Dataset adapters

These helpers hydrate the bundled fixtures plus publicly available power-system testbeds so the CLI has ready-to-run inputs for PF, OPF, SE, and TS commands.

RTS-GMLC

gat dataset rts-gmlc fetch --out data/rts-gmlc

Copies the staged MATPOWER network (grid.matpower) and telemetry CSV (timeseries.csv) into the destination directory. Follow up with gat import matpower and gat ts import dsgrid to build the Arrow/Parquet pipeline that every solver command consumes.

HIREN test cases

gat dataset hiren list
gat dataset hiren fetch case_big --out data/hiren

Lists the curated 9–240 bus cases and copies the selected MATPOWER file into the requested output. Use the fetched case as the input for gat import matpower when you need a realistic testbed.

dsgrid fixtures

gat dataset dsgrid --out data/dsgrid/demand.parquet

Writes the built-in dsgrid Parquet snapshot to the supplied path. It is a stand-in for a full HTTP fetch and lets you exercise telemetry joins without anything external.

Sup3r weather

gat dataset sup3rcc-fetch --out data/sup3r/weather.parquet

gat dataset sup3rcc-sample --grid grid.arrow --out data/sup3r/sample.parquet

sup3rcc-fetch copies the offline weather PoP dataset, while sup3rcc-sample ties resources to buses in an Arrow grid. The sampling command is helpful when coupling weather to OPF or contingency scenarios.

PRAS adequacy

gat dataset pras --path test_data/datasets/pras --out data/pras/pras.csv

Normalizes the Probabilistic Resource Adequacy Suite (PRAS) outputs (LOLE/EUE) for Gat scenarios. Use it to load region/season/hour forecasts before running adequacy or SE experiments.

Benchmark datasets

These datasets are used by gat benchmark commands to validate solver accuracy against known reference solutions.

PFΔ (Power Flow Delta)

Power flow perturbation dataset with reference bus voltages and angles for validation.

# Run PF benchmark against PFΔ test cases
gat benchmark pfdelta --pfdelta-root data/pfdelta --max-cases 100 -o pfdelta_results.csv

Output CSV contains: case_name, contingency_type, converged, max_vm_error, max_va_error_deg, solve_time_ms.

PGLib-OPF

Standard IEEE/ARPA-E MATPOWER test cases from the PGLib-OPF repository. Includes baseline objective values for OPF validation.

# Run OPF benchmark against PGLib cases
gat benchmark pglib --pglib-dir data/pglib --baseline baseline.csv -o pglib_results.csv

Output CSV contains: case_name, converged, objective_value, baseline_objective, objective_gap_rel, solve_time_ms.

OPFData (GridOpt)

Large-scale AC-OPF dataset with 300k+ solved instances per grid. Supports load perturbations (FullTop) and topology perturbations (N-1 line/gen/transformer outages). Uses GNN-format JSON.

Reference: arxiv.org/abs/2406.07234

# Run benchmark on OPFData samples
gat benchmark opfdata \
  --opfdata-dir data/opfdata/case118/group_0 \
  --max-cases 1000 \
  -o opfdata_results.csv

Output CSV contains: sample_id, file_name, converged, objective_value, baseline_objective, objective_gap_rel, num_buses, num_branches, solve_time_ms.