzzy numbers as
welI. In this chapter we look at such operations and examples of their use.
Fuzzy numbers have been successfulIy applied in expert systems, fuzzy
regression, and fuzzy data analysis methodologies (Kaufmann and Gupta,
1991; Terano et al., 1992). Fuzzy numbers have also been used in connection
with fuzzy equations, and alternative operations of fuzzy arithmetic have
been introduced for the purpose of reducing fuzziness in successive computa-
tions (Sanchez, 1993).
The universe of discourse on which fuzzy numbers are defined is the set of
real numbers and its subsets (e.g., integers or natural numbers), and their
membership functions ought to be normal and convexo We recalI from
Section 2.3 that a fuzzy set is calIed normal if there is at least one point in
the universe of discourse where the membership function reaches unity
[equation (2.3-11)]. But what is a "convex" fuzzy set? The intuitive meaning

77
78 FUZZV NUMBERS REPRESENTING FUZZV NUMBERS 79

of conoexity' is that the membership function of a convex fuzzy ~et does not
go "up-and-down" more than once. Conside~, for example, the fuzzy set~ A
j1(X) and B shown in Figure 4.1a. Fuzzy set A IS convex but not normal since
nowhere in the universe of discourse does its membership function reach
1 unity. Therefore it is not a fuzzy number. Fuzzy set B is normal but not
convex since its membership function goes "up-and-down" twice, and hence it
0.8
is also not a fuzzy number. On the other hand, consider the set C shown in
Figure 4.1b. It is both normal and convex and therefore may be considered a
0.6 fuzzy number. We will see in following sections that changing the shape of a
membership function results in a different number. "Shape" is what fuzzy
0.4
numbers are all about, and fuzzy arithmetic may be thought of as a way of
computing with "shapes" (arcas) instead of "points" (we consider crisp
0.2
numbers as "points".
Fuzzy numbers may also be defined on a multidimensional universe of
O X discourse that is a Cartesian product. Such fuzzy numbers are used, for
O 2 4 6 8 10
example, in connection with scene analysis and robotics to define the mean-
ing of a region in space, or a domain on the x-y plane, and also to add,
subtract, and multiply regions (PaI and Majumder, 1986). In this chapter,
however, we consider fuzzy numbers defined on a simple, one-dimensional
(a) universe of discourse. A very comprehensive treatment of fuzzy numbers,
including multidimensional ones, may be found in the book entitled Introduc-
tion to Fuzzy Arithmetic by Kaufmann and Gupta (1991).
j1(X)
l~--------+-----~~--r-----~--------+---------r
convex 'fui normal 4.2 REPRESENTING FUZZY NUMBERS

::r::I:II
I
We denote fuzzy numbers by boldfaced italics-for example, 3 or A-or by
referring to their membership function. As we said earlier, fuzzy numbers are

0.4 ·····························T··············· "!j j j

······T····························!·················· .
fuzzy sets used to represent the "about," "almost," or "nearly" qualities of
numerical data. We observe, however, that there are many possible meanings
to a statement such as "about 3." Therefore, several different sets may be
0.2 j L L l.. . used to represent "about 3." In the context of fuzzy arithmetic operations,
~ 1 l ;
~ ; : i however, at any given time we use only one meaning, chosen on the basis of
o-r--------~:----~---+--------~---------+--------~ X application-specific criteria and needs. Figure 4.2a shows a triangular mem-
O 2 4 6 8 10
bership function representing the fuzzy number 3. Another possible repre-
sentation is the bell-shaped membership function in Figure 4.2b. These are
two different 3's. If we start a computation using the triangular 3, we cannot
halfway through switch to the bell-shaped 3. Note that on both instances the
(b) shape of the membership function meets our normal and convex require-
Figure 4.1 (o) Two fuzzy sets that cannot be used as fuzzy numbers. (b) A fuzzy set
that may be used as fuzzy number. lThe notion of convexity is derived through references to geometrical objects. A body n in
Euclidean space is called convex if the line segment joining any two points of n lies in n.
Examples of convex bodies in three-dimensional space are the sphere, the ellipsoid, a cylinder, a
cube, and a cone.
 80 FUZZV NUMBERS REPRESENTING FUZZV NUMBERS 81

f.l(x) 3 Table 4.1 Tabular representaHon o, a fuzzy number 3

l;----+--~~--i---~--_+----~--+---~--~--~
0.8

a=l
0.6
a=O.9
0.4 a=O.8
a=O.7
0.2
a=O.6
a=O.5
1
x a=O.4

a=O.3
(a) a=O.2
a=O.l
a=O

f.l(x) 3
l;---~--~--~~--~ __+-__~ __~ __~ __~~ __~

0.8
ments. Another possible fuzzy number 3 is shown in Table 4.1, where the
0.6 shaded cells, the l's, indicate the shape of the number. Here 3 is defined
over the universe of natural numbers shown at the bottom of the table. In the
leftmost colurnn we list the values of a para meter, a, ranging between O and
1, used to parametrize the shape of the function (Kaufmann and Gupta,
1991). In fact, this is the same a we saw in connection with o-cuts (see
Section 2.6). The o-cuts of fuzzy numbers are very useful in fuzzy arithmetic
operations. Looking at Table 4.1 we see that the grade of membership of
crisp number 4 to the fuzzy number 3 is 0.7, and the grade of membership of
crisp 3 is 1.0. Although the fuzzy numbers shown in Figure 4.2 and Table 4.1
x are all different, we designate them with the same symbol (í.e., 3) since they
all peak at crisp 3 (Zimmermann, 1985; Kandel, 1986).
Fuzzy numbers, like any fuzzy set, may be represented by its a-cuts. We
(b) saw in Chapter 2 that a membership function may be parameterized by a
Figure 4.2 Two different fuzzy numbers: (a) triangular 3 and (o) bell-shaped 3. parameter a in a manner similar to the tabular representation of number 3
shown in Table 4.1. The parameter a is a number between O and 1 (i.e., in
the interval [O,1]). Parameterizing the shape of a fuzzy number by a offers a
 REPRESENTING FUZZV NUMBERS 83
82 FUZZV NUMBERS

