关键词: 交通安全；驾驶风格；K-Means++；NGSIM；PCA

Abstract: This article focuses on identifying driving styles among different drivers. Firstly, the NGSIM dataset is smoothed by symmetrical exponential moving average filtering algorithm. Secondly, by analyzing key indicators from domestic and international studies on characterizing driving styles, eight driving style feature variables are determined. The independence of these eight features is validated by calculating the rank of the driving style feature vector. Combining principal component analysis, three variables that characterize driving styles are identified. Then, the KMeans++ model is constructed to cluster driving styles into aggressive, moderate, and cautious types.For comparison and validation, K-means and gaussian mixture module (GMM) models are also established. The results show that the silhouette coefficient and algorithm runtime of the K-Means++ model are superior to those of the K-Means and GMM models. The driving style clustering method proposed in this paper can effectively classify the driving style of drivers, which is of great significance for improving traffic safety, traffic efficiency and promoting the development of intelligent transportation system.

Key words: traffic safety; driving style; K-Means++; NGSIM; PCA