Detection Method of Turn to Turn Insulation Short Circuit Fault of Dry-Type Air-Core Reactor Using Vibration Characteristics Based on Machine Learning

When the reactor produces turn-to-turn insulation fault, it will cause great damage to the long-term used reactor. Turn-to-turn short-circuit, as a common and frequent case of insulation failure, will cause a great disturbance to the stability of the power system if not detected in time. In this paper, a multi-factor turn-to-turn insulation short circuit fault detection method based on machine learning is proposed. This paper chooses the features from both time domain and frequency domain. The experimental platform has been established to simulate the occurrence of turn-to-turn insulation fault. The performances of different generated single-point classification models have been evaluated and discussed. The decision-tree-based model has a better performance compared with KNN and SVM model. The improved multi-point models has been proposed and evaluated. The results show that the neuro-network can also be selected as the classification model.