Research On Transformer Condition Assessment And Fault Diagnosis Method Based On Information Fusion

Power transformer is the most widely used hub equipment in power system,whose operation status directly affects the safe operation of the whole power system.Once the fault or accident occurs,it will cause great negative impact.Systematic and in-depth study of transformer condition assessment and fault diagnosis technology is of great theoretical and practical significance to prevent accidents,ensure the safe and stable operation of power system,and promote state maintenance.In this paper,the transformer condition assessment and fault diagnosis are studied as follows:Firstly,according to relevant regulations and standards,guided by information fusion technology and combined with engineering practice,the state evaluation index system of layered parts is constructed by optimizing the state indexes,which makes the state evaluation more targeted and effective.Aiming at the limitation of single index diagnosis,a fault diagnosis evaluation index system combining the information of DGA,electrical test,oil test and historical inspection records was constructed to improve the reliability of diagnosis results.According to the relevant transformer condition evaluation guidelines,the classification standards based on five state grades of good,normal,attention,abnormal and dangerous,the corresponding maintenance strategies for each state are established.The evaluation process of transformer condition evaluation and fault diagnosis was constructed,and the fault diagnosis is carried out according to the result of condition evaluation,which provides guidance for formulating more targeted and economical maintenance plan.Further,in view of the uncertainty of each state evaluation index,the transformer state was evaluated by the method of multi-source information fusion,whose evaluation model was constructed based on multi-dimensional information fusion.Aiming at the traditional evidence theory in the conflict information fusion runs counter to the conclusions and the actual problem,introducing the support probability distance function,which is effectively to quantify the degree of conflict between the body of evidence.On this basis,the conflict evidence is reasonably weighted and weakened to ensure that the evaluation can still be effective when an index is abnormal.The validity of the transformer condition evaluation model has been verified by an example,and compared with the classical evidence theory fusion method.The evaluation results of the improved fusion method presented in this paper are not only consistent with the actual state,but also more centralized and reliable.Finally,the extreme learning machine combined with improved evidence theory was introduced into transformer fault diagnosis,then a multi-dimensional information fusion transformer fault diagnosis model based on ELM and improved evidence theory was established.Aiming at the problem that the traditional network can only achieve label classification and may have hard decision,this paper adopts the method of a posteriori probability mapping to process the output results of ELM network and get the probability distribution of the corresponding labels.The probability distribution matrix is fused with the evidence theory based on supporting probability distance,so that the transformer fault classification is realized.The validity of the fault diagnosis model has been verified by an example,and compared with the diagnosis results of a single probability extreme learning machine,whose results show that the fault diagnosis method presented in this paper has a better classification and recognition effect,and the accuracy of the diagnosis results reaches 92.86%.The above research and conclusion in this paper provide a strong support for the formulation of effective and economic maintenance strategies and the realization of transformer condition maintenance.