1-1 Chapter 1 From Crisp to Fuzzy sets 1.1. Introduction ○ Uncertainty -- Arising from imprecision, vagueness non-specificity, inconsistency ○ Traditional view: -- Uncertainty is undesirable and should be avoided. Modern view: -- Uncertainty is not only an unavoidable plague, but it has a great utility. ○ Traditionally, three key characteristics are to be minimized in order to maximize system performance complexity incredibility uncertainty Although undesirable when considered alone, uncertainty becomes valuable when considered together with the other characteristics 1-2 e.g., allowing more uncertainty tends to (a) reduce complexity (b) increase credibility ○ Transition of views Newtonian mechanics- precise laws, analytic methods Statistical mechanics- probability theory Fuzzy mechanics- theories of uncertainty ○ Analytic methods- (a) Base on calculus (b) Involve a small number of variables that are related to one another in a predictable way. Statistical methods- (a) Base on probability theory (b) Specific manifestations of microscopic entities are replaces with their statistical averages, which are connected with appropriate macroscopic variables. 1-3 (c) Involve a large number of variables and a high degree of randomness. 1.2. Crisp Sets ○ Methods for describing sets 1. Enumeration A {a1 , a2 ,....., an } 2. Description A {x | p ( x)} 3. Characteristic function 1 x A mA ( x ) 0 x A 1-4 ○ Power sets of A︰P(A) Second order power set of A︰ P2 ( A) P( P( A)) 3 4 Higher order power set of A︰ P ( A), P ( A) ○ Relative complement (difference) ︰A-B Absolute complement U B B ○ General principle of duality , , U , , ○ Fundamental properties of set operations 1-5 Example: Show DeMorgan’s laws A B A B show 1) A B A B 2) A B A B (1) Show A B A B Let S A B and T A B Show A B A B show S T show x S x T Given x S => x A B => x A B => x A x B => x A x B => x A B T S T Similarly, T S (Assignment 1) so S=T (2) Show A B A B (Assignment 2) ○ Partial order Let X {a, b, c} Let P( X ) be the power set of X ignoring the empty set . 1-6 P( X ) {{a},{b},{c},{a, b},{a, c},{b, c},{a, b, c}} ( P( X ), ) forms a partial order, i.e., transitive , anti-symmetric (no loop) , reflective Arrows ( ) represent inclusion ( ) ○ Lattice – is formed by a partial ordering and for every pair A, B P ( X ) , exists an LUB (supremum , join , A B ) and a GLB (infimum , meet , A B ). ○ P ( X ) , = P( X ), , Boolean Lattice Boolean algebra e.g., ( A B iff A B B or A B A ) 1-7 ○ Partition: ( A) {Ai | i I , Ai A} i) Ai ii) Ai Aj iii) Ai A iI ○ Refinement relation : Let 1 , 2 be 2 partitions of A. 1 2 1 2 If Ai one and only one Aj s.t. Ai Aj => 1 is a refinement of 2 2 1 ○ Let ( A) be the set of all partitions of A ( A), forms a lattice (partition lattice). ○ Nested family: A = {A1 , A2 ,.............., An } if Ai Ai 1 A1 : innermost set, An : outermost set 1-8 ○ Convex set A: r , s A , and [0,1] t r (1 ) s A Any set defined by a single interval of real number is convex Any set defined by more than one separate interval can not be convex ○ For a partial ordering on A i) An upper bound r of A : x A, x r ii) A lower bound s of A : x A sx iii) r is a lowest upper bound (LUB) of A : iff (a) r is an upper bound of A (b) no a r is an upper bound of A s is a greatest