Power system load forecasting model can be divided into the single index model and information collection model.To insure the accuracy, it is necessary to avoid missing the important information and must collect the related indicators as much as possible. This paper used principal component analysis(PCA) to simplify information.Data correlation removed and data dimension reduced through normalization processing.By introducing the kernel function and the dual skills of the support vector machine algorithm, can effectively avoid the curse of dimensionality.Compared to the SVM method, the accuracy of load forecasting is effectively improved.