Regional sea surface temperature prediction method in combination with attention mechanism

The invention discloses a regional sea surface temperature prediction method combined with an attention mechanism, which comprises the following steps of: 1) processing SST data every day in a regioninto a matrix, and sequentially arranging the matrixes according to a time sequence to form a matrix sequence serving as the input of a CA-ConvLSTM model; 2) processing the SST matrix, and extractingdistribution characteristics of each recording point through a convolutional layer; 3) distributing attention weights for the obtained matrix features by utilizing an attention mechanism, and then multiplying the attention weights by the corresponding matrix features to obtain weighted features; and 4) finally, taking the weighted features as the input of a ConvLSTM model, and training the prediction model by using the ConvLSTM model to finally obtain a prediction result. The SST in the region is arranged into a matrix, and is input into the model as a whole, so that the time and space correlation of the SST can be conveniently extracted by the model.