The accuracy of wind power prediction can improve the security environment of largescale wind power integration. Under the background, a combined approach considering wind speed fluctuation characteristics for shortterm wind power forecasting is proposed in this paper. Firstly, five types of wind speed fluctuations are defined, and the fluctuation type of historical wind speed series in numerical weather prediction is divided to get the time period of different wind speed fluctuation types. Secondly, the historical wind power series corresponding to these time periods are classified, and the variational mode decomposition algorithm is used to deal with various types of wind power series for obtaining multiple submodalities with different characteristics and frequency bands. Thirdly, a prediction model based on the gated recurrent unit neural network is constructed for each submode, and the prediction results of each submode are added to obtain the predicted value of the wind power. Finally, the wind speed series of the forecasted day is identified from the types of fluctuation, and the matching power prediction model is selected to calculate the final prediction value. The wind speed data in the numerical weather prediction and power data of an actual wind farm are used for simulation analysis to verify the effectiveness of the proposed combined forecasting method.