Abstract:Fault diagnosis of main bearings is an important means to improve the reliability and availability of wind turbines. However, it is difficult to obtain sufficient training samples and comprehensive operational data from each unit in the early stages of wind farm establishment. Therefore, the key to solve the problem is to obtain a reliable main bearing fault warning model applicable to multiple wind turbines through transfer learning methods by taking advantage of the high similarity of structure and principle between different units. In this paper, a comprehensive method based on XGBoost and transfer learning is proposed to conduct main bearing fault early warning for different wind turbines in wind farms. Firstly, the data of a 6.8 MW wind turbine in a offshore wind farm in Guangdong province is collected ,to construct a set of main bearing fault warning model based on XGBoost, and then it is transferred to other wind turbines. Experiments show that compared with LSTM, GRU, Light and Random forest models, the model constructed in this paper has the highest accuracy, with an R2 as high as 0.995. And after transfer, it still maintains a high accuracy, the R2 of each unit is greater than 0.95.