Machine Learning and Deep Learning Introduction to Route Planning : data-driven maritime route planning methods, technologies, and applications

Abstract

This thesis explored the integration of Machine Learning (ML) and Deep Learning (DL) into maritime route planning, focusing on the use of Electronic Navigational Charts (ENC) in S-57 format. The work was driven by practical challenges encountered during voyage planning and the need to improve safety, efficiency, and compliance within modern maritime operations. A project-based methodology was used to develop a full pipeline for data extraction, processing, graph construction, and route calculation, supported by open-source tools such as GDAL, PostGIS, NetworkX, Uber’s H3 and Python libraries. Traditional pathfinding algorithms such as A* and Dijkstra were implemented and evaluated using real ENC data, and an experimental Q-Learning model was developed to assess the feasibility of Reinforcement Learning in maritime routing. The results showed that classical algorithms remain effective for structured graphs derived from ENC layers and that while ML and DL approaches offer potential for future development, they are not yet viable for standalone deployment due to challenges in data representation, training complexity, and route reliability. The practical prototype ‘Route Assistant’ was successfully implemented to demonstrate the capabilities of the developed system, including ENC visualization, feature extraction, and graph-based routing. The study concludes that while full automation is not currently achievable, the foundation laid by this work presents a valuable tool for supporting navigation officers, accelerating planning processes, and serving as a base for further development in autonomous and AI-assisted voyage planning systems.