Aiming at the mechanism of traffic conflicts in urban expressway weaving areas under different weather conditions, a traffic conflict identification algorithm based on vehicle bounding box information was proposed. The peak over threshold extreme value theory was employed to determine severe conflict thresholds under distinct weather conditions, and machine learning models combined with interpretability analysis methods were utilized to systematically investigate the impact of weather on traffic conflict characteristics. By leveraging high-precision trajectory data, the limitations of traditional conflict metric extraction methods, such as insufficient accuracy, subjectivity in threshold selection, and the lack of exploration into weather-related influences were addressed. The results demonstrate that the severe conflict thresholds for sunny and rainy conditions are 1.75 s and 1.85 s, respectively, with the proportion of severe conflicts in rainy conditions (39.1%) being significantly higher than in sunny conditions (9.7%). Through interpretability analysis, the following findings are revealed: ①Weather conditions significantly modulate the influencing factors of traffic conflicts; ②Under sunny conditions, spatial vehicle positions (particularly whether the conflicted vehicle is located in a ramp or weaving lane) dominate the severity of conflicts; ③In rainy conditions, the intensity of influencing factors increases markedly, and traffic flow characteristics emerge as the predominant factors. These findings provide a theoretical foundation for differentiated management and safety prevention and control in weaving areas.
针对交织区安全问题,现有研究主要采用两类数据:传统事故数据和替代性安全评价指标,如碰撞时间(Time to collision,TTC)。在传统事故数据分析领域,曹倩等[1]提出了考虑风险异质性的交通事故风险评估方法,通过融合道路类型和事故密度,提高了传统评估的准确性。Wang等[2]则结合事故数据、道路几何特征和天气因素,构建了多层次贝叶斯逻辑回归预测模型。
相较于事故数据的时空局限性,交通冲突数据凭借其采集便捷性、样本丰富性等优势,逐渐成为研究焦点,其与事故的关联性已经获得部分验证[3]。相关研究主要依赖实测车辆轨迹数据,仅少数采用仿真方法[4]。例如,Yuan等[5]、李佳硕等[6]、马菲[7]、卢启慧[8]以及Onelcin等[9]的研究中,均使用无人机拍摄或运动相机拍摄的方式获取原始轨迹数据。在从原始轨迹数据中提取交通冲突指标及交通流参数时主要依托3类技术:计算机视觉算法[10]、运动识别软件[8]或算法结合人工识别的方式[9]进行。而阈值确定方法包含:①经验法,如Xia等[10]直接采用TTC=4 s阈值;②累计频率法,马菲[7]以15%分位数划分冲突等级,卢启慧[8]选用85%分位数,赵涛等[11]验证了该方法的适用性;③极值理论法,穆松成[3]构建了侵占后时间(Post encroachment time,PET)和改进碰撞时间(Modified time to collision,MTTC)超阈值模型,通过平均剩余寿命图优化阈值选取。
为衡量交通冲突,学界已提出多种替代性安全评价指标,例如TTC、MTTC、避免碰撞减速度(Deceleration rate to avoid a crash,DRAC)等。本文研究对象为城市快速路交织区,主要冲突类型为交织车流之间产生的小角度换道冲突和车辆换道引发其他车辆制动避险行为导致的追尾冲突。此外,本文旨在利用高精度轨迹数据精确识别潜在碰撞的即时风险,因此选择TTC作为核心度量指标,并着重于改进其计算精度以适应交织区的冲突特点。该指标在时间层面上,反映了当前时刻下后车速度大于前车时,若双方维持原有的速度和行驶轨迹不变(即假定驾驶员不采取任何避险措施),从冲突发生到预计碰撞所需要的时间。以追尾冲突为例(见图2),假设时刻后车速度大于前车速度,两车处于相同车道且行驶方向相同,其车辆间空间距离为,若两车保持时刻的行驶状态,则可以得到经过的时间长度,在预计的时刻两车会发生追尾碰撞。
极值理论(Extreme value theory,EVT)是统计学和概率论的一个分支,其核心在于探索样本数据中极端数值(极大值或极小值)所遵循的概率统计规律。相比于经典统计学,极值理论主要聚焦于随机样本中异常值的分布特性,揭示一般事件分布背后极端事件的统计规律。因此,极值理论提供了更为精确和有针对性的极端事件分析工具。