where the symbol c denotes that the interval [a~a2l,a~a2l] is contained

convenient way for computing with fuzzy numbers because it essentially
transforms fuzzy arithmetic operations into operations of interval arithmetic. within the interval [a~all, a~all].
It is easy to see what we are talking about by looking at Table 4.1. At each We can uniquely describe two fuzzy numbers A and B as two collection of
level a we have a horizontal "slice," or interval of the membership function, intervals i.e., [a~al, a~al] and [b~al, b~al] respectively. We recall that the o-cuts
which is its o-cut. For example, at a = 0.5, the o-cut is the interval from 2 to of A and B (Section 2.6) were defined as the crisp sets
4, and at a = 0.2 it is the interval from 1 to 6. The tabular representation
( 4.2-2)
exemplifies the length of each o-cut; that is, it shows the number of cells and
thus the length of the membership function at level a.
Consider the fuzzy number A shown in Figure 4.3. The membership and
function of A is parameterized by the parameter a. With each a we identify ( 4.2-3)
an interval [a~al, a~al]. As may be seen from the figure, we indicate by a~al
the left endpoint of the interval ("left" is denoted by the subscript "I") and The o-cuts in equations (4.2-2) and (4.2-3) are simply intervals on the x axis,
by a~al the right endpoint of the interval ("right" is denoted by the sub- and hence for each a we have
script "2"). Requiring that the membership function of a fuzzy number be
convex and normal is another way of saying that the intervals that comprise A
ai'
= [a(al a(a
2
l] ( 4.2-4)
the interval representation of A should be nested into one another as we
move from the bottom of the membership function to the top (Klir and and
Folger, 1988; Terano et al., 1992). In other words, when ai < a2' as shown
B = [b(al b(al] ( 4.2-5)
in Figure 4.3, we have ai' 2

Thus the fuzzy numbers A and B can be described (using the resolution
(4.2-1 ) principle-see Section 2.7) as collections of intervals, that is,

A= V a'Aa = V a' [a~al,a~al] ( 4.2-6)

a, tüx) O~a~l O~a~l
A
1 ---------------t------. and

B= V a·B" = V a' [b~al, b~al] ( 4.2-7)

O~a~l O~a~l

a2t----------JL--~~--------------t- To simplify matters, we will not use the rather awkward representation of the
two numbers given by equations (4.2-6) and (4.2-7) but will, instead, use
equations (4.2-4) and (4.2-5), which we call the a-cut or interval representation
of A and B (with the understanding that the number is the colIection of all
"slices," all o-cuts as a varies from O to 1).
alt----------~I_--4_----_+----~~------------------t- Having two different ways of representing fuzzy numbers, through mem-
bership functions and through a-cuts or intervals, gives us the choice of
defining arithmetic operations either through the extension principie
(i.e., through a fuzzification of arithmetic operations on crisp numbers) or,
equivalently, through the operations of interval arithmetic. This last approach
is often more practical and straightforward as we will see in several examples.
Let us go next to the definition of addition, subtraction, multiplication, and
division with fuzzy numbers. Although we will define operations for two
numbers A and B, they are generally true for more than two numbers. A
Figure 4.3 Nested íntervots (o-cuts) ossocioted with a fuzzy number A. word of caution: Some of the properties of crisp numbers-for example,
 ADDITION 85
84 FUZZV NUMBERS

(7 -i- 3) x 3 = 7.-may not be valid for arithmetic operations involving fuzzy . (43 4) tells us that to compute the grade of membership of a
Equa.tlOn. . - b to the fuzzy number C we take the maximum of the
numbers. We will see that usually when fuzzy numbers are involved we have rtalll cnsp num er Z '. dd
that (1 -i- 3) X 3 may not equal 7. ce. f h des of membership of all pairs x and y which a up to z.
nUlll
ma o t. e g(r: 3-4) works will be seen in Example 4.2, where a rather
How equatlOn . .' '11 b d
of carrying out the max-rnm operations WI e presente .
simple ta b u 1ar wa Y
4.3 ADDITION
Addition of Discrete Fuzzy Numbers. Let us compute the sum
Examp Ie 4 . 1 .
When adding two fuzzy numbers A and B we seek to compute a new fuzzy C of two fuzzy nurnbers A = 3 and B = 7 defined as
number C = A + B. The new number C is uniquely described when we
obtain its membership function, ILc(Z) == ILA+B(Z), with Z being the crisp A = 3 = 0.3/1 + 0.7/2 + 1.0/3 + 0.7/4 + 0.3/5 + 0/6 (E4.1-1)
sum of x and y, the elements of the universe of discourse of A and B. The
B = 7 = 0.2/5 + 0.6/6 + 1.0/7 + 0.6/8 + 0.2/9 + 0/10 (E4.1-2)
addition of A and B may be defined in terms of addition of the o-cuts of the
two numbers as follows:
and seen in Table 4.2. We compute C by adding the o-cuts of A, B in
accordance with equation (4.3-2). We see from Table 4.2 that when a = 004,
( 4.3-1)
for example, the OA-cuts of A and B are

where [a~a), a~a)l is the collection of intervals representing the fuzzy number (E4.1-3)
A, and [b~a), b~a)l is the collection of intervals representing the fuzzy number
B. Intervals are added by adding their corresponding left and right endpoints,
and
and therefore equation (4.3-1) becomes
(E4.1-4)
A +B -- [a(a)
1
+ b(a)l' a(a)
2
+ b(a)]
2
( 4.3-2)
The intervals in equations (E4.1-3) and (E4.1-4) are shown as shaded "slices"
Equation (4.3-2) indicates that the new number is also a collection of of cells in Table 4.2. According to equation (4.3-1) the OA-cut of C is the sum
intervals with endpoints obtained from the endpoints of A and B. of the two intervals given by (E4.1-3) and (E4.1-4)-that is,
Another way of defining fuzzy addition is through the extension principie
(Section 2.5). We give here a cursory description of how this is done; more COA = [a(OA)
l'
a(OA)]
2
+ [b(OA)
l'
b(0.4)]
2
detailed treatments may be found in Dubois and Prade (1980) and in Terano
et al. (1992). Suppose we want to add two crisp numbers x and y. The result _
-
[a(OA)
1
+ b(OA)
l'
a(OA)
2
+ b(OA)]
2
is another crisp number z = x + y. Now, if x and y are variables, obviously
their sum may be thought of as a function of x and y; that is, = [2 + 6,4 + 8]
= [8,12] (E4.1-5)
z(x,y)=x+y ( 4.3-3)
We can obtain the same result from Table 4.2 simply by adding the endpoints
Fuzzifying x and y-that is, defining fuzzy sets on x and y-results in a of the shaded rows. We repeat this for each a to compute the entire sumo
fuzzified function, Z = f(x, y). We saw in Section 2.5 how we can use the We start from the bottom of the table and go up in a row-by-row manner
extension principle to obtain the fuzzy set C on Z = f(x, y), Suppose that we identifying the corresponding intervals of the two numbers and adding the~
have two fuzzy numbers, A and B, defined over x and y (the universe of up. The result is the number shown in Table 4.3. The OA-cut of C IS
discourse of real numbers), According to the extension principie, their sum is indicated as a shaded group of cells in the table. As seen from the table, the
a fuzzy set on Z denoted as C, whose membership function is new fuzzy number reaches unity at crisp number 10 (in the universe of
discourse shown at the bottom) and therefore we think of it as a fuzzy
ILC(Z) == V [ILA(X) /\ ILB(y)] (4.3-4 ) number 10. Thus we see that the sum is 7 + 3 = 10, as would ais o be the
z=x+y case with crisp numbers. O
86 FUlZY NUMBERS ADDITION 87

Table 4.2 Fuzzy numbers 3 and 7 In Example 4.1 Table 4.3 Sum or ruzzy numbers 3 and 7 In Example 4.1

3: 10:
0.3 0.7 1 0.7 0.3 O O O O

a=1.0 1
O O O o O 0.2 0.3 0.6 0.7 1.0 0.7 0.6 0.3 0.2 O O
a=O.9 1
a=1.0 1
a=0.8 1
a=0.9 1
a=O.7 1 1 1
a=0.8 1
a=O.6 1 1 1
a=0.7 1 1 1
a=O.5 1 1 1
a=0.6 1 1 1 1 1
a=O.4
a=0.5 1 1 1 1 1
a=O.3 1 1 1 1 1
a=O.4
a=O.2 1 1 1 1 1
a=0.3 1 1 1 1 1 1 1
a=O.l 1 1 1 1 1 a=0.2 1 1 1 1 1 1 1 1 1
a=O.O 1 1 1 1 1 1 1 1 1 a=O.l 1 1 1 1 1 1 1 1 1

1 2 3 4 5 6 7 8 9 a=O.O 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
7:
o O O O 0.2 0.6 1.0 0.6 0.2 O O

a=1.0 1

a=O.9 1 Example 4.2 Addition of Fuzzy Numbers Through the Extension Principie.
In this exampIe we compute the sum of the two numbers A and B of
a=O.8 1 ExampIe 4.1 using the alternative definition of addition through the exten-
a=O.7 1 sion principIe, nameIy, equation (4.3-4). At first glance, equation (4.3-4) Iooks
somewhat esoteric. We present here a rather simple technique for using it.
a=O.6 1 1 1 The same technique may be used with other fuzzy arithmetic operations as
a=O.5 1 1 1
well (Kaufmann and Gupta, 1991). Let's repeat equation (4.3-4) here:

a=O.4
JLA+B(Z) == V [JLA(X) 1\ JLB(Y)] (E4.2-1)
a=O.3 1 1 1 z=x+y

a=O.2 1 1 1 1 1
A convenient way to compute the sum according to equation (E4.2-1) is to
a=O.l 1 1 1 1 1 create a tabIe as shown in TabIe 4.4. We take the support of B and make as
a=O.O 1 1 1 1 1 1 1 1 1 1 1 many columns in the tabIe as there are eIements in the support; and simiIarly
we take the support of A and make as many rows in the tabIe as there are
1 2 3 4 5 6 7 8 9 10 11 eIements in the support of A. We recall that the support is the part of the
universe of discourse that has nonzero membership. A and B can be
 AUUI I lU1\1 CSY

