Abstract: A data-driven tower vibration monitoring method considering the influence of wind pulsation and turbine control action on tower sproposed for solving the problem of tower vibration monitoring. Firstly,the operation conditions of wind turbine are identified based on K-means clustering,and the effects of various state parameters and operation conditions on tower are analyzed. Secondly,the tower vibration trend under different operation conditions is modeled and predicted based on XGBoost algorithm;a tower vibration monitoring method based on Wasserstein distance is proposed according to the difference of preicted residual errors of different towers in the same wind farm. Finally,the actual data of the wind farm is used for verification,taking the sum of squared error as the evaluation index,and the results show that the method considering turbine operation conditions improves the accuracy of the tower vibration prediction by 34.6%. The data-driven method effectively identifies the tower with frequent abnormal vibration in the wind farm and improves the efficiency of operation and maintenance.