Fuzzy Logic and Chaos Theory in Time Series Forecasting

This paper presents a method of time series forecasting based on the integration of fuzzy logic and chaos theory. The proposed method has two stages. On the first stage, we consider the time series as a dynamic system and using the methods of mutual information and false nearest neighbors, as a part of applied chaos theory, we reconstruct the phase portrait corresponding to the original time series. On the second stage, we are learning the neuro fuzzy network as a model of time series forecasting using the vectors points of reconstructed phase portrait. We consider all the formalisms necessary for understanding the method and present the results of two computer experiments proving the ability of fuzzy inference accuracy increasing using the selection of optimal parameters of time delay and phase portrait dimension.