Considering the instability and strong nonlinearity of wind power, in order to improve the accuracy of wind power forecasting, the original wind power sequence is decomposed into a series of sub-sequences. Then each sequence is forecasted 24 h ahead by combination forecasting model. And the weight coefficients of each sequence are identified using variance reciprocal method through a virtual forecasting method. Consequently, all the subsequence forecasting outputs are superposed to obtain the final forecasted results. At the end, a wind farm in Guangdong is chosen to validate the feasibility of the proposed model. Through seven consecutive days of forecasting of the wind power, the results indicate that the wavelet packet transform can grasp the variation law of wind power effectively, and the combination forecasting method can obtain a better forecasting result than single method.