Fuzzy Control and Neural Networks

Dr. Nam H. Nguyen

Department of Automatic Control
School of Electrical Engineering

https://sites.google.com/view/n2c

9/15/2020 2:52 PM Dr. Nam H. Nguyen 1

 References

• [1]. Lý thuyết điều khiển mờ, Phan Xuân Minh và Nguyễn

Doãn Phước.
• [2]. Fuzzy Logic Toolbox User’s Guide, R2014b.
• [3]. Fuzzy Logic – Controls, Concepts, Theories and
Applications, Elmer P. Dadios
• [4]. Neural network design, Martin Hagan.
• [5]. Neural Network Toolbox User’s Guide.
• [6]. Fuzzy control and neural network

9/15/2020 2:52 PM Dr. Nam H. Nguyen 2

 Contents

• I. Fuzzy logic
• II. Fuzzy controller
• III. Neural networks and applications

9/15/2020 2:56 PM Dr. Nam H. Nguyen 3

 Chapter 1 Fuzzy set and fuzzy logic

• 1.1 Fuzzy set and membership function

• 1.2 The basic operation of fuzzy set
• 1.3 Linguistic variable and fuzzy rules
• 1.4 Defuzzification methods

9/15/2020 2:52 PM Dr. Nam H. Nguyen 4

 1.1 Fuzzy set and membership function

• 1.1.1 History
• 1.1.2 Set and membership function
• 1.1.3 Operations
• 1.1.4 Fuzzy set

9/15/2020 2:52 PM Dr. Nam H. Nguyen 5

 1.1.1 History

• Fuzzy set theory has been developed since 1965 by Lotfi A.

Zadeh [1].
• First fuzzy controller was made in 1975 by E. H. Mamdami
[2].

[1] http://www.cs.berkeley.edu/~zadeh/ [2] http://www.iis.ee.ic.ac.uk/~amamdani/

9/15/2020 2:52 PM Dr. Nam H. Nguyen 6

 1.1.2 Crisp Set

• A classical (crisp) set is normally defined as a collection of

elements or objects.
• If an element 𝑥 belongs to the set A then we denote 𝑥 ∈ 𝑨.
• If an element 𝑥 does not belong to the set A then we denote
𝑥 ∉ 𝑨.

9/15/2020 2:52 PM Dr. Nam H. Nguyen 7

 1.1.2 Crisp Set

• A classical (crisp) set can be described in different ways:

➢ Enumerate the elements that belong to the set
Example 1.1: A set of female students in a class 𝐴 =
𝐻𝑜𝑎, 𝑀𝑎𝑖, 𝑃ℎượ𝑛𝑔 … , 𝑇ℎ𝑎𝑛ℎ
➢ State conditions for membership
Example 1.2: A set of positive error that is lesser than 0.01
𝐵 = 𝑒|0 < 𝑒 < 0.01
➢ Use the characteristic function
1, 𝑖𝑓 𝑥 > 0
Example 1.3: 𝑓(𝑥) = ቊ
0, 𝑖𝑓 𝑥 ≤ 0

9/15/2020 2:52 PM Dr. Nam H. Nguyen 8

 1.1.2 Crisp Set

• Empty set: Φ
A set has no element.
Example 1.4: A set of real roots of the equation 𝑥 2 + 1 = 0
• Subset: 𝐴 ⊂ 𝐵 ⟺ ∀𝑥, 𝑥 ∈ 𝐴 ⟹ 𝑥 ∈ 𝐵
- If A is not included in B, then we represent as 𝐴 ⊄ 𝐵
- 𝐴 = 𝐵 ⟺ 𝐴 ⊂ 𝐵 𝑎𝑛𝑑 𝐵 ⊂ 𝐴 (equal sets)
Example 1.5: The set of integers is a subset of the real number
set

9/15/2020 2:52 PM Dr. Nam H. Nguyen 9

 1.1.2 Crisp Set

• Properties:
✓ A ⊂ A for every A.
✓ If A ⊂ B and B ⊂ C then A ⊂ C
✓ Φ ⊂ A for every A.
• Membership function: Figure 1.1 Subset property
1, 𝑖𝑓 𝑥 ∈ 𝐴
𝜇𝐴 𝑥 = ቊ
0, 𝑖𝑓 𝑥 ∉ 𝐴
Example 1.6: A set of cool temperatures
𝐴 = 𝑥|20 𝑜𝐶 ≤ 𝑥 ≤ 25 𝑜𝐶
1, 𝑛ế𝑢 𝑥 ∈ 𝐴
𝜇𝐴 𝑥 = ቊ
0, 𝑛ế𝑢 𝑥 ∉ 𝐴 Figure 1.2 Membership function for
the set A
9/15/2020 2:52 PM Dr. Nam H. Nguyen 10
 1.1.3 Operation of crisp sets
• A) Intersection: 𝐴 ∩ 𝐵 = 𝑥|𝑥 ∈ 𝐴 𝑎𝑛𝑑 𝑥 ∈ 𝐵
𝜇𝐴∩𝐵 𝑥 = 𝜇𝐴 𝑥 . 𝜇𝐵 𝑥 or
= min(𝜇𝐴 𝑥 , 𝜇𝐵 (𝑥))
Example 1.7: Given 𝐴 = 𝑥|1 < 𝑥 < 3 and B = 𝑥|2 < 𝑥 < 4
𝐴 ∩ 𝐵 = 𝑥|2 < 𝑥 < 3

