车辆轨迹预测是自动驾驶技术的关键任务之一，准确预测车辆轨迹对于智能网联环境下的车辆行驶安全与通行效率具有重要意义。为解决复杂交互场景的车辆轨迹预测问题，本文提出一种融合时空关联嵌入关系的轨迹预测网络。该网络由时空关联嵌入、融合特征提取以及轨迹预测三部分组成。首先，时空关联嵌入模块基于双层卷积融合局部轨迹时序信息与空间关系。其次，在编码器中引入旋转位置编码（Rotary Positional Embedding，RoPE）机制捕捉长时序轨迹的全局时空依赖关系，最后通过多模态解码器进行轨迹并行预测，降低预测累计误差。特征表示方面，采用自车特征、相对关系特征以及交通环境特征提升对场景的表征能力。最后，基于NGSIM数据集开展实验，结果表明，与LSTM（Long Short-Term Memory）、TCN（Temporal Convolutional Network）、Transformer等方法相比，本文方法表现最优，在平均终点位移误差上降低了0.296m。

Vehicle trajectory prediction is one of the key tasks in autonomous driving technology. Accurate prediction of vehicle trajectories is of great significance for driving safety and traffic efficiency in intelligent connected environments. To address the problem of vehicle trajectory prediction in complex interactive scenarios, this paper proposes a trajectory prediction network that integrates spatio-temporal correlation embedding relationships. The network consists of three components: spatio-temporal correlation embedding, fused feature extraction, and trajectory prediction. First, the spatio-temporal correlation embedding module employs a dual-layer convolutional structure to integrate local trajectory temporal information with spatial relationships. Second, a Rotary Positional Embedding (RoPE) mechanism is introduced into the encoder to capture global spatio-temporal dependencies in long-term trajectories. Finally, a multi-modal decoder is utilized for parallel trajectory prediction to reduce cumulative prediction errors. In terms of feature representation, ego-vehicle features, relative relationship features, and traffic environment features are employed to enhance scene representation capabilities. Experiments are conducted based on the NGSIM dataset. The results demonstrate that the proposed method outperforms LSTM, TCN, Transformer, and other methods, achieving a reduction of 0.296 meters in average final displacement error.