比較季節性時間序列預測模型－台灣地區能源消費之實證研究

A comparison of seasonal time series models for forecasting the energy consumption in Taiwan

黃千珊(Chian-Shan Huang)

指導教授 ： 伍志祥

淡江大學/理學院/數學學系碩士班/碩士(2008年)

https://doi.org/10.6846/TKU.2008.00727

摘要

近年來，全球能源價格節節上升，需求量及消費量大增，由鑑於此，本文將針對台灣地區能源消費量進行預測，希望能掌握未來消費量趨勢，所以將採用四種預測模型，分別為季節性整合自我迴歸移動平均模型(SARIMA)、季節性時間數列迴歸模型(RMTSE)及倒傳遞類神經網路(BPN)，第四種模型將混和SARIMA與BPN(SARIMABP)，並探討此混和性模型是否能改善其預測結果。研究結果發現，當時間序列之資料圖形震盪較為明顯採用BPN能得到較好預測，反之，資料圖形震盪較為平穩，則採用SARIMA能得到較好預測，且採用混合性模型更能改善預測誤差。

關鍵字

時間序列 ； 能源 ； 能源消費量

並列摘要

Recently, the energy price keeps increasing.Both the demand and the consumption are on the rise.Due to these scenarios,this essay will try to predict the energy consumption in Taiwan,hoping to get a better grasp of the future trend.We will use the following four models for prediction,and they are Seasonal Autoregressive Integrated Moving Average Models(SARIMA),Regression Models with Time Series Errors (RMTSE),Back-propagation Network(BPN),and hybrid SARIMA and BPN(SARIMABP).The findings discovered that,at that time series of graph the sequence shook obviously uses BPN to be able to obtain a better
forecast,otherwise,the graph shook steadily, used SARIMA to be able to obtain a better forecast,and adopt the mixing model to be able to improve the forecast error.

並列關鍵字

Time series ； energy ； energy consumption

參考文獻

[11] 劉蘇萍，「台灣能源需求之結構性變化、影響因素及節能政策效果」，中原大學國際貿易學系，民國九十一年六月。

[2] Bowerman Bruce L.,O'Connell Richard T.and Anne B.Koehler,Forecasting,Time Series,

[3] R.E Abdel-Aal and A.Z.Al-Garni(1996),Forecasting Monthly Electric Energy Consumption In Eastern Saudi Arabla Using Univariate Time-Series Analysis.

[4] S.Gonzales Chavez,J.Xiberta Bernat and H.Llaneza Coalla(1998),Forecasting of energy production and consumption in Asturias(northern Spain),Energy24(1999)183-198.

[5] Fang-Mei Tseng,Hsiao-Cheng Yu and Gwo-Hsiung Tzeng(1999~2000),Combining neural network model with seasonal time series ARIMA model,Technological For- Ecasting & SocialChange69(2002)71-87.