数据新鲜度驱动的协作式无人机联邦学习智能决策优化研究  

A Research on Collaborative UAVs Intelligent Decision Optimization for AoI-driven Federated Learning

作　　者：范文 韦茜 周知 于帅 陈旭[1] FAN Wen;WEI Qian;ZHOU Zhi;YU Shuai;CHEN Xu(School of Computer Science and Engineering,Sun Yat-Sen University,Guangzhou 510006,China)

机构地区：[1]中山大学计算机学院,广州510006

出　　处：《电子与信息学报》2022年第9期2994-3003,共10页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(U20A20159,61972432)。

摘　　要：联邦学习是6G关键技术之一,其可以在保护数据隐私的前提下,利用跨设备的数据训练一个可用且安全的共享模型。然而,大部分终端设备由于处理能力有限,无法支持复杂的机器学习模型训练过程。在异构网络融合环境下移动边缘计算(MEC)框架中,多个无人机(UAVs)作为空中边缘服务器以协作的方式灵活地在目标区域内移动,并且及时收集新鲜数据进行联邦学习本地训练以确保数据学习的实时性。该文综合考虑数据新鲜程度、通信代价和模型质量等多个因素,对无人机飞行轨迹、与终端设备的通信决策以及无人机之间的协同工作方式进行综合优化。进一步,该文使用基于优先级的可分解多智能体深度强化学习算法解决多无人机联邦学习的连续在线决策问题,以实现高效的协作和控制。通过采用多个真实数据集进行仿真实验,仿真结果验证了所提出的算法在不同的数据分布以及快速变化的动态环境下都能取得优越的性能。Federated learning is one of the key technologies of 6G,which can use cross-device data to train a usable and safe sharing model on the premise of protecting data privacy.However,most end devices have limited processing capabilities and can not support complex machine learning model training processes.In the framework of Mobile Edge Computing(MEC)in a heterogeneous network convergence environment,multiple Unmanned Aerial Vehicles(UAVs)are used as aerial edge servers to move flexibly within the target area in a collaborative manner,and collect fresh data in time for federated learning and local training to ensure real-time data learning.Multiple factors,such as data freshness,communication cost and model quality,are considered,and the flight trajectories of UAVs,the communication decisions with the user equipment,and the collaborative work between UAVs are comprehensively optimized.Moreover,a priority-based decomposable multi-agent deep reinforcement learning algorithm is used to solve the continuous online decision-making problem of multiple UAVs federated learning to achieve effective collaboration and control.By using multiple real data sets for simulation experiments,simulation results verify that the proposed algorithm can achieve superior performance under different data distributions and in rapidly changing complex dynamic environments.

关 键 词：移动边缘计算 联邦学习 深度强化学习 无人机 信息年龄 

分 类 号：TN915[电子电信—通信与信息系统]