You are on page 1of 9

IIE Transactions (1997) 29, 459467

A probabilistic alternative to fuzzy logic controllers


J. DOUGLAS BARRETT and WILLIAM H. WOODALL
University of Alabama, Tuscaloosa, AL 35487-0226, USA
Received April 1995 and accepted September 1996

Fuzzy logic controllers have been used increasingly in industrial applications. We introduce the probabilistic controller as an
alternative to fuzzy logic controllers. The probabilistic controller is a `universal' controller with a structure closely analogous to a
popular type of fuzzy logic controller, but it is not based on fuzzy logic.

1. Introduction
Fuzzy logic controllers are currently being used in microwave ovens, train systems, robots, appliances, and
many other applications [1, 2]. According to Kosko and
Isaka [1], the Ministry of International Trade and Industry estimated that Japan produced about two billion
dollars worth of fuzzy products in 1992.
The fuzzy logic controller was originally designed, in
part, to mimic the thought process of a human operator
[3, 4]. Ebrahim Mamdani, creator of an early fuzzy logic
controller, asserted in some applications that the fuzzy
controller outperformed human operators and slightly
improved on the conventional controller ([5], p. 117).
Cheeseman has argued, however, that ``If these (controller operators) were trained in probability theory, they
might get better performance'' ([5], p. 184). In this paper
we propose a probabilistic controller that does not use
fuzzy logic, but does use similar rules for process control.
The possibility of developing such a controller has been
mentioned by Myron Tribus ([6], p. 3) and Bart Kosko
([5], p. 207). In the probabilistic approach one can choose
any probability measure that makes sense or, if one so
chooses, construct an empirical probability measure.
Thus, characterization of uncertainty by using probability is just as exible as fuzzy characterizations, but it
oers the additional advantage that the theory of probability and statistics can be invoked if one wishes.
This paper begins in Section 2 with a discussion of the
denitions used in fuzzy logic. We present an example in
Section 3 of the most basic fuzzy logic controllers. In
Section 4 the basic probabilistic controller is introduced.
In Section 5 probabilistic controllers based on two input
variables are discussed. In Section 6 we show that the
probabilistic controller is a `universal approximator'.
Equivalent performance of the probabilistic controller and

0740-817X

1997 ``IIE''

a popular type of fuzzy logic controller is established in


Section 7. In Section 8 our conclusions are given.

2. Fuzzy logic denitions


Zadeh [7] proposed fuzzy set theory as an alternative to
traditional set theory. In traditional set theory, an object is
either in a set or it is not. In fuzzy logic, the object is given
membership in possibly several fuzzy sets. The membership value is a value between zero and one. A membership
close to zero implies `weak' membership in the given fuzzy
set and memberships close to one imply `strong' membership. For example, a temperature of 78  F might have a
membership of 0.7 in the fuzzy set `hot' and 0.1 in the
fuzzy set `mild'. In this example, the terms `hot' and `mild'
are linguistic variables. Note that the memberships of an
object in all fuzzy sets need not sum to one. However, the
membership of an object in a given fuzzy set and the
membership in the complement of the fuzzy set must sum
to one ([5], p. 37). For more information on fuzzy sets, see
Zadeh [7, 8], Yager and Zadeh [9], or Cox [10].
The membership function mA is dened as the function
that maps the set of possible values of a variable X to the
interval [0,1], i.e., mA : X ! 0; 1, where A is a given fuzzy
set. The `union' operation of two fuzzy sets is usually
dened as mA[B maxfmA ; mB g [3]. The `intersection'
operation is usually dened by the fuzzy set mA\B
minfmA ; mB g. These union and intersection operators
are referred to as maximum and minimum operators,
respectively. Other operators are described by Bonissone [11] and Cox [10]. Non-fuzzy values are said to be
`crisp'.
There has been a sometimes heated debate over the
relationship between fuzzy logic and probability theory.
Proponents of fuzzy logic argue that there are types of

460
uncertainty for which probability theory is inadequate.
Interested readers are referred to Laviolette and Seaman
[12].

