严重冲突场景下山区双车道公路货车跟驰风险预测  

Real-time prediction of accident risk of truck following on two-lane mountain highway in severe conflict scenes

作　　者：戢晓峰[1,2] 耿昭师 曹瑞 普永明 JI Xiaofeng;GENG Zhaoshi;CAO Rui;PU Yongming(Faculty of Transportation Engineering,Kunming University of Science and Technology,Kunming 650500,China;Yunnan ModernLogisticss Engineering Research Center,Kunming 650500,China)

机构地区：[1]昆明理工大学交通工程学院,昆明650500 [2]云南省现代物流工程研究中心,昆明650500

出　　处：《安全与环境学报》2023年第9期2975-2983,共9页Journal of Safety and Environment

基　　金：国家自然科学基金项目(52062024,52002161)。

摘　　要：为预测严重冲突场景下山区双车道公路货车跟驰事故风险,基于视频轨迹数据、实测交通流数据、道路线形数据及交通事故数据,采用双变量冲突极值模型、轻量型梯度提升机,构建严重冲突场景下山区双车道公路货车跟驰事故风险实时预测(Truck Following Accident Risk Real-Time Prediction,TFARRP)模型,并利用耦合度模型分析变量耦合程度。研究显示:碰撞时间(Time to Collision)为5.418 s,后侵入时间(Post Encroachment Time)为0.512 s,这是严重冲突场景阈值;TFARRP模型准确率高达95.000%;平均车头时距(AHD)、两车间距(TCD)、货车平均速度(TAS)、及货车横向偏移(TLO)对TFARRP模型的重要度都超过了10%;当监测到AHD<21.11 s、TCD<35.00 m、TAS<29.00 km/h或TLO>0.25 m时,应及时发出预警信号,以保证车辆安全行驶。结合4个变量的耦合状况,研究结果可用于双车道公路货车跟驰的短临动态预警系统设计。Trucks following behind vehicles on two-lane mountain highways bring serious traffic safety hazards,so realtime prediction of truck accident risks on two-lane mountain highways has become a critical and realistic concern for traffic managers.The high-precision trajectory data and measured traffic flow data were extracted based on drone video and MC5600 barometric tube equipment,and road alignment data and traffic accident data are integrated into this paper.The bivariate traffic conflict extreme value model was used to select the threshold of identifying serious conflict scenes on a two-lane mountain highway.Based on the selected threshold,the light gradient boosting machine and coupling model was used to construct in severe conflict scenes the truck following accident risk real-time prediction model(TFARRP)on the two-lane mountain highway,and the coupling degree of variables was quantitatively analyzed.The result shows that TTC is 5.418 s and PET is 0.512 s are the thresholds for identifying the severe conflict scene of the truck following on the two-lane mountain highway.Based on the thresholds,the accuracy rate of the TFARRP model is 95.000%,AUC is 0.972,and the false alarm rate is 2.778%.The variables with an importance value of more than 10%in the TFARRP model are the average headway,the distance between two vehicles,the average speed of the truck,and the lateral offset of the truck.When the average headway is.less than 21.11 s,the distance between two vehicles is less than 35.00 m,the average speed of the truck is higher than 29.00 km/h or the lateral offset of the truck is more than 0.25 m.The probability of a truck following the accident on a two-lane mountain highway increases significantly in severe conflict scenes.Besides,early warning signals should be issued in time to enhance the risk awareness of the truck driver and ensure the safe driving of the vehicle.The coupling level between the average speed of the truck and the lateral offset of the truck is the highest.The average headway,the average spee

关 键 词：安全工程 货车 风险预测 机器学习算法 双车道公路 动态预警 

分 类 号：X951[环境科学与工程—安全科学]