Short Term Ship Motion Prediction from Deep Learning Models and Past Inertial Measurements

In the strive for always more autonomous, efficient or safe maritime operations, the ship motion prediction plays a pivotal role and remains by far one of the most challenging problem to solve. The random nature of ship motions at sea due to the exciting waves conditions directly the success of many marine operations such as the landing on aircraft carriers. Elaborating accurate short term forecasts related to both ship attitude and position is of great interest for operational decision-making. To this aim, this paper describes an original training strategy, based on representative simulations of ship dynamics, which permits to design deep learning models as potential problem solution. The proposed methodology is assessed and validated by inferring the trained models on real sea trials data collected on a generic vessel.