| 引用本文: | 赵黎媛,张鹏举,赵志刚,李幕松,杨竣博.基于深度强化学习的考虑燃气动态掺氢的多能微网低碳经济调度[J].电力系统保护与控制,2026,54(01):168-178.[点击复制] |
| ZHAO Liyuan,ZHANG Pengju,ZHAO Zhigang,LI Musong,YANG Junbo.Low-carbon economic dispatch of multi-energy microgrids considering dynamic hydrogen blending in gas based on reinforcement learning[J].Power System Protection and Control,2026,54(01):168-178[点击复制] |
|
| 摘要: |
| 针对多能微网运行过程中面临的氢能利用单一、源荷不确定性等挑战，提出了一种基于深度强化学习的考虑燃气动态掺氢的多能微网低碳经济调度方法。首先，构建燃气掺氢比及热电比动态可调的氢能多元化利用结构，提升系统运行灵活性。然后，建立考虑阶梯型碳交易机制的多能微网低碳经济调度模型，将其转化为随机环境下的强化学习框架，设计基于超限截断保障下的动作空间，确保智能体动作在安全范围内。最后，采用双延迟深度确定性策略梯度算法对调度问题进行求解，实现多能耦合约束下智能体对不确定因素的自适应响应。算例结果表明，所提方法能够有效处理源荷不确定因素，降低系统运行成本及碳排放水平。 |
| 关键词: 多能微网 燃气动态掺氢 氢能多元利用 深度强化学习 低碳经济调度 |
| DOI：10.19783/j.cnki.pspc.250374 |
| 投稿时间：2025-04-08修订日期：2025-09-17 |
| 基金项目:河北省自然科学基金项目资助(E2025202270，F2024202005)；天津市教委科研计划项目资助(2022KJ088)；河北省教育厅科学研究项目资助(BJK2024151)；天津市自然科学基金项目资助(23JCQNJC01060) |
|
| Low-carbon economic dispatch of multi-energy microgrids considering dynamic hydrogen blending in gas based on reinforcement learning |
| ZHAO Liyuan,ZHANG Pengju,ZHAO Zhigang,LI Musong,YANG Junbo |
| (State Key Laboratory of Intelligent Power Distribution Equipment and System,
Hebei University of Technology, Tianjin 300401, China) |
| Abstract: |
| To address challenges in the operation of multi-energy microgrids, such as the single hydrogen energy utilization mode and uncertainties in supply and demand, this paper proposes a deep reinforcement learning-based low-carbon economic dispatch method for multi-energy microgrids considering dynamic hydrogen blending in gas. First, a diversified hydrogen utilization model with dynamically adjustable thermoelectric ratio and hydrogen blending ratio is constructed to enhance system operational flexibility. Next, a low-carbon economic dispatch model of the multi-energy microgrid incorporating a tiered carbon trading mechanism is formulated and transformed into a reinforcement learning framework under a stochastic environment. An action space with safety guarantees based on limit truncation is designed to ensure that agent actions remain within safe operational ranges. Finally, the dispatch problem is solved using the twin delayed deep deterministic policy gradient algorithm, realizing adaptive response of the agent to uncertainties under multi-energy coupling constraints. Case study results demonstrate that the proposed method can effectively address uncertainties in supply and demand, significantly reducing both operational costs and carbon emissions. |
| Key words: multi-energy microgrid dynamic hydrogen blending in gas diversified hydrogen utilization deep reinforcement learning low-carbon economy dispatch |
