Wind turbines are affected by a variety of factors during operation, such as extreme weather conditions, power abandonment strategies and sensor failures. The wind speed and power data recorded by wind power SCADA (Supervisory Control and Data Acquisition) equipment contains a large number of anomalous scattered points with different causes and uneven distribution. Traditional data cleaning methods are not fully effective in dealing with these anomalous data. In order to clean the anomalous data, especially the stacked anomalous data, a combination of kernel density estimation method and DBSCAN (Density-based spatial clustering of applications with noise) algorithm is used in this paper. Through this method, we successfully removed the stacked anomaly data, making the cleaned data more suitable for WTG (wind turbine generator) data modelling. This provides a reliable data base for further research on the operation law of wind turbines, improvement of wind energy utilization and optimization of wind farm strategies.