In order to improve the accuracy of load forecasting for the demand side, this paper proposes an improved short-term load forecasting method based on demand-side load characteristics. It completes the two-segment classification for demand-side load according to the load characteristic and the attribute clustering algorithm, then establishes the forecasting model for load forecasting by using MPSO-RBF and LS-SVM. This method is used to forecast the power load of an industrial park, and is compared with the actual load data, the traditional forecasting model and the single model prediction method. It shows that the method is effective and practical, it can get accurate prediction results and control load for the users pertinently, and it is convenient to macroly control the load aggregation for manager. What’s more, it can promote the development of the energy internet effectively.