You are on page 1of 48

8.

1
(fuzzy system)
(pattern recognition)(decision analysis)

8.1
(1)(2)(3)
(4)

8.1

8.1.1

(fuzzifier)

(crisp)

1. x0
A
1 x x0
A ( x)
0 x x 0

2. x= x0 1 x x0

x x0 2

A ( x ) exp
2

8.1.2 (fuzzy rule base) (1 )


Mamdani

R ( j ) : If x1 is A1j andand xn is Anj


Then y is B j

R ( j ) : If x1 is A1j andand xn is Anj


Then y j is f j ( x1 ,, xn )
(1)

R ( j ) : If x1 is A1j andand xn is Anj


Then y j is c0j c1j x1 cnj xn
(2)

R ( j ) : If x1 is A1j andand xn is Anj


Then y j is c0j

Tsukamoto Bj
(monotonical)

8.1.2 (fuzzy rule base) (2)


8.1.3

(premise) 1x is A
(premise) 2If x is A, Then y is B
-----------------------------------------------------------y is B

B A R A ( A B)

:
-:
B ( y ) maxmin A ( x), AB ( x, y )
x

B ( y ) max A ( x) AB ( x, y )
x

: B ( y ) max[ A ( x ) AB ( x, y )]
x

: B ( y ) max A ( x ) AB ( x, y )
x

R p , RM , RD , RL , RG

8.1.4

(1)




(
)
C
B A ( A B) B
1. (center of gravity defuzzifier or center of area defuzzifier)
(1)
C ( y ) ydy

*
Y
y
Y C ( y )dy
(2)

C ( yi ) yi

y * i 1L

C ( yi )

i 1

(2)

8.1.4

2. (mean of maxima defuzzifier)


1 N
*
y yj
N j 1
C ( y j ) max C ( y ) height(C ), j 1,, N
yY

3. (modified mean of maxima defuzzifier)


max{ y j } min{ y j }
j
j
y*
2

(
y
)

height
(
C
)
C j
4. (modified center average defuzzifier)
J

y
y
*

j 1
J

j (y )

C j ( y )
j 1

(3)

8.1.4

5. (modified center average defuzzifier)


J

y
y
*

j 1
J

j ( y ) /

C j ( y

) /

j 1

j j

j j
(firing strength)yj j
(weighted average method)
J

j y j
y*

j 1
J

j
j 1

8.2 (1)

x is x0 and y is y0
If x is A1 and y is B1
If x is A2 and y is B2

Then z is C1
Then z is C2

x0 y0
A B
C A B A1 B1 C1 A2 B2 C2
A B A1 B1 C1 A B A2 B2 C2
A A1 C1 B B1 C1

A A2 C2 B B2 C2

8.2 (2)
RM
-
C A B A1 B1 C1 A B A2 B2 C2

C ( z ) A ( x0 ) B ( y0 ) A1 ( x ) B1 ( y ) C1 ( z )

x, y

A ( x0 ) B ( y0 ) A2 ( x ) B2 ( y ) C2 ( z )
x, y

A ( x0 ) A1 ( x ) B ( y0 ) B1 ( y ) C1 ( z )
y

A ( x0 ) A2 ( x ) B ( y0 ) B2 ( y ) C2 ( z )
y

A1 ( x0 ) B1 ( y0 ) C1 ( z ) A2 ( x0 ) B2 ( y0 ) C2 ( z )

RM
8.2-

8.2 (3)

C A A1 C1 B B1 C1
A A2 C2 B B2 C2

C ( z ) A ( x0 ) A1 ( x ) C1 ( z ) B ( y0 ) B1 ( y ) C1 ( z )

x, y


A1 ( x0 ) B1 ( y0 )C1 ( z ) A2 ( x0 ) B2 ( y0 )C2 ( z )

A ( x0 ) A2 ( x ) C2 ( z ) B ( y0 ) B2 ( y ) C2 ( z )
x, y

A1 ( x0 ) C1 ( z ) B1 ( y0 ) C1 ( z ) A2 ( x0 ) C2 ( z ) B2 ( y0 ) C2 ( z )

