T-S fuzzy modeling based on support vector learning

Abstract

This paper presents a satisfactory modeling method for data-driven fuzzy modeling problem based on support vector regression and Kalman filter algorithm. Support vector learning mechanism has been utilized to partition input data space to accomplish structure identification, then the complex model can be constructed by local linearization represented as T-S fuzzy model. For the ensuing parameter identification, we proceed with Kalman filter algorithm. Compared with previous works, the proposed approach guarantees the good accuracy and generalization capability especially in the few observations case. Numerical simulation results and comparisons with neuro-fuzzy method are discussed in order to assess the efficiency of the proposed approach. © Springer-Verlag Berlin Heidelberg 2006.