This system mines the historical daily loading which has the same meteorological category as the forecasting day in order to compose data sequence with highly similar meteorological features, with this method it can decrease SVM training data and overcome the disadvantage of very large data and slow processing speed when constructing SVM model. With the advantage of data mining technology in processing, it can reduce the large data and eliminate redundant information. Comparing with single SVM and BP neural network in short-term load forecasting, this new method can achieve greater forecasting accuracy. It is denoted that the SVM learning system has advantage when the information preprocessing based on data mining technology.