Scholarlink

In modern power systems, efficient energy scheduling is essential for ensuring system stability and optimizing resource allocation. At present, Deep Reinforcement Learning (DRL) algorithms perform well in continuous action spaces, but they face challenges such as high online learning costs and poor offline learning performance. To address these challenges, this paper proposes a Dynamic Dual-Strategy Update (DDSU) DRL that combines Twin Delayed Deep Deterministic Policy Gradient (TD3) and Generative Adversarial Imitation Learning (GAIL) technologies. By dynamically updating the agent’s strategy, DDSU enables parallel training of offline and online reinforcement learning, thereby enhancing the efficiency and accuracy of energy system scheduling. Experimental results show that DDSU significantly improves the convergence speed and final performance in the energy system scheduling task compared to the TD3 algorithm, in terms of reward acquisition, cost control, and system stability. These results demonstrate the potential of DDSU in enhancing the efficiency and accuracy of power system scheduling.

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 Dual-Strategy Update of Offline-to-Online Reinforcement Learning for Optimal Energy System Scheduling

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhiyao Lin,
  • AiHui Yang,
  • Li Liu,
  • Lin Li

摘要

In modern power systems, efficient energy scheduling is essential for ensuring system stability and optimizing resource allocation. At present, Deep Reinforcement Learning (DRL) algorithms perform well in continuous action spaces, but they face challenges such as high online learning costs and poor offline learning performance. To address these challenges, this paper proposes a Dynamic Dual-Strategy Update (DDSU) DRL that combines Twin Delayed Deep Deterministic Policy Gradient (TD3) and Generative Adversarial Imitation Learning (GAIL) technologies. By dynamically updating the agent’s strategy, DDSU enables parallel training of offline and online reinforcement learning, thereby enhancing the efficiency and accuracy of energy system scheduling. Experimental results show that DDSU significantly improves the convergence speed and final performance in the energy system scheduling task compared to the TD3 algorithm, in terms of reward acquisition, cost control, and system stability. These results demonstrate the potential of DDSU in enhancing the efficiency and accuracy of power system scheduling.