Maritime traffic probabilistic prediction based on ship motion pattern extraction

This paper proposes a novel maritime traffic prediction method based on ship motion pattern extraction, considering ship destination prediction and ship trajectory prediction within a specific route. To extract ship motion patterns from historical Automatic Identification System data, traffic departure-arrival areas are determined based on the Order Points to Identify the Clustering Structure algorithm and ship trajectories following the same itinerary are clustered. A maritime traffic network abstraction consisting of nodes that represent waypoint areas and navigational legs is constructed to represent the maritime traffic at a larger scale. Multinomial Logistic Regression and Gaussian Process regression models are developed and applied for predicting probabilistically the ships' destinations and their trajectories along the ship route, respectively. Based on these models, the uncertainty on the ship's future position can be estimated given its current state. The proposed method is capable of long-term ship position prediction and provides information on the maritime traffic 10, 30 and 60 min ahead when the method is applied to all ships navigating in a study area. The presented method may assist maritime authorities to improve the efficiency of maritime traffic surveillance and to develop strategies to improve navigation safety.