基于支持向量机的车辆换道决策模型研究

Research on vehicle lane-changing decision model based on support vector machine

车辆换道行为是微观交通流中的典型驾驶行为之一。研究车辆换道决策模型,可以帮助无人驾驶车辆正确进行换道决策。以NGSIM数据集为依据,采用SG滤波器对NGSIM数据集进行平滑处理,筛选平滑处理后的数据得到训练集和测试集;选择影响车辆换道决策的7个因素作为模型输入,建立基于粒子群优化算法的支持向量机(PSO-SVM)车辆换道决策模型和标准支持向量机(标准SVM)车辆换道决策模型;对训练集和测试集进行归一化处理,利用归一化处理的数据进行模型的训练和测试。测试集数据分类验证结果表明,建立的PSO-SVM车辆换道决策模型的决策准确率为94.67%,相比于标准SVM车辆换道决策模型提高6%,能有效实现无人驾驶车辆的换道决策。

Vehicle lane-changing behavior is one of the typical driving behaviors in microscopic traffic flow.The study of vehicle lane-changing decision model is conducive to unmanned vehicles to correctly make lane-changing decision.Based on the NGSIM data set,this study uses the SG filter to smooth the NGSIM data set.The smoothed data is screened to obtain training set and testing set.7 factors that affect the vehicle lane-changing decision are selected as the model input,and a support vector machine vehicle lane-changing decision model based on the particle swarm optimization(PSO-SVM)and a standard support vector machine(SVM)vehicle lane-changing decision model are established.The training and testing set are normalized.The normalized data are used to train and test models.The testing set data classification verification results show that the decision accuracy of the PSO-SVM vehicle lane-changing decision model is 94.67%,which is 6%higher than that of the standard SVM vehicle lane-changing decision model,and it can effectively realize the lane-changing decision of unmanned vehicles.