Академический Документы
Профессиональный Документы
Культура Документы
253
C O M P A R I S O N OF F U Z Z Y R E A S O N I N G M E T H O D S *
Masaharu MIZUMOTO
Information Science Center, Osaka Electro-Communication University, Neyagawa, Osaka 572,
Japan
Hans-Jfirgen ZIMMERMANN
Lehrstuhl fiir Unternehmensforschung, RWTH Aachen, IV. Germany
Received March 1981
Revised June 1981
L.A. Zadeh, E.H. Mamdani, and M. Mizumoto et al. have proposed methods for fuzzy
reasoning in which the antecedent involves a fuzzy conditional proposition 'If x is A then y is
B,' with A and B being fuzzy concepts. Mizumoto et al. have investigated the properties of
their methods in the case of 'generalized modus ponens'.
This paper deals with the properties of their methods in the case of 'generalized modus
tollens', and investigates the other new fuzzy reasoning methods obtained by introducing the
implication rules of many valued logic s)~tems. Finally, the properties of syllogism and
contrapositive are investigated under each fuzzy reasoning method.
Keywords: Fuzzy reasoning, Fuzzy conditional inference, Generalized modus ponens,
Generalized modus tollens, Syllogism, Contrapositive.
1. Introduction
In o u r d a i l y life w e o f t e n m a k e i n f e r e n c e s w h o s e a n t e c e d e n t s a n d c o n s e q u e n c e s
contain fuzzy c o n c e p t s . S u c h an i n f e r e n c e can n o t b e m a d e a d e q u a t e l y b y t h e
methods w h i c h a r e b a s e d e i t h e r o n classical two v a l u e d logic o r o n m a n y v a l u e d
logic. In o r d e r to m a k e such an i n f e r e n c e , Z a d e h [1] s u g g e s t e d an i n f e r e n c e r u l e
called ' c o m p o s i t i o n a l r u l e of i n f e r e n c e ' . U s i n g this i n f e r e n c e rule, he, M a m d a n i
[2] and M i z u m o t o et al. [ 3 - 6 ] s u g g e s t e d s e v e r a l m e t h o d s for fuzzy r e a s o n i n g in
which t h e a n t e c e d e n t c o n t a i n s a c o n d i t i o n a l p r o p o s i t i o n with fuzzy c o n c e p t s :
Antl:
A n t 2:
IfxisAthenyisB
xisA'
Cons':
y is B '
(1)
* This work was attained with the assistance of the Alexander yon Humboldt Foundation.
0165-0114/82/0000-0000/$02.75
1982 N o r t h - H o l l a n d
254
where A, A ' , B, B' are fuzzy concepts. An example of the fuzzy reasoning is the
following.
Ant 1:
Ant 2:
Cons:
(2)
In [4-6] we have pointed out that for the type of fuzzy reasoning in (1) called
'generalized modus ponens', the consequences inferred by Zadeh's and Mamdani's methods are not always reasonable and suggested several new methods R~,
Rg, R~, Rsg, Rg, and Rgg which coincide with our intuition with respect to several
criteria.
As continuation of our studies, this paper investigates the properties of their
fuzzy reasoning methods in the case of 'generalized modus tollens'. Moreover, by
introducing the implication rules of many valued logic systems [7-9], we discuss
the newly obtained fuzzy reasoning methods in the cases of generalized modus
ponens and generalized modus tollens. Finally, we discuss the properties of
syllogism and contrapositive under each fuzzy reasoning method.
IfxisA
xisA'
Cons:
y is B'
then y i s B
(3)
where x and y are the names of objects, and A, A', B and B' are fuzzy Concepts
represented by fuzzy sets in universes of discourse U, U, V and V, respectively.
This form of inference may be viewed as a generalized mod,s ponens which
reduces to modus ponens when A ' = A and B ' = B .
Moreover, the following form of inference is also possible which also contains a
fuzzy conditional proposition.
Ant 1:
Ant 2:
IfxisA
yisB'
Cons:
x is A '
then y i s B
(4)
Comparisono[[uzzy reasoningmethods
255
A=~txa(u)/u,
B=~ttu(v)/v
(5)
and let x, U, f-I, --1 and 9 be cartesian product, union, intersection, complement and bounded-sum for fuzzy sets, respectively. Then the following fuzzy
relations in U x V can be derived from the fuzzy conditional proposition "If x is
A then y is B " in Ant 1 of (3) and (4). The fuzzy relations R,,, and R, were
proposed by Zadeh [1], Rc by Mamdani [2], and R~, Rg, R,g, Ugg, Rg~ and R~s are
by Mizumoto et al. [3-6].
