The phase-resolved wave prediction is of vital importance for improving the safety and operations of structures in ocean engineering, such as floating wind turbines and wave energy converters. However, it is challenging for traditional physical methods to predict wave elevation accurately and efficiently in high sea states where the nonlinearity plays an important role. Inspired by the capability of the Koopman theory in solving nonlinear problems, a phase-resolved wave prediction model by combining deep learning with Koopman operator, referred to as DKWP, is designed. The high-order spectral method is employed for providing training and test data. To illustrate the effectiveness of the proposed method, we compare it with another four phase-resolved wave prediction models, including the models based on artificial neural network, long short-term memory, U-Net and linear wave theory. The models are trained under the sea states 3–5 with wide ranges of significant height, peak period and peak factor. Comparisons between the predicted wave elevation and true wave elevation indicate that DKWP outperforms the other models in terms of accuracy. The generalization of the DKWP is validated when it predicts the wave elevation under the sea states 6–7. In addition, to improve the interpretability of the DKWP, the eigenvalues of the Koopman operator in the proposed model are further analyzed.
扫码关注我们
求助内容:
应助结果提醒方式:
