There is a problem of blade icing in the operation of wind turbine under low temperature environment

but the operation data of different working conditions of different wind turbine are quite different in the actual operation process. A single model cannot meet actual forecasting requirements for a single operating condition

a method for predicting wind turbine blade icing based on transfer learning is proposed. Effectively utilize the similarity between the historical data of the wind turbine and the current working condition data

supplemented by the transfer algorithm to achieve partition of similar working conditions

and use XGBoost to adaptively select the optimal features and training models for different working conditions

thus achieving multiple working conditions and multiple model wind turbine blade icing prediction. Experimental results based on data acquisition by SCADA system indicate that this modeling method can effectively mine historical data with the large amount of historical data of No. 21 fan and a small amount of new data of No. 15 fan

and realize rapid modeling under different working conditions. After the transfer algorithm

the accuracy of No. 21 fan model in predicting No. 15 fan is increased by 10 %

which verified the effectiveness of the modeling method.