To meet refined wind speed forecasting needs for wind farms, using output numerical prediction products of the MM5 model, measured meteorological data of two offshore wind towers around the coastal areas of Fujian, and the stepwise regressive statistical method, hourly wind speed prediction equations for the given layers of these two towers are fitted with both MM5 numerical prediction factors and measured wind tower factors being introduced into the regression equation. The implementation process of this method is described in detail. Based on forecasting performance metrics in one year and the effect of independent sample test for 15 days， the result shows certain value of this method to hourly wind speed forecasting.