Figure 1.3 Membership functions

9/15/2020 2:52 PM Dr. Nam H. Nguyen 11
 1.1.3 Operation of crisp sets

• B) Union
𝐴 ∪ 𝐵 = 𝑥|𝑥 ∈ 𝐴 𝑂𝑅 𝑥 ∈ 𝐵
𝜇𝐴∪𝐵 𝑥 = 𝜇𝐴 𝑥 + 𝜇𝐵 𝑥 − 𝜇𝐴 𝑥 . 𝜇𝐵 𝑥 or
= max(𝜇𝐴 𝑥 , 𝜇𝐵 (𝑥))
Example 1.8: 𝜇𝐴∪𝐵 𝑥 = 𝑥|1 < 𝑥 < 4

Figure 1.4 Membership function

9/15/2020 2:52 PM 12
Dr. Nam H. Nguyen
 1.1.3 Operation of crisp sets
• C) Complement
Assume 𝐴 ⊂ 𝑈, the complement set of 𝐴 is defined as
𝐴𝑐 = 𝑥|𝑥 ∈ 𝑈 𝑎𝑛𝑑 𝑥 ∉ 𝐴 (c: complement)
1, 𝑛ế𝑢 𝑥 ∉ 𝐴
𝜇𝐴𝑐 𝑥 = ቊ = 1 − 𝜇𝐴 𝑥
0, 𝑛ế𝑢 𝑥 ∈ 𝐴
Example 1.9: 𝐴𝑐 = 𝑥|𝑥 < 1 𝑜𝑟 𝑥 > 3

Figure 1.5 Membership functions

9/15/2020 2:52 PM Dr. Nam H. Nguyen 13
 1.1.3 Operation of crisp sets

• Some operational characteristics of intersection

✓ Commutativity: 𝐴 ∩ 𝐵 = 𝐵 ∩ 𝐴
✓ Associativity: 𝐴 ∩ 𝐵 ∩ 𝐶 = 𝐴 ∩ 𝐵 ∩ 𝐶
✓ Distributivity: 𝐴 ∩ (𝐵 ∪ 𝐶) = (𝐴 ∩ 𝐵) ∪ (𝐴 ∩ 𝐶)
✓ De Morgan’s law: (𝐴 ∩ 𝐵)𝑐 = 𝐴𝑐 ∪ 𝐵𝑐
• Some operational characteristics of union
✓ Commutativity: 𝐴 ∪ 𝐵 = 𝐵 ∪ 𝐴
✓ Associativity: 𝐴 ∪ 𝐵 ∪ 𝐶 = 𝐴 ∪ 𝐵 ∪ 𝐶
✓ Distributivity: 𝐴 ∪ 𝐵 ∩ 𝐶 = (𝐴 ∪ 𝐵) ∩ (𝐴 ∪ 𝐶)
✓ De Morgan’s law: (𝐴 ∪ 𝐵)𝑐 = 𝐴𝑐 ∩ 𝐵𝑐

9/15/2020 2:52 PM Dr. Nam H. Nguyen 14

 1.1.4 Fuzzy set

• Definition: Given a universal set U, a fuzzy set F is described

by a membership function:
𝜇𝐹 𝑥 : 𝑈 → 0; 1 , 𝑥 ∈ 𝑈
• Example 1.10: A fuzzy set F has a membership function as
follows

−100𝑥 + 1, 𝑖𝑓 0 ≤ 𝑥 < 0.01

• 𝜇𝐹 𝑥 = ቐ100𝑥 + 1, 𝑖𝑓 − 0.01 < 𝑥 < 0
0, 𝑜𝑡ℎ𝑒𝑟𝑠

Figure 1.6 Membership function

of the fuzzy set F

9/15/2020 2:52 PM Dr. Nam H. Nguyen 15

 1.1.4 Fuzzy set

• Types of membership functions:

Trimf (TRIangular Membership Function)

0, 𝑖𝑓 𝑥 ≤ 𝑎
𝑥−𝑎
, 𝑖𝑓 𝑎 ≤ 𝑥 < 𝑏
𝜇𝐹 𝑥 = 𝑏𝑐 −
−𝑥
𝑎
, 𝑖𝑓 𝑏 ≤ 𝑥 < 𝑐
𝑐−𝑏
0, 𝑖𝑓 𝑐 ≤ 𝑥

9/15/2020 2:52 PM Dr. Nam H. Nguyen 16

 1.1.4 Fuzzy set

• Trapmf (TRAPezoidal Membership Function)

0, 𝑖𝑓 𝑥 ≤ 𝑎
𝑥−𝑎
, 𝑖𝑓 𝑎 ≤ 𝑥 < 𝑏
𝑏−𝑎
𝜇𝐹 𝑥 = 1, 𝑖𝑓 𝑏 ≤ 𝑥 ≤ 𝑐
𝑑−𝑥
, 𝑖𝑓 𝑐 ≤ 𝑥 < 𝑑
𝑑−𝑐
0, 𝑖𝑓 𝑑 ≤ 𝑥
• 𝜇𝐹 𝑥 = 𝑡𝑟𝑎𝑝𝑚𝑓(𝑥, [𝑎; 𝑏; 𝑐; 𝑑])

