Global weight refinement of fuzzy production rules using neural network

Abstract

In order to capture and represent more useful and important information via a multi-level fuzzy production rule (FPR) system, the concept of global weight is introduced to represent the degree of importance of each rule in an inference path leading to a final goal. However, the proper and accurate assignment of a global weight to each rule in an inference path by a domain expert is a difficult task. Usually the initial global weight values acquired during the knowledge acquisition phase are rough estimates, and they require fine-tuning or refinement later on. It is a time-consuming and repetitive task to refine the global weights manually. In this paper we propose using a neural network approach to address this knowledge refinement problem. It is shown that each type of FPR with its knowledge representation parameters can be mapped into a neural network to create a fuzzy expert network, which provides the full representation of the knowledge base of the FPR system and its inferencing capability as well. A new learning algorithm is developed to enable the network to adapt to the new architecture and to support the inferencing function. Our approach has the advantages that the refinement of global weights could be done automatically, and that refined global weights would allow the fuzzy expert system to achieve better performance in terms of increasing the degree of certainty of the final goal being drawn.