The abnormal data of power system can affect the safe and stable operation of power system. Traditional abnormal data detection methods can no longer achieve effective identification and judgment of massive power system operation data. In this paper, a strategy fusion method of grid operation outlier detection is proposed, which combines machine learning algorithm and statistical algorithm to quickly determine the time period when the abnormal data appears through machine learning, and then uses statistical algorithm to effectively judge the grid operation outliers. The proposed method is applied to grid operation data containing a variety of electricity consumption ends, and the experimental results are compared with those of three traditional methods. The experimental results show that the F1-score of the proposed strategy fusion outlier detection method is higher than the other three compared methods, and is significant compared to the other three methods. The method proposed in this paper can identify grid operation anomaly data quickly and effectively, and the effect is significantly higher than the other three traditional methods.