電力負載量的預測準確性在電力公司扮演著重要的角色。如果高估的話，會造成不必要的能源浪費，另一方面，低估會造成供電不足而跳電。本研究我們將探討兩種預測方法，第一種方法為直接預測負載量；第二種方法為各別預測每天總量以及各時刻之所佔比例，並使用模型整合來合併這兩種方法。在建立模型的過程中，利用預測日的前四週作為訓練樣本，預測當天與未來七天的負載量，計算出新加坡在2013 年至2015年的平均絕對比例誤差、尖峰絕對比例誤差與離峰絕對比例誤差，呈現短期負載預測方法的有效性。另外因為負載量會受國定假日的影響，我們將使用不同的方式去做預測分析。

Accuracy of electricity load forecasting plays an important role for electric utilities and regulators. Overestimation of electricity load demand will cause waste on energy. On the other hand, underestimation may cause the electrical power off due to lack of power supply. In this work, we investigate two forecasting methods, one works directly with the original load observed at each time period, the other works with daily total and corresponding proportions on each time period. We also use model averaging method to combine the two forecasts. The forecasting results will be evaluated by the mean absolute percentage errors (MAPE), peak absolute percentage errors (Peak APE), valley absolute percentage errors (Valley APE) for the three years (2013-2015) on the Singapore load data, where for the forecasting day and the next seven days, based on four weeks training data before the forecasting day, are provided to demonstrate the effectiveness of the newly proposed model averaging short-term load forecasting methodology. As Loads are affected by the national holidays, we will discuss different ways to do the analysis for the special days.

論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
List of Tables vii
List of Figures viii
1 Introduction 1
2 Data description 2
2.1 Electricity load data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Electricity load data visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Meteorological data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.1 Temperature effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5
3 Short-term load forecast 7
4 Semi-parametric regression model 8
4.1 B-spline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8
4.1.1 Calendar effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10
4.1.2 Model selection criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11
4.1.3 Serial correlation of the residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12
5 Estimation method 13
5.1 Hybrid estimation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13
5.1.1 Long-term load estimation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14
5.1.2 Short-term load estimation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16
5.2 Proportional estimation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18
5.2.1 Daily total model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18
5.2.2 Proportional model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
5.3 Model averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21
6 Special days analysis 23
6.1 Special case: National day in 2015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .26
6.2 Update on today . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27
7 Forecasting results 28
7.1 Forecasting results for the ordinary days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28
7.2 Forecasting results for the special days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30
8 Conclusions 31
9 References 33
10 Appendix A 34
10.1 Appendix A-I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .34
10.2 Appendix A-II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38
10.3 Appendix A-III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
11 Appendix B 40

1. Cho, H., Goude, Y., Brossat, X. and Yao, Q. (2013). Modeling and Forecasting Daily Electricity
Load Curves: A Hybrid Approach. Journal of the American Statistical Association,
108, 7-21.
2. Fan, S. and Hyndman, R. J. (2012). Short-Term Load Forecasting Based on a Semi-
Parametric Additive Model. IEEE Transactions on Power Systems, 27, 134-141.
3. Hong, T., Gui, M., Baran, M. E. and Willis, H. L. (2010). Modeling and Forecasting
Hourly Electric Load by Multiple Linear Regression with Interactions. Power and Energy
Society General Meeting, IEEE, 1-8.
4. Harvey, A., and Koopman, S. J. (2008). Forecasting Hourly Electricity Demand Using
Time-Varying Splines. Journal of the American Statistical Association, 88, 1228-1236.
5. Steadman, R. G. (1984). A universal scale of apparent temperature. Journal of Climate
and Applied Meteorology, 23, 1674-1687.
6. Tsai, P. S., Lin, Y. H. and Chou, L. W. (2007). Constrained Optimization Method for
Collision-Free B-spline Trajectory Planning. Journal of China Institute of Technology, 36,
247-253.
7. Tsay, R. S. (2010). Analysis of Financial Time Series 3rd Edition. Wiley.
8. Energy Market Authority. https://www.ema.gov.sg/index.aspx.
9. Heat index. https://en.wikipedia.org/wiki/Heat_index.