基于轨迹重构数据的不同反应时间跟驰模型
Car-following Models with Different Reaction Time Based on Trajectory Reconstruction Data
耿志军 1程陆 2李阳 2柏海舰 2汪雪松3
作者信息
- 1. 安徽省综合交通研究院股份有限公司,安徽 合肥 230088
- 2. 合肥工业大学 汽车与交通工程学院,安徽 合肥 230009
- 3. 认知智能国家重点实验室,安徽 合肥 230088
- 折叠
摘要
基于NGSIM-I80 数据集,首先通过人工标注跟驰车辆轨迹曲线的刺激点与反应点,提取每条轨迹的反应时间,得到4 种不同反应时间的跟车数据,然后采用零反应时间跟驰模型重构刺激点后的跟驰车轨迹数据,得到任意不同反应时间的重构跟驰轨迹样本,最后采用LSTM神经网络架构,基于重构样本建立不同反应时间的跟驰行为模型.通过仿真发现:反应时间越小,和前车保持的车头间距越小,达到稳定状态的时间越短,反应时间的大小与仿真交通流稳定后的速度大小呈负相关.
Abstract
Based on the NGSIM-I80 dataset,the stimulus point and reaction point of the following vehicle trajec-tory curvesare manually marked,the reaction time of each trajectory is extracted,and the following data of four different reaction time is obtained.Then,the LSTM-0RT is used to reconstruct following vehicle trajectory data after the stimulus point,resulting in reconstructed following trajectory samples with arbitrary reaction time.Final-ly,an LSTM neural network architecture is used to establish car-following behavior models with different reaction time based on the reconstructed samples.Simulation results indicate that a smaller reaction time leads to a shorter headway with the preceding vehicle and a shorter time to reach steady state.The reaction time is negatively corre-lated with the speed after the simulated traffic flow is stabilized.
关键词
自动驾驶车/跟驰模型/轨迹重构/反应时间/LSTMKey words
autonomous vehicle/car-following model/trajectory reconstruction/reaction time/LSTM引用本文复制引用
基金项目
安徽省自然科学基金面上项目(JZ2022AKZR0413)
认知智能国家实验室开放基金(W2022JSKF0504)
出版年
2024
NETL
NSTL
维普
万方数据