Page 1
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 1/26
FUZZY RULE-BASED
SYSTEMS
Lecture 15
from: Fuzzy Rule-Based Modeling with
Applications to Geophysical, Biological and
Engineering Systems, by A. Bardossy and
L. Duckstein, CRC Press, 1995
A. Fuzzy Sets
• Dealing with uncertainty
–Probability theory and statistics –Bayesian statistics
–Fuzzy sets—nonfrequentistapproach
• Fuzzy set
–Set of objects without clearboundaries or not well defined—partial membership
1
• “Crisp” set
• Example:
–the set of “young persons” is afuzzy set
2
[ ]{ }( , ( ); , ( ) 0,1 A A x x x x µ µ = ∈ ∈A X
( ) 0 or 1 A x =
1 25
40 -( ) 25 40
25
0 > 40
A
x
x x x
x
µ
≤
= ≤ ≤
1
025 40
linear
x
3
Page 2
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 2/26
• defines the “truth value” of
• Learning—process of removing orreducing fuzziness
• Cardinality
4
( ) A x
x ∈ A
{ }
1
1 1
( )
of ( , ),..., ( , )
I
i
i
I I
car
a a
µ
µ µ
=
=
=
∑A
A
• h level set
“too at least a degree”
so:
• compliment:
5
( ) | ( ) A A h x x h µ = ≥
1 2 1 2( ) ( ) if A h A h h h⊂ >
(0 1) A A h≡ < <cC A=
( ) 1 ( )C A x x µ = −
• intersection
• union:
6
D A B= ∩
( )( ) min ( ), ( ) D A B x x x µ µ µ =
B
x
AD
B
x
A E E A B= ∪
( )( ) max ( ), ( ) E A B x x x µ µ µ =
: algebraic productab
• Note: not necessarily
• t norm (intersection operator)
7
c A A∩ 0/
let D A B= ∩
( ) ( ) , ( ) D A B x t x x
a b
µ µ µ
=
(1 )(1 )
ab ab
a b ab a a b+ − + − −( )max 0, 1 ... many othersa b+ −
Page 3
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 3/26
0 170 cm
185 -( ) 170 185 cm15
1 > 185
A
x
x x x
x
µ
≤
= < ≤
• t co-norm
• linguistic variables
– words, phrases, expressions
– e.g., “tall”
8
( , ) 1 (1 ,1 )
( , ) 1 (1 ,1 )
c x y t x y
t x y c x y
= − − −
= − − −
• Linguistic modifiers:
– VERY MOSTLY PRACTICALLY
– SORT OF NOT INDEED – SOMEWHAT ROUGHLY … etc.
• e.g.
– called “concentration”
– since actually has effectof decreasing membership
9
2VERY ( ) ( ) x x µ µ =
[ ]0,1 µ ∈
• “dilation” modifier
• “contrast” modifier
– to translate back to natural language,find closest Euclidean distance to
membership function of statement 10
MORE or LESS( ) ( ) x x µ µ =
( )
2
INDEED 2
2 ( ) 0 ( ) 0.5( )
1 2 1 ( ) 0.5 ( ) 1.0
x x x
x x
µ µ µ
µ µ
≤ <=
− − ≤ <
B. Fuzzy Numbers
• Special case of fuzzy sets
• Involves arithmetic operations
• A fuzzy subset A of real nos. is a
fuzzy number if:
– at least 1 z such that:[normality]
–for all
–[convexity]
∃ ( ) 1 A z =
( )
( ) min ( ), ( ) A A A
a c b
c a b µ µ µ
< <
≥
11
Page 4
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 4/26
• Convexity insures h levels sets and
the support of the fuzzy no. are
intervals
• Membership value of real no. is“likeliness” of occurrence of that no.
– not “likelihood” 12
{ }supp( ) | ( ) 0 A A x x µ = >
1 1
convex nonconvex
• Level sets: different nos. with given
minimum “likeliness”
• Zadeh: is “possibility” of x .impossible totally possible
• A “crisp” number is a fuzzy no. with a
single point
13
( ) A x
• Triangular fuzzy nos. [symmetric or
asymmetric]
14
1
1
1 22 1
32 3
3 2
3
0
( )
0
A
x a
x a
a x aa a x
a xa x a
a a
a x
µ
≤
− < ≤−
= − < ≤
−
<
1
a1
xa2 a3
1 2 3( , , )T a a a
• Trapezoidal
fuzzy no.