9/15/2020 2:52 PM Dr. Nam H. Nguyen 17

 1.1.4 Fuzzy set

Gbellmf (Generalized BELL Membership Function)

1
𝜇𝐹 𝑥 =
𝑥 − 𝑐 2𝑏
1+ 𝑎
a – width
𝑏 > 0 slope
c – center

9/15/2020 2:52 PM Dr. Nam H. Nguyen 18

 1.1.4 Fuzzy set

• Gaussmf (GAUSSian Membership Function)

1 𝑥−𝑐 2
−
𝜇𝐹 𝑥 = 𝑒 2 𝜎
c – center
𝜎 – width

9/15/2020 2:52 PM Dr. Nam H. Nguyen 19

 1.1.4 Fuzzy set

• Gauss2mf
1 𝑥−𝑐1 2
− ,𝑖𝑓 𝑥≤𝑐1
𝑒 2 𝜎
𝜇1 𝑥 = ቐ 1

1, 𝑖𝑓 𝑥 > 𝑐1
1 𝑥−𝑐2 2
− ,𝑖𝑓 𝑥≥𝑐2
𝑒 2 𝜎2
𝜇2 𝑥 = ቐ
1, 𝑖𝑓 𝑥 < 𝑐2
𝜇𝐹 𝑥 = 𝜇1 𝑥 *𝜇2 𝑥

9/15/2020 2:52 PM Dr. Nam H. Nguyen 20

 1.1.4 Fuzzy set

• Sigmf (SIGmoid Membership Function)

1
𝜇𝐹 𝑥 = = 𝑠𝑖𝑔(𝑥, 𝑎, 𝑐), a – shape
1+𝑒 −𝑎(𝑥−𝑐)
• Dsigmf ∶ 𝜇𝐹 𝑥 = |𝑠𝑖𝑔 𝑥, 𝑎1 , 𝑐1 − 𝑠𝑖𝑔 𝑥, 𝑎2 , 𝑐2 |
• Psigmf ∶ 𝜇𝐹 𝑥 = 𝑠𝑖𝑔 𝑥, 𝑎1 , 𝑐1 ∗ 𝑠𝑖𝑔 𝑥, 𝑎2 , 𝑐2 |

9/15/2020 2:52 PM Dr. Nam H. Nguyen 21

 1.1.4 Fuzzy set
• Zmf (Z-shaped curve MF)
1, 𝑖𝑓 𝑥 ≤ 𝑥1
2
𝑥 − 𝑥1 𝑥1 + 𝑥0
1−2 , 𝑖𝑓 𝑥1 ≤ 𝑥 <
𝑥0 − 𝑥1 2
𝜇𝐹 𝑥 = 2 = 𝑧𝑚𝑓(𝑥1 , 𝑥0 )
𝑥 − 𝑥1 𝑥1 + 𝑥0
2 , 𝑖𝑓 ≤ 𝑥 ≤ 𝑥0
𝑥0 − 𝑥1 2
0, 𝑖𝑓 𝑥0 ≤ 𝑥

• Smf (S-shaped curve MF): 𝜇𝐹 𝑥 = 1 − 𝑧𝑚𝑓(𝑥1 , 𝑥0 ) =

s𝑚𝑓(𝑥1 , 𝑥0 )
• Pmf (Pi-shaped curve MF): 𝜇𝐹 𝑥 = s𝑚𝑓 𝑥1 , 𝑥2 ∗ z𝑚𝑓(𝑥3 , 𝑥4 )

9/15/2020 2:52 PM Dr. Nam H. Nguyen 22

 1.2 Operation of fuzzy set

• Complement
μ𝐴𝑐 x = 1 − μA x (Zadeh)
𝑣 1/𝑣
μ𝐴𝑐 x = 1 − μA x (Yager)

9/15/2020 2:52 PM Dr. Nam H. Nguyen 23

 1.2 Operation of fuzzy set

• Example 1.11:
1
𝜇𝐴 𝑥 =
1 + 𝑥2
= Gbellmf (1, 1, 0)
𝑥2
𝜇𝐴𝑐 𝑥 =
1 + 𝑥2

9/15/2020 2:52 PM Dr. Nam H. Nguyen 24

 1.2 Operation of fuzzy set

• Union
μA⋃B x = max μA x , μB x for every x, Zadeh

μA⋃B x = μA x + μB x − μA x . μB x

μA⋃B x = min 1, μA x + μB x , Lukasiewicz

9/15/2020 2:52 PM Dr. Nam H. Nguyen 25

 1.2 Operation of fuzzy set

• Example 1.12: Union of two fuzzy sets using Zadeh’s method

9/15/2020 2:52 PM Dr. Nam H. Nguyen 26

 1.2 Operation of fuzzy set

• Intersection
μA∩B x = min μA x , μB x , ∀x ∈ U, Zadeh

μA∩B x = μA x . μB x , ∀x ∈ U

μA∩B x = max(μA x + μB x − 1,0), ∀x ∈ U, Lukasiewicz

9/15/2020 2:52 PM Dr. Nam H. Nguyen 27

 1.2 Operation of fuzzy set

Example 1.13:

