This paper presents a short term load forecasting method that based on the data mining technique.We introduced the SVM method into the field of short-term load forecasting.By randomly selected historical load data to update the regression function, so you can fully guarantee the computing speed and higher forecast accuracy. The paper proposes to use the historical load data of Song-Yuan areas as training samples，by comparison with the traditional BP neural network prediction model, comparing the predicted resul a certain degree.