Power Transformer Condition-based Maintenance Base On Biomimetic Pattern Recognition

As Murphy’s Law, a place where there may be problems with certain problems. We can’t completely avoid the transformer failure, the only thing to do is to reduce the transformer failure caused by the loss. The study conducted in this paper is for this purpose. The transformer condition-based maintenance by monitoring the state of the transformer, the potential failure of the transformer can be found hidden in a timely manner to eliminate hidden dangers, to extend the line transformer running time, increase operating efficiency. Transformer failure has occurred, the right to judge the type of fault can quickly determine the site of failure may occur, rapid repair, reduce downtime running time. With the development of China’s power industry, grid capacity is increasing, increasing supply reliability requirements, timely and accurate discovery of the transformer internal fault, reasonable arrangements for maintenance, has become a very urgent and important task.This paper first describes the current development of the condition-based maintenance, a brief introduction to the evolution of the maintenance strategy. Be able to detect early failure of the transformer in the transformer oil dissolved gas analysis, the second chapter details the mechanism of the gas dissolved in transformer oil and internal fault type and the relationship of the dissolved gas content, and the electricity sector analysis of transformer failure discuss principles and methods of the ratio method. The traditional three-ratio method lack of fault finding out coding with the development of artificial intelligence, many diagnostic methods based on artificial intelligence algorithms in transformer fault diagnosis, and provide a new avenue of research for transformer fault diagnosis.Traditional fault diagnosis methods are established on the basis of "segments", according to the classification criteria to find the optimal classification interface, so the algorithm are inclined to different things,"difference", that is, the distinction between a sample with limited class of known samples. In fact, this is contrary to the process of human understanding of things. In the process of understanding things is a class of a class, a focus on the contact of similar things, details of the distinction. Therefore, the Institute of Semiconductors, Chinese Academy of Sciences Academician Wang Shoujue, proposed the concept of the biomimetic pattern recognition, pattern recognition from the point of view of human understanding of things. The algorithm has a better application in the field of speech recognition and image processing. In the third chapter is given in detail basic mathematical theory and implementation of biomimetic pattern recognition. This paper attempts to apply this method to transformer fault diagnosis, proven to achieve better results, especially in the case of small sample size, better than probabilistic neural network recognition effect.At lase, the classification of the current transformer condition assessment, given the state of a power supply bureau overhaul instance.