For effective planning and scheduling and for Wind Power Prediction (WPP) at 70 meters above the ground and 0-4h super short term wind speed forecasting, this paper uses the NWP wind speed of MM5 grids from the National Meteorological Center to analyze the prediction error at the wind tower height in a wind farm which is located off the coast. Based on the meteorological data from the wind tower and after data statistical analysis, it is found that the numerical forecast wind speed errors have correlations with themselves and the prediction errors are caused by the elements of sustainability. A method using earlier observation errors and turbulent index to revise the wind speed forecasting of MM5 is discussed and an ANN dynamic modification model for super short term forecasting is set up. The results show that after correction of the forecast wind speed, the mean absolute error is reduced and the prediction accuracy is improved effectively. It is also shown that the error index decreases about 40%, and the prediction curve can better reflect the high frequency of wind speed fluctuations, which better agrees with the measured wind speed curve. Update can be done once every four hours, satisfying the requirements of power grid dispatching. The method is simple and economic and can be used widely in small and medium-sized wind farms. It will help effective use of wind power as well as safe operation of power companies.