储能科学与技术 ›› 2024, Vol. 13 ›› Issue (3): 858-869.doi: 10.19799/j.cnki.2095-4239.2023.0658

• 储能系统与工程 • 上一篇    下一篇

基于强化学习的新能源场站储能一次调频自适应控制策略

孙冉1(), 王建波1, 马彦钊2, 张小科3, 胡怀中2()   

  1. 1.国网河南省电力公司,河南 郑州 450000
    2.西安交通大学自动化科学与工程学院,陕西 西安 710100
    3.国网河南省电力公司电力科学研究院,河南 郑州 450052
  • 收稿日期:2023-09-26 修回日期:2023-11-23 出版日期:2024-03-28 发布日期:2024-03-28
  • 通讯作者: 胡怀中 E-mail:persiasun@126.com;huhuaizhong@xjtu.edu.cn
  • 作者简介:孙冉(1979—),女,硕士,高级工程师,主要从事电力系统稳定分析与控制研究,E-mail:persiasun@126.com
  • 基金资助:
    国网河南省电力公司科技项目(521702230011);国家重点研发计划资助项目(2018YFB1700104)

Adaptive control strategy for primary frequency regulation for new energy storage stations based on reinforcement learning

Ran SUN1(), Jianbo WANG1, Yanzhao MA2, Xiaoke ZHANG3, Huaizhong HU2()   

  1. 1.State Grid Henan Electric Power Company, Zhengzhou 450000, Henan, China
    2.School of Automation Science and Engineering, Xi'an 710100, Shaanxi, China
    3.Electric Power Research Institute of State Grid Henan Electric Power Company, Zhengzhou 450052, Henan, China
  • Received:2023-09-26 Revised:2023-11-23 Online:2024-03-28 Published:2024-03-28
  • Contact: Huaizhong HU E-mail:persiasun@126.com;huhuaizhong@xjtu.edu.cn

摘要:

针对当前储能参与一次调频时不同控制策略配合存在的缺陷,在传统虚拟惯性和虚拟下垂控制方法的基础上,考虑新能源出力特征,提出一种基于强化学习算法的新能源场站储能一次调频自适应控制策略。所提策略中,强化学习智能体负责根据系统频差和频差变化率实时波动来动态调整储能通过虚拟惯性控制方法参与一次调频的出力占比,进而由储能一次调频自适应控制器计算出虚拟下垂控制方法的出力占比,并获取储能总一次调频出力指令。考虑到新能源场站有功扰动的随机性,本工作中的强化学习智能体通过在特定新能源场站出力扰动下学习获取,其中新能源场站出力扰动由风速扰动通过风电机组模型获取。所提控制策略能充分发挥虚拟惯性和虚拟下垂控制方法在调频前后期的不同优势,实现两种控制方法的最优结合,挖掘储能参与一次调频的潜力。最终在Matlab/Simulink中搭建了区域电网频率响应模型,基于新能源发电突变和新能源发电连续波动两种扰动工况来模拟新能源场站极端工况和实际运行场景,并通过仿真验证了所提控制策略的有效性。结果表明,本工作所提策略能在调频过程中合理调整虚拟惯性出力和虚拟下垂出力的占比,减少电池储能的动作深度,有效缓解新能源出力波动给电网带来的频率波动,提升频率质量。

关键词: 电池储能, 一次调频, 新能源场站, 强化学习, 自适应控制

Abstract:

Based on traditional virtual inertia and virtual sag control methods, and considering the characteristics of new energy output, we propose a new energy storage adaptive control strategy for primary frequency regulation using a reinforcement learning algorithm. Herein, reinforcement learning agent dynamically adjusts the output proportion of the energy storage participating in primary frequency regulation through the virtual inertia control method, responding to real-time fluctuations in system frequency and frequency difference change rate. Subsequently, energy storage primary frequency regulation adaptive controller calculates the output proportion of the virtual sag control method, thereby obtaining the total primary frequency regulation output command for the energy storage. Considering the inherent randomness of active power output in new energy stations, reinforcement learning agent learns from the output disturbances, which are influenced by wind speed disturbances through the wind turbine model in the new energy station. The proposed control strategy maximizes the distinct advantages of virtual inertia and virtual sag control methods before and after frequency regulation. It achieves an optimal combination of the two control methods, unlocking the full potential of energy storage in participating in primary frequency regulation. To validate the effectiveness of the proposed strategy, we constructed a regional power grid frequency response model in Matlab/Simulink. This model simulated extreme working conditions and actual operational scenarios of the new energy station. The simulation results confirm that the strategy can appropriately adjust the proportion of virtual inertial output and virtual sagging output during the frequency modulation process. This adjustment reduces the operation depth of battery energy storage, effectively mitigates frequency fluctuation caused by variations in new energy output to the power grid, and enhances overall frequency quality.

Key words: battery energy storage, primary frequency regulation, new energy station, reinforcement learning, adaptive control

中图分类号: