Neuro-Fuzzy System Modeling With Density-Based Clustering And Its Application In Chaotic Time Series Prediction

Neuro-Fuzzy is widely used for nonlinear system modeling Previous ways of fuzzy system modeling suffer from several drawbacks, including difficulty to determine the number of rules,rule redundancy and effects of noise pollution, which hinder the application of fuzzy system. This paper presents a new approach to neuro-fuzzy system modeling based on DENCLUE using dynamic threshold and similar rules merging (DDTSRM). We first introduce DDT, which is DENCLUE using dynamic threshold in density-attractors merging. DDTSRM is not only good at determining the number of rules because DDT is not sensible to the input parameters, but also able to obtain a better performance because DDT can find clusters of any density in arbitrary shape. After that we merge similar rules and identify the noise by considering similarity measures between fuzzy sets. Finally, BP method is used to precisely adjust the parameters of the fuzzy model. We employed Sugeno and Yasukawa's model and applied DDTSRM to a nonlinear function and gas furnace data of Box and Jenkins. Experimental results show that DDTSRM has overcome the drawbacks with a good performance. Moreover, we introduce the application of density-based Neuro-fuzzy modeling in chaotic time series prediction. DDTSRM is performed on Mackey-Glass data set to illustrate the efficacy.