Dynamic Graph Neural Network Vehicle Trajectory Prediction Based on Diffusion Probabilistic Model

To address the challenges of vehicle trajectory prediction, including environmental changes and inter-vehicle interactions that lead to lower prediction accuracy over long-term horizons, a dynamic graph neural network(DGN) model based on a diffusion probability model(DPM) is proposed. Local outlier factor (LOF) is used to remove outliers and smooth the trajectory data. Experiments on the NGSIM dataset demonstrate that the model performs well, particularly in terms of vertical displacement error. Analysis of the predicted ADE and FDE results shows that the proposed model outperforms existing models, validating both the model's and the algorithm's effectiveness.