x "" 2, adding y = 6 and x = 3 and so on. Both elements for each addition
Table 4.4 Addlng tuzzy numbers through the extenslon principie
are found inside a cell. These are the shaded cells shown in Table 4.4.
Equation (E4.2-l) says that for z = 9 we need to take the maximum of the
S u p P o r t o f B núnima of the three pairs of grades of membership inside the shaded cells.
y=l y= 2 y=3 y=4 y=5 y=6 y=7 y=8 y=9 First we find the minimum of the grades of membership inside each cell-that
Y=10
B
is,
A
J-LA(1) 1\ J-LB(8) = 0.3 1\ 0.6 = 0.3
,""", ...,}'.
0.0 0.0 0.0 0.0 0.2 0.6 1.0 0.( 0.2 0.0 J-LA(2) 1\ J-LB(7) = 0.7 1\ 1.0 = 0.7
x=1 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
,I ...,,,,,,\,}
0.0 0.0 0.0 0.0 0.2 0.6 0.6
J-LA(3) 1\ J-LB(6) = 1.0 1\ 0.6 = 0.6
0.2 0.0
S x=2 0.7 0.7 0.7 0.7 0;7 0.7 .0.7 0.7 0.7 J-LA(4) 1\ J-LB(S) = 0.7 1\ 0.2 = 0.2
u 0.0 0.0 0.0 0.0 0.2 I",,',"i':' 1.0 0.6 0.2 0.0 J-LA(S) 1\ J-LB( 4) = 0.3 1\ O= O
P x=3 1.0 1.0 1.0 1.0 1.0 11:& 1.0 1.0 1.0 1.0
J-LA(6) 1\ J-LB(3) = O 1\ O = O
P 0.0 0.0 0.0 0.0 0.6 1.0 0.6 0.2 0.0
.".
O x--4 0.7 0.7 0.7 0.7 ··\i 0.7 0.7 0.7 0.7 0.7 J-LA(7) 1\ J-LB(2) = O 1\ O = O
r 0.0 0.0 0.0 1 <Q:O 0.2 0.6 1.0 0.6 0.2 0.0 J-LA(8) 1\ J-LB(1) = O 1\ O = O (E4.2-2)
x=5 0.3 0.3 0.3 ·0'::3/' 0.3 0.3 0.3 0.3 0.3 0.3
1.0
Now if we look only at the shaded part of the table, we can replace the
0.0 0.0 0.0 0.2 0.6 0.6 0.2 0.0
o x=6 0.0 0.0
~~;~, contents of each cell with the minima found in equations (E4.2-2)-that is,
0.0 0.0 0.0 0.0 0.0 0.0 0.0
0.0 .,,' }V'j 1.0 y=8
}}~ ri
f 0.0 0.0 0.2 0.6 0.6 0.2 0.0
x=7 0.0 B )'.'.'
.. )
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
A li<·p,g 0.0 0.0 0.0 0.2 0.6 1.0 0.6 0.2 0.0
x=8 O.Oi)) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.7
0.0 0.0 0.0 0.0 0.2 0.6 1.0 0.6 0.2 0.0
x=9 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6
0.0 0.0 0.0 0.0 0.2 0.6 1.0 0.6 0.2 0.0
x=lO 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2

O
interchanged in terms or rows and columns, but for the moment let's make
columns from the support of A and make rows from the support of A. In O
every cell of the table we put at the lower left comer the grade of mernber-
ship of x to A and put in the upper right comer the grade of membership of O
y to B. Thus we have J-LA(X) in the lower left comer and J-LaCy) in the upper
right comer as shown in Table 4.4. Now, let's take another look in the
x=8
equation above. It calls for taking the maximum of pairs of singletons that
add up to a certain z. For example, suppose that have z = 9. There are rhree
different ways to get z = 9: adding y = 8 and x = 1, adding y = 7 and
90 FUZZV NUMBERS SUBTRACTION 91

Next, we take the maximum of these numbers, which in this case is 0.7; this is Example 4.3 Subtracting Fuzzy Numbers as Intervals. Let us compute a
the maximum with respect to z = 9 in equation (E4.2-l). At this point we fuzzy number C = 7 - 3, where the fuzzy numbers 7 and 3 are as defined in
have completed the entire operation on equation (E4.2-l) for z = 4-that is, Table 4.2 (Example 4.1):
J.LA+B(9) = [(0.3) V (0.7) V (0.6) V (0.2) V (O) V (O) V (O) V (O)]
A = 3 = 0.3/1 + 0.7/2 + 1.0/3 + 0.7/4 + 0.3/5 + 0/6 (E4.3-1)
= 0.7 (E4.2-3)
B = 7 = 0.2/5 + 0.6/6 + 1.0/7 + 0.6/8 + 0.2/9 + 0/10 (E4.3-2)
This is the grade of membership of z = 9 to the sum C = A + B. We repeat
this procedure for a11other ce11sto obtain the membership function of C. Subtracting the two numbers is the same as interval subtraction at each a.
The result is From Table 4.2 we see that when a = 0.3, for example, the 0.3-cuts of the
two numbers are
C = 0/5 + 0.2/6 + 0.3/7 + 0.6/8 + 0.7/9 + 1.0/10
+ 0.7/11 + 0.6/12 + 0.3/13 + 0.2/14 + 0/15 A 0.3 = [a(O.3)
l'
a(0.3)] =
2
[1, 5] (E4.3-3)

which is the same number as the one we found by the interval approach in and
Example 4.1-that is, the number shown in Table 4.3. O
B 0.3 = [b(0.3)
l'
b(0.3)]
2
= [6, 8] (E4.3-4)

