Scholarlink

The smart grid integrates distributed energy and energy storage technologies, but energy fluctuations and privacy requirements increase the difficulty of optimization. This paper proposes a smart grid edge computing energy storage control method (FDRL) based on deep reinforcement learning and federated learning. It collects and processes multidimensional data from edge nodes to generate high-quality inputs suitable for model training. Deep reinforcement learning is used to optimize local energy storage strategies, enhancing the autonomous control capabilities of nodes. Federated learning aggregates local models to achieve global collaborative optimization and privacy protection, dynamically deploying the global optimization model to control energy storage devices in real time, adapting to complex environmental changes. Experimental results demonstrate that FDRL significantly outperforms comparison methods in terms of energy efficiency, response speed, load demand satisfaction rate, and energy storage device utilization, verifying its technological advantages in distributed energy storage control for smart grids.

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

A Smart Grid Edge Computing Energy Storage Control Method Based on Deep Reinforcement Learning and Federated Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yang Liu,
  • Qiuyu Lu,
  • Wei Zhong,
  • Yue Chen,
  • Weihua Wang,
  • Pinping Xie

摘要

The smart grid integrates distributed energy and energy storage technologies, but energy fluctuations and privacy requirements increase the difficulty of optimization. This paper proposes a smart grid edge computing energy storage control method (FDRL) based on deep reinforcement learning and federated learning. It collects and processes multidimensional data from edge nodes to generate high-quality inputs suitable for model training. Deep reinforcement learning is used to optimize local energy storage strategies, enhancing the autonomous control capabilities of nodes. Federated learning aggregates local models to achieve global collaborative optimization and privacy protection, dynamically deploying the global optimization model to control energy storage devices in real time, adapting to complex environmental changes. Experimental results demonstrate that FDRL significantly outperforms comparison methods in terms of energy efficiency, response speed, load demand satisfaction rate, and energy storage device utilization, verifying its technological advantages in distributed energy storage control for smart grids.