Abstract: In response to the issues of insufficient accuracy and poor adaptability in existing vehicle trajectory prediction methods when dealing with complex traffic environments and multi-vehicle interactions, this paper proposes a vehicle trajectory prediction method based on graph convolutional networks (GCN) and attention mechanisms. First, long short-term memory (LSTM) is used to initially encode the historical trajectories of vehicles. Then, GCN is introduced to model spatial relationships and interaction effects between vehicles, updating vehicle features. Subsequently, multi-head self-attention mechanisms further capture global interaction information between vehicles. Finally, residual decoders generate future trajectory predictions. Experimental results show that the model proposed in this paper demonstrates superior vehicle trajectory prediction capabilities on the Argoverse dataset.

Key words: vehicle trajectory prediction; GCN; LSTM; attention mechanism

摘要： 针对现有车辆轨迹预测方法在处理复杂交通环境和多车辆交互时存在精度不足和适应 性差的问题，文章提出一种基于图卷积神经网络（GCN）和注意力机制的车辆轨迹预测方法。 首先使用长短时记忆网络（LSTM）对车辆的历史轨迹进行初步编码；再引入 GCN 建模车辆 之间的空间关系和交互影响，对车辆特征进行更新；然后通过多头自注意力机制进一步捕获 车辆之间的全局交互信息；最后通过残差解码器生成未来的轨迹预测。实验结果表明，文章 提出的模型在 Argoverse 数据集表现出优越的车辆轨迹预测能力。

关键词: 车辆轨迹预测；GCN；LSTM；注意力机制