为研究中国交通场景下L4级自动驾驶汽车(autonomous vehicle,AV)交通事故的致因规律,并提升其运行安全性,该文首先收集了国内某AV示范区2021—2024年发生的60起典型AV事故案例,提取了涵盖环境、道路、车辆交互和管理等维度的22项致因要素;随后,针对数据样本量小、事故致因具有层级化和强耦合等特点,融合系统理论事故模型与过程(systems theoretic accident model and processes,STAMP)、Apriori关联规则和Bayes网络(Bayesian network,BN),构建了AV事故致因链分析模型,反向推理各类事故的关键致因链条,并分析了伤人事故节点的敏感度。结果表明:当追尾事故发生时,车辆间制动距离不足的后验概率提升了22%,该事故直接诱因是传统车辆驾驶人的危险操作行为,深层根源可追溯至驾驶人安全教育缺失;当伤人事故发生时,AV与非机动车碰撞的后验概率提升了18%,且碰撞类型多为侧碰事故;多条致因链共同揭示,AV系统尚未充分适应人类驾驶员的激进驾驶行为,安全员培训不足等管理因素虽不直接引发碰撞,但在事故演化过程中起基础性作用。该文研究结果验证了中国交通场景下,AV事故跨层级耦合的致因机制,可为制定区域化的安全管理策略和技术改进方案提供参考。
Abstract
[Objective] As the scale of autonomous vehicle (AV) testing has continued to expand, the overall frequency of associated traffic accidents has also increased. Identifying the underlying patterns of causation behind these accidents is vital for enhancing AV operational safety. However, existing research on AV accident causation has exhibited several critical limitations, including a restricted geographical scope of data sources, insufficient attention to the role of management-related factors, and inadequate analysis of the causal logic among accident variables. To investigate the causation patterns of level 4 (L4; AV) accidents in Chinese traffic scenarios, this study analyzed 60 typical AV accident cases that occurred between 2021 and 2024 in a domestic AV demonstration zone, systematically extracting 22 causal factors across multiple dimensions, including environment, road conditions, vehicle interaction, and management. [Methods] Given the small sample size and the hierarchical, strongly coupled nature of these accident factors, this study employed the systems theoretic accident model and processes (STAMP) framework to categorize the causal factors into four hierarchical levels: system, control, interaction, and outcome. Although the STAMP model provided a structured causal logic across these levels, it lacked the capacity to quantitatively evaluate the intrinsic relationships among accident variables. Consequently, this study integrated it with the Apriori association rule and Bayesian network methodologies. Through network topology construction and expectation-maximization parameter learning, this study established an AV accident causation chain analysis model. Leave-one-out cross-validation demonstrated that the model maintained robust discriminative ability under small-sample conditions. This study identified key causation chains for various accident categories using backward inference and performed sensitivity analysis on injury-related accident nodes. [Results] The results indicated that: (1) In rear-end collisions, the posterior probability of “insufficient braking distance” increased significantly by 22%; the direct trigger was typically a hazardous maneuver by a human driver, whereas the underlying root cause was often linked to a lack of human driver safety education. (2) In sideswipe accidents, the posterior probability of an AV encountering a lane-changing conflict increased significantly by 30%, with the primary triggers being delayed takeover by the AV safety operator or AV system perception failure; such incidents frequently occurred during peak traffic hours. (3) In injury accidents, the posterior probability of a collision with a non-motorist increased significantly by 18%, and the collision type was predominantly broadside. (4) In the sensitivity analysis of injury accidents, “human driver distraction” exhibited a sensitivity value of 0.018, and “inadequate AV safety operator training” showed a sensitivity value of 0.016, making them the most sensitive variables within their respective hierarchical levels. (5) Multiple causation chains collectively revealed that AV systems had not yet fully adapted to the aggressive driving behaviors of human drivers, while management factors, such as inadequate AV safety operator training, played a foundational role in accident evolution, despite not directly triggering collisions. [Conclusions] This study demonstrates that AV accident patterns are closely associated with the regional traffic environment. In Chinese urban settings, characteristics such as mixed traffic involving non-motorized vehicles and high peak-hour volumes are particularly influential. Management factors serve as deep-seated precursors in the evolution of AV accidents, exerting a hierarchical influence on crash outcomes. The proposed methodological framework provides a robust paradigm for analyzing causal propagation paths under small-sample constraints and offers a reference for formulating region-specific safety management strategies for AV operations.
关键词
自动驾驶汽车 /
事故致因分析 /
系统理论事故模型与过程 /
Bayes网络 /
关键致因链
Key words
autonomous vehicle /
accident causation analysis /
systems theoretic accident model and processes /
Bayesian network /
critical causation chains
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] National Highway Traffic Safety Administration, U.S. Department of Transportation. ADS incident report data[EB/OL]. (2025-06-16) [2025-12-16]. https://www.nhtsa.gov/laws-regulations/standing-general-order-crash-reporting.
