Scholarlink

<p>The Optimal Power Flow (OPF) problems of today’s interconnected power system in multi-region involve efficient and scalable solutions for the modulation of power flow. There are many centralized optimization methodologies proposed in the literature that have encountered some issues, such as scalability problems, slow convergence, and inability to adapt to the current grid status, meaning they cannot be used in real-time control. To tackle the above challenges, in this research, a Deep Reinforcement Learning (DRL)-Based Distributed Optimization Framework for multi-region power systems is proposed. The overall power flow control strategy is designed to utilize a multi-agent DRL scheme where each area is regarded as an autonomous agent attempting to control local power flow while being consistent with the overall result of all the agents via a reward function. The method reaches a 25% decrease in total generation costs, a 10% improvement in load matching, and voltage deviations less than 0.015 per unit, over 30 milliseconds per iteration. The approach offers convergence that is significantly faster than the convergence offered in conventional applications of the standard gradient. The results obtained from MATLAB, MATPOWER, and OpenAI Gym environments evidence the potential of this DRL approach for decentralized power system reconfiguration.</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

Deep Reinforcement Learning Based Distributed Optimization Framework for Optimal Power Flow in Multi-Region Power Systems

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Shuning Zhou

摘要

The Optimal Power Flow (OPF) problems of today’s interconnected power system in multi-region involve efficient and scalable solutions for the modulation of power flow. There are many centralized optimization methodologies proposed in the literature that have encountered some issues, such as scalability problems, slow convergence, and inability to adapt to the current grid status, meaning they cannot be used in real-time control. To tackle the above challenges, in this research, a Deep Reinforcement Learning (DRL)-Based Distributed Optimization Framework for multi-region power systems is proposed. The overall power flow control strategy is designed to utilize a multi-agent DRL scheme where each area is regarded as an autonomous agent attempting to control local power flow while being consistent with the overall result of all the agents via a reward function. The method reaches a 25% decrease in total generation costs, a 10% improvement in load matching, and voltage deviations less than 0.015 per unit, over 30 milliseconds per iteration. The approach offers convergence that is significantly faster than the convergence offered in conventional applications of the standard gradient. The results obtained from MATLAB, MATPOWER, and OpenAI Gym environments evidence the potential of this DRL approach for decentralized power system reconfiguration.