Abstract: To address the problem of assessing driving risk levels for intra-city logistics distribution vehicles, extreme value theory and floating car data were employed to carry out the study. Firstly, six types of risky and safe driving behaviors were extracted from distribution data, and their velocity safety entropy was calculated. Secondly, a generalized Pareto model for velocity safety entropy was established based on extreme value theory. The threshold (0.189 8~0.199 8) was determined by combining the graphical method with sliding time window method, and the model parameters were solved via Bayesian estimation. The proposed model could be used for predicting and assessing driving risk levels. The reproducibility level theory was adopted to validate the effectiveness of the proposed model. The research results show that the approximate linear goodness-of-fit for the minimum threshold model and the maximum threshold model are 0.827 3 and 0.855 9, respectively, with the maximum threshold model demonstrating superior predictive performance for risky driving behaviors. The proposed model establishes a correlation between non-risky driving behaviors and risk levels, enabling the prediction and assessment of risk levels based on non-risky behavior data, thereby expanding assessment methods. It provides support for establishing an operational risk assessment and early warning system for intra-city distribution vehicles, discriminating risky driving behaviors and enhancing driving safety.

Key words: traffic engineering; intra-city logistics; risky driving behavior; prediction evaluation; extreme value theory

摘要： 针对同城物流配送车辆驾驶风险水平评估问题，基于极值理论和浮动车数据展开了研究。利用配送数据提取6类风险与安全驾驶行为，并计算其速度安全熵；基于极值理论建立速度安全熵的广义帕累托模型，结合图解法与滑动时间窗法确定阈值(0.189 8~0.199 8)，并通过贝叶斯估计求解模型参数，该模型可用于驾驶风险水平的预测与评估；采用重现水平理论验证模型有效性。研究结果表明：最小与最大阈值模型的近似线性拟合优度分别为0.827 3和0.855 9，且最大阈值模型的风险驾驶行为预测性能更优。该模型建立了非风险驾驶行为与风险水平的关联，支持基于非风险行为数据预测评估风险水平，拓展了评估方法；对建立同城配送车辆运行风险评估预警系统、界定风险驾驶行为和提升驾驶安全性具有支持作用。

关键词: 交通工程；同城物流；风险驾驶行为；预测评估；极值理论