For the nonlinear, high redundancy and other characteristics of wind turbine SCADA monitoring data, this paper puts forward a wind turbine blade cracking fault prediction method based on the Restricted Boltzmann Machine (RBM) and the Support Vector Machine (SVM). The RBM's excellent feature learning ability is used as feature extractor to obtain the more expressive data features in the SCADA system data of a wind turbine. RBM's output is used as the input to the SVM to construct the combined prediction model of RBM+SVM. The prediction model is constructed and parameters are fine-tuned by using a training set and a validation set. To verify the effectiveness of the proposed model, the prediction results are compared with those of RBM+Logistic regression, SVM and Logistic regression. The experiments show that the prediction accuracy of RBM+SVM is 93.08%, which has obvious advantages over the three other compared models. The results can provide an important reference for the prediction of wind turbine blade cracking. This work is supported by Shandong Province Key Research and Development Plan (No. 2016GGX101024) and China Huadian Corporation LTD. 2019 Annual Science and Technology Project (No. CHDKJ18-02-52).