多车道交织区车辆跟驰行为风险判别与冲突预测

Risk Discrimination and Conflict Prediction of Vehicle-following Behavior in Multi-lane Weaving Sections

车路协同系统(IVICS)是保障安全高效出行的新兴技术之一,将高精度车辆轨迹数据与机器学习方法相结合,提出一种可应用于IVICS的多车道交织区的潜在风险判别与冲突预测方法。首先,基于无人机视频,从广域视角提取交织区交通矢量位置、速度等信息,并划分上下游、交织影响区等多个分区;然后,考虑决策行为(车车边缘距离、接近率)与车辆行为(横纵向速度、加速度、速度角度)构建风险判别模型,以单位面积冲突次数、持续时间、冲突密度等指标评估风险;最后,基于朴素贝叶斯模型与logistic回归模型分别进行交通冲突预测,与实测数据相比,预测准确率分别为74.86%、87.10%,Area Under Curve分别为0.84、0.88,表明logistic回归模型具有更好的预测性能。研究成果有助于交管部门制定与优化交通管控方案,可应用于IVICS动态预警。

Intelligent Vehicle Infrastructure Cooperative Systems(IVICS)is one of the emerging technologies to ensure travel security and efficiency.Combining high-precision vehicle trajectory data with machine learning method,a potential risk identification and conflict prediction method for multi-lane interweaving zones is proposed,which can be applied to IVICS.First,the information about the traffic vector position and speed in the interweaving area was extracted from the wide-area perspective based on UAV video.Several partitions were divided,such as the upstream and downstream sections,and the interweaving influence area.Then,the risk discrimination model was constructed by considering the decision behaviors(vehicle-vehicle edge distances,approach rates)and vehicle behaviors(the transverse and longitudinal velocity,acceleration,velocity angle).Risks were evaluated from the number of conflicts per unit area,duration of conflicts,conflict density and other indicators.Finally,traffic conflicts are predicted according to the Naive Bayes model and the Logistic regression model.Compared with the measured data,the prediction accuracy rates of the two models were 74.86% and 87.10% respectively,and the AUC were 0.84 and 0.88 respectively.It showed that the Logistic regression model has better prediction performance.The corresponding results are helpful for traffic management departments to formulate and optimize traffic control schemes and can be applied to IVICS dynamic early warning.