15
1
11 2
2 1
2 3
43 4
4 3
4
0
1( )
0
A
x a
x aa x a
a a
a x a xa x
a x aa a
a x
µ
≤
− < ≤−
< ≤= −
< ≤−
<
1
a1 xa2 a3 a4
Page 5
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 5/26
• Fuzzy mean
–we often need to “defuzzify” a fuzzy
set—i.e., replace with a crisp value –could take element with highest
membership value—but may not be
unique or representative of fuzzy set
16
1
3 x
6 9
e.g.
(3,3,9)T
• Fuzzy mean:
for
17
( )
( )
( )
A
A
t t dt
M A
t dt
µ
µ
+∞
−∞
+∞
−∞
=
∫
∫1 2 3( , , )T A a a a=
1 2 3( )
3
a a a M A
+ +=
Note: not all fuzzy sets have fuzzy
means: e.g., 1 2( , , )T a a ∞
• Fuzzy mean for piecewise linear
fuzzy membership function
18
( )
( )
1
1 11 1
0
11
0
( )
2 2( ) ( )6 6
( ) ( )
2
L
L
M A
x x x x x x x x
x x x x
µ µ
µ µ
−
+ ++ +
=
++
=
=
+ + − +
+−
∑
∑
0 1for L x x x< < ⋅⋅ ⋅ < breakpoints
• Advantage of fuzzy mean:
–continuous—i.e., differences in
membership functions produce
smooth changes in fuzzy mean• Fuzzy median:
–disadvantage: not continuous—also,not necessarily UNIQUE
19
( )
( )
( ) ( )
m A
A A
m A
t dt t dt µ µ +∞
−∞
=∫ ∫
Page 6
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 6/26
–advantage: good for fuzzy sets
defined on DISCRETE ordered
sets X • Expressing similarity between fuzzy
sets:
–similar to Euclidean distance—butonly applicable to TRIANGULAR
membership functions20
( )
2
2 2 21 1 2 2 3 3
( , )
( ) ( ) ( )
d A B
a b a b a b
=
− + − + −
C. Fuzzy Rules
• Problem of imprecise information and
measurements• Crisp systems tolerate no exceptions
and no errors
• Wide differences in opinion
–a statement AND its opposite may
both be true to a certain degree
21
• Structure of fuzzy rule:
or simplified:
[different rules with different
consequences can be applied to samepremises]
22
1 1 2 2IF: is is is
THEN:
i i k ik
i
a A a A a A
B
⋅ ⋅ ⋅
{ AND OR XOR
1 2 THEN:i i ik i A A A B⋅⋅⋅
consequencepremises
• Degree of fulfillment [DOF]
–Product inference
23
1
1 2
1 2
1 2
1 2
1 2
1 11 2 1 2
1 2 1 2
1 2
1 2 1 2
1 2
(not ) 1 ( )
( AND ) ( ) ( )
( OR ) ( ) ( )
( ) ( )
( XOR ) ( ) ( )
2 ( ) ( )
A A A
A A
A A
A A
A A
A a
A A a a
A A a a
a a
A A a a
a a
µ µ
µ µ
µ µ
µ µ
µ µ
= −
= ⋅
= +
− ⋅
= +
− ⋅
Page 7
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 7/26
–min-max inference
[both reproduce Boolean truth table
for crisp or 0-1 case]24
( )( )( )
( )
1 2
1 2
1 2
1 2
1 2 1 2
1 2 1 2
1 2 1 2
1 2
( AND ) min ( ), ( )( OR ) max ( ), ( )
( XOR ) max min 1 ( ), ( ) ,
min ( ),1 ( ) )
(
A A
A A
A A
A A
A A a a A A a a
A A a a
a a
µ µ µ µ
µ µ
µ µ
=
=
= −
−
• Example:
IF: (1,2,3)T AND
((1,2,6)T
