A deep operator network method for high-precision and robust real-time ocean wave prediction.

This article focuses on the development and evaluation of a real-time wave prediction model based on the Deep Operator Network (DON-WP), which learns nonlinear operators to map historical wave heights to future wave heights. Using experimental wave tank data from multiple sea states, the DON-WP model demonstrated superior generalization, accuracy, and robustness compared to a Long Short-Term Memory (LSTM)-based probabilistic prediction model (Deep-WP), achieving up to 80.86% and 77.11% reductions in mean absolute error (MAE) and normalized root mean squared error (NDRMSE), respectively. The DON-WP model requires training on only one sea state to generalize effectively to others, unlike Deep-WP which needs retraining for each condition, and it provides rapid predictions suitable for real-time offshore renewable energy optimization and floating structure safety. The study also identifies optimal model hyperparameters and historical time windows for different prediction horizons, highlighting the model's potential for advancing offshore wave dynamics modeling while noting future improvements such as integrating attention mechanisms and physical constraints for enhanced interpretability.