驾驶风格聚类与识别研究

Research of Clustering and Recognition for Driving Style

为实现驾驶人驾驶风格的准确聚类和有效识别,以NGSIM数据为依据,提取包含换道行为的16 s车辆行驶轨迹,选取可以全方位表征驾驶人驾驶特性的评价指标并借助因子分析法实现指标的降维;在此基础上选用k-means聚类算法对所得样本数据进行聚类分析,把驾驶风格分为谨慎型、普通型和激进型;最后对比了SVM和ANN驾驶风格识别模型。结果显示:车辆行驶速度、加速度和横向速度更能表征驾驶人驾驶风格;SVM模型和ANN模型在降维前后数据中的识别精度都能达到90%以上,而SVM模型的识别精度更高,说明SVM驾驶风格识别模型在小样本上的有效性。

In order to achieve accurate clustering and effective recognition of driving style,the 16 s complete vehicle trajectory data including lane changing behavior was extracted from NGSIM data.The driving style evaluation indices which can comprehensively characterize the driver’s driving characteristics were selected,and the dimension of the evaluation indices was reduced by factor analysis algorithm.Based on that,the k-means clustering algorithm was used to cluster the sample data,and the driving style was classified three categories:calm,moderate and aggressive.Finally,SVM and ANN driving style recognition model were compared.The results show that for driving style clustering,vehicle speed,acceleration and lateral speed can make a good representation of driving style;for driving style recognition,the recognition accuracy of SVM model and ANN model in the data before and after dimension reduction can reach higher than 90%,and the recognition accuracy of SVM model is higher than the recognition accuracy of ANN model,which indicates that SVM driving style recognition model is effective in small sample.