面向任务卸载的鲁棒多智能体联邦强化学习  

Robust Multi-agent Federated Reinforcement Learning for Task Offloading

作　　者：严地宝 文红 侯文静[1,2,3] 王永丰 马文迪 孙凡 YAN Dibao;WEN Hong;HOU Wenjing;WANG Yongfeng;MA Wendi;SUN Fan(School of Aeronautics and Astronautics,UESTC,Chengdu Sichuan 611731,China;Aircraft Swarm Intelligent Sensing and Cooperative Control Key Laboratory of Sichuan Province,Chengdu Sichuan 611731,China;Sichuan Provincial Engineering Research Center of Communication Technology for Intelligent IoT,Chengdu Sichuan 611731,China)

机构地区：[1]电子科技大学航空航天学院,四川成都611731 [2]电子科技大学飞行器集群智能感知与协同控制四川省重点实验室,四川成都611731 [3]电子科技大学四川省智慧物联通信技术工程研究中心,四川成都611731

出　　处：《通信技术》2024年第8期850-854,共5页Communications Technology

基　　金：国家自然科学基金(U23B2021,61901089)。

摘　　要：随着各种移动智能终端的普及,边缘计算和计算卸载技术为这些智能终端提供算力支持来提升终端设备的服务质量。为了使边缘计算设备能够学习多种复杂任务卸载场景下的任务卸载策略,提出了一种基于多智能体深度强化学习算法的联邦学习策略算法,旨在将多个边缘计算设备的训练策略聚合起来,并学习更优的任务卸载策略以适应更多样的场景,同时对于恶意节点有一定的防护能力。以奖励翻转攻击为例,该算法展示了对于恶意节点的有效识别与抵御能力。实验结果显示,所提算法不仅能够有效防御恶意攻击,还能够学习各节点的任务卸载策略。With the proliferation of various mobile smart terminals,edge computing and computational offloading technologies have emerged as pivotal support mechanisms,enhancing the service quality of these terminals.To facilitate the learning of task offloading strategies across diverse and complex scenarios,this paper proposes a federated learning strategy algorithm based on multi-agent deep reinforcement learning,which is designed to aggregate the training strategies of multiple edge computing devices,thereby enabling the synthesis of superior task offloading strategies tailored to a wide array of environments while also providing safeguards against malicious nodes.Taking the reward inversion attack as an example,the algorithm illustrates its ability to identify and counteract malicious nodes effectively.Experimental results validate that the proposed algorithm not only robustly defends against malicious attacks,but also adeptly learns the task offloading strategies pertinent to each node.

关 键 词：任务卸载 恶意节点检测 联邦学习 多智能体深度强化学习 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]