| Summary: | 碩士 === 國立成功大學 === 高階管理碩士在職專班(EMBA) === 102 === The Adaptive Network-Based Fuzzy Inference System (ANFIS) is widely applied to classification and numerical forecasting problems nowadays. Although the ANFIS is developed to obtain the optimal results of FIS with artificial neural networks by adapting the initial antecedent parameters in FIS, the profiles of fuzzy membership functions, how to well set the initial antecedent parameters still can affect the learning results. Focusing on the numerical forecasting problems, this study develops a systematic procedure, which employs the Fuzzy C-means with the fuzzy silhouette coefficients to identify the locations of observations with the most suitable clustering sizes, and then based on which the mega-trend-diffusion technique is taken to estimate the membership functions as the initial antecedent parameters for ANFIS. In the experiment, a real case taken from one of the leading TFT-LCD (thin-film transistor liquid-crystal displays) manufacturers is examined. The experiment results reveal that the predictive preciseness of ANFIS is improved when the initial antecedent parameters are set with the proposed procedure.
|
|---|
