DigiShape seedmoney project
In 2022 and 2023 Witteveen+Bos, Deltares and TU Delft worked together to build a machine learning model that can detect anomalous ship behavior. The project was supported by DigiShape seed money (2022) and carried out by TU student Bart van Engelen.
The project is based on a historical AIS dataset, which contains data on vessel behavior. These data are owned by the Department of Public Works and are confidential, but can be used under certain conditions to analyze ship behavior, for example.
During the study, ‘trips’ (i.e. trajectories of ships) were analyzed from the AIS data, using machine learning algorithms to create different clusters of ship behavior. Think of ships that sailed in a certain direction, had a certain distance to a structure or used a certain underpass.
Because this “normal” vessel behavior has been visualized using machine learning, there is now also potential to find anomalous behavior in the data. Consider near misses due to ships having passed too close to a work of art. During the research, two such anomalous situations were found, in which an accident or near-accident is suspected to have occurred but was not reported to the authorities. Because machine learning allows you to analyze much more data at once than traditional modeling, it is also possible to go back in the data for these two specific cases to see what kind of behavior the vessels exhibited just before the suspicious situation. Think about making an anomalous turn or braking quickly.
Faster identification of near accidents and other anomalous behavior
So by analyzing historical AIS data, near misses, which often go unreported, can be identified more quickly in the future. This helps managers to identify potentially dangerous situations on the waterway in a timely manner. The effects of measures on vessel behavior, such as the placement of new speed signs along the waterfront, can also be visualized in this way.
In order to actually apply the machine learning algorithms that have been developed in practice, some strokes are still needed. The foundation has now been laid (late 2023) with the graduation of Bart van Engelen. In December 2023, a new student will continue with the research, focusing on the North Sea.
Report and code
- More information about the research and the developed code can be requested through Lex de Boom of Witteveen+Bos.
- Also read the interview with Lex de Boom and Bart van Engelen dd February 15, 2023
