FUZZY
SYSTEMS
ARE
UNIVERS.4L
APPROXIMATORS
Li-Xin
Wang
Signal
and
Image
Processing Institute
Department
of
Electrical Engineering-Systems
University of Southern California
Los
Angeles, CA
90089-2564.
Abstract
In this paper, the Stone-Weierstrass
Theorem
is
used
to
prove
that
fuzzy systems with product
inference, centroid defuzzification, and Gaussian membership function
are
capable
of
approximating
any
real
continuous function
on
a
compact
set.
to
arbitrary
accuracy.
This result
can
be viewed
as
an existence
theorem
of
an
optimal
fuzzy
system
for
a
wide
variet.y
of
problems.
1
INTRODUCTION
Fuzzy systems have been successfully applied to
a
wide variety of practical problems. Notable applications
of fuzzy systems include the control
of
warm wat>er
[SI,
robot
[5,20],
heat exchange
[15],
traffic junction
[lG],
cement kiln
[lo],
activated sludge
[GI,
automobile speed
[13],
automatic train operation systems
[29],
model-
car parking and turning
[19],
turning
[lS],
aircraft
[2],
water purification
[2S],
automatic container crane
operation systems [30], elevator
[3],
automobile t.ransmission
[TI,
and power systems and nuclear reactor
[l].
Recent advances of fuzzy memory devices and fuzzy chips
[21,27]
make fuzzy systems especially suitable for
industrial applications.
A
very fundamental theoretical question about fuzzy systems remains unanswered, namely: “Why does
a
fuzzy system have such excellent, performance for such a wide variety of applications
?”
Existing explanations
are qualitative, e.g., “fuzzy systems can utilize linguistic information from human experts,” “fuzzy systems
can simulate human thinking procedure,” “fuzzy systems capture the approximate and inexact nature of the
real world,” etc.. In this paper, we try to answer this fundamental question by proving that fuzzy systems are
universal approximators, i.e., they are capable of approximating any real continuous function on
a
compact
set to arbitrary accuracy. We use the famous Stone-Weierstrass Theorem
[lT]
to prove this fundamental
result. This result can be viewed as
an
existence theorem of an optimal fuzzy system for
a
wide variety of
problems
.
2
DESCRIPTION
OF FUZZY SYSTEMS
We first define some terminology.
A
u7izverse
of
dzscourse
U
is
a
collection of objects which can be discrete
or continuous.
A
fuzzy
set
F
[31] in
a
universe of discourse
U
is
characterized by
a
membership function
p~
:
U
-
[0,
11,
and is labelled by
a
linguistic term, where
n
lzngtizstzc term
is
a
word such
as
“small”,
“medium”, “large”, “very large”, etc..
Then, we
can define three fuzzy sets in
U,
namely,
“slo”,
“mediuni”, and “fast”, which are characterized by the
membership functions shown in Fig.
1.
In
this paper, we use
(F,pp)
to denote
a
fuzzy set, where
F
is the
linguistic term labelling the fuzzy set, and
p~
is
its membership function.
The basic configuration of
a
fuzzy system
is
shown in Fig.
2.
Note that there are four principal elements
in
a
fuzzy system: fuzzification interface, fuzzy rule base, fuzzy inference machine, and defuzzification
interface. In this paper, we consider
multi-input-single-output
(MISO) fuzzy systems
f:
U
c
R”
-
R,
because
a
multi-output system can always be separated into a collection of single-output systems.
For example, let
U
be the values of speed of a car.
0-7803-0236-2
/92
$3.00
0
1992
IEEE
1163