Boiler wall temperature prediction method of multi-time map convolution attention network

The invention provides a boiler wall temperature prediction method based on a multi-time map convolution attention network. The method comprises a data screening and decomposition module and a hybrid prediction module. According to the method, the boiler temperature in historical data and related variables influencing the temperature serve as input, after data screening and decomposition, a part of components are input into a gating circulation unit, a Transform and a time convolution network for prediction, and the prediction result is obtained. And the other part of components are used for drawing a graph in a graph form of 400 * 400 pixel points by using the data of the components, and the graph is input into a residual channel attention network and ShuffleNet for prediction. And finally, adding the predicted values with the weight matrix to obtain a final output predicted value by taking the minimum loss function as a target. According to the method, the function of accurately predicting the temperature o