Aiming at the low-frequency oscillation problem frequently occurring in the weak grid connected with virtual synchronous generator (VSG), an intelligent control method for VSG is proposed, which integrates dynamic inertia-damping cooperative adjustment and multi-modal twin delayed deep deterministic policy gradient algorithm. An enhanced VSG model including dynamic inertia-damping adjustment mechanism is constructed. Based on the real-time monitoring of the standard deviation and change rate of frequency fluctuation, a continuous parameter adaptive algorithm is designed to realize the dynamic cooperative optimization of the inertia constant H and the damping coefficient D. The oscillation-aware deep deterministic policy gradient algorithm (TD3) is designed. The dual-state experience replay buffer structure is adopted to embed the low-frequency oscillation feature vector into the training samples, and a multi-objective reward function including frequency deviation penalty, voltage offset suppression and oscillation energy constraint is constructed. The simulation and actual case results show that the strategy can realize the online fast and accurate evaluation of VSG low-frequency oscillation, enhance the system damping and inertia, reduce the low-frequency oscillation risk, and improve the system stability.
VSG在有效改善系统惯量的同时,会引入类似SG的机电暂态过程[8],其虚拟功角动态易激发0.1~2.5 Hz频段的低频振荡[9]。电力系统稳定器(power system stabilizer,PSS)虽在传统同步机低频振荡抑制领域中具备扎实的理论与应用基础[10],但该技术基于相位补偿的线性化设计,难以适配高比例电力电子设备带来的强非线性、宽频域耦合特性,以及VSG本身参数动态适应性需求[11]。文献[12]基于Phillips-Heffron模型分析低频振荡机制,通过相位补偿法设计虚拟电力系统稳定器(virtual power system stabilizer,VPSS)抑制VSG低频振荡,但VPSS依赖线性化模型设计,难以适应高比例电力电子设备引入的强非线性特征。文献[13]提出基于自抗扰的VSG有功附加阻尼控制策略来抑制低频振荡,但其自抗扰参数需手动整定,对多运行工况的鲁棒性不足。文献[14]提出一种惯性自适应控制,根据频率变化动态调整惯性时间常数,抑制频率扰动引发的振荡,但控制后的高惯性会削弱系统阻尼特性,延长振荡衰减时间。
深度强化学习(deep reinforcement learning,DRL)凭借免模型特性和端到端优化决策能力,为电力系统控制提供了新思路[15]。DRL在微电网频率调节[16]、多源协同调频[17]等场景得到应用。然而,将DRL应用于VSG低频振荡抑制时,面临低概率-高风险事件捕捉困难和多时间尺度协同难两个关键挑战。传统DRL算法的经验回放机制多采用均匀随机采样,难以高效学习振荡这类暂态事件特征,严重影响策略性能。系统低频振荡事件发生概率较低,一旦发生,后果严重[18]。VSG控制涉及秒级机电暂态和微秒级电磁瞬态,现有DRL研究多聚焦单一时间尺度,尤其以机电暂态为主[19],缺乏能够有效整合并协调两种不同时间尺度动态信息的统一架构。文献[20]模型过度复杂化,高阶非线性模型包含12个状态变量,导致控制器实时计算负担过大。文献[21]提出结合离线强化学习的VSG参数自适应调整框架,通过混合优化算法(coyote optimization algorithm with self-adaptive differential evolution,COA-jDE)离线训练,动态优化虚拟惯量和阻尼系数,但离线训练需精确的电网模型,实际系统参数偏差可能导致控制失效。
针对上述问题,本文提出一种VSG智能控制策略,通过构建环境模型实现对下一时刻状态的预测,提升调节响应速度。引入基于动作-评价(actor-critic,AC)框架的算法,同时融合动态惯量-阻尼协同调节机制与多模态双延迟深度确定性策略梯度(twin delayed deep deterministic policy gradient,TD3)算法,实现VSG低频振荡的在线快速准确评估。此策略等效提高系统阻尼与惯量,减少低频振荡风险,同时改善系统的稳定性。
本文将LSTM的输出目标重映射为附加阻尼转矩增量,通过在线调整VSG 的 D 实现等效控制,替代原虚拟电阻指令;同时将SVG无功指令依据系统短路容量等效为阻尼转矩,与叠加后输入VSG阻尼系数整定模块。所有可调参数针对算例独立优化,包括EMD/KPCA超参数、LSTM结构、增益系数 及SVG控制器参数,采用网格搜索与敏感性分析完成优化过程。
FENGYuyao, FENGNan, XIONGXuejun, et al. Calculation method for dynamic reactive power reserve evaluation of AC/DC hybrid power grid based on adaptive dynamic dimension reduction[J]. Advanced Technology of Electrical Engineering and Energy, 2025, 44(5): 34-46.
CHENGQiyun, LIFeng, LUOShuxin, et al. Research on the planning methodology framework and key supporting technologies for new power systems[J]. Power System Technology, 2025, 49(6): 2219-2231.
