Wind power generation with network running is the main way to realize the exploitation and utilization of wind energy in large-scale. Accurate real-time prediction of wind power is an effective means to achieve rational management of wind energy in large-scale. A multi sampling scale multi-step prediction model based on time series method is established for wind power. For the periodic component, mean method of modeling domain is adopted to predict; For the remaining component, multi sampling scale multi-step prediction model based on time series method is adopted to predict after stationarity test. By means of the measured dates of two different wind farms, we perform the wind power prediction in real-time at different times, the prediction results show the effectiveness of the prediction model compare with the methods of persistence and rolling multi-step model based on time series method.