OR (4,5,7)T)
AND (2.5,4,4.5)T
THEN: (0,1,2)T
If: a1=1.5 0.5
a2=4 0.5
a3=4.8 0.8
a4=3 0.333product gives DOF=0.150
min-max gives DOF=0.33325
0.8
–min-max: accounts for only limiting
or extreme arguments
–product: accounts for fulfillment of
ALL arguments
26
• Degree of fulfillment for “AND” coupling
• Degree of fulfillment for “OR” coupling
2 clauses
multiple clauses—recursive “OR”
27
,
1 1
( )i k
K K
i A k k
k k
D a ν = =
= =∏ ∏
,1 ,2 ,1 ,21 2 1 2( ) ( ) ( ) ( )
i i i ii A A A A D a a a a µ µ µ = + − ⋅
1 1 1( ,..., ) ( ( ,..., ), )i K i i K K D D Dν ν ν ν ν −=
Page 8
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 8/26
• Degree of fulfillment for “MOST OF”
• Degree of fulfillment for “AT LEAST AFEW”
p=1: perfect compensation
p=∞: no compensation
p=2: good compromise 28
( ),
1
1
1
1 1 ( )i k
K p p
i A k k
D a K µ =
= − − ∑
( ),
1
1
1( )
i k
K p p
i A k
k
D a K
µ =
=
∑
p-norm
• Example: suppose we have the
following 5 rule arguments for given
set of facts :
29
1
2
3
4
5
1 1
2 2
3 3
4 4
5 5
( ) 0.9
( ) 0.9
( ) 0.5
( ) 0.1
( ) 0
A
A
A
A
A
a
a
a
a
a
ν µ
ν µ
ν µ
ν µ
ν µ
= =
= =
= =
= =
= =
1 2 3 4 5, , , ,a a a a a
For “AND” coupling
For “OR” coupling
30
0.9 0.9 0.5 0.1 0.0 0i D = ⋅ ⋅ ⋅ ⋅ =
( )( )( )
( )
0.9 0.9 0.9 0.9
0.5 0.9 0.9 0.9 0.9 0.5
0.1 0.9 0.9 0.9 0.9
0.5 0.9 0.9 0.9 0.9 0.5 0.1
0.9955
(
)
i D = + − ⋅
+ − + − ⋅ ⋅
+ − + − ⋅
+ − + − ⋅ ⋅ ⋅
=
For “MOST OF” coupling ( p=2)
For “AT LEAST A FEW” coupling
31
( ) ( ) ( ) ( ) ( )1
2 2 2 2 2 2
1
1 0.9 1 0.9 1 0.5 1 0.1 1 0
5
i D
0.355
= −
− + − − − −
=
( ) ( ) ( ) ( ) ( )1
2 2 2 2 2 20.9 0.9 0.5 0.1 0
5
i D
0.613
+ + + + =
=
Page 9
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 9/26
D. Combination of Fuzzy Rule
Responses
• With fuzzy rules (unlike “crisp” rules),several rules can be applied from the
same premise vector
• Need to be able to specify an “overall
response”
• DOF of rule i:
1( ,..., ) K a a
1( ,..., ) (response)i i K i D a a Bν = →
32
• Problem: define fuzzy set B based on
the individual
i.e.,[combination operator]
• Assume all rules have consequences
which are fuzzy subsets of same set
• 3 methods
– minimum
– maximum
– additive 33
( , )i i B ν
( ) ( )( )1 1, ,..., , I I B C B Bν ν =
• Minimum combination
• Cresting minimum combination
disadvantage: requires much care—can
easily get
• Maximum combination:
34
0( ) min ( )
i
i
B i B x xν
ν µ >
=
( ){ }0
( ) min min , ( )i
i
B i B x xν
µ ν µ >
=
( ) 0 B x =
1,...,( ) max ( )
i B i Bi I
x xν µ =
= ⋅
• Cresting Maximum combination
tolerates disagreements—but does notemphasize agreement--vague
35
