A short-term wind power confidence interval prediction method based on variational mode decomposition (VMD) and Gaussian process (GP) is proposed to describe the randomness, fluctuation and uncertainty of wind power. Firstly, the wind power sequence is decomposed into a series of modes with different bandwidths to reduce its nonlinearity by using the variational mode decomposition algorithm (VMD). Then the Gaussian process regression model is established for all the sub-modes. Finally, the prediction results of each sub-mode are added to obtain the forecasting confidence interval of wind power. The results of the example show that compared with other conventional decomposition method, this combined model can effectively improve the prediction accuracy and prediction interval coverage, and reduce the prediction interval width, which has certain practical value.