Abstract: Aiming at the problem of improving the grid-connected friendliness of wind farms,a wind farm generation capacity evaluation scheme considering uncertain wind speed prediction and uncertain active power characteristics of wind turbines is proposed.By analyzing the wind speed ultra-short-term prediction error and the wind turbine output characteristics in each wind speed range and establishing two probability distribution models,a double uncertainty prediction model of the wind turbine output in the ultra-short term is constructed using Bayesian network. Based on the proposed prediction model for each wind turbine,the power distribution strategy of the wind farm is designed with the objective of tracking the dispatching command with maximum probability. The analysis shows that the proposed prediction model with double uncertainty is more accurate in describing the probability distribution of the wind turbines’ active power,and the proposed model can effectively improve the completion level of the grid power command in the wind farm control.