3. Fuzzy logic controllers


The fuzzy logic controller (hereafter referred to as FLC)
consists of four basic operators: (1) the fuzzier, (2) the
knowledge (rule) base, (3) the decision-making logic, and
(4) the defuzzier [3]. The fuzzier converts a crisp
reading or input measurement to membership(s) in fuzzy
set(s). The rule base is a set of rules that dictate what
action should be taken given the input fuzzy set. The
decision-making logic operator is used when more than
one rule is activated to some extent in the rule base. The
output from the decision-making is a fuzzy set dened on
the control variable. Finally, the defuzzier converts the
fuzzied result in decision-making logic to a crisp action
to be taken [3]. For presentations of similar fuzzy logic
controllers, the reader is referred to Buckley [1315], Lee
[16, 17], Mamdani [18], and Sugeno [19].
3.1. Fuzzication
During fuzzication, the reading x0 of the input variable
is taken. This reading is then fuzzied, i.e., the crisp
reading is matched against the membership functions.
Using the example of Kosko ([6], pp. 162166), suppose
we wish to control an air conditioner. The reading taken
is the current temperature. The temperature X has
possible membership in ve fuzzy sets which are cold
A1 , cool A2 , just right A3 , warm A4 , and hot A5 .
In Fig. 1 the membership functions are shown for this

Barrett and Woodall


example. The action to be taken is an adjustment of the
motor speed. A motor speed z has membership in ve
sets: stop B1 , slow B2 , medium B3 , fast B4 , and
blast B5 .
Suppose, for example, that the temperature is taken to
be 68  F. The memberships of 68  F in the sets `cold',
`cool', and `hot' are zero. In the sets `just right', and
`warm', 68  F has memberships of mA3 0:4 and mA4
0:3, respectively.
3.2. The rule base
The rule base given by Kosko ([6], p. 163) for this example is given by the following ve rules:
Rule 1: if the temperature is cold, the motor speed
stops.
Rule 2: if the temperature is cool, the motor speed
slows.
Rule 3: if the temperature is just right, the motor speed
is medium.
Rule 4: if the temperature is warm, the motor speed is
fast.
Rule 5: if the temperature is hot, the motor speed
blasts.
The rules serve the purpose of mapping the input fuzzy
sets to the output fuzzy sets. As shown in Fig. 2, 68  F has
positive memberships 0.4 in `just right' and 0.3 in `warm'.
Thus, rules 3 and 4 have been activated to some extent.
Since two rules have been activated, decision-making
logic must be used.
3.3. Decision-making logic
In decision-making logic, the strength of a rule must be
computed. In the univariate case, the strength of a rule is
simply the given membership in the corresponding fuzzy
set. Here, the strengths of rules 3 and 4 are given by
a3 0:4 and a4 0:3, respectively.
The strengths ai must be used to compute the control
output of the rule. Several methods can be used to calculate the resulting output membership function. The two
most common methods are correlation minimum encoding and correlation product encoding ([6], p. 309).
For correlation minimum encoding, the control output
of rule k is given by
m0Bk z ak \ mBk z;

Fig. 1. Membership functions for (a) given temperatures and


(b) given motor speeds.

where ak is the strength of the rule, and mBk is the


membership of the action variable Z in fuzzy set Bk . The
control output of the rule is the membership function that
results when we `trim' the portion of the output membership function in which the memberships exceed the
strength of the rule. If two rules are activated, say rules 1
and 2, the conict resolution process then produces the
following membership function:

Probabilistic alternative to fuzzy logic controllers

461

Fig. 2. Strengths for 68  F.

mB z m0B1 z [ m0B2 z a1 \ mB1 z [ a2 \ mB2 z


[3]. This function represents the resulting membership
function of the control action to be taken. Thus, the
portion of the action membership functions above the
corresponding strength is removed, creating a new gure
(here it is a trapezoid). The gures are then superimposed
by using the maximum operator to create the membership
function of the control action. The resulting gures in our
example are shown in Fig. 3. The gure in Fig. 3c represents the membership function of motor speed. Other
methods are also used to nd the resulting membership
function for the conict resolution process. One common
alternative is to replace the maximum operator in the
conict resolution with a summation operator ([6], p.
385). Any of these methods can be used with correlation
minimum encoding or correlation product encoding. In
correlation product encoding, each action membership
function is multiplied by the strength of the corresponding rule, which creates a `scaled-down' membership
function ([6], p. 384). Mizumoto [20], and others, state
that correlation product encoding with the summation
operator is preferable to correlation minimum encoding
with the maximum operator.

if Z is continuous, where mB z is the


R membership function of
the
control
action.
In
Fig.
3,
z zmB zdz 942:35
R
and z mB zdz 14:95. Thus, the center of gravity is
63.033 and the motor speed would need to be set to
63.033 r.p.m.

