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

Published in Journal of Machine Learning for Modeling and Computing (2024).

This work introduces a hybrid chance-constrained optimal power flow (HCC-OPF) framework that integrates enhanced multi-fidelity graph neural networks (EMF-GNN) as fast and accurate surrogate power flow solvers under high-dimensional load and renewable generation uncertainty.

The proposed EMF-GNN architecture fuses low-fidelity DC power flow simulations with high-fidelity AC simulations through a residual multi-fidelity structure, significantly reducing training cost while improving voltage magnitude and phase angle prediction accuracy across IEEE benchmark systems.

To ensure operational reliability, a hybrid optimization mechanism selectively invokes the exact AC power flow solver near constraint thresholds, preventing surrogate-induced constraint violations while preserving computational efficiency.

Extensive experiments on IEEE 9-, 14-, 30-, and 118-bus systems demonstrate:

• Improved surrogate accuracy compared to single-fidelity and prior multi-fidelity GNN models
• Reduced constraint violation probability under uncertainty
• Robust performance under N-1 security contingencies
• Significant computational speedup relative to full AC-based chance-constrained OPF

Methods: Multi-fidelity graph neural networks, chance-constrained optimization, hybrid surrogate-based OPF, Monte Carlo uncertainty modeling, N-1 contingency analysis

Recommended citation: Khayambashi, K., Hasnat, M. A., & Alemazkoor, N. (2024). Hybrid chance-constrained optimal power flow under load and renewable generation uncertainty using enhanced multi-fidelity graph neural networks. Journal of Machine Learning for Modeling and Computing, 5(4), 53–76.
Download Paper