DUNEDIN NEW ZEALAND Hybrid Neuro-Fuzzy Inference Systems and their Application for On-line Adaptive Learning of Nonlinear Dynamical Systems

In this paper, an adaptive neuro-fuzzy system, called HyFIS, is proposed to build and optimise fuzzy models. The proposed model introduces the learning power of neural networks into the fuzzy logic systems and provides linguistic meaning to the connectionist architectures. Heuristic fuzzy logic rules and input-output fuzzy membership functions can be optimally tuned from training eramples by a hybrid learning scheme composedof two phases: the phase of rule generation from data, and the phase of rule tuning by using the error backpropagation learning scheme for a neural fuzzy system. In order to illustrate the performance and applicability of the proposedneuro-fuzzy hybrid model, extensive .simulation studies of nonlinear complex dynamics are carried out. The proposed method can be applied to on-line incremental adaptive learning for the purpose of prediction and control of nonlinear dynamical systems. Keywords-Neuro-fuzzy systems, Neural networks, Fuzzy logic, Parameter and structure learning, Knowledge acquisition, Adaptation, Time series. *TR-99-65, Department of Information Science, University of Otago, PO Box 56, Dunedin, New Zealand. Submitted and revised to Neural Networks, 15 March, 1999. H yF IS: Adaptive Neuro-Fuzzy Inference Systems 2