无人机群场景下边端协同计算卸载技术  

Research on edge to end collaborative computing offloading technology in unmanned aircraft cluster scenarios

作　　者：黄子祥 张新有[2] 邢焕来[2] 冯力 Huang Zixiang;Zhang Xinyou;Xing Huanlai;Feng Li(Tangshan Research Institute,Southwest Jiaotong University,Tangshan Hebei 063000,China;School of Computer&Artificial Intelligence,Southwest Jiaotong University,Chengdu 610000,China)

机构地区：[1]西南交通大学唐山研究院,河北唐山063000 [2]西南交通大学计算机与人工智能学院,成都610000

出　　处：《计算机应用研究》2024年第5期1515-1520,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(62172342)。

摘　　要：当今全球频繁出现自然灾害,针对一种无人机协同下的应急救灾计算卸载场景,提出一种带有协调器的边-端架构。综合考虑场景中的时延、能耗与无人机之间的负载均衡作为系统总代价,采用改进的深度强化学习算法APPO(advanced proximal policy optimization),以最小化系统总代价为目标进行卸载优化。任务的部分卸载相比二进制卸载可以更大程度上降低系统的总代价,APPO算法针对不同的任务情况可以找到合适的卸载比例与无人机进行卸载。仿真与实验结果表明,该算法与全本地处理相比,系统总代价降低了约50%,与较先进的A2C相比,系统总代价降低了约14%。展现了所提策略在该场景下的优越性。This paper proposed a coordinated edge-to-edge architecture for an emergency response scenario assisted by unmanned aerial vehicles(UAVs),taking into consideration the frequent occurrence of natural disasters globally.The architecture incorporated a coordinator and aimed to minimize the overall system cost by considering factors such as latency,energy consumption,and loaded balancing among UAVs.And it utilized an improved deep reinforcement learning algorithm called advanced proximal policy optimization(APPO)to optimize the offloading process.Compared to binary offloading,partial offloading of tasks could effectively reduce the overall system cost.The APPO algorithm enabled users to determine suitable offloading ratios and allocate tasks to UAVs based on different task scenarios.Simulation and experimental results demonstrate a reduction of approximately 50%in the overall system cost compared to full local processing,and about 14%compared to the advanced A2C algorithm,showing the superiority of the proposed strategy in this specific scenario.

关 键 词：边缘计算 深度强化学习 边端协同 无人机协同 计算卸载 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]