A multi-source benchmark dataset for day-ahead electricity-price forecasting in the Guangdong spot market
広東省スポット市場における日前電力価格予測のためのマルチソースベンチマークデータセット (AI 翻訳)
Liu, Hongyan, Zhang, Qianqi, Tu, Tianyi, Wen, Fange, Lu, Liangji, Lan, Xiaping, Xu, Zeling, Zou, Yuntao
🤖 gxceed AI 要約
日本語
この論文は、中国広東省の電力スポット市場における日前価格予測のためのマルチソースデータセットを公開する。既存のベンチマークは欧米市場に偏っており、入札時点でのデータ利用可能性が不明確なため、将来情報の混入リスクがある。本データセットは17テーブル・152カラムを統合し、気象、燃料価格、炭素価格、台風近接性、ニュースセンチメントなどを含む。各カラムの入札時点での実利用可能性を明示し、再現可能な予測ベンチマークを提供する。
English
This paper releases a multi-source benchmark dataset for day-ahead electricity price forecasting in the Guangdong spot market, China's largest. It addresses the lack of open data from Chinese markets and prevents look-ahead bias by labeling each column's availability at bid time. The dataset integrates 17 tables (152 columns) covering market, weather, external drivers, and news sentiment, including weather forecasts, carbon prices, and typhoon proximity. It provides a reproducible pipeline and a neutral baseline task.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本においてもJEPXスポット市場の再生可能エネルギー比率上昇に伴う価格変動が課題となっている。本データセットが採用する将来情報混入防止設計やマルチソース統合手法は、日本の市場データ整備や予測モデル開発の参考となる。
In the global GX context
As global electricity markets integrate more renewables, spot price forecasting becomes critical. This dataset offers a rigorous, reproducible benchmark from a major Chinese market, filling a gap in open data outside Europe and the US. Its design principles—preventing look-ahead bias and integrating diverse drivers—are applicable to any market undergoing energy transition.
👥 読者別の含意
🔬研究者:Provides a rigorous, reproducible benchmark for day-ahead price forecasting in Chinese electricity markets, enabling honest back-testing and cross-market comparisons.
🏢実務担当者:Useful for firms participating in Guangdong's spot market or developing forecasting models for renewable revenue assessment and storage arbitrage.
🏛政策担当者:Illustrates the data infrastructure and methodological standards needed to support transparent and efficient spot markets under high renewable penetration.
📄 Abstract（原文）
China's installed renewable capacity overtook coal-fired capacity at the end of 2024, and from 2025 renewable generation is being brought fully into the market with prices set by trading, so that spot-price volatility, in the form of zero, negative and spike prices, has become routine. Day-ahead and real-time price forecasting is therefore a shared need for retailer bidding, storage arbitrage and renewable-revenue assessment, yet the open data underpinning such research remain weak: existing benchmarks come mostly from mature European and U.S. markets, and most do not record whether each column was actually available at bidding time, which invites the accidental use of future information in back-tests. Using Guangdong (a first-batch pilot, the largest spot market by traded energy, renewable-heavy, and located on a coastal typhoon corridor that makes prices weather-sensitive) as a representative Chinese market, we release a dataset for day-ahead price forecasting. On a unified Beijing-time hourly grid it organises 17 tables and 152 documented columns across market, weather, external-driver and news domains, targeting province-wide day-ahead and real-time settlement prices and shipping official day-ahead boundary forecasts, 24/48-hour-ahead weather forecasts for 21 cities, nodal-price signals, international fuel and carbon prices, typhoon proximity and a large-language-model news-sentiment signal. Its design has three distinctive features. First, every column is labelled with its real availability at bidding time, with as-of-bid columns kept separate from ex-post and settlement columns, so that an honest day-ahead forecast provably reads no future information. Second, the multi-source coverage spans supply and demand, constraints, cost and sentiment. Third, the pipeline is reproducible end to end, with per-source provenance, raw-response checksums, and an accompanying direction-prediction benchmark task with a neutral baseline. The dataset provides a time-aligned, caliber-transparent and reproducible empirical basis for spot-price-forecasting research in China.
🔗 Provenance — このレコードを発見したソース
- Zenodo https://zenodo.org/records/20698386first seen 2026-06-18 04:17:24 · last seen 2026-06-20 04:28:54
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。