| Author: | Chu, Nana |
| Title: | Research on improving terminal traffic flow efficiency and airport capacity through data-driven and deep learning-based dynamic aircraft wake separation prediction |
| Advisors: | Ng, Kam K. H. (AAE) Hsu, Li-ta (AAE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Department: | Department of Aeronautical and Aviation Engineering |
| Pages: | 1 volume (various pagings) : color illustrations |
| Language: | English |
| Abstract: | Improving traffic flow efficiency and runway throughput has always been crucial to dealing with heavy traffic. The trajectory-based operational concept provides an alternate option to achieve highly collaborative air traffic management, among which four-dimensional flight trajectory prediction and time-based separation build the foundation. With the development of computer technology and the accumulation of massive historical flight information and other separation-related data, emerging data-driven machine learning techniques have gained great popularity for civil air traffic operations. Despite extensive deep learning implementations in flight traffic prediction and optimisation problems, they generally focus on model development rather than the applicational scenarios, and the flight procedures and distance-based separation requirements are conservative and static, with restrictions on the efficiency of traffic dispatch. Therefore, this thesis intends to investigate the potential of deep learning in terminal traffic flow optimisation and runway scheduling, particularly from the perspective of predicting dynamic wake separation. Novel deep learning approaches are developed to improve efficiency and safety in terminal approach management and enhance their reliability and trustworthiness for these safety-critical air traffic operations. As the lifetime and movement characteristics of aircraft wake turbulence are crucial determinants of dynamic wake separation and they are highly related to aircraft weight and meteorological conditions, deep convolutional neural networks are developed for near-real-time vortex locations and strength recognition using Light Detection and Ranging (LiDAR) -scanned wake images in the first stage. Data pre-processing, analysis, and pattern learning based on machine learning (ML) and deep learning techniques are involved and pinpointed to identify wake pairs. The first step consists of vortex core locating utilising the Convolutional Neural Network (CNN), and the second step predicts vortex strength within the Region of Interest (ROI), derived from raw images based on the initial core locating results. This study primarily processes the wake features and builds preconditions for the second-stage wake prediction. In the second stage, the dynamic time-based flight separation in the final approach for avoiding aircraft wake turbulence encounters is studied through long-term wake evolution prediction regarding the spatial-temporal attributes of aircraft wake vortices in their future decay and transport process, utilising the probabilistic sequential prediction models. First, the wake vortex sequences are sectored with the relevant aircraft information, such as the flight speed, heading and aircraft type, and ambient weather information mapped in the final approach. In the offline model training phase, the Attention-based Temporal Convolutional Networks (ATCN) are built and trained using the historical LiDAR dataset to achieve optimal performance. Next, the trained CNN and wake prediction models are fused to achieve long-term wake decay forecasts in actual flight scenarios. Finally, the dynamic aircraft separation minima in relation to wind conditions in the final approach is assessed with wake encounter safety analysis. Furthermore, the model decision processes are explained by feature relevance analysis of both image-based and sequential prediction-based models to enhance the trustworthiness of the deep learning model, and the prediction uncertainty of the model is estimated to improve the robustness of deep learning models. The third part assesses the effect of time-based dynamic wake separation on runway capacity improvement at both the theoretical and operational levels. The dynamic wake vortex separation model that adjusts separation criteria in response to varying atmospheric conditions and flight pairs is trained. The runway sequencing and scheduling model is constructed and solved with the Branch and Bound algorithm under dynamic wake separation matrices. Furthermore, the performance improvement of time-based wake separation is verified under the First-come, First-serve strategy and runway optimisation and compared with the traditional RECAT-EU wake separation standard. The last part of this thesis develops an integrated approach for optimising runway and terminal traffic management by performing dynamic wake separation matrixes. The study evaluates the contributions of time-based and weather-related wake separation in runway operation efficiency and assesses its implications for terminal area operations during high traffic density without compromising safety and ensuring conflict-free airspace. The overall research presents new attempts at improving traffic flow efficiency and safety of the terminal approach and enhancing the runway capacity using reliable and trustworthy data-driven deep learning-supported models and algorithms. We expect to find a comprehensive understanding of the trade-offs between efficiency gains of aircraft separation reduction under dynamic wind situations and safety considerations. This research may facilitate the development of dynamic flight separation indicators in the final approach and the robust, proactive decision-support tool for runway and terminal approach operations to benefit air traffic controllers and airport managers. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13756