Abstract: On account of the SCADA data has been deficiently studied via the existing fault prediction,this paper proposed a PCA-PSO/GS-SVM integrated model of fault prediction based on the standard support vector machine（SVM）method,principal components analysis（PCA）,particle swarm optimization（PSO）and grid search（GS）. Compared with the standard SVM,the proposed method establishes the correlation between variables accurately with the PSO and GS methods. Meanwhile,SCADA monitoring data of the 2 MW wind turbine gearboxes in north China is as an example for analysis. The results show that the mean prediction absolute error of the gearbox output power after using PSO algorithm is 3.0647 times that of the GS algorithm. It is more reasonable to use PSO algorithm to optimize the parameters for the prediction of the temperature. Moreover,eliminating singular points in training sample data can improve the prediction accuracy and generalization ability of the model.