Case 1

Jointly predict active power, reactive power, and voltage magnitude at buses plus active and reactive flow in both directions on every branch. The headline accuracy is the fraction of operating samples satisfying all stated error thresholds.

Case I starts from the IEEE 39-bus, 46-branch system. Local Topology Slicing expands 5,000 selected operating scenarios into one million training subgraphs; separate 100,000-sample test, downscaled, and upscaled sets probe unseen operating conditions.

Nodes represent buses with electrical quantities and edges represent transmission elements with network parameters. Inputs, targets, units, slack-bus treatment, and feasibility status follow the benchmark case generator.

The catalog records these protocols or variants: Standard split. Evaluation uses Accuracy (higher is better). Exact masks or folds must come from the cited release.

Only converged, feasible operating points should be compared under the documented policy. Per-unit conventions, topology changes, solver tolerances, and bus-type handling can silently alter labels; these cases do not establish transfer to real grids.