Anda di halaman 1dari 10

Fuzzy Set

𝐴 = { (𝑥, 𝜇 𝐴 (𝑥))|𝑥 ∈ 𝑋 }

Tuples: 𝑥, 𝜇 𝐴 (𝑥)

Membership Function: 𝜇 𝐴 (𝑥) = 𝐷𝑒𝑔𝑟𝑒𝑒(𝑥 ∈ 𝐴) => non- fuzzy

Representations:
𝜇 𝐴 (𝑥1)⁄ 𝜇 𝐴 (𝑥2)⁄ 𝜇 𝐴(𝑥𝑛 )⁄ 𝜇 𝐴 (𝑥)⁄
𝐴={ 𝑥1 , 𝑥2 , … 𝑥𝑛 } = { 𝑥 |𝑥∈𝑋}

𝜇 𝐴 (𝑥1 )⁄ 𝜇 𝐴 (𝑥𝑛 )⁄ 𝜇 𝐴 (𝑥𝑖 )⁄


= 𝑥1 + ⋯ + 𝑥𝑛 = ∑ 𝑥𝑖
for continuous domain
𝜇 𝐴 (𝑥)⁄
𝐴 = ∫𝑋 𝑥

Properties
height ℎ𝑒𝑖𝑔ℎ𝑡 (𝐴 ) = max 𝜇 𝐴 (𝑥𝑖 ), normal fuzzy set (ℎ𝑒𝑖𝑔ℎ𝑡(𝐴 ) = 1) vs subnormal fuzzy set (ℎ𝑒𝑖𝑔ℎ𝑡(𝐴 ) < 1)
𝑥𝑖

normalized fuzzy set 𝐴 ′ = 𝑛𝑜𝑟𝑚 (𝐴) ⟺ 𝜇 𝐴′ (𝑥𝑖 ) = 𝜇 𝐴 (𝑥𝑖 )/ℎ𝑒𝑖𝑔ℎ𝑡(𝐴)

support: 𝑠𝑢𝑝𝑝(𝐴 ) = { 𝑥 ∈ 𝑋 |𝜇𝐴 (𝑥) > 0 } singleton: 𝑠𝑢𝑝𝑝(𝐴 ) = {𝑥 0 }

core: 𝑐𝑜𝑟𝑒(𝐴 ) = { 𝑥 ∈ 𝑋 |𝜇 𝐴 (𝑥) = 1 }

α-cut 𝐴 𝛼: { 𝑥 |𝜇 𝐴 (𝑥) > 𝛼 }

fuzzy set 𝐴, 𝛼1 , 𝛼2 ∈ [0, 1], 𝛼1 < 𝛼2 >> 𝐴 𝛼1 ⊇ 𝐴 𝛼2, 𝐴 𝛼1 ∩ 𝐴 𝛼2 = 𝐴 𝛼2, 𝐴 𝛼1 ∪ 𝐴 𝛼2 = 𝐴 𝛼1


convex fuzzy set: each of its α-cut is a convex set

Operations
𝐴 ∨ 𝐵 = 𝜇 𝐴∪𝐵 (𝑥) = max [𝜇 𝐴 (𝑥), 𝜇 𝐵 (𝑥)]
𝐴 ∧ 𝐵 = 𝜇 𝐴∩𝐵 (𝑥) = min [𝜇 𝐴 (𝑥), 𝜇 𝐵 (𝑥)]
!! prove T-norm

Fuzzy intersection: t-norm with different T – probabilistic intersection: 𝜇 𝐴∩𝐵 (𝑥) = 𝜇 𝐴 (𝑥) ⋅ 𝜇 𝐵 (𝑥)

Lukasiewicz(bold) interstn: 𝜇 𝐴∩𝐵 (𝑥) = max(0, 𝜇 𝐴 (𝑥) + 𝜇 𝐵 (𝑥) − 1)

Fuzzy Union, t-conorm: algebraic sum 𝜇 𝐴∪𝐵 (𝑥) = 𝜇 𝐴(𝑥) + 𝜇 𝐵 (𝑥) − 𝜇 𝐴 (𝑥) ⋅ 𝜇 𝐵 (𝑥)

Lukasiewicz 𝜇 𝐴∪𝐵 (𝑥) = min (1, 𝜇 𝐴 (𝑥) + 𝜇 𝐵 (𝑥))

Complement/Negation 𝐴:̅ 𝜇 𝐴̅ (𝑥) − 1 − 𝜇 𝐴(𝑥)

Subset 𝐴 ⊆ 𝐵: 𝜇 𝐴 (𝑥) ≤ 𝜇 𝐵 (𝑥) for all 𝑥

DeMorgan’s Law: ̅̅̅̅̅̅̅


𝐴 ∨ 𝐵 = 𝐴 ̅ ∧ 𝐵̅, ̅̅̅̅̅̅̅
𝐴 ∧ 𝐵 = 𝐴 ̅ ∨ 𝐵̅
Cartesian Product: 𝜇 𝐴1×𝐴2× … ×𝐴𝑛 (𝑥1 ,𝑥 2 ,… , 𝑥 𝑛) = min{ 𝜇 𝐴𝑖 (𝑥𝑖 ) }
Relations
Cartesian product -> degree of association

>> Binary Fuzzy Relation

𝑅 = { ( (𝑥, 𝑦) , 𝜇 𝑅 (𝑥, 𝑦) )|(𝑥, 𝑦) ∈ 𝑋 × 𝑌 } >> fuzzy matrix

domain: 𝑑𝑜𝑚 𝑅(𝑋, 𝑌) with 𝜇 𝑑𝑜𝑚𝑅 (𝑥) = max 𝜇 𝑅 (𝑥, 𝑦)


