Research On Transformer Fault Diagnosis Method Based On Information Fusion Technology

Transformer plays an important role in the whole power transmission system and distribution system,and its running state is directly related to the stable operation of the whole power equipment system.The analysis of characteristic gas in transformer oil is one of the most common and effective methods for transformer fault diagnosis.Most transformer fault diagnosis methods are more accurate for single fault diagnosis,but not for comprehensive fault diagnosis.In this paper,the BP neural network has the advantages of self-learning habit,self-adaptability,fault tolerance and D-S evidence theory,through mapping between transformer faults and characteristic gases,judge the specific types of transformer internal faults,so as to improve the accuracy of fault diagnosis.First of all,the electrical and thermal faults of the transformer are summarized.According to the corresponding relationship between the characteristic gas generated when the transformer fails and the fault type,the transmission process of BP neural network is studied.The hidden layer parameters are determined by Matlab simulation.The 5-15-6 BP neural network model based on characteristic gas input and 3-7-6 BP neural network model based on three ratio input were established,and the training effect of the network was verified,and the performance of the network was analyzed.The original data of characteristic gas in transformer fault were constructed as three characteristic vectors,namely characteristic gas,gas content ratio and ratio of three,as the input of the preliminary diagnosis layer,and three probability outputs were obtained to carry out the preliminary diagnosis of transformer fault.Second,the research will be D-S evidence theory is applied to the feasibility in transformer fault diagnosis,the BP neural network combined with D-S evidence theory,based on information fusion diagnosis method of transformer model,using the fusion diagnosis D-S evidence theory,the probability distribution of different fault types of transformer,determine the actual transformer fault type.At the same time,the transformer diagnosis method based on information fusion technology is compared with the primary diagnosis result of BP neural network,and the conclusion is drawn that the transformer diagnosis method based on information fusion technology can diagnose transformer comprehensive faults.Finally,through the verification of a large number of sample data,the accuracy of transformer fault diagnosis based on information fusion technology is close to94%,which greatly improves the reliability and accuracy of diagnosis,reflecting the superiority of information fusion technology.