Fast Identification of Static Voltage Weak Nodes in Large-Scale Power Grids Using GATv2-JK

Rapid identification of static voltage weak buses in large-scale power grids is hindered by the computational burden of bus-by-bus continuation power flow (CPF) verification. This study proposes a GATv2–Jumping Knowledge (GATv2-JK) framework that screens operating conditions by system voltage stability margin and uses CPF-derived bus load margins to supervise weak-bus scoring. GATv2 attention, multi-hop feature fusion, and a local ranking constraint prioritize high-risk buses; a validation-calibrated adaptive strategy selects candidates for CPF re-verification. A dataset of 400 operating conditions with system margins below 15% was constructed from a practical 2475-bus grid containing 333 positive-load buses. Buses with CPF load margins below 10% were labeled as weak. On the test set, the model achieved an area under the precision–recall curve (AUC-PR) of 0.8002 and a Recall@20 of 0.9478. Under the adaptive CVaR0.95 setting, it selected 17.22 buses per condition on average (5.17% of positive-load buses) and attained a sample-averaged recall of 0.9863. These results indicate that the framework prioritizes weak buses while substantially reducing the number of CPF re-verifications required under the studied low-margin conditions.