Water Resources and Power

In order to improve the accuracy of transformer fault diagnosis,a hybrid model which combines the FRVM with the depth belief network(DBN)was proposed.The method established the relationship between gas and fault types using the ration of dissolved gas as input parameter.Considering that DBN needed to extract huge amount of feature information,the FRVM was used to separate the discharge and overheating faults to reduce the feature information that DBN needed to extract.Then DBN was used to realize further fault diagnosis.The output was the probability of the corresponding fault types.The proposed method was compared with wavelet neural networks and SVM.The results show that the proposed method has the highest accuracy,and can analyze the uncertainty of the problem,as well as has the ability to diagnose multiple faults.