Compressing and improving fuzzy rules using genetic algorithm and its application to fault detection

Abstract

The fuzzy rule sets, which have been derived from the hybrid neural network model, called the O-EGART-PR-FIS, is an integration of the Adaptive Resonance Theory (ART) into Generalized Regression Neural Network (GRNN), display substantial redundancy and low interpretability that leads to time-consuming prediction process. The O-EGART-PR-FIS approach can achieve the highest accuracy rate among all, however the extracted rules are less compact. Hence, in this paper, we propose a genetic algorithm based method with the inclusion of the “Don't Care” antecedent (hereafter denoted as DC-GA) to the foundation of the O-EGART-PR-FIS, with the aim of further optimizing the existing fuzzy rules. The improved model is applied to two benchmark problems, and the rules extracted are analyzed, discussed and compared with other published methods. From the comparison results, it is observed that the improved model is attested to be statistically superior to other ANN models. Therefore, it reveals the efficacy of DC-GA in eliciting a set of compact and yet easily comprehensible rules while sustaining a high classification performance.