Fuzzy Functional Inference Method

Hirosato Seki, Member, IEEE and Masaharu Mizumoto

Abstract—This paper proposes a fuzzy functional inference where x1 , x2 , . . . , xn are variables of the antecedent part,
method in which the consequent parts of the T–S inference A1i , A2i , . . . , Ani fuzzy sets, and fi (x01 , x02 , . . . , x0n ) functions
method is generalized to fuzzy functions. Moreover, this paper of the consequent part, i = 1, 2, . . . , M and M is the total
addresses equivalence of the proposed fuzzy inference method.
Namely, the fuzzy functional inference method is shown to number of fuzzy rules. The inference result y 0 for inputs
be equivalent to the weighted T–S inference method whose x01 , x02 , . . . , x0n is given as
consequent parts have weights, by adopting as the weight the
M

area or the base length of fuzzy set obtained from the fuzzy
function in the consequent part. hi fi (x01 , x02 , . . . , x0n ) n
i=1

I. I NTRODUCTION y0 = M
, hi = Aji (x0j ) (2)
 j=1
The fuzzy inference methods have powerful capability hi
in nonlinear systems, control, and so on since Zadeh [1] i=1

introduced fuzzy sets and Mamdani [2] applied the concept where hi is the degree of the ith rule and M is the total
of fuzzy inference to steam engine experimental device. number of the fuzzy rules.
Especially, researches on the T–S inference method [3], [4],
which is widely used as fuzzy control method and so on, are B. Min–Max–Gravity method
reported in many papers.
We secondly explain the min–max–gravity method as
Although the antecedent parts of the T–S inference method
Mamdani’s fuzzy inference method [2] for the fuzzy inference
are fuzzy sets, the consequent parts are functions. However,
form (see Fig. 1). The rules of the min–max–gravity method
it may be completely natural that the consequent parts are
are given as follows:
also fuzzy. From this reason, Type-2 fuzzy inference method

[5], whose both the antecedent parts and consequent parts are x1 = A1i , x2 = A2i , . . . , xn = Ani
fuzzy sets, is proposed. Therefore, this paper proposes a fuzzy Rule Ri = (3)
−→ y = Bi
functional inference method in which the consequent parts of
the T–S inference method is generalized to fuzzy functions, where x1 , x2 , . . . , xn are variables of the antecedent part,
and shows the equivalence [6]–[15] of the proposed method A1i , A2i , . . . , Ani fuzzy sets, Bi fuzzy sets of the consequent
and the weighted T–S inference method, whose consequent part, i = 1, 2, . . . , M and M is the total number of rules.
parts have weights, by adopting as the weight the area or Each inference result Bi which is infered from the fact
the base length of fuzzy set from the fuzzy function in “x01 , x02 , . . . , x0n ” and the fuzzy rule “A1i , A2i , . . . , Ani −→
the consequent part. Moreover, from the above results, it Bi ” is given in the following.
is shown that the functional fuzzy inference method can The degree of fitness, hi , of the fact “x01 , x02 , . . . , x0n ” to
treat complex problem as compared with conventional T–S the antecedent parts “A1i , A2i , . . . , Ani ” is given as
inference method.
hi = min{A1i (x01 ), A2i (x02 ), . . . , Ani (x0n )} (4)
II. F UZZY I NFERENCE M ETHODS
In this section we review the T–S inference method, min– Thus, the inference result Bi is given as
max–gravity method and product–sum–gravity method. Bi (y) = min{hi , Bi (y)} (5)
A. T–S Inference Method
The final consequence B  of (3) is aggregated from
At first, we shall explain the T–S inference method. B1 ,B2 , . . . , BM

by using the max. Namely,
Rules in the T–S inference method [3], [4] are constituted
as B  (y) = max{B1 (y), B2 (y), . . . , BM

(y)} (6)

x1 is A1i , x2 is A2i , . . . , xn is Ani
Rule Ri = (1) The representative point y0 for the resulting fuzzy set B 
−→ y = fi (x01 , x02 , . . . , x0n )
is obtained as the center of gravity of B  :
This work was partially supported by the NEC C&C Foundation. 
H. Seki is with the Department of Technology Management, Osaka y · B  (y)dy
Institute of Technology, 5-16-1, Omiya, Asahi-ku, Osaka, 535-8585, Japan
(email: h [email protected]). y0 =  (7)
M. Mizumoto is with the Department of Engineering Informatics, Osaka B  (y)dy
Electro-Communication University, 18-8 hatsu-cho, Neyagawa-shi, 572-
8530, Japan (email: [email protected]).

