抄録
This paper discusses a self-tuning method of fuzzy controllers. The self-tuning procedure consists of two parts. One is a real time tuning by δ-rule which is a basic learning law in neural networks. The δ-rule of neural network is used for minimizing integral of squared error between a reference signal and output of controlled object. The other is an iterative adjustment based on evaluation of transient characteristics using compensation rules. The compensation rules are derived from the viewpoint of effectively improving frequency characteristics, that is, gain and phase characteristics, of control system. We point out from simulation results for several controlled objects that two self-tuning mechanisms are useful and that the self-tuning fuzzy controller constructed by combining these self-tuning mechanisms is more effective.
