In order to increase the accuracy and improve the forecasting system of electricity sales, a combined forecasting model of Elman neural network combined with historical similar monthis proposed. Combined with the characteristics of rapid identification among historical data, a set of historical data similar to the forecasted month is found by analyzing and processing the detailed data and external influencing factors of electricity sales objects.This set of historical data is used as the input data of Elman neural network to complete the prediction of such sales objects. Then, the forecast data of each electricity sales object is combined to get the total monthly forecast electricity sales. The simulation result shows that compared with the single Elman neural network, the combined prediction method has higher prediction accuracy, better convergence performance and a good application prospect.