9/15/2020 2:52 PM Dr. Nam H. Nguyen 28

 1.3 Linguistic variable

• Linguistic variables: A linguistic variable is characterized by

a quintuple (y, T(y),U,Gr,Ru)
- y: the name of the variable (for example, temperature, level,
speed)
- T(y): a set of names of linguistic values of y (small, big, …)
- U: a universal set, which is associated with the base variable u
- Gr: a syntactic rule for generating the names of linguistic
values (terms)
- Ru: a semantic rule for associating with each term a fuzzy set

9/15/2020 2:52 PM Dr. Nam H. Nguyen 29

 1.3 Linguistic variable

9/15/2020 2:52 PM Dr. Nam H. Nguyen 30

 1.3 Linguistic variable

• Example 1.14:
• y: error
• T(y): very big, big, small, very small, very very small
• U: 0; 10
• Gr: very
• Ru: associate with each value of linguistic variable a fuzzy set
μ𝑣𝑒𝑟𝑦 𝑏𝑖𝑔 𝑥 = 𝑡𝑟𝑖𝑚𝑓 𝑥, 7,5 10 12
μ𝑏𝑖𝑔 𝑥 = 𝑡𝑟𝑖𝑚𝑓 𝑥, 5 7,5 10
μ𝑠𝑚𝑎𝑙𝑙 𝑥 = 𝑡𝑟𝑖𝑚𝑓 𝑥, 2,5 5 7,5
μ𝑣𝑒𝑟𝑦 𝑠𝑚𝑎𝑙𝑙 𝑥 = 𝑡𝑟𝑖𝑚𝑓 𝑥, 0 2,5 5
μ𝑣𝑒𝑟𝑦 𝑣𝑒𝑟𝑦 𝑠𝑚𝑎𝑙𝑙 𝑥 = 𝑡𝑟𝑖𝑚𝑓 𝑥, −2 0 2,5
9/15/2020 2:52 PM Dr. Nam H. Nguyen 31
 1.3 If-then rules

• If-then rule
R: If 𝜒 is A Then 𝛾 is B
• The if-then rule is called the fuzzy implication or fuzzy
conditional statement.
• χ is A: called the antecedent or premise
• γ is B: called the consequence or conclusion
• The if-then rule can be abbreviated as
𝑅: 𝐴 ⟶ 𝐵

9/15/2020 3:07 PM Dr. Nam H. Nguyen 32

 1.3 If-then rules

• R can be viewed as a fuzzy set with a two-dimensional

membership function
μ𝑅 𝑥, 𝑦 = 𝑓(μ𝐴 𝑥 , μ𝐵 𝑦 )
𝑓: called the fuzzy implication function

𝑓 μ𝐴 𝑥 , μ𝐵 𝑦 = min( μ𝐴 𝑥 , μ𝐵 𝑦 𝑀𝑎𝑚𝑑𝑎𝑛𝑖

𝑓 μ𝐴 𝑥 , μ𝐵 𝑦 = μ𝐴 𝑥 . μ𝐵 𝑦 Larsen - PROD

9/15/2020 3:07 PM Dr. Nam H. Nguyen 33

 1.3 Compositional Rule of Inference

• R: If 𝜒 is A Then 𝛾 is B
• A•⊂ X, B ⊂ Y, x ∈ 𝑋 and y ∈ 𝑌
• When the input A’ is given to the inference system, the output
B’ is obtained through the inference operation denoted by the
composition operator “o”.
B’ = A’ o R
• This inference procedure is called as the “compositional rule
of inference”

9/15/2020 3:07 PM Dr. Nam H. Nguyen 34

 1.3 Compositional Rule of Inference

• Case 1: Singleton input, 𝑥 = 𝑥0

• Let 𝛼0 = μ𝐴 𝑥0
• 𝑎𝑛𝑑 μ𝑅 𝑥0 , 𝑦 = μ𝐵′ 𝑦 .
• 𝛼0 : called matching degree
• μ𝐵′ 𝑦 = min(𝛼0 , μ𝐵 𝑦 ) using Mamdani

• Case 2: Fuzzy input, A’

• 𝛼0 = max(min μ𝐴 𝑥 , μ𝐴′ 𝑥 )
• μ𝐵′ 𝑦 = min(𝛼0 , μ𝐵 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 35

 1.3 Compositional Rule of Inference

• Using Larsen’s method.

9/15/2020 3:07 PM Dr. Nam H. Nguyen 36

 1.3 Compositional Rule of Inference

• Fuzzy inference with rule base

𝑅1 : If 𝜒 is 𝐴1 Then 𝛾 is 𝐵1 OR
𝑅2 : If 𝜒 is 𝐴2 Then 𝛾 is 𝐵2 OR
…
𝑅𝑛 : If 𝜒 is 𝐴𝑛 Then 𝛾 is 𝐵𝑛

• B’ = A’ o ⋃𝑛𝑖=1 𝑅𝑖 = ⋃𝑛𝑖=1 A’ o 𝑅𝑖
• μ𝑅 𝑥, 𝑦 = ⋃𝑛𝑖=1 μ𝑅𝑖 𝑥, 𝑦

9/15/2020 3:07 PM Dr. Nam H. Nguyen 37

 1.3 Compositional Rule of Inference

• The connective OR can be implemented as a MAX or SUM

operations
• The composition operator maybe:
❑ MAX – MIN
❑ MAX – PROD
❑ SUM – MIN
❑ SUM - PROD

9/15/2020 3:07 PM Dr. Nam H. Nguyen 38

 1.3 Compositional Rule of Inference

• Example 1.15:
• μ𝐴1 𝑥 = 𝑡𝑟𝑖𝑚𝑓(𝑥, [1 2 3]),
• μ𝐴2 𝑥 = 𝑡𝑟𝑖𝑚𝑓(𝑥, [2 3 4]),
• μ𝐵1 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [1 3 5]),
• μ𝐵2 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [3 5 7]),
• Rule base:
𝑅1 : If 𝜒 is 𝐴1 Then 𝛾 is 𝐵1 OR
𝑅2 : IF 𝜒 is 𝐴2 Then 𝛾 is 𝐵2
• Determine μ𝑅 𝑥0 , 𝑦 using the composition operation MAX –
MIN, given 𝑥0 = 2.75.

