A self-healing method for faults in low and medium voltage substations during large-scale power outages

To address the coordination gap between distributed generation anti-islanding protection and feeder automation, this study proposes a self-healing method for low/medium-voltage faults during large-scale outages. A multi-domain (time, frequency, wavelet) detection approach was developed using zero-sequence current analysis and a single-phase grounding fault model. Feature selection via random forest and fault zone identification with LightGBM enhance fault response efficiency. A distributed self-healing strategy based on smart terminal collaboration achieves autonomous fault isolation via associated feeder partitioning. Combining switch transfer and DG islanding restores power to non-fault zones. A topological adjacency matrix and load-based repair priority model optimise fault location, network reconfiguration, and repair scheduling. Experimental results demonstrate 50ms fault detection (0.5%-1% false positives) and 95%-99% load recovery, improving grid resilience.