Abstract: With the rapid development of the automotive industry, the quantification and classification of driver behavior characteristics have become key factors in improving testing efficiency and driving safety. This paper proposes a driver style identification model that relies solely on raw vehicle dynamics data from the onboard controller area network (CAN) bus, without the need for additional sensors. By collecting key parameters such as vehicle speed, throttle opening, longitudinal and lateral acceleration, steering angle, and combining descriptive statistical analysis with interval distribution methods, driver behavior is clustered under three typical driving conditions: urban, highway, and mountain roads. The study results indicate that aggressive drivers have significantly higher acceleration standard deviation, proportion of extreme maneuvers, and steering frequency than stable and cautious drivers. By constructing a two-dimensional threshold system based on standard deviation and interval proportions, automated classification of driver styles can be achieved, with high accuracy. The model has been validated with actual fleet data, providing practical technical support for driver selection, safety warnings, and driving behavior interventions. This paper also points out the current model's limitations in complex road conditions and multi-vehicle adaptability, and suggests future improvements by introducing weight adaptive mechanisms and automatic condition recognition modules.

Key words: driver style; CAN bus; acceleration; fleet management

摘要： 随着汽车行业快速发展，驾驶员行为特征的量化与分类成为提升试验效率与行车安 全的关键环节。文章提出一种无需额外传感器、仅依赖车载控制器局域网（CAN）总线原始 动力学数据的驾驶员风格识别模型。通过采集车速、油门开度、纵向与横向加速度、方向盘 转角等关键参数，结合描述性统计分析与区间分布方法，在城市、高速、山路三种典型工况 下对驾驶员行为进行聚类分析。结果表明，激进型驾驶员在加速度标准差、极端区间占比及 转向频率上显著高于平稳型与谨慎型驾驶员；通过构建基于标准差与区间占比的双维度判定 阈值体系，可实现对驾驶员风格的自动化分类，准确率高。模型已在实际车队管理中验证， 为驾驶员选取、安全预警与驾驶行为干预提供了可落地的技术支撑。文章亦指出当前模型在 复杂路况与多车型适配性方面的局限，并提出未来需引入权重自适应机制与工况自动识别模 块的改进方向。

关键词: 驾驶风格；CAN 总线；加速度；车队管理