ACTUALIZATION OF THE DISTRIBUTED KNOWLEDGE BASE OF ERGATIC SYSTEM USING THE METHOD OF FUZZY CLASSIFICATION

In the article a method of actualization the distributed knowledge base of ergatic system using the method of fuzzy classification is proposed. As an example we consider the request choice formation of an alternative of decision-making from the knowledge base, according to the values of the input parameters. Genetic algorithm is used for finding optimal solutions. For automation of calculations MATLAB software package was used.


Introduction
One of the most important components of ergatic decision support systems, at the management of complex technical objects, is a Knowledge Base (KB), which is realized as a special kind of database, developed for operating knowledge. KBs should contain structured information covering some area of knowledge for using with a specific purpose. Modern KBs contain not only factual information, but also the rules of inference, reasoning about newly inputted facts, meaningful information processing but also work together with the information retrieval systems and have classification structure and format of knowledge representation.
The most important and time-consuming affair in creating knowledge base is to support its relevance. Considered in [2][3][4][5]8] the ways of supporting the relevance of KB in our opinion are quite complex, involving technical knowledge, moreover, presupposes the existence of large arrays of address databases received from various sources. In [1,7] there are considered the methods of examination and diagnosis in order to maintain the relevance of KB.
All considered expert methods are widely used and is described in detail in the modern literature.
The disadvantages of expert methods are subjectivism, the limited application, the high costs of their conduct, so these methods are appropriate to use at the initial stage of filling KB. This paper proposes a method for updating knowledge base to build alternatives to decision-making in real time, taking into account the cognitive state of users based on fuzzy classification method.

Methods
In a distributed KB there is stored information in the form of alternatives, which maintains records of direct and indirect data influencing the process of making relevant decisions. Full information structure of formation of the alternatives presented in figure 1, where:  X1 -information on the characteristics of the external environment, which directly affects DM (noise, temperature, low-frequency vibration, light level, etc).  X2 -information about psychological characteristics of DM (test results, physiological state of DM, pulse, pressure, etc.);  X3 -information about technological process, as well the range of deviations.  Table 1.
It is required to select an alternative for any set of parameter values x 1 , x 2 , x 3 , wherein the values of some or all of the parameters do not belong the intervals, specified in Table 1.
To solve this problem we use the method of fuzzy classification [8]. Classification problem consists in assigning an object, specified by the vector of informative signs V = (x 1 , x 2 , ..., x n ), to one of advance certain classes {A 1 , A 2 , ..., A m }, that is, consists in performing of mapping the form: Classification based on fuzzy inference is made on the knowledge base in the form: where u j {A 1 ,A 2 ,…,A m }consequent value of j-th rule; ij a -a fuzzy term, valuating the criteria ( ̅̅̅̅̅ ) in the j-th rule.
The degree of membership of classification object, informative characteristics is given by a vector ( ), classes A j from knowledge base, is calculated as: where ( ) -is a degree of membership values of fuzzy therms ̃ ; Λ -operation of finding the minimum.
As a solution, there is selected class with a maximum degree of membership: 1 2 { , ,..., } . 12 * max( ( *), ( *),..., ( *)) arg In the problem under consideration, vector of informative features of object classification is V=( x 1 , x 2 , x 3 ), and alternatives A 1 ,A 2 ,…,A 27 -classes of solutions. The fuzzy knowledge base of mapping V→ U {A 1 ,A 2 ,…,A 27 } we build, taking into account dependency, presented in Table 2, where by L, A, H are denoted the terms "low", "average", "high". Membership functions of terms of the input variable x 1 are shown in Figure 2. The membership functions of the terms of input variables x 2 , x 3 , we will choose similar to the corresponding functions of the variable x 1 . This will reduce the total number of parameters of membership functions of the terms of all the variables from 24 to 8, which is significantly while minimizing function (7).
As a criterion of training fuzzy classifier [8] let us choose the simplest criterion: where: ) an error of classification of the object V k ; N -the number of pairs of input-output (V k , u k ), ̅̅̅̅̅ training sample; P is the vector of the parameters of the membership function of the fuzzy terms of knowledge base (1); F(P, V k ) is the result of the classification on the fuzzy basis with parameters P if the input value is V k . Training fuzzy classifier, therefore, is to find the vector P that minimizes the distance between the results of the logical inference and experimental data from the sample (V k , u k ).
Choose from the k-th row of table 1 random values x 1 , x 2 , x 3 , belonging to the respective intervals of changing the values of variables in the considered string. Get input vector V k . According to Table 1, it belongs to the class u k =A k . Considering all rows we will receive 27 pairs of "input-output" (V k , u k ), ̅̅̅̅̅̅ training sample.
Using (2), (3), produce a classification based on fuzzy input data V k . The calculation is made in MATLAB environment; assume the weights w j equal to zero. While classification result F (P, V k ) on the fuzzy basis with the parameters P 0 when the input value V k differs from a clear exit u k 15% of the trials.
To increase the accuracy of the fuzzy classification we will use the criterion of (4).
As the minimized function is an integer, the most appropriate is the genetic algorithm for finding extreme. For its realization we use "gatool" MatLAB function.
Below are the results of the fuzzy selection of alternatives for the vector of parameters of membership functions (5) and some input values x 1 , x 2 and x 3 :

A
In the case, for example, when V = (22, 5; 7, 410) selection of alternatives is ambiguous: either A 14 or A 17 . As a solution we should accept the new formed alternative that after assessing the relevance will be recorded into a knowledge base: . 17 14

Results
The proposed method allows increasing the efficiency of the process of actualization the distributed knowledge base through automation fuzzy classification components of alternatives in multilevel ergatic decision support systems in the management of complex technical objects in the real time.