基于BP神经网络的智能车辆换道决策模型研究

Research on Lane-Changing Decision Model for Intelligent Vehicles Based on BP Neural Network

为了正确刻画智能网联环境下的车辆换道行为,提出基于BP神经网络的车辆换道决策模型。分析了交通流中车辆换道行为,以HighD自然驾驶数据集为数据来源,筛选出1 900组车辆换道和未换道信息作为模型的训练与验证,利用高斯滤波方法拟合目标车辆换道轨迹和横向位移轨迹,选择影响车辆换道决策的7个参数作为模型输入,建立BP神经网络换道决策模型,利用筛选出的数据进行模型的训练和测试。结果表明,选择7个参数建立的BP神经网络模型决策精度达到96.2%,能有效识别车辆换道行为。

In order to correctly describe the lane-changing behavior of vehicles in the intelligent networked environment,a vehicle lane-changing decision model based on BP neural network is proposed.Initially,the lane-changing behavior in traffic flow was analyzed,and then the HighD natural driving dataset was used to select 1 900 samples with or without lane-changing behavior for model training and testing.The Gaussian filtering was applied to fit the lane-changing trajectories and the lateral displacement trajectories.7 parameters that affect the lane-changing decisions were chosen as the model input.Finally,the BP neural network based lane changing decision model was trained and tested.The results show that the decision accuracy of the proposed BP neural network model reaches 96.2%,which can effectively identify the lane changing behavior of vehicles.