Transformer Fault Diagnosis Based On Hierarchical Inference Of Bayesian Network

Transformers are very important devices in the power system,and their operating conditions are closely related to the safety performance and economic benefits of the power grid.At present,the research of transformer fault diagnosis technology is not perfect enough.Finding a method that can not only efficiently process data information but also can improve the efficiency of transformer fault diagnosis is a breakthrough point to solve the technical bottleneck of transformer fault diagnosis.Because Bayesian networks can qualitatively and quantitatively deal with uncertainties,it provides a new idea for transformer fault diagnosis research.The main work and innovations of the paper are as follows:(1)Transformer fault model learning based on Bayesian network.Firstly,the factors that cause transformer faults are analyzed,and the transformer fault types are classified by means of discrete coding.Then,the HBN structural model of transformer fault diagnosis is obtained by hierarchical Bayesian network structure learning algorithm,and the relationship between variables is qualitatively described.Finally,the conditional probability of transformer fault diagnosis is obtained based on the maximum likelihood estimation,and Dirichlet distribution is used as the prior probability distribution to quantify the relationship between the variables of the model quantitatively.(2)For the diagnosis of transformer fault diagnosis model with HBN framework,a hierarchical reasoning algorithm for Bayesian network is proposed.By separating the HBN aggregation nodes,the network size is reduced,and the joint tree algorithm is used for reasoning to reduce the amount of computation.Compared with the classical variable elimination algorithm and message propagation algorithm,the simulation experiment proves the correctness and efficiency of the proposed hierarchical reasoning algorithm.(3)Comparison of three transformer fault diagnosis models.By comparing the BBN structural model of transformer fault diagnosis with the traditional naive Bayesian network(NBC)model and tree-like Bayesian network model,the rationality of the HBN(TAN)structural model is verified and the precision is high.When a transformer fault occurs,the cause of the fault can be accurately determined and the fault diagnosis efficiency of the transformer is improved.