Unintended stopping conflict risk prediction for high-level autonomous vehicles based on CatBoost and SHAP

doi:10.3969/j.issn.1674-8484.2025.01.017

Abstract

Abstract:

For the traffic conflicts and environmental impacts arising from unanticipated parking of high-level autonomous vehicles, existing research lacks the capture and interpretability assessment of risk characteristic interactions. This study proposed a risk prediction and interpretation model utilizing Categorical Boosting (CatBoost) and SHapley Additive exPlanations (SHAP), constructing a conflict risk prediction framework by analyzing takeover events in urban centers, residential areas, and suburban transportation networks. The results show that the number of takeovers reaches 161, 227, and 164 instances in urban centers, residential areas, and suburbs, respectively, with the maximum single-road-section takeover frequencies being 11, 11, and 16 times. The prediction accuracy of the model exceeds 93%. SHAP analysis reveales that the relative speed and position between preceding and following vehicles significantly influence collision risk. These findings hold important implications for enhancing the reliability and safety of autonomous vehicles.

Key words: conflict risk, traffic emission, high-level autonomous vehicles, Categorical Boosting (CatBoost), Shapley Additive exPlanations (SHAP)

CLC Number:

U491.31

LIU Qingchao, WANG Ruihai, CAI Yingfeng, WANG Hai, CHEN Long. Unintended stopping conflict risk prediction for high-level autonomous vehicles based on CatBoost and SHAP[J]. Journal of Automotive Safety and Energy, 2025, 16(1): 170-180.

Figures/Tables 12

References 23

[1]	YUAN Chen, LI Ye, HAUNG Helai, et al. Using traffic flow characteristics to predict real-time conflict risk: A novel method for trajectory data analysis[J]. *Anal Method Accid Re*, 2022, 35: 100217-100232.
[2]	HU Yuping, LI Ye, YUAN Chen, et al. Modeling conflict risk with real-time traffic data for road safety assessment: A copula-based joint approach[J]. *Transport Safe Enviro*, 2022, 4: 17-25.
[3]	袁泉, 高岩, 裘晨璐. 基于人-机-环境因素的未来交通事故风险研究[J]. 系统仿真学报, 2019, 31(3): 566-574. doi: 10.16182/j.issn1004731x.joss.17WJP-012
	YUAN Quan, GAO Yan, QIU Chenlu. Research on future traffic accident risk based on human-machine-environment factors[J]. J Syst Simul, 2019, 31(3): 556-574. (in Chinese)
[4]	LI Pei, Abdel-Aty M, YUAN Jinghui. Real-time crash risk prediction on arterials based on LSTM-CNN[J]. *Accid Anal Pre, 2020, 135*: 105371-105379.
[5]	Halim Z, Kalsoom R, Bashir S, et al. Artificial intelligence techniques for driving safety and vehicle crash prediction[J]. *Artif Intel Re*, 2016, 46: 351-387.
[6]	LIU Qian, WANG Xuesong, WU Xiangbin, et al. Crash comparison of autonomous and conventional vehicles using pre-crash scenario typology[J]. *Accid Anal Pre, 2021, 159*(4): 106281-106293.
[7]	Mahdinia I, Mohammadnazar A, Arvin R, et al. Integration of automated vehicles in mixed traffic: Evaluating changes in performance of following human-driven vehicles[J]. *Accid Anal Pre, 2021, 152*: 106006-106017.
[8]	Fitzpatrick C, Rakasi S, Knodler M. An investigation of the speeding-related crash designation through crash narrative reviews sampled via logistic regression[J]. *Accid Anal Pre*, 2017, 98: 57-63.
[9]	Kim J, Kum D. Collision risk assessment algorithm via lane-based probabilistic motion prediction of surrounding vehicles[J]. *IEEE Trans Intel Transport Sys*, 2018, 19(9): 2965-2976.
[10]	GUO Miao, ZHAO Xiaohua, YAO Ying, et al. A study of freeway crash risk prediction and interpretation based on risky driving behavior and traffic flow data[J]. *Accid Anal Pre, 2021, 160*: 106328-106338.
[11]	LIU Qingchao, LI Chun, JIANG Haobin, et al. Transfer learning-based highway crash risk evaluation considering manifold characteristics of traffic flow[J]. *Accid Anal Pre, 2022, 168*: 106598-106599.
[12]	Matthias S. A multivariate analysis of environmental effects on road accident occurrence using a balanced bagging approach[J]. *Accid Anal Pre, 2019, 136*: 105398-105409.
[13]	JING Mengxia, LU Guangquan, CHEN Fachegn, et al. Modeling takeover behavior in level 3 automated driving via a structural equation model: Considering the mediating role of trust[J]. *Accid Anal Pre, 2021, 157*: 106156-106170.
[14]	LI Pei, Abdel-Aty M, YUAN Jinghui. Real-time crash risk prediction on arterials based on LSTM-CNN[J]. *Accid Anal Pre, 2019, 135*: 105379-105387.
[15]	LI Guofa, YANG Yifan, ZHANG Tingru, et al. Risk assessment based collision avoidance decision-making for autonomous vehicles in multi-scenarios[J]. Transport Res Part C: Emerg Tech, 2021, 122: 102820-102836.
[16]	LU Qinglong, YANG Kui, Antoniou C. Crash risk analysis for the mixed traffic flow with human-driven and connected and autonomous vehicles[C]// Proc 2021 IEEE Int'l Intel Transport Syst Conf (ITSC). Indianapolis, USA, 2021: 19-22.
[17]	ZHU Jie, Tasic I. Safety analysis of freeway on-ramp merging with the presence of autonomous vehicles[J]. Accid Anal Prev, 2021, 152: 105966-105980.
[18]	刘擎超, 徐天宇, 熊晓夏, et al. 城市环形交叉口智能汽车接管过程事故风险预测[J]. 中国安全科学学报, 2022, 32(12): 150-157. doi: 10.16265/j.cnki.issn1003-3033.2022.12.2650
	LIU Qingchao, XU Tianyu, XIONG Xiaoxia, et al. Accident risk prediction for smart car takeover process at urban roundabouts[J]. *Chin Safe Sci* , 2022, 32(12): 150-157. (in Chinese)
[19]	朱小平, 张丽英, 刘静, 等. 基于XGBoost的自动驾驶汽车事故风险预测研究[J]. 时代汽车, 2024, 6: 187-189.
	ZHU Xiaoping, ZHANG Liying, LIU Jing, et al. Research on accident risk prediction of self-driving cars based on XGBoost[J]. Autom Time, 2024, 6: 187-189. (in Chinese)
[20]	马小翔, 陈丰, 张霖. 预期接管场景下接管绩效及接管风险研究[J]. 中国公路学报, 2022, 35(1): 159-168. doi: 10.19721/j.cnki.1001-7372.2022.01.014
	MA Xiaoxiang, CHEN Feng, ZHANG Lin. A study of takeover performance and takeover risk in anticipation of takeover scenarios[J]. *Chin J Highw Transpor*, 2022, 35(1): 159-168. (in Chinese)
[21]	Milanés V, Shladover S, Spring J, et al. Cooperative adaptive cruise control in real traffic situations[J]. *IEEE Trans Intel Transport Sys*, 2013, 15(1): 296-305.
[22]	Kesting A, Treiber M, Helbing D. Enhanced intelligent driver model to access the impact of driving strategies on traffic capacity[J]. Philosoph Trans Roy Soc A: Mathem, Phys Engi Sci, 2010, 368(1928): 4585-4605.
[23]	Prokhorenkova L, Gusev G, Vorobev A, et al. CatBoost: Unbiased boosting with categorical features[J/OL]. (2017-01-28) https://arxiv.org/abs/1706.09516.

