Problem outline
The growth of shipping on Flemish waterways is good for the economy and for the environment, but it also poses challenges, such as saturation. Moreover, scheduling in terms of infrastructure and personnel is becoming more complex. The responsible authorities have therefore developed systems that help to optimise scheduling. However, unlike route planning for cars, these do not use machine learning.They use statistical averages to predict arrival times at the quayside, which can lead to large discrepancies between the predicted and actual arrival times. These deviations can result in economic losses.
Central research question
How can we more accurately predict arrival times of inland vessels, by drawing up a model that takes several parameters into account?
Research method
We use machine learning to find connections between historical ship movements (AIS data) on the one hand and predefined parameters on the other hand. Based on this, we develop an algorithm that predicts the arrival time. We compare this arrival time with the prediction of a route planner without AI (VisuRIS) on the one hand, and with the actual arrival time on the other hand, by analysing the AIS data of the route travelled after the arrival of the ship.
Expected output
The desired output of the research project is a specific algorithm (e.g. a Bayesian network) for ship route planning, which allows more accurate predictions than the existing route planner of the Flemish government (VisuRIS), which does not use AI.