Phase-resolved wave prediction with varying buoy positions in the field using machine learning-based methods

Surface wave predictions for several wave periods in advance are crucial for optimizing a wide range of offshore applications. This work focuses on the potential application in active control of wave energy converters (WECs), which can dramatically enhance the efficiency of power generation. Field tests were conducted in the Southern Ocean of Albany, Western Australia. We compared two prediction models: a physics-based algebraic model and a machine learning-based Artificial Neural Network (ANN) model. Although the standard ANN model is found to achieve better prediction accuracy than the algebraic model for highly directional spreading waves, the prediction performance of the model is greatly reduced due to varying buoy positions, leading to phase offset in predictions. To overcome this limitation, phase correction and partition methods have been incorporated with physical insights into a purely data-driven ANN model. The modified ANN model significantly reduced the prediction error in peaks and troughs compared to the standard ANN model. This study demonstrates that both physics-based and machine learning-based models can work in parallel to provide more accurate predictions, thereby enhancing the practical value of wave prediction for WECs.