Scholarlink

<p>In view of the surge in computing demand for ground mobile devices (MDs) and insufficient ground network coverage, this paper proposes a collaborative edge computing architecture based on a LEO-MEO dual-layer satellite network to solve the computing offloading challenges in remote areas and disaster scenarios. LEO satellites have the advantages of wide coverage and low latency, while MEO satellites can provide more stable computing resources. To this end, this paper innovatively constructs a dynamic task scheduling mechanism to prioritize latency-sensitive tasks to LEO satellites and offload computing-intensive tasks to MEO satellites to achieve a balance between resource efficiency and service quality. In order to optimize task offloading decisions in a dynamic network environment, this paper proposes a D3QN-LSTM intelligent offloading algorithm, combines the LSTM module to predict the MEO load trend, uses Dueling network and Double Q-learning to improve action selection accuracy and learning stability, and designs a multi-objective reward function to coordinate the constraints of task survival time, energy consumption and delay. In addition, in view of the communication limitations of satellite networks, a distributed learning framework is designed to reduce communication overhead through a parameter sharing mechanism and enhance decision reliability when the link is interrupted. Simulation results show that the proposed scheme is significantly superior to traditional methods in terms of computing delay, energy consumption and system stability, providing an efficient solution for space-air collaborative computing.</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

Mobile edge computing task offloading strategy for two-layer satellite network based on deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Heng Du,
  • Ligang Cong,
  • Xu Liu,
  • Qingyun Liang

摘要

In view of the surge in computing demand for ground mobile devices (MDs) and insufficient ground network coverage, this paper proposes a collaborative edge computing architecture based on a LEO-MEO dual-layer satellite network to solve the computing offloading challenges in remote areas and disaster scenarios. LEO satellites have the advantages of wide coverage and low latency, while MEO satellites can provide more stable computing resources. To this end, this paper innovatively constructs a dynamic task scheduling mechanism to prioritize latency-sensitive tasks to LEO satellites and offload computing-intensive tasks to MEO satellites to achieve a balance between resource efficiency and service quality. In order to optimize task offloading decisions in a dynamic network environment, this paper proposes a D3QN-LSTM intelligent offloading algorithm, combines the LSTM module to predict the MEO load trend, uses Dueling network and Double Q-learning to improve action selection accuracy and learning stability, and designs a multi-objective reward function to coordinate the constraints of task survival time, energy consumption and delay. In addition, in view of the communication limitations of satellite networks, a distributed learning framework is designed to reduce communication overhead through a parameter sharing mechanism and enhance decision reliability when the link is interrupted. Simulation results show that the proposed scheme is significantly superior to traditional methods in terms of computing delay, energy consumption and system stability, providing an efficient solution for space-air collaborative computing.