Scholarlink

<p>Under the framework of smart grid development, electric power communication networks face challenges in real-time operations, transmission reliability, and intelligent management. This paper proposes a self-convergent optimization control strategy, RiskQuant-GRL (RQGRL), which integrates risk quantization and graph deep reinforcement learning. By combining Graph Neural Networks (GNN) with Deep Q-Learning (DQN), RQGRL aims to enhance network intelligence in resource scheduling and path planning. Specifically, it embeds a Graph Convolutional Network (GCN) into a Deep Reinforcement Learning (DRL) framework to monitor the network state, make intelligent decisions, and provide optimal routing. By quantifying risk as edge weights and using weighted GCN, the model dynamically optimizes path selection and resource allocation. Simulation results show that RQGRL improves bandwidth allocation efficiency by 6.6% over state-of-the-art DRL solutions and excels in link failure recovery, maintaining efficient network operation even under high concurrency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Riskquant-grl: a self-optimal control method for power network communication based on graph deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月14日

查看全文
  • Yingjie Jiang,
  • Yongjing Wei,
  • Chao Sun,
  • Anqi Tian,
  • Yan Zhang,
  • Youxiang Zhu

摘要

Under the framework of smart grid development, electric power communication networks face challenges in real-time operations, transmission reliability, and intelligent management. This paper proposes a self-convergent optimization control strategy, RiskQuant-GRL (RQGRL), which integrates risk quantization and graph deep reinforcement learning. By combining Graph Neural Networks (GNN) with Deep Q-Learning (DQN), RQGRL aims to enhance network intelligence in resource scheduling and path planning. Specifically, it embeds a Graph Convolutional Network (GCN) into a Deep Reinforcement Learning (DRL) framework to monitor the network state, make intelligent decisions, and provide optimal routing. By quantifying risk as edge weights and using weighted GCN, the model dynamically optimizes path selection and resource allocation. Simulation results show that RQGRL improves bandwidth allocation efficiency by 6.6% over state-of-the-art DRL solutions and excels in link failure recovery, maintaining efficient network operation even under high concurrency.