Abstract: To mitigate the impact of uncertainty and volatility of wind power on the power grid, a gate recurrent unit neural network prediction model is proposed based on extreme-point symmetric mode decomposition and wavelet packet decomposition. Firstly, the original wind power output sequence is decomposed several times to thoroughly excavate the patterns among the sequences. Then, a gate recurrent unit neural network is employed to predict the decomposed sub-modes. Finally, the prediction results of the integrated model are compared with those of the gate recurrent unit neural network and BP neural network. Taking a wind farm in Altay, Xinjiang as an example, the validity of the model is verified. It has been demonstrated that the proposed model can substantially enhance the forecasting accuracy of short-term wind power generation.