8.3
R

8.1

If x is small
If x is medium
If x is large

Then y is large
Then y is medium
Then y is small

8.4xy

8.5

8.3

Sugeno TSK
cij

(1)(2)

x is x0 and y is y0
If x is A1 and y is B1
If x is A2 and y is B2

Then z is
Then z is

f1 ( x, y )
f 2 ( x, y )

i A ( x0 ) Ai ( x ) B ( y0 ) Bi ( y ) i
f i ( x0 , y0 ) i
z*

1 f1 ( x0 , y0 ) 2 f 2 ( x0 , y0 )
z
1 2

8.2 (1)
If x is small
Then y = 2x
If x is medium Then y = -x +3
If x is large
Then y = x - 1

8.6/

8.2 (2)

(1)(2)
(3) Takagi
Sugeno

y( x)

j 1

y / j
j

j 1

J j

y j c0j c1j x01 cnj x0 p


p x 0 [ x01, x02 ,, x0 p ]
j x 0 j
j

x j 2

j 2
p
1

0
i
i
1 x0i i
j j ( x0i ) exp

exp

Ai
j

2 i j
i 1
2

i 1

8.2 (3)

A j ( x0i ) x0i Ai
i
(nonlinear approximator)
(universal approximator)
c j
0

c0 j
y ( x)

j 1
J

j
j 1

cj

j 2
n

x
1
c j exp 0i j i

2 i 1 i
j 1

c0j
J

j
j 1

Radial Basis Function Network


(RBFN) RBFN

8.4 Tsukamoto (1)

TsukamotoTsukamoto
()

x is x0 and y is y0
R1If x is A1 and y is B1 Then z is C1
R2If x is A2 and y is B2 Then z is C2

i A ( x0 ) Ai ( x ) B ( y0 ) Bi ( y ) i
zi i i
z*

1 f1 ( x0 , y0 ) 2 f 2 ( x0 , y0 )
z
1 2

8.7Tsukamoto

8.5 (1)
R1If x is A1 and y is B1
Then z is C1
R2If x is A2 and y is B2
Then z is C2
x0 y0 x y A1 A2 B1 B2
C1 C2
x 2
x 3
2

5
3 y 6
3
3
A1 ( x)
, A2 ( x)
,
8 x
9 x

5 x8
6 y9
3
3
y 5
y4
5

8
4 y7
3
3
B1 ( y )
, B2 ( y )
,
11 y
10 y

8 y 11
7 y 10
3
3
z 1
z 3
1

4
3 z 6
3
3
C1 ( z )
, C2 ( z )
,
7 z
9 z

4 z7
6 z 9
3
3

8.5 (2)
y0 (t1 ) 8
x0 (t1 ) 4

x0 (t1 ) y0 (t1 )

2
A1 ( x0 4)
B1 ( y0 8) 1
3
2
1

(
y

8
)

B2 0
A2 ( x0 4)
3
3

1 min A1 ( x0 ), B1 ( y0 ) min ,1

2
2
3 3
1 2
1
2 min A2 ( x0 ), B2 ( y0 ) min ,
3 3 3

1 8.8
2
8.8
(max)

8.8

8.5 (3)
C
z 1
C (z )
1 z 3
3
2
3 z 5

3
7 z
C ( z )
5 z 6
3
1
6 z 8
3
9 z
8 z9
3
3 z 1
52
67 z
81
99 z
zdz

zdz

zdz

zdz

1 3
3 3
5 3
6 3
8 3 zdz

3 z 1
52
67 z
81
99 z
1 3 dz 3 3 dz 5 3 dz 6 3 dz 8 3 dz
28 16 49 28 25

18 3 18 6 18
2 4 1 2 1

3 3 2 3 6
4.7
(1)

8.5 (4)
(2) 1,2,...,9 9

1
2
2
2
1
1
1
10 2 3 4 5 6 7 8 90
3 3 3 3 3 3 3
z*
4.7
1 2 2 2 1 1 1

3 3 3 3 3 3 3
(3)
34 5

zj

j 1

3 4 5
4.0
3

(4)

53
4.0
2

(5)

z*

4 2 61

3
3
2 1
3 3

14
4.7
3

8.6 (1)

1.

2. (1) (2)
/

numerical data

1. 1
(2)
(correlation)

2.