4.4 SUBTRACTlON The a-cut of C at a = 0.3 is the difference of the a-cuts in by (E4.3-3) and
(E4.3-4)
The difference C of two fuzzy numbers A, B may be defined either through
interval subtraction utilizing the a-cut representation of the two numbers or C = [b(0.3) b(0.3)] _ [a(0.3) a(0.3)]
0.3 l' 2 l' 2
through the e.xtension principie. Using a-cuts we subtract them as fo11ows
= [b(0.3) _ a(0.3) b(0.3) _ a(0.3)]
1 2' 2 1
(4.4-1)
= [6 - 5,8- 1]
where is the co11ection of closed intervals representing A, and
[ala), a~a)]
[bla), b~a)] is the
co11ection of closed intervals representing B. Two intervals = [1,7] (E4.3-5)
are subtracted by subtracting their left and right endpoints, and thus equa-
tion (4.4-1) becomes shown as a "slice" of shaded ce11s in Table 4.5. In a similar manner we
compute the o-cuts of C at the other levels of a and obtain the fuzzy
(4.4-2) number

The alterna tive way to define the difference of fuzzy numbers A and B is C = 0.2/0 + 0.3/1 + 0.6/2 + 0.7/3 + 1.0/4
through the extension principle-that is, by fuzzifying a function z = x - y.
Fuzzification means that we define fuzzy sets on the universes of discourse + 0.7/5 + 0.6/6 + 0.3/7 + 0.2/8
where the crisp elements x and y are found. As a result, z gets fuzzified as
we11;that is, there is a fuzzy set C over the universe of discourse of the z's, which is also shown in Table 4.5. As may be seen from Table 4.5, C can be
considered a fuzzy 4. O
which is the result of fuzzifying the function z = f(x, y) = x - y. The mem-
bership function of C = A - B can be computed from
Example 4.4 Subtracting Fuzzy Numbers with Continuous Membership
J.LA-B(Z) == V [J.LA(X) 1\ J.LB(y)] (4.4-3) Functions. Consider the two triangular fuzzy numbers A and B shown in
z-x-y Figure 4.4. We want to compute their difference-that is, find a fuzzy
num~er C = A - B. When continuous (or piecewise continuous) membership
Equation (4.4-3) gives, of course, the same number C obtained through functions are used, we subtract them by parameterizing their membership
(4.4-2).
functions by a and subtracting their o-cuts. The membership functions of
 SUBTRACTION 93
92 FUZZV NUMBERS

Table 4.5 Difference of fuzzy numbers 7 and 3 In Example 4.3 A, B are

Ji..A(X) = O, x~7
4:
=x - 7, 7~x~8
0.2 0.3 0.6 0.7 1 0.7 0.6 0.3 0.2 O = -x + 9, 8~x~9
a=1.0 1 = 0, x;:=:9 (E4.4-1)

a=O.9 1
and
a=0.8 1
Ji..B(X) = 0, x~4
a=0.7 1 1 1
=x - 4, 4~x~5
a=O.6 1 1 1 1 1
= -x + 6, 5~x~6
a=O.5 1 1 1 1 1 x;:=:6 (E4.4-2)
= 0,
a=O.4

a=O.3
li ~
Let us parameterize them by a. Ta simplify matters, consider the left and
right side of each membership function separately. There is one equation for
the left side and another for the right side of the membership function of A,
a=O.2 1 1 1 1 1 1 1 1 1
and likewise for B. Thus, we have a total of four equations ta parameterize.
a=O.l 1 1 1 1 1 1 1 1 1 From equations (E4.4-l) we take the part that describes the left side of A,
Ji..;(x) = x - 7, and write it in terms of a. We note that the value of a is the
a=O.O 1 1 1 1 1 1 1 1 1 1
same as the value af the membership function at the left endpoint a~e ) of an
O 1 2 3 4 5 6 7 8 9
o-cut, and a~"') is the value of x at that point, Thus we have for the left side
of A,

a = a~"') - 7 = a~"') = a + 7 (E4.4-3)

a, J.l(x) /lc(X) where a~"') is the left endpoint of the "slice" of A at level a.
11---~--~---'~--+---~--~--~--~~--~--~ Similarly for the right side of A we parameterize the right endpoint a~"')
,:l'.,
: of each o-cut in terms of a as
a=0.8
a = - a~"') + 9 = a~o ) = - a +9 (E4.4-4)

a=0.6
Using equations (E4.4-3) and (E4.4-4) the e-cut representation of A is
written as
a=0.4
(E4.4-5)
a=02

The membership function of the number B is parameterized in terms of a in

x a similar fashion. We express the left endpoint b~"') in terms of a by
o 1 2 3 4 5 6 7 8 9 10
(E4.4-6)
Figure 4.4 Subtracting two fuzzy numbers, C = A - B.
 94 zz:y NUMBERS
MULTIPLICA TlON 95

The right endpoint 0'f[') is given as a function of a by

The number C described by equations (E4.4-11) is shown in Figure 4.4. Note
that C has its peak at crisp 3, and therefore it can be considered as a fuzzy
a = -b~a) +6 = b~a) = - a +6 (E4.4-7)
number 3 (as expected since, 8 - 5 = 3). O
From equations (E4.4-6) and (E4.4-7) the interval representation of B is

B = [b(a) b(a)1 = [ 4 4.5 MULTlPLlCATION

I , 2 a + ,- a + 6] (E4.4-8)

