Predicting a chaotic time series using a fuzzy neural network

In this paper the authors present an alternative neurofuzzy architecture for application to chaotic time series prediction. The architecture employs an approximation to the fuzzy reasoning system to considerably reduce the dimensions of the network as compared to similar approaches. The application considered is the chaotic Mackey-Glass differential equation. Simulation results for single and multi-step predictions were obtained using the MATLAB neural network toolbox and these are compared with both traditional neural network implementations and other fuzzy reasoning approaches. The work not only demonstrates the advantage of the neurofuzzy approach but it also highlights the advantages of the architecture for hardware realisations. (C) 1998 Elsevier Science Inc. All rights reserved.