MEC辅助的多基站系统中基于多代理的联合任务卸载和资源分配  

Joint Task Offloading and Resource Allocation Based on Multi-agent for MEC-enabled System with Multi-BS

作　　者：王嫦 孙彦赞[1] 张舜卿 陈小静 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《工业控制计算机》2025年第1期63-65,共3页Industrial Control Computer

摘　　要：在对下一代通信网络架构的广泛研究中,对于有多密集基站(Base Station,BS)的移动边缘计算(Mobile Edge Computing,MEC)系统来说,任务卸载和资源分配问题变得越来越复杂,首先将这一多维动态问题表述为一个优化问题。其次,为了使系统开销最小化,提出了一种基于注意力机制的多代理近端策略优化(Attention-based Multi-agent Proximal Policy Optimization,A-MAPPO)算法,该算法采用集中式训练和分布式执行(Centralized Training and Decentralized Execution,CTDE)框架,并利用注意力机制来促进Critic网络的收敛,从而提高算法的性能。最后,实验结果表明,与其他基准算法相比,A-MAPPO算法可以最多降低28.3%的系统成本。In an extensive study of next-generation communication network architectures,the problem of task offloading and resource allocation has become increasingly complex for mobile edge computing(MEC) systems with many dense base stations(BSs),and we first formulate this multidimensional dynamic problem as an optimization problem.In order to minimize the system cost,this paper proposes an attention-based multi-agent proximal policy optimization(A-MAPPO) algorithm,which employs centralized training and decentralized execution(CTDE),and exploits the attention mechanism to facilitate the convergence of the critic networks,thus improving the performance of the algorithm.Simulation experiments show that the A-MAPPO algorithm can reduce the system cost by 28.3% compared to other baseline algorithms.

关 键 词：移动边缘计算 多代理深度学习 资源分配 任务卸载 

分 类 号：TN929.5[电子电信—通信与信息系统] TP18[电子电信—信息与通信工程]