In this paper, we present an innovative approach to the identification of non-linear systems. The proposed neuro-fuzzy system identifier employs a hybrid clustering and least mean squared error (LMS) algorithm. The neuro- fuzzy system under consideration is implemented as an two- layer FHRCNN (fuzzy hyperrectangular composite neural network). The SDDL (supervised decision-directed learning) algorithm is used to find a set of hyperrectangles defined by the parameters of hidden nodes while the LMS algorithm estimates the connection weights from hidden nodes to output nodes. Furthermore, based on the hybrid learning rule, the fuzzy neural networks can evolve automatically to acquire a set of fuzzy if-then rules for approximating the input/output functions of considered systems. A highly nonlinear system is used to test the proposed neural-fuzzy systems. The simulation results demonstrate its feasibility and robustness.
Proceedings of 1994 International Symposium on Artificial Neural Networks，頁495-500