Terminal area flight trajectory clustering based on deep autoencoding gaussian mixture model

Abstract

Terminal airspace trajectory clustering is a significant research topic, providing crucial decision-making support for airspace design, traffic management, and route planning. Currently, an advanced method for flight trajectory clus-tering involves using the latent space of autoencoders. While autoencoders excel in handling low-dimensional, simple data, the high dimensionality, complexity, and noise of terminal area flight trajectories pose challenges for data recon-struction. Therefore, this paper introduces the Deep Autoencoding Gaussian Mixture Model (DAGMM) to establish a new flight trajectory clustering model, aiming to uncover terminal airspace flow patterns. Using departure trajectories in the terminal airspace of Guangzhou Baiyun Airport in China as a case study, the effectiveness of the proposed clustering framework is validated through visualization of various trajectory distributions and the use of the t-SNE method to visualize trajectories in latent space.