臺鐵事故持續時間特性與模式之研究

Abstract

臺灣鐵路運量隨著鐵路捷運化政策的實施，讓旅運量持續穩定的成長，可說是我國一重要之運輸動脈。然而，臺鐵長年來的問題─準點率，依然無法達到歐洲、日本等先進國家的水準，通常造成鐵路誤點原因大部分可歸因為鐵路事故之發生，因此對於事故之深入分析有其必要性。本研究目的旨在藉由整理臺鐵運務處所提供之97年~99年因事故造成列車誤點的行車事故報告資料，先對可取得知資料範圍坐變數篩選，以SPSS之敘述性統計對臺鐵列車事故作基本的特性分析，了解各種類的事故發生頻率、對於後續班次所造成的影響以及事故本身的持續時間，再以SAS軟體用存活分析模式來做時間面之探討，檢視因事故所造成軌道上列車停駛到排除事故並恢復行駛這段持續時間長短之相關因子的顯著程度，藉以模式化來觀察各類事故之持續時間的特性。經由敘述性統計方法之研究結果顯示：臺鐵鐵路事故平均每件事故之事故持續時間為「47.07」分鐘、平均每件事故會影響「5.00」班列車車次總計共造成「88.87」分鐘的總誤點時間。經由存活分析統計方法之研究結果顯示：事故持續時間的分配偏向Lognormal、Loglogistic分配，因此以存活分析之參數模型會有較佳之適合度。由模式之顯著性發現，整體而言臺鐵發生在4月~6月、7月~9月之事故會相較於1月~3月、10月~12月有較長的事故持續時間；發生在站間事故會顯著比車站事故的事故持續時間還來得長；禮拜日皆會相較有較長的事故持續時間。研究結果所提供之事故分析資訊乃可作為臺鐵未來對於事故之討論、調查、改善之方向與參考。

By implementing the Metro Style policy of TRA, passenger volume grows rapidly and steadily. Railway system obviously becomes one of the most important transportation in Taiwan. However, a problem exists over years and has no significant improvement, punctuality rate, which can not reach to the level of developed countries like Europe or Japan. Generally speaking, most of the delays can refer to railway accidents and thus are reasonable to have a deeper analysis to the accidents of TRA. This study synthesizes three-year-data(2008~2010) of railway accidents providing by TRA and analyzes the characteristics of railway accidents by descriptive statistics using SPSS to explore the frequency of accident, how long the durations in average are and how the railway accidents affect. Duration model of survival analysis is then applied to analyze the critical factors of railway accident duration. The result of descriptive statistics shows: the average accident duration of TRA accidents is ”47.07” minutes, accidents affect average “5.00” numbers of runs for train and average totally ”88.87” minutes. The result of survival analysis shows: the Lognormal and Loglogistic distributions come with proper fit to data and are selected as appreciate parametric model. The model estimation results show that a wide variety of factors affect accident duration. In general, accidents in Apil-June and July-September has significantly longer accident duration than in January-March and Octorber-December; accidents occurring between stations has significantly longer accident duration than at stations; Sunday is also found to have significantly longer accident duration than other days of a week. The findings of this study provide a demonstration of method to assess accident management programs.