In the field of wind power, the icing phenomenon of fans working in cold regions is serious. Changes in material and structural properties and load changes caused by low temperature environment threaten the power generation and safe operation of the fan. Paper proposes a method for monitoring the icing of wind turbine blades combined with random forest(RF) and support vector machine(SVM). The recursive feature elimination(RFE)-RF feature selection method is mainly used to select effective features from the original fan dataset, SVM train the dataset after feature selection. Finally, the SVM model and the RF model are merged by the Stacking combination strategy. The test results show that The method of combining RFE-RF feature selection and SVM is improved by 9.64% on the classification accuracy than the SVM model without feature selection.; Stacking combined strategy to fuse SVM model and random forest model, Fusion model has the best accuracy of 99.05% and best generalization performance. This method can effectively predict the icing of the fan and is understandable., It has guiding significance for wind farm operators to maintain fans.