3.4. Defuzzication
Defuzzication must be used to translate the resulting
ouput membership function to a specic crisp action to be
taken. Several methods have also been given for defuzzication (see [3, 4, 6, 16] and [21], pp. 203209). Here we
summarize the most common method ([6], p. 309), the
center of area (gravity) method. The center of area is
given by
,Z
Z
mB zdz ;
1
COA zmB zdz
z

Fig. 3. (a) Control output process for medium motor speed;


(b) control output process for fast motor speed; (c) motor speed
membership function after conict resolution.

462

Barrett and Woodall

3.5. Other methods


The air conditioner example can also be used to illustrate
an FLC that uses correlation product encoding and the
summation operator. This type of controller represents
an `additive' fuzzy system, which has been shown to
possess some attractive properties [22]. The fuzzication
stage and rule base are identical to the previous example.
The decision making logic is the only stage that is
aected.
In decision making logic, the strengths are computed as
before. For a temperature of 68  F, the strengths remain
a3 0:4 and a4 0:3. In correlation product encoding,
the control output of rule k is given by
m0Bk z ak mBk z;
which is simply the product of the strength and the activated membership function of rule k. The strength acts
as a scaling constant that rescales the membership function. By using the summation operator, the following
membership function is produced if, for example, rules 1
and 2 are activated:
mB z m0B1 z m0B2 z a1 mB1 z a2 mB2 z :
Here the activated membership functions have been
scaled and added together to produce the output membership function. The resulting gures for our example
are shown in Fig. 4.
R The center of area Ris found by using (1). Here,
z zmB zdz 620 and
z mB zdz 10, resulting in a
centroid of 62 r.p.m. Note that this diers slightly from

the centroid of 63.033 found by using correlation minimum encoding with the maximum operator.
The type of controller shown in our example is referred
to as a Type II controller. This seems to be the most
prevalent controller in the fuzzy literature. However, Jang
[23] categorizes all fuzzy controllers as Type I, Type II, or
Type III controllers. The Type II and Type III controllers
are also referred to as Mamdani and Sugeno controllers,
respectively. The outputs of Type I and Type III controllers are weighted averages of functions of the input
variables with the weights based on input memberships.
For more information on Type I and Type III controllers,
see Terano et al. ([24], pp. 160165) or Jang [23]. For a
general discussion of all aspects of fuzzy controllers see
Driankov et al. [21].
An important characteristic that should be noted regarding the basic FLC is that the fuzzy logic controller
does not allow variability in the control action for a given
value of the crisp input. A given crisp input always yields
the same crisp output value. However, Yager and Filev
([25], pp. 342343) have suggested an ad hoc method for
incorporating variability into the control action by scaling the output membership function to create a probability density function. A random control setting is then
generated from this density function.

4. A probabilistic controller
The probabilistic controller, discussed briey by Laviolette et al. [26], consists of a structure very similar to
the FLC. Specically, there are several classications in
which the input variable X can be placed. There is a
family of probabilities of classication in the classes indexed by X x0 . Note that we are imposing the condition
X
Pclass i; X x0 1 8 x0 :
i

Fig. 4. (a) Scaled activated membership functions; (b) resulting


membership function and centroid.

Note that these probabilities could represent the probability that a human operator would classify the input as
being in a given class. The use and advantages of probabilities in place of the membership functions is discussed
and justied in detail by Hisdal [2730].
Similarly to a fuzzy controller, the probabilistic controller consists of three stages: coding, the knowledge
(rule) base, and the conict resolution. The initial stage,
which we will refer to as coding, consists of measuring the
input variable. Once the measurement x0 is obtained, the
classication probabilities are calculated. For example,
suppose we have a situation similar to that of the FLC
example given earlier. The classication probabilities are
given in Fig. 5a. This particular choice of probabilities
was made to emulate somewhat the membership functions of Kosko given in Fig. 1a. Any temperature that
would be given only one non-zero membership in the
FLC is given a probability of one in the corresponding

