Predicting the trajectories of fishing vessels
The fishing industry is very competitive and highly regulated. The Danish Fisheries Agency handles mandatory reporting from Danish vessels, perform inspections and has the authority to impose penalties in response to irregularities. Bringing down the irregularity rate will benefit the fish stocks and the fishing industry as a whole. An element in bringing down irregularity rate is trajectory prediction as this can improve existing risk profile scores, make vessel localization easier for inspection activities and help predict fishing activities of a vessel. The challenge is to predict the trajectory of a vessel a number of time steps ahead in order to allow for proactive actions.
The solution is an interactive sea map displaying an active vessel's travelled trajectory and predicted trajectory on demand. This relies on the steady flow of locational data from each vessel, provided by the AIS system. The predictions indicate the most likely trajectory along with an illustration of the uncertainty of the prediction. The already travelled route is used as the input on which the predictions are based.
In order to predict the trajectories of fishing vessels a deep learning model for handling time series data is employed. The type of neural network suitable for time series data is a Recurrent Neural Network which learns typical fishing vessel trajectories by being exposed to an abundance of past travelled trajectories and trying to predict the next step in each of these.
The problem of trajectory prediction is a general one which arises in many contexts. Other domains with time-varying data could employ this approach in order to learn from typical patterns in the data to do predictions.
Predicting the trajectories of fishing vessels has the potential to increase the relevancy of the inspections performed, identify critical situations before they occur, act proactively and thus bring down the irregularity rate. The trajectories can also be used in predicting the current and future exhaustion of the relevant quota and therefore a sharper deadline for when the fishing activities for a given species need to stop.