Accelerating Security-Constrained Optimal Power Flow with Graph Neural Networks for Renewable Energy Integration

Accelerating Security-Constrained Optimal Power Flow with Graph Neural Networks for Renewable Energy Integration (Proposals Track)

Zhenhua Zhang (UC San Diego)

Cite

Power & Energy Climate Science & Modeling Hybrid Physical Models

Abstract

The rapid deployment of variable renewable energy (VRE) resources presents significant electric grid management challenges due to their intermittent nature and the computational complexity of solving real-time AC Optimal Power Flow (AC-OPF) problems under uncertainty. In this study, we propose a Graph Neural Network (GNN) approach to achieve computationally efficient and robust security-constrained AC-OPF decisions. We will train different GNN architectures (SimpleGCN, GCN, GAT) to learn optimal power dispatches and locational marginal prices under N-0 contingencies, while systematically evaluating methods to ensure physical feasibility. We will then test model performance under N-1 contingencies. Preliminary results on DC-OPF show SimpleGCN outperforms attention-based models, although generalizing to unseen N-1 topologies remains a key challenge for future work.