A two-stage approach to self-learning direct fuzzy controllers (Q1603304)

The paper proposes, using a two-stage algorithm, an approach to self-tuning a direct fuzzy controller with very limited information on the system to be controlled. No off-line pre-training of the system and its parameters is assumed. During the first stage, a coarse tuning is achieved based on the system output error. In the second stage, the fine tuning of the fuzzy rules is accomplished by using the controller output error as the information source and a gradient-based method as the optimization tool. Several examples are presented to validate the proposed approach.