Wind speed forecasting is the base and precondition of wind power prediction of wind farm. The Numerical Weather Prediction (WRF) model is used to predict wind speed. In order to improve the accuracy of WRF model, the Least Square Support Vector Machine (LSSVM) is used to correct the wind speed of the output of the WRF model. At the same time, in order to improve the accuracy of the LSSVM model and reduce the complexity of the fitting process, Particle Swarm Algorithm (PSO) is used to optimize the parameters. Experimental results show that the LSSVM can further reduce the error of WRF model in predicting wind speed sequence, and the relative root mean square error and the relative to the average absolute error are reduced by 5%~10%, the RMS error decreased by 0.5 m/s. Compared with without optimized LSSVM and ELM, PSO-LSSVM has a better correction effect in wind speed predicting by WRF to improve the accuracy of wind speed forecasting. This work is supported by National Natural Science Foundation of China (No. 41675156).