Aiming at short-term wind power focasting, a kind of combining forecasting method for short-term wind power based on empirical wavelet transform (EWT) and extreme learning machine with kernels (KELM) is proposed. firstly the EWT method is used to decompose the wind speed data and extract the different modes which have a compact support Fourier spectrum. Secondly, different KELM forecasting models are constructed for the sub-sequences formed by the each mode component. Simultaneously, the ultimate wind speed forecasting results can be obtained by the superposition of the corresponding forecasting model, the forecast value of wind power is calculated by the wind power characteristic curve. In order to verify the effectiveness of the proposed methods,it is applied to some wind farms in China for short-term wind power forcasting. The experiments are also implemented in the ELM method, KELM method, SVM method and BP methodin the same condition respectively. The comparing experimental results show that proposed method have higher forecasting accuracy and superior application potential.