Efficient Prediction of Mooring Forces of a Floating Offshore Wind Turbine Using LSTM-Based Deep Learning

Abstract The relationship between rigid body motion and mooring line forces of moored floating offshore structures is highly nonlinear. Physical experiments and numerical simulations can be used to determine the relationship between the floater motions and mooring forces. However, they are typically time-consuming and expensive for design space exploration. We present an efficient Long Short Time Memory (LSTM)-based deep-learning technique to predict the mooring forces on a floating offshore wind turbine (FOWT) with a catenary mooring system. Dropout, leaky linear rectification, and swish layers approximate the mapping between the motion time series and the mooring force time series. The FOWT is subjected to irregular waves based on the JONSWAP spectrum and the measured surge motion is fed into a deep neural network. The LSTM is trained using a stochastic gradient descent approach to forecast the time series of mooring forces of the FOWT, and the results are compared with the simulation data. Within the error threshold, the computational time of our deep network approach, albeit using few computational resources, is shortened by almost two orders of magnitude as compared to the physics-based approach. We also present a novel approach to obtaining real-time predictions for all considered sea-states without modifying the LSTM neural network. Overall, the suggested LSTM-based approach could be used as a proxy for parametric design and digital twinning of FOWTs.