Airfusion: a machine learning framework for balancing air traffic demand and airspace capacity through dynamic airspace sectorization

• Main Authors: Zhou, Wei, Pham, Duc-Thinh, Alam, Sameer
• Other Authors: School of Mechanical and Aerospace Engineering
• Format: Conference or Workshop Item
• Language: English
• Published: 2023
• Subjects: Engineering; Air Traffic Control; Machine Learning Framework
• Online Access: https://hdl.handle.net/10356/171345; https://2023.ieee-itsc.org/
• Institution: Nanyang Technological University

Balancing air traffic demand and airspace capacity is a key challenge in airspace management. This task requires situational awareness among air traffic controllers, necessitating the use of interpretable traffic forecasts and visual tools to facilitate well-informed decision-making processes. This paper proposed AirFusion, a machine learning framework designed to balance airspace demand and capacity through Dynamic Airspace Sectorization (DAS). DAS is a concept that involves the dynamic change of the sectors configuration in response to fluctuations in traffic demand. The proposed framework includes four key components: (i) demand and capacity prediction, leveraging the Temporal Fusion Transformer (TFT) - a high-performing multi-horizon prediction model that offers interpretable insights into temporal dynamics, enabling traffic demand and airspace sector capacity prediction with a 4-hour look ahead and 6-hour look back window; (ii) major flow identification, using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to effectively learn traffic patterns and identify major traffic flows; (iii) DAS, optimizing airspace sector capacity by employing graph-based partitioning method to split the sector when predicted demand exceeds capacity; and (iv) visual interface, providing an interactive platform that presents the sector splitting boundary and key influencers for demand and capacity predictions, enabling well-informed timely DAS for air traffic controllers. To validate the proposed AirFusion framework, air traffic data from four selected sectors of Singapore Flight Information Region (FIR) in December 2019 is used for training and evaluation. The experimental results demonstrate the model's high accuracy, with a mean absolute error of 0.0234 for traffic demand prediction and 0.0291 for airspace sector capacity prediction. Furthermore, the R-squared values indicate high predictive performance, with an average of 0.9133 for traffic demand and 0.9605 for airspace sector capacity.