R,, = ( A x B) U (-hA x V)
[
(I~A(tt)AtXI3(V))V(I--I.tA(tt))/(tt, V).
(6)
"O xV
R,~ = (-hA x V ) ~ ( U x B )
[
1^(1-txa(u)+t.tu(v))/(u , v).
(7)
9i o x V
Rr = A x B
(8)
Rs=AxV=), UxB
S
= Iu [m,(u) ~ u. (v)]/Cu.v).
(9)
where
~A(u)<~u.(v),
~A(u) > u.(v).
Rg=AxVOU
g
= Iu [~(.) ~
(10)
u.(vl]l(u, v),
where
~(u) 7u.(v)=
~.(v)
U~(U)>U.(V).
R~g = ( A x V O U x B ) N ( T A
s
x V ~ Ux--1 B)
g
(11)
Rgg=(A V ~ U x B ) N ( " n A x V ~ U x ~ B )
g
(12)
256
R~, = ( A x V ~ U x B ) N ( - 1 A x V O U x ~ B )
g
= L
(13)
R~ = ( A x V ~S U x B ) f q ( ' - n A x V ~$ U x ~ B )
= Iu
(14)
Rb=(~AxV)U(UxB)
= [
au xv
(1--/XA(U))V~t~(V)/(U, V).
(15]
Ra=A215
Iu
(16
where
1
~A(U)<~ttB(V),
UxB
A
(17
where
/"tA(I'l) ~ ~B(I)) = [/"J"A(") ~ /IB('0)]A [1 - - l-tB(I))"~' 1 - P,A(u)]
"
m,(0
1 A ~ A
l-,~,(u)
ViA(u) 1-/zB(v)
,1
R,=AxV~,
p.A(u)>O,
1 - - / . t B (U) > O,
IZA(U)----O or 1--m](v)=O.
UxB
(1~
257
where
/XA(U) .--->/~B(V) = 1 -- ~A (U) +/~A (U)~ZB(V).
R,z=AX V ~ U x B
(19)
where
tLa (U) -~ tLB(V) = (/XA(U) ^/XB (V)) V (1 -- tLA(U) A 1 -- tLB(V))
V (tLB(V) ^ 1 --/xa (u))
= (1 - txa (u) v lab (v)) ^ (txa ( u ) v 1 - - / x a ( u ) )
R r n = A x V ~[] U x B
(20)
where
{~
OA(U)-~ ~B(V)=
~ A ( u ) < l or /ZB(V)=I,
/~A(U)= i, OB(V)<I
Fig. 1 shows the graphs of the 15 fuzzy relations obtained in (6)-(20). In each
graph the symbols P'A and bib are used instead of /ZA(U) and tie(v) for simplicity.
Each left figure with parameter btB will be found to be useful to discuss the
generalized modus" ponens in (3), and each right figure with parameter P-A is
useful to analyze the generalized modus tollens in (4).
In the generalized modus ponens of (3), the consequence B' in Cons can be
deduced from Ant 1 and Ant 2 using the max-rain composition %' [1] of the fuzzy
set A' and the fuzzy relation obtained in (6)-(20). For example, we can have
B~ = A' o Rm
(21)
(22)
(23)
B c' - A ' o Rc
(24)
=A'
o (A x B),
(25)
Similarly, in the generalized modus tollens of (4), the consequence A ' in Cons
can be deduced using the composition 'o' of the relation and the fuzzy set B'.
258
pB =
1
~A =
PA =
.1
\,.."
.8
.7
.6
.5
95
"---.6
---,5
.4
3
.2
.8
99
~A
.2
.3
"---~~ B
!
~B =
1
.9
8
7
6
5
.4
.3
.2
PA =
0
.2
.3
.4
.5
.6
.l
i
~' P B
PB =
1
///
//
~A =
1
.7
.6
.5
//
.4
.3
.1
/
PA
,/
//
.6
.5
.4
.3
,/
//
.7
laB
.2
.1
I
259
i l l
I
I
I
I I I
lug;=
, ~, ~
iF~',- !, ','
'
r!
I I
BATM
,,
; :
.,I . / l~'
..1.
e91
.i 4~.5:.6:.7:.~,.9
!i i i ! , ,I ; I! :
I
!
!
I
i~
,
'
i
:'!
i
i
i
'I. ', ;
!; l',' ,. l', ,
, ! ; ' '
i ! ', ! '
i ! ', ! i
~I
I I I
'
;
i
',
! i' i
'
"~- PA
~" P B
PB =
I
.?