9/15/2020 3:07 PM Dr. Nam H. Nguyen 39

 1.3 Compositional Rule of Inference

• 𝛼1 = μ𝐴1 𝑥0 = 3 − 2,75 = 0,25

• μ𝑅1 2,75, 𝑦 = min(0,25, μ𝐵1 𝑦 )
• 𝛼2 = μ𝐴2 𝑥0 = 2,75 − 2 = 0,75
• μ𝑅2 2,75, 𝑦 = min(0,75, μ𝐵2 𝑦 )
• μ𝑅 𝑥, 𝑦 =
𝑀𝐴𝑋{min 0,25, μ𝐵1 𝑦 , min 0,75, μ𝐵2 𝑦 }
• problem_fuzzy_rule.m

9/15/2020 3:07 PM Dr. Nam H. Nguyen 40

 1.3 Compositional Rule of Inference
1 1
A1 B1
A2 B2
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1 1
A1 Mu1
Do thoa man B1
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1 1
A2 Mu2
Do thoa man B2
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1
MuR

0.5

0
0 1 2 3 4 5 6 7 8

9/15/2020 3:07 PM Dr. Nam H. Nguyen 41

 1.3 Compositional Rule of Inference

1 1
A1 B1
A2 B2
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1 1
A1 Mu1
Do thoa man B1
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1 1
A2 Mu2
Do thoa man B2
0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8

1
MuR

0.5

0
0 1 2 3 4 5 6 7 8

9/15/2020 3:07 PM Dr. Nam H. Nguyen 42

 1.3 Compositional Rule of Inference

• The IF THEN rule for MISO system.

𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 THEN 𝛾 is 𝐵1
This is equivalent to
𝑅1,1 : IF 𝜒1 is 𝐴1,1 THEN 𝛾 is 𝐵1 AND
𝑅1,2 : 𝜒2 is 𝐴2,1 THEN 𝛾 is 𝐵1
where the connective “AND” is an intersection operator, which
works as a MIN function
Thus:
μ𝑅1 𝑥1 , 𝑥2 , 𝑦 = 𝑀𝐼𝑁(μ𝑅1,1 𝑥1 , 𝑦 , μ𝑅1,2 𝑥2 , 𝑦 )
Let μ𝑅1 𝑥1 , 𝑥2 , 𝑦 = μ𝑅1 𝑦 = μ𝐵1′ 𝑦

9/15/2020 3:07 PM Dr. Nam H. Nguyen 43

 1.3 Compositional Rule of Inference

• Singleton inputs: 𝑥1 = 𝑥1,0 𝑎𝑛𝑑 𝑥2 = 𝑥2,0

Let 𝛼1,1 = μ𝐴1,1 𝑥1,0 𝑎𝑛𝑑 μ𝐵′1,1 𝑦 = μ𝑅1,1 𝑥1,0 , 𝑦 .
μ𝐵′1,1 𝑦 = min(𝛼1,1 , μ𝐵1 𝑦 ) ---- Zadeh

Let 𝛼1,2 = μ𝐴2,1 𝑥2,0 𝑎𝑛𝑑 μ𝐵′1,2 𝑦 = μ𝑅1,2 𝑥2,0 , 𝑦 .

μ𝐵′1,2 𝑦 = min(𝛼1,2 , μ𝐵1 𝑦 ) ---- Zadeh

μ𝐵1′ = 𝑀𝐼𝑁( μ𝐵′1,1 𝑦 , μ𝐵′1,2 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 44

 1.3 Compositional Rule of Inference

μ𝐵1′ = 𝑀𝐼𝑁[min(𝛼1,1 , μ𝐵1 𝑦 ), min 𝛼1,2 , μ𝐵1 𝑦 ]

=𝑀𝐼𝑁[𝑚𝑖𝑛(𝛼1,1, 𝛼1,2 ), μ𝐵1 𝑦 ] ------ Zadeh

• Conclusion:
✓ The matching degree (MISO) is
𝛼𝑖 = 𝑀𝐼𝑁(𝛼𝑖,1 , 𝛼𝑖,2 ,…𝛼𝑖,𝑗 , …)
𝛼𝑖,𝑗 = μ𝐴𝑖,𝑗 (𝑥𝑗 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 45

