A Deep Clustering and Generative Approach for Large-Scale Air Traffic Trajectory Data

Precise recognition of spatio-temporal patterns in air traffic trajectory data plays an important role in ensuring the safety and efficiency of air traffic management. Clustering is a widely used method for identifying spatio-temporal patterns in traffic trajectories. However, the high dimensionality of trajectory data often complicates the analysis, especially in terms of both temporal and spatial aspects. Here, we introduce an efficient deep clustering model using a generative approach for trajectory clustering, called Air Traffic Trajectory analysis via Latent feature Clustering (ATTLC). It features distinctive characteristics: ATTLC combines a variational autoencoder (VAE) model with a clustering algorithm to effectively learn latent features, accurately represent air traffic trajectory data, and cluster objects within the latent space. ATTLC optimizes both clustering and network loss. We demonstrate the effectiveness of ATTLC in clustering tasks on large-scale air traffic trajectory data covering China. The experiments demonstrate that ATTLC can address high-dimensional trajectory data. Most importantly, ATTLC can help analyze spatio-temporal patterns in large-scale air traffic trajectory data, potentially revealing the evolution of the trajectory networks and identifying core hub nodes.
© 2026 IEEE.