This paper proposes a PSO-SVM short-term power load forecasting method based on the introduction of particle swarm algorithm SVM learning process. To improve SVM kernel function by selecting a portfolio, so you can fully guarantee the computing speed and high prediction accuracy. In this paper, historical load data Jilin region as training samples, with the traditional SVM prediction model by comparing the predicted results were compared with the actual data to prove that the combination forecasting method based on kernel function to some extent able to guarantee the accuracy of short-term load forecasting.