Support vector machine (SVM) method is already very common in the field of short-term load forecasting. Traditionally, historical data is usually selected from the date around the forecast day when the model is trained, without considering the effects of the meteorological conditions, the types of the week and the holidays, therefore the established model does not fully reflect the characteristics of the forecast day. A support vector machine load forecasting method based on data mining technology is proposed in this paper. It gives a novel method to select the sample for prediction model. Firstly, hierarchical clustering method is used to cluster the historical load. A decision tree is set up by the classification result. According to the properties of the forecast date, the historical load for support vector machine prediction model is obtained through the decision tree, then establish support vector machine model to predict the daily load. Data from a specific area in Zhejiang province was used to verify ninety-six points load forecasting, the performance of algorithm was compared with SVM without data mining, In this paper, we propose a method to solve the problem that the traditional SVM method cannot properly select the dates which reflect the characteristics of the day.