| 基于轨迹数据的山区危险性弯道路段交通事故风险动态预测 |
| |
| 引用本文： | 戢晓峰,谢世坤,覃文文,杨文臣,胡澄宇.基于轨迹数据的山区危险性弯道路段交通事故风险动态预测[J].中国公路学报,2022,35(4):277-285. |
| |
| 作者姓名： | 戢晓峰 谢世坤 覃文文 杨文臣 胡澄宇 |
| |
| 作者单位： | 1. 昆明理工大学 交通工程学院, 云南 昆明 650504;2. 云南省现代物流工程研究中心, 云南 昆明 650500;3. 云南省交通规划设计研究院 陆地交通气象灾害防治技术国家工程实验室, 云南 昆明 650200 |
| |
| 基金项目： | 国家自然科学基金项目(52062024,52002161);国家重点研发计划项目(2017YFC0803906);陆地交通气象灾害防治技术国家工程实验室开放基金项目(NEL-2019-05) |
| |
| 摘 要： | 针对山区双车道公路危险性弯道路段交通事故多发的现实问题，提出主动评估短时交通流状态下的交通事故风险，以降低交通事故发生率。采用无人机高空拍摄弯道路段交通流运行状态，利用计算机识别技术提取高精度的车辆轨迹和交通流数据，结合山区双车道公路弯道路段危险驾驶行为特征表征交通冲突，以距离碰撞时间为交通冲突量化指标，提出山区车道公路弯道路段交通冲突严重程度类型的阈值划分标准。在此基础上，选取统计分析(LR模型)和非参数数据挖掘技术(RF和SVM模型)构建山区双车道公路弯道路段交通事故风险动态预测模型，以混淆矩阵和AUC为评价指标，对比分析各模型的整体预测能力；最后，选择预测精度最好的模型定量分析特征变量与交通事故风险之间的关系。研究结果表明：在车辆轨迹和交通流实测数据的环境下，采用RF算法构建的山区双车道公路弯道路段交通事故动态风险预测模型准确度最高，准确率达到78.9%，AUC为0.846 2，较LR模型和SVM模型精度分别提高了21.8%和15.93%，可以较为准确的评估山区双车道公路弯道路段交通事故风险的高低；车头时距(AHD)和货车混入率(RT)是风险动态预测模型中相对重要度最高的2个特征变量，且在AHD小于20 s、RT达到0.6时事故风险发生概率会大幅增大。研究成果可应用于山区双车道公路的短临动态预警系统的设计，指导安全管理对策制定。
|
| 关 键 词： | 交通工程 事故风险动态预测 机器学习 弯道路段 车辆轨迹数据 山区双车道公路 |
| 收稿时间： | 2020-06-15 |
Dynamic Prediction of Traffic Accident Risk in Risky Curve Sections Based on Vehicle Trajectory Data |
| |
| JI Xiao-feng,XIE Shi-kun,QIN Wen-wen,YANG Wen-chen,HU Cheng-yu.Dynamic Prediction of Traffic Accident Risk in Risky Curve Sections Based on Vehicle Trajectory Data[J].China Journal of Highway and Transport,2022,35(4):277-285. |
| |
| Authors: | JI Xiao-feng XIE Shi-kun QIN Wen-wen YANG Wen-chen HU Cheng-yu |
| |
| Affiliation: | 1. Faculty of Transportation Engineering, Kunming University of Science and Technology, Kunming 650504, Yunnan, China;2. Yunnan Modern Logistics Engineering Research Center, Kunming 650500, Yunnan, China;3. National Engineering Laboratory for Surface Transportation Weather Impacts Prevention, Yunnan Institute of Transportation Planning and Design, Kunming 650200, Yunnan, China |
| |
| Abstract: | In view of the problem of numerous in two-lane mountain highway. It is proposed to actively assess the risk of traffic accidents in the short-term traffic flow state, to reduce the incidence of traffic accidents. The UAVs were used to collect traffic flow from high altitudes in curve sections, and computer recognition technology was used to extract high-precision space-time traffic trajectories and traffic flow data. This study combined the characteristics of dangerous driving behavior to indicate traffic collision, used the time to collision as a quantitative index, and constructed the threshold classification standard of traffic collision severity of mountain road corner section. Based on this, the statistical (LR model) and nonparametric data mining techniques (RF and SVM model) were used to construct a dynamic prediction model of traffic accident risk in two-lane mountain highway curve sections. The confusion matrix and AUC as evaluation indicators, and model prediction accuracy was compared and analyzed. Finally, the most accurate prediction model was used to analyze the relationship between variables and accident risk. The results reveal that, the dynamic risk prediction model of traffic accident in two-lane mountain highway curve sections based on RF is the most accurate under the background of vehicle trajectory and traffic flow measurement data, and the accuracy rate is 78.9%, the AUC is 0.846 2, which is 21.8% and 15.93% higher than the accuracy of LR model and SVM model respectively. In addition, it is found that the average road headway (AHD) and truck mixing rate (RT) are the most important feature variables in the risk dynamic prediction model. And when AHD is less than 20 s or RT reaches 0.6, the probability of accident occurrence increases sharply. The research results can be further applied to the design of short-term dynamic early warning system of two-lane mountain highway, and guide the development of security management countermeasures. |
| |
| Keywords: | traffic engineering dynamic prediction machine learning curve sections vehicle trajectory data two-lane mountain highway |
|
| 点击此处可从《中国公路学报》浏览原始摘要信息 |
|
点击此处可从《中国公路学报》下载全文 |
|
