Mpc Based Anomaly Detection of Vessel Routes Using Ais Data

Detecting anomalies in maritime vessels routes can anticipate potential collisions and prevent illicit activities. Common methods in the literature employ Automatic Identification System (AIS) data to build models of normal vessel behavior. The predictions of the models are then compared with actual data to detect discrepancies in vessel motion patterns. Datadriven or geometrical approaches can be used to build such models, requiring however a large set of historical data and lacking in interpretation. In this paper, we propose a Model Predictive Control (MPC)-based anomaly detection framework that aims at overcoming the aforementioned issues. Results on publicly available real AIS data with artificially simulated anomalies show the benefits of the proposed physics-based MPC approach when compared with a geometrical-based approach.