基于近端策略优化的无人机辅助移动边缘计算  

UAV-assisted mobile edge computing based on proximal policy optimization

作　　者：胡静 李君 李正权 徐钰龙 张圣 HU Jing;LI Jun;LI Zheng-quan;XU Yu-long;ZHANG Sheng(School of Electronics and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;College of Electronic Information Engineering,Wuxi University,Wuxi 214105,China;State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]无锡学院电子信息工程学院,江苏无锡214105 [3]北京邮电大学网络与交换技术全国重点实验室,北京100876

出　　处：《计算机工程与设计》2025年第4期974-982,共9页Computer Engineering and Design

基　　金：网络与交换技术全国重点实验室(北京邮电大学)开放课题基金项目(SKLNST-2023-1-13)。

摘　　要：为更高效利用移动边缘计算(MEC)系统中的有限资源,提出一种基于近端策略优化(PPO)的无人机(UAV)辅助边缘卸载优化算法。通过联合优化任务卸载率、用户通信资源分配、无人机飞行角度和飞行速度,结合状态归一化算法,以最小化最大处理延迟为目标。实验结果表明,所提算法能够快速收敛到最优且稳定性较强。与基线算法深度确定性策略梯度(DDPG)和Actor-critic(AC)相比,PPO算法在处理延迟方面较DDPG提升22%,较AC提升近48%。To ensure more efficient utilization of limited resources in mobile edge computing(MEC)systems,a proximal policy optimization(PPO)based optimization algorithm for unmanned aerial vehicle(UAV)-assisted edge offloading was proposed.The task offloading rate,user communication resource allocation,UAV flight angle and flight speed were jointly optimized,and a state normalization algorithm was combined,the objective was to minimize the maximum processing delay.Experimental results show that the proposed algorithm can quickly converge to the optimum and is highly stable.Compared with the baseline algorithms deep deterministic policy gradient(DDPG)and Actor-critic(AC),the PPO algorithm improves the processing latency by 22%and nearly 48%respectively compared to DDPG and AC.

关 键 词：移动边缘计算 无人机 近端策略优化 计算卸载 深度强化学习 资源分配 状态归一化 

分 类 号：TN929.5[电子电信—通信与信息系统]