Tugboat operations manage the safe and efficient handling of vessel movements in and around ports. Accurate detection of these operations is necessary for the monitoring, planning, coordination and optimization of port-related activities. As maritime traffic continues to grow, there is a need for scalable and automated methods to detect and analyze these interactions based on available AIS data. The ability to systematically map tug activity across ports worldwide has commercial relevance. For maritime service providers, this facilitates large-scale analysis, operational planning, resource allocation and market development. In particular, identifying patterns of inefficient tug use offers actionable input for fleet optimization tools, such as those provided by KOTUG Optiport.


% Gap / Problem Statement
Existing detection methods based on AIS (Automatic Identification System) data rely on heuristic, rule-based logic and typically require predefined port zones. This limits their use in unstructured environments and prevents global deployment. Most AIS-based machine learning models analyze vessels in isolation and do not capture ship-to-ship interactions, which are central to identifying tugboat operations. This work addresses that gap by explicitly modeling the interaction between tug and assisted vessel as input, enabling the detection of relational behavior. In addition, it targets the specific trajectory segments where towing occurs, rather than labeling entire vessel tracks.


% Purpose / Objective
This thesis aims to develop a scalable, data-driven machine learning pipeline that can detect and classify tugboat operations worldwide by explicitly modeling ship-to-ship interactions using AIS data. The goal is to move beyond static, zone-based rules and build a method that uses dynamic behavioral data to identify interactions. The final objective is to implement this pipeline as a system that runs continuously and can be used in any maritime region across the world.