Wind speed forecasting is the basis of wind power forecasting, and its accuracy affects the efficiency of the wind farm seriously. In order to improve the accuracy of shortterm wind speed forecasting, the WRF mesoscale numerical model was used to predict the wind speed of a wind farm in the eastern coasts of China. Besides, the Extreme Learning Machine (ELM) algorithm was used for further correction. The results show that, the WRF model has a better effect on the wind speed, wind direction and other meteorological elements. After correcting the wind speed forecasting of the WRF model, with the ELM algorithm, the error of wind speed forecasting becomes smaller, and the relative root mean square error and relative mean absolute error are reduced by 20%-30%. Thus, the ELM algorithm is qualified to have better correction capability for the wind speed of WRF model forecasting compared with other intelligent algorithms (BP neural network, SVM algorithm), and can improve the accuracy of wind speed forecasting effectively.