1. Fuzzy sets, fuzzy relational calculus, linguistic approximation

Size: px

Start display at page:

Download "1. Fuzzy sets, fuzzy relational calculus, linguistic approximation"

Audra Robbins
6 years ago
Views:

1 1. Fuzzy sets, fuzzy relational calculus, linguistic approximation 1.1. Fuzzy sets Let us consider a classical set U (Universum) and a real function : U --- L. As a fuzzy set A we understand a set of pairs (x, A (x)) : A = (x, A (x)) A (x) L, x U. (1.4) Function A is called a membership function and is L is usually L = 0,1. Note: If L s degenerated to its border points 0,1 then A is a characteristic function and fuzzy set A is an ordinal set of ordered pairs. Then we speak about a "crisp" set and about a "crisp" value of a membership function. Example 1.4: The expression of fuzzy set by enumeration of its elements (extensional), i.e., by ordered pairs (x, A (x)), x U: A = 0.5/x 1, 0.8/x 2, 0.0/x 3,, 1.0/x k, 0.7/x n, The introduced form of expression for A was often used in early works of prof. Lofti Zadeh. However it is only expressioning convention (x, A (x)) = ( A (x)/x). The advantage of extensional expression is this: it is explicitly given which members belongs to fuzzy set and with which quantity of membership function (e.g., x 2 belongs to U with 80 % certainty; on the contrary x 3 does not belong to U). Height of fuzzy set A - hgt(a): hgt(a) = sup A (x 1 ),, A (x n ), (2.4) x 1,..., x n U Normal fuzzy set: U is ordered set of real numbers, hgt(a) = 1.

2 Normal and a convex fuzzy number: x 1,x 2 U, 0,1, A ( x 1 + (1- ) x 2 ) min A (x 1 ), A (x 2 ), (3.4) Ex: In Fig is a normal convex fuzzy number approximately 9. A as an - cut of fuzzy set A: For 0,1 and ordered subset of U: A = (x, A (x)) A (x), x U, (4.4) E.g. A 0.85 = (8.0, 0.85), (8.5, 0.9), (9.0, 1.0), (9.5, 0.9),... - Fig.1.4. Fig.1.4. Note: Graphs of membership functions may have very strange shapes (bell-functions, sigmoid functions, ). However in case when they represent some language terms they are not absolutely optional. E.g., in case when they represent intensities of some properties that are measured on universum U of real numbers the most of such functions has non monotonic form. (E.g., when we say that the consumption of gasoline of our car is standard then we mean that for interval (0, 3 ) [l/100 km] the value of membership function is zero, then is gradually growing till 5 [l/100 km], where it has value 1, it continues constant till value 6.5 [l/100 km] and then is decreasing till zero in 10 [l/100 km]. Second point of view is the simplicity of the membership function shape. The shapes could be acceptable for manipulations and they would be derived from some not too large library. 91

3 In addition of known bell and sigmoid functions is often used quadruplet notation (parameterization) a,b,, - see. Fig , for x a - -1 (x - a + ), for x a-,a A (x) = 1, for x a,b (5.4) -1 (b + - x), for x b,b+ 0, for x b+. A (x) 1 a- a b b + x U Fig Basic operation with fuzzy sets Let us consider fuzzy sets A, B, C defined gradually on universes U 1, U 2, U 1 U 2 and let us assume L = 0,1. Binary operations union ( ), intersection ( ), complement (Compl) and set subtraction (-) are described by the following relations: C = A B = (x, A B (x)) x (U 1 U 2 ), A B (x) = A (x) (x, A B (x)) x (U 2 U 1 ), A B (x) = B (x) (6.4) (x, A B (x)) x (U 1 U 2 ), A B (x) = max A (x), B (x). C = A B = (x, A B (x)) x (U 1 U 2 ), A B (x) = min A (x), B (x). (7.4) C = Compl A = (x, Compl A (x)) Compl A (x) = 1 - A (x), x U 1. (8.4) 92

4 C = A B = (x, A-B (x)) x U 1, A-B (x) = min A (x),1- B (x). (9.4) Example 2.4: Let us consider the following fuzzy sets A, B and let us construct complement (Compl A), union C = A B and the difference C = A B. A = 0.5/x 1, 0.8/x 2, 0.0/x 3, 0.5/x 4, 0.6/x 5, 1.0/x 6, 0.7/x 7, B = 0.6/y 1, 0.7/y 2, 0.3/x 3, 0.8/x 4, 0.7/x 5, 0.5/y 6, 0.0/y 7, C = A B = 0.5/x 1, 0.8/x 2, 0.2/x 4, 0.3/x 5, 1.0/x 6, 0.7/x 7. x 1 o x 2 o x 6 o x 7 o A x 5 o x 4 o o x 3 o y 6 B o y 2 o y 1 y 7 o Fig. 3.4 Theory of fuzzy sets allows introducing completely new algebraic operations by means so called induction principle: Let us consider fuzzy sets A 1,..., A n defined on universes U 1,...,U n by means of membership functions A1,..., An and n-ary function f, f : U 1 x... x U n Fuzzy set B on V (resp. Its membership function B ) is then induced following way: --- V. by the x 1,...,x n U 1 x... x U n, y V, f(x 1,...,x n ) = y Sup min A1 (x 1 ),..., An (x n ), pro f -1 (V) x 1,..., x n U 1 x... x U n B (y) = (10.4) 0, pro f -1 (V) =. Example 3.4: Let us consider fuzzy numbers A1 = 2, 2, 1, 1 a A2 = 4, 4, 1.5, 1 defined on universum R 1 and function f: R 1 R 1 R 1 defined as an ordinary function 93

