Wave-dependent predictability of floating offshore wind turbine responses : a BiLSTM study based on fully coupled CFD simulations

Abstract

Reliable short-term prediction of floating offshore wind turbine (FOWT) responses under complex wave conditions remains challenging due to nonlinear aero–hydro–mooring interactions and transient wave-induced effects. This study evaluates the predictability of coupled FOWT responses using a Bidirectional Long Short-Term Memory (BiLSTM) framework trained on high-fidelity datasets generated from a fully coupled aero–hydro–mooring computational fluid dynamics (CFD) model of the National Renewable Energy Laboratory (NREL) 5 MW OC4 semi-submersible system. Two excitation conditions are examined: regular waves representing periodic steady-state behavior and focused waves representing transient amplified responses. The model simultaneously predicts platform motions, mooring-line tensions, aerodynamic power, and total thrust. Hyperparameter optimization is performed to ensure stable convergence and robust model performance. Predictability is assessed across multiple prediction-ahead times (PATs). Results show that regular-wave responses maintain high accuracy at longer horizons (R² > 96% at 2.5 s and 5.0 s), whereas focused-wave cases exhibit decreasing accuracy with increasing PAT, achieving R² values above 95%, 90%, and 85% at 0.5 s, 1.0 s, and 1.5 s, respectively. These findings demonstrate that forecasting performance strongly depends on wave type, emphasizing the need to consider wave conditions when predicting coupled FOWT dynamic responses.