Abstract Accurate in-situ wave measurement is crucial for the safe and efficient operation of offshore floating platforms. However, the presence of a platform significantly perturbs the local wave field through complex wave-structure interactions, including wave diffraction and radiation, making direct measurement of the undisturbed incident waves a significant challenge. The relationship between the platform's hydrodynamic responses (air-gap and 6-DOF motion responses) and the incident wave field constitutes a complex, nonlinear inverse problem. Traditional linear methods often struggle with the strong nonlinearities inherent in this relationship, especially under severe sea states. This paper proposes an improved methodology for decoupling undisturbed incident waves from near-field measurements based on a dual-frequency convolutional neural network (CNN). By separating the hydrodynamic responses into wave-frequency and high-frequency components and training the network with distinct datasets (irregular waves and white noise waves) for each, our approach significantly enhances the model's generalization capabilities and accuracy. The proposed dual-frequency CNN effectively reconstructs the incident wave time series, with the standard deviation of the wave-frequency components achieving accuracies within 3% of the ground truth. Furthermore, the model demonstrates robustness against measurement noise, highlighting its potential for practical deployment on operational floating platforms for near-field wave sensing.
Zhang et al. (Mon,) studied this question.