Optimization for a class of self-adaptation neuro-fuzzy models and its application to CSTR modeling

We study the model structure and parameter optimization for Takagi-Sugeno type neuro-fuzzy models based on the statistical information criteria, matrix singular value decomposition, rule elimination method and rule merging methods. Some novel parameter self-adaptation algorithms are presented in this paper, which can be used to modify cluster center values, cluster radius values and parameters of consequent functions of the neuro-fuzzy models. The neuro-fuzzy model and methods presented have been successfully applied to modeling a continuous stirred tank reactor (CSTR).