A mesoscale numerical weather prediction model and the BP neural network forecasting model are used to predict wind farm wind power. By using the WRF numerical model, the meteorological elements from June 2008 to June 2009 of wind are calculated, and results show that the correlation coefficient between forecast and measured wind speed is 072 The forecasts of wind direction, temperature, humidity, and air pressure are also relatively accurate. The BP neural network forecast model of wind power for 40 wind turbines are established, and the influences of the data standardization method and the number of hidden neurons on prediction are analyzed. The results of trial prediction show that the relative RMS error of a single wind generator is 248% to 326%; the correlation coefficient is 045 to 068; and those for the whole wind farm are 215% and 074.