In order to improve the accuracy of wind power prediction, a new short-term wind power prediction method based on the variational mode decomposition (VMD) and the improved least squares support vector machine (LSSVM) is proposed. Firstly, VMD is used to decompose the historical power data into the trend component, detail component and random component to reduce the complexity and instability of original data. Then the IBA-LSSVM prediction model is established, the parameters of the least squares vector machine are optimized by the improved bat algorithm (IBA), and each sub mode is predicted separately. The final predicted wind power is obtained by superimposing the prediction results of sub modes. Finally, the power of a wind power plant in Ningxia is predicted by the proposed prediction method. The effectiveness of the model is proved by the error analysis of the prediction results. The comparison of different prediction methods verifies that the proposed model has higher prediction accuracy.