Earlier detection of latent faults in a power transformer is of great importance to security and stability of power grid. In this study

by making use of the transformer oil chromatography and preventive test data

a synthetic fault diagnosis method for power transformers is proposed based on neural network and evidence reasoning theory. According to the output of neural network

the unknown part of evidence fusion credit distribution is divided into two parts-the unknown part due to weight distribution and the unknown part due to lack of information-for fusion of conflict information

which can objectively reduce the uncertainty of diagnosis and optimize the evidence reasoning algorithm. Simulation proves the superior diagnostic performance of the proposed diagnosis method.