.3
!
I
I
I
:~-9
'
' ;
I
I
i
~A=.~
.7
.6
' "'*5
.4
!
!
'.3
.2
.I
O
!
PA
l'
, "-'~PB
,
1
PI='I'2
9
.3 .4.
--~ p B = . l
o.2
o.3
o.4
o.5
o.6
~.7
o.8
i
' PA
.6
7 .8 9 1
I
I
I
I
I
I
I
I
I
PB
260
9B =
.1 --?
~-.9
|. . . . . . . . .
-~B =
.2 , - - F - - - ?
I
I
,o , ', .7
.3
i
I
i
, , , o
~ :' :' : . 6
.4
i
f
i
!
!
.5 -- I I i i ~
o'
,5
"'7-'", , .
4' ~ : ''
.4
FI
1I
--,
.6,,I:
7 "-i---,
'
' ~'
, ,
"'T--~
.9
"
"--"-'~
.@
,.!
'
laA,.,
.i',.2,.
.3
.2
,
I
l,
9B = . 9 o-,7
.6 ~
AI
I
I
I
I
I
I
I
I
l
I
l
I
:
I
I
I
I
I
i, /
.5 o
.4 o
.3Q
I
I
I / lIl
I-"
[
~A
I-~
I
I
I
I
I
I
I
I
I
I
I
I
I
I
l
I
l
I
I
I
I
!
i
' '
I
I
lo
I
I~
I
/
I
I
l
I
'
~Y
I
I
I
I
I
l
I
I
I
I
r
I
I
I
I
I
I
9I,.2,.3'=.4~ 5 I.6..7,.8,.
9,1
=
I
~A .i'.2
I
I
I
I
l
.31.4
!
!
!
l
I
!
!
I
1
~A
~> laB
1
~
.8
.7
\\
.6
.5
.4
o2
.1
I~A=
0
.1
.2
.3
.4
.5
.6
.7
261
.8
.9
0
Fig. l(xii). R j : / x a ~ gs (see (17)).
262
.8
.2
.5
94
.2
", ~PA
.8
kk~ "I
1
laA~
.I, .9
.2, .8
93, .7
94, .6
.5
.8
oPA,=1
/.---~
._e...i, .9
.2
~llll.
2 , . 8
.7
.6
.5
~ .%. . . . . . . .
%-
a3
~"
.4, .6
.5
95
o6 . . . . . . . .
.is
-S S
e 7 ......
sS
PA
"~-- 9
"~
"" lab
}XB=l
laA~ 1
laA=I
laB< i
~'PA
0
Fig. l(xv). Rm: taA El" tan (see (20)).
lab
263
Namely,
A mt - R,,, o B
(26)
(27)
(28)
(29)
--
A 2= II
(30)
P'A(it)2[ll'
(31)
(32)
(33)
(34)
(35)
(36)
264
(37)
B ' = B = i , I.tB ( v ) / v .
The consequences inferred by all the fuzzy reasoning methods are summarized
in Table 1 (The case of generalized modus ponens) and Table 2 (The case of
generalized modus tollens), in which #B and /xA stand for p.B(v) and /xa(u),
respectively. These results are also depicted in Fig. 2 in order to make their
comparison more transparent. Each left figure shows the results inferred by R,,,
Re, Rs, Rg, R~g, Rgg, Rg,, R~,, Rb and Re, and each right figure shows the results
by R,, Ra, R , and R.. The figure of Rm is omitted because of its simplicity.
In Table 1 the inference results by the method R,t at A ' = A , very A and mor~
or less A can be more precisely rewritten as follows.
3-,,/5
__
"/~B A 2 J /1B"=
~
2 - Ixa
(= 0 . 3 8 1 9 . . . ) ,
3-'~
~B ~> 2
/x~ 3
~LI213A I x / 5 - 4/-tB-1] 2 /
L" 2-~--~B) J = ~ [ x / 5 " - 4 t x B - 1 ]
/.tB <~bo,
very A
more or less A
not A
R~
0.5v~
3-,./5
~ - ~ v/.tB
,,,/5-1
"~----V ~tn
Ra
R~
l+gB
2
~tB
3+2~tB-~
2
~B
"J'5~/aa r- 1
.2
/xB
0.5 ^ ttb
Rg
R~
Rag
Ra~
R.
