At present, there is no mature prediction model for the shutdown caused by icing in wind farms. There is a gap between the traditional prediction of icing thickness calculated by the icing growth model and the actual capacity of icing shutdown needed by power dispatching departments. Therefore, the weather parameters that play a key role in the icing process are studied, and the prediction model of the ice-covered shutdown capacity of wind farms is built based on the machine learning algorithm, to effectively warn wind farms before the arrival of cold waves and provide data support for the provincial power dispatching department. The algorithm results show that the prediction accuracy of the shutdown capacity of the wind farm in Jiangxi province is more than 80%.