建構高速公路交通事故延時與占用車道數之聯立模式

Abstract

高速公路之事故延時與占用車道數係造成事故衝擊車流之主要因素，過去有許多研究利用不同分析方法預測交通事故延時及探討影響事故延時之因素，且在不同的研究中，發現不同地區的研究結果，其事故延時模式最適分配也有所差異，而目前占用車道數多為研究中的解釋變數，未有研究探討占用車道可能亦係交通事故之重要結果變數，且事故占用車道數越多事故延時越長，占用車道數對於事故延時可能存在內生性，據文獻回顧所知，目前未有研究考慮占用車道數對於事故延時之內生性。本研究以聯立方程建構事故延時與占用車道數模式，其中占用車道數為事故延時的內生變數。本研究共考慮三大類變數，包括：(1)事故特性；(2)道路及環境特性；(3)時間空間特性，並透過加速失效模型(Accelerated failure time, AFT) 建構交通事故延時模式，且同時建構占用車道數之次序性反應模式。 本研究利用2011年的事故資料進行推估，並以2013年的中區事故資料進行驗證。在個別的交通事故延時與占用車道數模式推估結果顯示，交通事故延時模式以伽瑪脆弱對數羅吉斯特加速失效模式為最佳模式，而占用車道數模式則以異質選擇模式為最適模式。聯立方程模式推估結果指出占用車道數對於事故延時具有內生性，必須以聯立方程同時推估交通事故延時及占用車道數之參數值，避免於個別模式參數推估所造成之偏誤。事故延時模式之MAPE值為54.49%，事故延時模式於20-40分之MAPE值高達22.26%。在聯立模式之事故占用車道數模式中，事故車輛數越多、死傷事故、大型車輛事故、事故發生在隧道及第二、三、四警察隊管轄範圍發生的事故，其占用車道數相對較多；而聯立模式之事故延時模式中，若事故具有死亡、受傷、事故占用車道數越多、事故車輛數越多、牽涉大型車輛、起火事故、翻車事故及非追撞事故，其事故延時相對較長。此外，若事故發生在晨峰、夜間及星期三，事故延時也越長。

Freeway accident duration and number of lanes blocked are the primary causes of traffic congestion. Past studies have applied various approaches to estimate accident duration and a number of factors affecting accident duration were successfully identified; however, few studies examined the factors contributing to the number of lanes blocked. Especially, the number of lanes blocked may be regarded as an important factor affecting to accident duration. To our best knowledge, none of studies have considered the endogeneity of number of blocked lanes when modeling incident duration regardless of the fact that more blocked lanes normally indicate longer duration. Accordingly, this study proposed a simultaneous equation model where number of blocked lanes was an endogeneous variable in explaining accident duration. Three sets of explanatory variables for both duration and number of lanes blocked were considered, including accident characteristics, road and environmental features, and spatial and temporal characteristics of accidents. Accident duration was modeled with accelerated failure time (AFT) models while number of lanes blocked was modeled by ordered response models. The models are developed with 2011 Taiwan freeway accident data and validated with 2013 data. The estimation results of separate single equation models of accident duration and number of lanes blocked show that the AFT model with a log-logistic distribution and Gamma-distributed frailty is the best fitted model for accident duration; while heterogeneous choice model is the best fitted model for number of blocked lanes. The estimation results of the simultaneous equation model indicated the existence of endogeneity of number of lanes blocked in modeling accident duration, suggesting that the proposed simultaneous equation model could avoid biased estimations in separate models. The MAPE for accident duration is 54.59%, where MAPE reaches 22.26% for the accidents with 20-40 minutes duration. The estimated model suggests that positive effects on number of lanes blocked are caused by injured or fatal accidents, large vehicles involved accidents, tunnel accidents, number of vehicles involved, and accidents in police jurisdiction zones 2, 3 and 4. As to the accident duration, the longer duration is caused by injury or fatal accidents, more number of lanes blocked by accidents, number of vehicles involved, large vehicles involved accidents, fire accidents, rollover accidents and non-rear-end collision accidents. Additionally, significantly longer duration can also be found during peak periods, nighttime, and Wednesday.