Selforganizing neural network method is applied to classify weather p atterns based on daily NWPs of ECMWF from Jan. 1,2003 to Dec. 31,2006. It shows that the weather pattern is similar over China from November to March. Dynamic di agnosis is applied to daily NWPs of ECMWF in November to March in year 2004-2007 to pick up predictors which have good correlation coefficients with daily maxim um wind speed at 16 coastal weather stations. An updated KNN method is used to s et up wind speed forecast models for November to March. Daily wind speed forecas t for January to March of 2007 is carried out. Results show that KNN method is of good ability in daily maximum wind forecast.