In order to improve the accuracy of fault diagnosis of power transformer

quantum particle swarm optimization（QPSO）is proposed to optimize relevance vector machine（RVM）of the transformer fault diagnosis.Four binary RVM classifiers are used to solve the problem of transformer fault diagnosis multi-classification.Multikernel function of RVM can be fused to a variety of transformer running characteristic information

thus provides a effective method for transformer fault diagnosis model which with nonlinear and limited samples.RVM diagnostic model parameter can fast get the best value by QPSO algorithm

and effectively improve the generalization ability to diagnose models through the cross validation（CV）fitness function.Study of practical cases indicate that

compared with IEC three ratio

BPNN algorithm or PSO-SVM algorithm

the coupling algorithm correct diagnosis was 91.1%

can effectively improve the accuracy of fault diagnosis of transformer.