lower bound (GLB) of A 1-9 iff (a)’ s is a lower bound of A (b)’ no b s is a lower bound of A iv) Supremum r: sup A iii) (a), (b) and r A Infimum s: inf A iii) (a)’, (b)’ and s A 1.3 Fuzzy Sets: Basic Types ○ Fuzzy sets -Sets with vague boundaries -Membership of x in A is a matter of degree to which x is in A ○ Utilization of fuzzy sets (1) Representation of uncertainty (2) Representation of conceptual entities e.g., expensive, close, greater, sunny, tall ○ Fuzzy Sets membership Crisp Sets characteristic function A : X [0,1] function mA : X { 0 , 1 } 1-10 e.g., 1 i) “close to 0” : A ( x) 1 10 x 2 1 ( x ) 2 ii) “very close to 0” : A 1 10 x 2 1 iii) “close to a” : A ( x) 1 10( x a)2 ○ Difference between crisp, random, and fuzzy variables: Crisp variable: a uniform probability distribution Random variable: a probability distribution Fuzzy variable: a membership function is associated with its domain. 1-11 ○ Generalization i) Ordinary fuzzy sets: A : X [0,1] Abbreviated as A : X [0,1] . i.e., Each element of X is assigned a particular real number (i.e., precise membership grades). ii) L-fuzzy sets: A : X L , where L is a partial order set. iii) Interval–valued fuzzy sets: A : X ([0,1]) , where ([0,1]) is the family of all closed interval in [0,1]. 1-12 iv) Fuzzy sets of type-K -- Interval–valued fuzzy sets possess fuzzy Intervals (a) Type-2: A : X ([0,1]) , where ([0,1]) : fuzzy power set of [0,1], the set of all ordinary fuzzy sets defined on [0,1]. (b) Type-3 1-13 v) Level-K fuzzy sets -- Elements in a universal set are themselves fuzzy sets. (a) Level-2: A : ( X ) [0,1] e.g., fuzzy set “x is close to r” x : a fuzzy variable r : a particular number , e.g., 5. (b) Level 3: vi) Combinations of interval-valued, L, type-K, level-K fuzzy sets. 1-14 1.4 Fuzzy Sets: Basic Concept ○ Example: 3 fuzzy sets defined on age. A1 : “young”, A2 :”middle-aged”, A3 : ”old” Membership functions: 1 A1 ( x) (35 x) /15 0 0 ( x 20) /15 A2 ( x) (60 x) /15 1 0 A3 ( x) ( x 45) /15 1 x 20 20 x 35 x 35 x 20 or x 60 20 x 35 45 x 60 35 x 45 x 45 45 x 60 x 60 1-15 ○ -cut A: A {x | A( x) } If 1 2 1 A 2 A Strong -cut A: A {x | A( x) } e.g., A1 [0,35 15 ] A2 [15 20, 60 15 ] (0,1] A3 [15 45,80] A1 (0,35 15 ) A2 (15 20, 60 15 ) [0,1) A3 (15 45,80) ○ Level set ( A) ︰ ( A) { | x X , s.t { | A } or A( x) } 1-16 e.g., Continuous case --( A1 ) ( A2 ) ( A3 ) [0,1] Discrete case --( D1 ) {0 , 0.13 , 0.27 , 0.4 , 0.5 , 0.67 , 0.8 , 0.93 , 1} ○ Support︰ S ( A) = [x X | A( x) 0] S ( A) 0 A , e.g., S ( D2 ) {22,24, ,58} ○ Core︰ 1 A (i,e, 1 - cut) 1-17 ○ Hight h(A)︰the largest membership grade h( A) sup A( x) xX ○ Normal︰h(A) = 1 Subnormal︰h(A) < 1 ○ Convex fuzzy set︰ (0,1] , -cut is convex 1-18 ○ Theorem 1.1: A convex fuzzy set on R iff x1 , x2 R , [0,1] , A( x1 (1 ) x2 ) min [ A( x1 ), A( x2 )] Proof︰ i, ( ) Given A : convex , x1, x2 , Let a = min[A(x1 ), A(x2 )] x1, x2 a A A : convex a A convex [0,1], x x1 (1 ) x2 a A (definition of convex set) A( x) a min[ A( x1 ), A( x2 )] ii, ( ) x1, x2 , Given A( x1 (1 ) x2 ) min[ A( x1 ), A( x2 )] (Show that (0,1] , A : convex ) x1, x2 , , s,t. A(x1 ) , A(x2 ) (i,e., x1,x2 A) (1) [0,1] A( x1 (1 ) x2 ) min[ A( x1), A( x2 )] min( , ) , i.e., x1 (1 ) x2 A (2) (1), (2) A : convex A : convex 1-19 ◎ Fuzzy Set Operations ․Standard complement︰ A( x) 1 A( x) ․Equilibrium points︰ A( x) A( x) A( x) A( x) 1 A( x) 2 A( x) 1 A( x) A( x) 0.5 ․Standard intersection︰ ( A B)( x) min[ A( x), B( x)] ․Standard union: ( A B)( x) max[ A( x), B( x)] ․Difference A B A B min( A( x),1 B( x)) Symmetric difference A B ( A B) ( B A) 1-20 ○ Example: 1-21 ○ Example: A1 A3 ? A1 A3 ︰not young and not old A2 ︰middle age 1-22 ○ Any fuzzy power set P(X) with form a lattice, referred to as a De Morgan lattice (De Morgan algebra) In such a lattice , A, B P( X ) , join︰ A B (LUB, supremum) meet︰ A B (GLB, infimum) This lattice possesses all the properties (Table 1.1) of the Boolean lattice (or Boolean algebra) except the laws of contradiction ( A A ) and exclusive middle ( A A X ) ․Verify A A (law of contradiction) is violated for fuzzy sets, i.e., Show x min{ A( x),1 A( x)} 0 e.g., A( x) 0.3 1 A( x) 0.7 min{0.3, 0.7} 0.3 0 1-23 ․Verify A ( A B) A (law of absorption) i.e., Show x max{ A( x),min{ A( x), B( x)}} A( x) x i, if A(x) B(x) , min[A(x) , B(x)] = A(x) and max[A(x ), B (x )] = A(x ) ii, if A(x) > B(x) , min[A(x) , B(x)] = B(x) and max[A(x ) , B (x )] = A(x ) ◎ Fuzzy set inclusion (subset) A B iff x , A( x) B( x) A B A, A B B 1-24 ○ Description of fuzzy sets with finite supports i, Finite universersal set X (discrete case) A or A a a1 a2 .... n x1 x2 xn ai , ai A( xi ) xi Supp ( X ) xi ii, X is an interval of real numbers (continuous case) A X A( x) x ◎Scalar cardinality (or sigma count) | A | | A | A( x) xX 1-25 ◎ Fuzzy cardinality ․Fuzzy number: convex, normalized fuzzy set ․Fuzzy cardinality | A | --- a fuzzy number define on N whose membership function is A | A | (| A |) or | A | A | A | | A | ︰ the degree to which fuzzy set A contain the number of members , | A | , is 1-26 ○ Example X︰crisp universal set X = {5 , 10 , 20 , 30 , 40 , 50 , 60 , 70 , 80} Fuzzy sets labeled as “infant” , “adult” , “young” , “old” 1-27 Consider Fuzzy set labeled “old” Scalar cardinaliity: |old| 0 0 0.1 0.2 0.4 0.6 0.8 1 1 4.1 Fuzzy cardinality: {0,0.1,0.2,0.4,0.6,0.8,1} old when = 0.1, old {20,30, 40,50, 60, 70,80} 0.1 | 0.1 old| 7 = 0.2, 0.2 old {30, 40,50, 60, 70,80} | 0.2 old| 6 = 0.4, 0.4 old {40,50, 60, 70,80} | 0.4 old| 5 = 0.6, 0.6 old {,50, 60, 70,80} | 0.6 old| 4 = 0.8, 0.8 old {60, 70,80} | 0.8 old| 3 = 1, old {70,80} 1 | 1 old| 2 | old | 0.1 0.2 0.4 0.6 0.8 1 7 6 5 4 3 2 1-28 ◎ Degree of subsethood , S(A,B) , of A in B S ( A, B) | A B | | A| 1 (| A | max{0, A( x) B ( x)}) | A| x X 1 = (| B | max{0, B ( x) A( x )}) | A| x X 1 = ( min{ A( x), B ( x )}) | A | xX S ( A, B ) 1-29 ◎ Distances between fuzzy sets X︰universal set containing n elements A, B︰fuzzy sets defined on X an a1 a1 A x1 x2 xn bn b1 b2 B x1 x2 xn 0 ai , bi 1 From A PA (a1 , a2 , , an ) From B PB (b1 , b2 , , bn ) In an n-D space , d ( A, B) | A( x) B( x) | xX d ( A, B) d ( B, A) The n-cube represents the fuzzy power set (X) The vertices represents the crisp power set P(X) 1-30 ※ Scalar cardinality |A| = d(A,Φ): Probability distributions are represented by sets whose cardinality is 1 ( P 1) i the set of all probability distributions is represented by a (n-1)-D simplex of the n-cube ( P 1) i ◎ Assignment︰ 1.8 , 1.10 , 1.11 (1996), 1.4 , 1.6 , 1.13 (1997) 1.12 , 1.14 1.7 , 1.9 , 1.15 (2000) (1998),