Abstract: Real-time monitoring of wind turbine output and timely detection of the unit problems can maximize the economic benefits of wind farms. Using adaptive Density-Based Spatial Clustering of Applications with Noise（DBSCAN） algorithm to extract the datas required to establish the performance model of wind turbines under normal conditions in a wind speed-power（v-P） coordinate system. In the monitoring phase, after dividing horizontal power intervals, Mahalanobis distance is used to measure residual errors between the monitoring data and the performance model, and the sliding window method is used to continuously obtain the residual sequence into cloud model for fuzzy evaluation, obtaining the operating status of wind turbines. The changes in clouds show that the anomaly monitoring method based on cloud models can accurately and objectively reflect the operating status of the unit, providing effective guidances and suggestions for maintenance work.