Abstract: Aiming at the very short-term electric load forecasting, a combined model is proposed, which uses EEMD(ensemble empirical mode decomposition )and SE(sample entropy) to preprocess the original data, and then applies SA(simulated annealing) to optimize the deep belief network for forecasting. In order to reduce the time-series data of the predicted value behind the real value caused by the autocorrelation of the data, the original sequence is decomposed by EEMD, the sequence is reconstructed according to the SE of each decomposed sequence, and the reconstructed sequence is predicted separately by SA-DBN model composed of SA optimizing the number of nodes in each hidden layer of DBN, and the predicted results of each sequence are superimposed to obtain the final prediction curve. The experimental results show that compared with other prediction models, this model can eliminate the lag of prediction curve, the predicted error indexes MAPE,MAE and RMSE are reduced to 1.9 592%, 9.423 9 and 11.977 1 respectively, and the prediction accuracy of the model is increased to 96.435%.

Key words: very short-term electric load forecasting, ensemble empirical mode decomposition, sample entropy, simulated annealing algorithm, deep belief network