Motion Response Prediction of Deep-Sea Floating Wind Turbines With the NLinear Model Optimized by the Optuna Framework

Abstract Floating wind turbines are core equipment for developing deep-sea wind energy resources. Accurately predicting their six degrees-of-freedom (6DOF) motion response is crucial for load fluctuation control and proactive structural health monitoring. This study proposes an Optuna-based hyperparameter optimization framework for the NLinear model (OP-NLinear) to perform multi-step prediction of the motion response for a 15 MW deep-sea floating wind turbine. Twelve typical operating conditions are selected to construct a 6DOF motion response dataset. The NLinear model is employed to extract trends and patterns from the motion response time series, while Optuna enhanced its generalization capability by optimizing model hyperparameters. Results demonstrate that when predicting the 6DOF motion response over the next 15 time-steps (15 s), OP-NLinear outperforms benchmark models such as OP-Bi-LSTM, OP-DLinear, and OP-XGBoost across all operating conditions, achieving an average coefficient of determination (R2) of 0.922. Even when predicting motion responses over 20 time-steps, the average R2 value remained at 0.895. Furthermore, the model exhibits strong noise immunity, demonstrating “intelligent robustness” against 20 dB colored noise. The proposed OP-NLinear method provides a reliable and efficient solution for predicting the 6DOF motion response of deep-sea floating wind turbines, laying the foundation for developing related active control strategies.