数字孪生辅助UAV网络计算卸载优化  

Offloading Optimization in Digital Twin-Aided UAV Networks

作　　者：缪家辉 郑镐 谢正昊 赖健鑫 蒋丽 MIAO Jiahui;ZHENG Hao;XIE Zhenghao;LAI Jianxin;JIANG Li(Guangdong Key Laboratory of Internet of Things Information Technology,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学物联网信息技术广东省重点实验室,广州510006

出　　处：《北京邮电大学学报》2022年第6期133-139,共7页Journal of Beijing University of Posts and Telecommunications

基　　金：国家重点研发计划项目(2020YFB1807801);移动通信教育部工程研究中心开放研究项目(cqupt-mct-202003)。

摘　　要：针对动态时变无人机(UAV)网络环境下智能终端有限计算资源不能满足资源密集型任务计算需求的问题,使用数字孪生技术构建了UAV网络的孪生模型,并设计了智能终端计算卸载策略。将计算任务卸载过程建模为马尔可夫决策过程,建立联合UAV悬停点选择、计算卸载决策、UAV计算资源分配的优化模型。考虑到孪生模型与真实UAV网络的虚实映射误差,提出了基于近端策略优化的计算卸载优化算法。仿真结果表明,所提算法在适应虚实映射误差方面优于传统的深度强化学习算法。To address the problemthat limited resource cannot meet the computing requirements of resource-intensive tasks in dynamic and time-varying unmanned aerial vehicle(UAV) networks, a digital twin technology is leveraged to construct the twin model of UAV networks, and a scheme of computing tasks offloading is developed for smart terminal. Then, the problem of computing offloading is modeled as a Markov decision process, and an optimization model is established for jointly optimizing UAV hovering point selection, computing tasks offloading decision, and UAV computing resource allocation. Considering the virtual-real mapping error between the twin model and the real UAV networks, a computing tasks offloading approach is designed based on proximal strategy optimization. Numerical results illustrate that the proposed approach can better adapt to the virtual-real mapping error comparing with the traditional deep reinforcement learning.

关 键 词：数字孪生 UAV网络 计算任务卸载 虚实映射误差 近端策略优化 

分 类 号：TN929.5[电子电信—通信与信息系统]