Nowadays, multivariate regression algorithm is mostly adopted in the mid-long term load forecasting, but it lacks good theoretical foundation for the selecting of the impact factors and the past years when modeling. It is difficult to make a balance between considering more affecting factors, historical data and reducing the regression model error, which leads to the inaccuracy of the multivariate regression algorithm in actual load forecasting. Rough set theory and D-S evidence theory are applied to the multivariate regression algorithm. First, rough set theory is used to sort the importance of influencing factors, and then the impact factors and the historical years are clustered respectively, so that several multiple regression models can be built. Additionally, the weights of different models are fused by using D-S evidence theory. In this way, the final combination forecasting model based on the multivariate regression analysis method can be built. According to the valid example, it can be concluded that the final model can preferably balance the selection of impact factors and past years while effectively improve the accuracy of the multiple regression algorithm in the mid-long term load forecasting, which provides stronger applicability at the same time.