經濟指標對短期電力負載需求預測之性能影響研究

A Performance Evaluation Study of Economic Indices on Short-Term Load Forecasting

林政廷(Cheng-Ting Lin)

指導教授 ： 周立德

國立中央大學/資訊電機學院/資訊工程學系/博士(2014年)

摘要

2008-2009年間，由於美國雷曼兄弟破產之金融海嘯事件所導致的全球經濟衰退現象，引發電力負載需求量的驟減。由於此事件來得突然，傳統的電力負載需求預測技術無法有效反應此種突如其來的經濟事件，因此會造成預測值過高之問題。
為了解決此問題，本論文提出五種導入經濟指標作為影響參數之短期電力負載需求預測方法，該方法係混合各種不同經濟指標，並利用各種不同的景氣預測及經濟分類方法，以有效反應經濟的變動，並進而提高電力負載需求預測的準確性。
本論文之目標係研究包括台灣景氣指標(Taiwan business indicator, TBI)、台灣加權股票指數(Taiwan Stock Exchange Capitalization Weighted Stock Index, TAIEX)等在內的經濟指標對短期電力負載需求預測性能的影響。本論文選用2008-2011年台灣全國每小時的用電負載需求作為研究的範圍，選用平均絕對誤差比例值(Mean Absolute Percentage Error, MAPE)，及平均絕對誤差值(Mean Absolute Error, MAE)作為電力負載預測之性能指標，並選用不含任何經濟指標作為預測影響參數之傳統支持向量回歸方法(Support Vector Regression, SVR)作為性能比較的基準方法。
研究結果顯示導入TBI指標作為K-means分群法的分類參數及景氣條件判斷參數，並加入TAIEX作為SVR電力負載預測模型之輸入參數的混合經濟模型式的短期電力負載需求預測方法之預測性能優於其他四種方法。與傳統的SVR方法比較，在此最佳的條件下，於2008-2011年的預測時間範圍內，其四年總MAPE值及MAE值分別提升了16.19%及13.75%；而在受全球經濟衰退現象影響的時間範圍內，其MAPE值及MAE值則分別提升了57.59%及54.05%。
本論文係為全球第一個將股票指數作為電力負載預測模型之輸入參數，以預測短期電力負載需求之學術論文，其優越的電力負載預測性能亦呈現於本文。

關鍵字

短期電力負載需求預測 ； 支持向量回歸方法 ； K-means分群法 ； 經濟指標 ； 股票指數 ； 景氣指標 ； 經濟衰退 ； 移動平均

並列摘要

The global economic recession during 2008 and 2009, which was spurred by the bankruptcy of Lehman Brothers, sharply reduced the demand for electricity load. Traditional load forecasting approaches were unable to respond to sudden changes in the economy, because these approaches do not consider the effect of economic factors. Therefore, the over-prediction problem occurred.
To overcome this problem, this dissertation proposes five alternative economy based short-term load forecasting approaches. These approaches incorporate different kinds of economic index into load forecasting model by way of different business situation judgment algorithm and economy classification algorithm, to reflect to the economic variations and improve forecasting performance.
The objective of this dissertation is to study the influence of different Taiwan economic indices including Taiwan business indicator (TBI) and Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) on load forecasting performance. The Taiwan island-wide electricity load demands from 2008 to 2011 were employed as the case study for performance testing, and the load forecasting performance in terms of mean absolute percentage error (MAPE) and mean absolute error (MAE) by traditional support vector regression (SVR) load forecasting approach, which does not include any economic index in the regression model, is used as a comparative benchmark.
The results demonstrate that the hybrid economic models approach which combines TBI based K-means economic features classifier model, business situation judgment algorithm, and TAIEX included support vector regression model outperforms other proposed approaches. Compared to the traditional SVR approach, in the best condition, the MAPE is improved by 16.19%, and 57.59% respectively in the overall and economic recession affected period of 2008-2011; while the MAE is improved by 13.75% and 54.05% in the corresponding periods, respectively.
This dissertation is the first trial for introducing the stock index as an exogenous variable for short-term load forecasting, and the superior forecasting performance is exhibited.

並列關鍵字

Short-term load forecasting ； Support vector regression ； K-means ； Economic index ； Stock index ； Business indicator ； Economic recession ； Moving average

參考文獻

[2] S. Fan, L. Chen, W.J. Lee, “Machine learning based switching model for electricity load forecasting”, Energy Conversion and Management, Vol. 199, Issue 6, pp.1331–1344, June 2008.

[3] C.C. Hsu, C.Y. Chen, “Regional load forecasting in Taiwan－applications of artificial neural networks”, Energy Conversion and Management, Vol. 44, Issue 12, pp.1941–1949, July 2003.

[4] M. Meng, D. Niu, “Annual Electricity Consumption Analysis and Forecasting of China Based on Few Observations Methods”, Energy Conversion and Management, Vol. 52, Issue 2, pp.953–957, February 2011.

[5] P. Bunnoon, K. Chalermyanont, C. Limsakul, “Multi-substation control central load area forecasting by using HP-filter and double neural networks (HP-DNNs)”, International Journal of Electrical Power and Energy System, Vol. 44, Issue 1, pp.561–570, January 2013.

[6] M. Meng, D.X. Niu, “Annual electricity consumption analysis and forecasting of China based on few observations methods”, Energy Conversion and Management, Vol. 52, Issue 2, pp.953–957, February 2011.