双碳背景下GCN和DRL的含高比例可再生能源消纳约束的主动配电网经济调度策略

雍少华; 马耀东; 蒋斌文; 陈雨; 秦英

雍少华,马耀东,蒋斌文,陈雨,秦英.双碳背景下GCN和DRL的含高比例可再生能源消纳约束的主动配电网经济调度策略[J].电测与仪表,2026,63(8):36-43.
yongshaohua,mayaodong,jiangbinwen,chenyu,qinying.Economic dispatching strategy of active distribution network with high proportion of renewable energy consumption constraints for GCN and DRL under dual carbon background[J].Electrical Measurement & Instrumentation,2026,63(8):36-43.

双碳背景下GCN和DRL的含高比例可再生能源消纳约束的主动配电网经济调度策略

Economic dispatching strategy of active distribution network with high proportion of renewable energy consumption constraints for GCN and DRL under dual carbon background

DOI：10.19753 / j.issn1001-1390.2026.08.004

中文关键词: 主动配电网经济调度深度强化学习 GCN DDPG

英文关键词:active distribution network, economic dispatch,deep reinforcement learning, GCN, DDPG

基金项目:国家自然科学基金资助项目(92267104); 国网宁夏电力有限公司科技项目(B329ZW230002)

作者中文名	作者英文名	单位
雍少华	yongshaohua	国网宁夏电力有限公司中卫供电公司
马耀东	mayaodong	国网宁夏电力有限公司中卫供电公司
蒋斌文	jiangbinwen	国网宁夏电力有限公司中卫供电公司
陈雨	chenyu	国网宁夏电力有限公司中卫供电公司
秦英	qinying	国网宁夏电力有限公司中卫供电公司

摘要点击次数: 8

中文摘要:

为了实现双碳目标，提高主动配电网（active distribution network, ADN）中可再生资源的消纳率，增强配电系统调度的经济性，文中提出一种融合图卷积网络（graph convolutional network, GCN）和深度确定性策略梯度算法（deep deterministic policygradient, DDPG）的GCN-DDPG策略。该策略首先以ADN日运行成本最小为目标，构建光伏、负荷、能量存储系统的经济调度模型；其次，将ADN调度问题描述为深度强化学习（deep reinforcement learning, DRL）中的马尔可夫决策（markov decision process, MDP）过程，并定义系统的状态空间、动作空间以及奖励函数；进一步的，在DRL中应用DDPG算法，同时为了改善DDPG算法采用数据驱动，忽略ADN拓扑结构的缺陷，将GCN融合其中，提高DRL智能体对配电网图数据的表征能力；最后，在改进的IEEE 33仿真系统中，验证了该模型的有效性和鲁棒性。

英文摘要:

To achieve the dual carbon goals, improve the integration of renewable resources in Active Distribution Networks (ADN) and enhance the economic efficiency of distribution system scheduling, this paper introduces a GCN-DDPG strategy that combines Graph Convolutional Networks (GCN) with Deep Deterministic Policy Gradient (DDPG) algorithms. The strategy aims to minimize the ADN""s daily operating costs by developing aneconomicdispatch model for photovoltaic systems, loads, and energy storage systems. Subsequently, the ADN scheduling problem is discribed as a Markov Decision Process (MDP) within the framework of Deep Reinforcement Learning (DRL), defining the state space, action space, and reward function of the system. Additionally, it applies the DDPG algorithm in DRL, To address the limitations of DDPG, namely its data-driven nature and disregard for ADN topological structure, it integrates GCN to improve the DRL agent""s ability to represent and process graph data of the distribution network. Finally, the proposed model""s effectiveness and robustness are validated through simulations using an enhanced IEEE 33 benchmark system.

查看全文查看/发表评论下载PDF阅读器

关闭