 1.3 Compositional Rule of Inference
• The rule base for MISO system.
𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 AND … AND 𝜒𝑚 is 𝐴𝑚,1
THEN 𝛾 is 𝐵1 OR
𝑅2 : IF 𝜒1 is 𝐴1,2 AND 𝜒2 is 𝐴2,2 AND … AND 𝜒𝑚 is 𝐴𝑚,2
THEN 𝛾 is 𝐵2 OR
…
𝑅𝑛 : IF 𝜒1 is 𝐴1,𝑛 AND 𝜒2 is 𝐴2,𝑛 AND … AND 𝜒𝑚 is 𝐴𝑚,𝑛
THEN 𝛾 is 𝐵𝑛
𝑛

μ𝑅 𝒙, 𝑦 = ራ μ𝑅𝑖 𝑦
𝑖=1
𝒙 = x1 x2 … x𝑚
9/15/2020 3:07 PM Dr. Nam H. Nguyen 46
 1.3 Compositional Rule of Inference

• μ𝑅𝑖 𝑦 = min 𝛼𝑖 , μ𝐵𝑖 𝑦 --- Zadeh

• 𝛼𝑖 = 𝑀𝐼𝑁(𝛼𝑖,1 , 𝛼𝑖,2 ,…𝛼𝑖,𝑚 ) – matching degree for the rule 𝑅𝑖
• 𝛼𝑖,𝑗 = μ𝐴𝑗,𝑖 (𝑥𝑗 )
• μ𝑅 𝒙, 𝑦 = 𝑀𝐴𝑋 μ𝑅1 𝑦 , μ𝑅2 𝑦 , … , μ𝑅𝑚 𝑦

9/15/2020 3:07 PM Dr. Nam H. Nguyen 47

Example
μ𝐴1,1 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [1 2 3]),
μ𝐴1,2 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [2 3 4]),
μ𝐴2,1 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−3 − 2 − 1]),
μ𝐴2,2 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−4 − 3 − 2]),
μ𝐵1 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [1 3 5]), μ𝐵2 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [3 5 7]),
𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 THEN 𝛾 is 𝐵1 OR
𝑅2 : IF 𝜒1 is 𝐴1,2 AND 𝜒2 is 𝐴2,2 THEN 𝛾 is 𝐵2
Find μ𝑅 𝒙𝟎 , 𝑦 using composition operator MAX - MIN,
Given 𝒙𝟎 = [2.75; −2.5]

9/15/2020 3:07 PM Dr. Nam H. Nguyen 48

 Example

• μ𝐴1,1 𝑥1 = 3 − 𝑥1 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼1,1 = μ𝐴1,1 𝑥1,0 = 3 − 2.75 = 0.25
• μ𝐴2,1 𝑥2 = 𝑥2 + 3 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼1,2 = μ𝐴2,1 𝑥2,0 = −2.5 + 3 = 0.5
• 𝛼1 = min 𝛼1,1 , 𝛼1,2 = min 0.25,0.5 = 0.25
• μ𝑅1 𝒙𝟎 , 𝑦 = min(0.25, μ𝐵1 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 49

 Example

• μ𝐴1,2 𝑥1 = 𝑥1 − 2 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼2,1 = μ𝐴1,2 𝑥1,0 = 2.75 − 2 = 0.75
• μ𝐴2,2 𝑥2 = −2 − 𝑥2 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼2,2 = μ𝐴2,2 𝑥2,0 = −2 − (−2.5) = 0.5
• 𝛼2 = min 𝛼2,1 , 𝛼2,2 = min 0.5,0.75 = 0.5
• μ𝑅2 𝒙𝟎 , 𝑦 = min(0.5, μ𝐵2 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 50

 1.4 Example
• μ 𝑅 𝒙𝟎 , 𝑦 =
𝑀𝐴𝑋(min(0.25, μ𝐵1 𝑦 ), min(0.5, μ𝐵2 𝑦 ))
1 1 1
A11 A21 Mu1
0.8 alpha11 0.8 alpha12 0.8
B1
0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 1 2 3 4 5 -5 -4 -3 -2 -1 0 0 1 2 3 4 5 6 7 8

1 1 1
A12 A22 Mu1
0.8 alpha21 0.8 alpha22 0.8
B2
0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 1 2 3 4 5 -5 -4 -3 -2 -1 0 0 1 2 3 4 5 6 7 8

1
MuR
0.8

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8

9/15/2020 3:07 PM Dr. Nam H. Nguyen 51

 1.4 Defuzzification

• The process of finding a non-fuzzy value y’ from μ𝑅 𝒙, 𝑦 is

called defuzzification.
• Three methods of defuzzification:
✓Maximum
✓Centroid (Center of area)
✓Bisector of area
• Maximum
𝐻 = max(μ𝑅 𝒙, 𝑦 )
𝐺 = 𝑦 ⊂ 𝑌|(μ𝑅 𝑦 = 𝐻
Let 𝑦1 = min(𝐺) and 𝑦2 = m𝑎𝑥(𝐺).