978-1-4244-8126-2/10/$26.00 ©2010 IEEE

 B'1
B1 , B2 , . . . , BM

by using the algebraic sum (+). Namely,
A11 a A 21
b B1
min{a,b} (= h 1 ) B  (y) = B1 (y) + B2 (y) + · · · + BM

(y) (10)
The representative point y0 for the resulting fuzzy set B 
x1 x2 y is obtained as the center of gravity of B  :
A2
1
A 22 B2 
B'2
y · B  (y)dy
y0 =  (11)
x10 x1 x02 x2 y B  (y)dy

III. F UZZY F UNCTIONAL I NFERENCE M ETHOD AND

W EIGHTED T–S I NFERENCE M ETHOD
B' = max{B'1 , B'2 }
A. Fuzzy Functional Inference Method
y
y0 Although the antecedent parts of the conventional T–S
inference method are fuzzy sets, the consequent parts are
Fig. 1. Min–Max–gravity method. constant. However, this structure is hard to understand from
the different of antecedent parts and consequent parts. Rather,
it may be completely natural that the consequent parts are
C. Product–Sum–Gravity method also fuzzy.
We thirdly explain a fuzzy inference method called In this subsection, we propose a fuzzy functional inference
product–sum–gravity method [16]–[19] for the fuzzy infer- method in which the consequent parts of the T–S inference
ence form (see Fig. 2). The rules of the product–sum–gravity method are generalized to fuzzy function. This method can
method are also given as (3). easily understand intuitive concept.
Each inference result Bi which is infered from the fact For example, fuzzy functions can represent the following:
“x01 , x02 , . . . , x0n ” and the fuzzy rule “A1i , A2i , . . . , Ani −→ • Gold has leaped in value

Bi ” is given in the following. • Sea surface temperatures have risen due to global warm-

The degree of fitness, hi , of the fact “x01 , x02 , . . . , x0n ” to ing

the antecedent parts “A1i , A2i , . . . , Ani ” is given as • In late 2008s, the economy worsened rapidly

Conventiona function can not represent the above expression

hi = A1i (x01 ) · A2i (x02 ) · · · Ani (x0n ) (8) like “leaped in value,” “have risen,” and “worsen rapidly.”
Therefore, fuzzy function is more suitable for the above case
where · stands for the algebraic product. Thus, the inference
than the conventional function.
result Bi is given as
The rules in the fuzzy functional inference method are
Bi (y) = A1i (x01 ) · · · Ani (x0n ) · Bi (y) constituted as

= hi · Bi (y) (9) x1 is A1i , x2 is A2i , . . . , xn is Ani
Rule Ri = (12)
−→ y = Fi (x01 , x02 , . . . , x0n )
The final consequence B  of (3) is aggregated from
where x1 , x2 , . . . , xn are variables of the antecedent part,
A1i , A2i , . . . , Ani fuzzy sets, and Fi (x01 , x02 , . . . , x0n ) are fuzzy
A11 A 21 B1
functions. The inference result y 0 for inputs x01 , x02 , . . . , x0n
a b B'1 is given as
ab (= h1 ) M

hi Fi (x01 , x02 , . . . , x0n )
x1 x2 y n

i=1
1
A2 A 22 B2 y0 = M
, hi = Aji (x0j ) (13)
B'2  j=1
hi
i=1
x 0
1 x1 x 0
2
x2 y where hi is the degree of the ith rule and M is the total
number of the fuzzy rules.
Since the consequent parts Fi (x1 , x2 , . . . , xn ) constitutes
B' = B'1 + B'2 fuzzy sets, the final output y0 of the fuzzy functional infer-
y ence method should be obtained from Mamdani type fuzzy
y0 inference method. For example, the inference result y0 is
obtained as follows when the product–sum–gravity method
Fig. 2. Product–sum–gravity method. is used (See Fig. 2).
 F1 (x0 )
Given the rules of (12), inputs x01 , x02 , . . . , x0n and each
antecedent part Ai1 , Ai2 , . . . , Ain , the degree hi is obtained A1
F1 (x0 )’
as (8), and the inference result Fi (x1 , x2 , . . . , xn ) of each h1
fuzzy rule is obtained as
x y
Fi (x1 , x2 , . . . , xn ) (y) = hi · Fi (x1 , x2 , . . . , xn ) (14) F2 (x0 )
h2
where · is the algebraic product. A2
The final output F  of (3) is obtained by the summation F2 (x 0 )’
of each inference result Fi (x1 , x2 , . . . , xn ) . Namely,
x0 x y
Fi (x) (y) = F1 (x1 , x2 , . . . , xn ) (y) + · · ·
+Fn (x1 , x2 , . . . , xn ) (y) (15)
F’ = max{F1 (x 0 )’ , F2 (x0 )’ }

