Ship trajectory prediction using encoder–decoder-based deep learning models

Abstract

ABSTRACT Accurate prediction of ship trajectories can be an important capability for various maritime transport applications, such as vessel traffic services (VTS), traffic flow assessment, and collision avoidance systems. The widespread availability of AIS (Automatic Identification System) data and the progress made in the deep learning methods in the last decade motivate us to attack this problem from a data-driven perspective. This paper presents the results of a study where various encoder–decoder architectures were applied to the ship trajectory prediction problem using AIS data that were collected from the Rotterdam port approach area. The models were trained with the AIS data along four routes belonging to different ship types/lengths without any clustering or filtering. An average position RMSE of 1.6 km was obtained when the best-performing model predicts ship positions 30 min into the future using 60 min of historical data.