Abstract: The present invention relates to a method for screening a correlated seed turbine for wind direction prediction. The method includes the following steps: (1) modeling a yaw event of a wind turbine based on a wind direction, a wind speed and a yaw parameter, and obtaining a wind turbine yaw event flag of each wind turbine in a wind farm during a modeling period; (2) classifying and counting the wind turbine yaw event flag, and obtaining a yaw correlation coefficient of other wind turbines each with a target wind turbine in the wind farm; and (3) screening a seed turbine based on the yaw correlation coefficient. Compared with the prior art, the method of the present invention has the advantages of high discriminant validity of the seed turbine, small error, high correlation, and close wind speed characteristics.