摘 要：本文选取NGSIM数据集中的运营车辆行驶数据作为研究对象，采用小波降噪方法处理干扰数据，进而提出一种基于Transformer-MISTM的车辆换道轨迹预测模型。该模型集成了Transformer模块的多头注意力机制和高效并行处理能力，可对周边车辆运动轨迹特征进行权重计算；MLSTM模块在捕获长期依赖的同时，加强各层间的信息传递，从而增强模型的整体表达能力。结果表明：在Transformer-MLSTM联合模型最优超参数组合下，预测车辆横向和纵向轨迹变化的RMSE值分别达到0.364和1.492,该模型的预测速度和准确度均优于采用LSTM、MLSTM和Transformer等方法建立的单一网络模型。
关键词：行驶数据；NGSIM；换道轨迹预测；ISTM；Transformer
中图分类号：U491.1+4；TP183 文献标志码：A DOI：10.15917/j.cnki.1006-3331.2026.01.003
Prediction of Operating Vehicle Driving Trajectory Based on Transformer-MLSTM Joint Model
DU Yucheng,ZHU Liwei,LI Huimin,CHEN Fanghua
Abstract: This paper selects operational vehicle trajectory data from NGSIM data set as the research subject,employs a wavelet denoising method to address data noise,and proposes a Transformer-MLSTM-based model for predicting vehicle lane-changing trajectories.The model integrates the multi-head attention mechanism and efficient parallel processing ability of the Transformer module to calculate the weight of the motion trajectory characteristics of the surrounding vehicles.Simultaneously,the MLSTM module captures long-term dependencies in trajectory sequences,enhancing information transfer and further improving model performance.The results show that under the optimal hyper-parameter combination of the Transformer-MLSTM joint model,the RMSE values for predicting the lateral and longitudinal trajectory changes of the vehicle reach 0.364 and 1.492,respectively.The prediction speed and accuracy of the model are better than the single network model established by ISTM,MLSTM and Transformer.
Key words: driving data; NGSIM; lane change trajectory prediction; ISTM; Transformer
