Data-driven network modeling for airspace optimization through airway re-structurization and traffic flow prediction

• Main Author: Ma, Chunyao
• Other Authors: Sameer Alam
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Aeronautical engineering; Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/164436
• Institution: Nanyang Technological University

The increasing air traffic demand is leading to airspace congestion which not only elicits traffic delays causing substantial economic losses but may also exceed the airspace capacity causing potential safety and inefficiency issues. Accommodating the future air traffic demand becomes a major challenge confronted by air navigation service providers (ANSPs). This problem is further compounded due to air traffic's constantly evolving nature, owing to weather disturbances, frequent flight rerouting, Special Use Airspace (SUA) management, and tactical Air Traffic Flow Management (ATFM). However, according to a consensus in the aviation community, the projected increase in traffic demand can be accommodated through improved and effective airspace management, which urges research to optimize the airspace structure and flow for better air traffic management. In the literature, research efforts on airspace optimization mainly focus on sector structure optimization and air traffic flow optimization. Sector structure optimization usually leads to a redesign of the entire airspace structure, which is difficult to implement due to operational constraints, while the effectiveness of air traffic flow optimization is limited by the constantly changing nature of air traffic due to airspace capacity constraints, weather, and varying traffic demand. Therefore, it is essential to develop innovative approaches that can effectively manage the airspace by making minimal changes to the underlying network structure and can provide timely and accurate predictions about traffic flow conditions for effective flow management. In view of this, this thesis aims to seek answers to the following research questions which can contribute to airspace optimization: 1) how to identify the structural bottlenecks in a given airspace such that minimal changes can be made to the airspace network structure to improve the traffic flow? 2) how to exploit the structural and flow features of an airspace to make effective predictions of air traffic capacity and flow? To this end, this thesis concentrates on three specific research topics: airspace structure optimization, airspace capacity overload identification, and air traffic flow prediction. Note that most traditional methods in airspace optimization are based on analytical models derived from empirical analysis or traffic simulations, which may not accurately represent the complex, dynamic, and interdependent nature of real-world air traffic. With the widespread availability of ADS-B air traffic surveillance data, data-driven approaches offer a new paradigm for solving the problem of airspace optimization. Compared to analytical models, data-driven approaches can be more effective in representing the complex relationships among the system variables without explicit knowledge of the physical behavior of the air traffic system. Therefore, this thesis aims to develop data-driven airspace optimization approaches based on real-world air traffic data. In airspace structure optimization, this thesis proposes to improve airway network performance based on complex network theories. Based on the network percolation and centrality theories, this thesis proposes a complex network approach for critical link detection in airway networks to identify airway links that are vital for the structural integrity and performance of the networks in accordance with time-varying traffic situations. Moreover, inspired by the Braess's Paradox (BP) phenomenon, a scenario where an alteration to a traffic network to improve traffic conditions actually has the reverse effect, this thesis provides a ``counter-intuitive'' perspective towards airspace optimization by removing airways/links from a given airway network to improve the air traffic flow efficiency. In terms of airspace capacity overload identification, this thesis proposes to use airspace collision risk patterns as an indicator of capacity overload. With air traffic data and airspace configurations, the collision risk distributions inside an airspace are organized into different patterns, such as normal patterns and overload patterns, based on the density and intensity of the collision risk. The capacity overload status can be identified by recognizing the best-fit pattern of the collision risk distribution, which can indicate the anticipated saturation in the airspace. As for air traffic flow prediction, this thesis proposes to predict short-term en-route traffic flows based on graph convolutional networks. A graph representation of air traffic structure is built from real-world air traffic data, and the dynamic spatial-temporal features of air traffic flow are explored from a graph perspective. Such prediction can help plan and execute tactical measures for better air traffic management. Feedback from subject-matter experts demonstrates that the complex network theory-based method for critical link identification proposed in this thesis can dynamically identify links, both structurally and operationally, critical under time-evolving air traffic scenarios. Research results presented in the thesis show that optimizing the airway network structure by removing some links can save up to 3.8\% of the travel time for a one-day traffic sample at a given flight level. Moreover, results of airspace capacity overload identification demonstrate that distinct collision risk patterns exist for the normal state, transition phase, and capacity overload state. The proposed method can identify the capacity situation of a given traffic situation by recognizing its best fit pattern. The results of air traffic flow prediction show that the proposed graph convolutional network-based method outperforms the well-established Long Short-Term Memory (LSTM) model. It exhibits a better capability to predict rapid changes in traffic flow and has a relatively smaller decrease in prediction accuracy as the prediction window increases. Through airspace optimization, i.e., airway network optimization to improve air traffic flow efficiency and capacity overload identification/air traffic flow prediction to advise ATFM to manage anticipated capacity/air traffic flow circumstances, this thesis paves the way to a better accommodation of the increasing air traffic.