台灣國道一號服務區營收預測

Abstract

摘要 本研究旨在預測台灣國道一號沿線服務區的月營收情況，涵蓋中壢、湖口、泰安、西螺、新營、仁德及六站平均，研究資料起訖時間為2014年1月至2024年4月，研究數據的樣本內期間為2014年1月至2023年4月，樣本外預測期間則為2023年5月至2024年4月，關注各服務區在這段時間內的月營業額變化。特別是COVID-19疫情期間的波動。由於時間數列數據中呈現異常點，本研究使用介入分析捕捉這些結構性改變，所採用三種模式配適和預測，分別是ARIMA與介入分析、時間數列迴歸法與指數平滑法。分析過程中，針對每個服務區使用四種評估準則和三種預測模式，挑選出各站的最佳模式，並評估預測表現。其中評估準則包括均方誤差(Mean-square error，MSE)、平均絕對誤差(Mean absolute error，MAE)、平均百分比誤差(Mean Percentage Error，MPE)和平均絕對百分比誤差(Mean Absolute Percentage Error，MAPE)。數據顯示，COVID-19疫情開始後，有出現顯著結構性改變，這些波動主要受防疫措施影響，包括邊境管制、限制大型聚會和三級警戒等。隨著疫情逐步緩解和疫苗接種推進，服務區營收逐漸回升，特別是在2023年農曆新年期間達到高峰。比較結果顯示國道一號各站的最佳預測法，中壢站、泰安站、西螺站、仁德站為指數平滑法，湖口站、新營站為ARIMA 介入分析。分析結果表明，在面對複雜且多變的環境時，平均營收的方法在預測精度上更具優勢。這一發現對於理解疫情和假日效應對交通和旅遊業的影響，以及制定有效的管理和經營策略具有重要意義。

Abstract This study aims to forecast the monthly revenue of service areas along Taiwan's National Freeway No. 1, covering the Zhongli, Hukou, Tai'an, Xiluo, Xinying, and Rende stations, as well as their average performance. The research data span from January 2014 to April 2024, with the in-sample period from January 2014 to April 2023 and the out-of-sample forecasting period from May 2023 to April 2024. The focus is on the monthly revenue fluctuations in these service areas during this time, with particular attention to the impact of the COVID-19 pandemic. Due to the presence of anomalies in the time series data, this study employs intervention analysis to capture these structural changes. Three modeling and forecasting approaches are used: ARIMA with intervention analysis, time series regression, and Holt-Winters exponential smoothing method. We evaluate each service area using four criteria and three forecasting models/methods, selecting the best model based on forecasting performance. The evaluation criteria include Mean-Square Error (MSE), Mean Absolute Error (MAE), Mean Percentage Error (MPE), and Mean Absolute Percentage Error (MAPE). Significant structural changes occurred after the COVID-19 pandemic, driven by public health measures like border controls and Level 3 alerts. As the pandemic subsided and vaccinations progressed, peaking during Lunar New Year 2023. The comparison results show that exponential smoothing is the best forecasting method for Zhongli, Tai'an, Xiluo, and Rende stations, while ARIMA with intervention analysis performs best for Hukou and Xinying stations. Our results suggest that averaging monthly revenue across service areas enhances forecasting accuracy. These findings are important for understanding the impact of the pandemic and holiday effects on transportation and tourism, helping to inform more effective management and operational strategies.