Fuzzy neural modeling using stable learning algorithm

In general, fuzzy neural networks cannot match nonlinear systems exactly. Unmodeled dynamic can lead parameters drive and even instability problem. Some robust modifications must be contained, in order to guarantee Lyapunov stability. In this paper input-to-state stability is applied to access robust training algorithm of the fuzzy neural networks. We state that the normal gradient descent law with a time-varying learning rate is stable in the sense of L_∞. The fuzzy neural networks approximation, which is suggested in this paper, needs no robust modification and is robust to any bounded uncertainty.