In order to improve the support vector machine (Support Vector Machine, SVM) performance, reduce the time cost, this paper proposes a SVM algorithm based on feature extraction, and uses it for fault diagnosis of steam turbine generator set. Firstly, Kernel Fisher Discriminant Analyst (KFDA) algorithm is used to extract the key features of the turbine generator set data, and then the SVM classifier is used to classify the feature data set. Experimental results show that the algorithm is feasible and effective, and it has been improved in classification performance and detection time.