EEMD-ConvLSTM: a model for short-term prediction of two-dimensional wind speed in the South China Sea

Accurate prediction of sea surface wind speed is crucial for marine activities such as marine search and rescue, marine shipping, and marine fishing. Because of the gustiness of sea surface winds, the wind speed data have strong non-stationarity and non-linearity, and it is still challenging to predict sea surface winds accurately and stably in a short time. Most previous studies have only considered the problem of non-smoothness of wind speed in single-point wind speed prediction, and this study extends the method to solve the non-smoothness of data in two dimensions. We proposed a hybrid model based on ensemble empirical mode decomposition (EEMD) and Convolutional long short-term memory network (ConvLSTM). Specifically, this study employs the Pearson correlation coefficient (PCC) to select the most periodic and most predictable subsequences of the EEMD signal decomposition, then fuse them into the spatio-temporal prediction of ConvLSTM. The method eliminates the influence of noise in the two-dimensional wind speed prediction and grasps the wind speed variation pattern more accurately. The proposed model has the best prediction effect based on the experimental findings, and this advantage becomes increasingly evident as time increases. This shows that EEMD signal decomposition is a good way to solve the short-term prediction problem for two-dimensional wind speed.

Automated summary

This study proposes a hybrid model based on ensemble empirical mode decomposition and Convolutional long short-term memory network to solve the non-smoothness problem in sea surface wind speed prediction. Experimental findings show that this model has the best prediction effect, and this advantage becomes increasingly evident as time increases.