The invention discloses a wind power short-term interval prediction method based on an RT reconstructed EEMD-RVM combined model. Firstly, ensemble empirical mode decomposition is conducted on an original wind power sequence to obtain a stable intrinsic mode function (IMF) component and a remaining (RES) component with different features; by means of the runs-test method (RT), fluctuation degree detection is conducted on the components, and the similar components are reconstructed into three new components, with typical features, including a random component, a detail component and a trend component according to the fine-to-coarse sequence; then, a relevance vector machine (RVM) is adopted to the new components to build interval prediction models respectively; finally, prediction results of the new components are superposed to obtain a total interval prediction result under a certain confidence level. By means of the method, prediction precision of the models and the interval coverage are improved, the interval width is obviously reduced, and accordingly the prediction result is remarkably improved.