A Method of Faulted Section Location in Distribution Network Based on Siamese Network Under Small Sample

To address the issue of inaccurate fault location in data-driven fault location methods for distribution networks under small sample, a Siamese network-based fault location method is proposed. Firstly, the fault signals of the global transient zero-sequence current are collected and subjected to continuous wavelet transform to obtain time-frequency images. A convolutional neural network model is introduced to achieve fault section location discrimination. Furthermore, cross-pairing of fault samples is carried out to recombine them, resulting in a significant expansion of the small samples. Meanwhile, a Siamese network model consisting of two sub-models is designed for the expanded samples. Finally, a dedicated classifier is designed for the end of the Siamese network to achieve fault location in small sample distribution networks. The effectiveness of the proposed model is validated through testing in a 10kV distribution system constructed on the MATLAB/Simulink platform.