| 引用本文: | 柳伟,胡添欢,汤程烨.基于深度强化学习的孤立微电网分布式牵制协同控制[J].控制理论与应用,2026,43(2):325~334.[点击复制] |
| LIU Wei,HU Tian-huan,TANG Cheng-ye.Distributed pinning collaborative control of islanded microgrids based on deep reinforcement learning[J].Control Theory & Applications,2026,43(2):325~334.[点击复制] |
|
| 基于深度强化学习的孤立微电网分布式牵制协同控制 |
| Distributed pinning collaborative control of islanded microgrids based on deep reinforcement learning |
| 摘要点击 130 全文点击 18 投稿时间：2023-11-23 修订日期：2025-09-20 |
| 查看全文 查看/发表评论 下载PDF阅读器 HTML |
| DOI编号 10.7641/CTA.2024.30762 |
| 2026,43(2):325-334 |
| 中文关键词 孤立微电网 二次控制 深度强化学习 牵制控制 分布式协同控制 |
| 英文关键词 islanded microgrids secondary control deep reinforcement learning pinning control distributed collaborative control |
| 基金项目 国家自然科学基金项目(52077103)资助. |
|
| 中文摘要 |
| 微电网能够组合运用源网荷储协调弥补分布式电源(DGs)出力不确定性的不足, 并在孤岛状态下最大程度
保障用户的供电可靠性. 以孤立微电网为研究对象, 为解决一次下垂控制引起的电压频率偏差问题, 提出了一种基
于深度强化学习的孤立微电网分布式牵制协同控制策略, 能够有效抑制负载扰动或系统拓扑结构变化时频率和电
压的波动. 首先, 针对牵制一致性算法中牵制目标值难以准确预设和改变的问题, 采用深度强化学习算法中双重深
度Q学习(DDQN)算法对牵制目标值进行自适应修正, 利用DDQN的适应性以及泛化能力, 提升牵制一致性算法的
适应性和有效性; 其次, 设计了基于DDQN的牵制控制器结构, 并完成了状态空间、动作空间和奖励函数的定义; 最
后, 仿真结果表明, 所提控制方法与传统的牵制控制和基于强化学习的牵制控制相比, 在抑制频率电压波动方面更
具优势, 并且能更好的适应系统网络拓扑参数和结构的变化. |
| 英文摘要 |
| Microgrids can comprehensively utilize various energy sources and coordinate energy storage systems to
compensate for the uncertainties in the output of distributed generators (DGs). a distributed pinning collaborative control
strategy based on deep reinforcement learning for isolated microgrids is proposed to solve the problem of voltage and
frequency deviation caused by droop control in isolated microgrids, which can effectively suppress frequency and voltage
fluctuations under load disturbance or system topology changes. Firstly, aiming at the problem that the pinning target value
in the pinning consistency algorithm is difficult to accurately preset and change, the double deep Q learning (DDQN) is
used to adaptively modify the pinning target value by leveraging the adaptability and generalization capability of DDQN.
It improves the adaptability and effectiveness of the pinning consistency algorithm. Furthermore, the structure of the
distributed pinning collaborative controller based on DDQN is designed, and the definition of state space, action space,
and reward function is completed. Finally, Simulation results demonstrate that the proposed control method outperforms
traditional pinning control and reinforcement learning-based pinning control in responding to changes in network topology
parameters and structure. |
|
|
|
|
|