5 of addition, i.e., f(x 1, x 2 ) = x 1 + x 2. The application of expression (10.4) is for the main quantities introduced in the following table: Row X 1 x 2 y A1 (x 1 ) A2 (x 2 ) B (y) The expressions of application of (10.4) introduced in the table is possible to compare with results of arithmetic operation + (from the section 1.2.1). By the Induction principle is very easy to define the set of algebraic operations f F on the set of fuzzy sets or for the definition of so called approximation of fuzzy sets Let us consider fuzzy sets A, A 1,..., A n A, the operation f F appropriate metric function d: 0,1 x 0, ,1. As an approximation of the set B, that is formed as a result of function f, B = f(a 1,..., A n ), is understood a fuzzy set A, for which holds: and some d( B (x), A (x)) = min d( B (x), C (x)). (11.4) C A By application of induction principle on concept of multiplication of fuzzy relations is possible to construct new fuzzy relations: Let us consider classical Relations R, S (for the simplicity let us consider Binary relations) defined on Cartesian products of universes U 1, U 2, U 3, i.e., R U 1 x U 2, S U 2 x U 3. Fuzzy relations A R, A S are defined as fuzzy sets on sets R, S. The product of fuzzy relations A R, A S is defined by means of induction principle: RoS (x 1, x 3 ) = Sup min AR (x 1, x 2 ), AS (x 2, x 3 ), (12.4) (x 1,x 2,x 3 ) U 1 x U 2 x U 3 94

6 where RoS is the product of relations R, S. By means of (12.4) it is very simply to define some other useful properties, as it is, e.g., transitivity of fuzzy ordering (fuzzy transitivity): R S, U x U, R o R = R, A R = A S, (x 1,x 2,x 3 ) U x U x U o (x 1, x 3 ) = max min (x 1, x 2 ), (x 2, x 3 ), (13.4) (x 1, x 2, x 3 ) U x U x U Example 4.4.: Let us consider universum U = a, b, c, d on which is defined fuzzy relation ordering by the following way: x, y U, (x,y) = 0.1/(a b), 0.5/(a d), 0.4/(b c), 0.9/(c d). (14.4) Membership function (x,y) is expressed by individual pairs where the number above the symbol "/" gives the certainty of the statement that is given below "/". The computation of membership function value for the remaining two pairs on U - ((a,c), (b,d)) uses the transitivity of relation and its description by (14.4). Applying (8.4) x,y U, (y,x) = 1 - (x,y), (15.4) we obtain values (a,c) a (b,d). (a,c) = max min (a,b), (b,c), min (a,d), (d,c) = = max min 0.1, 0.4, min 0.5, 0.1 = max 0.1, 0.1 = 0.1. (b,d) = max min (b,c), (c,d), min (b,a), (a,d = = max min 0.4, 0.9, min 0.5, 0.5 = max 0.4, 0.5 =

7 1.2.1 A basic arithmetics of parametrically defined fuzzy numbers Let us consider two fuzzy numbers Q(x) = a,b,, = m, Q(y) = c,d,, = n. The constructions of arithmetic operations f are given in Table 1.4. The symbol 0 is the cut by a partial ordering for = 0. Tab.1.4. Operation f Computation formula Conditions n ( d, c,, ) n,, ( d ), ( c ) c n 0 0, n 0 d c d 0 m n ( a c, b d,, ) m n ( a d, b c,, ) ( ac, bd, a c, b d ) m, n 0 m n , n 0 0 0, n 0 0 0, n 0 ( ad, bc, d a, b c ) m 0 ( bc, ad, b c, d a ) m 0 ( bd, ac, b d, a c ) m Linguistic variable and the space of fuzzy values Though we have studied the above concepts (and operations with them) in general levels (similarly as in mathematics, where we are not interested in units of contents and variables) in the development of the space of fuzzy values in real cases is always considered a certain system or real environment. In other words spaces of fuzzy values are semantically bonded. This fact is included in the concept of linguistic variable that contains with fuzzy space also the name of the variable and the structuralisation of participating universes. As a linguistic variable LV is understood a triplet LV = J, A, U, (16.4) where J is the name of the variable (e.g., the temperature in Prague in June"), A is the set fuzzy linguistic values of the variable J (e.g., under long time standard (PDN), low (Nízká), middle (Střední),"above long time standard (NDN)",...) and U is the structure of universes, on which are the linguistic variable and its fuzzy 96

