Fuzzy control system design by fuzzy clustering and self-organization

Proposes a two-step method for designing fuzzy rules when no plant model or control surface table is available. The first step learns heuristic fuzzy rules by performing online adaptive control via a trial-and-error method. One simple rule is to choose the control y(t) such that both the plant-state error x(t) and the change of error Δx(t) move toward zero at the same rate, up to some constant factor. The second step applies fuzzy clustering to the rule data generated by the first step to obtain more general and robust fuzzy control rules. Our experiments with an inverted pendulum problem show a good performance