Robust agglomerative clustering algorithm for fuzzy modeling purposes

This paper addresses the Takagi-Sugeno-Kang (TSK) fuzzy model identification. An enhanced algorithm that uses clustering techniques for the approximation of nonlinear systems from the data is presented. The algorithm combines the parallel axis version of the Gustafson-Kessel (GK) algorithm with the fuzzy C-regression models (FCRM) in order to maintain the interpretability and improve the global accuracy of the model. A low sensibility to noise and automatic detection of the number of clusters is achieved by using robust statistic and competitive agglomeration techniques similar to the techniques developed in the robust competitive agglomeration (RCA) algorithm. Finally, two numeric examples concerning to static and dynamic nonlinear systems are shown to demonstrate the effectiveness of the proposed algorithm.