Construction of Sample Data for Line Disconnection Faults in Distribution Networks Aimed at Deep Learning Fault Diagnosis

Fault diagnosis in distribution lines is crucial for the safe operation of power systems, especially for the timely identification and handling of line faults. This paper focuses on the application of deep learning in the diagnosis of line faults in distribution networks, discussing the significance of fault type identification and analyzing the characteristics of line faults, including both source-side grounding and load-side grounding faults. When a fault occurs, the three-phase symmetrical structure of the distribution network transforms into a threephase asymmetrical structure, and the zero-sequence current, negative-sequence current, zero-sequence voltage, and negativesequence voltage signals generated in the line contain rich fault information. To support the effective training of fault diagnosis models based on deep learning, this paper proposes a method for constructing sample data for line faults in distribution networks. Considering the high data requirements of deep learning, this study conducts PSCAD simulation experiments by building a distribution network structure similar to that of a physical test site to expand the limited real test data. To ensure the validity of the simulation experiments, the paper also verifies through comparisons of simulation waveforms with physical test waveforms and simulation data with theoretical values. The research findings will provide an important data foundation for the application of deep learning models in the diagnosis of faults in distribution lines.