9/15/2020 3:07 PM Dr. Nam H. Nguyen 52

 1.4 Defuzzification

• Three options to determine y’:

➢ MOM – mean value of maximum
′
𝑦1 + 𝑦2
𝑦 =
2
➢ SOM – smallest value of maximum
𝑦′ = 𝑦1
➢ LOM – largest value of maximum
𝑦′ = 𝑦2

9/15/2020 3:07 PM Dr. Nam H. Nguyen 53

 1.4 Defuzzification

• Centroid
Let S = 𝑦 ∈ 𝑌|μ𝑅 𝑦 > 0

′ ‫𝑦 𝑆׬‬μ𝑅 𝑦 𝑑𝑦
𝑦 =
‫ 𝑆׬‬μ𝑅 𝑦 𝑑𝑦

• BISECTOR
y’ satisfies the condition
𝑦′ 𝑏
න μ𝑅 𝑦 𝑑𝑦 = න μ𝑅 𝑦 𝑑𝑦
𝑎 𝑦′
a = min(𝑆) và b = m𝑎𝑥(𝑆)

9/15/2020 3:07 PM Dr. Nam H. Nguyen 54

 1.4 Defuzzification
• Example:
μ𝐴1,1 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [1,2,3]),
μ𝐴1,2 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [2,3,4]),
μ𝐴2,1 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−3, −2, −1]),
μ𝐴2,2 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−4, −3, −2]),
μ𝐵1 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [1,3,5]), μ𝐵2 𝑦 = 𝑡𝑟𝑖𝑚𝑓(𝑦, [3,5,7])
𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 THEN 𝛾 is 𝐵1 OR
𝑅2 : IF 𝜒1 is 𝐴1,2 AND 𝜒2 is 𝐴2,2 THEN 𝛾 is 𝐵2
Find y’ using composition operator MAX - MIN,
given 𝒙𝟎 = [2.75, −2.5]

9/15/2020 3:07 PM Dr. Nam H. Nguyen 55

 1.4 Defuzzification

• μ𝐴1,1 𝑥1 = 3 − 𝑥1 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼1,1 = μ𝐴1,1 𝑥1,0 = 3 − 2.75 = 0.25
• μ𝐴2,1 𝑥2 = 𝑥2 + 3 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼1,2 = μ𝐴2,1 𝑥2,0 = −2.5 + 3 = 0.5
• 𝛼1 = min 𝛼1,1 , 𝛼1,2 = min 0.25,0.5 = 0.25
• μ𝑅1 𝒙𝟎 , 𝑦 = min(0.25, μ𝐵1 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 56

 1.4 Defuzzification

• μ𝐴1,2 𝑥1 = 𝑥1 − 2 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼2,1 = μ𝐴1,2 𝑥1,0 = 2.75 − 2 = 0.75
• μ𝐴2,2 𝑥2 = −2 − 𝑥2 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼2,2 = μ𝐴2,2 𝑥2,0 = −2 − (−2.5) = 0,5
• 𝛼2 = min 𝛼2,1 , 𝛼2,2 = min 0.5,0.75 = 0.5
• μ𝑅2 𝒙𝟎 , 𝑦 = min(0.5, μ𝐵2 𝑦 )

9/15/2020 3:07 PM Dr. Nam H. Nguyen 57

 1.4 Defuzzification
• μ 𝑅 𝒙𝟎 , 𝑦 =
𝑀𝐴𝑋(min(0.25, μ𝐵1 𝑦 ), min(0.5, μ𝐵2 𝑦 ))
1 1 1
A11 A21 Mu1
0.8 alpha11 0.8 alpha12 0.8
B1
0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 1 2 3 4 5 -5 -4 -3 -2 -1 0 0 1 2 3 4 5 6 7 8

1 1 1
A12 A22 Mu1
0.8 alpha21 0.8 alpha22 0.8
B2
0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 1 2 3 4 5 -5 -4 -3 -2 -1 0 0 1 2 3 4 5 6 7 8

1
MuR
0.8

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8

9/15/2020 3:07 PM Dr. Nam H. Nguyen 58

 1.4 Defuzzification

Maximum:
H = max(μR 𝐱 𝟎 , y )=0.5
G = y ⊂ Y|(μR y = H = y ⊂ Y 4 ≤ y ≤ 6
y1 = min G = 4
y2 = max G = 6
Thus, y’ can be:
y’ = 4, or
y’ = 6, or
Y’ = 5
Centroid:

9/15/2020 3:07 PM Dr. Nam H. Nguyen 59

 1.4 Defuzzification

• S = 𝑦 ∈ 𝑌|μ𝑅 𝑦 > 0 = 𝑦 ∈ 𝑌|1 < 𝑦 < 7

7
• 𝑆𝑑 = ‫ 𝑆׬‬μ𝑅 𝑦 𝑑𝑦 = ‫׬‬1 μ𝑅 𝒙𝟎 , 𝑦 𝑑𝑦
• = 0.5 ∗ 0.5 ∗ 0.25 + 2 ∗ 0.25 + 0.5 ∗ 0.25 + 0.5 ∗ 0.5 + 2 ∗
0.5 + 0.5 ∗ 0.5 = 2
7
• 𝑆𝑛 = ‫ 𝑆׬‬yμ𝑅 𝑦 𝑑𝑦 = ‫׬‬1 𝑦μ𝑅 𝑦 𝑑𝑦 =
1.5 3.5
‫׬‬1 𝑦 ∗ 0.5 ∗ (𝑦 − 1)𝑑𝑦 + ‫׬‬1.5 𝑦 ∗ 0.25𝑑𝑦 +
4 6
‫׬‬3.5 𝑦 ∗ 0.5 ∗ (𝑦 − 3)𝑑𝑦 + ‫׬‬4 𝑦 ∗ 0.5𝑑𝑦 +
7
‫׬‬6 𝑦 ∗ 0.5 ∗ (7 − 𝑦)𝑑𝑦 = 8.625
• y’ = Sn/Sd = 4.3125

