考虑到车辆在变道过程中的安全性和舒适性,提出基于改进PET算法的车辆变道交通冲突识别模型和基于特征类(Eigen class)分类器的变道行为判别模型。首先,从城市快速路交织区的公开数据集(City Sim)中提取微观驾驶数据,建立改进的后侵入时间(Post - encroaching Time)模型,以计算变道特征。然后,在相关性分析的基础上,利用Eigen class分类方法识别车辆变道风险的重要特征变量,构建以改进PET为指标的车辆变道风险判别模型。最后,通过与K-近邻算法(k-nearest neighbor classification,KNN)、多支持向量机(multiple support vector machine,MSV)等算法的分类预测性能进行对比分析,评价车辆变道风险判别模型的精度。结果表明:变道风险判别模型的判别准确率为99.73%,优于KNN算法(99.71%)、MSVM算法(99.50%)和DT算法(99.61%)。

Considering the safety and comfort of vehicles during lane changing, a vehicle lane changing traffic conflict recognition model based on the improved PET algorithm and a lane changing behavior discrimination model based on an Eigen class (Eigen class) classifier are proposed. Firstly, micro-driving data are extracted from a publicly available dataset (City Sim) of urban expressway intersection area, and an improved Post - encroaching Time (PET) model is established to compute lane-changing features. Then, on the basis of correlation analysis, Eigen class classification method is used to identify the important characteristic variables of vehicle lane changing risk, and to construct a vehicle lane changing risk discrimination model with improved PET as the indicator. Finally, the accuracy of the lane-change risk discrimination model is evaluated by comparing and analyzing the classification prediction performance with that of K-nearest neighbor classification (KNN) and multiple support vector machine (MSV) algorithms. The results show that the discrimination accuracy of the lane change risk discrimination model is 99.73%, which is better than the KNN algorithm (99.71%), MSVM algorithm (99.50%) and DT algorithm (99.61%).

U491???????