Fuzzy-neural model for nonlinear systems identification

A state space representation of both continuous and discrete time mathematical models of Recurrent Neural Networks (RNN) are given in two layer Jordan canonical architecture, and a new improved Back-Propagation (BP) type learning method, is proposed. Some topology improvements, are suggested. The proposed RTNN model is linear in small and nonlinear in large, which permits to apply all well known state- and output linear systems design methods. The obtained RTNN model is incorporated in a rule-based fuzzy system, giving the possibility to approximate and to identify a complex nonlinear plants. Simulation results of nonlinear systems identification by the proposed fuzzy-neural system, using RNN BP learning, are given. Copyright (C) 1998 IFAC.