Fro~ the a~cut.representations of A and B (equations (E4 4- As in the case of addition and subtraction, fuzzy number muItiplication may
we find their difference by subtracting their corr di . .5) and (E4.4-8», be defined either as o-cut multiplication or through the extension principie.
a, that is, espon lllg llltervals at each
Using the a-cut representation of two numbers A, B, their product is
defined as
c=A - B = [a~a) - b}a), a~a) - b~a)1
( 4.5-1)
= [( a + 7) - (- a + 6), (- a + 9) - (a + 4)1
= [2a + 1, -2a + 5] In general, the product of two intervals is a new interval whose left endpoint
(E4.4-9) is the product of the left endpoints of the two intervals and the right
Therefore, C is endpoint is the product of the right endpoints of the two intervals. Thus,
equation (4.5-1) is
C= [c(a) (a)1 - [2
I'C2 - a+1,-2a+5] (E4.4-1O) A .B = faCa) . b(a) a(a). b(a)]
I I' 2 2 ( 4.5-2)
We note that the left. and right endpoints of C are functions of
the fuzzy number C III terms of a membershi functí ": To express Altematively, we define the product of A and B through the extension
we
for the left and right side of C. The left endpo~nt c(at~~~, e d~nve equatio~s principie by fuzzifying the function z(x, y) = x . y. The extension principIe
equal to the value of x when the left- id b I h' qua.tlOn (E4.4-1O) IS tells us that their product is a fuzzy set on z, denoted as A· B, whose
S·lIllilarly,
.
the right endpoint
SI e mem ers lp funcríon'
C}a) is equal to th I
1 .
s va ue IS a. membership function is
right-side membership funetion is Th th e ~a ue of x when the
obtained by setting c~a) = x and re:~llin uS ; eq~atI~n of the left side is J.LA·B(Z) == V [J.LA(X) /\ J.LB(Y)] ( 4.5-3)
th
the left-side membership function for C. ~e :av: - ·J.Lc(x), where J.L~(x) is Z=X'Y

Of course, equations (4.5-2) and (4.5-3) are equivalent in that they give us the
x = 2J.L~(x) + 1 = J.L~(x) = i(x - 1) (E4.4-11)
same number C = A . B.
In a similar manner we obtain an equation fo R() . A special case of fuzzy multiplication is the product of fuzzy number by
mem?ership function of C, and solve it to obtai~ ~he x , t~ ngh~t side ?f the crisp number. Let k be a crisp positive real number and A a fuzzy number
the nght side-that is, mem ers lp functíon of defined over the universe of discourse of positive real numbers also. We
define the product of k with A either as interval multiplication or through
x = - 2 J.L~( x) + 5 = J.L~( x) = - i( x - 5)
(E4.4-12)
the extension principIe. Crisp number k may be viewed as an interval also,
a trivial interval whose left and right endpoints are the same+-that is,
From equations (E4.4-11) and (E4.4-12) we obtain k = [k, k]. We use equations (4.5-1) and (4.5-2) to obtain the product of k
with A as
J.Lc(x) = 0, x:s1
k·A =
-, [k k]· faCa)
I' ara)]
2
= t(x-1),
= [ka(a)I ,
ka(a)]
2 ( 4.5-4)
= -i(x - 5),

= 0, x~5 Altematively, we define the product of fuzzy number A with a crisp number
(E4.4-13) k, k· A, through the extension principie. It may be shown using equation
 MULIIl-'lI\...-I-\IIVI\I 71

7Q r-ULLY I\lUMtlt:ló

(4.5-3) that the membership function of k .A is and

x:o::;O
1
= 2:x, O:o::;x:o::;2
( 4-5-5)
= -tx + i, 2:o::;x:o::;6
x~6 (E4.5-2)
where equations (4.5-4) and (4.5-5) give the same result. =0
Next, we parameterize the membership functions in equations (E4.5-l) and
Example 4.5 Multiplication of Two Fuzzy Numbers. Consider the triangular (E4.5-2) in terms a (a procedure of renaming the left and right side of the
fuzzy n.um~ers A = 8 and B = 2 defined over the positive real numbers as membership functions and thus the endpoints of all intervals in terms of a).
sho,,:n m Figure 4.~ (since both numbers are defined over the same universe Let us take the left and right side of each membership function separately
~~ dlSCOurS~ we simply use x to indicate an element of the universe of and rewrite it in terms of a. It should be noted that a given value of a is the
ls~ou~se, instead of x, y, etc.). We want to compute a fuzzy number C same as the value of the membership function at that leveI. From equation
which IS th.e ~ro~uct of A and B-that is, C = A .B. Let us do this throu h (E4.5-l) we have that the left and right endpoints of A are
a-cut mU~hp~catlOn.-that is: by parameterizing their membership functio~s
1= = 4( a + 1) (E4.5-3)
and .mulhplym~ their a-cuts. in the manner indicated by equation (4.5-2). a -- laCa) -
4 I
a(a)
1
First, we wnte the analytical expressions for the membership functions of
A and B: and
a = - ta~a) +3= a~a) = -4( a - 3) (E4.5-4)

x:o::;4
Using equations (E4.5-3) and (E4.5-4) we obtain A as
= lx - 1, (E4.5-5)
4
A = [a\a), a~a)] = [4( a + 1), -4( a - 3)]
8:0::; x:o::; 12 Similar1y, we parameterize the membership function of B and write its left
and right endpoints at each a as
= O, x ~ 12 (E4.5-1)
(E4.5-6)

J.l(x) B and
A (E4.5-7)
a -- _lb(a) + 12 = b(a) = -4(a - 1)
ltT--rr~~----~--+---~---+--~----L 42 2 2

Thus, from equations (E4.5-6) and (E4.5-7) the interval representation of

0.8
B is
0.6 (E4.5-8)

0.4 Having the endpoints of A and B in terms of a, we multiply the two

numbers using equation (4.5-2) and obtain
0.2
C= A .B = l-r": b\a), a~a). b~a)]

°lo~~~---j~--1---~----4---~--~~---130 40 50 60 70 80
= [4( a + 1) ·2 a, - 4( a - 3) . (- 4( a - ~))]

= [8a2 + 8a, 16a2 - na + n] (E4.5-9)

Figure 4.5 The product C = A . 8 of numb ers A = 8 and 8 = 2 in Example 4.5.
x
 DIVISION 99
98 FUZZV NUMBERS

