Back-propagation neural-network based fuzzy controller with a self-learning teacher

Abstract

By considering a previous study (Nie and Linkens 1992) as a first step towards integrating a rule-based fuzzy controller with neural networks from a viewpoint of functional equivalence, this paper continues the process by making a crucial assumption that neither control experts nor teacher signals are available for the multivariable control problem. In response to this challenge, we present a novel and systematic approach capable of learning and extracting required control rules automatically from the controlled environment for use by back-propagation neural networks (BNN)-based fuzzy controllers. Three possible controller structures are suggested with some comparative studies. Some pertinent points relating the present method to other traditional ones are discussed. Simulation results of blood pressure control demonstrate the utility and feasibility of the proposed approach in solving relatively complex control problems, in particular, those problems where neither control experts nor mathematical models of the controlled process are available.