The representative point y0 of conclusion F is obtained
as follows:

y · Fi (x) (y)dy
y0 =  (16)
 y
Fi (x) (y)dy
y0
Note that it is possible to apply the min–max–gravity Fig. 4. Case of min–max–gravity method.
method by replacing algebraic product with min in (14) and
algebraic sum with max in (15).
Example 1: We consider a simple example as one- case of triangular-type fuzzy sets, the fuzzy function values
dimensional case of the fuzzy functional inference method, as F1 (x) and F2 (x) are expressed as follows:
shown in Fig. 3. Then, the rules of fuzzy functional inference
method are given as F1 (x) =< 0, 1, 2 >
F2 (x) =< (x − 5)/2, (x + 1)/2, (x + 7)/2 > (18)
Rule 1 : A1 → F1 (x)
Rule 2 : A2 → F2 (x) where the symbol < a, b, c > expresses triangle, and a, b and
(17)
Fact : x0 c are left periphery, center and right periphery, respectively.
Cons. : y0 Therefore, the widths of fuzzy functions F1 and F2 are 2
and 6, respectively. From the above, we have the results as
where consequent parts F1 (x) and F2 (x) are fuzzy func- shown in Fig. 5. As can be seen from an example of Fig. 5,
tions. Therefore, given input x0 , function values F1 (x0 ) and the inference result by the product–sum–gravity method is
F2 (x0 ) are fuzzy sets. For example, the inference result y0 widely different from the inference result by the min–max–
is obtained as in Fig. 4 when the min–max–gravity method gravity method because the overlap of fuzzy functions in the
is used. min–max–gravity method strongly operate.
Example 2: We next explain the fuzzy functional infer- Therefore, it turns out that the fuzzy functional inference
ence method with concrete descriptions (See Fig. 5). In the method can obtain ingenious results by changing width and
form of fuzzy functions in the consequent parts.
F2 (x) B. Weighted T–S Inference Method
y
Next, we consider a weighted T–S inference method
as follows. This method is an extension version of fuzzy
F1 (x 0 )
singleton-type inference method [20].

x1 is A1i , x2 is A2i , . . . , xn is Ani
Rule Ri =
F1 (x) −→ y = wi (x01 , . . . , x0n )/fi (x01 , . . . , x0n )
F2 (x 0 ) (19)
where the weight wi (x01 , x02 , . . . , x0n ) is a function value
x0 x such that wi (x01 , x02 , . . . , x0n ) ≥ 0. wi (x01 , x02 , . . . , x0n ) ≥
0
1 means that the corresponding rule A1i , A2i , . . . , Ani −→
h1
A1 A2 wi (x01 , x02 , . . . , x0n )/fi (x01 , x02 , . . . , x0n ) is emphatic, and 0 ≤
h2 wi (x01 , x02 , . . . , x0n ) < 1 means that the rule is suppressive.
1
The consequence y0 by the weighted T–S inference
Fig. 3. Fuzzy functional inference method. method is obtained as follows (see Fig. 6). The degree of
 y Fi (x)
F'i
3.5 Min-max-gravity method hi
f 2 (x) = (x+1)/2 S'i
f’i
Product-sum-gravity method
1 Fig. 7. Area Si and gravity fi of Fi .
f 1 (x) = 1
1 6 x
0
We have the following theorem regarding the fuzzy func-
A1 A2 tional inference method and weighted T–S inference method.
1 Theorem 1: Let the area Si (x01 , x02 , . . . , x0n ) of the con-
clusion part Fi of the fuzzy functional inference method of
Fig. 5. Inference results in the case of fuzzy function of (18). (12) be interpretted as the weight wi (x01 , x02 , . . . , x0n ) of the
weighted T–S inference method of (19). Then, two fuzzy
1 w1 ( x10, x02) inference methods are equivalent.
A1 A 21 Proof: Let us consider again the method (16) of obtaining
a b abw1 the representative point y0 for the fuzzy functional inference
(= h1w1) method.
The inference result Fi (x01 , x02 , . . . , x0n ) in (14) is obtained
x1 x2 f1 ( x10, x02) y
as the shape of fuzzy sets, where the center of gravity
w2 ( x10, x02)
1
A2 A 22
h2w2 of Fi (x01 , x02 , . . . , x0n ) is fi (x01 , x02 , . . . , x0n ) , and the area
is Si (x01 , x02 , . . . , x0n ) (See Fig. 7). We use the following
symbol for short:
x10 x1 x02 x2 f2 ( x10, x02) y
Fi = Fi (x01 , x02 , . . . , x0n ) (21)
h2w2 fi = fi (x01 , x02 , . . . , x0n ) (22)
h1w1
Si = Si (x01 , x02 , . . . , x0n ) (23)