IxB
~
/.tB
~B
~n
x/~a
0.5 vl-t~
~
~
~
~
qc5-1
TVl-tB
1- ~
1 -/.t B
1 -/x~
1 - ~tn
Rb
~tn
~
/.t~
t'tB
~
3-x/5
Tv~B
R,
"~~
R,
1
2-#a
R,~
0.5vg~
Rn
[ . ~ -
112
B tt2B/3^L 2(1-g~) J
+5qz.a- 1 1
tt~/3"~//a'~21xa2
gBv (1-ga)^
~-BV(1--g.B)
where
3~
265
31
bo = 1 - ~/~+--]-~--- ~/~
(38)
~83 =0.3177...
t-tB ~<b~,
{/'t h/3
4tLZ--2I~B +5 + ~B-- 1
2
where
311 + 3 , , / ~
bo = ~ - 3
t-
(39)
] - 0.4301 . . .
not very B
Rra
0.5 v ( 1 - IX'*)
(1 - IX'*)V
3-4~
2 V(1-IXA)
Ra
I IXA
2
I - 2/.ta + +~-+'~a
1
2
3-~/I +4/-ta
2
0.5 ^ IX'*
~ 2
,g-1
3-4g
Rc
(_~ ^IX"*)
^IX'*
2 "^IX'*
~A
R~
0.5 v ( 1 - IXA)
~ 2 - 1 v (1 - IX~")
3 24~"v (1 --"/-~-~A)
R,g
1--ix'*
1 -IX~
1 - "/-~a
0.5vixA
R,~
0.5v(1- IXA)
"J-5~-1v ( 1 - IX~)
3 ~'f5v (1 -,]~A)
0.5vixA
R,,
"/'5~-1v(1 - p.~)
3 2"/5v (1 -,f-~A)
IXA
R,~
1-ix A
1 -ix~
1 --'~A
IXA
Rb
0.5 v ( 1 - IX'*)
n/5-1v(1-ix'*)
2
---v(I-IXA)
2
1.J]--~p.~ - 1
2ix~
2+IXa - - J - ~ + 4ix'*
Rzx
RA
R,
1
1 + V-A
1 + IX'*
1
1+
IX'*
Ix'*(IXA+ 2 - ~ )
2
3-4~
see (42)
R.
0.5v(1- IXA)
(I--~A)V Ixa^
T '
IXAv(l - IXA)
R~
{; IXA<I
IXA= 1
{10 IXA<I
IXA= 1
{~ IX'*<I
Ix,, = 1
M. M i z u m o t o , H.-J. Z i m m e n n a n n
266
FB,
PB'
95
Rm,Rb,Re
.5
Re,Rn,Rg,Rsg,
, s,Rss
O
0
I P3
1 " 0
.5
~B
~/~ .5
laB,
RcBREt
REg,
Rgs
Rs,RsE,Rss
95
IFB~
P~
.5,
bo-O. 3177..of(3B)
laB,
PB'
1
Rs,Rg,Rsg,
REg,RES.Rss
.5
R/RI
.5
Put
.I5
w P3
1 "
~5
b&-0.4301..o~'(~9)
laB
I
~A'
/i
267
Ra ,Rs,Rg,Rb,RA,RA~R~
\,\
"\
... __a_~
95
/'/
"NN'I~"~rt
Rm
RctRgs,RsI
0
.5
Fig. 2(d). At
Ilal,
95
A'=notA
(= ~ A ) ; (e) at
B'=B.
Rs,Rsg,Rss
95 "
~,
eS
Rm,Rg,Rgg,
Rgs,Rb,RI
.5
RC
\
\
ills
FA~
0
.5
9 t
.; ~
not B (= ~B).
ii ~A'
Rs,Rsg,Rss
~~~. ~,~,~
.5
"9
95
. / ~ - - R;
\
i
t
//, //'/
95
Fig. 2(g). At B =
ao-0.5549..o1"(41)
not very B
( = -"IB2).
PA
268
,5
""
95
.5
a&-.6920..ot(43)
not very
x/1 +4/x~- 1
2~
1-4co
^
1-
[ ~ - !
/
21x2
I -
IXA~ao,
(40
4 cos
where
2[
/0'+4~r'~l
ao=g 1 + 4~ c o s ~ - - - ~ ) j = 0.5549...
(41
with
0' = c o s _,/v~
~-~].
2 + txa - x//x2
] + 4/~'~A [1
2 -- ?
- 2x/6(1 - 3p'0 c~
2 + CA - ~lx~ + 4/xa
P~A ~ a ; ,
(4:
p.A ~ a~,
269
where
q~= c o s _ l (
I
ao =
3x/6(1-/,a)),8
2q~-i cos(]q/) - 3
= 0.6920...
