The control of a chaotic process is a natural application for artificial intelligence techniques because the properties of a chaotic system typically result in an imprecise process model. In this paper we consider only one aspect to the problem, namely the formulation of an adequate model. The ability of an artificial intelligence technique to predict a chaotic time series is used to determine the "adequacy". Neural networks have shown some success at predicting chaotic time series but little is known of the predictive ability of a fuzzy logic technique although fuzzy logic controllers appear to be more acceptable to the engineering community. A comparison is made on the predictive capability of a neural network and fuzzy rule based approach for chaotic time series data. The two approaches are applied to the Lorenz system, the logistic equation and to a set of experimental fluidized bed data. Both one step ahead and long term prediction are assessed based on the mean error. Sensitivity of the results to the specific chaotic data set are investigated.

Return to publications and abstracts or Dr. Qammar's home page.