Abstract: Aiming at the problem of low wind power prediction accuracy due to abnormal data of wind turbines, a wind power farm data cleansing method based on density-based spatial clustering of applications with noise (DBSCAN) algorithm coupled with least absolute residual (LAR) method is proposed. Firstly, DBSCAN is used to identify dispersed abnormal data, and then LAR is used to construct a model for identifying stacked abnormal data, which realizes the cleansing of dispersed abnormal data and stacked abnormal data of wind power farms. Finally, the effect of the proposed method is verified by Pearson correlation coefficient and back propagation neural network prediction model. The results show that the wind power farm data cleansing method based on DBSCAN+LAR can effectively reduce the wind power prediction error.