Abstract: An improved short-term power load forecast model that uses Support Vector Machine (SVM) was developed. The new model, called the PSO-SVM forecast model, is based on Particle Swarm Optimization (PSO) parameters. In traditional SVM models, penalty factor C and kernel function parameter σ are generally dependent on particle experience. When power load forecast data change, however, obtaining satisfactory forecast precision using these empirical values is difficult to accomplish. Therefore, this study used PSO to optimize the parameter selection methods of SVM in accordance with training data and improved SVM forecast precision. PSO-SVM is generalizable and easily expandable. To verify the validity of the model, this study selected and analyzed integral point data on Fujian Province in October 2011. Data for October 1-25 were used for training and those for October 26-30 were employed for testing. The PSO-SVM model was then employed to forecast and analyze the October 31 data. Results show that the forecast efficiency of PSO-SVM was better than that of traditional SVM. In contrast to the forecast efficiency of GA-SVM, PSO-SVM was slightly better. In addition, PSO-SVM exhibited better operational performance than did GA-SVM.