In view of the accuracy requirement of wind power prediction, combined with the power characteristic curves of wind turbines and the nonlinear fitting of support vector machines, a particle swarm optimization (PSO) algorithm based on empirical mode decomposition (EMD) has thus been proposed to optimize the short-term prediction model of support vector machine (SVM) power. Based on the prediction of each wind speed sequence component after EMD decomposition by PSO-SVM model, the predicted values of each component will be superimposed, thus obtaining the predicted value of the wind speed. With the value input into the power conversion curve, the result of the final wind power prediction can be worked out, and the prediction of the power of wind turbines can be realized. A comparative analysis of the simulation error of the actual wind speed in a certain area verifies the effectiveness and feasibility of the combined forecasting mode, which effectively improves the prediction accuracy of the short-term wind power.