A Multi-task Learning Approach for Facilitating Dynamic Airspace Sectorization

Abstract

Dynamic Airspace Sectorization (DAS) is a key pathway for enabling advanced demand capacity balancing (DCB) in modernizing Air Traffic Management (ATM). By splitting and merging the sectors, DAS allows airspace to accommodate the evolving air traffic situations for improving the utilization of airspace in response to different air traffic demands, airspace capacity, weather events and other factors. This research aims at supporting the decision-making on when-to-do such DAS from a deep learning perspective. To this end, this paper proposes a multi-task learning (MTL) approach which is able to predict sector traffic flow and airspace capacity simultaneously using a shared neural network architecture. Specifically, the proposed model predicts the demand-capacity imbalance and identifies the opportunity for sector split/merge implementation. To validate the feasibility of the proposed model, a case study has been carried out in Singapore en-route airspace using the Automatic Dependent Surveillance – Broadcast (ADS-B) data and meteorology data in December 2019. Experimental results explicitly show the capability of the proposed MTL model in predicting flow and capacity. Based on predicted results along with a pre-defined rule, the proposed model predicts the demand-capacity imbalance across multiple timescales and explores the potential to facilitate DAS in terms of tactic, pre-tactic and strategic ATM operations.