Deterministic prediction of vessel motion in real-time using Artificial Neural Network

This paper describes the development of a framework for deterministic vessel motion prediction in real-time using an Artificial Neural Network (ANN), with the measurements of vessel motion and upstream wave elevation as the inputs. Instead of relying on sensitivity analysis to select the ANN training dataset, the location and duration of upstream wave elevation is estimated based on group velocity of the incident waves, given a userspecified prediction time horizon for vessel motion. This method is a form of physics-informed feature extraction that only uses correlated information to train the ANN in order to reduce computational effort. The developed framework is promising, as the ANN trained using physics-based numerical results can provide motion predictions with an error down to 6% for surge, heave and pitch motions of a turret-moored Floating Production Storage Offloading (FPSO) vessel. By applying the ANN training using numerical results onto experimental results of the FPSO, an error of 22% was obtained for pitch motions. At an instant of time, up to 120 s of vessel motion can be predicted in real-time. This physics-informed feature extraction method can be automated as part of the framework for the ANN training and testing pipeline in the future.