Wind power prediction is crucial for wind power plan to plan power dispatch and maintenance. This paper decomposes wind power sequence a series of frequency subsequences, making use of the capability of the improved wavelet packet on processing mixing information, then forecasts each subsequence by the combination model of genetic algorithm neural network, and finally get the forecasting outputs by reconstructing the forecasting results with wavelet packet algorithm. A wind power plant in Anhui is chosen to validate the feasibility of the proposed model. The simulation results indicate that this combination algorithm achieves good forecasting effect.