交通运输系统工程与信息 ›› 2026, Vol. 26 ›› Issue (2): 203-211.DOI: 10.16097/j.cnki.1009-6744.2026.02.019

• 系统工程理论与方法 • 上一篇    下一篇

城市道路交叉口人车冲突极值模型与风险评估

张文会* ,吕家乐,陈德启,孟凤威   

  1. 东北林业大学,土木与交通学院,哈尔滨150040
  • 收稿日期:2025-11-09 修回日期:2025-12-15 接受日期:2026-01-12 出版日期:2026-04-25 发布日期:2026-04-20
  • 作者简介:张文会(1978—),男,黑龙江哈尔滨人,教授,博士。
  • 基金资助:
    国家自然科学基金(52572369)。

Extreme Value Model and Risk Assessment for Pedestrian-Vehicle Conflicts at Urban Road Intersections

ZHANG Wenhui*, LV Jiale, CHEN Deqi, MENG Fengwei   

  1. School of Civil Engineering and Transportation, Northeast Forestry University, Harbin 150040, China
  • Received:2025-11-09 Revised:2025-12-15 Accepted:2026-01-12 Online:2026-04-25 Published:2026-04-20
  • Supported by:
    National Natural Science Foundation of China (52572369)。

摘要: 为探究城市信号交叉口人车冲突的形成机制,并评估碰撞风险,以后侵入时间(PET)作为冲突替代指标,构建基于贝叶斯极值理论的主动安全评估模型。本文利用实地采集的3个交叉口人车冲突数据,采用广义帕累托分布对PET尾部进行建模,并在道路结构与交通参与者行为特征中选用机动车道数量、人车之间最小间距、行人速度、机动车速度与过街行人数量作为协变量连接尺度参数,实现对冲突极值非稳定性和异质性的捕捉。分别构建稳态模型、非稳态全模型与非稳态显著模型,采用MCMC(Markov Chain Monte Carlo)方法进行贝叶斯参数估计,并基于DIC(Deviance Information Criterion)进行模型比较。结果显示,非稳态显著模型DIC值为(1811.92),低于稳态模型(1827.68)和非稳态全模型(1817.17),拟合与预测性能最优;模型能够定量反映特征变化对应的风险水平差异,在碰撞数据稀缺的情况下实现了尾部风险分析,并提供了一种可用于主动安全评估的量化方法,为交叉口安全管理与风险干预提供科学依据。

关键词: 城市交通, 事故预测, 极值理论, 信号交叉口, 人车冲突

Abstract: To explore the underlying mechanisms and evaluate the potential collision risks of pedestrian-vehicle conflicts at urban signalized intersections, this study adopts Post Encroachment Time (PET) as a surrogate safety indicator and establishes a proactive safety assessment framework grounded in Bayesian Extreme Value Theory. With the conflict data which collected from three typical signalized intersections, the tail behavior of PET is characterized by the Generalized Pareto Distribution. Multiple traffic and behavioral covariates including the number of motorized lanes, minimum pedestrian-vehicle distance, pedestrian speed, vehicle speed, and the number of crossing pedestrians are incorporated into the scale parameter to capture the non-stationary and heterogeneous characteristics of extreme conflicts. Three hierarchical models including a stationary model, a non-stationary full model, and a non-stationary significant model are constructed, and Bayesian parameter estimation is performed by using the MCMC (Markov Chain Monte Carlo) algorithm. Model comparison by the DIC (Deviance Information Criterion) reveals that the non-stationary significant model (DIC is 1 811.92) yields the best fitting and predictive performance, which outperforms both the stationary (DIC is 1 827.68) and non-stationary full (DIC is 1 817.17) models. The proposed model quantitatively captures variations in risk levels driven by different influencing factors and enables tail risk estimation under the limited crash data conditions. This approach provides a robust quantitative framework for proactive safety assessment and then delivers scientific insights to support intersection safety management and risk intervention strategies.

Key words: urban transportation, accident prediction, extreme value theory, signalized intersections, pedestrian and vehicle conflict

中图分类号: