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

Phase-resolved real-time ocean wave prediction with quantified uncertainty based on variational Bayesian machine learning

“…Various NN structures in deterministic ML can also be used in the BML framework to take advantage of the underlying data structure, such as convolutional NN for image classification [52] and recurrent NN for language processing [53]. In this work, as was in previous NN-based wave prediction works [36,37], the fully-connected NN is employed, which makes full connection from the model input to the model output. The input-output mapping f𝑊 , which is illustrated in Fig.…”

“…In addition, the predictable zone cannot be obtained. The previous NN-based works [36,37] all employed LWT for the predictable zone determination, as described in Section 3.2.…”

“…The comparison results showed that the proposed method clearly outperformed the LWT-based approach. To summarize, the works in [36,37] both demonstrated the great potential of the ML-based approaches in nonlinear wave predictions.…”

“…The proposed method was evaluated using numerical simulations and the results showed that it performed much better than the LWT-based approach. In [37], a similar NN-based wave prediction method was proposed for phase-resolved wave forecasting. The proposed method was evaluated using wave tank experiments and a set of input-output strategies were tested to improve the prediction accuracy.…”

“…(1) Phase-resolved real-time nonlinear wave prediction with quantified uncertainty (including both aleatory and epistemic uncertainties) is achieved for the first time. Because the previous ML-based works [36,37] achieved deterministic phase-resolved wave predictions without quantifying the prediction uncertainty, it is difficult to use them for the control of WECs in realworld conditions, as it may lead to catastrophic damage. This work, therefore, brings vital opportunities for the safe and optimal operation of WECs.…”