Robust Method Based on XGBoost for Inter-Turn Faults Detection in Air-Core Dry-Type Shunt Reactor

This paper presents the development and evaluation of a new scheme based on a machine learning classifier named eXtreme Gradient Boosting (XGBoost) for detecting inter-turn faults in air-core dry-type shunt reactors. The proposed scheme was developed to protect a 230 kV air-core reactor installed on the Penedo’s bar of the "Nossa Senhora do Socorro - Penedo" transmission line in Brazil. To achieve this, the XGBoost model was trained and tested for a set of different internal faults in the reactor, generated in Matlab, using the time domain solution. Furthermore, different levels of noise were added to the dataset, making the method more robust. The results show that the proposed solution based on XGBoost is effective in detecting interturn faults in the reactor, even for the most challenging cases in which the overcurrent protections fail to operate due to their low sensitivity.