Probabilistic alternative to fuzzy logic controllers

463
Note that for a given input value x0 , the p.d.f. of the
adjustment variable is given by
f z p1 f1 z p 2 f2 z . . . pr fr z ;
where pk P(class k; x x0 and there are r rules in the
rule base. This is true because we are making adjustment
k with probability pk . Thus, Ez and Varz are easily
obtained by using elementary rules for means and variances. For example, if two rules (rules m and n) are
activated, then the expected control action is
Ez pm lm 1 pm ln :

Fig. 5. (a) Probabilities for temperature classes; (b) motor


speed density functions.

temperature category. Note that each probability represents the probability that a controller would choose the
temperature category given the specic temperature.
These probabilities can be determined subjectively, estimated by direct observation of an operator or a group of
operators, or by other performance-based methods commonly applied to determine memberships. The interpretation of these probabilities is simple. For example, if
the temperature is 68  F, then 40% of the time the temperature is classied as `just right' and 60% of the time it
is classied as `warm'.
The second stage, the rule base, consists of setting the
specic control rules, just as in the FLC. Using the previous example, our controller would have the same ve
ifthen rules as the FLC. The logic behind the rules is that
if P(class i; X x0 > 0 then make adjustment i with
probability P(class i; X x0 ). Again, two or more rules
may be activated.
The adjustment is a random variable that is based on
the value of the input variable. In our example, there is a
probability density function (p.d.f.) associated with each
class of motor speed. For each class j of the adjustment,
the p.d.f. is given by fj z. In our example, adjustment
groups 1, 2, 3, 4, and 5 correspond to stop, slow, medium,
fast, and blast, respectively. The p.d.f.'s for this example
are given in Fig. 5b. It is required that the area under each
p.d.f. of the control action be equal to one by denition.
This is not usually the case for the membership functions
of the FLC. For our example, the family of indexed
probabilities and the p.d.f.'s of the control action have
been set to mimic somewhat the FLC of Kosko [6].
The nal stage, conict resolution, is much easier to
apply with the probabilistic controller than with the FLC.

where lm and ln are the means of the distributions fm z


and fn z, respectively. Thus, to resolve the conict, one
need only take the mean of the resulting distribution of Z
to obtain the necessary control action. We could use
other location measures such as the median or mode of
the resulting distribution if desired as alternative methods. Use of the mode corresponds somewhat to the height
method of defuzzication described by Driankov et al.
([21], pp. 203209).
In our example, again suppose a temperature of 68  F
is obtained. For 68  F we have P(cold; 68  F 0, P(cool;
68  F 0, P(just right; 68  F 0:4, P(warm; 68  F
0:6, and P(hot; 68  F 0. Thus, the p.d.f. for the
motor speed adjustment is given by
f z 0:4f3 z 0:6f4 z
as shown in Fig. 6. Here, l3 50 and l4 70. Thus, our
chosen motor speed is
0:450 0:670 62 :
We would set the motor speed to 62 r.p.m. Note that the
resulting motor speed diers slightly from the speed of
63.033 obtained by using the FLC with correlation minimum encoding and the maximum operator, but matches
exactly the output of the FLC with correlation product
encoding and the summation operator.
Note that, because the probability distribution of
the adjustment is known, we need not always make the
same adjustment for a given crisp reading x0 . We can
incorporate variability in the control process, if desired,
by randomly generating an observation from this
distribution and making the appropriate adjustment.

Fig. 6. Resulting motor speed density function and centroid.

464
Yager and Filev ([25], pp. 342343) have recently suggested a similar method for the FLC. This method makes
the probabilistic controller non-deterministic to some
extent, and also allows the probabilistic controller to
imitate a human controller more realistically, if this is
desired. For many applications it is unlikely that a person
would always make exactly the same control action for a
given value of the crisp input x0 . There appears to be,
however, no advantage in incorporating this variability
into an automated control system unless, as Yager and
Filev ([25], pp. 342343) state, the other types of
defuzzication method lead to control actions that violate
constraints on the control action.
In this paper we do not show a probabilistic analog to
the Type I or Type III fuzzy controllers. However, if we
replace the input membership functions with a family if
indexed probability functions, the extension is straightforward. The output would simply be a weighted average
of a function of the input value(s) with the weights given
by the indexed probabilities.

