To increase the accuracy of short-term load forecasting, a short-term load combination forecasting model based on adaptive noise-based full set empirical mode decomposition (CEEMDAN)-sample entropy (SE) and deep belief network (DBN) is proposed. Firstly, CEEMDAN-sample entropy is used to decompose the original load sequence into multiple sub-sequences with different characteristics. The sample entropy of each sub-sequence is calculated, and the sub-sequences with similar entropy values are recombined to obtain a new sequence, which reduces the accuracy of the original non-stationary sequence. The impact is reduced and the computational scale is reduced. Then, considering the periodic characteristics and influencing factors of each new sequence, different DBN prediction models are constructed for each new sequence,Using DBN prediction model to overcome the problem of insufficient feature extraction and initial parameters of shallow neural network and finally the prediction results are superimposed to obtain the final prediction value. The simulation results show that the combined prediction model effectively improves the prediction accuracy.