Fuzzy rule extraction for controller designs

This paper presents an innovative method for extracting fuzzy rules directly from numerical data for controller designs. Conventional approaches to fuzzy systems assume there is no correlation among features and therefore involve dividing the input and output space into grid regions. However, in most cases, it is likely that features are highly correlated. Therefore, we propose to use an aggregation of hyperspheres with different sizes and different positions to define fuzzy rules. The genetic algorithm is used to select the parameters of the proposed fuzzy systems. The inverted pendulum system is utilized to illustrate the efficiency of the proposed method for finding fuzzy control rules