Vous êtes sur la page 1sur 16

Intelligent Control Systems

EPM643
Module 6: Fuzzy Logic Modelling

Oct 2017 Dr. Khaled El-Metwally, CU 1


Today Objectives
• Part I
– Optimization of Fuzzy Controller gains with PSO
– Fuzzy basis functions
• Part II
– Fuzzy Modelling Using Local Approximation in
Fuzzy Partition Space
– Fuzzy Modelling Using Sector Non-linearity
Principle

Oct 2017 Dr. Khaled El-Metwally, CU 2


PSO Tuning of FLC
SSE=S e^2

Oct 2017 Dr. Khaled El-Metwally, CU 3


Fuzzy basis functions
Two input one output system using TS defuzzifcation we get :
Ri: If x1is Mi and x2 is Mi Then Ui
𝑛
Rules memberships using product t-norm 𝜇𝑅𝑖 (𝑥) = 𝑤𝑖 (𝑥) = 𝑗=1 𝜇𝑗𝑖 (𝑥𝑗 )

𝑟 𝑟
𝑖 𝜇𝑅𝑖 (𝑥)𝑈𝑅𝑖 𝑖 𝑤𝑖 (𝑥)𝑈𝑅𝑖
𝑈𝑓 = 𝑟 = 𝑟
𝑖 𝜇𝑅𝑖 (𝑥) 𝑖 𝑤𝑖 (𝑥)
𝑤𝑖 (𝑥)
𝜂 𝑥 = ℎ𝑖 (𝑥) = 𝑟
𝑖 𝑤𝑖 (𝑥)

𝑈𝑓 = ℎ𝑖 (𝑥) 𝑈𝑅𝑖
𝑖

Oct 2017 Dr. Khaled El-Metwally, CU 4


Fuzzy Modelling Using Local Approximation in
Fuzzy Partition Space
The basic idea of the fuzzy modelling using local approximation is to represent the
plant by a set of linear models corresponding to different ranges of x, then blending
these models to get the net linear model at any specific value of x. This method results
in better linearization over a wide range of x.
Given :
𝑥=𝑓 𝑥
Then a lineartized model can be obtained as:

𝜕𝑓
𝑥 = 𝑓 𝑥0 + (𝑥 − 𝑥0 )
𝜕𝑥 𝑥0

𝑥 = 𝐴𝑥 + 𝐹

Oct 2017 Dr. Khaled El-Metwally, CU 5


Fuzzy Modelling Using Local Approximation in Fuzzy
Partition Space
where:
𝜕𝑓 𝜕𝑓
𝐴 = 𝜕𝑥 and 𝐹 = 𝑓 𝑥0 − 𝜕𝑥 𝑥0
𝑥0 𝑥0

𝑥0 is a fixed point around which linearization is to be operated This linearized model is


only valid for 𝑥 − 𝑥0 is small enough
1. Define the universe of discourse for each xi , i=1,..,n.
2. Assign for each universe of discourse group of membership functions
3. Construct the rule base using the (Takagi-Sugeno) TS fuzzy model:
Ri: If x1(t) is M1i and .. and xn is Mni Then 𝑥 = 𝐴𝑖 𝑥 + 𝐹𝑖
i=1,..,r

Oct 2017 Dr. Khaled El-Metwally, CU 6


Fuzzy Modelling Using Local Approximation in
Fuzzy Partition Space
1. Using TS defuzzifcation we get :
𝑟

𝑥= ℎ𝑖 (𝑥)(𝐴𝑖 𝑥 + 𝐹𝑖 )
𝑖

𝑤𝑖 (𝑥)
ℎ𝑖 (𝑥) = 𝑟
𝑖 𝑤𝑖 (𝑥)

𝑤𝑖 (𝑥) = 𝜇𝑗𝑖 (𝑥𝑗 )


𝑗=1

Oct 2017 Dr. Khaled El-Metwally, CU 7


Example 1

Oct 2017 Dr. Khaled El-Metwally, CU 8


Oct 2017 Dr. Khaled El-Metwally, CU 9
Oct 2017 Dr. Khaled El-Metwally, CU 10
Oct 2017 Dr. Khaled El-Metwally, CU 11
Fuzzy Modelling Using Sector Non-linearity Principle

Oct 2017 Dr. Khaled El-Metwally, CU 12


Oct 2017 Dr. Khaled El-Metwally, CU 13
Oct 2017 Dr. Khaled El-Metwally, CU 14
Example 1

Oct 2017 Dr. Khaled El-Metwally, CU 15


Exercises

Oct 2017 Dr. Khaled El-Metwally, CU 16

Vous aimerez peut-être aussi