A multi-source benchmark dataset for day-ahead electricity-price forecasting in the Guangdong spot market

GD-Spot v1.1 provides a multi-source dataset and reproducible analysis code for day-ahead and real-time electricity-price forecasting in the Guangdong spot market. It contains 12 data tables: eight market tables, two ICON Global forecast vintages for 21 cities, a holiday calendar and a typhoon-proximity table. The tables contain 138 documented columns, including repeated keys and provenance fields, and 638,288 stored rows counted across tables rather than independent time points. Market inputs include settlement-price targets, nodal-price reference signals, day-ahead forecasts and ex-post power fundamentals, congestion, West-to-East transmission and features derived from 15-minute records. The common analysis window is 1 January 2025–30 April 2026; individual table and column coverage differs and is documented in the data dictionary and coverage records. Dates and hourly keys use Beijing time (UTC+08:00). The authors identify the two settlement targets as the demand-side day-ahead and real-time uniform settlement-point prices in CNY/MWh. The record includes a machine-readable dictionary, coverage audits, source and transformation records, file hashes, archived weather and cyclone inputs, station-validation records, an ERA5 reference for retrospective weather substitution, and build, benchmark and figure code. Bidding-time availability is evaluated at 12:30 Beijing time on the day before delivery. ICON Global d2 is the primary forecast vintage under an explicit initialization-plus-six-hour availability convention; actual historical publication times are not independently verified. Ex-post inputs are stored separately. Public third-party hourly comparisons and workbook-to-CSV checks provide numerical corroboration of the archived settlement fields, but official records of the same fields were not available to the authors for reconciliation and final-settlement-version verification remains incomplete. The processed market inputs were obtained from a data provider under a signed agreement with the research team. The agreement permits public downloading and CC BY or CC0 reuse of the supplied and author-derived market tables; this version distributes the eight market tables under CC BY 4.0. No news records, indices or annotations, fuel-price values, carbon-price values or national CEA source pages are included. Optional news, fuel and CEA workflows supply methods, code and synthetic examples or source/alignment documentation for users who obtain their own inputs; they do not establish additional forecasting benefit. Licence scope: CC BY 4.0 applies to the dataset materials designated in RIGHTS.md, subject to the component-specific terms recorded there. Open-Meteo/DWD forecasts and archived API responses, GeoNames coordinates, ERA5 via Open-Meteo, and the authors’ derived records retain their respective attribution requirements. IBTrACS files remain subject to the IBTrACS v04r01 product terms, including the product guidance on commercial use; no blanket CC0 designation is made for those files. The Figure 5 boundary subset is under CC BY 3.0 IGO. Original GHCNh observations are CC0 1.0, while the included author-derived validation records are designated CC BY 4.0. Only the author software and software documentation explicitly listed in code/LICENSE_SCOPE.md are MIT-licensed. The calendar table is designated CC BY 4.0; the generating chinesecalendar package is MIT-licensed. See RIGHTS.md and code/NOTICE for full attribution, modifications and component boundaries.