Fuzzy Logic Control Systems

(FLCS)

CHAPTER ELEVEN

Approximation Properties of
Fuzzy Systems II
Chapter 11. Approximation
Fall 2005 Properties of Fuzzy Systems II ١
 Approximation Properties of Fuzzy
Systems II

In Chapter 10 we saw that by using the bound in Theorem 10.1 a large number of rules
are usually required to approximate some simple fractions. We note that the bound is a
linear function of hi . Since hi are usually small, if a bound could be a linear functions
2
of hi , then this bound would be much smaller. That is, if we can
obtain tighter bound, we may use less rules to approximate the same function with the
same accuracy.
In approximation theory, if g( x ) is a given function on U and U i1i2 ( i1 = 1,2,K, N 1 , i2 = 1,2,K
is a partition of U, then f ( x ) is said to be the k’th order accurate approximator for g( x ) if
g − f ∞ ≤ M g hk , where M is a constant that depends on the function g , and h is
g
the module of the partition that in our case is max(h1 , h2 ) .

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٢
11.1 Fuzzy Systems with Second-Order Approximation Accuracy

We first design the fuzzy system in a step-by-step manner and then study its
approximation accuracy. As in Chapter 10, we consider two-input fuzzy systems for
notational simplicity; the approach and results are still valid for n-input fuzzy systems.
The design problems is the same as in Sections 10.2.

Design of Fuzzy System with Second-Order Accuracy:

1 2 Ni
• step 1. Define N i ( i = 1,2) Fuzzy sets Ai , Ai ,K, Ai in [α i , β i ] , which are normal,
consistent, and complete with the triangular membership functions
µ A1 ( xi ) = µ A1 ( xi ; ei1 , ei1 , ei2 )
i i

µ A j ( xi ) = µ A j ( xi ; eij −1 , eij , eij +1 )

i i

For j = 2,3,K, N i − 1 , and

µ A N i ( xi ) = µ A N i ( xi ; eiN i −1 , eiN i , eiN i )
i i

Where i = 1,2 , α i = ei1 < ei2 < K < eiN i = β i . Fig. 11.1 shows an example with
N 1 = 4, N 2 = 5,α 1 = α 2 = 0
and β1 = β 2 = 1 .
Chapter 11. Approximation
Fall 2005 Properties of Fuzzy Systems II ٣
 Fig. 11.1

• Step 2 and 3. The same as Steps 2 and 3 of the design procedure in section 10.2.

Since the fuzzy system designed from the above steps is a special case of the fuzzy
system designed in section 10.2, Theorem 10.1 is still valid for this fuzzy system.

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٤
Theorem 11.1. Let f ( x ) be the fuzzy system designed through the above three steps. If
the g( x ) is twice continuously differentiable on U, then
1  ∂2g 2 ∂2g 2 
g − f ∞ ≤  2 h1 + h2 
8  ∂x1 ∞ ∂x22 ∞ 
Where hi = max eij +1 − eij ( i = 1,2) .
1≤ j ≤ N i − 1

From Theorem 11.1 we see that if we choose the particular triangular membership
functions, a second-order accurate approximator can be obtained.

Example 11.1. Same as Example 10.1 except that we now use the new bound. Since
∂2g − ≤
1
< ε . Therefore, we
= 1 , if we choose h = 1 , then we have g f ∞
∂x 2 ∞
2
8
7 fuzzy sets A j with e = −3 + ( j − 1) for j = 1,2,K,7 . The designed fuzzy
j
define
7
system is
∑ sin( e j )µ A j ( x )
f ( x ) = j =1 7
∑ µA j ( x)
j =1

We see that the number of rules is reduced from 31 to 7,

but the accuracy remains the same.
The f ( x )is plotted in Fig. 11.2 against g ( x ) = sin( x ).

