台鐵東部幹線列車座位需求預測

Abstract

台灣鐵路列車整體運能不足一直是台鐵營運上的重要課題之一，103年6月台鐵花東鐵路電氣化通車後，傾斜式自強號列車成為東部幹線主力車種，吸引大量觀光旅客搭乘東部幹線火車，導致東部幹線民眾返鄉車票一票難求情形益加嚴重。且起迄站座位分配不均導致增加訂票困難度。因此本研究目的為針對東部幹線旅運需求，且依據時段、星期、月份及季節等波動性，提出運量預測方法。 本研究首先蒐集民國103年7月至民國104年7月間台北-花蓮及台北-台東區間之實際旅運料，並針對不同時段別、星期別、月份別等特性變數進行旅運需求趨勢分析，利用時間變數進行多元迴歸方法建立旅運需求預測模式。 由實證結果發現此預測模式對於傾斜式自強號列車旅運量預測效果良好，總平均誤差率結果均小於 20%;且和其他車型列車相較之下，本研究發現預測傾斜式自強號列車旅運量有較好之預測效果。從預測結果顯示時段是影響旅運需求最大的因素，緊接著是星期和月份，有助呈現旅客因時間而異的喜好程度，本研究亦針對座配分配提出策略方法，為了強化此模型之預測正確性，未來可考量加入其他影響變數，例如:氣候、人口、經濟成長或大型活動等，做為未來研究之方向。

Insufficient transport capacity remains one of the major challenges of the Taiwan Railways Administration (TRA) recent years. Since the section between Hualien and Taitung was electrified in June 2014, the tilting express has become the major train type serving the East Link and a large number of travelers have been attracted to visit East Taiwan. As a result, the difficulty in reserving back-home train tickets for the residents living in the East Taiwan becomes more severe. Improper allocation of OD seat capacity will even worsen the difficulty. Therefore, this study aims to prospose a ridership forecast method for the East Link based on real travel demand may based upon time-of-day and day-of-week and month-of-year, and seasonal factors. To do so, this study firstly collected the actual ridership data of the section of Taiper-Hualien and Taipei-Taitung during July, 2014 and July, 2015. Afterwards, travel demand distribution patterns across time-of-day and day-of-week and month-of-year were analyzed. A multiple regression model was then estimated by regressing the ridership on time variables. According to the training and validation results, the prediction accuancy of the estimated multiple regression model for tilting express are lower and 20%, suggesting a good fit of the medel. Additionally, the prediction accuracy for titling express performs better than those of other types of trains. The estimation results also show that the time-of-day factor has the largest effect on travel demand, following by day-of-week and month-of-year, showing the time preference of travelers. At least, the seat allocation strategies are then proposed accordingly. To further enhace the prediction accuracy of the proposed model, more influential factors, such as wether, population, economy growth, and larges events an be further incorporated in future studies.