The invention discloses a method for predicting short-term power of a wind farm on the basis of a BP (back propagation) neural network. The method includes the steps of a, acquiring historical records of meteorological element data of a location of the wind farm and output power corresponding to each record; b, correcting the meteorological element data into fan hub height data; c, applying the corrected meteorological element data as input data to be input into the BP neural network, and applying the output power corresponding to the meteorological element data as input of the BP neural network to train the BP neural network; d, acquiring meteorological element data of the location of the wind farm according to numerical weather prediction data in a prediction period, correcting the meteorological element data into fan hub height data and generating corrected meteorological element data; and e, inputting the corrected meteorological element data obtained in step d into the BP neural network, and outputting data which is generation output power of the wind farm in the prediction period. The method is simple, easy and highly accurate.