Converter is one of the core equipment in HVDC systems

and converter faults are complex and difficult to diagnose. Based on singular value decomposition（SVD）

the fault information is indicated by sin-gular values. The characteristic matrix of the largest singular value is selected to train the support vector machine（SVM）diagnotor. When faults occur in converters

characteristic matrix of the largest singular value can be calculated and then be input into the trained support vector machine classifier to diagnose. Simulation shows that the method can extract features of different converter faults

and distinguish different converter faults effectively and correctly.