A novel fuzzy system modeling approach: multidimensional structure identification and inference

Abstract

A new fuzzy system modeling approach that uses an inference mechanism working in the input-output space is proposed. The new inference mechanism eliminates the need to identify the membership functions on each separate system variable axis and avoids the problems due to the projection step of some popular fuzzy system modeling approaches. In the new method, inputs and outputs are first clustered together by means of the fuzzy c-means (FCM) algorithm, with several levels of fuzziness, m, and numbers of clusters, c. Instead of a cluster validity index, the system output error is used as our performance index while selecting the best (m,c) pairs. Then, a modified version of the classical simulated annealing algorithm is used to identify the relative weights of the system input variables.