驾驶模型	CACC		IDM
驾驶模型	实际值	典型值	实际值	典型值
颜色	蓝色	-	绿色	-
最大加速度/ (m·s^-2)	2.5	2.0~3.5	3.5	1.0~3.5
最大减速度/ (m·s^-2)	6.0	4.0~6.0	8.0	7.0~9.0
安全车头时距/ s	1.0	0.8~1.5	1.0	1.0-2.0
最大速度/ (m·s^-1)	30.0	20.0~30.0	33.0	25.0~33.0
最小间距/ m	2.0	1.0~2.5	2.0	2.0~4.0
间距控制系数	0.2	0.2~0.3	-	-

驾驶模型	CACC		IDM
驾驶模型	实际值	典型值	实际值	典型值
颜色	蓝色	-	绿色	-
最大加速度/ (m·s^-2)	2.5	2.0~3.5	3.5	1.0~3.5
最大减速度/ (m·s^-2)	6.0	4.0~6.0	8.0	7.0~9.0
安全车头时距/ s	1.0	0.8~1.5	1.0	1.0-2.0
最大速度/ (m·s^-1)	30.0	20.0~30.0	33.0	25.0~33.0
最小间距/ m	2.0	1.0~2.5	2.0	2.0~4.0
间距控制系数	0.2	0.2~0.3	-	-

	预测为正 Positive	预测为负 Negative
真实为正True	真正 True Positive (TP)	假负 False Negative (FN)
真实为负False	假正 False Positive (FP)	真负 True Negative (TN)

	预测为正 Positive	预测为负 Negative
真实为正True	真正 True Positive (TP)	假负 False Negative (FN)
真实为负False	假正 False Positive (FP)	真负 True Negative (TN)

高级别自动驾驶车辆比例	m_co / kg		m_co2 / t		m_HC / kg		m_NO_x / kg		m_PM_x / g
高级别自动驾驶车辆比例	住宅区	郊区	住宅区	郊区	住宅区	郊区	住宅区	郊区	住宅区	郊区
0.1	296.21	251.59	12.75	10.39	1.66	1.40	5.31	4.31	280.5	226.5
0.2	271.37	232.88	11.83	9.84	1.52	1.30	4.93	4.08	259.5	213.8
0.3	249.90	219.16	10.76	9.29	1.40	1.22	4.49	3.85	236.3	201.7
0.4	218.83	206.80	9.86	8.59	1.23	1.15	4.10	3.56	215.1	186.7
0.5	202.55	179.79	9.15	7.77	1.14	1.01	3.80	3.21	199.6	167.7