Designing fuzzy models for nonlinear discrete-time systems with guaranteed performance

We propose an identification scheme for nonlinear discrete-time systems using fuzzy systems that takes the specific internal structure of the fuzzy system explicitly into consideration. We first design a fuzzy system and give the upper bound of its approximation error to a general nonlinear function. Then we propose a projection algorithm to turn the parameters in the fuzzy identification model and prove that the parameter error is nonincreasing and the identification error converges to an error band which can be made arbitrarily small by properly designing the fuzzy system. We also determine conditions under which the parameters converge