Aiming at the problem that there are a large number of horizontal or vertical distribution outliers in the wind speedpower data collected by SCADA system when wind turbine is in abnormal operation, an abnormal data processing method based on median absolute deviation method (MADM) and quartile method (QM) is proposed to solve it, namely MADM –QM algorithm. Firstly, based on the relationship model of wind speedpitch angle, the wind speedpitch angle data outside of ±4.5 MAD are discarded by solving the median absolute deviation (MAD) in the wind speedpitch angle data set of the wind speed interval. Secondly, based on the wind speedpower relationship model, the abnormal values in the wind speedpower data set of the power interval are eliminated, and then the abnormal values in the wind speedpower data set of the wind speed interval are eliminated to complete the abnormal data processing. Finally, the actual operation data of wind turbine under complex working conditions of a wind farm are taken as examples for verification, and comparison with MADM, QM and densitybased spatial clustering (DBSCAN) method. The results indicate that the proposed method can not only effectively identify abnormal data but also efficiently and stably clean them. Compared with the other three methods, to a certain extent, it proves that MADMQM can achieve good efficiency of abnormal data processing and optimal cleaning quality on the abnormal data.