Short-term load prediction method based on time sequence prediction and deep learning complementation

The invention discloses a short-term load prediction method based on time sequence prediction and deep learning complementation, and relates to the technical field of load prediction. The method comprises the following steps: firstly, carrying out stability test on a historical load data sequence by utilizing ADF test and KPSS test, and if the historical load data sequence is not stable, carrying out differential processing until a signal is stable; and secondly, a periodic mutation mechanism is introduced into a dung beetle optimization algorithm to improve the optimization ability of the algorithm, and the improved dung beetle optimization algorithm is utilized to perform order determination on parameters p and q of the differential integration moving average autoregression model to realize load prediction by the ARIMA. And then, predicting the nonlinear residual error sequence predicted by the ARIMA by using LSTM (Long Short Term Memory). And finally, the obtained load prediction result is verified, so tha