| 引用本文: | 吴瑶,占志彬,张宜知,周继彪,刘佩.智能网联车辆事故严重程度影响因素可解释性分析[J].哈尔滨工业大学学报,2026,58(7):97.DOI:10.11918/202412028 |
| Wu Yao,Zhan Zhibin,Zhang Yizhi,Zhou Jibiao,Liu Pei.Explainability analysis of influencing factors of accident severity of connected-automated vehicles[J].Journal of Harbin Institute of Technology,2026,58(7):97.DOI:10.11918/202412028 |
|
| 摘要: |
| 为了研究外界环境对智能网联车辆事故严重程度的相关影响因素,采用美国2016—2023年DMV官网公开的智能网联车辆事故数据资料。首先,使用Boruta方法对交通事故数据进行处理,确定时间、地点、限速、天气等14个维度为输入变量集。其次,构建CatBoost多分类模型,并利用TPE优化方法调整超参数,得到了相对优超参数组合下的评估模型,该拟合模型的平均AUC值达到了0.86。最后,引入了SHAP解释模型,揭示显著影响因素对事故严重程度的全局影响及贡献程度,并探究了变量之间的交互效应。结果表明:道路类型、季节、车辆运动状态和交通事故类型与事故严重程度密切相关;追尾事故、恶劣的天气条件和路况会增加严重交通事故的风险。研究结果为智能网联车辆的安全评估提供了TPE-CatBoost-SHAP分析框架,有助于评估和改进当前智能网联车辆的安全性能。 |
| 关键词: 智能交通 事故严重性 TPE-CatBoost-SHAP模型 智能网联车辆 机器学习 |
| DOI:10.11918/202412028 |
| 分类号:X703.1 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(52202382) |
|
| Explainability analysis of influencing factors of accident severity of connected-automated vehicles |
|
Wu Yao1,Zhan Zhibin1,Zhang Yizhi1,Zhou Jibiao2,Liu Pei 3
|
|
(1.School of Modern Posts, Nanjing University of Posts and Telecommunications, Nanjing 210003, China; 2.Ningbo High-grade Highway Construction Management Center, Ningbo 315100, Zhejiang, China; 3.School of Transportation,Southeast University,Nanjing 211189, China)
|
| Abstract: |
| To investigate the factors influencing the severity of accident in CAVs(connected-autonomous vehicles), this study utilizes publicly available accident data from the DMV(U.S. Department of Motor Vehicles) website between 2016 and 2023. The Boruta method was first applied to process the traffic accident data, identifying 14 input variables such as time, location, speed limit, and weather conditions. A CatBoost multi-class classification model was then developed, and hyperparameters were optimized using the TPE(Tree-structured Parzen Estimator ) method. The evaluation model achieved an average AUC(Area Under the Curve) of 0.86 under the optimal hyperparameter combination. Furthermore, the SHAP(SHapley Additive exPlanations explanation) model was utilized to reveal the global impact and contribution of significant factors to accident severity and explore the interactions between variables. The results show that road type, season, vehicle motion state, and accident type are strongly correlated with accident severity. Rear-end collisions, adverse weather, and poor road conditions increase the risk of severe accidents. This research provides a TPE-CatBoost-SHAP analysis framework for safety assessment in CAVs, offering insights into improving their safety performance. |
| Key words: intelligent transportation accident severity TPE-CatBoost-SHAP model connected-automated vehicles machine learning |