A decomposition approach for fuzzy systems identification

This paper develops a novel approach to building Sugeno-type models. This approach consists of two steps: First, a fuzzy discretization technique is used to determine the membership functions of input variables, which is the most difficult aspect in constructing a Sugeno-type model. Second, an iterative algorithm, known as the EM algorithm, is used to estimate the parameters of linear regression models in the consequent part of the model. The approach has two salient features: 1) The premise identification and the consequence identification of the model can be separated through use of the fuzzy discretization technique, while these are mutually related in previous methods. This greatly simplifies the process of model construction. 2) The complex multiparameter optimization problem essential for building the model can be decomposed into L smaller-scale optimization problems by means of the EM algorithm, where L is the number of fuzzy rules. Hence, the complexity of this approach is essentially unaffected by the number of fuzzy rules in the model. Moreover, Because of the clear separation in algorithmic structure, the proposed approach can also be easily implemented on a parallel computer.