高速公路合流区冲突极值建模与交通事故预测  

Conflict Extremum Modeling and Traffic Accident Prediction for Expressway Merging Areas

作　　者：郑来[1] 侯芹忠 郭延永[3] 杨晓冬[4] ZHENG Lai;HOU Qin-zhong;GUO Yan-yong;YANG Xiao-dong(School of Transportation Science and Engineering,Harbin Institute of Technology,Harbin Heilongjiang 150090,China;School of Automotive Engineering,Harbin Institute of Technology(Weihai),Weihai Shandong 264209,China;School of Transportation,Southeast University,Nanjing Jiangsu 210096,China;Harbin Vocational&Technical College,Harbin Heilongjiang 150081,China)

机构地区：[1]哈尔滨工业大学交通科学与工程学院,黑龙江哈尔滨150090 [2]哈尔滨工业大学(威海)汽车工程学院,山东威海264209 [3]东南大学交通学院,江苏南京210096 [4]哈尔滨职业技术学院,黑龙江哈尔滨150081

出　　处：《公路交通科技》2022年第10期132-140,共9页Journal of Highway and Transportation Research and Development

基　　金：国家自然科学基金面上项目(52072097);国家自然科学基金青年基金项目(71701046)。

摘　　要：为了更好地利用交通冲突技术开展高速公路合流区交通安全研究,提出了一种基于贝叶斯层级超阈值理论的冲突极值建模与交通事故预测方法。以辽宁省7个高速公路合流区的交通运行视频和交通事故数据为基础,利用后侵入时间(PET)识别了潜在合流冲突,进而构建了融合超阈值极值理论和贝叶斯层级结构(含数据层、过程层和先验层)的贝叶斯层级超阈值极值模型,提出了基于分位数回归的阈值选取方法,最后利用标定的最优模型对合流区交通事故进行了预测并对事故影响因素进行了分析。结果表明:调查时段内合流区共发生898次合流冲突;在构建的稳态模型、非稳态显著模型和非稳态全模型中,考虑冲突极值非稳态性和异质性的非稳态显著模型最优;最优模型的事故预测精度较高,其预测事故数与观测事故数的平均误差和平均绝对误差分别为1.0次/a和2.1次/a;通过建模发现加速车道长度、合流车辆类型、主线车辆类型、加速车道小时平均交通量对交通事故有显著影响,其中加速车道越长、主线车辆车型越大,交通事故发生的概率越小;合流车辆类型越小、加速车道小时平均交通量越大,交通事故发生的概率越大。所提出的贝叶斯层级超阈值方法解决了冲突极值的稀少性、非稳态性和异质性问题,实现了基于短期观测冲突数据的交通事故可靠预测。To better use the traffic conflict technique to investigate the safety of expressway merging area, the methods for conflict extremum modeling and traffic accident prediction based on Bayesian hierarchical supra-threshold theory are proposed. On the basis of traffic operation videos and traffic accident data collected from 7 expressway merging areas in Liaoning Province, the potential merging conflicts are identified using the PET. The Bayesian hierarchical supra-threshold extremum model, which integrated hierarchical supra-threshold theory and Bayesian hierarchical structure(including data layer, process layer and prior layer), is constructed. An approach for threshold selection based on quantile regression is developed. Finally, the traffic accidents of the merging areas are predicted and the influencing factors of accidents are analyzed by using the calibrated optimal model. The result shows that(1) There are 898 merging conflicts in the merging areas during the investigation.(2) In the constructed stationary model, non-stationary significant model and non-stationary full model, the non-stationary significant model which accounts for the non-stationarity and heterogeneity in conflict extremums is the optimal model.(3) The accident prediction accuracy of the optimal model is high, the average error and average absolute error of the predicted accidents and the observed accidents are 1.0 times/a and 2.1 times/a, respectively.(4) It is found through modeling that the length of acceleration lane, type of merging vehicle, type of mainline vehicle, and average hourly traffic volume on acceleration lane have significant influence on the accident occurrence. The longer the acceleration lane and the larger the size of the mainline vehicle, the smaller the probability of accident occurrence, while the smaller the size of merging vehicle and the larger the average hourly traffic volume on acceleration lane, the higher the probability of accident occurrence. The proposed Bayesian hierarchical supra-threshold method overcome

关 键 词：交通安全 交通事故预测 极值统计 高速公路合流区 交通冲突 

分 类 号：U268.6[机械工程—车辆工程]