The Study on Self-Constructing Least Wilcoxon-Generalized Radial Basis Function Neural-Fuzzy Inference System

• Main Authors: Cheng-Han Tsai, 蔡承翰
• Other Authors: Tsung-Ying Sun
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/2m8uxz

碩士 === 國立東華大學 === 電機工程學系 === 96 === The purpose of this thesis is to develop Self-Constructing Least Wilcoxon-Generalized Radial Basis Function Neural-Fuzzy Inference System (LW-GRBFNFIS) based on Radial Basis Function Neural-Fuzzy Inference System (RBFNFIS) network and applied to non-linear function approximation and chaos time sequence prediction. First, the hidden layer parameters of the most traditional RBFNFIS antecedent part were decided in advance and the output weights of the consequent part are evaluated by least square estimation manner. The fixed hidden layer structure of the RBFNFIS is lack of flexibility and can not adjust the structure of the RBFNFIS effectively according to the dynamic behavior of the system. The use of least square estimation for output weights values, the accuracy of the weights values will be affected by the outliers of the nonlinear function and the approximation performance is reduce for RBFNFIS. This thesis uses our previous research result: Self-Constructing RBFNFIS hidden layer structure and parameter with particle swarm optimizer (PSO) and use Least Wilcoxon norm to increase the accuracy of the output weights with outliers. This thesis uses many non-linear function approximation problems to verify the proposed method. The experiments results shows proposed method not only effetely solve outliers’ problems, but also increase the calculate time.