Prediction of ship trajectories is an important AI application that can support vessel traffic management, especially given the digitization of shipping traffic and its accompanying data. Via machine learning techniques, intelligent systems can learn from historical trajectories and conditions. The trained prediction models can provide vessel traffic operators, captains and pilots on ships, insights into upcoming changes in the traffic situation, and into possible conflicts.

Machine learning models

For safe and effective use of such systems, it is essential to understand these techniques and the accuracy of the predictions. In the data-driven trajectory prediction project, we therefore built and tested machine learning models that predict ship trajectories 30 minutes into the future, based on the track positions of the last 60 minutes. The models all follow an encoder-decoder architecture. We focused on a versatile traffic hotspot: the approach area to the Port of Rotterdam. The figure above shows actual ship trajectories in the approach area.

More information on the models can be found in the following paper:
Ship trajectory prediction using encoder–decoder-based deep learning models
https://doi.org/10.1080/17489725.2024.2306339.

For a complete project description please see download below.