Chaotic Time Series Prediction using Improved ANFIS with Imperialist Competitive Learning Algorithm

This paper presents an improved adaptive Neuro-fuzz y inference system (ANFIS) for predicting chaotic time s ries. The previous learning algorithms of ANFIS emphasized on g radient based methods or least squares (LS) based methods, but gradient computations are very computationally and difficult in each stage, also gradient based algorithms may be trapped into local optimum. This paper introduces a new hybrid learning algorithm based on imperialist competitive algorithm (ICA) for training the antecedent part and least square estimation (LSE) me thod for optimizing the conclusion part of ANFIS. This hybrid method is free of derivation and solves the trouble of fallin g in a local optimum in the gradient based algorithm for trainin g the antecedent part. The proposed approach is used in o rder to modeling and prediction of three benchmark chaotic me series. Analysis of the prediction results and comparisons wi th recent and old studies demonstrates the promising performa nce of the proposed approach for modeling and prediction of no nlinear and chaotic time series.