基于门控循环单元和Transformer的车辆轨迹预测方法  

Vehicle Trajectory Prediction Method Based on GRU and Transformer

作　　者：王庆荣[1] 谭小泽 朱昌锋[2] 李裕杰 Wang Qingrong;Tan Xiaoze;Zhu Changfeng;Li Yujie(Institute of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070;Institute of Transportation,Lanzhou Jiaotong University,Lanzhou 730070)

机构地区：[1]兰州交通大学,电子与信息工程学院,兰州730070 [2]兰州交通大学,交通运输学院,兰州730070

出　　处：《汽车技术》2024年第7期1-8,共8页Automobile Technology

基　　金：国家自然科学基金项目(71961016,72161024);甘肃省教育厅“双一流”重大研究项目(GSSYLXM-04)。

摘　　要：为增强自动驾驶车辆对动态环境的理解能力及其道路行驶安全性,提出基于门控循环单元(GRU)和Transformer的车辆轨迹预测模型STGTF,使用GRU提取车辆的历史轨迹特征,通过双层多头注意力(MHA)机制提取车辆的时空交互特征,生成预测轨迹。试验结果表明,预测结果的均方根误差(RMSE)平均降低7.3%,STGTF在短期预测和长期预测方面均有不同程度的提升,验证了模型的有效性。In order to enhance the understanding of the dynamic environment of autonomous vehicles and to improve road driving safety,this article proposed a vehicle trajectory prediction STGTF model based on the Gated Recurrent Unit(GRU)and Transformer that used the GRU to extract the historical trajectory features of vehicles,and used a two-layer Multi-Headed Attention(MHA)mechanism to extract the spatio-temporal interaction features of vehicles,generating the predicted trajectories.The experimental results show that the Root-Mean-Square Error(RMSE)of the predicted results decrease by 7.3%on average,STGTF model has different degrees of improvement compared with other existing methods for both short-term prediction and long-term prediction,proving validity of this model.

关 键 词：车辆轨迹预测 门控循环单元 TRANSFORMER 车辆交互 多头注意力机制 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]