Waves that remain at sea or come from other sea areas after the wind weakens,stops,or turns are called swells.Swell has a lower wave level than the wind wave,but the period is larger.Swell is easy to resonate with floating structures such as ships,which affects the normal operation.In order to avoid the damage caused by swells,it is necessary to have a method that can predict swells efficiently and accurately.A neural network was built based on the Convolutional LSTM(ConvLSTM)model to process the wave height distribution of swells was a two-dimensional image.The network combines the image feature capture ability of convolution operation and the time series prediction ability of Long Short Term Memory(LSTM)model,and considers both the spatial propagation characteristics of swell and the temporal variation characteristics in the network learning process.The ERA5 reanalysis dataset was used to train the network,and the significant wave height of the swell in the East China Sea(21°N—34°N,114°E—131 °E)was predicted,and the prediction results were in good agreement with the dataset,with a maximum correlation of 0.997.At the same time,compared with the model that does not consider the spatial propagation characteristics of swells,the prediction effect is improved.The method used in this paper provides a new idea for the research on swell forecasting.
万方数据
维普
