摘要
由于车辆高速移动引发位置突变,深度学习模型难以学习预测函数,导致现有方案在长期预测中位置误差较大,对此提出了基于车流特性的长期车辆轨迹预测方案。通过分析城区场景中车流的聚集性和潮汐性,分析各个位置出现的频次,确定了不同位置相应的权重值;利用长短期记忆(long short-term memory,LSTM)网络的“记忆”特性,设计了基于LSTM的编码器-解码器预测模型,实现长期车辆轨迹预测。仿真实验结果表明,所提方案与消融实验方案对比,在工作日长期预测中的平均位置误差降低了1.4%,最终位置误差降低了1.1%,均方根误差降低了0.9%,且具有较好的泛化性。
Due to the sudden change of position triggered by the high-speed movement of vehicles,it is difficult for the deep learning model to learn the prediction function,which leads to the problem of large position error in the long-term prediction of existing schemes.A long-term vehicle trajectory prediction scheme based on traffic flow characteristics was proposed.By analyzing the aggregation and tide of traffic in urban scenes,the frequency of each location was counted,and the corresponding weight values of different locations were determined.The encoder-decoder prediction model based on the long short-term memory(LSTM)network was designed for long-term traffic trajectory prediction by using the“memory”feature of the network.The experimental results of simulation show that the proposed scheme has a 1.4%lower average position error,1.1%lower final position error,and 0.9%lower root mean square error as well as a better generalization in long-term prediction compared with the ablation experimental scheme.
作者
李新
蔡英
张猛
李汶锦
范艳芳
LI Xin;CAI Ying;ZHANG Meng;LI Wenjin;FAN Yanfang(Computer School,Beijing Information Science&Technology University,Beijing 100101,China)
出处
《北京信息科技大学学报（自然科学版）》
2022年第6期32-37,共6页
Journal of Beijing Information Science and Technology University
基金
北京市自然科学基金-海淀原始创新联合基金资助项目(L192023)。
关键词
车辆轨迹预测
长短期记忆网络
车流特性
vehicle trajectory prediction
long short-term memory network
traffic flow characteristics
