Multivariable time sequence prediction method based on spatio-temporal embedding independence

The invention provides a multivariable time series prediction method based on spatio-temporal embedding independence, and belongs to the technical field of multivariable time series prediction. The method comprises the following steps: acquiring multivariable time sequence data as sample data, constructing an independent multivariable time sequence prediction model based on space-time embedding, training the prediction model by using the multivariable time sequence data, updating the weight of the model, inputting power load data of the multivariable time sequence into the prediction model, and outputting a prediction result. According to the method, the accuracy of multivariable time sequence prediction is remarkably improved; the calculation overhead of the model is remarkably reduced, and meanwhile, the problem of overlarge state vector scale caused by mutual coupling between cross-time dependent modeling and cross-dimension dependent modeling and the problem of overfitting caused by explicit cross-dimensi