196 THE SITUATION CATEGORIZATION FOR SUPPORT DECISION MAKING V.L. Tokarev1 1Tula state university, E-mail: [email protected] The problem of computer aided support of the situations recognition for a conclusion making is discussed . For decision of this problem is offered the method using building of the linguistical model. Introduction The problem statement One of the first problems of computer support decision making is a situations recognition, in which is found system, comparatively which is taking the decision. It is expected that system S, functioning in surrounding ambience E, has: 1) the uncontrolled but measuring (or estimating) entry x; 2) the immeasurable and indefinite entry w; 3) the controlled (chosen) entry u, depended from decision d (or directly being realization of the decision d) of the person coming to a conclusion (PCC); outputs y depending from system status (the Fig. 1). ЛПР The interaction LPR, systems S and ambiences E is shown on the Fig.1, 2. The collection of the conditions of the system S and ambiences E at the time of k decision making dD form the situation x, y, z, u, k . Here k - discrete time, D - a discrete ensemble of the possible decisions. Метасистема x y z u Информационноизмерительная система u ЛПР Система S x w y Среда E z Fig.1. The surrounding ambience interacts with system, giving on it importance of variable x, w, depended from ambience status , partly hung from y. The surrounding ambiences usually are considered as "a nature", when actions of the ambience do not carry the goal-directed nature, or counteracting side ("opponent"), when actions of the ambience have a purpose to beat PPC. Fig.2 We expect that ensemble situation ={} is divide into classes ai, i=1,…,n, (some classes can consist of one situation). Each class situation i under any purposes gG corresponds to the nonblank ensemble of the decisions ={d}D. The number of the classes n known moreover classes all ranked on separate sign from normal situation (source) before threatening. The Relationship of the observations y with condition also approximate (within -differentially of the decisions dD) also known: y H v . Here y, v - a vectors to one dimensionality. So 197 class situation possible to match, identical him classes aiA output yY. That is to say shall suppose that if it is chose variant of the partition j, j= 1, ..., n, that ensemble A subdivides on n classes, that is. ai j x, u, z ; i 1,...,n . It is required on information on class i and information on operation of metasystem {x,y,z,u, k=1,…,N} for determined length of time (N counting out) split the ensemble of importance {x,u,z} on n classes that is build in space {x,u,z} categorization tree, allowing predict the attribute of the observations {x,u,z} to that or other class categorical hung variable y depending on corresponding to importance one or several variables. Rational categorization In report is offered method of decision of this problem founded on building models tree forming rough model of metasystem bi=F(x,u,z). The tree contains three levels of the linguistically models - a models, handling importance linguistically variable. On lower level are found models of the ambience, chosen by importance L(z), on rally-blowing level - a models, chosen value L(x); on the one third level - a models, chosen by importance L(u), but their output is an instruction of a certain class aiA. The Rational categorization ensemble А on n classes is identified n- measured vectorfunction (у) =(1(у) ,..., n(y)), (i(y) - a function accessories to class ai), satisfying conditions : first, (y) - measurable on measure result functions accessories model output y beside and, in second, for any yY importance i(y) satisfies standardization condition n i y 1, 0 i y 1. i 1 As criterion quality to categorizations is functional n n J b j , ai b j yi i 1 j 1 (1) where b j , ai - certain measure to vicinity between element ensemble B (the thermo linguistically variable L(y)) and element ensemble classes А; - monotonous increasing function , displaying length [0,1] on itself moreover (0) =0 and (1) =1. Linguistically models, forming tree, are built by iteration procedure [1] before achievement of the minimum criterion J ò (e M ) ln( st ( )) N kB - (c M /N) ln ( k ji k ji /k B ) (2) i 1 j 1 where st() - a factor to stability estimation criterion, calculated on sample data; k ji ρ(b ji ) - a distance in given metrics ( ) subset bj in iline of the matrix of the observations WN from begin area of importance output variable y; k B - a number subset bjВ; cM - a parameter of regular, intended for increasing her(its) regular, calculated on one of the expressions C M 1 n x /N x or C M 1 - n S /N S , where nx, Nx accordingly, number taken into account by model factor and total number possible factor; nS, NS - accordingly, number elementary structured element, comprised of model, and the total number source structured element, determined from a priori information. The Lemma. Distribution of importance function accessories b y , 1 1,...,n , got by i means of models bi=F(z,x,u), carries унимодальный nature. From lemma follows that maximum b y , 1 1,...,n under some condition can i point to attribute of the situations class ai. The Theorem. If linguistically model bi=F(z,x,u) answers the minimum a criterion (2), that such model provides the rational categorization a situation in the sense of criterion (1). As a result adjusted thereby linguistically model becomes the qualifier ai=f(z,x,u), herewith in dug the degree to validity to current categorization emerges resulting importance of the function accessories b y , 1 1,...,n . i 198 Conclusion It has been Shown that computer support of the decision of the problem of situation categorizations is possible to realize by buildings of linguistical model describing metasystem, when ambience, surrounding system is "a nature" that is to say actions of the ambience casual and inexpediently. The Offered method allows to automate the initial stage of computer support of decision making. References 1. Tokarev V.L. Bases of the decisions rationality provision theory. Tula: Tula state university. – 120 p. (in Russian).
© Copyright 2026 Paperzz