AirTraj-Diff: Generating Aircraft Trajectory With Conditional Diffusion Probabilistic Model

doi:10.1109/TITS.2025.3622626

Return

AirTraj-Diff: Generating Aircraft Trajectory With Conditional Diffusion Probabilistic Model

2025-11-10

PRE

DOI：10.1109/TITS.2025.3622626

Original

Original request for help

Save

Abstract

En 中文

Probabilistic aircraft trajectory models in the terminal area that reflect real-world distributions can significantly facilitate conflict detection, performance analyses, and risk assessments. However, due to the inherent uncertainties in air traffic, complex spatiotemporal correlations within terminal area, and the influence of diverse weather conditions, modeling these distributions and generating realistic trajectories pose substantial challenges. In this paper, we present a diffusion model-based method to learn the distribution of aircraft trajectories in the terminal area, enabling the generation of high-quality trajectories that resemble real data distributions. This is achieved by effectively combining the generative capability of diffusion models with the ability of extracting spatiotemporal features embedded in real trajectories. Specifically, we propose an airplane trajectory diffusion network structure, which integrates a UNet deep neural network to capture multi-level noise estimation and accurately model the uncertainty inherent in trajectory data. Additionally, we employ a conditional generation module that incorporates meteorological information, allowing the model to learn the correlation between weather conditions and trajectory distributions. Experiments with real-world terminal-area datasets demonstrate that our model can generate realistic trajectories that closely resemble the actual distribution of terminal area flight paths. Comparative results present improvements over existing methods across various evaluation metrics.

Keywords:

Air traffic management

trajectory probability model

generative model

diffusion model

trajectory generation