Faulty Feeder Detection Method for Distribution Networks Based on Time-Frequency Image Mapping

Rapid and reliable detection of faulty feeder is critical for power system when a single-phase-to-ground fault occurs in neutral non-solidly grounded distribution system. This paper introduces a novel faulty feeder detection method based on time-frequency image mapping of zero-sequence current. Firstly, the zero-sequence current is mapped into grey scale figure in time-frequency domains with short-time Fourier transform. All the fault features, not just one of them, can be used to build the detection scheme in this way. Then the time-frequency images of each feeder are sent to a convolutional neural network (CNN), which is the classifier to detect the faulty feeder from healthy ones. The training set is generated by digital simulations on PSCAD/EMTDC with a typical distribution network. The validation set is from other simulation models with different topology and parameters to verify the generalization of proposed method. The results show that the proposed method has good performance with higher accuracy and better generalization. And it can be put into practice directly with no additional equipment investment.