3
,
-,[
3-v/'~"~
with q~ =cos ~ - - - - ~ , ] .
(43)
Using Fig. 1 we shall show how to obtain the results of Tables 1 and 2.
However, we shall discuss only the case of Rb (15) at A ' = very A and B ' = not
very B because of limitations of space. The methods of obtaining the consequences for R,~, R . . . . . . R,g and Rgg of (6)-(12) in the case of generalized modus
ponens are found in 1-4-6]. The other consequences can be obtained in the same
way as in the case of Rb, though we must solve a cubic equation, particularly in
the case of R~, in (17).
(i) The case of Rb at A ' = very A
The consequence B;, which is inferred by taking the composition of A ' and Rb
as in (21)-(25), is given by
B b' --A '
(44)
tl
(45)
is indicated by the broken line , _ _ _ 2 in Fig. 3(a) whose figure comes from the
left figure of Fig. l(x). The value of ttn, at ttB = 0.2 becomes (3-x/5)/2 by taking
the maximum of this line (i.e. (45)) by virtue of (44). Thus, in general, we can
have
/XBb,=---~
at gB
3-2
(=0.3819...).
On the other hand, when ttn= 0.7 (I> (3-x/5)/2), (45) is shown by the dotted line
' - ' - " and then the value ttBr (the maximum value of this line) is 0.7 when
tts= 0.7. Thus, in general
P'Bb' = /'tB
at gB I > - -
270
lab='7
Fig. 3. T h e w a y of o b t a i n i n g B[, and A[,: (a) ~s~ a t / x A . = Ix2; (b) P-Af, at gs' = 1 - Ix2.
Therefore, we obtain
3-~
g"' =
3-x/5
2
'
3 -',/5
P.s ~ - ~ ,
/-ts
which leads to
3-4"5
/aBe'=
.vp.s.
B'.
(46)
(471
I 1 -/x a
--
/xA 4 .
,/g-1
2
~.LAI~ T
271
Namely,
~"~b' =
x/-5-1 v (1 _ ~A).
2
The same way is applicable to B ' = not B, not more or less B, and B.
Example. Using Tables 1 and 2 and Fig. 2, we shall present a simple example of
fuzzy reasoning in Fig. 4. Fig. 4(a) shows fuzzy sets A and B, and Fig. 4(b)
includes fuzzy sets 'not A ' , 'not very A ' , . . . . ' u n k n o w n ' in order to compare with
the inference results of Fig. 4(c)-(j).
In the forms of fuzzy conditional inferences (3) and (4), it seems according to
our intuitions that the relations between A ' in Ant 2 and B' in Cons of the
generalized modus ponens (3) ought to be satisfied as shown in Table 3. Similarly,
the relations between B' in Ant 2 and A ' in Cons of the generalized modus
tollens (4) ought to be satisfied as in Table 4.
Relation I in Table 3 corresponds to the modus ponens. Relation I1-2 has a
consequence different from that of Relation II-1, but if there is not a strong causal
relation between " x is A " and " y is B " in the proposition "If x is A then y is
B", the satisfaction of Relation II-2 will be permitted. Relation IV-1 asserts that
when x is not A , any information about y is not conveyed from Ant 1. T h e
satisfaction of Relation IV-2 is demanded when the fuzzy proposition "If x is A
then y is B " means tacitly the proposition "If x is A then y is B else y is not B " .
Although this relation may not be accepted in ordinary logic, in our daily life we
often encounter the situation in which this relation can hold. Relation V corresponds to modus tollens. Relation VIII is discussed as in the case of Relation IV.
In Table 5, the satisfaction (0) or failure (x) of each criterion in Tables 3 and 4
under each fuzzy reasoning method is indicated by using the consequence results
of Tables 1 and 2.
Under these criteria it is found that Rm and Ra are neither very suitable for the
fuzzy conditional inference in the case of generalized modus tollens nor in the
case of generalized modus ponens. Rc is not bad. R~, Rg, R~g. . . . . R~ are
satisfactory. Rb . . . . . Rrn are not very good.
PA
1 j PB
o
1
272
unknown
unknown
not very A - - ~ ~
/ F//~-~ess
/ / not more
A or
I
Fig. 4(b). not A , not very A , not more or less A , not B, very B, more or less B, and unknown.
~B'
iL~aB,
//~BS.Bw
./~B~g,B~s,B~s
.5
laB ,
.5
3~
.5
~'~'B~B~'~2,~gs
3
qaB ,
lab,
//
\.\ B~g,B~s,Bw
//o-\\
.5
273
PB'
~A'
.5
.5
',~," ~Bw
z \ B2s /
',.
