| 摘要: |
| 为了满足物联网（Internet of Things,IoT）系统的用户设备计算需求，使用携带移动边缘计算服务器的无人机（Unmanned Aerial Vehicle,UAV）协同用户设备进行任务卸载。针对用户设备存在的数据隐私和数据开销问题，引入联邦学习进行模型训练。针对计算资源分配不合理的现象，考虑了用户公平性，基于Jain公平指数引入了公平性因子。通过联合优化每架无人机的飞行轨迹和卸载决策，共同最大化系统的总能耗和覆盖范围内的用户公平性，并提出了一种结合Actor-Critic网络的联邦强化学习算法（Federated Reinforcement Learning Combined with Actor-Critic Network,FRLACN），使用Actor-Critic为每个设备生成最优决策动作，进行了更精确的梯度更新，充分利用其异构资源。仿真结果表明，所提FRLACN算法相比传统联邦学习算法降低了11%的总能耗，数据传输成本方面减少了8.7%，并提高了用户公平性。 |
| 关键词: 多无人机协同 任务卸载 移动边缘计算 联邦强化学习 公平性 |
| DOI：10.20079/j.issn.1001-893x.241230005 |
|
| 基金项目:重庆市自然科学基金创新发展联合基金（中国星网）（CSTB2023NSCQ-LZX0114） |
|
| Cooperative Trajectory Optimization and Task Offloading of UAVs for Fairness |
| TIAN Xu,WANG Huahua,LIAO Fujian,ZHENG Shaojie |
| (School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065,China) |
| Abstract: |
| In order to meet the computing needs of user devices in Internet of Things(IoT) systems,unmanned aerial vehicles(UAVs) carrying mobile edge computing servers are used to collaborate with user devices to offload tasks.In order to solve the data privacy and data overhead problems existing in user devices,federated learning is introduced for model training.In view of the irrational allocation of computing resources,the user fairness is considered,and the fairness factor is introduced based on the Jain fairness index.By jointly optimizing the flight trajectory and unloading decisions of each UAV,the total energy consumption of the system and the user fairness in the coverage area are jointly maximized,and a Federated Reinforcement Learning combined with Actor-Critic Network(FRLACN) is proposed.The algorithm uses Actor-Critic to generate the optimal decision action for each device,and performs a more accurate gradient update to make full use of its heterogeneous resources.The simulation results show that the proposed FRLACN algorithm reduces the total energy consumption by 11% and the data transmission cost by 8.7% compared with the traditional federated learning algorithm,and improves the user fairness. |
| Key words: multi-UAV collaboration task offloading mobile edge computing federal reinforcement learning fairness |
