Abstract: To explore the characteristics and driving style identification method of different kinds of drivers in steady car-following situations, twenty drivers were recruited to carry out real-vehicle driving tests. The indicators such as self-driving speed, following distance, and following time interval under natural driving conditions were collected, and the extraction rules of steady car-following events were determined based on radar data, etc. The distribution pattern and following characteristics of driving behavior indicators under stable following conditions were analyzed by coupling analysis. The following distance, following time interval together with opening degree of accelerator pedal were selected as clustering indicators, and K-means clustering algorithm was used to perform clustering analysis on stable following events in section 301. According to the frequency and proportion of different style types, drivers were divided into three types of styles (conservative, general, and radical). Finally, the clustering results were verified by the CART decision tree method, and the transfer characteristics of driving style under long-term steady car-following states were furtherly analyzed. The research results show that the following distance and the opening degree of accelerator pedal tend to increase as the self-driving speed increases, and have significant differences among different speed ranges. There is no significant change in the mean value of car-following time interval with different speed ranges, and it is stably distributed between 2.57~2.72 s. The CART decision tree verifies that the overall coincidence rate of driving style clustering results is 99.7%. There are significant differences in speed, opening degree of accelerator pedal, the following distance and following time interval for drivers with different types. Over time, conservative drivers tend to be more conservative, aggressive drivers tend to be more aggressive, and average drivers are relatively stable. The research results can provide technical support for the formulation of car-following control strategies and parameters of high-level automatic driving systems.

Key words: traffic engineering; steady car-following states; driving style; K-means clustering; naturalistic driving; driving behavior analysis; transfer characteristics

摘要： 为探究高速工况稳定跟车状态下不同类型驾驶人的跟车特性及驾驶风格识别方法，选取20名驾驶人开展实车驾驶试验，采集自然驾驶状态下的自车速度、跟车间距、跟车时距等指标，基于雷达数据等确定稳定跟车事件提取规则。通过耦合分析稳定跟车状态下的驾驶行为指标分布规律及跟车特性，选取跟车间距、跟车时距及加速踏板开度为聚类指标，使用K均值聚类算法对301段稳定跟车事件进行聚类分析，并根据不同风格类型出现的频数及所占比例将驾驶人划分为3种风格类型（保守型、一般型、激进型）。最后通过CART决策树方法对聚类结果进行验证，进一步分析长时间稳定跟车状态下驾驶风格的迁移特性。研究结果表明：随自车速度增大，跟车间距与加速踏板开度亦呈现增大趋势，且在不同速度区间下均具有显著性差异。不同速度区间下的跟车时距均值无明显变化，稳定分布于2.57~2.72 s。CART决策树验证驾驶风格聚类划分结果总体吻合率达99.7%。不同风格驾驶人的车速与油门踏板开度、跟车间距与跟车时距均存在显著性差异。随时间推移，保守驾驶人更加趋于保守，激进驾驶人更加趋于激进，一般驾驶人则相对较为稳定。研究结果可为高级别自动驾驶系统跟车控制策略及参数的制定提供技术支持。

关键词: 交通工程；稳定跟车状态；驾驶风格；K均值聚类；自然驾驶；驾驶行为分析；迁移特性