Dynamic Compensatory Pattern Matching in a Fuzzy Rule-Based Control System

In the recent years, fuzzy rule-based control systems have been receiving more and more attention because of their ability to capture the approximate nature of the real world. Moreover, the use of linguistic-valued rules largely reduced the size of rule bases and decreased the number of loopholes owing to human experts´ episodic heuristics. In brief, fuzzy rule-based approach provides us a way to achieve comprehensibility, simplicity and robustness at the same time. Two common problem in fuzzy rule-based control systems are addressed in this paper. The first one regards to incomplete information which results from a noisy environment or a real-time constraint. The other one concerns about earning because an intelligent system should be able to adjust itself when more information is provided. The power of traditional matching techniques is limited when faced with these problems. We propose a dynamic pattern matching mechanism which has a sound semantic interpretation and is computationally efficient to provide a better solution. We first briefly review the basic matching techniques involved in rule-based control systems. They are divided into three phases: matching of pattern elements, aggregation of partial matching results, and interpolative matching. Within this framework, we identify the issues of incomplete information and learning. A learning mechanism based on generalized neural networks is introduced next. Lastly, we describe the proposed compensatory matching mechanism.