Prediction of mooring dynamics for a semi-submersible floating wind turbine with recurrent neural network models

Abstract

Offshore wind technology emerges as a promising solution to tap into wind resources in shallow and deep waters. For floating wind turbines, mooring systems provide station-keeping functionalities. Efficient monitoring of mooring loads is critical for safe and cost-effective operation and maintenance. As direct measurement of the dynamic mooring line tension is expensive, alternative means are needed. To this end, this article focuses on predicting mooring tensions based on accessible motion data of the floating platform and proposes utilizing deep learning algorithms, i.e., recurrent neural networks (RNNs). We selected three RNN algorithms: Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM). These algorithms were trained with various load cases of a 10-MW floating wind turbine and their efficacy was assessed using unseen data. Among the three algorithms, the BiLSTM algorithm performed best, achieving $R^2$ of more than 0.97 under various load conditions. Further, a statistical assessment of the tension time series demonstrates an excellent agreement between the predicted and the measured mooring line tensions with a percentage difference in the peak tension of less than 0.5%. The outcomes of this paper contribute to real-time prediction and structural health monitoring of floating wind turbines.