基于深度强化学习的电力系统紧急切机稳控策略生成方法  

Policy generation method for power system stability control during emergent tripping of unit based on deep reinforcement learning

作　　者：高琴 徐光虎[1] 夏尚学[2] 杨欢欢 赵青春[2] 黄河[1] GAO Qin;XU Guanghu;XIA Shangxue;YANG Huanhuan;ZHAO Qingchun;HUANG He(China Southern Power Grid Co.,Ltd.,Guangzhou 510663,China;NR Electric Co.,Ltd.,Nanjing 211102,China)

机构地区：[1]中国南方电网有限责任公司,广东广州510663 [2]南京南瑞继保电气有限公司,江苏南京211102

出　　处：《电力科学与技术学报》2025年第1期39-46,共8页Journal of Electric Power Science And Technology

基　　金：中国南方电网有限责任公司科技项目(000005KK52220027)。

摘　　要：电力系统快速发展的同时也改变着电力系统的结构,使得系统稳定机理变得更加复杂。为解决新能源电力系统存在的功角稳定问题,提出基于深度强化学习的电力系统紧急切机稳控策略生成方法。首先,归纳并提出电力系统紧急控制切机动作策略以及涉及的安全约束,并将电力系统稳控模型转换为马尔科夫决策过程,再采用特征评估与斯皮尔曼(Spearman)等级相关系数方法筛选出最典型的特征数据;随后,为提高稳控策略智能体的训练效率,提出基于深度确定性策略梯度(deep deterministic policy gradient,DDPG)算法的稳控策略训练框架;最后,在IEEE 39节点系统和某实际电网中进行测试验证。研究结果显示,所提方法能够根据系统的运行状态和对故障的响应,自动调整生成切机稳控策略,在决策效果和效率方面都表现出更好的性能。The rapid development of the power system has been changing its structure,making the system stability mechanism more complex.To ensure power angle stability in the new energy power system,a policy generation method for power system stability control during emergent tripping of units based on deep reinforcement learning is proposed.Firstly,the policies for emergent tripping of units of the power system are summarized,as well as the security constraints involved.The power system stability control model is then transformed into a Markov decision process.Next,the most typical feature data are selected by feature evaluation and the Spearman rank correlation coefficient method.To improve the training efficiency of the intelligent agent of the stability control policy,a training framework for the stability control policy based on the deep deterministic policy gradient(DDPG)is put forward.Finally,tests are performed in the IEEE 39 node system and a real-life power grid for validation.The results show that the proposed method can automatically adjust and generate a stability control policy for tripping of units according to the system’s running states and fault responses,confirming its enhanced decision-making effect and efficiency.

关 键 词：新能源电力系统 稳控策略 强化学习 深度确定性策略梯度算法 马尔科夫模型 

分 类 号：TM723[电气工程—电力系统及自动化]