5. The two-input case


In Sections 3 and 4 we considered the simplest case of the
probabilistic controller and of the FLC, i.e., the one input
one output case. However, multiple input multiple
output controllers would be applicable in many settings.
In this section, we discuss the case of the two input one
output FLC and probabilistic controller.
For the FLC, the stages aected by the extra input
variable are the fuzzication, rule base, and decision
making logic. In our previous example, suppose that we
use a second input variable, humidity Y . Note that
another variable, such as change in temperature, could
also be used together with temperature. During coding,
we would take measurements on the temperature and
humidity. For example, we might have a temperature of
68  F and a humidity of 60%. Now suppose humidity
level has ve classications: low, medium-low, just right,
medium-high, and high. Each classication is represented
by a membership function. Thus, two sets of memberships for temperature and humidity must be determined.
The rule base becomes more extensive in the two-input
case. There are 25 52 combinations of temperature and
humidity classications. Thus, there are up to 25 rules in
the rule base. We do not state all of them, but for cold
temperatures, we could have:
Rule 1: if temperature is cold and humidity is low, then
motor speed stops.
Rule 2: if temperature is cold and humidity is mediumlow, then motor speed stops.
Rule 3: if temperature is cold and humidity is just
right, then motor speed slows.

Barrett and Woodall


Rule 4: if temperature is cold and humidity is mediumhigh, then motor speed is medium.
Rule 5: if temperature is cold and humidity is high,
then motor speed is medium.
Note that the rules discussed here are of the form:
If
If
If
If

x0
x0
x0
x0

is
is
is
is

xa
xa
xb
xb

and y0
and y0
and y0
and y0

is yc , then make adjustment zm .


is yd , then make adjustment zn .
is yc , then make adjustment zo .
is yd , then make adjustment zp .

Each temperature class will have ve rules. Again, multiple rules will probably be activated to some extent.
The decision making logic stage is aected by the addition of a second input. To compute the strength of a
rule, a minimum membership or product of memberships
is usually taken for the FLC. For example, suppose a
temperature has membership of 0.6 in `cold' and the
humidity has membership of 0.3 in `just right'. Thus, rule
3 has been activated to some extent. It is common to take
the minimum of 0.3 and 0.6 as the membership in `motor
speed slows'.
If another rule is also activated, we would take the
minimum input membership or take a product of memberships as above. Once the strength of each rule is determined, the process of determining the membership
function of the control action is equivalent to that for the
one-input controller. The defuzzication is completed by
using one of the methods such as the center-of-area
method discussed in Section 3.
For the probabilistic controller, there is little eect on
the approach when a second input is added. The most
profound eect is on the size of the rule base, just as with
the FLC. We now consider the previous example with the
probabilistic controller.
In coding, the temperature and humidity measurements
would be taken just as shown with the FLC. The rule base
would also be the same set of 25 ifthen rules. The difference lies in the conict resolution stage.
For each of the ve humidity classes, a family of indexed probability functions can be dened analogously to
the set shown for the temperature classes. The interpretations of these probabilities are as discussed for temperature. During coding, the categorization probabilities
are computed for temperature and humidity. Several rules
may be activated to some extent, which will require the use
of conict resolution. Independence of the classication of
the two inputs is assumed here, and for most cases this
assumption is reasonable. If the classication of temperature, for example, depended on the humidity, this could
be incorporated into the probabilistic controller.
The expression for the p.d.f. of the action function is
somewhat more complex than for the one-input case.
Note that for each rule, two categorization probabilities
temperature and humidity are computed. These probabilities are multiplied together for each action function

Probabilistic alternative to fuzzy logic controllers

465

under the assumption of independence of classication.


Thus, when four rules, m, n, o, and p, are activated for
temperatures (classes a and b) and humidity (classes c and
d), then the resulting p.d.f. of the control action is given
by
f z Pa Pc fm z Pa Pd fn z Pb Pc fo z Pb Pd fp z ;
2
where Pa P class a; X x0 , Pb P class b; X x0 ,
Pc P class c; Y y0 , and Pd P class d; Y y0 . Note
that the sum of the coecients in (2) is one.
Here, we have
Ez Pa Pc l1 Pa Pd l2 Pb Pc l3 Pb Pd l4 ;
where l1 , l2 , l3 , l4 are the means of the distributions
f1 z, f2 z, f3 z, and f4 z, which correspond to rules m,
n, o, and p, respectively.
The conict resolution is carried out easily as before.
One can simply take Ez as the adjustment to be made.
Extension of the probabilistic controller to multiple inputs is straightforward.

