Abstract: Wind power forecast is highly beneficial for the diapatch and stable operation of power system with large-scale of wind power.The model structure of a back propagation artificial neural network（BP-ANN） is introduced for the wind power ultra-short-term forecasting system in operation since October,2009.By analyzing three months’ forecasting results of the model,the system error is calibrated and the impact of wake effects on the forecasting results is also corrected by statistical approach,thus the forecasting precision can be improved.The root mean square errors of modified model are 6% lower and mean absolute errors are 7% lower than those of original model,indicating a significant improvement of the forecasting model.The forecasting results coincide with the field measured results and demonstrate the referential significance for the wind farm dispatch.