Evaluation of the Properties of Fuzzy Control Systems in the Stage of Formation of the Knowledge Base

Abstract

The paper is devoted to the study of the influence of the procedure of formation of the knowledge base on the characteristics of the fuzzy logic controller (FLC). The source of information in the construction of fuzzy controllers is expert knowledge. Constructing of membership functions of terms of the input and output linguistic variables and the fuzzy matching between the antecedent and consequent spaces is formalizes this knowledge. This entails loss of information, because there is no unique translation from a qualitative entity to a quantitative representation except for some special cases. Therefore, as a rule, it is necessary to correct the knowledge base and parameters of algorithm of fuzzy inference in order to achieve the required quality of the system. The main problem of the organization of the correction procedure lies in the complexity of purposeful changes of certain parameters of the algorithm, since the relationship between the settings of FLC and its dynamic properties are still not well studied. Thus, the task of complex research of FLC, allowing an analysis and synthesis system from the standpoint of the classical theory of automatic control, is relevant. The algorithm of the fuzzy inference consists of several stages such as formation of a knowledge base, fuzzification, aggregation, activation, accumulation and defuzzification. Creation of a knowledge base is perhaps the most critical step because it requires the involvement of experts and formalizing their knowledge to further computer processing. The formation of knowledge base in the algorithms of fuzzy inference based on relational models is discussed in the paper. We suggest considering three typical models of decision-making in compiling the rule base is "strong", "uncertain" and "balanced" - and estimate their influence on the control surface of FLC and the transient response of the fuzzy control system. We concluded that, from the point of view of influence on the dynamics of fuzzy control systems, specific analytical description of the membership functions is not essential; however, their degree of tension-compression has a significant effect on both the control surface, and system behavior. The conducted research allows goal-seeking tuning of FLC, providing required quality metrics of control system.