Data cleaning，feature selection，and the estab lishment of a prediction model are indispensable steps to realize anomaly warning of the wind turbine generator system based on data collection and supervisory control and data acquisition （SCADA）.Firstly，isolated forest （iForest） and density-based spatial clustering of applications with noise （DBSCAN ） algorithm are combined to effectively clean the data outliers of SCADA，and random forest （RF） and Pearson correlation coefficient method are used to optimize the input parameters of the model.Based on the categorical boosting （CATBoost ） algorithm optimized by Optuna，a prediction model of gearbox oil pool temperature in the wind turbine generator system under normal operating conditions is established.Then，the state evaluation index is constructed with the sliding window method，and the interval estimation theory is employed to determine its critical threshold for anomaly discrimination of oil temperature.Finally，the anomaly warning of oil temperature is realized.The real historical fault data of oil temperature anomaly in the SCADA system of the wind turbine generator system are used to verify the effectiveness of the method.