Compared to power system loads, transformer loads have strong randomness, instability, and nonlinearity, making it relatively difficult to predict transformer loads. In order to achieve accurate prediction of transformer load and make transformer electrical load more susceptible to external environmental influences, on the basis of a deep belief network model optimized by the Supplementary Overall Empirical Mode Decomposition (CEEMD) algorithm, this article quantifies variables such as temperature, weather, and humidity as new input parameters, and obtains a CEEMD-DBN model with higher prediction accuracy. The experimental results indicate that by increasing meteorological factors, the prediction accuracy of the CEEMD-DBN model has been improved to varying degrees. The CEEMD-DBN model based on humidity factors (H-CEEMD-DBN) has the best prediction performance.