[2] ABDEL-ATY M, DING S X. A matched case-control analysis of autonomous vs human-driven vehicle accidents[J]. Nature Communications, 2024, 15(1): 4931.
[3] XIANG J, WANG Z W, CHEN Y B, et al. Joint analysis of crash injury severities for autonomous and conventional vehicles in mixed traffic environments: Application of random parameter bivariate probit model[J]. Accident Analysis & Prevention, 2025, 215: 108017.
[4] LIU P, GUO Y Y, LIU P, et al. What can we learn from the AV crashes?—An association rule analysis for identifying the contributing risky factors[J]. Accident Analysis & Prevention, 2024, 199: 107492.
[5] 阎莹, 王玉莹, 周旋, 等. 基于混合模型的自动驾驶车辆事故严重程度影响因素分析[J]. 交通运输工程学报, 2025, 25(1): 184-196. YAN Y, WANG Y Y, ZHOU X, et al. Analysis of autonomous vehicle accident severity factors based on hybrid model[J]. Journal of Traffic and Transportation Engineering, 2025, 25(1): 184-196. (in Chinese)
[6] PAN R X, YUAN Q, CAO J M, et al. Sentence-resampled BERT- CRF model for autonomous vehicle crash causality analysis from large-scale accident narrative text data[J]. Accident Analysis & Prevention, 2025, 221: 108184.
[7] 王秀杰, 田浩浩. 自动驾驶事故数据挖掘与典型危险场景构建[J]. 交通运输研究, 2025, 11(2): 65-80. WANG X J, TIAN H H. Autonomous driving accident data mining and typical dangerous scenario construction[J]. Transport Research, 2025, 11(2): 65-80. (in Chinese)
[8] TAMAKLOE R, KHORASANI M, DAS S, et al. Pattern recognition in crash clusters involving vehicles with advanced driving technologies[J]. Accident Analysis & Prevention, 2025, 218: 108072.
[9] CHANNAMALLU S S, ALMASKATI D, KERMANSHACHI S, et al. Autonomous vehicle safety: An advanced bagging classifier technique for crash injury prediction[J]. Multimodal Transportation, 2025, 4(1): 100189.
[10] KUTELA B, DAS S, DADASHOVA B. Mining patterns of autonomous vehicle crashes involving vulnerable road users to understand the associated factors[J]. Accident Analysis & Prevention, 2022, 165: 106473.
[11] FU H L, YE S, FU X W, et al. New insights into factors affecting the severity of autonomous vehicle crashes from two sources of AV incident records[J]. Travel Behaviour and Society, 2025, 38: 100934.
[12] SUN Z W, LIN M, CHEN W T, et al. A case study of unavoidable accidents of autonomous vehicles[J]. Traffic Injury Prevention, 2024, 25(1): 8-13.
[13] 刘福聚, 王辰, 陈桂华, 等. 智能网联汽车交通事故研究[J]. 中国司法鉴定, 2024(5): 34-41. LIU F J, WANG C, CHEN G H, et al. The research on traffic accidents of intelligent and connected vehicles[J]. Chinese Journal of Forensic Sciences, 2024(5): 34-41. (in Chinese)
[14] 赵天亮, 王冰山, 台发强, 等. 基于改进DEMATEL-ISM-BN的人因视角下煤矿事故致因研究[J]. 安全与环境工程, 2025, 32(1): 91-99. ZHAO T L, WANG B S, TAI F Q, et al. Study on the causes of coal mine accidents from a human cause perspective based on improved DEMATEL-ISM-BN[J]. Safety and Environmental Engineering, 2025, 32(1): 91-99. (in Chinese)
[15] 蒋子怡, 肖仲明, 冯胤伟, 等. 基于HFACS-BN模型的船舶引航事故人为因素分析[J]. 大连海事大学学报, 2025, 51(1): 63- 70. JIANG Z Y, XIAO Z M, FENG Y W, et al. Human factor analysis of ship pilotage accidents based on HFACS-BN model[J]. Journal of Dalian Maritime University, 2025, 51(1): 63-70. (in Chinese)
[16] 霍小森, 杜爽, 谭琪麟, 等. 基于数据驱动贝叶斯网络的地铁施工事故致因差异化分析[J]. 铁道科学与工程学报, 2025, 22(4): 1802-1814. HUO X S, DU S, TAN Q L, et al. Differential analysis of causes of subway construction accidents based on data-driven Bayesian networks[J]. Journal of Railway Science and Engineering, 2025, 22(4): 1802-1814. (in Chinese)
[17] 胡越宁, 赵丹. 基于事故序列的自动驾驶汽车事故场景提取与分析[J]. 科学技术与工程, 2023, 23(11): 4908-4916. HU Y N, ZHAO D. Extraction and analysis of autonomous vehicle crash scenarios based on crash sequences[J]. Science Technology and Engineering, 2023, 23(11): 4908-4916. (in Chinese)
[18] 傅贵, 韩梦, 贾琳, 等. 事故致因理论及其发展综述[J]. 安全, 2023, 44(3): 1-8. FU G, HAN M, JIA L, et al. A review on accident cause theory and its development[J]. Safety & Security, 2023, 44(3): 1-8. (in Chinese)
[19] 张江石, 李泳暾, 吴静茹, 等. 煤矿事故原因智能分析方法研究与应用[J]. 清华大学学报(自然科学版), 2025, 65(3): 555-568. ZHANG J S, LI Y T, WU J R, et al. Research and application of intelligent methods for analyzing the causes of coal mining accidents[J]. Journal of Tsinghua University (Science and Technology), 2025, 65(3): 555-568. (in Chinese)
[20] WANG X S, ZHOU Y R, FILTNESS A, et al. Applying human factors analysis and classification system for commercial vehicles crashes investigation and critical failure routes analysis [J]. Accident Analysis & Prevention, 2025, 217: 108047.
