[4K3-J-13] AI application: sea and rivers

[4K3-J-13-05] Development of wave prediction method using deep convolutional neural network

〇Tomoaki Itaya¹, Nobuyuki Iwamae¹ (1. Kajima Corporation)

Keywords:wave prediction, weather, convolutional neural network

Wave prediction is important for maritime construction works. This study investigated the possibility of wave prediction by deep convolutional neural networks using numerical weather prediction model data as input. This method predicts wave heights from grid data of mean sea-level pressure p, wind velocity u and v. 4 networks were trained for different forecast lead times 0, 6, 12, 24 hours. As a result, 12, 24 hours networks, in spite of their longer lead times, gave good prediction in terms of RMSE and sensitivity for detection of high waves whose heights were over 1.0 m.