,~ s , A w
0
i PA'
A~,A~.A~g,
A~s,AD,A~
.5
_~
~A'
A,Aw
Aft" ~ A / ~ , A ~
.S
.5
/
O
1
9r
~A'
A~,A~g,A~s ~ . ~ - -
AV
~A'
~-l
o
.5
/~"~
"'
9 ~----- AA
.4.fi,Aii
95
sJ
/
Afg,Afs/
A
",,Aft
st
~.
~I
U,
VA'
}~A'
.5
A~,A~g,A~s
.5
3-E
/~-~ AA
A~,Aw
1
N%A~..
s#
~J
I
2
Fig. 4(j). Inference results at B' = not more or less B.
I~L M i z u m o t o ,
274
Table
3. R e l a t i o n s
H.-J. Zinnnermann
y is B ' ( C o n s )
Relation I
(modus ponens)
R e l a t i o n II-1
R e l a t i o n 1I-2
Relation III-1
R e l a t i o n 1II-2
Relation IV-I
Relation IV-2
x is A
y is B
x
x
x
x
x
x
y
y
y
y
y
y
Table
4. R e l a t i o n s
Relation V
(modus tollens)
Relation VI
Relation VII
Relation VIII-1
Relation VIII-2
is
is
is
is
is
is
very A
very A
m o r e o r less A
m o r e o r less A
not A
not A
is
is
is
is
is
is
1 for the
Ant
very B
B
m o r e o r less B
B
unknown
not B
Ant
1 for the
y is B ' ( A n t 2)
x is A ' ( C o n s )
yisnotB
x is n o t A
y
y
y
y
not very B
x is n o t v e r y A
n o t m o r e o r less B
x is n o t m o r e o r l e s s A
B
B
x is u n k n o w n
x is A
is
is
is
is
T a b l e 5. S a t i s f a c t i o n o f e a c h R e l a t i o n in T a b l e s 3 a n d 4 u n d e r e a c h m e t h o d
Ant 2
Cons
Rm Ro l~
R,
R,
R.~ 5m
~'eryA
veryA
rery/3
B
x
x
x
x
x
0
0
x
x
0
0
x
x
0
x
0
0
x
x
x
x
x
x
x
x
x
x
x
x
x
more or
more or
less A
less B
Relation III-2
nlore or
Relation IV-1
Relation IV-2
less A
not A
not A
unknown
not B
0
x
0
x
0
x
0
x
x
0
x
0
0
x
0
x
0
x
0
x
0
x
not B
not A
Relation I
(modus ponens)
Relation It-1
Relation I1-2
Relation III-1
Relation V
(modus tollens)
Relation VI
Relation Vll
Relation VIII-I
Relation VIII-2
or l e s s A
unknown
275
If x is A then y is B,
If y is B then z is C,
If x is A then z is C,
I f x i s A then y i s B
If y is B then z is C
P3:
IfxisA
(48)
then z is C
(49)
That is to say,
PI:
P2:
If x is A then y is B ~ R ( A , B )
If y is B then z is C
--~ R(B, C)
P3:
(50)
where 'o' is the max-min composition of R(A, B) and R(B, C), and the membership function of R(A, B) o R(B, C) is given by
P-a(a.m~
(51)
Now we shall obtain R(A, B)o R(B, C) under each fuzzy reasoning method and
then show whether the syllogism holds or not.
We shall begin with the method R~. The fuzzy relations Ra(A, B) and R,(B, C)
are obtained from propositions P~ and P2 by using (7):
276
(52)
The function (i), i.e. 1 ^ ( 1 - txA(u)+ ~ ( v ) ) , can be depicted by using the parameter t-L^(U) as in Fig. 5(a) and the function (ii), 1 ^ ( 1 - ~t,(v)+ ~c(W)), is shown by
using the parameter t.tc(w) as in Fig. 5(b). These figures base on Fig. l(ii). From
these figures, the function (i)^(ii) in (52) with both parameters p.A(u)= a and
txc(w) = c will be shown by the broken line ' . . . . ' in Fig. 5(c) and its maximum
value (by virtue of (52)) is 0.5 + (1 - a + c)/2. On the other hand, if the parameter
txA(u) is taken to be a' as in Fig. 5(c), the maximum value of its line ' - . - . '
becomes 1. Therefore, in general, for any parameters a and c, the maximum value
of (i)A(ii) is shown to be 1 ^ ( 0 . 5 + ( 1 - a + c ) 1 2 ) .
