Wave Prediction with Convolution Neutral Network using Airy Wave Fields and Southwestern Coast Movie Clips of Korea

Abstract

In this paper, one of powerful machine learning techniques, the Convolution Neutral Network(CNN), was adopted in order to predict the sea conditions. Initially, the Airy wave fields were produced with linear wave superpositions. Various image filtering methods were applied in long-crested waves data and then the black and white filtering was confirmed as an effective and reliable technique in the research. Several short-crested waves conditions were then trained and predicted the wave height categories. While middle range wave height categories could not be predicted, the highest and lowest categories showed high prediction accuracy, which are representative groups in wave conditions. Lastly, the real ocean environmental images achieved from the wave energy converter of Korea Research Institute of Ships and Ocean engineering(KRISO) were trained with the sea state categories and the average wave height categories. The prediction results showed quite the high global accuracy even the complicated numerical schemes were not applied.