首页|邻域感知的分布式智能边缘计算卸载和资源分配算法

邻域感知的分布式智能边缘计算卸载和资源分配算法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
随着大量计算密集型和时延敏感型任务的出现，利用移动边缘计算(mobile edge computing，MEC)来提高用户体验并降低系统能耗已成为研究热点。然而，在密集部署的MEC网络场景下，无线网络状态复杂的空间相关性和动态性给卸载方案的制定带来了严峻挑战。本文针对多基站多用户MEC网络场景，研究了一种智能协作的计算卸载和资源分配算法。首先，提出了卸载决策、信道分配、传输功率分配和计算资源分配的联合优化问题，旨在用户时延约束下最小化系统的能耗。其次，由于该问题是一个混合整数非线性规划问题，本文提出了一种基于图注意力网络的混合动作多智能体强化学习算法(graph attention network-based hybrid-action multi-agent reinforcement learning，Gat-HMARL)，将基站作为智能体并配置该算法。Gat-HMARL算法通过图注意力网络捕捉无线网络状态之间潜在的空间相关性，使基站有选择性地关注邻域中其他基站的无线网络状态信息，从而学习更优的计算卸载和资源分配策略。最后，仿真结果表明Gat-HMARL与基准算法相比在性能上有明显提升。
Neighborhood-aware distributed intelligent computing offloading and resource allocation for edge computing
With the emergence of massive compute-intensive and delay-sensitive tasks,mobile edge computing(MEC)has become an active research field to improve user experience and reduce system energy consumption.However,in densely deployed MEC networks,the complex spatial relations and dynamics of the wireless network state present serious challenges for designing offloading schemes.In this paper,an intelligent collaborative computing offload and resource allocation algorithm is proposed for a multibase station and multiuser MEC network.First,we formulate a joint optimization problem of offloading decision,channel allocation,transmit power allocation,and computation resource allocation to minimize the system energy consumption under delay constraints.Then,because this problem is a mixed integer nonlinear programming problem,we propose a graph attention network-based hybrid-action multiagent reinforcement learning algorithm(Gat-HMARL),where each base station refers to an agent and configures the Gat-HMARL algorithm.Gat-HMARL adopts a graph attention network to capture the potential spatial relations of wireless network states,allowing the base stations to selectively attend to the wireless network state of other base stations to learn better computing offloading and resource allocation strategies.Finally,the simulation results demonstrate that the proposed Gat-HMARL algorithm exhibits a remarkable performance improvement than the benchmark algorithms.

mobile edge computingcomputing offloadingresource allocationmulti-agent reinforcement learninggraph attention network

李云、张剑鑫、姚枝秀、夏士超

展开 >

重庆邮电大学通信与信息工程学院,重庆 400065

重庆邮电大学软件工程学院,重庆 400065

移动边缘计算 计算卸载 资源分配 多智能体强化学习 图注意力网络

国家自然科学基金国家自然科学基金重庆市自然科学基金创新发展联合基金重庆市教委科学技术研究计划青年项目中国博士后科学基金

62071077623010992022NSCQ-LZX0191KJQN2023006382023MD734137

2024

10.1360/SSI-2023-0177

中国科学F辑
中国科学院,国家自然科学基金委员会

中国科学F辑

CSTPCD北大核心
影响因子：1.438
ISSN：1674-5973
年,卷(期)：2024.54(2)
  • 26