A self-learning neuro fuzzy controller

This paper describes a neuro-fuzzy controller that can mimic the way a human controller might function. The controller comprises an artificial neural network (ANN), a knowledge base and a fuzzy inference engine (FIE). Initially the controller learns the system dynamics, which it stores in its knowledge base. The FIE trains the ANN with the observed data, then the ANN controls the system during operation the ANN´s performance is continually evaluated by the FIE and retraining is performed as necessary. A feature of the approach is that the controller learns to control the system without any a priori mathematical model of its dynamics. The initial application of the approach to a hypothetical temperature controller is described