New Methodology for Structure Identification of Fuzzy Controllers in Real Time

This paper presents an innovative approach to self-adaptation of the structure of a neuro-fuzzy controller in real time. Without any off-line pretraining, the algorithm achieves very high control performance through the iteration of a three-stage algorithm. In the first stage, coarse tuning of the neuro-fuzzy rules (both rule consequents and membership functions of the premises) is accomplished using the sign of the dependency of the plant output with respect to the control signal and an overall analysis of the main operating regions. In stage two, fine tuning of the rules is achieved based on the controller output error using a gradient-based method. Finally, the third stage is responsible of modifying the structure of the controller, proposing that input variable which should get a new membership function in order to improve the control policy in an optimum way.