GNN ‐Based Partitioning Enhanced ADMM for Reactive Power and Voltage Optimization in Distribution Networks

Targeting voltage fluctuations in distribution networks with high distributed PV penetration and the high computational cost of centralized optimization, this paper proposes a decentralized reactive power and voltage optimization method using comprehensive index‐driven GNN partitioning and improved ADMM. First, based on voltage sensitivity, electrical distance, and node membership, a comprehensive partition evaluation index is developed to assess electrical coupling, partition rationality, and scale balance. Second, this index is integrated into GNN training and partition evaluation, enabling adaptive partitioning that considers both topology and electrical characteristics. Based on the obtained block results, a partition optimization model considering voltage deviation, reactive power regulation cost and resource regulation balance is constructed with the reactive power of photovoltaic inverter and the output of compensation device as control variables, and the improved alternating direction multiplier method is used to solve the problem. Simulations on the IEEE 33‐node system show that the proposed method achieves partitions with clear electrical coupling and good connectivity. It effectively improves node voltage profiles, reduces voltage deviation, and enhances distributed solution performance.