Research On Power Transformer Fault Diagnosis Technology Based On Fault Tree And Signal Analysis

Power transformer is the hub of power system,once failed,it may bring huge economic losses and even cause a security incident.In recent years the outage rate of power transformer increased year by year and reached 0.136 in 2015.Therefore,it is of great significance to study the fault diagnosis technology of power transformer to improve the efficiency and correctness of diagnosis,this thesis has researched power transformer fault diagnosis technology based on fault tree and characteristic signal.This thesis studies the structured expression of fault case an d establishes the case database of transformer failure.Based on the case database this thesis has researched Apriori algorithm to mine strong association rules,and obtained strong association rules between fault phenomena and phenomena,phenomena and cau ses,causes and causes.In order to make full use of the information,this thesis gives the definition of bridging information of power transformer failure,and studies the weak association rule mining method based on bridging information.The similarity model of fault cases based on strong and weak association rules is established,which could be used to similar case retrieval.The automatic modeling method of transformer fault tree based on similar cases is studied.In this thesis,a reachable matrix is e stablished for the detected anomalous sequence,and the fault tree hierarchy information is obtained by matrix boolean intersection calculation.The fault tree is automatically constructed with the help of hierarchy information,and the conflict resolution strategies are used to check and correct the logic and process errors in the fault tree.The fault tree of the power transformer winding is established by this method,and it is the same with the fault tree established by the detected anomalous sequence analysis,which verifies the validity of the method.The non-fault factors corresponding to the abnormal phenomena are summarized as the prerequisite condition of the fault tree reasoning to reduce the false alarm of the non-faulty transformer abnormal diagnosis and improve the accuracy of the fault diagnosis.The fault diagnosis method of power transformer based on characteristic signal clustering analysis is studied.The relationship between the dissolved gas in the oil and the states of the transformer is analyzed.A multi-layer iterative XGBoost algorithm is proposed to identify the transformer failure mode based on the dissolved gas in the oil.The experimental results show that the classification accuracy of the multilayer iterative XGBoost algorithm is higher than other methods.Aiming at partial discharge,the denoising method combining EMD and wavelet hard threshold method is studied.The experimental results show that the method can remove the interference signal in partial discharge signal effectively.Becasue partial discharge signal has large information and the characteristics is distinguishable,this thesis researches the feature extraction method which calculates the energy of each layer of wavelet packet decomposition as the energy characteristics.The multi-layer iterative XGBoost algorithm is used to identify the partial discharge based on energy characteristics.The effectiveness of the method has verified by experiments.This thesis develops a fault diagnosis system of power transformer and designs the architecture and function structure.The system has implied strong and weak rules mining,similarity modeling,fault tree automatic modeling and fault diagnosis based on information analysis.Applying a C-phase high-voltage partial discharge fault case as a test to verify the system.The research of this thesis can enrich the methodology of power transformer fault diagnosis and bring benefits to transformer manufacturers.In addition,the methods of this thesis can be applied to other complex e quipment fault diagnosis.