为充分对比晴天和雨天冲突严重程度影响因素的异同,针对晴天和雨天TTC数据分别建模。在模型训练过程中,首先,将晴天TTC数据和雨天TTC数据分别按照7∶3的比例分为训练集和测试集。然后,在训练集上使用递归特征消除(Recursive feature elimination,RFE)进行特征筛选,以LightGBM为基模型,逐步剔除对曲线下面积(Area under the curve,AUC)值贡献最小的特征,最终保留关键特征子集。在筛选后的特征空间上,使用随机搜索算法进行LightGBM模型超参数调优。针对每个模型,进行200次随机搜索,每次搜索时均使用5折交叉验证。最后,使用测试集数据检验模型预测效果。获取的各模型最优超参数如表4所示。
MuSong-cheng. Research on crash prediction model based on extreme statistics of traffic conflicts[D]. Weihai: School of Automotive Engineering, Harbin Institute of Technology, 2022.
[6]
姜雪. 城市快速路交织区交通冲突机理研究[D]. 大连: 大连交通大学交通工程学院, 2021.
[7]
JiangXue. Study on traffic conflict mechanism in urban expressway weaving areas[D]. Dalian: School of Traffic Engineering, Dalian Jiaotong University,2021.
[8]
YuanR, Abdel-AtyM, XiangQ. A study on diversion behavior in weaving segments: individualized traffic conflict prediction and causal mechanism analysis[J]. Accident Analysis & Prevention, 2024, 205: No.107681.
LiJia-shuo, ZhengZhang-ji, GuXin, et al. An analysis of the mechanism of traffic conflicts considering risky lane-changing behavior in weaving sections of expressways[J]. Journal of Transport Information and Safety, 2023, 41(3): 1-11.
[11]
马菲. 基于轨迹数据的快速路交织区交通风险评估[D]. 济南: 山东大学齐鲁交通学院, 2024.
[12]
MaFei. Risk assessment of expressway weaving sections based on trajectory data[D]. Jinan: School of Qilu Transportation, Shandong University, 2024.
LuQi-hui. Research on safety evaluation method of urban expressway weaving areas based on traffic conflict prediction model[D]. Nanjing: School of Transportation, Southeast University, 2024.
[15]
OnelcinP, AlverY. A new lane change index for lane change conflicts at weaving segments[J]. Traffic Injury Prevention, 2023, 24(7): 559-566.
[16]
XiaY, QinY, LiX, et al. Risk identification and conflict prediction from videos based on TTC-ML of a multi-lane weaving area[J]. Sustainability, 2022, 14: No.4620.
ZhaoTao, ZhangNing, WangXiao-chao, et al. A traffic conflict prediction method for merging areas based on trajectory prediction with graph neural network[J]. Journal of Shandong University (Engineering Science), 2024, 54(2): 36-46.
[19]
ZhengO, Abdel-AtyM, YueL, et al. CitySim: a drone-based vehicle trajectory dataset for safety oriented research and digital twins[J]. Transportation Research Record, 2023, 2678(4): 606-621.
[20]
史道济. 实用极值统计方法[M]. 天津: 天津科学技术出版社, 2006.
[21]
LundbergS M, LeeS I. A unified approach to interpreting model predictions[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, CA, USA, 2017: 4768-4777.
WenHui-ying, HeZi-qi, LiQiu-ling, et al. Traffic conflict prediction and influencing factors analysis of truck lane change on expressway[J]. Journal of Jilin University (Engineering and Technology Edition), 2024, 54(10): 2827-2836.