9/15/2020 3:07 PM Dr. Nam H. Nguyen 60

 1.5 Sugeno fuzzy model

𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 AND … AND 𝜒𝑚 is 𝐴𝑚,1

THEN 𝑦 = 𝑔1 (𝒙) OR
𝑅2 : IF 𝜒1 is 𝐴1,2 AND 𝜒2 is 𝐴2,2 AND … AND 𝜒𝑚 is 𝐴𝑚,2
THEN y = 𝑔2 (𝒙) OR
…
𝑅𝑛 : IF 𝜒1 is 𝐴1,𝑛 AND 𝜒2 is 𝐴2,𝑛 AND … AND 𝜒𝑚 is 𝐴𝑚,𝑛
THEN y = 𝑔𝑛 (𝒙)
𝒙 = x1 x2 … x𝑚
𝑔𝑖 (𝒙) polynomial functions of 𝒙

9/15/2020 3:07 PM Dr. Nam H. Nguyen 61

 1.5 Sugeno fuzzy model

The non-fuzzy output:

Weighted average – wtaver
′
𝛼1 𝑦1 + 𝛼2 𝑦2 + ⋯ + 𝛼𝑛 𝑦𝑛
𝑦 =
𝛼1 + 𝛼2 + ⋯ + 𝛼𝑛
𝑦𝑖 = 𝑔𝑖 (𝑥1 , 𝑥2 , … , 𝑥𝑚 )
• 𝛼𝑖 = 𝑀𝐼𝑁(𝛼𝑖,1 , 𝛼𝑖,2 ,…𝛼𝑖,𝑚 ) matching degree (firing strength)
• 𝛼𝑖,𝑗 = μ𝐴𝑖,𝑗 (𝑥𝑗 )
• Weighted sum - wtsum
𝑦 ′ = 𝛼1 𝑦1 + 𝛼2 𝑦2 + ⋯ + 𝛼𝑛 𝑦𝑛

9/15/2020 3:07 PM Dr. Nam H. Nguyen 62

 1.5 Sugeno fuzzy model

• Example
μ𝐴1,1 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [1,2,3]),
μ𝐴1,2 𝑥1 = 𝑡𝑟𝑖𝑚𝑓(𝑥1 , [2,3,4]),
μ𝐴2,1 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−3, −2, −1]),
μ𝐴2,2 𝑥2 = 𝑡𝑟𝑖𝑚𝑓(𝑥2 , [−4, −3, −2]),
𝑅1 : IF 𝜒1 is 𝐴1,1 AND 𝜒2 is 𝐴2,1 THEN 𝑦 = 𝑔1 𝑥1 , 𝑥2 OR
𝑅2 : IF 𝜒1 is 𝐴1,2 AND 𝜒2 is 𝐴2,2 THEN 𝑦 = 𝑔2 𝑥1 , 𝑥2
Find y’, given 𝒙𝟎 = 2.75, −2.5 , 𝑔1 𝑥1 , 𝑥2 = 3 and
𝑔2 𝑥1 , 𝑥2 = 5

9/15/2020 3:07 PM Dr. Nam H. Nguyen 63

 1.5 Sugeno fuzzy model

• μ𝐴1,1 𝑥1 = 3 − 𝑥1 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼1,1 = μ𝐴1,1 𝑥1,0 = 3 − 2.75 = 0.25
• μ𝐴2,1 𝑥2 = 𝑥2 + 3 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼1,2 = μ𝐴2,1 𝑥2,0 = −2.5 + 3 = 0.5
• 𝛼1 = min 𝛼1,1 , 𝛼1,2 = min 0.25,0.5 = 0.25

9/15/2020 3:07 PM Dr. Nam H. Nguyen 64

 1.5 Sugeno fuzzy model

• μ𝐴1,2 𝑥1 = 𝑥1 − 2 𝑓𝑜𝑟 2 < 𝑥1 < 3

• 𝑥1,0 = 2.75
• 𝛼2,1 = μ𝐴1,2 𝑥1,0 = 2.75 − 2 = 0.75
• μ𝐴2,2 𝑥2 = −2 − 𝑥2 𝑓𝑜𝑟 − 3 < 𝑥2 < −2
• 𝑥2,0 = −2.5
• 𝛼2,2 = μ𝐴2,2 𝑥2,0 = −2 − (−2.5) = 0,5
• 𝛼2 = min 𝛼2,1 , 𝛼2,2 = min 0.5,0.75 = 0.5

9/15/2020 3:07 PM Dr. Nam H. Nguyen 65

 1.5 Sugeno fuzzy model

• 𝛼1 = 0.25
• 𝛼2 = 0.5
𝛼1 𝑦1 +𝛼2 𝑦2 0.25∗3 + 0.5∗5 3.25
• 𝑦′ = 𝛼1 +𝛼2
= 0.25+0.5
= 0.75 = 4.33 – wtaver
• 𝑦 ′ = 𝛼1 𝑦1 + 𝛼2 𝑦2 = 3.25 - wtsum
• Find y’ when 𝑔1 𝑥1 , 𝑥2 = 𝑥1 + 𝑥2 − 1 and 𝑔2 𝑥1 , 𝑥2 = 𝑥1 −
𝑥2 + 1

9/15/2020 3:07 PM Dr. Nam H. Nguyen 66