𝑦∈𝑌

range: 𝑟𝑎𝑛 𝑅 (𝑋, 𝑌) with 𝜇 𝑟𝑎𝑛𝑅 (𝑥) = max 𝜇 𝑅 (𝑥,𝑦)


𝑥∈𝑋

height: 𝐻(𝑅) = 𝑠𝑢𝑝𝑦∈𝑌𝑠𝑢𝑝𝑥∈𝑋 𝜇𝑅 (𝑥, 𝑦)

Projection and Cylindrical Extension


𝑠𝑢𝑝𝑈2 𝜇 𝐴 (𝑢)
𝑝𝑟𝑜𝑗𝑈1 (𝐴 ) = { ⁄𝑢1 | 𝑢1 ∈ 𝑈1 }

𝑠𝑢𝑝𝑈3 𝜇𝐴 (𝑢)
𝑝𝑟𝑜𝑗𝑈1 ×𝑈2 (𝐴) = { ⁄(𝑢 , 𝑢 ) | (𝑢1 , 𝑢2 ) ∈ 𝑈1 × 𝑈2 }
1 2

for fuzzy relation:

𝑅𝑋 = ∫𝑋 [max 𝜇𝑅 (𝑥, 𝑦)] /𝑥


𝑦

𝑒𝑥𝑡𝑈 (𝐴) = { 𝜇 𝐴(𝑢1 )/𝑢 | 𝑢 ∈ 𝑈 }

Intersection on Cartesian Product Space


𝐴1 × 𝐴 2 = 𝑒𝑥𝑡𝑋2 (𝐴1 ) ∩ 𝑒𝑥𝑡𝑋1 (𝐴 2 )

𝜇𝐴1 ×𝐴2 (𝑥1 , 𝑥 2 ) = 𝜇𝐴1 (𝑥1 ) ∧ 𝜇𝐴2 (𝑥2 )

Operations on Fuzzy Relations

Composition of Relations

Min “series connection” && Max “parallel connection”

Composition of Fuzzy Relation with Fuzzy set


Fuzzy relation R in X x Y: (x is approximately equal to y)

Fuzzy subset B of Y can be introduced by A through composition of A and R

𝐵 = 𝐴 ∘ 𝑅 = 𝑝𝑟𝑜𝑗𝑌 (𝑅 ∩ 𝑒𝑥𝑡𝑋×𝑌(𝐴 ))
Combination (intersection) -> Projection

Hedges

Fuzzy Arithmetic
Interval of confidence [ 𝑠, 𝑠]
[ 𝑠1 , 𝑠1 ], [ 𝑠2 , 𝑠2 ]: Equality, Intersection, Union, Inequality, Inclusion

=>> conservative: intend to make the resulting interval to be large to avoid losing true solution

=>> cause unusual

>> addition, multiplication: commutative, associative, distributive

>> -> ( ) !!!! conservative!!!


>> ->

>> ->

Interval Evaluation

𝑓𝐼 ({𝑋𝑖 }; {𝐴𝑗 }) = [ min 𝑓({𝑥𝑖 };{𝑎𝑗 }) , 𝑚𝑎𝑥 … ]


𝑥 𝑠∈𝑋𝑠
𝑎𝑡∈𝐴𝑡

>> , , -->
𝑎𝑥 𝑎
>> always reduce number of intervals involved! →1 =>> best interval solution
1+𝑥 +1
𝑥

Operational on Fuzzy Sets


for any function

Example:

Solution:

So that:
Letting

So and
Solved:

Fuzzy System
Fuzzy parameter,

Fuzzy input/output

Fuzzy Rule Base

>> Rule-based fuzzy system: linguistic (Mamdani) fuzzy system, nonlinear mapping

Fuzzy Rule

“firing rule”

Fuzzy Reasoning
𝑋−→ 𝑅: 𝑋 × 𝑌−→ 𝑌
Cylindrical extension c(A) --> intersection of c(A) and R --> projection c(A)∩F

>> Graphical: w clipping of B


Fuzzy Inference Systems
 Fuzzification of the input variables
 Evaluation of output for each rule

firing level:

o Mamdani’s minimum fuzzy implication rule, “Clipping”:

o Larsen’s product fuzzy implication rule, “Scaling”:

 Aggregation of the rules' outputs


 Defuzzification
o Centroid Defuzzification
 Centre of Gravity (computationally intensive)
 Centre Average Defuzzifier
 Mean of maxima

Fuzzy Modeling
where
Eg.

1
2
1
3

Solution:

Given that , from the figure:

(i) The corresponding outputs:

(ii) The fuzzy set Y:

(iii) Fuzzy membership values of the outputs:


(iv) Average value of the final output

(v) The three membership functions, by α-cut operation:


Rule 1:
1 1
α = 1 − 𝑥1 ,α = 1 − 𝑥 2
15 5
(𝑋1 )α = [1, 15 − 15α ]
(𝑋2 )α = [1, 5 − 5α]
(𝑌2)α = (𝑋1 ) ∝ + (𝑋2 )∝ = [1, 20 − 20α]
𝑦1 = 1, 𝑦2 = 20 − 20α
𝑦
→ 𝜇 𝑆(𝑦) = − + 1
20
𝑒𝑡𝑐 …

Parameters Identification in Static Fuzzy Modeling


Eg.

Solution:

Then
Discrete-Time Dynamic Fuzzy Systems

Fuzzy Control Systems


Basic Fuzzy Control Schemes

 Direct (low-level, Mamdani) fuzzy control


 Fuzzy supervisory (high-level, Takagi-Sugeno) control
 Fuzzy model-based control

Anda mungkin juga menyukai