Evaluating the potential of Deep-Q Networks for weather avoidance in aviation

Posker, Adam (2025) Evaluating the potential of Deep-Q Networks for weather avoidance in aviation. Bachelor's Thesis, Computing Science.

Preview	Text bCS2025PoskerA.pdf Download (3MB) | Preview
	Text Toestemming.pdf Restricted to Registered users only Download (182kB)

Abstract

Adverse weather conditions pose considerable challenges to aviation, affecting safety, efficiency, and passenger comfort. The current approach to weather avoidance in aviation is dependent on pilots' interpretation of weather forecasts from various channels, potentially causing inconsistencies in the ways the weather is avoided. Deep Reinforcement Learning offers a promising alternative to traditional automation methods, enabling agents to learn optimal actions through repeated interactions with simulated environments. The project adapts the Deep Q-Network algorithm, building on agent architecture originally designed for aviation conflict resolution, to navigate simulated aircraft around inclement weather. During evaluation, the trained agents managed to avoid all adverse weather in the majority of simulation runs. Despite challenges with training stability and occasional unrealistic flight paths, the results demonstrate the potential of Deep Reinforcement Learning—specifically the Deep Q-Network algorithm—in automating aircraft navigation around inclement weather. Ultimately, the project provides a foundation for integrating the presented approach with other domains of air traffic control, such as conflict resolution, weather forecasting, and more.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Wilkinson, M.H.F. and Weerd, H.A. de
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	15 Jul 2025 06:43
Last Modified:	15 Jul 2025 06:43
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/35800

Actions (login required)

View Item