Hybrid Chance-Constrained Optimal Power Flow Under Load and Renewable Generation Uncertainty Using Enhanced Multi-Fidelity Graph Neural Networks

Power systems are transitioning toward renewable sources and electrification, introducing significant uncertainties in generation and demand that optimal power flow (OPF) methods must manage. Traditional deterministic methods struggle with these variabilities. Additionally, addressing uncertainty in OPF calculations incurs computational burdens due to the need for multiple evaluations across various scenarios. This necessitates the use of advanced surrogate models. However, these models require significant data for training, and surrogate-based optimization can yield unreliable results due to inaccuracies in constraint handling. To overcome these issues, this paper proposes a novel surrogate-based hybrid chance-constrained optimal power flow (HCC-OPF) methodology employing enhanced multi-fidelity graph neural networks (EMF-GNN) as power flow solver surrogates. This model integrates low-fidelity and high-fidelity simulations to significantly reduce training cost while maintaining high accuracy. We further enhance the robustness and accuracy of OPF solutions through a hybrid methodology that selectively uses exact power flow solver to correct surrogate inaccuracies near critical thresholds. Extensive testing on multiple IEEE systems under high-dimensional correlated uncertainty of load and generation shows the EMF-GNN model outperforms existing single-fidelity and multi-fidelity models. Furthermore, the proposed HCC-OPF methodology accurately solves OPF problems across various system sizes and conditions, exhibiting scalability and efficiency. Additionally, it effectively manages N-1 security constraints to further exhibit its robustness under operational challenges.