Abstract: Unmanned Aerial Vehicle (UAV) swarm, integrated with 5G networks, serves as a fleet of flying tools carrying computational resources, providing additional computational power support for Mobile Edge Computing (MEC) networks. In semi-connected networks, where there are challenges such as scarce infrastructure computing resources, massive task data, uneven distribution of mobile Internet of Things (IoT) devices, and complex communication scenarios using Orthogonal Frequency Division Multiple Access (OFDMA) technology, a Moving-aware Adaptive-parallel Task Offloading System (MATOS) was proposed. The system was comprised of a ground equipment layer, a UAV layer, and an edge computing layer, aiming to reduce task offloading latency and energy consumption, thus enhancing the task offloading success rate. The UAV swarm was utilized as Air Base Stations (ABS) to handle task offloading and relay services. Firstly, to improve the task transmission quality between ground devices and the UAV swarm, a task collaborative collection mechanism was proposed, which combined task attributes with mobile perception of regional service devices. Secondly, an Adaptive Parallel Genetic Ant Colony Optimization (AGACO) task offloading mechanism was proposed, which integrated UAV swarm trajectory planning to achieve load balancing for ABS and reduce task offloading latency. Finally, by jointly optimizing UAV swarm trajectory planning, task offloading latency, and task offloading energy consumption, the task offloading success rate was improved. Experimental results show that MATOS reduces flight energy consumption by 40% compares to the energy efficient edge Cloud Architecture (RESERVE), a Smart and Trusted Multi-UAV Task Offloading system (STMTO), UAV Edge Computing IoT Network (UECIN), Multi-UAV Assisted Offload System (MAOS), and Mobility-aware Online Task Offloading (MOTO). Compared with RESERVE, the task unloading delay is reduced by 38%, and the task unloading energy consumption is reduced by 31%, which verifies the superiority of MATOS. 

Key words: mobile edge computing, Unmanned Aerial Vehicle (UAV) swarm, task offloading, load balancing, track planning, cluster parallel computing

摘要： 无人机群(UAV swarm)结合5G网络成为携带计算资源的集群飞行工具，可以为移动边缘计算(MEC)网络提供额外算力支持。在半连接网络中，针对基础设施算力稀缺、海量任务数据、移动物联网(IoT)设备分布不均以及利用正交频分多址(OFDMA)技术进行通讯的复杂场景，提出由地面设备层、UAV层以及边缘计算层构成的移动感知自适应并行计算任务卸载系统(MATOS)，旨在降低任务的卸载时延和能耗，从而提升任务卸载的成功率。该系统利用UAV swarm作为空中基站(ABS)，完成任务卸载和任务中继服务。首先，为提升地面设备与UAV swarm任务传输质量，结合任务属性与区域服务设备移动感知思想提出了任务协同收集机制。其次，提出自适应并行遗传蚁群模型(AGACO)任务卸载机制，同时结合UAV swarm航迹规划思想，使ABS负载均衡并降低任务卸载时延。最后，以UAV swarm航迹规划、任务卸载时延、任务卸载能耗为联合优化指标，提升了任务卸载成功率。实验结果表明，MATOS与高能效边云架构(RESERVE)、智能可信任务卸载系统(STMTO)、自主预测动态框架(UECIN)、多无人机辅助卸载系统(MAOS)以及自适应负载均衡(MOTO)相比，在飞行能耗上降低了40%，在任务卸载时延上相较于RESERVE降低了38%，在任务卸载能耗上相较于RESERVE降低了31%，验证了MATOS的优越性。

关键词: 移动边缘计算, 无人机群, 任务卸载, 负载均衡, 航迹规划, 集群并行计算