y We begin with the inference result Fi given in (15). Let

y0 yi be the center of gravity of Fi and Si be the area of Fi as
f1 ( x10, x02) f2 ( x10, x02) in Fig. 7, then we have
 

Fig. 6. Weighted T–S inference method. y · Fi (y)dy y · Fi (y)dy
yi =  = (24)
Si
Fi (y)dy
fitness hi is given as (8), and the product hi ·wi of the fitness
hi and the weight wi (x01 , x02 , . . . , x0n ) of fi (x01 , x02 , . . . , x0n ) and thus

is regarded as the degree to which fi (x01 , x02 , . . . , x0n ) is y · Fi (y)dy = Si · yi (25)
obtained. Therefore, the final consequence y0 of (19) is
inferred as the weigted average of fi (x01 , x02 , . . . , x0n ) by the The center of gravity y0 of F  in (15) is given as
degree hi wi (x01 , x02 , . . . , x0n ): 
y · Fi (x) (y)dy
M
 y0 = 
hi wi (x01 , x02 , . . . , x0n )fi (x01 , x02 , . . . , x0n )
i=1
Fi (x) (y)dy
y0 = (20)
M
 
hi wi (x01 , x02 , . . . , x0n ) y · {F1 (y) + · · · + FM
(y)}dy
i=1 = 
{F1 (y) + · · · + FM
(y)}dy
IV. E QUIVALENCE OF F UZZY F UNCTIONAL I NFERENCE  
M ETHOD AND W EIGHTED T–S I NFERENCE M ETHOD y · F1 (y)dy + · · · + y · FM 
(y)dy
We show the equivalence of the fuzzy functinoal inference =  
method by using the product–sum–gravity method and the F1 (y)dy + · · · + FM 
(y)dy
weighted T–S inference method.
 S1 · f1 + · · · + SM
 
· fM
=   y
S1 + · · · + S M
M
 f i (x0 ) Fi (x0 )
hi · Si (x01 , x02 , . . . , x0n ) · fM


i=1
= M
(26)
 F i (x)
hi · Si (x01 , x02 , . . . , x0n ) Area
i=1 S i (x0 )
l i (x0 )
where the area Si of Fi is given as x0 xi
0

Si = hi · Si (x01 , x02 , . . . , x0n ) (27) Ai

1
It is found that the way of obtaining the representa-
tive point y0 has the same form as (19) with replacing (a) xi = Ai → yi = Fi (x).
wi (x01 , x02 , . . . , x0n ) with Si (x01 , x02 , . . . , x0n ) (See Figs. 8 and
9).
Thus, the fuzzy rules of the fuzzy functional inference y
method in (12) is transformed to the fuzzy rules of the
weighted T–S inference method as

x1 = A1i , x2 = A2i , . . . , xn = Ani f i (x)
Rule Ri =
−→ y = Si (x01 , . . . , x0n )/fi (x01 , . . . , x0n )
(with weight S i (x) )
(28)
Therefore, it turns out that the fuzzy functional inference (with weight l i (x) )
method is a special case of the weighted T–S infeernce
method. xi
0

Ai
Area
Fi (x 1 , x 2 ) 1
1 A1 1 A2 1 Si (x 1 , x 2 )

Center of gravity
fi (x1 , x 2 ) (b) xi = Ai → yi = Si (x)/fi (x)(= li (x)/fi (x)).
0 y
x1 x2 Fig. 9. Equivalent fuzzy rules in fuzzy functinoal inference rules and
Length of base weighted rules.
li (x 1 , x 2 )

(a) Fuzzy rules in fuzzy functional inference method.

