基于行驶轨迹的车辆换道行为预测方法研究  

Research on the Prediction Method of Vehicle Lane Changing Behavior Based on Driving Trajectory

作　　者：周立宸 邓建华[1] ZHOU Li-chen;DENG Jian-hua(College of Civil Engineering,Suzhou University of Science and Technology,Suzhou,Jiangsu 215011,China)

机构地区：[1]苏州科技大学土木工程学院,江苏苏州215011

出　　处：《黑龙江交通科技》2023年第1期142-144,共3页Communications Science and Technology Heilongjiang

摘　　要：车辆换道行为是机动车在道路上行驶时的一种常见行为,对其进行预测能有效减少交通事故。基于NGSIM项目数据,对换道车辆换道时刻的行为特性进行分析,运用随机森林算法筛选出表征换道行为的参数指标:横纵向速度、加速度、偏角、车头间距以及相对时距,提出一种级联LSTM模型预测车辆在行驶过程中是否会发生换道行为,并与传统单层LSTM模型对比。结果表明:级联LSTM模型预测准确率可达93.6%,具有良好的换道预测效果且高于单层LSTM模型,可为自动驾驶和深度学习提供一定的理论参考。Vehicle lane-changing behavior is a common behavior when motor vehicles are driving on the road, and predicting it can effectively reduce traffic accidents.Based on the data of NGSIM project, this paper analyzes the behavior characteristics of lane-changing vehicles at lane-changing time, and uses the random forest algorithm to screen out the parameters that characterize lane-changing behavior, such as horizontal and vertical speed, acceleration, corner, headway and relative time distance.A cascade LSTM model is proposed to predict whether the vehicles will change lanes during driving, and compared with the traditional single-layer LSTM model.The results show that the prediction accuracy of cascade LSTM model can reach 93.6%,which has a good lane change prediction effect and is higher than that of single-layer LSTM model, and can provide some theoretical reference for autonomous driving and deep learning.

关 键 词：行为特性 随机森林 神经网络 换道预测 

分 类 号：U492[交通运输工程—交通运输规划与管理]