應用向量自我迴歸及遺傳規劃法建構能源消耗預測模型

Abstract

隨著時代的進步及人口的增長，能源漸漸成為經濟發展不可或缺的物質，因此能源政策與國家的發展息息相關。在經歷過三次石油危機之後，國際能源局勢動盪不安，能源價格節節攀升，各國政府開始重視能源的相關議題。因此若能準確地預測每一年能源的消耗量，就能讓國家訂出好的能源政策以節約使用能源。許多中外文獻以時間序列方析來建構能源消耗預測模型，但單變數時間序列並不考慮除能源消耗量以外的其他變數，而能源消耗量受到其他變數影響，若僅由單變量時間序列分析預測能源消耗量可能會不準確。由於能源的消耗量與經濟的發展息息相關，由每年的經濟指標來預測能源消耗量可以提升單變量時間序列方法之預測。因此本研究之主要目的是以人口、出口總額等經濟指標來預測能源消耗量，本研究首先應用向量自我迴歸(Vector Autoregression)來建立能源消耗預測模型，以經濟指標為自變數並考慮各經濟指標之落後期數對能源消耗量之影響，再利用遺傳規劃法來建立殘差預測模型，最後整合向量自我迴歸模型與殘差預測模型，得到一個高準確度的能源消耗量預測模型。最後，本研究利用台灣1979年~2010年的能源消耗量資料來驗證所提出之預測模型的準確率。

After three oil crises, the international energy situation is now characterized by volatility, rising energy prices, and heightened societal attention to issues related to energy. Accurate predict is a nation's annual energy consumption would enable help government to develop a policy of energy resources. Most studies constructed energy consumption prediction models for using univariate time series model. However, univariate time series model does not consider other variables such as some economic indice. When energy consumption are affected significently by other variables, the result of univariate time series model will be inaccurate. Because the energy consumption is tied up with the economic growth, this study will focus on predict energy consumption using the economic Indicators. This study utilizes economic indicators, such as population and total exportd and their lag effects as input data, and the energy consumption data as the output variable to build a Vector Autoregression model for predicting energy consumption. In addition, genetic Programming is employed to build a residual prediction model to enhance the forecasting accuracy. Finally, a hybrid energy consumption prediction model is developed by combining the Vetor Autoregression model and residual prediction model. The energy consumption data in Taiwan from 1979~2010 are utilized to demonstrate the effectiveness of the proposed method.