Fuzzy identification and controller based on generalized fuzzy radial basis function networks

This paper first proposes a new kind of fuzzy neural networks-generalized fuzzy radial basis function networks (f-RBF), which combines the fuzzifying and defuzzifying processes into a united network structure. We then give the dynamic training rule and training strategy for the f-RBF. We further discuss several special features of this kind of networks that conventional neural networks do not have, and conclude that it can process both the fuzzy-valued and real-valued data simultaneously, and can achieve the minimum realization of fuzzy controller for nonlinear systems. Finally, using the f-RBF, we design a fuzzy controller for a nonlinear system regulation. Furthermore, we point out that any nonlinear control u can be decomposed into three parts: a fuzzy control u_f, a linear control u_l, and an error compensation u_e, i.e., u=u_f+u_l +u_e. The stability of the closed-loop system is also analyzed using sliding control techniques