The motion response of an offshore floating wind turbine (FOWT) platform is closely related to the control operation regarding the safety of a wind turbine. It is affected by various factors such as sea state environments and mooring systems. In practice, how to predict the motion response of the wind turbine platform in the short term has always been a concern of engineering practice. At present, the development of deep learning technology has brought some potential solutions to this problem. In this paper, a Multi-Input Long-Short Term Memory (MI-LSTM) neural network method is proposed to predict the short-term motion response of a floating offshore wind turbine platform. Specifically, the numerical simulation of the 5 MW Braceless platform is carried out under different environmental conditions, and the data of platform motion response, wave elevation, and mooring force are selected as input variables. Then the training and test groups are established after post-processing data. Subsequently, a Single-Input LSTM (SI-LSTM) model and a Multi-Input LSTM (MI-LSTM) model are established to learn the input data. After comparing the overall accuracy of the results, it is found that the additional mooring force and wave elevation positively affects the platform response prediction results. From the aspects of discreteness and overall accuracy, it is verified that the established MI-LSTM model is also applicable, considering the influence of second-order hydrodynamics. Lastly, compared with the prediction results obtained by the multi-input one-dimensional convolutional neural network (MI1D-CNN), the advantages of the two different models are expounded from the perspectives of training time and accuracy, which provides ideas for the optimization of the FOWT motion response prediction model. This study sheds insights on the short-term motion response forecast and platform positioning of a FOWT. Short-term forecasts of a FOWT can be achieved under various sea conditions by combining the global positioning system.
Read full abstract