To further enhance the accuracy of power load forecasting, a hybrid model utilizing Support Vector Machine (SVM) and Particle Swarm Optimization (PSO) is proposed. Firstly, the forecasting models of SVM and traditional PSO-SVM have been established. Based on the premature convergence problem inherent in the traditional particle swarm optimization algorithm, this paper improves PSO optimization algorithm by controlling the population characteristics through diversity metrics to avoid the population from falling into local optimization prematurely. Finally, an improved PSO-SVM short-term load forecasting model is established. The historical load and meteorological data from a specific area in Tianjin in 2018 are utilized for example analysis. The MATLAB simulation results show that the improved PSO-SVM model has higher power load forecasting accuracy than the single SVM and the traditional PSO-SVM.