1
(modeling)
2

8.10

8.6 (2)

/

( x1 is A1j ) AND ( x2 is A2j )AND AND ( x p is Apj )


If
Then y is B j

If
x is A j
Then y is B j

8.11

8.8 (1)
:
(1) (adaptive network-based fuzzy
inference system ANFIS)
(2) (fuzzy adaptive learning control
network FALCON)
(3) (backpropagation fuzzy system)

(AND)
(OR)

: (Fuzzy
HyperRectangular Composite Neural Network
FHRCNN)[17],[18]

8.8.1 (1)
x , yz
Sugeno

R1 x is A1 and y is B1, Then z f1 p1 x q1 y r1


R 2 x is A2 and y is B2 , Then z f 2 p2 x q2 y r2

8.12(a) (b) ANFISJ.-S. Roger Jang et. al. [12]

8.8.1 (2)

O1,i Ai ( x ), for i 1,2

A ( x)

O1,i Bi 2 ( y ), for i 3,4


1

x ci
1
ai

2bi

O2,i wi Ai ( x) Bi ( y ), i 1,2

O3,i wi

wi
, i 1,2
w1 w2

8.8.1 (3)

O4,i wi f i wi ( pi x qi y ri ), i 1, 2

O5,i wi f i
t

wi f i
i

wi
i

(ANFIS)

8.13ANFISJ.-S. Roger Jang et. al. [12]

8.14(a)Tsukamoto(b)TsukamotoANFISJ.-S. Roger Jang et.


al. [12]

8.8.2 (1)
FALCON

FALCON

8.15FALCONC.T. Lin and C.S. George Lee [14]

8.8.2 (2)

f ui(1)

a f

(ui( 2 ) mij ) 2

ij2

a exp( f )


(AND)

f min( u1( 3) , u2( 3) ,, u (p3) )

a f

8.8.2 (3)
1
2

(OR)

f ui( 4 )

a min(1, f )


12

a f
f yi
f wij(5) uij(5)
( mij ij )uij(5)
j

ijuij(5)
j

8.8.3 (1)
1
2Rp3


l 2
J l p
x x i
y exp i l
i
l 1 i 1

f ( x)


l 2
J p
x x i
exp i l
i
l 1 i 1

l l

p J yxi il

8.16L.-X. Wang [16]

8.8.3 (2)

1
ek2 [ f ( x k ) d k ]2
2
l
y
l

y ( k 1) y ( k ) a

e
l

y( k )

J l

a yzl
l 1

b zl
l 1

f a
l
a l
y
y
1
( f d ) zl
b
( f d)


j 2
p
x x
l
z exp ki j i
i 1
i

y (k 1) y (k )

f d l
z
b

8.8.3 (3)
x l
l

x (k 1) x (k ) a

e
l

x (k )
e

f zl
( f d ) l l
l
z
xi
xi
l

l
2 xki xi
y f l

( f d)
z
b
i2l
l

2
x

x
(
k
)

ki
l
l

l
f d

xi ( k 1) x ( k )
( y f )zl l2
b
i (k )

8.8.3 (4)
il

il ( k 1) il (k )

e
il ( k )

2 xki xi ( k )

f d

il ( k )
y f zl
b
il 3 (k )


z l (l 1,, J ) ab f
l l
y
x
l
i

8.8.4 (1)

If ( x is A j )and and( x is A j )
1
p
p
1

Then y is c0j

If x is A

Then y is c0j

FHRCNN
FHRCNN

If x is ( HR j HR j )
1
k

Then y is c0j

8.17

8.18(a)HRCNN(b)FHRCNN

8.8.4 (2)
(HyperRectangular Composite Neural
Networks HRCNNs)
J

Out ( x ) f Out j ( x )

j 1

Out j ( x) f net j ( x)
net j ( x ) f ( M ji xi )( xi m ji ) p
p

i 1

1
f ( x) {
0

M ji , m ji R

if x 0
if x 0
j

8.8.4 (3)

(supervised decision-directed learning algorithm
SDDL algorithm)HRCNN
(overlap)
100%
JJ

If x [ml1, M ll ] [mlp , M lp ] Then Out( x ) 1,

If x [mJ 1 , M Jl ] [mJp , M Jp ] Then Out( x ) 1.

8.8.4 (4)
(FHRCNN)
(HyperRectangular Composite Neural
Networks HRCNNs)
m j (x ) f(x)

FHRCNN
Out( x )

w j m j ( x)
j 1

m j ( x ) exp s 2j per j ( x ) per j 2

per j M ji m ji
p

i 1

per j ( x ) max M ji m ji , xi m ji , M ji xi
p

i 1

8.8.4 (5)

If x is HR1
Then Out( x ) is w
1

If x is HRJ

Then Out( x ) is wJ .

8.19

You might also like