Abstract: The drivers driving style of electric city bus has an important impact on the safety of urban public transport. In order to identify the driving style of the driver, a feature extraction network composed of convolutional layer and pooling layers was used to deeply fuse and automatically extract multi-channel feature data, by using the onboard CAN data in the process of the drivers natural driving electric city buses entering stations. It seamlessly output to fully connected neural networks for inbound driving style recognition, and an electric city bus driver inbound driving style recognition model was constructed. The research shows that the proposed model can effectively integrate the corresponding time series data of driving behavior and vehicle operation status during the entry process, and automatically extract higher-order features of driving behavior, achieving effective recognition of the driving style of electric city bus drivers entering the station, with an accuracy rate of 98.2%. The research results help to identify drivers with aggressive driving styles, so as to carry out targeted driving safety education and thereby reduce driver-induced electric city bus traffic accidents.

Key words: traffic and transportation engineering; driving style; convolution neural network (CNN); CAN data; electric city bus; driving safety

摘要： 电动城市客车驾驶员的驾驶风格对城市公共交通安全有重要影响。为识别驾驶员的驾驶风格，利用驾驶员自然驾驶电动城市客车进站过程的车载CAN数据，采用由卷积层和池化层组成的特征提取网络对多通道特征数据进行信息深度融合和特征自动提取，无缝输出给全连接神经网络进行进站驾驶风格识别，构建出电动城市客车驾驶员进站驾驶风格识别模型。研究表明，采用建立的模型可以有效融合进站过程中驾驶行为和车辆运行状态对应时序数据并自动提取驾驶行为高阶特征，实现电动城市客车驾驶员进站驾驶风格的有效识别，准确率达到98.2%。研究成果有助于识别出激进型驾驶风格的驾驶员，以便针对性开展驾驶安全教育，进而降低驾驶员致因的电动城市客车交通事故。

关键词: 交通运输工程；驾驶风格；卷积神经网络（CNN）；CAN数据；电动城市客车；驾驶安全