Optimal Matching of Ring and Radial Wiring Modes with Three-Level Feeder Automation Configuration in 10 kV Urban Distribution Networks Based on Graph Attention Networks

The matching between wiring modes and feeder automation (FA) levels in 10 kV urban distribution networks lacks systematic methods, particularly neglecting communication infrastructure constraints. This paper proposes a two-stage matching model based on graph attention networks (GAT) that fuses electrical, topological, and information and communication technology (ICT) features. A 360-scenario dataset is constructed from IEEE 33/69-node test systems, covering four wiring modes (single-radial, multi-section radial, two-supply ring, three-supply ring) and three FA levels (FA-I/II/III). The first stage employs GAT for wiring mode identification, achieving 91.7% accuracy. The second stage recommends optimal FA configurations through feature fusion with embedded ICT feasibility hard constraints, attaining 83.3% Top-1 matching accuracy. Transfer validation on 72 operational feeders from three Nanjing districts yields 87.5% wiring mode accuracy and 79.2% Top-1 FA matching. Communication features contribute a 3.1 percentage-point gain to matching accuracy, and the constraint mechanism effectively prevents infeasible FA recommendations in low-coverage scenarios. Ablation studies confirm the superiority of the two-stage cascaded architecture over end-to-end alternatives.