In order to meet the requirements of accurate and efficient short-term power system load forecasting in the power market, a method is proposed which is an improved support vector machine short-term load forecasting. The load characteristic factors of daily load data are obtained by using principal component analysis method. The relationship between the load characteristic factors and influencing factors is established by grey correlation analysis. By constructing the weighted difference degree value of influencing factors between different days, the reciprocal is introduced into the support vector machine model as the weight coefficient, thus a SVM load forecasting model considering the weight characteristics of influencing factors is established. The simulation results show that the average forecasting accuracy of the improved SVM model is about 97.01%, and the daily average relative error fluctuates in the range of 0.32%–3.57%. The accuracy and stability of the proposed model are better than the conventional SVM model and LS-SVM model.