Life Prediction of Dry Reactor Sensor Based on Deep Neural Network

Abstract

In order to solve the problem of increasing the number and service life of a dry-type air core reactor and frequent interturn insulation faults, this paper proposes a life prediction method of a dry-type reactor sensor based on the deep neural network. On the basis of summarizing the research status of turn-to-turn insulation-related problems, this method studies the switching overvoltage generated in the process of breaking the dry-type air core reactor, the deterioration law of turn-to-turn insulation under the cumulative action of switching overvoltage, the influence of thermal aging on the Switching Overvoltage Withstand characteristics of turn-to-turn insulation, and the electrical aging life of turn-to-turn insulation under the power frequency overvoltage. Based on the deep neural network, the electrical aging life model of turn-to-turn insulation of the dry-type air core reactor under power frequency overvoltage is obtained. The results are as follows: with the increase of the applied voltage amplitude, the deterioration speed of the turn-to-turn insulation of the model sample accelerates. When the applied voltage amplitude reaches a certain value, the maximum discharge amount and pulse discharge power of the partial discharge pulse increase rapidly, and the image coincidence degree reaches 85%. The electric aging life curve of the modified interturn insulation model sample of the dry-type air core reactor has a high correlation with the measured aging life data, and the performance is more than 95%. The research results of this paper lay a practical foundation for further research on the deterioration mechanism of interturn insulation under the combined action of multiple factors and provide theoretical support for the risk and life assessment of the dry-type air core reactor.