In order to solve the problems of poor stability and premature phenomenon caused by the initially randomly generated population in traditional particle swarm optimization (PSO) algorithm, a good-point set particle swarm optimization (GSPSO) was proposed. Based on that, an efficient prediction model (GSPSO-SVM) was constructed by combining support vector machine (SVM). The good-point set was utilized to make the initial particles uniformly distributed in PSO algorithm, and then GSPSO was used to optimize the penalty factor C and radial basis function parameter g of SVM to obtain the best parameter for improving the accuracy and stability of classification of SVM. Finally, the model was successfully applied to the drought forecasting. The simulation results showed that the model have achieved good results in average accuracy and variance. Compared with PSO and genetic algorithm (GA) to optimize SVM model, the performance has been improved.