( ){ }1,...,
( ) max min , ( )i
B i Bi I
x x µ ν µ =
=
Page 10
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 10/26
• Additive combination:
1) Weighted sum:
A rule is better if its consequence ismore specific—i.e., not vague
36
1I
i=1
( )
( )
max ( )
i
i
I
i B
i B
i Bu
x
x
u
ν µ
µ
ν µ
=
⋅
=
⋅
∑
∑
2) Normed weighted sum:
37
1( ) [continuous]
1car( ) ( ) [discrete]
i B
i
i ii j
x dx
B j
µ β
µ β
∞
−∞
=
= =
∫
∑
1
=1
( )
( )
max ( )
i
i
I
i i B
i B I
i i Bu
i
x
x
u
ν β µ
µ
ν β µ
=
⋅ ⋅
=
⋅ ⋅
∑
∑
3) Cresting weighted sum:
38
( )
( )1
=1
min , ( )
( )
max min , ( )
i
i
I
i B
i B I
i Bu
i
x
x
u
ν µ
µ
ν µ
==
∑
∑
4) Cresting normed weighted sum:
rules with “crisper” answers carry
more weight
39
( )
( )1
=1
min , ( )
( )
max min , ( )
i
i
I
i i B
i B I
i i Bu
i
x
x
u
β ν µ
µ
β ν µ
=
⋅
=
⋅
∑
∑
Page 11
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 11/26
• Example:
minimum combination:
40
1 1
2 2
Rule 1: 0.4 (0,2,4)
Rule 2: 0.5 (3,4,5)
T
T
B
B
ν
ν
= =
= =
4min 0.4 ,0.5 ( 3)
2
4 23 0.4 0.5 ( 3) @
2 7
x x
x x x
− ⋅ ⋅ −
− ⋅ = ⋅ − =
41
( )
230.5( 3) 3
74 23
0.4 42 7
B
x x
x x
x µ
− < ≤
= − < <
cresting minimum
42
( )
4 4min 0.4,
2 2
min 0.5, ( 3) 34 10
3 @2 3
x x
x x x
x x
− − =
− = −−
= − =
( )
10( 3) 3
3
4 10 42 3
B
x x
x x x
µ
− < ≤
= − < <
1
1 x2 3 4 5 6
0.5 crestmin
mincomb
1
1 x2 3 4 5 6
0.5crest
max
maxcomb
43
Page 12
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 12/26
weighted sum combination
44
0.4 0< 224
0.4 2 3( ) 2
4 0.4 +0.5( -3) 3 4
20.5(5 ) 4 5
x x
x x
f x
x x x
x x
≤
− < ≤
= −
< ≤
− < ≤Sum of membership functions · ν i
imax ( ) 0.5
so... ( ) 2 ( )
i
i
B
B
u
x f x
ν µ
µ
⋅ =
=
∑
normed weighted sum:
11
22
1area under : 2
1area under : 1
B
B
β
β
=
=
45
0.2 0< 22
40.2 2 3
( ) 24
0.2 +0.5( -3) 3 42
0.5(5 ) 4 5
x x
x x
f x x
x x
x x
≤
− < ≤
= −< ≤
− < ≤
( ) 2 ( ) B x f x=
46
1
1 x2 3 4 5 6
0.5norm. wtd.
sum
wtd.sum
1
1 x2 3 4 5 6
0.5
crest.wtd.sum
crest.norm.
wtd. sum
47
Page 13
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 13/26
• Advantage of additive methods for
combination
– same support
• Disadvantage of additive methods
– for repeating rules, gives different
µ
48
E. Defuzzification
• Often necessary to replace fuzzy
consequence with single crispconsequence
• Example: prediction for forecasting
or control decision:
• Methods
1) maximum2) mean
3) median
Bb
( ) f b D B=
49
• Defuzzification by maximum
– not necessarily unique
• Defuzzification by mean
50
( ) max ( ) B B x
b x µ =
( ) ( ) f b D B M B= =
1
1
1( )
( )1
I
i iii
I
iii
B
M B
ν β
ν β
=
=
=
∑
∑
– for weighted
sum comb.
easy!
