A Spatiotemporal Hybrid Model for Airspace Complexity Prediction
Wenbo Du, Biyue Li, Jun Chen, Yisheng Lv, Yumeng Li
Abstract
Airspace complexity is a key indicator that reflects the safety of airspace operations in air traffic management systems. Furthermore, to achieve efficient air traffic control, it is necessary to accurately predict the airspace complexity. In this article, we propose a novel spatiotemporal hybrid deep learning model for airspace complexity prediction to efficiently capture spatial correlations as well as temporal dependencies pertaining to the airspace complexity data. Specifically, we apply convolutional networks to discover the short-term temporal patterns and skip long short-term memory networks to model the long-term temporal patterns of airspace complexity data. Furthermore, it is observed that the graph attention network in our proposed model, which emphasizes capturing the spatial correlations of the airspace sectors, can significantly improve the prediction accuracy. Extensive experiments are conducted on the real data of six airspace sectors in Southwest China. The experimental results show that our spatiotemporal deep learning approach is superior to state-of-the-art methods.