The interval representation of C is 4.6 DIVISION

fi d th quotient of two fuzzy numbers A and B either through
We can lD e ... I f hei
di .. or by the extension principle. n terms o t err o-cut
interva 1 ivtston .
tation we write the quotient of the two numbers as
represe n ,
where the left and right endpoints in equation (E4.5-10) are functions for a.
We can obtain the membership function of C as well. Equation (E4.5-10) ( 4.6-1)
provides us with left and right endpoints of each o-cut. The equation for the
left-side membership function JLÉ(x) is obtained by setting c~a) = x and
recalling that a = JLÉ(x). Thus, we obtain an equation involving JLÉCx), In general, the quotient of two intervals is a new interval given by
which is
a~a) a~a)l
[a~a), a~a)] -ê- [b~a), b~a)] == [ b~a)' b~a)
(E4.5-11)

Solving quadratic equation (E4.5-11) for JLÉ(x), we obtain two solutions and Hence, provided that b~a) "* O and b~a) "* O, the quotient of A, B is
accept only the value of JLÉ(x) in [0,1], ignoring the other one. The result is

( 4.6-2)
JLÉ( x) = - ~ + ~J 1 + ~x (E4.5-12)

Similarly we obtain an equation for JL~(x), the right side of the membership Altematively, we find the quotient of A and B through the extension
function of C, and solve it, keeping the solution which is within [0,1]. The principle by fuzzifying the function z(x, y) = x + y, where x an? y ar~ c~isp
result is elements of the universe of discourse of A and B. The extension principle
tells us that A -i- B is a fuzzy set with membership function
JL~(x) = ~(4.5 - ';(4.5)2 - 4(4.5 - ftx)) (E4.5-13)
JLA+B(Z) == V [JLA(X) /\ JLB(Y)] ( 4.6-3)
z~x+y
The membership function of C is
The results obtained through equations (4.6-4) and (4.6-2) are of course the
JLc( x) = O, x::; O same. Equation (4.6-3) may be used in the manner shown in Example 4.2. We
construct a table such as Table 4.4 and proceed as outlined in the example. A
O sx s 16 word of caution: Fuzzy number division is not lhe reverse of multiplication;
that is, generally it is not true that (A B) x = A. -ê-

= ~(4.5 - ';(4.5)2 - 4(4.5 - ftx)), 16 sx ::; 72

Example 4.6 Division of Fuzzy Numbers. Consider the triangular fuzzy
O, numbers A = 8 and B = 2 used in Example 4.5. Let us find C = A -i- B
= x ~ 72 (E4.5-14)
using interval division. The analytical expressions for the membership func-
tions of A and B are given in Example 4.5 [equations (E4.5-l) and (E4.5-2)],
as shown in Figure 4.5. It should be noted that C has its peak point at crisp
and their parameterized interval representation is found in equations (E4.5-5)
16 and therefore may be considered a fuzzy number 16. It should ais o be
and (E4.5-8), which for convenience we repeat here:
noted that multiplying two fuzzy numbers results in a new number whose
shape has been considerably changed, no longer having a triangular member-
ship function with linear sides but in this case parabolic sides. Multiplication
A = [a~a),a~a)] = [4(a + 1), -4(a - 3)] (E4.6-1)
in general has the effect of "fattening" the lower part of the membership
B =[b~a),b}a)] = [2a, -4(a - ~)] (E4.6-2)
functions involved. D
 100 FUlZY NUMBERS MINIMUM AND MAXIMUM 101

Thus their quotient C = A -i- B is obtained using equation (4.6-2): membership function of C is

C=A -;- B =
a(a) a(a)
b~a)' b~a)
1 JLc( x) = O, x::;; O
[
~x - 1
0::;;x::;;4
_ [
-
4( a + 1)
( -4 ( 0'-2 3))'
- 4( a - 3) 1 _
(E4.63)
x + 1 '
20' 6
4::;; x::;; 72
The «-cut representation of C is x + 2'
= O, x ~ 72 (E4.6-7)
C = [(a)
C1 'C2
(a)] = [_
(3)' + 1) _ ~--..:..
(a 2( 0'- 3) 1 (E4.6-4) It should be noted from equation (E4.6-7) that the quotient is a new fuzzy
0'-2 a
number that no longer has a triangular shape with linear sides. As may be
where the left and right endpoints are functions of a. We may also express C seen from the figure, the fuzzy number C only asymptotically reaches zero
in terms of a membership function by deriving equations for the left and right and hence we may consider the use of a levei fuzzy set (Chapter 2) in order to
sides of the membership function as we did in Example 4.5. Equation limit and exclude trivially small grades of membership-for example, less
(E4.6-4) gives us the endpoints of the interval of each a-cut. The equation of than 0.2. O
the left side is obtained by setting cla) = x and recalling that a = JL~(x),
where, JL~(x) is the left side membership function for C. The· result is
4.7 MINIMUM ANO MAXIMUM
lx - 1
JLL(X) = _2 __
+1 (E4.6-5)
C X The minimum and maximum of two fuzzy numbers A, B result in finding the
Similarly we obtain an equation for JL~(x), the right side of the membership smallest and the biggest one, respectively, and may be defined either through
function of C, and solve it to obtain their interval representation or by the extension principie. In interval arith-
metic the minimum of two intervals is a new interval whose left endpoint is
6 the minimum of the left endpoints af the original intervals and whose right
JL~(x) = -- (E4.6-6)
x+2 endpoint is the minimum of the right endpoints of the two intervals. Thus the
The quotient is shown in Figure 4.6, and the analytical description of the minimum of A, B is a new number, A /\ B, given by
A /\ B == [a\a), a~a)] /\ [bla), b~a)]
J.l(x)
B A - [a(a) /\ b(a) a(a) /\ b(a)] (4.7-1)
- 1 l' 2 2

