基于典型工况和GMM算法的驾驶风格辨识

郭智刚，闫立冰，张红倩，申宗

潍柴动力股份有限公司，山东 潍坊  261061

摘要：为提高车辆控制策略中驾驶风格信息的利用率、燃油经济性及行车安全性，在大量车辆行驶数据的基础上，应用K-Means算法提取车辆行驶典型工况，采用高斯混合模型（Gaussian mixture model，GMM）对驾驶风格进行识别，利用轮廓系数评价基于典型工况GMM算法和普通GMM算法的识别效果。结果表明：采用K-Means算法，以车速均值、车速标准差、怠速占比3个参数可有效提取拥堵、市区、高速3种工况；设计的基于典型工况的GMM算法可以将驾驶风格区分为稳健型、普通型、激进型3类，不同驾驶风格的区分度较好，识别效果优于普通GMM算法。

关键词：驾驶风格；K-Means；GMM；工况提取

Driving style identification based on typical working conditions and GMM algorithm

GUO Zhigang, YAN Libing, ZHANG Hongqian, SHEN Zong

Weichai Power Co., Ltd., Weifang 261061, China

Abstract: To improve the utilization rate of driving style information in vehicle control strategies, fuel economy, and driving safety, K-Means algorithm is applied to extract typical driving conditions based on a large amount of vehicle driving data.Gaussian mixture model (GMM) is used to recognize driving style, and contour coefficients are used to evaluate the recognition effect of GMM algorithm based on typical driving conditions and ordinary GMM algorithm. The results show that using the K-Means algorithm, three parameters including mean vehicle speed, standard deviation of vehicle speed, and idle ratio can effectively extract three typical working conditions: congestion condition, urban area condition, and high-speed condition. The GMM algorithm designed based on typical working conditions can distinguish driving styles into three categories: robust, normal, and aggressive. The differentiation of different driving styles is good, and the recognition effect is better than that of ordinary GMM algorithms.

Keywords: driving style; K-Means; GMM; condition extraction