Abstract: Aiming at the problem of strong randomness of wind power and difficulty in modeling time series correlation, a combination of variational mode decomposition (VMD) and weight sharing gate recurrent unit neural network (WSGRU) The integrated VMD-WSGRU integrated learning and prediction method. Case studies show that the proposed model can effectively track the change of wind power and has high short-term prediction accuracy. The model uses variational modal decomposition to non-recursively decompose the original wind speed sequence into sub-components with a predetermined number of layers at first. The modal function components at different frequencies represent different characteristics of the power load, while reducing the original sequence. Stationary degree, then use WSGRU to quickly and accurately model and predict all the analyzed sub-components as a whole, and finally use ANN to modify and obtain the prediction result of wind power. The calculation results show that, compared with the traditional single model prediction method, the proposed integrated prediction model can better grasp the trend of wind power and has better prediction accuracy. Compared with other common combined prediction methods, the training of this method is more accurate and efficient.