Abstract: For problems of small samples, nonlinear, high dimensions and the local minimum of electric power load, a modeling method based on the least square support vector machine was proposed to forecast short-term power load by taking historical load, temperature and humidity data as inputs. For parameter selection problem of the least square support vector machine in modeling, an improved particle swarm optimization algorithm was used to optimize two parameters of penalty factor and kernel function of the least square support vector machine model, which guides initial population selection and avoids premature convergence of the particle according to diverse information of population. The simulation results showed that the method based on the improved particle swarm optimization algorithm and the least square support vector machine model has higher accuracy for predicting short-term load in comparison with method of the least square support vector machine model and method of the particle swarm optimization algorithm and the least square support vector machine model.