Integration of Multi AI Paradigms for Intelligent Operation Support Systems – Fuzzy Rule Extraction from a Neural Network

Abstract

Some artificial intelligence (AI) paradigms have been applied to water and wastewater treatment systems. An artificial neural network (ANN), which can learn historical data of a plant, provides operational guidance for plant operators, and a fuzzy system (FS) provides a framework to put operators' heuristics into practical use as fuzzy rules in a fuzzy rulebase. In application, however, the practical problems remain that the ANN is a blackbox model which is unfamiliar to plant operators, and the FS usually requires much time-consuming work by system engineers and operators for knowledge acquisition and rulebase maintenance. The authors think that integration of the paradigms can give appropriate solutions to these problems. As one method which realizes such integration, an automatic fuzzy rule extraction method using an ANN is proposed. Simulation results of the proposed method using full-scale plant data demonstrated that an FS whose rulebase was modified automatically with extracted rules had better performance than a conventional FS whose rulebase included only operators' heuristics. This effect is thought to be realized by enhancement of knowledge source with the proposed method.