Fault-Line Selection for Rural Distribution Networks Based on a Spatio-Temporal Fusion Neural Network

Abstract

With the large-scale integration of distributed generation, rural distribution networks are increasingly affected by long feeders, tree-contact faults, and high-resistance grounding. These conditions make transient fault characteristics weak and highly complex. This paper proposes a fault-line selection method for rural distribution networks based on a spatio-temporal fusion neural network. The proposed method directly takes raw zero-sequence voltage and branch-current waveforms as inputs. A temporal convolutional network adaptively decouples time-domain features and reduces information loss, while a graph attention network dynamically aggregates spatial topological features. Experimental results show that the proposed method maintains high selection accuracy and strong generalization under 1500-ohm high-resistance grounding, strong noise interference, and different data structures in complex rural distribution networks.