The Study On Multi-information Diagnosis Method Of Power Transformer Winding Mechanical State

With the development of the smart grid construction and the improvement of the electric power equipment state examination and repairmen, the requirements of the operation reliability of power transformer and comprehensive state maintenance is higher and higher.The basis of finding the transformer’s potential failure is to assess accurately and diagnose the working transformer state. It is also the important guard of the safety operation of power system. The winding mechanical state diagnosis is the important core part of the transformer state maintenance management. But nowadays, the existing state diagnosis guidelines and methods of winding machine have some problems, such as the off-line judgments, diagnosis signal monoculture, inaccurate results of small fault diagnosis and incomprehensive data source. Therefore, this dissertation makes further study combining the theory analysis with practical application from the views of the state assessment diagnosis method, simulation modeling and system development, using the S11-M-500/35 type oil immersed power transformer as the research object and combing the multi-state and multi-parameter signal analysis with information fusion theory. This dissertation is based on the state grid’s basic and forward-looking cooperation project State Evaluation Method of Power Transformer Online Multi Information Fusion with academicians team. The main research work is as follows:(1) Based on the characteristics theory study of transformer winding mechanical state,this dissertation establishes vibration simulation model of the multi-field coupling winding and the finite element model of the short-circuit reactance. It proposes the short circuit reactance on-line calculation method. This dissertation calculates each phase winding’s induced voltage and current, magnetic flux leakage, displacement distributions and vibration acceleration etc. It gains the characteristic relationship between clamping force and winding vibration, body and the oil tank surface vibration. This dissertation also discusses the relationship between the winding deformation and the variety of the short-circuit reactance and builds the on-line calculation model of short circuit reactance. And then, this dissertation verifies the accuracy of the model according to the actual measured data. This dissertation establishes an evidence body criterion knowledge base of the transformer each phase winding fundamental frequency vibration and short-circuit reactance, based on the numericalsimulation, actual measured data and experts’ experience.(2) The dissertation proposes the characteristics value extraction method of power transformer”s short-circuit reactance and windings’ frequency vibration. For the vibration signals of transformer oil tank surface windings in different states, the dissertation extracts feature using wavelet packet entropy spectrum. And it uses fuzzy c-means clustering algorithm to verify the classification validity of this method. Ultimately, the dissertation draws the vibration signal change characteristics regular pattern of transformer windings in different states.(3) This dissertation establishes the diagnosis method model of winding mechanical state based on evidence theory and support vector machine theory. The dissertation puts forward the classification state diagnosis concepts, which assesses the state first and then diagnoses it. Besides,the dissertation also designs the logical process of the diagnosis model.The dissertation puts the information entropy, fuzzy theory, D-S evidence theory and support vector machines theory into state diagnosis model organically. As for the state level assessment part, the mechanical state of the winding is divided into three states, normal,attention, and ordinary faults. This dissertation puts forwards a double-layer evaluation system which includes the factor layer and decision layer and combines two evidence bodies which are the fundamental vibration and short-circuit reactance. And then, this dissertation assesses the current winding mechanical state grade comprehensively. As for fault diagnosis part, this dissertation proposes the particle swarm optimization(PSO) and the multi-class support vector machine(SVM) of fault diagnosis model. This dissertation uses the multi-classification concept to combine and encode the second grade support vector machine classifier. Meanwhile, taking the advantage of particle swarm optimization algorithm, this dissertation optimizes the penalty factor of diagnosis model and kernel function. Comparing the model diagnosis results with neural network diagnosis results, this dissertation consumes that the classification and accuracy of the proposed method are higher than that of the traditional fault diagnosis methods, which provides a solution to diagnosing winding mechanical state accurately.(4) This dissertation designs, researches and develops the multi-parameter data acquisition system and fault diagnosis system of the transformer winding mechanical state.The dissertation gives the overall design and implementation scheme of the system. And the hardware device and software system platform have already applied for the nationalinvention patent. This dissertation also gets through the national metrology station and software test center’s detection, and validates the reliability of the system. This dissertation builds winding fault diagnosis experimental platform and conducts verification test to the diagnosis system. The dissertation diagnoses the four states of the transformer windings respectively, which are normal, inter turn insulation shedding, low voltage horizontal variation compression and the winding looseness under the impact of sudden short circuit state. And compared with the existing reactance method diagnosis results, it is showed by the results that state diagnosis method which is proposed in this dissertation can effectively compensate the disadvantage of reactance method to the winding slight loosening and fault type which can not be detected or judged. It is proved that the great accuracy and advantages of winding state diagnosis method. Meanwhile, this dissertation verifies the practicability of the diagnosis system.