Mid-term and long-term load forecasting forecasting is the premise of electric power system planning, operation and control. Improving forecasting accuracy is of great significance to the safety, economy and environmental protection of electric power system. The multiple linear regression algorithm, structured by the small sample dates, has the heteroscedasticity impact. With regard to this deficiency, this paper puts forward an improved multiple linear regression algorithm based on Adaboost. This algorithm uses Adaboost algorithm to dynamically adjust the weighting factors corresponding to different samples, coordinates and combines various multiple linear regression models and improves the generalization ability of the algorithm. The improved algorithm of this paper is taken as an example of load forecasting on the data set of electric power utilization in Jinxian county, which verifies the effectiveness and practical value of the improved algorithm.