The invention relates to the field of wind power output power prediction technology, in particular to a short-term wind power prediction method based on EWT-ESN. The method comprises the steps that first, EWT is adopted to disintegrate original wind power with strong nonlinearity and non-stationarity to obtain N AM-FM components, wherein F0 refers to an experience dimension component, and F1 refers to an experience wavelet component; second, an ESN prediction model for all the components is established in combination with characteristics of all the disintegrated components from F0 to FN; third, prediction results of all the components are superposed to obtain a final prediction result; and fourth, prediction result analysis is performed according to error evaluation indexes. According to the method, first, the EWT algorithm is utilized to disintegrate an original sequence of the non-stationary wind power into the AM-FM components with compact support Fourier spectrum characteristics, then ESN is utilized to perform prediction on all the AM-FM components, and superposition is performed to obtain the final prediction result. Through the method, wind power prediction precision can be effectively improved.