8 values defined. (In our case - "the temperature in Prague in June is a universum of temperature scale (however it could be in general Cartesian product of some sub universes U = U 1 x... x U n.)) In Fig is illustrated linguistic variable "T 6 temperature in Prague in June". Fig In the context of application of fuzzy sets for modeling of systems is very natural to assume that exists a certain intuitive image about a term system variable (temperature, density, aggresivity, ). By means of these variables is possible to describe the system and its behavior qualitatively, e.g., on the base of observation (without an exact numerical model) in terms of qualitative values and their qualitative derivations (temperature is high, the density is low, the temperature is increasing, the aggresivity is decreasing, ). Fuzzy qualitative value may be then understood as a fuzzy number with certain semantic content. The set of all qualitative values related to all considered variables of the modeled system is called as a space of fuzzy values - Q F. Mainly for technical purposes was done a certain unification of fuzzy spaces. One of examples of fuzzy spaces is in Fig (for case U = -1,1 ). Q F = - Velká (Large), - Střední (Middle), - Malá (Small), Nula (Zero), + Malá (Small), + Střední (Middle), + Velká (Large) Fig

9 1.4. Linguistic approximation With help of operations from Table 1.4. we may compute values of fuzzy variables in space Q F. Results are new fuzzy values B (fuzzy numbers), that might not be none value from Q F. However no inter-values in Q F we have. For the linguistic determination of computed number we need special operation, so called linguistic approximation. This operation select from the values in Q F the nearest (A) value to B. (Speaking about distance we need some metric function d(b, A).) The method of linguistic approximation will be now explained by example for computed fuzzy number Q(z) = c,d,, = B, its approximation A = a,b,, and metric function (17.4). d(b, A) = 1/2 2(a-b-c+d) /2 c + d - a - b 2 1/2, (17.4) Example 5.4.: Let us consider the space Q F given in Fig.5.4 with the following fuzzy numbers: Nula (Zero) = 0, 0, 0, 0, Malá (Small) = 0, 0.2, 0, 0.2, Střední (Middle) = 0.4, 0.7, 0.1, 0.2, Velká (Large) = 0.9, 1, 0.1, 0, Determine semantic content of fuzzy number B = Small + Large. a) Arithmetic operation + from Table 1.4: B = p, q, r, s + j, l, m, n = p + j, q + l, r + m, s + n B = 0, 0.2, 0, , 1, 0.1, 0 = 0.9, 1.2, 0.1, 0.2 = B. b) Linguistic approximation using (17.4): We have to compute: d(b, Zero) =, d(b, Small) =, d(b, Middle) = 0.5, d(b, Large) =

10 Result: Small + Large Large. Check examples: 1) For fuzzy value "NDN" of varibale Temperature in Prague in June construct the cut for = 0.75 by means of enumeration of some quantities. 2) Find at least two cuts (in Fig. 4.4) for which holds value (A1) = value (A2). 3) Let us consider fuzzy sets A, B on universum U = x 1,...,x 4 defined by enumeration of some elements: A = 0.32/x 1, 0.45/x 2, 0.63/x 3, 0.11/x 4, B = 0.75/x 1, 0.12/x 2, 0.27/x 3, 0.81/x 4. Construct fuzzy set C = Compl (A B)! 4) Let us consider universum U = a,b,c,d, on which is defined fuzzy relation equivalence " " by the following way: x,y U, (x,y) = 0.1/(a b), 0.5/(a d), 0.4/(b c), 0.9/(c d). (Relation equivalence has the property «transitivity» as relation ordering and it is possible to use expressions (12.4), (13.4) and method for computation of relation of order for pairs (a,c), (b,d). Equivalence is symmetric and instead of (15.4) is necessary to use x,y U, (x, y) = (y, x). 5) Let us consider the space Q F from Figure obr Compute linguistic approximation B = Small Middle. 99

11 The list of recommended literature: [1] Zadeh, L., A: Fuzzy Sets. Information and Control. No.8., 1965, str [2] Leitch, R. and Shen, Q.: Fuzzy Qualitative Simulation. IEEE Trans. on Syst., Man and Cybernet., No.4., str [3] Sugeno, M. and Yasukawa, T.: A Fuzzy-Logic-Based Approach to Qualitative Modelling. Trans. on Fuzzy Systems, Vol. 1., No. 1., str [4] Bíla, J., Šmíd, J., Král, F. a Hlaváč, V.: Informační technologie. Databázové a znalostní systémy. ČVUT v Praze,

CHAPTER 5 FUZZY LOGIC CONTROL

64 CHAPTER 5 FUZZY LOGIC CONTROL 5.1 Introduction Fuzzy logic is a soft computing tool for embedding structured human knowledge into workable algorithms. The idea of fuzzy logic was introduced by Dr. Lofti