"Physics-Aware Heterogeneous Graph Neural Networks for AC Optimal Power Flow"

"This dataset contains experimental results for physics-aware heterogeneous graph neural networks applied to AC Optimal Power Flow (OPF) prediction. It includes trained model checkpoints for three IEEE test systems (case30, case57, and case118) with 5 random seeds each, totaling 15 mainline experiments. The dataset comprises PyTorch model checkpoints (.pt.tar.gz), processed graph datasets, YAML configuration files, and comprehensive experimental metrics in JSON format. Key metrics include feasible rate, flow MSE, KCL violations, and thermal limit violations. Additionally, the package contains baseline comparisons (MLP, Homo-GNN, Hetero-Sup-Only), ablation study results (no-box, no-edge-attr, no-edge-decoder), and scale\/OOD robustness evaluations. All experiments use the PyTorch Geometric library for heterogeneous message passing. Researchers can utilize this data to reproduce our results, train new OPF solvers, or conduct further analysis on physics-informed neural networks for power systems. The dataset is released under CC BY 4.0 license."