Effective wind power prediction on wind farm can not only guarantee safe operation of wind farm, but also increase wind power storage and utilization efficiency. This research combines mesoscale numerical weather prediction model with BP neural network model for the use of wind power prediction. WRF model is used to recalculate the meteorological elements of trial wind farm from Jun. 2008 to Jun. 2009, and the accuracy check result shows that the correlation coefficient between predicted value and corresponding measured value of wind speed reaches 0.72. Predictions accuracy of wind direction, air temperature, humidity and air pressure are also precise, which meets the requirement of building BP neural network prediction model. The BP neural network prediction models of output power of 40 wind turbines are established on trial wind farm one by one, to analyze the influence of data normalization method and neuron number at the hidden layer on prediction accuracy. The prediction test every 10 minutes, with the actual effect of 24 hours, is done for 26 days, and prediction accuracy test is conducted by using independent samples. The result shows that relative root mean square error of the output power of the single wind turbine from 24.8% to 32.6%, and the correlation coefficient between predicted value and measured value is from 0.45 to 0.68; relative root mean square error of the whole wind farm is 21.5%, and the correlation coefficient between predicted value and measured value is 0.74.
