An end-to-end framework for flight trajectory data analysis based on deep autoencoder network

摘要

In order to jointly solve the tasks of abnormal trajectory detection and flow pattern recognition for flight trajectory data analysis, an end-to-end framework based on deep autoencoder network is proposed in this paper. Considering the coupling relationship between the two tasks, a structured sparsity-inducing norm is introduced into the reconstruction-based loss function to separate abnormal trajectories from the whole and extract low-dimensional and outliers-free representations from the remaining normal trajectories. On this basis, cluster assignment hardening is applied to further learn cluster-friendly representations as well as cluster assignment for each trajectory. The effectiveness and efficiency of the framework are validated on flight trajectories arriving at Hong Kong International Airport. Experimental results show that the proposed framework not only detects typical spatial anomalies, including holding patterns and rerouting patterns, but also identifies fine-grained cluster structures. Furthermore, it surpasses current state-of-the-art methods in terms of anomaly detection performance and cluster quality, with an improvement of 13.56% and 22.82%, respectively. With parallel computing, its time cost can be reduced to less than 1 second, which helps to perceive traffic situations and monitor abnormal behaviors in real time.