LIJiaxu, WUJunyong, LILusu, et al. Review on frequency safety assessment and emergency control of power system[J]. High Voltage Engineering, 2025, 51(4): 1817-1833.
ZHANGXiang, YULing, ZHANGQiang, et al. Inertia adaptive control and secondary adjustment control of grid forming hybrid energy storage system considering SOC constraints[J]. Journal of Liaoning Technical University(Natural Science),2025,44(2):211-218.
[9]
CHENS M, SUNY, HOUX C, et al. Quantitative parameters design of VSG oriented to transient synchronization stability[J]. IEEE Transactions on Power Systems, 2023, 38(5): 4978-4981.
[10]
MAY Z, XUJ, GAOC X, et al. Low-frequency oscillations and resonance analysis of VSG-controlled PMSG-based wind generation systems[J]. Journal of Modern Power Systems and Clean Energy, 2025, 13(1): 115-127.
HEPing, WANGMingyang, WENFushuan, et al. Damping characteristic analysis and control of energy storage virtual synchronous generator based on virtual torque compensation[J]. Automation of Electric Power Systems, 2025, 49(11): 102-113.
WANGHaixin, LIUMingqi, DONGHenan, et al. Review on analysis and suppression of low-frequency oscillation in power system with high penetration of renewable energy sources[J]. Electric Power Automation Equipment, 2023, 43(9): 152-163.
[15]
QINB S, XUY H, YUANC, et al. A unified method of frequency oscillation characteristic analysis for multi-VSG grid-connected system[J]. IEEE Transactions on Power Delivery, 2022, 37(1): 279-289.
XIEXiaorong, LIHaozhi, ZHANGDan, et al. Preliminary study on stabilizer for new power systems[J]. Automation of Electric Power Systems, 2024, 48(23): 156-166.
HANYingduo, LUChao, SONGWenchao, et al. Review and prospects of online calculation and analysis techniques for power system security and stability[J]. Proceedings of the CSEE, 2024, 44(17): 6733-6761.
KONGF N, ZHANGH C, WUZ L, et al. Nonlinear adaptive disturbance rejection control strategy based on VSG for photovoltaic-storage grid-connected inverter[J]. IEEE Access, 2023, 11: 139534-139545.
[22]
WUH Y, DIQ, PEIX Y, et al. Analysis of low-frequency oscillation in dual virtual synchronous generators parallel system[C]//2024 IEEE International Conference on Energy Internet. November 1-3, 2024, Zhuhai, China. IEEE, 2025: 294-298.
LIPeikai, FUYuan, PANLianrong, et al. Deep reinforcement learning-based long and short-term coordinated scheduling for wind-solar-storage hybrid energy systems[J/OL]. Proceedings of the CSU-EPSA, 2025: 1-12[2025-06-18].
ZHAOWeiguang, ANBaijie, YANGYing, et al.Decentralized reactive power and voltage control strategy of distribution network with sources based on stability constraints for deep reinforcement learning[J/OL]. Southern Power System Technology,2025:1-11[2025-06-22].
YANGYue, WANGDan, HUBo, et al. Multi-source optimal joint frequency modulation control strategy based on improved multi-agent Q-learning [J]. Power System Protection and Control, 2022, 50(7): 135-144.
PENGZiran, HEZhenyu, XIAOShenping, et al. Electric vehicle brake energy recovery strategy based on deep reinforcement learning model TD3 optimization and improvement[J]. Control and Decision, 2025, 40(8): 2361-2372.
LIUXiaoming, LIUJun, YAOHongwei,et al. Deep reinforcement learning method for frequency regulation of wind-solar-hydro-thermal-storage systems based on virtual synchronous generator [J]. Automation of Electric Power Systems, 2025, 49(9): 114-124.
[33]
LIX L, FENGP, TIANZ, et al. Transient stability analysis for grid-tied VSG considering high-order nonlinear interactions between active and reactive power control loops[J]. IEEE Transactions on Power Electronics, 2024, 39(6): 6974-6988.
[34]
YAMEENM Z, LUZ G, EL-SOUSYF F M, et al. Improving frequency stability in grid-forming inverters with adaptive model predictive control and novel COA-jDE optimized reinforcement learning[J]. Scientific Reports, 2025, 15: 16540.
ZhipengLYU, SHENGWanxing, ZHONGQingchang, et al. Virtual synchronous generator and its applications in micro-grid[J]. Proceedings of the CSEE, 2014, 34(16): 2591-2603.
ZHANGXu, XUXin, DONGChengwu, et al. Sub-synchronous oscillation suppression method for DFIG wind power systems based on EMD-KPCA-LSTM and SVG damping control [J]. Journal of Electrical Engineering, 2025, 20(2): 54-67.
[39]
YANL Q, ZHAOZ, ULLAHZ, et al. Optimization of optical storage VSG control strategy considering active power deviation and virtual inertia damping parameters[J]. Ain Shams Engineering Journal, 2024, 15(12): 103131.