Abnormal data detection of wind turbine is of great significance for maintaining the stable operation of wind power equipment.In order to solve the problem of clustering randomly assigned initial points of K-means algorithm and abnormal wind turbine data,this paper proposes an improved K-means algorithm to detect abnormal wind turbine data. The improved method first selects the median of the data sample as the first initial clustering center.When selecting the next cluster center,the point farther away from the current n cluster centers will have a higher probability of being selected as the n+1 cluster center.And then achieve the goal that the cluster centers are far away from each other. Based on this,the operation data of wind turbines are clustered to detect outliers and outliers,so as to ensure the stable operation of wind power equipment.