| 摘要: | 企業之競爭優勢來自於準確之預測與有效之策略規劃，因此，各種預測技術陸續被發展出來。本研究之目的是應用適應性類神經模糊推論系統(ANFIS)模型與組合預測模型對觀光旅客數量、電力負載、能源消費等不同領域之資料作預測，同時驗證這兩種模型之預測績效。 本研究首先比較多個傳統預測方法、灰色預測模式及人工智慧預測法，分別針對台灣區域性電力負載與來台觀光旅客人數作預測分析，並評估其預測績效。其次，嘗試使用倒傳遞類神經網路(BPNN)將預測績效極佳之ANFIS法與其他預測模式組合，並分別針對世界能源消費最多之前三大國家及台灣能源消費(按部門別為基礎)作預測分析。 將本研究之實證案例作訓練與測試，結果發現ANFIS對於線性與非線性資料均能準確預測，且歸屬函數之配適情形為：高斯歸屬函數較適合用於台灣區域性電力負載需求預測，而鐘形歸屬函數適合用於預測來台觀光旅客人數、世界與台灣能源消費之預測。 根據常用之預測績效評估指標，包括絕對百分比差(APE)、平均絕對值差(MAE)、平均絕對百分比差(MAPE)、根均方差(RMSE)和統計結果，可知ANFIS模型比其他單一模型有更好之預測績效，而將ANFIS模型之預測結果與其他單一模型之預測結果以倒傳遞類神經網路模型組合而成之預測模型比任一單一模型有更好之預測績效。
We know that accurate forecasting and effective strategy planning can create the competitive advantage of an enterprise. So, various forecasting techniques have been developed. The purpose of this study is to apply the adaptive network-based fuzzy inference system (ANFIS) model and the combined forecasting model to forecast the data of various fields, such as the volume of tourists, electricity load and energy consumption, and demonstrate the forecasting performance of these two models. First, we use traditional forecasting methods, grey forecasting model and artificial intelligent forecasting methods to forecast the regional electricity load in Taiwan and the volume of tourists visiting Taiwan, and evaluate the forecasting performance. Second, we use the combined forecasting model, combining the results of ANFIS model and another forecasting model by back-propagation neural network (BPNN), to forecast the volume of energy consumption of world, top three countries and Taiwan. Via training and testing, we found that ANFIS can forecast accurately for both linear and nonlinear data. We also found that Gaussian membership function is suitable for the data of electricity load in Taiwan, and bell membership function is suitable for the volume of tourists and energy consumption. According to the evaluation index, including absolute percentage error (APE),mean absolute error(MAE), mean absolute percentage error(MAPE), root mean square error(RMSE) and statistical results, we can see that ANFIS model has better forecasting performance than other single model and the combined forecasting model has better forecasting performance than any single model. |
