A new Gaussian Process based model for non-linear wave loading on vertical cylinders

We aim to establish a fast and accurate model for fast prediction of nonlinear loading on vertical cylinders such as are typically used for fixed offshore wind turbines. We follow a ‘Stokes-type’ force model and approximate the amplitude of the higher harmonics of force by relating these to the linear force time series raised to appropriate power through amplitude and phase coefficients. We reanalyse previous experimental data and perform new experiments to expand the parameter space and establish a force coefficients database for engineering applications. A machine learning model is used to interpolate the database and make predictions on the higher order force coefficients. The machine learning model also provides a cross-validated confidence interval to indicate the prediction uncertainty and reflect model reliability. We further extend the prediction capability to unidirectional random waves with a novel force segmentation method, which localised wave groups from the random background. The new Stokes-Gaussian Process (Stokes-GP) model developed can provide engineering predictions of nonlinear wave loading on a cylinder for individual wave groups and random seas, which are straightforward to apply and fast to compute and the important higher-order loading components are considered. This will significantly improve the accuracy of the loading prediction and the ease of application for force predictions.