Faulty feeder detection of single phase-to-ground fault for distribution networks based on improved K-means power angle clustering analysis

•Compared with the faulty feeder detection method based on single fault feature quantity, the weak fault information is amplified gradually with the increase of the dimension of fault features.•The boundary between the fault condition and non-fault condition is generated automatically by analyzing the historical feature samples with clustering algorithm.•The proposed method can ensure a higher correct rate of the fault recognition even for intermittent arc ground fault and high-impedance single phase-to-ground fault.•Based on the spatial density and maximum neighbor distance principle, the calculation efficiency of large-scale fault data can be improved while ensuring the calculation accuracy.