A Self-Constructing General Type-2 Scheme for Neuro-Fuzzy System Modeling

• Main Authors: Wen-Hau Jeng, 鄭文豪
• Other Authors: Shie-Jue Lee
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/v2z966

碩士 === 國立中山大學 === 電機工程學系研究所 === 97 === We propose a self-constructing general type-2 fuzzy neural network for system modeling. The problems of constructing a general type-2 fuzzy neural network include type reduction, structure identification, and parameter identification. Regarding the type reduction, an α-planes strategy is used to decompose a type-2 fuzzy set into several interval type-2 fuzzy sets, and then apply the Karnik-Mendel algorithm to do type reduction to interval type-2 fuzzy sets. After getting both the lower and upper bound of the output for each α-plane, a crisp output value is obtained by the weighted average method. Since the amount of time required by this method is more demanding, an efficient strategy is proposed to solve this problem. Based on type reduction, a type-2 fuzzy neural network for fuzzy inference can be built. Regarding the structure identification, an incremental similarity-based fuzzy clustering method is used to partition the dataset into several clusters and a local regression model is obtained for each cluster, and then a type-2 fuzzy rule is extracted from each cluster. A hybrid learning algorithm which combines particle swarm optimization and recursive least squares estimator is adopted in the parameter identification to refine the antecedent and consequent parameters, respectively, of fuzzy rules. Simulation results show that our proposed method runs faster in type reduction without deterioration of the forecasting performance and the resulting networks obtained are robust against outliers.