Abstract
Real-time wave prediction is used to determine the feasibility of marine construction work, but the numerical models currently in use have the disadvantage of being computationally expensive. Recently, a prediction model using machine learning has been proposed, which enables prediction at low computational cost. However, because observed data is used as the teacher data for machine learning, it is not possible to make predictions at locations different from the observation points, making it difficult to use machine learning to make decisions on marine construction work, which requires wave prediction at arbitrary points. Therefore, we developed a real-time wave prediction method that can predict waves at arbitrary locations by using a data assimilation method to create highly accurate spatial wave distributions from numerical model results and using them as teacher data. The results of wave prediction in Tsuruga using this method showed that the method was able to quickly obtain predictions that were more accurate than numerical models, confirming the usefulness of the method.
