Abstract: Driving style is used to represent the behavior characteristics of drivers. It is of great value to develop automatic driving technology and formulate personalized driving strategies. Based on the vehicle driving state data in the American NGSIM database, this paper selects nine statistics such as the mean of the absolute value of transverse speed, the standard deviation of the absolute value of transverse speed and the standard deviation of the absolute value of longitudinal speed as the characteristic variables, and uses the dimension reduction algorithm of principal component analysis and K-means clustering algorithm to classify the driving style of the driving data, and divides the driving styles into conservative, general, radical three types. The data analysis shows that the statistics of the mean transverse speed, the mean longitudinal acceleration and the mean transverse and longitudinal impact of the conservative driver are the minimum of the three types, while the corresponding statistics of the radical driver is the maximum of the three, and the general driver is in the middle, which verifies the rationality of this clustering result.

Key words: Data processing; Feature selection; Driving style clustering; NGSIM database

摘要： 驾驶风格用于表征驾驶人的行为特性，对发展自动驾驶技术、制定个性化的驾驶策略 具有重要价值。文章基于美国 NGSIM 数据库中的车辆行驶状态数据，选取横向速度绝对值均 值、横向速度绝对值标准差、纵向速度绝对值的标准差等九个统计量作为特征变量，利用主 成分分析降维算法及 K-means 聚类算法对行驶数据进行驾驶风格分类研究。将驾驶风格分为 保守型、一般型及激进型三个类别，数据分析表明，保守型驾驶人的横向速度均值、纵向加 速度均值、横纵向冲击度均值等统计量均为三种类型中的最小值，而激进型驾驶人的对应统 计量为三者中的最大值，一般型驾驶人居中，验证了本次聚类结果的合理性。

关键词: 数据处理；特征选择；驾驶风格聚类；NGSIM 数据库