Regional distributed power generation prediction method based on federal cloud model

The invention discloses a regional distributed power generation prediction method based on a federal cloud model, and relates to the technical field of power data prediction, and the method comprises the following steps: constructing a historical data feature set of a distributed power supply in a regional manner, carrying out the feature similarity screening of data features in the historical data feature set of the distributed power supply in a regional manner through a cloud model, and carrying out the prediction of the feature similarity. Weighting feature sets of different dimensions by adopting a multi-dimensional cloud feature weighting method, screening out five days with the highest similarity with a prediction day feature set as a training set, and then predicting data features in the training set by utilizing an LSTM (Long Short-Term Memory) long-short-term neural network to obtain a distributed power supply output prediction value; the total server sends a federation training request, each region