Scholarlink

<p>Existing mobile edge computing caching strategies face several challenges, including privacy leakage due to centralized training, the lack of consideration for the age of cached content, and issues such as poor content freshness, low cache hit rate, and high system energy consumption, which arise from the randomness and high dynamics of user requests in emergency scenarios. To address these issues, a federated reinforcement learning-based multi-unmanned aerial vehicle cooperative dynamic caching replacement strategy is proposed. Firstly, a comprehensive utility objective function is developed, considering user access delay, the age of cached content, and system energy consumption. Secondly, BiLSTM is employed to capture the dynamic dependencies in user request time series, combined with an attention mechanism to focus on key features. Federated learning is used to achieve distributed training across UAVs, ensuring user privacy while improving the accuracy of content popularity prediction. Subsequently, a weighted caching scoring mechanism based on content popularity and information age is proposed to dynamically adjust the cache replacement order. Finally, the cache replacement and user association problems are modeled as a Markov decision process, and the TD3 reinforcement learning algorithm is applied, utilizing double <i>Q</i>-networks and a delayed policy update mechanism, to derive the optimal cache replacement strategy. The experimental results demonstrate that the proposed method outperforms comparison approaches in terms of cache hit rate, access delay, and content freshness, demonstrating strong performance advantages and application potential.</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

Federated reinforcement learning-based multi-UAV cooperative dynamic caching replacement strategy

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Yanpei Liu,
  • Haoyang Zhao,
  • Yanqiang He,
  • Weiwei Zhang,
  • Liang Zhu,
  • Hongchan Li

摘要

Existing mobile edge computing caching strategies face several challenges, including privacy leakage due to centralized training, the lack of consideration for the age of cached content, and issues such as poor content freshness, low cache hit rate, and high system energy consumption, which arise from the randomness and high dynamics of user requests in emergency scenarios. To address these issues, a federated reinforcement learning-based multi-unmanned aerial vehicle cooperative dynamic caching replacement strategy is proposed. Firstly, a comprehensive utility objective function is developed, considering user access delay, the age of cached content, and system energy consumption. Secondly, BiLSTM is employed to capture the dynamic dependencies in user request time series, combined with an attention mechanism to focus on key features. Federated learning is used to achieve distributed training across UAVs, ensuring user privacy while improving the accuracy of content popularity prediction. Subsequently, a weighted caching scoring mechanism based on content popularity and information age is proposed to dynamically adjust the cache replacement order. Finally, the cache replacement and user association problems are modeled as a Markov decision process, and the TD3 reinforcement learning algorithm is applied, utilizing double Q-networks and a delayed policy update mechanism, to derive the optimal cache replacement strategy. The experimental results demonstrate that the proposed method outperforms comparison approaches in terms of cache hit rate, access delay, and content freshness, demonstrating strong performance advantages and application potential.