• Defuzzification by normed weightedsum:
• For weighted sum combination andmean defuzzification, rules can be:
51
1
1
( )
( )
I
i i
i
I i
i
B
M B
ν
ν
=
=
⋅
=
∑
∑
,1 ,IF AND ANDTHEN: ( ( ), )
i i K
i i
A M B β
⋅⋅ ⋅
Page 14
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 14/26
• That is, we only have to represent
fuzzy consequence by its mean
and area under• For normed weighted sum
combination and mean
defuzzification, only need
• Defuzzification by median:
52
i B
( )i B : i µ β
( )i B
( ) ( ) f b D B m B= =
• Best methods are:
– product inference
– additive combinations – fuzzy mean defuzzification
53
F. Rule Systems
• Rule system
–this does not imply that allarguments used for every rule:
–e.g., assign for all
arguments k not used in rule i
ℜ
,1 ,2 ,IF THEN
(for 1,..., )
i i i K i A A A B
i I
⋅ ⋅ ⋅
=
,( ) 1
i k A x =
, fuzzy subsets offuzzy subsets of
i k k
i
A X B Y
54
• Numerical Rule System :
• To calculate response of rule system:
–inference method
–combination method
–defuzzification method
• complete if for every premise vector
, responseis nonempty fuzzy set
55
,IF: and fuzzy numbersi k i A B
ℜ
ℜ
( )1,..., K a a ∈ A ( )1,..., K a a
Page 15
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 15/26
• Rule system with maximum or
additive combination methods is
complete on Q iff:
[minimum does not work]
56
1
rule such that
( ,..., ) 0i k
i
D a a
∀ ∃
>
a
ℜ • Example:
not complete under minimum
combination
for support [2,3]
• It is complete for maximum or additive
57
1,1 1
2,1 1
(1, 2,3) (1,2,3)
(2,3,4) (3,4,5)
T T
T T
A B
A B
= =
= =
( )1 2
1 2min ( ), ( ) 0 B B x xν µ ν µ ⋅ ⋅ =
• For maximum and additive
combinations, all we need for
completeness is:
58
1
supp( )
e.g., above 1 3; 2 4
1 4
I
i
A i
x x
x
=
⊂
≤ ≤ ≤ ≤
⇒ ≤ ≤∪
∪
• Rule system is nondegenerate if:
• If nondegenerate and complete—then mean defuzzification response is
continuous function of A.
59
ℜ
i,k , Aand and ,
are continuous
i k ii A B∀ ∀
ℜ
Page 16
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 16/26
• Example:
complete and nondegenerate on
interval [1, 5]
Use normed weighted sum
combination with mean defuzzification
60
if (1, 2,3) then (0,1,2)
if (3, 4,5) then (2,3,4)if (0,3,6) then (0,2, 4)
T T
T T
T T
61
for 1 2
values are ( 1),0, resp.3
for 2 3 values are (3 ), 0, resp.
3for 3 4
6 values are 0, ( 3), resp.
3for 4 5
6 values are 0, (4 ), resp.
3
i
i
i
i
x x
D x
x x D x
x x
D x
x x
D x
≤ ≤
−
≤ ≤−
≤ ≤−
−
≤ ≤−
−
1( 1) 3(0) 25 33
for 1 2 :4 3
( 1) 03
1(3 ) 3(0) 293
for 2 3 :9 2
(3 ) 03
x x
x x
x x x
x x
x x
x x
− + +
− ≤ ≤ =
− − + +
− + + −
≤ ≤ =−
− + +
62
61(0) 3( 3) 2
7 153for 3 4 :
6 2 30 3
36
1(0) 3(5 ) 257 113
for 4 5 :6 21 4
0 53
x x
x x
x x x
x x
x x
x x x
− + − +
− ≤ ≤ =
− − + − +
−
+ − + −
≤ ≤ =− −
+ − +
63
Page 17
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 17/26
0
0.5
1
1.5
2
2.5
3
0 1 2 3 4 5
x
y
“defuzzified”response
• 3 crisp rules would give 3 different values
as consequences
• Objects in discrete categories (e.g., HIGH,
MEDIUM, LOW) can be given continuous
representation64
( )
1 1
1
1
If: = , ,
and ( , ..., ) is a continuousfunction on , then for any 0,
any inference combinations and any
defuzzification method, a rule system
such that: ,...,
K K
K
K
a a a a
f a a
f a a
A
A ε
− + − + ×⋅⋅⋅×
>
∃
ℜ − ℜ( )
( )1
1
,...,
,...,
K
K
a a
a a
ε <
∀
• Proposition:
65
• Corollary:
Every nondegenerate rule system,
inference and combination method
with mean defuzzification can bereplaced by a rule system using:
– AND operator
– product inference
– weighted combination
– fuzzy mean defuzzification
66
• Use of closed form functions:
– have specific shape
• e.g., linear
– parameters assessed with specifictechnique
• e.g., least-squares
– try to approximate observed data
– parameters may not have physical
meaning
67
Page 18
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 18/26
• Advantage of rule systems:
– rules can define function in
specific neighborhoods – errors in coefficients of polynomial
can be disastrous
– errors in a single rule only
influence function on the support
of that rule
68
G. Membership Functions in
Rule Systems
• What is a good rule?
• How crisp should arguments and
responses be?
