Prediction of dynamic mooring responses of a floating wind turbine using an artificial neural network

Abstract

Numerical simulations in coupled aero-hydro-servo-elastic codes are known to be a challenge for design and analysis of offshore wind turbine systems because of the large number of design load cases involved in checking the ultimate and fatigue limit states. To alleviate the simulation burden, machine learning methods can be useful. This article investigates the effect of machine learning methods on predicting the mooring line tension of a spar floating wind turbine. The OC3 Hywind wind turbine with a spar-buoy foundation and three mooring lines is selected and simulated with SIMA. A total of 32 sea states with irregular waves are considered. Artificial neural works with different constructions were applied to reproduce the time history of mooring tensions. The best performing network provides a strong average correlation of 71% and consists of two hidden layers with 35 neurons, using the Bayesian regularisation backpropagation algorithm. Sea states applied in the network training are predicted with greater accuracy than sea states used for validation of the network. The correlation coefficient is primarily higher for sea states with lower significant wave height and peak period. One sea state with a significant wave height of 5 meters and a peak period of 9 seconds has an average extreme value deviation for all mooring lines of 0.46%. Results from the study illustrate the potential of incorporating artificial neural networks in the mooring design process.