RT info:eu-repo/semantics/doctoralThesis
T1 Improving efficiency of trajectory-based operations for air traffic management using deep learning
A1 Silvestre Vilches, Jorge
A2 Universidad de Valladolid. Escuela de Doctorado
K1 Sistemas informáticos
K1 Deep learning
K1 Aprendizaje profundo
K1 Air traffic management
K1 Gestión del tráfico aéreo
K1 Estimated time of arrival
K1 Estimación tiempo de llegada
K1 Data integration
K1 Integración de datos
K1 1203.1 Informática
AB Trajectory-Based Operations (TBO) are the cornerstone of contemporary Air Traffic Management (ATM) systems, aiming to enhance operational efficiency, predictability, and safety through the strategic management of flight trajectories. TBO define flight trajectories as 4D-trajectories, where each point in the flight path is defined through its latitude, longitude, altitude and time. Hence, the use of airspace capacity can be optimized by strategically planning the flights not only from the perspective of the aircraft position, but also of the time. 4D trajectories can be enriched with additional flight information to provide a more holistic representation of flights. However, this information is scattered and harmonizing it in an integrated representation is yet to be achieved.This thesis addresses the challenge of leveraging heterogeneous data sources to generate high-quality 4D trajectories, and explores their potential to improve ATM operations through a data-driven approach based on deep learning.  We first design a conceptual data model tailored to represent the most critical elements of ATM flight operations. This model encompasses multiple domains, including surveillance data, flight plans, meteorological information, and airport infrastructures, ensuring a unified representation of the foundations of ATM. A data integration architecture is developed to process data from heterogeneous sources and reconstruct enriched 4D trajectories conforming to the defined data model. The data quality is thoroughly evaluated to ensure that the data is complete, precise and reliable, with special attention to the position and time where it is missing or incorrect.The thesis further investigates the applicability of these enriched 4D trajectories in addressing critical operational challenges within ATM, and does so by proposing specific deep learning architectures that model 4D trajectories as time series. Two primary use cases are identified: the estimation of arrival times (ETA) for en-route flights and the prediction of flight trajectories, since both represent core elements of TBO and are vital for improving air traffic flow management, reducing delays, and optimizing resource allocation in congested airspace. We focus on flights arriving at the Madrid Barajas-Adolfo Suárez Airport (Spain) as our case study to exemplify and validate our proposal. The time of arrival at the destination airport for en-route flights can be reliably estimated based on the current state of the flight, together with other factors influencing over the flight, such as weather and schedule data. Our approach based on Long Short-Term Memory networks (LSTM) has improved the current state of the art, providing longer term and more accurate predictions, at any point in the trajectory, than existing methods with a mean error of 2.67 minutes versus the 3.67 minutes of the state of the art. A similar approach can be applied to predict the trajectory of a flight, as the future positions can be predicted based on the past positions of the flight. The state of the art methods provide results with a mean error of 0.005 and 0.01 nautical miles (0.92 and 1.85 kilometres) for latitude and longitude predictions, respectively, but does so by focusing on a single route (repetitive flights between two airports). In contrast, our solution apply Temporal Fusion Transformers to achieve a similar precision at predicting the 2D position for all flights arriving at a specific airport, regardless of their origin.The outcomes of this thesis not only satisfy our research hypothesis and objectives, but also contribute to advancing the state of the art by demonstrating how enriched 4D trajectories, derived from a comprehensive data integration pipeline, can serve as the foundation for innovative, data-driven solutions for critical Trajectory Based Operations in the increasingly complex context of the global Air Traffic Management.
YR 2025
FD 2025
LK https://uvadoc.uva.es/handle/10324/81920
UL https://uvadoc.uva.es/handle/10324/81920
LA eng
NO Escuela de Doctorado
DS UVaDOC
RD 26-ene-2026