應用灰色馬可夫及基因規劃法建構能源消耗預測模型

Abstract

科技的日新月異以及人口的快速成長，驅動了人類對於能源的需求，其影響層面日漸廣泛，包含了經濟發展、民生需求、環境保護等，且由於能源資源並非取之不盡，因此對於能源消耗量的預測以及管理已是各個國家現在與未來必須非常重視的議題。現有文獻大多使用傳統的統計方法來建構能源消耗量預測模型，但能源消耗量資料通常取得不易，因此資料筆數少，且可能呈非線性而未必能符合統計假設(即常態性、同質性與獨立性)，造成預測模型之準確度不高。目前已有研究提出使用灰色預測模型GM(1,1)與殘差修正GM(1,1)模型來預測能源消耗量，這些灰色預測模型的優點為僅需四筆或以上的資料即可建模且資料不需服從統計假設。然而，由於能源消耗量會受到各種隨機波動因素如：經濟發展、科技技術突破、國家政策改變等的影響，而GM(1,1)適配的函數為一條平滑的指數曲線函數，對於隨機性大的數據適配度不佳，預測準確度也不高。因此本研究之主要目的是針對能源消耗量資料，結合GM(1,1)、馬可夫鏈以及基因規劃法三者建立一個能源消耗量預測模型，以改善GM(1,1)模型之缺失。本研究最後應用1990年至2005年之美國能源消耗量資料驗證了本研究所提出之能源預測模型比現有文獻所提出之預測模型具有更佳的預測準確度。

The rapid growth of technology and population drives the human needs for energy. It influences widely on many national issues, including economic development, environmental protection, etc. However, because of energy resources are not inexhaustible, so the forecasting and management of energy consumption have become an increasingly important issue now and future. The conventional statistical methods are typically utilized to construct the prediction model, but the data collected on energy consumption are often very few, non-linear or do not meet the statistical assumptions, such as normality, homogeneity and independence, resulting poor prediction accuracy of forecasting models. Some studies have proposed GM(1,1) and residual modification GM(1,1) models to predict the energy consumption. The advantages of these GM(1,1) models can be constructed for at least four observations and they do not require any statistical assumptions. However, the energy consumption is influenced by random fluctuations of various factors, such as: economic development, technological breakthroughs, changes in national policy, and GM(1,1) is a smooth function of exponential curve, which has poor prediction accuracy when the data is random. Therefore, the objective of this study is to combined GM(1,1), Markov chains and genetic programming to establish the improved prediction models on energy consumption data. Finally, the energy consumption data set from 1990 to 2005 of US was utilized to demonstrate the effectiveness of the proposed energy consumption forecasting model.