Research On Fault Diagnosis For Power Transformer Based On Improved Gray Relational Analysis Theory And Elman Neural Network

The safe and stable operation of power transformer is to ensure the normal power supply and the basis of social production and life. It has the important practical significance to raise the maintenance level of the transformer that discovering the potential failure of transformer promptly and accurately. Based on the principle of dissolved gas analysis in oil, applying the improved gray relational analysis theory and Elman neural network into transformer fault diagnosis, to a certain extent, the fault recognition rate has been improved.The basic idea of gray relational analysis is according to similarity between curves to determine the correlation between factors. By use of the gray relational analysis theory, we can judge the fault by interrelatedness between transformer standard fault mode and test mode. Being aimed of the weakness of traditional method, at first, this paper makes interrelatedness resolution coefficient better to reflect integrity and relevance of the system. Then we introduce the distance analysis method, which can seek interrelatedness by use of non-uniform weight form, and also can make up weakness such as the tendencies of local point and information loss in traditional methods. The improved gray relational algorithm applied in transformer fault diagnosis can enhance the diagnosis rate effectively.Nowadays, artificial neural networks have been used in the field of fault diagnosis widely. Hence, in this paper, Elman network with internal feedback connection is used to seek fault diagnosis. While due to the feedback connection, Elman network has sensitivity to the data of history state. Because Elman network adopts BP algorithm to amend the weight and threshold value, which are easy to converge to local minimum points, Elman network diagnosis model with momentum term is built, namely, momentum term is added during network amend weight. In addition, considering the fast speed convergence characteristic of Levenberg-Marquardt (LM) algorithm by using of approximate second derivative information, so LM algorithm is introduced to the Elman neural network weights and thresholds for the study as a trial, and the transformer fault diagnosis model is established based on LM-Elman neural Network. Finally, by testing trained neural network, the results show that the improved diagnosis model of Elman neural network can check out the transformer fault types well compared with BP neural network. So it has advantages in transformer fault diagnosis.In this paper, we have made beneficial attempts for the transformer failure diagnosis research, and the new scheme has certain practical significance to improve the safe and economic operation level of transformer and the reliability of power system operation.