Fuzzy Model Identification

Takagi-Sugeno Fuzzy Model illustrates the local-rule for every local area with a linear equation and achieves global nonlinearity based on local linearity by fuzzy inference. Considering the merits above of the T-S Fuzzy Model, one kind of hierarchical evolutionary programming is proposed in this paper to train model configuration and parameters by changing chromosome into the hierarchical structure including controlling genes and parameter genes.After analyzing the astringency of the evolutionary programming and comparing with genetic algorithm, the configuration and parameters of the model are trained based on the hierarchical evolutionary programming. In order to quicken the process of convergence, the configuration and premise parameters of the fuzzy model is identified with the method above-mentioned to reduce the number of parameter genes, and the consequential parameters of the model are acquired by least square method.Then, since the function equivalence exists between the T-S Fuzzy Model and the RBF Neural Network in certain conditions, the topology and weights of RBF Neural Network are trained with the hierarchical evolutionary programming, and then configuration and parameters of the T-S fuzzy model are ascertained. The simulation results prove the method, and which of different methods are compared and analyzed finally.