[21] 郑世博, 樊运晓, 李振明. 基于HFACS的货车交通事故致因研究[J]. 安全与环境工程, 2020, 27(6): 133-139. ZHENG S B, FAN Y X, LI Z M. Causal factors of truck traffic accidents based on HFACS model[J]. Safety and Environmental Engineering, 2020, 27(6): 133-139. (in Chinese)
[22] 刘倩, 王雪松. 交叉口自动驾驶车辆事故前场景生成与致因分析[J]. 中国公路学报, 2024, 37(4): 297-309. LIU Q, WANG X S. Pre-crash scenario generation and causation analysis for autonomous vehicle crashes at intersections[J]. China Journal of Highway and Transport, 2024, 37(4): 297-309. (in Chinese)
[23] 胡雁宾, 周文辉. 辅助驾驶汽车交通事故综合因素调查方法研究[J]. 中国安全生产科学技术, 2024, 20(9): 181-187. HU Y B, ZHOU W H. Research on investigation methods for comprehensive factors of traffic accidents of assisted driving vehicles[J]. Journal of Safety Science and Technology, 2024, 20(9): 181-187. (in Chinese)
[24] WANG L Y, GRIFFIN M, HE J H, et al. Revealing the components of the risk profile of miners' unsafe behaviors: An integration of data-driven and theory-driven approaches[J]. Safety Science, 2025, 192: 107004.
[25] LIU R L, CHENG W M, YU Y B, et al. Human factors analysis of major coal mine accidents in China based on the HFACS-CM model and AHP method[J]. International Journal of Industrial Ergonomics, 2018, 68: 270-279.
[26] YANG J F, WANG P C, LIU X Y, et al. Analysis on causes of chemical industry accident from 2015 to 2020 in Chinese mainland: A complex network theory approach[J]. Journal of Loss Prevention in the Process Industries, 2023, 83: 105061.
[27] 田延飞, 乔慧, 滑林, 等. 海上船舶碰撞贝叶斯网络模型及其应用[J]. 海军工程大学学报, 2023, 35(6): 28-33. TIAN Y F, QIAO H, HUA L, et al. Bayesian network model for ship collisions at sea and its applications[J]. Journal of Naval University of Engineering, 2023, 35(6): 28-33. (in Chinese)
[28] 牛毅, 李振明, 樊运晓. 基于数据挖掘的高速公路货车交通事故影响因素关联分析研究[J]. 安全与环境工程, 2020, 27(4): 180-188. NIU Y, LI Z M, FAN Y X. Correlation analysis of influencing factors of truck traffic accidents on expressways[J]. Safety and Environmental Engineering, 2020, 27(4): 180-188. (in Chinese)
[29] 国汉君, 康荣学, 王勇, 等. 基于数据驱动贝叶斯网络模型的起重作业关键致因要素分析[J]. 中国安全生产科学技术, 2025, 21(4): 86-93. GUO H J, KANG R X, WANG Y, et al. Analysis on key causative factors of lifting operation based on data-driven Bayesian network model[J]. Journal of Safety Science and Technology, 2025, 21(4): 86-93. (in Chinese)
[30] 孙少楠, 李博宇, 聂相田. 基于贝叶斯网络的水轮机故障诊断研究[J]. 水电能源科学, 2023, 41(3): 190-194. SUN S N, LI B Y, NIE X T. Research on fault diagnosis of hydraulic turbine based on Bayesian network[J]. Water Resources and Power, 2023, 41(3): 190-194. (in Chinese)
基金
国家重点研发计划项目(2023YFB4302703);中央高校基本科研业务费专项资金项目(2026JKF03ZK01)