• What shape for membership
functions?
• If we are using normed weighted sumcombination and mean
defuzzification—DOESN’T MATTER!69
• Arguments with narrow (i.e., crisp)
supports
– need many rules
• Very wide supports – nonspecific responses
• Triangular or trapezoidal
membership function are the most
popular
70
H. Fuzzy Rule Construction
• Assessment of rules: knowledge
and/or available data encoded into
rules• Ways:
1) rules known by experts*
2) rules from experts—but updated from
data
3) not known—but variables specified4) only observations available*
71
Page 19
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 19/26
• Example: algae growth model
x : available nutrients
y : algae biomass[use scale between 0 and 1]
States described by
high (H )
low (L)
Experts: HH LH LL HL HH
[ represents time transition]72
nutrientsalgae biomass
Expert Rules: Starting @ HH – if
algae high, nutrients reduced to L @
t +1; then since nutrients insufficient,
algae becomes L in next period; as
algae breaks down, nutrients
replenished and become H again;
then algae becomes H again in thenext time period.
73
74
ˆ ˆ
ˆ
ˆ
T
T
Let and be triangular
fuzzy numbers:
(0.4,1.0,1.0)
(0.0,0.0,0.7)
H L
H =
L =
1
0.4 0.5 0.7 1.0
H L
ˆ
ˆ
Calculate means:
0.4+1.0+1.0 2.4= = 0.8
3 3
0.0+0.0+0.7 0.7= = 0.233
3 3
Construct state vector trajectory :
Let initial state be :
( (0), (0)) = (0.5,0.6)
M(H)=
M(L)=
x y
75
Page 20
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 20/26
{ }2
ν
ε ν µ µ
⋅
L
H
L
H
2(0.5) =
for 7
1 nutrients x(0.5) =
61
(0.6) =for 7
2 algae y(0.6) =
6
Use for DOF: AND rule fulfillment grade
for is: i j
µ
µ
µ
µ
(i, j) :
(i, j) L,H (i, j)= (0.5) (0.6)
product infere nce
76
2/492/21
1/421/18(0.5)H
µ
(0.5)L µ
(0.6)H µ (0.6)L µ
DOF:
1/42HL HH
2/49LL HL
2/21LH LL
1/18HH LH
DOFRule
Transition table:
77
there are
really 8 rules:
HHL AND H[x] [y]
⋅ ⋅ + ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅
⋅ ⋅ + ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅
1 0.7 2 0.7 2 2.4 1 2.40.35 0.35+ 0.3 + 0.318 3 21 3 49 3 42 3(1)=
1 2 2 10.35 + 0.35+ 0.3+ 0.3
18 21 49 42
=
1 2.4 2 0.7 2 0.7 1 2.40.3 0.35+ 0.35+ 0.3
18 3 21 3 49 3 42 3(1) =
1 2 2 10.3 + 0.35+ 0.35+ 0.318 21 49 42
x
y
0.4319
= 0.4222
Fuzzy mean combination of rules:
replaced with mean value, and:( )i B µ x
= =1 1
0.3 0.35H L β β
78
1
0.60.5
nutrients x (t )algae y (t )
Same procedure used to obtain x (t +1),
y (t +1) as function of x (t ), y (t ), etc.
oscillatory behavior
79
t
same as solution ofcoupled diff. eqs.
Page 21
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 21/26
I. Deriving Rule Systems from
Data Sets
• Training set:
• Fuzzy rules can be used to:
– simplify complex models
– can generate synthetic training
sets
80
( ){ }1( ),..., ( ), ( ) ; 1,..., K a s a s b s s S T = =
( ), , , ,
,
, , for each
1where mean ( )
i
i k i k i k i k T
i k k i s R
A
a s
N
α α α
α
− +
∈
= ∑
• Counting algorithm:
1. Define support:
2. Assume
81
( ), , ,, for eachi k i k i k Aα α − +
{
( )1
, ,
th
where ( ),..., ( ), ( )
such that ( ) , ,
1,...,
denotes the set of all those premisevectors that fulfill at least the
rule; forms a subset of the trainingse
i K
k i k i k
i
R a s a s b s
s
k K
R
i
in part
T
α α α − +
= ∈
∈
=
t ; is the number of elementsin .
i
i
N R
T
82
3. Define support:
83
( ), ,i i iT
β β β − +
min ( )
1
( )
max ( )
i
i
i
i s R
ii s R
i s R
b s
b s N
b s
β
β
β
−
∈
∈+
∈
=
=
=∑
( ) ( )( )1 1 1
Rule System is:
IF , , AND , ,
THEN: , ,
i i i iK iK iK T T
i i iT
α α α α α α
β β β
− + − +
− +
⋅ ⋅ ⋅ ⋅ ⋅ ⋅
Page 22
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 22/26
• Questions:
– How to define supports?
– Use split sampling methods?• Weighted counting algorithm
– considers minimal DOF
– calculate DOF’s
– For each rule i :
84
( )i sν
( )min ( )i
i s
b sν ε
β −>
=
( )
( )
( )
( ) ( )
( )
max ( )
i
i
i
i s
i
i s
i s
s b s
s
b s
ν ε
ν ε
ν ε
υ
β
υ
β
>
>
+
>
⋅
=
=
∑
∑
85
The higher the ε value, the fewer
elements used to define the responses,and the crisper are the assessed rules
• Least Squares Method:
86
( )
( )
2
1
11
1
min ( ),..., ( ) ( )
( ) ( )
( ),..., ( )
( )
K s
I
i i
i K I
ii
R a s a s b s
s M B
R a s a s
s
ν
ν
=
=
−
⋅
=
∑
∑
∑
Rule response fornormed weighted
sum combination:
If we are using normed weighted sum
combination and mean
defuzzification, then the unknowns
are:Shapes of premise membership
functions are assumed in order to get
the
1( ),..., ( ) I B M B
iν
87
Note: the other methods can be used
for any combination and defuzzification
Page 23
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 23/26
Example:
• For comparison of the algorithms
• Training set with 25 sets ofobservation data (a1(s),b(s)),
s=1,…,25
• Develop rule system for the interval
[0,8]
• Highly variable, nonlinear behavior
88
TrainingSet
89
• Rule system from counting algorithm
• 7 rules; supports of equal length
• 2 or 3 rules applicable to each a(s)
90
• Rule system from weighted counting
algorithm
• For ε = 0.5; smaller supports; crisper
91
Page 24
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 24/26
• Least-squares algorithm
• For ε = 0.5; smaller supports; crisper
( )( )1( ) ( ) cos 12
L x R x xπ = = +
1
L R
L-R fuzzy nos. –
smooth
approximations
to triangular
fuzzy numbers
92
• Rule system from least-squares
algorithm
93
x training setcounting algorithm
weighted countingalgorithm
94
x training setleast-squares
6th
order polynomial
Least-squares
[correlation = 0.96]
6th order
polynomial
[correlation = 0.89]
95
Page 25
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 25/26
J. Application: Reservoir
Operation*
• Premises:
– Reservoir pool elevations
– Inflows [net]
– Forecasted demands (power)
– Time of year
96
•Shresha, B., L. Duckstein, and E. Stakhiv. (1996).“Fuzzy Rule-Based Modeling of Reservoir
Operation,” ASCE Journal of Water Resources
Planning and Management , 122(4), 262-269.
• Consequence:
– Actual release
• Typical Rule:IF pool elevation is Ai ,1 AND
Net inflows is Ai ,2 AND
Forecast demand is Ai ,3 AND
Time of year is Ai ,4
THEN release is Bi
97
• Uses split sampling
• Calibration:
– Training set
– Weighted counting algorithm• Storage elevation premises
calculated from mass balance:
St+1=St + I t – R t - Lt
98
consequence
• Following constraints imposed:
St,min < St < St,max
R t,min < R t < R t,max
• Applied to Tenkiller Lake inOklahoma
• Project purposes:
– flood control
– water supply
– hydropower
99
– recreation
– habitat
Page 26
7/17/2019 DSS- Fuzzy (literature)
http://slidepdf.com/reader/full/dss-fuzzy-literature 26/26
• Capacity: 371,000 AF
• Daily data from 1980 to 1992
• Training set uses 1989• 9 TFN’s cover elevation 620 ft. to
677.2 ft.
• 8 TFN’s: inflows from 0 to 180,000
day-second-ft
• Supports cover entire range—25%overlap
100
• Training for each month separately
• Power demands: low, medium-low,
medium, medium-high, high• Product inference used [danger of
incompleteness
• Additive combination
101
102 103