跳转至

文章背景与核心概要

极端天气和波动的批发电力市场给居民消费者带来了严重的财务风险。虽然需求响应（Demand-Response）项目可以通过在电价高峰期提供财务信贷来缓解这一问题，但通过强化学习来优化这一连续决策过程仍然十分困难。公开可用的离线数据无法捕捉到电力公司定价信号与客户采纳率之间的互动反馈循环。

为了解决这一痛点，作者推出了 DR-Gym，这是一个专为电力公司打造的开源、在线且兼容 Gymnasium 的环境。与设备级别的模拟器不同，DR-Gym 对市场级别的电力公司环境进行了建模，其核心特性包括：专为极端事件校准的区制转换批发价格模型、基于物理特性的建筑电力需求剖面，以及支持多样化训练目标的可配置多目标奖励函数。

Towards Affordable Energy: A Gymnasium Environment for Electric Utility Demand-Response Programs

arXiv: 2605.12462 [cs.AI]
Authors: Jose E. Aguilar Escamilla, Lingdong Zhou, Xiangqi Zhu, Huazheng Wang
Submitted: 12 May 2026 (Last revised 2 September 2026)

arXiv: 2605.12462 [cs.AI]
Authors: Jose E. Aguilar Escamilla, Lingdong Zhou, Xiangqi Zhu, Huazheng Wang
Submitted: 12 May 2026 (Last revised 2 September 2026)

📌 Summary

Extreme weather and volatile wholesale electricity markets put residential consumers at severe financial risk. While demand-response programs can help by offering financial credits during high-price periods, optimizing this sequential decision-making process via reinforcement learning remains difficult. Publicly available offline data fails to capture the interactive feedback loop between utility pricing signals and customer adoption.

To solve this, the authors introduce DR-Gym, an open-source, online Gymnasium-compatible environment built for electric utilities. Unlike device-level simulators, DR-Gym models the market-level utility setting, featuring: * A regime-switching wholesale price model calibrated for extreme events * Physics-based building demand profiles * A configurable, multi-objective reward function to support diverse training goals

📌 Summary

Extreme weather and volatile wholesale electricity markets put residential consumers at severe financial risk. While demand-response programs can help by offering financial credits during high-price periods, optimizing this sequential decision-making process via reinforcement learning remains difficult. Publicly available offline data fails to capture the interactive feedback loop between utility pricing signals and customer adoption.

To solve this, the authors introduce DR-Gym, an open-source, online Gymnasium-compatible environment built for electric utilities. Unlike device-level simulators, DR-Gym models the market-level utility setting, featuring: * A regime-switching wholesale price model calibrated for extreme events * Physics-based building demand profiles * A configurable, multi-objective reward function to support diverse training goals

📋 Paper Metadata

Field Details
Primary Subject Artificial Intelligence (cs.AI)
Cross-List Subjects Computers and Society (cs.CY), Computer Science and Game Theory (cs.GT), Machine Learning (cs.LG)
DOI 10.48550/arXiv.2605.12462
License Creative Commons Attribution 4.0 license icon

📋 Paper Metadata

Field Details
Primary Subject Artificial Intelligence (cs.AI)
Cross-List Subjects Computers and Society (cs.CY), Computer Science and Game Theory (cs.GT), Machine Learning (cs.LG)
DOI 10.48550/arXiv.2605.12462
License Creative Commons Attribution 4.0 license icon