Weakly Nonlinear Surface Wave Prediction Using a Data-Driven Method With the Help of Physical Understanding

Abstract

Accurate surface wave prediction can potentially improve the safety and efficiency of various offshore operations, such as heavy lifts and active control of wave energy converters and floating wind turbines. Prediction of surface waves, even if only for a few periods in advance, is of value for decision-making. This study aims to predict weakly nonlinear surface waves (up to the second-order) in real-time using a data-driven model based on artificial neural networks (ANNs), where the application of physics is investigated to aid the development of a data-driven model. Based on numerically synthesized nonlinear wave records calculated using exact second-order theory, ANN models were trained to separate the nonlinear bound components at an up-wave location, propagate the linear waves, and reintroduce the nonlinear components as a correction to the prediction at a down-wave location. Our findings indicate that the optimal approach is to predict each stage separately following the basic physical structure of weakly nonlinear water waves using a series of ANN rather than direct prediction in a single step using ANN. Furthermore, we examined the generalization of the models across different sea states and investigated the impact of the second-order bound waves on prediction accuracy.