Abstract: At present, the load characteristics of the power grid are becoming more and more complex, and in order to effectively support the steady progress of the electricity market, it is urgent to carry out research on continuous forecasting of electricity. In order to solve the problem that the current continuous forecasting is inaccurate due to the lack of future data and the propagation of errors, a two-step electricity forecasting framework is proposed, which makes full use of the advantages of ensemble learning and radial basis function neural network (RBFNN). In the first stage, the stacking model is established for forecasting in the first half of the cycle, in the second stage, the forecasting results of the first stage are fused, and RBFNN is used to predict the second half of the cycle, and finally the effectiveness of the proposed framework is proved by experiments with the actual electricity consumption data of a province in western China.