Phase-resolved wave prediction model for long-crest waves based on machine learning

Abstract

Phase-resolved wave prediction is important to a vessel for both predicting deterministic motion and assisting decision-making. In this study, both linear and nonlinear physical-based wave models are addressed for achieving phase-resolved wave prediction. These physical-based models are developed based on the fast Fourier transform (FFT) technique. Although nonlinear wave models such as high order spectrum models are capable of nonlinear wave modeling, they still experience nonstationary difficulties. The FFT technique requires wave motions to be linear and stationary. Accordingly, inspired by the capability of machine learning in solving nonlinear dynamic problems, a machine learning method is applied in phase-resolved wave prediction. Thus, an artificial neural network-based wave prediction (ANN-WP) model is proposed. Experimental wave data sets are employed in model training optimization, verification, and comparison studies. The prediction accuracy of the ANN-WP model is verified and the input–output strategy is explored for training optimization of the ANN-WP model. A comparison between the ANN-WP and linear wave prediction (LWP) models is presented using the nonlinear experimental wave data. Furthermore, a preliminary discussion of the extreme wave predictive ability of the ANN-WP model is presented. The comparison results indicate that the ANN-WP model performs better than the LWP model. Although selected problems remain in extreme wave prediction, the proposed ANN-WP model provides a feasible and effective approach for achieving accurate nonlinear phase-resolved wave prediction.