SISO nonlinear system identification using a fuzzy-neural hybrid system

This paper describes a fuzzy-neural hybrid system for the identification of nonlinear dynamic systems with unknown parameters. The proposed model takes the form of a context-sensitive module in which a fuzzy system is used as a function module and a multilayer neural network is used as a context module. Fuzzy-neural hybrid systems with a decomposed structure reduce complexity and thus accelerate the learning process. Also, the parameters of a fuzzy system have clear physical meanings, which makes it possible to incorporate a priori knowledge into the selection of initial parameter values and constraints among parameter values. Since hybrid systems correspond to networks, it is feasible to construct fast, parallel devices to implement these models for practical applications. The gradient descent method for the adjustment of parameters in hybrid systems is discussed. Simulations demonstrate that the hybrid identification models suggested here for SISO dynamic systems are quite effective.