Research Of Fault Diagnosis For Large Power Transformer Based On Information Fusion

Large power transformer is one of the most important and expensive equipments in power industry. Its operation reliability involves the security of power supply and has significant impact on national economy development and people's lives. With the development of power system in the direction of extra-high voltage, large power grids, large capacity and automation, the research of large power transformer's condition monitoring and fault diagnosis technology is carried out to detect the latent failure of power transformer insulation. This plays great theoretical significance on the safe, reliable, stable operation for transformer and even the entire power system, and accelerates maintenance of electrical equipment to transit from regular maintenance to predict maintenance.The power transformer faults are used as diagnosis objects in this thesis. The basic large power transformer faults and methods of fault diagnosis are briefly analyzed. According to the feature and existing state of power transformer fault diagnosis, information fusion technology is introduced into the field of transformer fault diagnosis. A hierarchical decision fusion fault diagnosis model with transformer molt-fault information's based on wavelet neural network and D-S evidence theory with dissolved gases in oil and electrical test dates is proposed according to the general frame of information fusion fault diagnosis.Firstly, the structure, common faults and fault diagnosis approach of power transformers are simply described in this paper.Secondly,the evidence decision layer fusion is put forward for the uncertainty of power transformer fault diagnosis. A belief distribution way based on evidence importance and evidence sufficiency is put forward for the objectivity of evidence theory belief distribution, which causes the reduction of BPA hold by uncertain information and conflict factor. The unknown part of evidence fusion credible distribution is composed of two parts,named unknown parts by the weight distribution and lack of information for the fusion of conflict information, which objectively reduces the uncertainty of diagnosis and optimizes the evidence reasoning algorithm. The fusional result satifies objectively the needs of diagnosis.Wavelet neural network optimized by adaptive genetic algorithm based on real value encoded and binary encoding is presented in this paper according to that BP network used in character hierarchy is easy to fall into local minimum and uneasy to converge. It is proved that the rate of convergence, the accuracy of convergence and the ratio of diagnostic accuracy of proposed network are superior to BP network.Finally, the fault diagnosis model based on wavelet neural network optimized by adaptive genetic algorithm and D-S evidence theory is proved to be effective and reliable through two examples.