Fault Cause Identification of Single‐Phase‐to‐Ground Faults in Distribution Networks Based on Semantic Segmentation of Zero‐Sequence Currents

ABSTRACT

Accurate identification of single‐phase‐to‐ground (SPG) fault causes is important for fault inspection and condition‐based maintenance in medium‐voltage distribution networks. However, identification remains challenging because different grounding media and discharge processes can produce complex and similar zero‐sequence‐current (ZSC) waveform patterns. To address this problem, this paper proposes a method based on semantic segmentation of ZSC waveform images. First, the causes are grouped into overhead‐line wet‐medium grounding faults, overhead‐line dry‐medium grounding faults, and cable‐line arcing grounding faults. Their characteristics are then analysed according to grounding‐medium conductivity, zero‐crossing distortion and arc‐related transients. Based on these differences, three‐cycle ZSC segments are normalised into standardised greyscale images with pixel‐level labels for waveform traces, low‐current‐distortion regions and arcing‐waveform regions. An edge‐enhanced multi‐scale dilated residual U‐Net learns global waveform morphology and local fault‐related evidence, and pixel proportions in the predicted masks determine the final fault cause. Independent simulation and field tests show that the method achieves high accuracy. The results demonstrate that the method provides accurate, noise‐robust and interpretable evidence for post‐fault inspection and maintenance.