The invention provides a wind electricity power prediction method based on wind electricity weather classification feature selection. The method specifically includes the steps of: screening wind electricity weather sample segments of numerical weather forecasting by 10m wind speed of a wind measurement tower in a historical data set; clustering sample segments screened out, dividing wind electricity weather into n classes, and using the sample segments, which are not screened out, as an (n+1)th class; respectively selecting input features of each class of weather, reducing dimensions of W-dimensional features to di(i=1, 2, L, n+1) dimension, where di is the number of the features obtained by screening of the i-th class of weather; and establishing wind electricity power prediction modelsunder the n+1 classes of wind electricity weather; identifying a wind electricity weather type of each segment of the next 24 hours, and respectively using di(i=1, 2, L, n+1)-dimension features associated with each class of wind electricity weather as prediction input for wind electricity power prediction. According to the method, the wind electricity weather of different duration is classified, adata-driven feature selection method is employed to find associated weather features of wind turbine output under different weather backgrounds, the wind power prediction models are respectively established, and prediction accuracy and efficiency are effectively improved.