整合多變量灰預測與基因規劃法預測二氧化碳排放量

Abstract

由於科技與經濟的發展，將伴隨著化石燃料的使用量激增。全球暖化將會加速空氣汙染且引發洪水以及旱災，不僅影響到我們人類的安全也會對經濟方面產生很大的改變。故關於二氧化碳排放量的趨勢及造成其排放量成長的原因已經成為各國重視的議題。然而，相關研究中提出許多影響碳排放的因素如燃料燃燒、運輸交通排放廢氣或國家人口數等，但過去有關碳排放研究大多只考慮部分影響因素，缺乏同時考量三種以上的影響因素。其次，過去碳排放的研究其預測模型通常須符合一些假設及限制，例如：常態分配、需要大筆資料或未能將影響因素考慮至模型中。綜合上述，提出整合型預測方法建立模型，此方法分為兩個階段，第一階段是多變量灰預測方法，該方法的優點為能將影響預測值的因素導入預測模型之中，並且僅需四筆以上之樣本就能夠準確地預測。不過當數據資料為非線性趨勢時，灰預測模型可能會出現較差的預測值。因此，為提高預測準確度，本研究方法第二階段將灰預測結合基因規劃法建立誤差修正模型，以降低預測誤差，此兩階段預測模型將被稱為混和型多變量灰預測模型。實驗部分將以臺灣地區為例，進行2000至2015年二氧化碳排放量預測及分析，考慮實驗不同影響因素的組合，利用本研究提出混和型多變量灰預測模型和過去的方法相互比較預測準確度，以確認此方法在實際問題運用上的價值。

Due to the technology and economic development, the use of fossil fuels surge will be accompanied. Global warming will speed up air pollution and initiation floods and droughts, not only affecting the safety of human beings, but also cause change on the economic. Therefore, the trend of carbon dioxide emissions and the factors for the growth of its emissions has become the subject of national attention. However, related studies have suggested that many factors that affect carbon emissions such as fuel consumption, Transport emissions or national population, but in the past most of the carbon emissions research only consider some of the factors, the lack of consideration more than three factors. Forecasting carbon emissions using conventional statistical methods usually requires some assumptions and limitations such as normality, large dataset and not considered the factors into forecasting model. Consequently, this study investigated two-stage forecasting model which are Multi-variable grey forecasting model and genetic programming. However, the grey forecasting model have large forecasting errors when data is nonlinear. To overcome the above problem, this study use Genetic programming to establish the error correction model to reduce the prediction error. The experiment will take Taiwan as an example, to forecast and analyze the carbon dioxide emissions from 2000 to 2015. Considering the combination of different impact factors, this paper proposes forecasting models to compare the past methods accuracy. Compare our methods and past methods differences in forecast accuracy.