基于CNN-LSTM模型的车辆换道前跟驰研究
Research on vehicle following before lane changing based on CNN-LSTM model
潘公宇 1马斌2
作者信息
- 1. 江苏大学 车辆产品实验室,江苏 镇江 212013
- 2. 江苏大学 汽车与交通工程学院,江苏 镇江 212013
- 折叠
摘要
考虑换道车辆在换道前的跟驰行为与无换道意图的一般跟驰行为有明显的差异,为研究车辆在换道前的特殊跟驰行为,提出"换道前跟驰"阶段概念,将换道车辆的跟驰过程划分为"基本跟驰"与"换道前跟驰"两阶段,以主车在换道前斜率的第五八分位数作为"换道前跟驰"的终点,使用z检验法验证了换道车辆在换道前跟驰阶段运动状态的特殊性.搭建CNN-LSTM网络以车辆速度、加速度、相对距离、横向偏移量等为输入,利用CNN层提取输入层特征,再将提取出的特征作为LSTM网络的输入,利用LSTM网络实现跟驰车辆状态的预测.仿真结果表明,传统的IDM不适用于车辆换道前的特殊跟驰行为,搭建的CNN-LSTM模型在加速度精度上较传统IDM模型提升了 15.1%,更适用于车辆换道前跟驰状态的描述.
Abstract
Obvious differences exist between the car following before lane change and the car following without lane change.This paper proposes the"car following before lane change"to study the special car following before changing lanes.The lane change is divided into two stages:"basic car following"and"car following before lane change",with the fifth and eighth Quantile of the slope of the main vehicle before lane change as the end point of"car following before lane change".Z-test method is employed to verify the specificity of the motion state of lane changing vehicles before changing lanes.A Convolutional Neural-Long Short Term Memory network(CNN-LSTMnetwork)is built with vehicle speed,acceleration,relative distance and lateral offset as inputs.The CNN layer is employed to extract input layer features,which are then used as inputs to the LSTMnetwork.The LSTMnetwork is employed to predict the following vehicle status.The simulation results show the traditional IDMis not suitable for the special car following behavior before changing lanes.Our CNN-LSTM model improves the acceleration accuracy by 15.1%compared to the traditional IDMmodel,and therefore is more suitable for describing the car following before changing lanes.
关键词
换道前跟驰/车辆状态预测/CNN-LSTM融合神经网络/NGSIM数据集Key words
car following before lane change/vehicle status prediction/CNN-LSTM fusion neural network/NGSIM dataset引用本文复制引用
基金项目
国家自然科学基金面上项目(52072157)
出版年
2024
国家科技期刊平台
NSTL
万方数据