An adaptive neuro-fuzzy approach for system modeling

Abstract

In this paper, a novel adaptive neuro-fuzzy modeling system is proposed for solving system modeling problems. Two phases are included in our approach.. In the first phase, a merge-based fuzzy self-clustering algorithm is used to automatically partition the sample data set into fuzzy clusters. Initial clusters are generated rapidly and similar clusters are merged together gradually based on similarity and distortion measures. TSK-type fuzzy rules associated with generated clusters are extracted. Then, the obtained rules are refined by a fuzzy neural network in the second phase. To speed up the convergence of learning, we develop a hybrid learning algorithm which combines a recursive SVD-based least squares estimator and the gradient descent method. Experimental results have shown that our method is more efficient than other methods.