Abstract: To ensure vehicle operation safety under freezing rain conditions and improve the capability of pavement condition identification and risk early warning for expressways under freezing rain hazards, this paper takes the Xihan Section of the Beijing-Kunming Expressway as the research object. By integrating road environmental characteristics, meteorological characteristics and pavement monitoring data, an XGBoost-based identification model for freezing rain hazards is constructed, in which pavement conditions are classified into five categories: dry, wet, snow-covered, icy, and freezing rain. Through integrated analysis of meteorological elements, road environment and pavement condition data, intelligent identification and prediction of pavement conditions under different hazardous weather conditions are realized. The results show that the overall accuracy,weighted F1-score and area under curve (AUC) of the model reach 0.910 2, 0.912 0 and 0.932 6 respectively, demonstrating favorable performance in identifying freezing rain conditions. Analysis indicates that temperature, humidity, altitude and precipitation are the primary factors affecting freezing rain formation. High-risk areas are mainly concentrated in sections such as high-altitude bridges, long steep longitudinal slopes and tunnel exits, with higher risks from nighttime to early morning. The research findings can provide a reference for freezing rain hazard monitoring, risk early warning for key road sections and traffic safety management of expressways.

Key words: expressways; freezing rain hazards; pavement condition identification; spatio-temporal risk prediction; machine learning

摘要： 为保障冻雨天气下车辆运行安全，提高高速公路冻雨灾害条件下路面状态识别与风险 预警能力，文章以京昆高速西汉段为研究对象，融合道路环境特征、气象特征及路面监测数 据，构建了基于 XGBoost 的冻雨灾害识别模型，并将路面状态划分为干燥、潮湿、积雪、结 冰和冻雨五类。通过对气象要素、道路环境及路面状态数据进行融合分析，实现了不同灾害 天气条件下路面状态的智能识别与预测。结果表明，模型总体准确率达到 0.910 2，加权 F1 值达到 0.912 0，曲线下面积（AUC）达到 0.932 6，对冻雨状态具有较好的识别能力。分析发 现，温度、湿度、海拔及降水量是影响冻雨形成的主要因素，高风险区域主要集中于高海拔 桥梁、长大纵坡及隧道出口等路段，且夜间至凌晨时段风险更高。研究结果可为高速公路冻 雨灾害监测、重点路段风险预警及交通安全管控提供参考。

关键词: 高速公路；冻雨灾害；路面状态识别；时空风险预测；机器学习