Enhancing Wave Prediction Using Machine Learning: Modelling Nonlinear Components in 3D Wave Buoy Motion

Abstract Accurate wave prediction is essential for advancing predictive control in offshore renewable energy applications. Modern wave buoys provide semi-Lagrangian time histories in three dimensions, enabling precise wave-by-wave predictions. However, recent findings suggest that while wave buoy motion is predominantly linear, notable nonlinearities - particularly in the horizontal plane - can impact prediction accuracy. Effectively understanding these nonlinear components is critical for achieving real-time active control of devices such as wave energy converters and floating offshore wind turbines. In this study, we propose a novel method to extract high harmonic components from the measured signal, enhancing the accuracy in wave-by-wave predictions. By employing machine learning-based models, we predict second-order sum and difference components derived from linear wave signals. This work introduces a comprehensive machine learning framework to assess the feasibility of this approach, optimising data requirements and refining model structures to ensure robust results. Field data from a Datawell buoy are used to establish optimal data and modelling parameters, demonstrating that accurate, real-time wave prediction is achievable. These advancements hold significant potential for improving control strategies and operational efficiency in marine and renewable energy applications.