6. The probabilistic controller as a universal


approximator
Kosko [22], Buckley [13], Wang [31], Lee and Chae [32],
Mizumoto [20], and Ying [33] have argued that the fuzzy
logic controller can be used to approximate any continuous real-valued function to any desired degree of accuracy. Kosko ([6], p. 157) has stressed the importance of
this property as a major strength of fuzzy logic controllers. In this section we show that the probabilistic controller is also a universal approximator. We consider the
single input single output case here, although an extension to the multiple input case is possible.
The structure of the one input one output probabilistic controller is such that the value of an input variable
X is obtained, i.e., x0 . On the basis of the value of the
input, a resulting expected value of the output variable is
found, Ez; x0 . To simplify the notation we let w Ez; x0 .
Suppose we wish to approximate a continuous function
f : x ! w such that the function maps the input variable
onto the output variable. It is desirable to be able to
approximate the function f : x ! w with any chosen degree of accuracy. We need the following result:
Theorem 1. If f is a continuous function on the closed and
bounded interval a, b, then, for any e > 0, there is a
piecewise linear function F that approximates f uniformly
within e on the interval ([34], p. 70)
Theorem 2. A single input single output probabilistic
controller can uniformly approximate f: x ! z if f is continuous on a closed and bounded interval.

Proof: Consider any piecewise linear function F that is


linear between the points xi ; wi ; i 1; 2; . . . ; p. This
piecewise linear function can result from the probabilistic
controller with p classes for the input variable X such that
P(class i; X xi 1 and the mean control action for
classP
i is wi li . Note that the function F is of the form
F pi1 Pclass i; X x0 li , where li is the mean of
the distribution of the ith output. We let the indexed
probability functions for the classes be triangular or
trapezoidal such that classication in at most two adjoining classes is possible. Since any piecewise linear
function can be obtained by some probabilistic controller,
the use of Theorem 1 completes our proof.
j

7. Performance equivalence of probabilistic


controllers and additive FLCs
Theorem 2 shows that a probabilistic controller can approximate an additive fuzzy system to any desired degree
of accuracy. Theorem 3 below states that a probabilistic
controller can be constructed that can produce exactly the
same mapping function as any given additive fuzzy system based on correlation product encoding.
Theorem 3. For any additive fuzzy system with correlation
product encoding and centroid defuzzication, there exists a
probabilistic controller that can produce the same function
that relates the input variable(s) and the output variable.
Proof: For a given input value, the additive fuzzy system
produces
the output membership function mz
Pp
a
m
z,
where ak are the strengths of the p actik1 k k
vated membership functions. (In practice, p is usually 2
or 3.)
Substituting into (1), we have
Z
Z
p
X
a1 zm1 zdz
ak zmk zdz
COA

k2

Z
m1 zdz

a1

p
X

ak

k2

Z
mk zdz
z

Now, we dene c1 ; c2 ; . . . ; cp such that ci scales the rst


membership function so that its area is equal to the area
of the ith membership function. Note that the choice of
the `rst' membership function is arbitrary and that
c1 1. We also dene b1 ; b2 ; . . . ; bp such that bi ai ci .
Note that b1 a1 since c1 1. Substituting, we have
Z
Z
p
X
b1 zm1 zdz
ak zmk zdz
COA

