Wind power forecasting is very important to improve the power quality and the safe operation of power systems. Ultra short term forecasting of wind power date of a wind farm in Chifeng Inner Mongola based on time series analysis, by applying the Eviews software through the stationary test, the ARMA (Autoregressive Moving Average) model of time series is built. Through ARCH effect of the residual of ARMA model by Lagrange Multiplier (LM), the corresponding ARMA-GARCH model is set up. Through the analysis of the wind power forecasting by using ARMA model, ARMA-ARCH model and ARMA-GARCH model respectively. ARMA-GARCH model possesses higher accuracy on the residual sequence of the data has the long term correlation.