This paper proposes a trajectory prediction method based on graph attention network to accurately predict the trajectories of HDV (Human Drive Vehicles) around the ICV (Intelligent Connected Vehicles) under mixed traffic flow scenario on highways. Firstly, the vehicle trajectory data is filtered and smoothed to construct a trajectory prediction dataset containing map information. Secondly, the vehicle interaction relationship graph is constructed based on the position and behavior of vehicles. The high-dimensional spatial interaction relationship features between the target vehicle and surrounding vehicles are extracted using the graph attention network, which serves as input for the encoder-decoder model. Subsequently, an encoder-decoder model based on GRU (Gate Recurrent Unit) is employed to encode time-series features of vehicle trajectory data and generate future trajectories through decoding. Finally, experimental validation using NGSIM (Next Generation Simulation) datasets demonstrates that our proposed method achieves low displacement error in predicting vehicle trajectories compared to models such as GRU, and CNN-GRU (Convolutional Neural Network-Gate Recurrent Unit).

中文翻译：

---

基于图注意力网络的混合交通流下周边车辆轨迹预测

本文提出一种基于图注意力网络的轨迹预测方法，以准确预测高速公路混合交通流场景下HDV（人类驾驶车辆）围绕ICV（智能网联汽车）的轨迹。首先，对车辆轨迹数据进行滤波和平滑处理，构建包含地图信息的轨迹预测数据集。其次，根据车辆的位置和行为构建车辆交互关系图。使用图注意力网络提取目标车辆与周围车辆之间的高维空间交互关系特征，作为编码器-解码器模型的输入。随后，采用基于GRU（Gate Recurrent Unit）的编码器-解码器模型对车辆轨迹数据的时间序列特征进行编码，并通过解码生成未来轨迹。最后，使用 NGSIM（下一代模拟）数据集的实验验证表明，与 GRU 和 CNN-GRU（卷积神经网络门循环单元）等模型相比，我们提出的方法在预测车辆轨迹方面实现了较低的位移误差。