CBD: Nodal electricity price field tensors and forecasting pipeline (CAISO/MISO/NYISO, 2023–2026)

Derived tidy datasets and the complete experiment pipeline for the manuscript "Topology-free probabilistic forecasting of full nodal electricity price fields via congestion-basis decomposition" (submitted to Sustainable Energy, Grids and Networks). Contents: canonical hourly price-field tensors for CAISO (1,926 nodes, 2023-04 to 2026-06, day-ahead LMP with all five published components) and MISO (2,165 nodes, 2023-01 to 2026-06, day-ahead and real-time ex-post LMP with congestion and loss components); NYISO 15-zone tidy tables (2020-2025);official day-ahead load-forecast covariates (CAISO ISO forecast, EIA-930);frozen-protocol evaluation results including baseline-family comparisons and Diebold-Mariano tests; and the full Python pipeline with the frozen protocol document. All raw data originate from public sources (CAISO OASIS, MISO market reports, NYISO public CSV, EIA-930 API). See README.md for the data dictionary and reproduction steps. Data are released under CC BY 4.0; code under the MIT License. CBD: Nodal electricity price field tensors and forecasting pipeline (CAISO / MISO / NYISO, 2023–2026) Companion data-and-code deposit for the manuscript "Topology-free probabilistic forecasting of full nodal electricity price fields via congestion-basis decomposition" (submitted to Sustainable Energy, Grids and Networks). All raw data originate from public sources: CAISO OASIS API, MISO public market reports, NYISO public CSV archives, and the U.S. EIA-930 API. This deposit contains the derived tidy datasets and the complete experiment pipeline. Archive contents Archive Contents cbd_tensors_caiso.zip CAISO canonical hourly tensors (UTC axis), plus the cross-market missing_flags.parquet; see data dictionary below cbd_tensors_miso_da.zip MISO day-ahead tensors plus shared MISO metadata (time axis, node master, covariate) cbd_tensors_miso_rt.zip MISO real-time ex-post tensors (LMP, MCC, MLC) cbd_nyiso_processed.zip NYISO tidy parquet tables (5 series types, 2020–2025, UTC) cbd_results.zip Frozen test results, baseline-family comparisons, probabilistic references, DM tests (JSON + per-node daily-loss NPZ; see RESULTS_GUIDE.md inside) cbd_code.zip Full pipeline source (src/), method package (src/cbd/), frozen protocol document (D0-definitions.md), this README Data dictionary (cbd_tensors.zip) File Shape / type Description caiso_da_{lmp,mce,mcc,mcl,mghg}.npy (hours × 1926) float32 CAISO day-ahead LMP and its four published components; NaN = missing (0.09%) caiso_dayblocks.npz lmp/mcc (days × 24 × 1926), days CAISO trading-day blocks (OPR_DT, Pacific prevailing time; DST days logged) caiso_time_utc.npy, caiso_nodes.txt — Hour axis (UTC) and node identifiers caiso_cov_loadfcst.npy, caiso_cov_dayblocks.npy (hours,), (days × 24) CAISO official day-ahead system load forecast covariate miso_{da,rt}_{lmp,mcc,mlc}.npy (30648 h × 2165) float32 MISO day-ahead / real-time ex-post LMP and components (EST market, UTC axis) miso_time_utc.npy, miso_nodes.txt, miso_node_master.parquet — Time axis, node list, node master table miso_cov_loadfcst.npy (hours,) EIA-930 day-ahead demand forecast covariate missing_flags.parquet 3,661 rows All missing node-days, classified (all drop_node_day; no interpolation) *_log.md / *_log.json — Parse and QC logs Reproduction Environment: Python 3.10, NumPy 1.26, LightGBM 4.7, pandas, pyarrow (deep baselines additionally: PyTorch 2.x, chronos-forecasting). Pipeline order (all under src/): p1_*.py — raw → tensors (already materialized in this deposit) p2_validation_{caiso,miso}.py, p2_prob_routes.py — validation-window selection p3_caiso_day_tensor.py, p3_run.py — frozen test period (run once), p3_run_amend2.py, p3_amend1_zombie_filter.py — labelled amendments p3_nyiso_control.py, p3_dart.py — control and economic backtest a30_*.py — deep/foundation-model baselines (GPU) p4_*.py — merges, probabilistic reference, CRPS DM tests, figures The frozen evaluation protocol (D0) is included as D0-definitions.md. Licenses Data (tensors, parquet, results): CC BY 4.0 Code (cbd_code.zip): MIT License Underlying raw market data remain subject to the respective ISO/EIA terms of use; they are public and were accessed via documented public interfaces. Citation Zhang, J., Zhang, H., Shen, L., & Xu, Z. (2026). CBD: Nodal electricity price field tensors and forecasting pipeline (CAISO/MISO/NYISO, 2023–2026) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21522100