| 胥鹏,张悦,王蓓蓓,朱红,刘少君,许洪华.基于深度强化学习的配电网在线拓扑优化策略研究[J].电力需求侧管理,2022,24(3):09-14 |
| 基于深度强化学习的配电网在线拓扑优化策略研究 |
| Research on online topology optimization strategy of distribution network based on deep reinforcement learning |
| 投稿时间：2022-01-20 修订日期：2022-03-28 |
| DOI：10. 3969 / j. issn. 1009-1831. 2022. 03 . 002 |
| 中文关键词: 在线重构 配电网 深度强化学习 动作机制 在线拓扑优化 |
| 英文关键词: online reconfiguration distribution network deep reinforcement learning action mechanism online topology optimization |
| 基金项目:国家电网有限公司科技项目“电力物联背景下主动配电网电压控制深度强化学习策略研究”（SGTYHT/19-JS-215） |
|
| 摘要点击次数: 1809 |
| 全文下载次数: 520 |
| 中文摘要: |
| 分布式发电（distributed generation，DG）在电力系统中的应用日益广泛，会频繁地导致在线电压问题。在分布式发电产生较大预测误差时，依靠有限的电压调节装置逐渐不能满足在线电压调节的要求。针对这一问题，提出了一种灵活的拓扑控制方法，并采用一种深度强化学习算法对其进行建模与求解。该算法中的动作机制与图论中的分支交换概念相融合，有效简化了智能体的动作维度和动作空间。在标准IEEE 14节点系统上的算例表明，该算法具备较优秀的泛化能力；与传统算法相比，该算法在获得接近最优解的同时，显著提高了计算性能，满足在线电压调节的需求。 |
| 英文摘要: |
| The increasing use of distributed generation（DG）in power systems can result in frequent online voltage problems.In case of large prediction error of DG, the requirement of online volt-age regulation can not be met by limited voltage regulation devices.In order to solve this problem, a flexible topology control method is proposed, and a deep reinforcement learning algorithm is used to model and solve it. The action mechanism in this algorithm combines the branch exchange concept in graph theory, and effectively simplifies the action dimension and action space. The analysis of the calculation examples on the standard IEEE 14 node system shows that the algorithm has excellent generalization ability. Compared with the previous methods, the proposed algorithm can obtain the closest optimal solution, improve the performance of the calculation significantly and meet the needs of online voltage regulation. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |