A fast approach for automatic generation of fuzzy rules by generalized dynamic fuzzy neural networks

Generalized dynamic fuzzy neural networks (G-DFNN) based on ellipsoidal basis functions, which implement TSK fuzzy inference systems, are presented to extract fuzzy rules from input-output sample patterns. The salient characteristics of the approach are: (1) fuzzy rules can be gained quickly without using the backpropagation iteration learning; (2) the online self-organizing learning paradigm is employed so that structure and parameters identification are done automatically and simultaneously without partitioning the input space and selecting initial parameters a priori; (3) the sensitivity of fuzzy rules and input variables are analyzed based on the error reduction ratio method so that fuzzy rules can be recruited or deleted dynamically and the premise parameters of each input variable can be modified. Simulation studies and comprehensive comparisons with some other approaches demonstrate that the proposed scheme is superior in terms of learning efficiency and performance