高速鐵路延人公里變動趨勢研究

Abstract

中文摘要 本研究旨在探討台灣高速鐵路自民國 96 年通車以來至民國 114 年間，其整體運周轉量（延人公里）之長期變動趨勢與季節性特徵。研究採用長達 19 年、共 228 筆的月度時間序列資料，運用 ARIMA 模型、時間序列分解法（Winters Multiplicati ve）及指數平滑法（Winters Additive）三種預測方法進行配適與預測。由於實證資料在民國 110 年因 COVID-19 三級警戒遭受劇烈的外在衝擊，本研究特別於 ARIMA 模型中加入介入變數（Intervention Analysis）以吸收極端異常值。結果顯示，含介入變數的 ARIMA(0,1,1)(0,1,1)₁₂ 模型在各項誤差評估準則（MSE、RMSE、MAE、MAPE）中表現最佳，能最精準地捕捉高鐵運量的長期上升趨勢與年底的高峰季節性規律。本研究不僅證實了台灣高鐵具備強大的疫後復甦力與長期成長動能，亦透過最佳預測模型為未來高鐵的運能調度與風險控管提供具體的科學建議。

Abstract This study aims to explore the long-term trends and seasonal characteristics of the overall passenger turnover (passenger-kilometers) of the Taiwan High-Speed Rail (THSR) from its commencement of operations in 2007 to 2025. Using 19 years of monthly time series data comprising 228 observations, this research employs three forecasting methods for fitting and prediction: the ARIMA model, the Time Series Decomposition Method (Winters Multiplicative), and the Exponential Smoothing Method (Winters Additive). Because the empirical data suffered a severe external shock in 2021 due to the COVID-19 Level3 alerts, this study specifically incorporates intervention variables (Inter venation Analysis) into the ARIMA model to absorb extreme outliers. The results show that the ARIMA(0,1,1)(0,1,1)₁₂ model with intervention variables performs best across various error evaluation criteria (MSE, RMSE, MAE, MAP E), most accurately capturing the long-term upward trend and the year-end p eak seasonal patterns of THSR passenger volume. This study not only confirms THSR's strong post-pandemic resilience and long-term growth momentum but also provides concrete scientific recommendations for future capacity scheduling and risk management through the optimal forecasting model.