k2

Z
m1 zdz

b1
z

p
X
k2

Z
ak

mk zdz
z

466

Barrett and Woodall


Z
b1

zm1 zdz
Z

b1

m1 zdz
z

zm1 zdz
p
X

p
X

bk

p
X

bk

zmk zdz
z

Z
bk

m1 zdz
z

Z
ak

zmk zdz
z

!Z

k1

k1

p
X

k2

ak

k2

Z
b1

k2

p
X

m1 zdz

z
p
X

bk =ck

k1

zmk zdz
z

m1 zdz
z

zm1 zdz

zm2 zdz

b1 z
b2 z
Z
Z
p
p
P
P
m1 zdz
m2 zdz
bk
bk

k1

k1

zmp zdz



bp z
Z
:
p
P
mp zdz
bk

k1

P
Note that the coecients bi = pk1 bk sum to one and
the kth ratio of integrals is the centroid of the kth output
membership function. Therefore, the output value is
simply a weighted average of the centroids of the activated output membership functions. Recall that the
output of a probabilistic controller is a weighted average
of the centroids of the activated output distributions.
Thus, if we choose
Pan equivalent rule base, categorization
probabilities bi = pk1 bk , and output density functions
such that for each output class i, the centroid of the
output density function is equal to the centroid of the
corresponding output membership function; the resulting
probabilistic controller will produce the same output
value as the additive fuzzy system for each given input
value and, therefore, the same mapping function.
j
Theorem 3 is widely applicable because most fuzzy systems
are additive ([6], p. 207). No assumptions are made concerning the shapes of the membership functions of the
FLC. See Barrett ([35], p. 44) for an example illustrating
Theorem 3.

8. Conclusions
The probabilistic controller has the same structure as the
FLC, but does not require the use of fuzzy logic. The
same type of ifthen rules are used. Thus, the probabilistic controller can solve the same type of control problem as the FLC without the mathematics of fuzzy logic,
and similar potential applications exist. The probabilistic
controller is closely analogous to the FLC with correlation product encoding, the summation operator and
centroid defuzzication and, thus, lends a theoretical
justication to this type of approach. A probabilistic
approach is not more complicated than the FLC as stated
by Kosko ([5], p. 207). Because both types of controllers
are universal approximators, it is impossible to show in a
given application that the control performance of one is,
in general, superior to that of the other.
For a discussion of fuzzy controllers from a control
theory point of view, refer to Abramovitch [36]. For other
comparisons of fuzzy and probabilistic methods, see
Cheeseman [37] or Laviolette et al. [26].

Acknowledgement
We thank the editor for a number of helpful suggestions
on an earlier version of this paper.

References
[1] Kosko, B. and Isaka, S. (1993) Fuzzy logic. Scientic American,
July, 7681.
[2] Marks, R.J., III (ed.) (1994) Fuzzy Logic Technology and Applications, IEEE Technology Update Series, New York.
[3] Berenji, H.R. (1992) Fuzzy logic controllers, in An Introduction to
Fuzzy Logic Applications in Intelligent Systems, Yager, R.R. and
Zadeh, L.A. (eds), Kluwer Academic Publishers, Boston. pp. 69
96.
[4] Pedrycz, W. (1989) Fuzzy Control and Fuzzy Systems, Research
Studies Press, Somerset.
[5] McNeill, D. and Freiberger, P. (1993) Fuzzy Logic, Simon and
Schuster, New York.
[6] Kosko, B. (1993) Fuzzy Thinking: The New Science of Fuzzy
Logic, Hyperion, New York.
[7] Zadeh, L.A. (1965) Fuzzy sets. Information and Control, no. 8,
338353.
[8] Zadeh, L.A. (1973) Outline of a new approach to the analysis of
complex systems and decision processes. IEEE Transactions on
Systems, Man, and Cybernetics, SMC-3(1), 2844.
[9] Yager, R.R. and Zadeh, L.A. (1992) An Introduction to Fuzzy
Logic Applications in Intelligent Systems, Kluwer Academic Publishers, Norwell, MA.
[10] Cox, E. (1994) The Fuzzy Systems Handbook, AP Professional,
Boston.
[11] Bonissone, P.P. (1987) Summarizing and propagating uncertain
information with triangular norms. International Journal of Approximate Reasoning, 1, 71101.
[12] Laviolette, M. and Seaman, J.W., Jr (1994) The ecacy of fuzzy
representations of uncertainty (with discussion). IEEE Transactions on Fuzzy Systems, 2(1), 442.