Therefore, the fuzzy functional inference method is re-
duced to the weighted T–S inference method by interpreting
1 A1 A2
Si (x 1 , x2 ) the area Si (x01 , x02 , . . . , x0n ) of the conclusion part Fi as the
weight wi (x01 , x02 , . . . , x0n ) of the weighted T–S inference
method. Thus, Theorem 1 holds.
0 Remark 1: Note that fuzzy functional inference method
x1 x2 fi (x1 , x 2 ) y by using the min–max–gravity method is not generally
equivalent to the weighted T–S inference method.
(b) Fuzzy rule equivalent to (a). Remark 2: The fuzzy functional inference method is re-
duced to the weighted T–S inference method when the
areas Si (x01 , x02 , . . . , x0n ) of Fi (x01 , x02 , . . . , x0n ) are all equal,
1 A1 A2
li (x 1 , x 2 ) i.e., S1 (x01 , x02 , . . . , x0n ) = S2 (x01 , x02 , . . . , x0n ) = · · · =
SM (x01 , x02 , . . . , x0n ).
Remark 3: Note that the above discussion holds for any
0 types of fuzzy set Fi , though Fi in Fig. 7 is of a triangular
x1 x2 fi (x1 , x2 ) y type.
Moreover, since the area Si (x01 , x02 , . . . , x0n ) of fuzzy
(c) Fuzzy rule equivalent to (b). sets Fi (x01 , x02 , . . . , x0n ) in consequent part is proportional
to the length of base (See Fig. 8), the length of base
Fig. 8. Equivalent fuzzy rules. li (x01 , x02 , . . . , x0n ), i.e., width of fuzzy function, can be
replaced to the weight for the rules as follows: R EFERENCES
[1] L. A. Zadeh, “Fuzzy sets,” Information and Control, vol. 8, pp. 338–
wi (x01 , x02 , . . . , x0n ) = li (x01 , x02 , . . . , x0n ) (29) 353, 1965.
[2] E. H. Mamdani, “Application of fuzzy algorithms for control of simple
dynamic plant,” in Proc. IEE, vol. 121, no. 12, pp. 1585–1588, 1974.
Thus, the rules of the weighted T–S inference method can [3] M. Sugeno, “An introductory survey of fuzzy control,” Inf. Sci., vol.
be rewritten as 36, pp. 59–83, 1985.
[4] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its

