Multiple fuzzy systems for function approximation

Standard procedure for building a fuzzy model often involves trying several candidates with varying number of fuzzy rules and training parameters in order to achieve acceptable model accuracy. Typically, one of the candidates is chosen as best, while the rest are discarded. When the system being considered is highly nonlinear or includes a number of input variables, the number of fuzzy rules constituting the underlying model is usually large. This paper proposes an alternative approach to designing fuzzy systems. The essential scheme is to decompose the overall system into subsystems and then combine their individual outputs. This offers advantages of speed, reliability, and simplicity of design. The concept of competition developed in modular networks theory is used to derive identification algorithm. The utility of the proposed approach is illustrated by a nonlinear function approximation example.