Wide-Area Measurement System-Based Low Frequency Oscillation Damping Control Through Reinforcement Learning
Wide-Area Measurement System-Based Low Frequency Oscillation Damping Control Through Reinforcement Learning
复制标题
DOI：
10.1109/tsg.2020.3008364
复制
发表时间：
2020-01
影响因子：
9.6
通讯作者：
Yousuf Hashmy;Zhe Yu;Di Shi;Yang Weng
中科院分区：
文献类型：
--
作者：
Yousuf Hashmy;Zhe Yu;Di Shi;Yang Weng
由于负荷不确定性的快速增长和可再生能源渗透率的增加，确保电力系统的稳定性比以往任何时候都受到更多的关注。近年来，基于广域测量系统（WAMS）的集中控制技术提供了灵活性和更强大的控制，以保持系统的稳定。然而，基于WAMS的控制技术面临着长距离通信信道中的不规则延迟和设备对控制动作的后续响应的紧迫挑战。本文提出了一种新颖的控制策略，用于阻尼输电系统中的低频振荡。该方法使用强化学习技术来克服广域阻尼控制中的通信延迟和其他非线性的挑战。它将传统的振荡阻尼控制问题建模为一种新的基于更快探索的深度确定性策略梯度（DDPG-S）。一个有效的奖励函数的目的是捕捉必要的功能，使这种振荡及时阻尼振荡，即使在各种不确定性。详细的分析和系统设计的数值验证，证明了可行性，可扩展性，可解释性，和比较性能的建模低频振荡阻尼控制器。该技术的好处是，即使在负荷和发电的不确定性上升时，也能确保稳定性。
Ensuring the stability of power systems is gaining more attention today than ever before due to the rapid growth of uncertainties in load and increased renewable energy penetration. Lately, wide-area measurement system (WAMS)-based centralized controlling techniques are offering flexibility and more robust control to keep the system stable. WAMS-based controlling techniques, however, face pressing challenges of irregular delays in long-distance communication channels and subsequent responses of equipment to control actions. This paper presents an innovative control strategy for damping down low-frequency oscillations in transmission systems. The method uses a reinforcement learning technique to overcome the challenges of communication delays and other non-linearity in wide-area damping control. It models the traditional problem of oscillation damping control as a novel faster exploration-based deep deterministic policy gradient (DDPG-S). An effective reward function is designed to capture necessary features of oscillations enabling timely damping of such oscillations, even under various kinds of uncertainties. A detailed analysis and a systematically designed numerical validation are presented to prove feasibility, scalability, interpretability, and comparative performance of the modelled low-frequency oscillation damping controller. The benefit of the technique is that stability is ensured even when uncertainties of load and generation are on the rise.