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٥
Example 11.2. Same as Example 10.2 except that we now the new bound. Since
∂2g
= 0 ( i = 1,2) , we know that f ( x ) = g ( x ) for all x ∈ U . In fact, choosing hi = 2, ei1 = −1
∂x i e 2 = 1
2
And i for i = 1,2 (i.e., N 1 = N 2 = 2 ), we obtain the designed fuzzy system as
2 2
∑ ∑ g(e1 , e2 )( µ A ( x1 )µ A ( x2 ))
i i 1 2
i1 i2
i1 =1i 2 =1 1 2
f ( x) = 2 2
∑ ∑ ( µ A ( x1 )µ A ( x2 ))
i1 i2
i1 =1i 2 = 1 1 2

Where the membership functions are

1
µ A1 ( xi ) = µ A1 ( xi ;−1,−1,1) = (1 − xi )
i i 2
1
µ A2 ( xi ) = µ A2 ( xi ;−1,1,1) = (1 + xi )
i i 2
For i = 1,2 and x ∈ U.
g (e11 , e21 ) = g ( −1,−1) = 0.08, g (e11 , e22 ) = g ( −1,1), g (e12 , e21 ) = g (1,−1) = 0.4 and
g (e12 , e22 ) = g (1,1) = 0.84 . We obtain

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٦
 0.08 0.76 0.4
f ( x) = [
(1 − x1 )(1 − x2 ) + (1 − x1 )(1 + x2 ) + (1 + x1 )(1 − x2 )
4 4 4
0.84 1 1
+ (1 + x1 )(1 + x2 )] /[ (1 − x1 )(1 − x2 ) + (1 − x1 )(1 + x2 )
4 4 4
1 1
+ (1 + x1 )(1 − x2 ) + (1 + x1 )(1 + x2 )]
4 4
= 0.52 + 0.1 x1 + 0.28 x2 − 0.06 x1 x2
Which is exactly the same as g( x ) . In Example 10.2 we used 121 rules to construct the
fuzzy system, whereas in this example we only use 4 rules to achieve zero
approximation error. ∂ 2 g
=0
For any function with ∂xi2 ∞ , our fuzzy system f ( x ) designed through the three
steps reproduces the g ( x ) , i.e., f ( x ) = g ( x )for all x ∈ U .

Corollary 11.1. Let f ( x ) be the fuzzy system designed through the three steps in this
section. If the function g( x ) is of the following form:
1 1
g ( x ) = ∑ ∑ ak1k2 x1k1 x2k2
k1 = 0 k 2 = 0

Where ak1k2 are constants, then f ( x ) = g ( x ) for all x ∈ U .

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٧
11.2. Approximation Accuracy of Fuzzy Systems with Maximum Difuzzifier

In Chapter 9 we learned that. fuzzy systems with maximum defuzzifier are quite different
front those with center average defuzzifier. In this section, we study the approximation
properties of fuzzy systems with maximum defuzzifier.

Design of Fuzzy System with Maximum Defuzzifier:

Step 1. Same as Step 1 of the design procedure in Section 11.1.
Step 2. Same as Step 2 of the design procedure in Section 10.2.
Step 3. Construct a fuzzy system f ( x ) from the N 1 × N 2 rules using product inference
engine, singleton fuzzier, and maximum defuzzifier. According to Lemma 9.4, this fuzzy
system is
* * * *
f ( x ) = y i1 i2 = g (e1i1 , e2i2 )
* *
Where i1 i 2 is such that
µ i1* ( x1 )µ i2* ( x2 ) ≥ µ Ai1 ( x1 )µ Ai2 ( x2 )
A1 A2 1 2

For all i1 = 1,2,K, N 1 and i = 1,2,K, N .

