半月刊

ISSN 1000-1026

CN 32-1180/TP

+高级检索 English
基于强化学习的多微电网分布式二次优化控制
作者:
作者单位:

1.南京理工大学自动化学院,江苏省南京市 210094;2.国网江苏省电力有限公司电力科学研究院,江苏省南京市 211103;3.国网浙江省电力有限公司嘉兴供电公司,浙江省嘉兴市 314000

作者简介:

通讯作者:

基金项目:

国家自然科学基金资助项目(51607036);江苏省自然科学基金资助项目(BK20160674)。


Reinforcement Learning Based Distributed Secondary Optimal Control for Multiple Microgrids
Author:
Affiliation:

1.School of Automation, Nanjing University of Science and Technology, Nanjing 210094, China;2.Electric Power Research Institute of State Grid Jiangsu Electric Power Co., Ltd., Nanjing 211103, China;3.Jiaxing Power Supply Company of State Grid Zhejiang Electric Power Co., Ltd., Jiaxing 314000, China

Fund Project:

This work is supported by National Natural Science Foundation of China (No. 51607036) and Jiangsu Provincial Natural Science Foundation of China (No. BK20160674).

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
    摘要:

    针对微电网中分布式电源下垂一次控制产生的系统频率和电压静态偏差问题,提出了一种基于强化学习就地反馈方法的分布式二次优化控制,利用本地信息可兼顾频率恢复和电压调整的需求。首先,针对微电网经济性、频率及电压控制需求和各分布式电源综合性能(环境效益、经济效益、技术效益),定义了本地奖励,协调多微电网的频率恢复和电压调节。其次,针对电网实际运行情况,在满足供需平衡的同时,使用多智能体强化学习算法对全局奖励反馈优化修正,使各分布式电源协同出力渐近消除频率偏差,保证微电网的稳定运行。最后,通过实例验证了所提出控制的有效性和适应性。

    Abstract:

    Aiming at the static deviation problems of system frequency and voltage caused by primary control of distributed generators in microgrid, a distributed secondary optimal control based on reinforcement learning local feedback method is proposed, which addresses the need of frequency recovery and voltage adjustment by using the local information. Firstly, according to the demand of microgrid economy, frequency and voltage control and the comprehensive performance of distributed generators (environmental benefit, economic benefit and technical benefit), a local reward is defined to coordinate the frequency recovery and voltage regulation of multiple microgrids. Secondly, according to the actual operation of the power grid, the multi-agent reinforcement learning algorithm is used to optimize and modify the global reward feedback, while satisfying the balance between supply and demand, so that the frequency deviation can be eliminated asymptotically and the stable operation of the microgrid can be guaranteed. Finally, the effectiveness and adaptability of the proposed control are verified by the analysis of case study.

    表 4 Table 4
    表 5 Table 5
    表 1 Table 1
    表 6 Table 6
    图1 分布式多智能体强化学习框架Fig.1 Framework of reinforcement leaning for distributed multi-agent
    图2 强化学习流程图Fig.2 Flow chart of reinforcement learning
    图3 仿真系统和通信连接Fig.3 Simulation system and communication connection
    图4 场景1相关参数变化Fig.4 Related parameter changing in scenario 1
    图5 场景2频率偏差变化Fig.5 Frequency deviation changing in scenario 2
    图 分布式电源并网的等效线路Fig. Equivalent circuit of the distributed generation connecting to grid
    图 引入虚拟阻抗后的等效电路Fig. Equivalent circuit with virtual impedance
    图 各DG的发出有功有名值变化(场景1)Fig. Active nominal values variation of DGs (scenario 1)
    图 各DG发出的有功有名值变化(场景2)Fig. Active nominal values variation of DGs (scenario 2)
    图 各DG发出无功有名值变化(场景3)Fig. Reactive power change of each DG (scenario 3)
    图 控制参数电压变化(场景3)Fig. Control parameter change of voltage (scenario 3)
    表 2 Table 2
    表 3 Table 3
    参考文献
    相似文献
    引证文献
引用本文

沈珺,柳伟,李虎成,等.基于强化学习的多微电网分布式二次优化控制[J].电力系统自动化. DOI:10.7500/AEPS20190521007.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2019-05-21
  • 最后修改日期:2020-01-21
  • 录用日期:2019-10-08
  • 在线发布日期:
  • 出版日期: