Scholarlink

<p>Reconfiguring conventional distribution systems into islanded networked microgrids enhances system resilience against severe power outages. However, maintaining stable operation during islanded conditions requires rapid and intelligent reconfiguration under uncertain operating environments. Key challenges include frequency regulation, distributed resource coordination, and real-time decision-making. To address these challenges, this paper proposes a dynamic coordination framework based on Double Deep Q-Networks integrated with Convolutional Neural Networks. The proposed approach explicitly models frequency variations and employs an exponential epsilon-greedy strategy to improve learning efficiency and convergence stability. A modified backward/forward sweep algorithm is used for islanded power flow analysis, while a graph-based Breadth-First Search method enables feasible topology reconfiguration. Extensive simulations demonstrate that the proposed Exponential Epsilon DDQN achieves up to 93% improvement in mean Q-values compared with baseline deep reinforcement learning methods, while also providing faster convergence and improved training stability. Compared with conventional reconfiguration techniques, the proposed method achieves up to 20% higher restoration reward and approximately 750 times faster decision-making than genetic algorithm and mixed-integer linear programming benchmarks, completing reconfiguration decisions within 0.034&#xa0;s. Sensitivity analysis further validates the robustness of the framework, with the proposed method outperforming both Artificial Neural Network and Long Short-Term Memory architectures. The results demonstrate the effectiveness and practical applicability of the proposed framework for enhancing the resilience and operational reliability of islanded networked microgrids.</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

Dynamic coordination of islanded networked microgrids for power systems resilience: a double deep Q-network learning approach

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Mohammadjavad Bordbari,
  • Fuzhan Nasiri

摘要

Reconfiguring conventional distribution systems into islanded networked microgrids enhances system resilience against severe power outages. However, maintaining stable operation during islanded conditions requires rapid and intelligent reconfiguration under uncertain operating environments. Key challenges include frequency regulation, distributed resource coordination, and real-time decision-making. To address these challenges, this paper proposes a dynamic coordination framework based on Double Deep Q-Networks integrated with Convolutional Neural Networks. The proposed approach explicitly models frequency variations and employs an exponential epsilon-greedy strategy to improve learning efficiency and convergence stability. A modified backward/forward sweep algorithm is used for islanded power flow analysis, while a graph-based Breadth-First Search method enables feasible topology reconfiguration. Extensive simulations demonstrate that the proposed Exponential Epsilon DDQN achieves up to 93% improvement in mean Q-values compared with baseline deep reinforcement learning methods, while also providing faster convergence and improved training stability. Compared with conventional reconfiguration techniques, the proposed method achieves up to 20% higher restoration reward and approximately 750 times faster decision-making than genetic algorithm and mixed-integer linear programming benchmarks, completing reconfiguration decisions within 0.034 s. Sensitivity analysis further validates the robustness of the framework, with the proposed method outperforming both Artificial Neural Network and Long Short-Term Memory architectures. The results demonstrate the effectiveness and practical applicability of the proposed framework for enhancing the resilience and operational reliability of islanded networked microgrids.