融合K-means与高斯混合模型的驾驶风格聚类研究

Study on driving style clustering based on K-means and Gaussian mixture model

为研究驾驶员的跟车特性,探究驾驶员风格划分方法,采集50名驾驶员的实车试验数据,选取平均跟车时距和平均制动时距为二维向量,建立基于K-means聚类结果的高斯混合模型(GMM)并分析不同风格驾驶员的聚类结果。研究表明:样本数据聚为3类时的平均轮廓系数为0. 45,将驾驶员划分为冒进型、平稳型、保守型3类时聚类效果较好;冒进型驾驶员倾向于选择较小的跟车时距和制动时距,保守型驾驶员的跟车及制动时距则普遍较大,模型聚类结果更加柔性,样本区分度更高。

In order to study drivers ’ car-following characteristics and explore an effective method to classify driving styles,50 participants were recruited to carry out a real road driving test. A GMM with results of K-means clustering was established based on two-dimensional variables: average car-following time gap and average braking time gap. And then results of different types of drivers were analyzed. The research shows that clustering result is better with three categories( aggressive drivers,steady drivers,and conservative drivers) with an average contour value of 0. 45. It is found that aggressive drivers tend to choose a smaller car-following time gap or braking time gap while conservative drivers usually take a larger value,and a much softer clustering result with a high separability between samples would be achieved.