Scholarlink

Green Edge Cloud Computing (GECC) represents an emerging technology that integrates environmental sustainability and energy efficiency into edge computing architectures. In this domain, Hierarchical Federated Learning (HierFl) demonstrates extensive applicability by effectively reducing transmission energy consumption in conventional federated learning through its three-tier cloud-edge-end architecture. The energy consumption of systems operating under the HierFl framework exhibits a direct correlation with the scale of training data volume. Consequently, balancing the dual objectives of minimizing system energy consumption while enhancing model accuracy presents a significant research challenge. To address this challenge, this paper proposes a hierarchical federated learning framework based on deep reinforcement learning. Specifically, we employ Deep Q-Network (DQN) to optimize the training data allocation for individual client devices. The experimental results show that our method improves model performance by more than 15%.

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

Low Energy Consumption Hierarchical Federated Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shengaocheng Zhang,
  • Jigang Wen,
  • Xiaofan Zhou,
  • Kun Xie,
  • Yinchuan Cong,
  • Xiaoyan Chen,
  • Tianxiong Liu

摘要

Green Edge Cloud Computing (GECC) represents an emerging technology that integrates environmental sustainability and energy efficiency into edge computing architectures. In this domain, Hierarchical Federated Learning (HierFl) demonstrates extensive applicability by effectively reducing transmission energy consumption in conventional federated learning through its three-tier cloud-edge-end architecture. The energy consumption of systems operating under the HierFl framework exhibits a direct correlation with the scale of training data volume. Consequently, balancing the dual objectives of minimizing system energy consumption while enhancing model accuracy presents a significant research challenge. To address this challenge, this paper proposes a hierarchical federated learning framework based on deep reinforcement learning. Specifically, we employ Deep Q-Network (DQN) to optimize the training data allocation for individual client devices. The experimental results show that our method improves model performance by more than 15%.