Faulted Line Detection Using Sampled Data and GBDT for Non-solidly Earthed System

Single line-to-ground (SLG) faulted line detection has always been a key constraint of the distribution system reliability. This paper focuses on practicability and generalization of detection methods, and therefore a novel SLG faulted line detection method based on sampled value of zerosequence current is proposed. The normalized sampled values used in the method don't require complicated transformations while taking the correlation of lines into account. It uses gradient boosting decision tree (GBDT) as a classifier to tell whether a waveform is normal or not. Consequently, there is no need to retrain GBDT when the structure and parameters change in the operation of the distribution network. With much training work based on multiple data generated by PSCAD simulation, an adaptable GBDT structure is finally determined. The PSCAD and RTDS experiments verified its high performance of generalization ability and application prospects in different distribution systems.