Probabilistic alternative to fuzzy logic controllers


[13] Buckley, J.J. (1992) Theory of the fuzzy controller: an introduction. Fuzzy Sets and Systems, 51, 249258.
[14] Buckley, J.J. (1993) Sugeno type controllers are universal controllers. Fuzzy Sets and Systems, 53, 299303.
[15] Buckley, J.J. (1993) Controllable processes and the fuzzy controller. Fuzzy Sets and Systems, 53, 2731.
[16] Lee, C.C. (1990) Fuzzy logic in control systems: fuzzy logic controller, part I. IEEE Transactions on Systems, Man, and Cybernetics, 20(2), 404418.
[17] Lee, C.C. (1990) Fuzzy logic in control systems: fuzzy logic controller, part II. IEEE Transactions on Systems, Man, and Cybernetics, 20(2), 419435.
[18] Mamdani, E.H. (1977) Application of fuzzy logic to approximate
reasoning using linguistic synthesis. IEEE Transactions on Computers, 26(12), 11821191.
[19] Sugeno, M. (1985) An introductory survey of fuzzy control. Information Sciences, 36, 5983.
[20] Mizumoto, M. (1994) Fuzzy controls under product-sum-gravity
methods and new fuzzy control methods, in Fuzzy Control Systems,
Kandel, A. and Langhilz, G. (eds), CRC Press, Boca Raton, FL. pp.
276294.
[21] Driankov, D., Hellendoorn, H. and Reinfrank, M. (1993) An
Introduction to Fuzzy Control, Springer-Verlag, Berlin.
[22] Kosko, B. (1992) Fuzzy systems as universal approximators, in
IEEE International Conference on Fuzzy Systems, IEEE, pp. 1153
1162.
[23] Jang, J.R. (1993) ANIFS: adaptive-network-based fuzzy inference
system. IEEE Transactions on Systems, Man, and Cybernetics, 23
(3), 666669.
[24] Terano, T., Asai, K. and Sugeno, M. (1987) Fuzzy Systems Theory
and Its Applications, Academic Press, San Diego.
[25] Yager, R.R. and Filev, D.P. (1994) Essentials of Fuzzy Modeling
and Control, John Wiley and Sons, New York.
[26] Laviolette, M., Seaman, J.W., Jr, Barrett, J.D. and Woodall,
W.H. (1995) A probabilistic and statistical view of fuzzy methods
(with discussion). Technometrics, 37, 249292.
[27] Hisdal, E. (1986) Innite-valued logic based on two-valued logic
and probability. Part 1.1. Diculties with present-day fuzzy set
theory and their resolution in the TEE model. International
Journal of ManMachine Studies, 25, 89111.
[28] Hisdal, E. (1986) Innite-valued logic based on two-valued logic
and probability. Part 1.2. Dierent sources of fuzziness. International Journal of ManMachine Studies, 25, 113138.

467
[29] Hisdal, E. (1988) Are grades of membership probabilities? Fuzzy
Sets and Systems, 25, 325348.
[30] Hisdal, E. (1988) The philosophical issues raised by fuzzy set
theory. Fuzzy Sets and Systems, 25, 340356.
[31] Wang, L.-X. (1992) Fuzzy systems are universal approximators, in
IEEE International Conference on Fuzzy Systems, IEEE, pp. 1163
1169.
[32] Lee, J. and Chae, S. (1993) Completeness of fuzzy controller
carrying a mapping f : R1 ! R1 , in IEEE International Conference
on Fuzzy Systems, IEEE, pp. 231235.
[33] Ying, H. (1994) Sucient conditions on general fuzzy systems as
functions approximators. Automatica, 30(3), 521525.
[34] Buck, R.C. (1965) Advanced Calculus, McGraw-Hill, New York.
[35] Barrett, J.D. (1995) A probabilistic alternative to fuzzy logic
controllers. Unpublished Ph.D. dissertation, Applied Statistics
Program, University of Alabama.
[36] Abramovitch, D. (1994) Some crisp thoughts on fuzzy logic, in
Proceedings of the American Control Conference, Baltimore, MD,
pp. 168172.
[37] Cheeseman, P. (1986) Probabilistic vs. fuzzy reasoning, in Uncertainty in Articial Intelligence, Kanal, L.N. and Lemmer, L.F. (eds),
Elsevier Science Publishers, Amsterdam. pp. 8593.

Biographies
William H. Woodall holds a Ph.D. degree (1980) in statistics from
Virginia Polytechnic Institute and State University. He is currently
Russell Professor of Statistics in the Department of Management Science and Statistics and Director of the Applied Statistics Program at
the University of Alabama. His research interests are in control
charting methods and in probabilistic alternatives to fuzzy methods.
He is an associate editor of IIE Transactions on Quality and Reliability
and a member of the editorial review board of Journal of Quality
Technology.
J. Douglas Barrett holds a Ph.D. in Applied Statistics from the University of Alabama. Currently, he is Assistant Professor in the Department of Economics, Finance and Qualitative Methods at the
University of North Alabama. His research interests include statistical
process control and probabilistic alternatives to fuzzy methods.

You might also like