军事知识服务平台
基于深度强化学习的储能系统虚拟惯量控制
Deep Reinforcement Learning Based Optimal Virtual Inertia Control of Energy Storage Systems
摘要 在碳达峰碳中和的目标下,大量可再生能源接入电力系统替代传统同步发电机.然而,新能源带来的不确定性和低惯性的特点导致电力系统面临频率波动和振荡等挑战.储能系统结合逆变器技术可提供快速的频率响应,其中虚拟惯性控制能有效减少频率偏差及频率变化率.针对储能系统的虚拟惯性控制往往需要考虑控制成本设计最优控制器,但大部分现有最优虚拟惯性控制方法依赖线性化模型,难以处理复杂非线性系统.为此,提出基于李雅普诺夫的强化学习方案,将虚拟惯性控制建模为马尔可夫决策过程,利用李雅普诺夫理论设计控制器,确保闭环系统局部稳定性,并采用深度确定性策略算法优化控制器参数.仿真结果表明,所提方法显著提升了频率控制性能.
Abstract:
Under the objective of achieving carbon peak and neutrality,a substantial amount of renewable energy sources have been integrated into the power system to replace traditional synchronous generators.However,their inherent uncertainty and low characteristics inertia pose challenges,including frequency fluctuations and oscillations in power systems.Benefiting from the recent advancement in power electronic technology,energy storage systems are capable of providing rapid frequency response.Among these,virtual inertia control effectively mitigates frequency deviations and rate of change of frequency.The implementation of virtual inertia control for energy storage systems usually needs to consider control cost and design optimal controller.Nevertheless,most of existing methods rely on linearized models,which fail to adequately address complex nonlinear dynamics.To this end,a Lyapunov-based reinforcement learning scheme is proposed in this paper,formulating virtual inertia control as a Markov decision process and designing the controller to ensure local stability via Lyapunov theory.The deep deterministic policy gradient algorithm optimizes controller parameters.Simulation results indicate that the proposed method significantly enhances frequency control performance.
刊 名 山东电力技术 ISTIC
年,卷(期): 2026, 53(9)
分类号: TK02
机标分类号: TM732 TP391 TM614
在线出版日期: 2026年10月14日