Abstract:

Single-phase-to-ground fault (SPGF), being the most prevalent issue in distribution networks, significantly impacts the reliability and safety of the distribution system. Accurate identification of SPGF can enhance the level of refinement in handling grounding faults in distribution networks. Firstly, a set of candidate waveform features that effectively reflect various grounding fault causes is extracted from the fault waveforms. These features are then subjected to multivariate analysis of variance (MANOVA) to assess their correlation with grounding fault causes, thereby selecting effective features for identifying the root causes. Subsequently, fault cause identification models based on Extreme Learning Machine (ELM) and Support Vector Machine (SVM) are designed respectively. These models' recognition results are fused using Dempster-Shafer (D-S) theory of evidence fusion, establishing a comprehensive identification model for grounding fault causes. Finally, the validity of the established comprehensive identification model is verified based on field data, demonstrating its superiority over any single identification model and confirming its feasibility.

Key words: ground fault cause, single-phase-to-ground fault, extreme learning machine, support vector machine, D-S evidence theory.