Therefore, the membership
function /~R,(A,B~.Ro(~.c(U,W) of (52) becomes:
/xR,~A.B)~
~6(~)=
Vc (w)=
.i
.2
.3
.4
.5
.6
.7
,8
.9
liA(u)=a'
1~c(W)=C
~A(u)=a
, ]~B(v)
1
(c) l~
Comparisono[fuzzy reasoningmethods
277
1A(1--1"tA(It)q-I'gc(W))/(ll'
iV)).
7~ Ra(A, C ) ( = Iu
(53)
Hence, we can conclude that the fuzzy reasoning method R. does not satisfy the
syllogism.
Similarly, we can obtain R ( A , B ) o R ( B , C ) under other fuzzy reasoning
methods and we shall list them in the following.
R,,,(A, B) o R.,(B, C)
[JtJ
xW
7sRm(A, C) (= IUxw(~A(li)AI-tc(W))V(1--~A(lt))/(II,W)).
(54)
R~(A, B) o R~(B, C) = [
(55)
p,A(U)~ t-tc(W)/(U, W)
a l l x ~,V
= R, (A, C).
ltA(u) ~ tZc(W)/(u' w)
= Rg(A, C).
R,g(A, B) o R,~(B, C)
Ju
(56)
(57)
= R,g(A, C).
(58)
Rgg(A, B) o Rgg(B, C)
=[
Ju
= Rgg(A, C).
(59)
Rg,(A, B) o Rg~(B, C)
=[
Ju
[txA(u)g-->~c(W)]^[1-txA(u)~ 1 - tlc(W)]/(u, w)
(60)
hi. l~lizumoto,H.-J.Zimmennann
278
R~(A, B) oR~(B, C)
I~
(61)
B) o Rb(B, C)
0.5 v ( 1 - ~ . ( u ) ) v
t~c(w)/(u, w)
a t / x ~,v
Rb(A, C ) ( =
IUxW(1--bCA(ll))Vi.tc(W)/(II,W)).
(62)
RA(A, B) o R,,(B, C)
=[
[ ~ ( " ) 7 ~,~(w)]/(., w)
a t / ~,v
,Ra,A. C ) ( = Iu
[~A(U'a--'~"c(W)]/(U'W))'
(63)
where
~ ( . ) <- uc(w),
v.n ( . ) > t~c(W),
m~ (u) <~~c(W),
l~--~
~ (")> ~c(W).
RA(A, B) o R.(B, C)
[m~(u) ~ ~tc(w)]/(., w)
(64)
where
~A (u) ~ txc(w) =
~A(u)~(w)
..
txc(w) > I
IzA(u) = 0 or 1 - / x c ( w ) = 0
^~--~-~^l-t~(w)
m~(u)>0, 1 - t ~ c ( w ) > O ,
IxA(U) = 0 or 1 --
gc(W) = O.
279
B) o R,(B, C)
: Iu
1-.~(w)+v-..(,0~*c(W)/(.,
~ 1 - gc(W)+ p.a(u )
~ R.(A, C ) ( = Iu
w)
1--P"x(u)+P'a(U)l'tcCw)l(u' w))"
(65)
R#(A, B) o R#(B, C)
= Iux w (0"5 v 1 - bta(u) v/Xc(W)) A (P'a (U) V 1 -- P'a (U))
A
(1 - ~c(W) v I~c(w))l(',, w)
-~R#(A, C)
(=
Iux,v (1-p.A(U)VP.c(W))A(p.A(Z,)V1--p-a(U))
\
a (1 - ~c(w) v
RB(A, B) o R[a(B, C) = I
Ju
Uc(w))l(t,, w)).
(66)
= RE](A, C ) ,
(67)
where
IzA(u)~+~tc(W)=/l[0
fza(z')< or f x c ( w ) = l ,
P.A(U) = 1, g c ( V ) < 1.
Using these results the satisfaction or failure of the syllogism under each method
is listed in Table 6. The membership functions of R(A, B)o R(B, C) whose
method does not satisfy the syllogism are depicted in Fig. 6 using a parameter P.c
in order to make comparisons between R(A, B)o R(B, C) and R(A, C), where
the membership function of R(A, C) is obtained by replacing P-B with P-c in Fig. 1
(left figure).
Finally, we shall investigate the contrapositive of a fuzzy conditional proposition under each method.