2 2

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٨
Theorem 11.2. Let f ( x ) be the fuzzy system designed from the three steps above. If g ( x )
Is continuously differentiable on U = [α 1 , β 1 ] × [α 2 , β 2 ] , then
∂g ∂g
g− f ∞ ≤ h1 + h2
∂x1 ∞ ∂x 2 ∞
j +1 j
h
Where i = max e i − e i ( i = 1,2).
1≤ j ≤ N −1
i
From Theorem 11.2 we immediately see that fuzzy systems with product inference
engine, singleton fuzzifier, and maximum defuzzifier are universal approximators. In fact
by choosing the h1 and h2 sufficiently small, we can make g − f ∞ ≤ ε for arbitraryε > 0.
j
Example 11.3. Same as Example ∂g 10.1 except that we use the fuzzy system with A
maximum defuzzifier. Since ∂x = 1 , we choose h = 0.2and define 31 fuzzy sets
∞
. Let e j = −3 + 0.2( j − 1) , then the fuzzy system becomes
 sin( −3), x ∈ [ −3,−2.8]

f ( x ) = sin( e j ), x ∈ ( e j − 0.1, e j + 0.1}, j = 2,3,K,30
 sin( 3), x ∈ ( 2.8,3]

Which is plotted in Fig. 11.4.

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ٩
Example 11.4. Same as Example 10.2 except that we use the fuzzy system with
maximum defuzzifier. As in Example 10.2, we choose h1 = h2 = 0.2 and define 11
fuzzy sets A j on [− −1,1] . We construct the fuzzy system using
i +1
the 121
i +1
rules. For this
example, e = −1 + 0.2( j − 1) ( j = 1,2,K,11 ) and
j U = [e , e ] × [e , e ] ( i1 , i2 = 1,2,K. ,10
ii
12 i 1 1 i 1 1

As shown in Fig. 11.3 we further decompose U i i into U i i = U1p=0 U1q=0 U pqi i , where
12 12 12

1 1 1 1
U 00i i = [e1i , (e1i + e1i +1 )] × [e2i , (e2i + e2i +1 )],U 01i i = [e1i , (e1i + e1i +1 )] × [ (e2i + e2i +1 ), e2i +1 ]
12 1 1 1 2 2 2 12 1 1 1 2 2 2

2 2 2 2
1 1 1 1
U 10i i = [ (e1i + e1i +1 ), e1i +1 ] × [e2i , (e2i + e2i +1 )], U 11i i = [ (e1i + e1i +1 ), e1i +1 ] × [ (e2i + e2i +1 ), e2i +1 ]
12 1 1 1 2 2 2 12 1 1 1 2 2 2

2 2 2 2

Then the fuzzy system becomes

f ( x ) = g (e i + p , e i + q ), x ∈ U pqi i
1 2 12

Which is computed through the following two steps: (i) for given x ∈ U , determine
i1 , i2 , p, q such that x ∈ U pqi i , and (ii) the f ( x ) equals g (e i + p , e i + q ).
12 1 2

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ١٠
Example 11.5. Let g ( x ) = x on U = [0,1] and µ A ( x ) = µ A ( x; e , e , e ) where
i −1 i i +1
i i
i −1
e 0 = 0, e N +1 = 1, e i = , i = 1,2,K, N and N can be any positive integer. So, in this
case we have N rules and h =N − 1 1
. Let U = [e , e ] ( i = 1,2,K, N − 1)
i i i +1
and f ( x )
be the fuzzy system with maximum N − defuzzifier.
1 If x ∈ U , then
i

1 1
max g ( x ) − f ( x ) = max x − g ( e i ) or g (e i +1 ) = ( e i +1 − e i ) = h
i +1
i
x∈U i
x∈[ e ,e ] 2 2
1
Since h( = ) ≥ h2 for any positive integer N, this fuzzy system can not
N −1
approximate the simple function g ( x ) = x to second-order accuracy. Because of this
counter-example, we conclude that fuzzy systems with maximum defuzzifier can not
be second-order accurate approximators. Therefore, fuzzy systems with center
average defuzzifier are better approximators than fuzzy systems with maximum
defuzzifier.

Chapter 11. Approximation

Fall 2005 Properties of Fuzzy Systems II ١١