Fault line selection in small current grounding systems is challenging due to single-phase grounding faults. Existing methods struggle with feature extraction under varying fault conditions and temporal pattern recognition. To address this, we propose the GRU-S Transform Fault Line Selection (GRU-STF) method. This approach combines an enhanced S-transform with adjustable factors for adaptive time–frequency resolution optimization and a multi-scale analysis framework covering three frequency bands. A BP neural network with Gated Recurrent Units captures temporal dependencies, while dynamic sample balancing using K-means clustering addresses sample imbalance and overfitting. Experimental validation on simulation datasets and field data demonstrates that GRU-STF achieves 95.7% accuracy on simulation data and 91.8% on field data, with an average response time of 28 ms, outperforming existing baselines.
Liping Wen (Mon,) studied this question.