The main meteorological factors in Fangshan district of Beijing were selected as the input feature vectors of the support vector machine(SVM) model by grey relational analysis. The particle swarm optimization algorithm(PSO) is used to optimize the penalty vector C and the kernel function parameter δ of SVM. The short-term prediction model of grape downy mildew was established based on grey correlation and PSO-SVM. The model was applied to predict the grape downy mildew in the next day. Compared with the SVM model optimized by improved grid search method, the standard SVM with experience selection parameters and the BP network model optimized by different training functions and particle swarm optimization, the PSO-SVM model based on grey correlation analysis has the best prediction effect with the accuracy of 95.24% for the incidence of grape downy mildew. Compared with the PSO-SVM model based on all meteorological factors, the accuracy is increased by 1.19%, and the operation speed is 1.81 s faster.