Knowledge-assisted reinforcement learning for risk aware coupled electricity and carbon market trading
知識支援型強化学習によるリスクを考慮した電力・炭素市場連動取引 (AI 翻訳)
Y Y Li, Yu Zhang, Xuanang Gui, Hongyun Zhao, Ruihua Yu, Jin Zhao
🤖 gxceed AI 要約
日本語
この論文は、電力市場と炭素市場の連動取引において、ディープ強化学習と知識支援手法、条件付き生成対抗ネットワークを組み合わせたリスク管理フレームワークを提案する。物理市場モデルやルールベースの保護機構によりドメイン知識を注入し、危険回避型の学習アルゴリズムとリスク計測手法を開発。シミュレーションにより、高い収益性を維持しつつリスクを低減できることを示した。
English
This paper proposes a risk-aware coupled electricity-carbon market trading framework combining safe deep reinforcement learning, knowledge assistance, and Conditional GAN. Domain knowledge from physical market models and rule-based protection is injected. A new algorithm, Knowledge-Assisted Risk-Estimation PPO, and a GAN-based risk measurement method are developed. Experiments show the method maintains high profit while explicitly reducing risk.
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
日本では2023年度から試行的な排出量取引が始まり、今後本格的なカーボンプライシング導入が検討されている。本研究の知見は、電力事業者やトレーディング部門が排出権取引リスクを管理する際に有用である。
In the global GX context
Globally, coupled electricity and carbon markets are emerging as key mechanisms for decarbonization. This work advances reinforcement learning applications for risk management in these markets, relevant for traders and utilities in regions with active carbon markets like the EU ETS and China's national ETS.
👥 読者別の含意
🔬研究者:Offers AI method integrating domain knowledge for risk-sensitive trading in carbon markets.
🏢実務担当者:Provides a framework for electricity and carbon market participants to manage trading risks while maintaining profitability.
🏛政策担当者:Illustrates how AI can enhance market efficiency and risk management in carbon pricing systems.
📄 Abstract（原文）
The coupled electricity–carbon market trading leverages the temporal coupling of dispatch and the long-horizon compliance obligation to improve profitability and sustainability. However, the pronounced uncertainty stemming from market fundamentals and opponents’ behaviours makes it difficult to secure a profit under acceptable risk. To address this challenge, this paper proposes a knowledge-assisted risk-sensitive coupled electricity–carbon market trading framework, combining safe deep reinforcement learning, knowledge-assisted method and the Conditional Generative Adversarial Network. First, the knowledge-assisted module is applied, which injects domain priors via a physical market model, an evaluation model, and a rule-based protector. In addition, the Knowledge-Assisted Risk-Estimation Proximal Policy Optimisation algorithm is developed to improve the learning efficiency in the absence of direct risk signals. At last, a Conditional Generative Adversarial Network-based risk measurement method is proposed to provide the risk signal for the agent, with additional regularisation to avoid local convergence caused by inaccurate tail estimation. The proposed approach is tested on a MATPOWER-based coupled electricity–carbon market system with thermal and renewable units. Results show that the method maintains high profit while explicitly reducing risk and maintaining stable training compared with representative baselines.
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.1049/icp.2026.1611first seen 2026-07-13 05:34:47
🔔 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。