Abstract: A driving style recognition method for passenger drivers based on iForest and Biscting K-means model was proposed. In order to solve the problem that the clustering centroid seriously affected the clustering results in the Bisceting K-means model, the iForest was adopted to train the candidate set of clustering centers as the clustering centroid set. Through the investigation of 30 professional bus drivers on an intercity passenger line, the proposed model was verified by about 4 million bus driving data in more than 90 days under the straight road driving condition. The experiment shows that when the standard deviation of acceleration and overspeed tendency coefficient are used as clustering indexes, the driving styles of these 30 drivers cluster into three categories: cautious, ordinary and aggressive, among which 11 are cautious, 19 are ordinary and no aggressive.

Key words: traffic engineering, driving style, Bisecting K-means algorithm, iForest algorithm, big data

摘要： 提出了一种基于iForest+Biscting K-means模型的客运驾驶员驾驶风格辨识方法。该方法针对在Bisceting K-means模型中，聚类质心严重影响聚类结果问题，采用iForest模型训练聚类中心候选集作为聚类质心集的方法加以改进。通过考察某城际客运线路30位客车职业驾驶员，在直线道路行驶工况下，90余天约400万条客车行驶数据开展模型验证。试验表明：在加速度标准差和超速倾向系数作为聚类指标的情况下，客运驾驶员驾驶风格聚类为谨慎型、普通型和激进型3类，其中谨慎型11人，普通型19人，激进型0人。

关键词: 交通工程, 驾驶风格, Bisecting K-means算法, iForest算法, 大数据