For a fuzzy conditional proposition P~:
PI:
I f x i s A then y is B
If y is
Syllogism
Contraposifive
R., R~ R c R,
R,
R,
R~, R D
x
x
0
x
0
x
x
0
x
0
x
0
0
x
0
0
0
x
0
x
0
0
x
0
x
x
0
x
280
pC =
l
.9
~0 =
i
.8
.8
.7
.6
.6
.4
.2
~.5
b
I
PC=
I
.9
.8
.7
.6
- ~_. 5
c
I
> PA
PI = 9 1
.5
.9
P C 1=
.9
.5
f
40
e
!
281
~C=
pc =
o, 1
.1, .9
.2, .8
93, .7
.4, .6
.5
g
!
o
Fig. 6(g). Roe(A,B) *R~,(B,C).
R("IB,
(68)
-qA) = / ~ ( A , B)
R.('nB, hA)=
B).
1 ^ [ 1 - (1 - lis (v)) + (1 -
tta(u))]/(v, u)
= i~o(A, B).
Therefore, the contrapositive holds under the method R,.
For the method Rr,,
R,,('-aB, --aA)
=[
av x U
-~:.t~m(A, J~) (=
~xU(~I'A(12)
AI"I'B('D))V(1--~LA(L())](I)'[()~"
5. Conclusion
282
were proposed before and which are newly obtained by introducing the implica.
tion rules of m a n y valued logics. Moreover, the syllogism and contrapositive are
discussed under each method. From these results we can conclude that the
methods R,, and R a a r e not suitable for the fuzzy reasoning, since they do not
satisfy the criteria which are quite reasonable demands. Rc is not a bad method
and R~, R~ . . . . . R~, are suitable methods. The new methods Rb, R a . . . . . R E
which are based on the implication rules of m a n y valued logic systems are not
good.
We have discussed the compositional rule of inference which uses the max-min
composition. The possibility of using new compositions other than max-rain
composition is given in [10, 11].
The formalization of inference methods for the more complicated form ot
inference such as
I f x i s A then y is B else y is C
x isA'
y isD
If x is A t then y is Bt else
If x is A 2 then y is B 2 else
If x is A . then y is B .
x is A'
y isB'
would be the future subjects of investigation.
Acknowledgment
The authors acknowledge the invaluable help of Mr. U. Thole and the m e m b e r
of fuzzy research group at R W T H Aachen. T h e y are also grateful to Prof. K
T a n a k a (Osaka Univ.) and Mr. S. Fukami (N'VF) for their valuable advices durin~
the process of this study.
References
[1] L.A. Zadeh, Calculus of fuzzy restriction, in: L.A. Zadeh et al., Eds, Fuzzy Sets and Thei
Applications to Cognitive and Decision Processes (Academic Press, New York, 1975) 1-39.
[2] E.tt. Mamdani, Application of fuzzy logic to approximate reasoning using linguistic system!
IEEE Trans. Comput. 26 (1977) 1182-1191.
[3] M. Mizumoto, S. Fukami and K. Tanaka, Fuzzy conditional inference and fuzzy inference wit
fuzzy quantifiers, in: Proc. of 6th Int. Conf. on Artificial Intelligence (Tokyo, Aug. 20-23, 197c.
589-591.
[4] M. Mizumoto, S. Fukami and K. Tanaka, Some methods of fuzzy reasoning, in: M.M. Gupta c
al., Eds., Advances in Fuzzy Set Theory and Applications (North-Holland, Amsterdam, 197c.
117-136.
283
[5] M. Mizumoto, S. Fukami and K. Tanaka, Several methods for fuzzy conditional inference, in:
Proc. of IEEE Conf. on Decision & Control (Florida, Dec. 12-14, 1979) 777-782.
[6] S. Ftrkami, M. Mizumoto and K. Tanaka, Some considerations on fuzzy conditional inferences,
Fuzzy Sets and S~tems 4 (1980) 243-273.
i7] N. Rescher, Many Valued Logic (McGraw-Hill, New York, 1969).
[8] W. Bandler and L. Kohout, Fuzzy power sets and fuzzy implication operators, Fuzzy Sets and
S~tems 4 (1980) 13-30.
[9] R. Willmott, Two fuzzier implication operators in the theory of fuzzy power sets, Fuzzy Sets and
Systems 4 (1980) 31-37.
[I0] M. Mizumoto, Note on the arithmetic rule by Zadeh for fuzzy conditional inference, Cybernetics
and Systems 12 (1981) 247-306.
[11] M. Mizumoto, Fuzzy inferences using max-,x composition in the compositional rule of inference,
in: M.M. Gupta et al., Eds., Fuzzy Information and Decision Processes (North-Holland),
submitted.