| Large-scale commercial operations of The Sixth Generation(6G)mobile networks will lead to a variety of mobile applications,such as mobile online gaming,augmented reality and intelligent navigation.However,most of the new applications are resource-intensive and delay-sensitive,and the end devices are mostly low-power devices with limited resources.Therefore,it is challenging for users to perform computing tasks locally.Due to the advantages of the unmanned aerial vehicle(UAV),such as low cost,high flexibility and easy deployment,UAV network will become an important part of the future 6G communication network.Mobile Edge Computing(MEC)can deploy Computing and storage resources at the edge of the network to help terminal devices process high-computing applications and meet the low latency requirements of terminal devices.Research on edge computing based on UAV network has become a hot issue in the academic and industrial field.In the traditional UAV network scenario,many scholars have proposed a computing task offloading scheme based on the joint optimization of multi-dimensional resources such as communication and computing,but the dynamic network topology of UAV network is not considered.Therefore,the traditional computing task offloading scheme is difficult to adapt to the dynamic UAV network environment.Therefore,this thesis proposes an edge intelligently driven unmanned aerial vehicle network computing task offloading scheme.In addition,6G mobile network in meet the communication needs of the human mind at the same time,also to the user data security and privacy protection has brought great challenge,for some special computing tasks,such as machine learning tasks,to unload the primitive data of terminal equipment directly to unmanned aerial vehicle to calculate will reveal data privacy of terminal equipment.Therefore,this thesis considers the use of federated learning to improve the protection of user privacy data.the main research work of this thesis is as follows:(1)Firstly,aiming at the problem of computing task offloading and resource allocation in the UAV network,this thesis constructs an optimization problem from the perspective of UAV benefits.Considering the time-varying characteristics of the computing offloading environment,Double Deep Q-learning Learning(DDQN)algorithm is used to jointly optimizes the offloading decision,UAV hovering point,computing resources and other variables during the offloading process,and maximizes the UAV utility function while ensuring that the computing task of the ground terminal equipment can tolerate the delay.In addition,in order to enable the UAV network edge computing to provide optimal decisions in real time,this thesis proposes a digital twin-based DDQN algorithm framework.The simulation results show that the proposed scheme enables the UAV to obtain greater returns compared to the existing schemes.(2)Then,this thesis first proposes a multi-UAV-assisted federated learning scenario for data privacy protection and resource allocation in UAV network edge computing.From the perspective of delay,the iterative process of federated learning is mathematically modeled.Considering the time-varying characteristics of the UAV network,the Multi-Agent Deep Deterministic Policy Gradient(MADDPG)algorithm is used to optimize UAV trajectory and channel resources to minimize the iteration time of federated learning.In order to further improve the execution efficiency of multi-UAV federated learning and make optimal decisions in real time,this thesis proposes a MADDPG algorithm framework based on digital twin technology.Simulation experiments show that compared with existing schemes,the proposed algorithm has good convergence and can significantly reduce the iteration time of federated learning. |
PDF Full Text Request