New approach to adaptive control architecture based on fuzzy neural network and genetic algorithm

Abstract

A novel approach to an adaptive control architecture which formulates the adaptive behavior of a human mathematically is presented. It is made up of the two main modules: the Controller constructed by fuzzy neural network (FNN), and the Adapter composed of three components: the Performance Evaluator (PE), the Action Searcher (AS), and the Rule Constructer (RC). The Controller and Adaptor perform the offline and online learning to learn control strategy and to adapt variant environments. The PE evaluates the system's performance. If the control effect is satisfactory, the Controller keeps on its assignment. Otherwise, the genetic algorithm (GA) based AS will explore the new control actions. Then, the RC transforms these actions to the fuzzy rules and updates the corresponding fuzzy rules in Controller. An example of the path planning of a mobile robot is used to demonstrate the presented method.