111 A I ~ ! ! , I ! ! I Alternatively, the minimum of two fuzzy numbers may be obtained through
0.8-··· the extension principIe. The membership function of A /\ B is

JLAIIB(Z) == V [JLA(X) /\ JLB(Y)] (4.7-2)

o. z=x r.y

In an analogous manner we define the maximum of two fuzzy numbers A

0.4·
and B, recalling that in interval arithmetic the maximum of two intervals is a
new interval whose left endpoint is the maximum of the left endpoints of the
0.2 original intervals and whose right endpoint is the maximum of the right
endpoints of the two intervals. Thus the maximum A v B is given by
O,' I I ;
I
I ;
r
;
i
;
I r
! i
I
~
i
:1- A v B == [a\a), a~a)] v [b\a), b~a)]
o 10 20 30 40 50 60 70 80
x = [Q.(a)
1
V b(a)
l'
a(a)
2
v b(a)]
2 (4.7-3)
Figure 4.6 The quotient C= A -i- 8 of the fuzzy numbers A = 8 and 8 = 2 in Example
4.6.
Alternatively, by the extension principie the membership function of the
 PROBLEMS 103
102 FUZZVNUMBERS

, d F I T A Fuzzy Sets Uncertainty, and lnformation, Prentice-Hall,

p(x) Klir, G. J., an o ger, . ., ,
B A Englewood Cliffs, NJ, 1988. ,
l~----.r-----~----~----~----~--~----~-----L----~----4- ' mder D K D Fuzzy Mathematical Approach to Pattem Recogni-
Pai, S. K ., an d M aju ,'"
tion John Wiley & Sons, New York, 1986.
' E Non Standard Fuzzy Arithmetie, in Between Mind and Computer-Fuzzy
Sane h ez, ., d W Id S' ifi
Science and Engineering, P-Z. Wang, and K-F. Loe, e s. ar cientínc,
Singapore, 1993, pp. 271-282.
Asai ,.,K and Sugeno " M. Fuzzy Systems Theory and lts Applications,
T erano,.,T
Academic Press, Boston, 1992.
Zimmermann, H. J., Fuzzy Set Theory and its Applications, K.luwer-Nijhoff, Boston,
1985.

PROBLEMS
2 4 6 8 10 12 14 16 18 20
x
1. The fuzzy numbers A and B are given by
Figure 4.7 The minimum and maximum of the two numbers A = 8 and 8 = 2 used in
Example 4,6, A = + 0.67/7 + 1.00/8 + 0.67/9 + 0.33/10
0.33/6
B = 0.33/1 + 0.67/2 + 1.00/3 + 0.67/4 + 0.33/5
maximum of the two numbers A and B is
Subtract B from A to give fuzzy number C. Draw a sketch of C.
J.LAVB(Z) = V [J.LA(X) 1\ J.LB(Y)] (4.7-4) 2. Multiply fuzzy numbers A and B of Problem 1. Draw a sketch of C.
z=x v y
3. Divide fuzzy number A by fuzzy number B where the fuzzy numbers are
It should be noted that the maximum and minimum of two fuzzy numbers defined in Problem 1. Draw a sketch of C.
are different than the maximum and minimum of membership functions used
in connection with the union and intersection of two fuzzy sets. Let us 4. Modify Example 4.2 to subtract the two fuzzy numbers using the extension
illustrate this by finding the minimum of the numbers A = 8 and B = 2 used principie.
in Examples 4.5 and 4.6 and redrawn in Figure 4.7. Equations (4.7-1) or 5. Consider the fuzzy numbers A and B described by the membership
(4.7-2) do not give us the littIe wedge between A and B, which is the
functions:
intersection of A and B. They will simply give us the number B = 2 itself,
which is the smallest of the two fuzzy numbers. Similarly the largest of the J.LA(X) = O, x s 8,
numbers is found by using the maximum operation of either equation (4.7-3)
1 8
or (4.7-4), which is simply the number A = 8, as shown in Figure 4.7. For =-x-- 8 s x s 18,
more intricately overlapping membership functions the maximum or mini- 10 10'
mum may not simply be a number with the membership function of either A 1 32
or B, but may have a totally new shape (Kaufmann and Gupta, 1991). = - --x + -- 18 s x s 32,
14 14'
= O, X> 32,
REFERENCES J.LB(X) = O, x s -3,
1 1
Dubois, D., and Prade, H., Fuzzy Sets and Systems: Theory and Applications, Aeademic
= gX-+ 3' -3 s x s 6,
Press, Boston, 1980.
Kandel, A., Fuzzy Mathematical Techniques with Applications, Addison-Wesley, 1 4
Reading, MA, 1986. = --x +- 6 s x s 24,
18 3'
Kaufmann, A., and Gupta, M. M., lntroduction to Fuzzy Arithmetic, Van Nostrand
Reinhold, New York, 1991. = O, X> 24
104 FUZZY NUM:.'I'RS

Compute:
(a) A (+) B,
(b) A (-) B,
(c) A (-;-) B.

6. Repeat the computations in Problem 5 for the fuzzy numbers A and B

given below, and using C state and show the distributivity property (with
respect to addition and multiplication)

A = 0.6/1 + 0.8/2 + 1.0/3 + 0.6/4

B = 0.5/0 + 0.7/1 + 0.9/2 + 1.0/3 + 0.4/4
C = 0.7/1 + 0.8/2 + 1.0/3 + 0.3/4