Abstract: To address the deficiencies in existing research regarding vehicle car-following behavior and risk prediction under the combined effect of different environmental factors, this study relies on natural driving data collected from two-lane highways in Norway to systematically evaluate the impacts of precipitation type, precipitation intensity, road surface conditions, and lighting conditions on microscopic driving behavior. It establishes a four-level car-following risk classification system with time-to-collision as the core indicator, and employs multiple machine learning models for risk prediction to explore the improvement effect of environmental features on model performance. The results show that with the deterioration of environmental factors, drivers generally adopt a conservative strategy of reducing vehicle speed; when visibility decreases, drivers shorten the car-following distance to ensure that the preceding vehicle remains within the visual range; and the introduction of various environmental features effectively improves model performance, with a particularly significant enhancement observed in rainy and snowy weather. This study provides data support and model references for dynamic speed limit control and active safety warning in harsh environments.

Key words: adverse weather; road surface status; illumination conditions; car-following behavior; risk assessment

摘要： 针对现有研究对不同环境因素作用下的车辆跟驰行为与风险预测方面仍显不足的问题。 该研究基于挪威双车道公路的自然驾驶数据，系统评估了降水类型、降水强度、路面状态与 光照条件对微观驾驶行为的影响。构建以碰撞时间为核心的四级跟驰风险分类体系，使用多 种机器学习模型开展风险预测，以探究环境特征对模型性能的提升作用。结果表明，随着环 境因素的恶化，驾驶人普遍采取降低车速的保守策略；而在能见度下降时，驾驶人会降低跟 驰车距以确保前车在可视范围内；并且引入各类环境特征可有效提升模型性能，在雨雪天时 提升较为明显。该研究为恶劣环境的动态限速与主动安全预警提供了数据支持与模型参考。

关键词: 恶劣天气；路面状态；光照条件；跟车行为；风险评估