重庆理工大学学报2026，Vol.40Issue(7)：1-10,10.DOI:10.3969/j.issn.1674-8425(z).2026.04.001
融合多头注意力的时空交互感知车辆轨迹预测模型
Spatiotemporal interaction perception vehicle trajectory prediction model with multi-head attention
摘要
Abstract
High-precision vehicle trajectory prediction is crucial to improving both the safety and operational efficiency of autonomous driving systems.This paper proposes a spatiotemporal attention and bidirectional long short-term memory(STA-BiLSTM)trajectory prediction model that integrates multi-head attention mechanisms with bidirectional LSTM networks.By hierarchically extracting single-vehicle trajectory features and jointly modeling spatiotemporal interdependencies among multiple vehicles,the proposed model effectively enhances prediction accuracy in complex dynamic scenarios.First,the model employs a CNN-LSTM architecture to encode historical trajectory features of individual vehicles.Then,a multi-head attention mechanism is introduced to adaptively model spatial interactions among vehicles.Finally,a BiLSTM network is employed to capture the temporal evolution of these inter-vehicle dependencies.Experimental results demonstrate that the proposed model reduces the root mean square error(RMSE)at a 5s prediction horizon by9.8％compared with the Double Transformer baseline,while maintaining strong interpretability.By explicitly modeling the spatiotemporal dependencies between ego and surrounding vehicles,this study may provide an effective solution for trajectory prediction in complex traffic environments.关键词
车辆工程/轨迹预测/注意力机制/时空建模/双向长短时记忆网络/深度学习Key words
automotive engineering/trajectory prediction/attention mechanism/spatiotemporal modeling/bidirectional long/deep learning分类
交通工程引用本文复制引用
陈峥,靳云淇,郑嘉,魏福星,沈世全,郭凤香..融合多头注意力的时空交互感知车辆轨迹预测模型[J].重庆理工大学学报,2026,40(7):1-10,10.基金项目
云南省重大科技专项(202503AA080005) （202503AA080005）
云南省车路协同控制与运行安全创新团队(202505AS350024) （202505AS350024）
云南省"兴滇英才支持计划"云岭学者专项(KKRC202402005) （KKRC202402005）
