In view of the nonlinear and non-stationary characteristics of wind power time series, this paper presents a modified ensemble empirical mode decomposition (MEEMD)-sample entropy (SE)-ARMA wind power ultra short term combined forecasting model. The white noise signal added in the EEMD decomposition is changed to two groups positive and negative white noise signal which have the equal absolute value, and the EMD steps of the decomposition process of MEEMD is improved with the endpoint extension and three piecewise Hermite interpolation, forming a modified EEMD decomposition algorithm (MEEMD). The wind power time series is decomposed into a series of complex wind power generation by MEEMD-SE, and the ARMA forecasting model is built for each different sub sequence, and the final wind power forecast value is obtained. Through numerical analysis and comparison with other forecasting models, the results show that the MEEMD-SE-ARMA combination forecasting model can effectively improve the accuracy of the ultra short term forecasting of wind power generation. This work is supported by National Science & Technology Pillar Program during the 12th Five-year Plan Period (No. 2012BAJ26B00).