x1 is A1i , x2 is A2i , . . . , xn is Ani Applications to modeling and control,” IEEE Transactions on Systems,
Rule Ri = Man, and Cybernetics, vol. SMC-15, no. 1, pp. 116–132, 1985.
−→ y = li (x01 , . . . , x0n )/fi (x01 , . . . , x0n ) [5] Q. Liang and J. M. Mendel, “An introduction to type-2 TSK fuzzy
(30) logic systems,” in Proc. 1999 IEEE International Conference on Fuzzy
Systems, vol. 3, pp. 1534–1539, Seoul, Korea, 1999.
[6] J. R. Jang and C. T. Sun, “Functional equivalence between radial basis
Remark 4: The inference result Fi in (14) is given by function networks and fuzzy inference systems,” IEEE Transactions on
using algebraic product (·). More generally, Fi can be Neural Networks, vol. 4, no. 1, pp. 156–159, 1993.
[7] K. J. Hunt, R. Haas, and M. Brown, “On the functional equivalence
obtained by using two kinds of product operations ∗1 and of fuzzy inference systems and spline-based networks,” Int. J. Neural
∗2 as follows: Syst., vol. 6, no. 2, pp. 171–184, 1995.
[8] M. Mizumoto, “Product–sum–gravity method = fuzzy singleton-type
Fi (y) = [A1i (x01 ) ∗1 · · · ∗1 Ani (x0n )] ∗2 Fi (y) (31)
reasoning method = simplified fuzzy reasoning method,” in Proc. 1996
IEEE International Conference on Fuzzy Systems, vol. 3, pp. 2098–
2102, New Orleans, USA, 1996.
[9] J. M. Benitez, J. L. Castro, and I. Requena, “Are artifical neural
networks black boxes?,” IEEE Transactions on Neural Networks, vol.
More generally, n product operations ∗1 , ∗2 , . . . , ∗n would 8, no. 5, pp. 1156–1164, September 1997.
be used as follows: [10] J. L. Castro, C. J. Mantas, and J. M. Benitez, “Interpretation of artifical
neural networks by means of fuzzy rules,” IEEE Transactions on
Neural Networks, vol. 13, no. 1, pp. 101–116, January 2002.
Fi = [A1i (x01 ) ∗1 A2i (x02 ) ∗2 · · · ∗n−1 Ani (x0n )] ∗n Fi (y) (32) [11] E. Kolman and M. Margaliot, “Are artificial neural networks white
boxes?,” IEEE Transactions on Neural Networks, vol. 16, no. 4, pp.
844–852, July 2005.
Such a fuzzy inference method as in (31) is written as [12] C. J. Mantas and J. M. Puche, “Artificial neural networks are zero-
“∗1 /∗2 –sum–gravity method.” For example, product–sum– order TSK fuzzy systems,” IEEE Transactions on Fuzzy Systems, vol.
gravity method is represented as (product/product)–sum– 16, no. 3, pp. 630–643, June 2008.
[13] H. Seki, H. Ishii, and M. Mizumoto, “On the generalization of single
gravity method. The fuzzy functional inference method input rule modules connected type fuzzy reasoning method,” IEEE
as “∗1 /∗2 –sum–gravity method” can be represented by a Transactions on Fuzzy Systems, vol. 16, no. 5, pp. 1180–1187, October
weighted T–S inference method as long as ∗2 is alge- 2008.
[14] H. Seki and M. Mizumoto, “On the equivalence of single input
braic product, even if ∗1 is arbitrary product operations type fuzzy inference methods,” in Proc. 2009 IEEE International
such as min, bounded–product or, more generally, t-norms. Conference on Systems, Man, and Cybernetics, pp. 692–697, San
In other words, the fuzzy functional inference method Antonio, Texas, USA, October 2009.
[15] H. Seki, H. Ishii, and M. Mizumoto, “On the monotonicity of fuzzy in-
as (∗1 /product)–sum–gravity method is equivalent to the ference methods related to T–S inference method,” IEEE Transactions
weighted T–S inference method in which ∗1 operation is a on Fuzzy Systems, vol. 18, no. 3, 2010. (in press)
t-norm. [16] M. Mizumoto, “Fuzzy controls under various fuzzy reasoning meth-
ods,” Inf. Sci., vol. 45, pp. 129–151, 1988.
[17] M. Mizumoto, Fuzzy controls under product–sum–grvity method and
new fuzzy control methods, Fuzzy Control Systems (ed. A. Kandel and
V. C ONCLUSIONS G. Langholz), CRC Press, pp. 275–294, 1993.
[18] B. Kosco, Neural Networks and Fuzzy Systems, Prentice Hall, 1992.
Although the antecedent parts of the usual T–S inference [19] B. -G. Hu, G. K. I. Mann, and R. G. Gosine, “A systematic study
method are fuzzy sets, the consequent parts are constant. of fuzzy PID controllers—function-based evaluation approach,” IEEE
Transactions on Fuzzy Systems, vol. 9, no. 5, pp. 699–712, October
Therefore, this paper have proposed the fuzzy functional 2001.
inference method in which the consequent parts of the T– [20] M. Mizumoto, “Fuzzy controls by fuzzy singleton-type reasoning
S inference method is generalized to fuzzy functions, and method,” in Proc. 5th IFSA World Congress, Seoul, Korea, pp. 945–
948, 1993.
have shown the equivalence of the proposed method and
the weighted T–S inference method, whose consequent parts
have weighting fanctions, by adopting as the weight, say, the
area or the base length of fuzzy set from the fuzzy function
in the consequent part.
From the above results, it turns out that the functional
fuzzy inference method can treat complex problem as com-
pared with conventional T–S inference method.
Further studies are required to apply the fuzzy functional
inference method to the real systems. Moreover, investigation
on the optimization of the fuzzy function and its parameters
is needed.