Studies On The Regional Ocean Wave Prediction Based On Deep Learning

Ocean waves have long affected human activities such as navigation and trade,ocean engineering,and fishing.Accurate forecasting of waves can effectively reduce the probability of marine accidents and economic losses.At present,wave forecast mainly relies on evolving numerical models.In recent years,with the continuous integration of artificial intelligence and the ocean field,deep learning has provided a new approach for wave forecast.This paper selected two typical sea areas,namely Domain1(130～160°E,0～30°N)in the Northwest Pacific Ocean and Domain2(100～130°E,0～30°N)in the offshore waters.Based on the wave,10 m wind field and water depth topographic data from 1981 to 2020,a 3-day forecast of regional waves based on deep learning was launched.For Domain1 in the open sea area,we improved the convolutional long short-term memory neural network(Convl STM)from the physical process,and proposed the more interpretable Opensea-Conv LSTM to simulate the 3-day development process of the initial waves driven by wind,whose prediction effect was better than Conv LSTM.The annual mean absolute error(MAE),root mean square error(RMSE),mean relative error(MRE)and correlation coefficient(CC)of the Opensea-Conv LSTM forecast of significant wave heights above 1 meter are 0.201 m,0.302 m,10.4% and respectively0.880,the annual MAE of forecast mean wave period and mean wave direction is 0.49 s and 17.8°.For Domian 2 in the offshore area,considering the important role of water depth based on the Opensea-Conv LSTM,Offshore-Conv LSTM was proposed to simulate the3-day development process of initial waves under the influence of wind and water depth topography.The prediction effect is better than Opensea-Conv LSTM.Offshore-Conv LSTM predicts the annual MAE,RMSE,MRE and CC for significant wave heights above 1 meter to be 0.224 m,0.323 m,12.6% and respectively 0.904.The annual MAE for forecast mean wave period and mean wave direction is 0.44 s and 17.7°.In addition,when using deep learning to forecast ocean waves in the open sea and offshore,the effect is best in winter,followed by spring and summer,and worst in autumn.