| With the rapid development of network technology in recent years,mobile edge computing has been widely used in various IoT scenarios because of its excellent computing power and rapid deployment capability.Meanwhile various new applications are emerging,such as navigation,video processing,face recognition and driverless.The computationally intensive,time-delay sensitive of these applications and the dynamic movement of users pose a new round of challenges to traditional edge computing networks.How to improve the flexibility of edge computing,the service extensibility of edge computing devices and the optimization of computing offloading strategy have become the key to improve the service quality of edge computing users.However,the traditional offloading strategy based on mathematical programming has poor performance in dynamic scenarios and cannot meet the requirements of large-area multi-mobile terminals.In order to better adapt to the new scene requirements,this dissertation establishes the edge computing framework with the assistance of multiple UAVs as mobile relays.Firstly,the offloading strategy based on deep reinforcement learning is proposed in the user terminal part.By mapping the status of mobile terminal and UAV to offloading action,the corresponding strategy is developed.At the same time,the neural network model is used to realize the optimal offloading method in the multi-task scenario in the UAV swarm.The appropriate UAV is selected to perform offloading action through centralized control,and the task is offloaded to the edge server.Finally,two offloading strategies of mobile terminal and UAV are combined to optimize task offloading.The main work of this dissertation is summarized as follows:1.We propose a task offloading strategy for multi-user terminal scenarios.Firstly,multiple system models such as task queue,energy queue and data communication are constructed in the multi-user terminal task offloading scenario,and the problem of minimizing energy consumption of mobile terminals in this scenario is proposed.Secondly,according to the optimization objectives,the offloading strategy based on DDPG deep reinforcement learning is proposed,and then we propose the improvement method for offloading strategy in collaborative scenarios.Finally,experiments are carried out to verify the effectiveness of the proposed method in the offloading scenario of mobile terminal tasks.2.We propose a task offloading strategy in multiple UAVs scenario.Firstly,in the task offloading scenario assisted by multiple UAVs,system models such as data communication and MEC calculation are constructed,and the optimization objective of maximizing the task offloading amount of UAVs in this scenario is proposed.In order to select the offloading UAV under this objective,this dissertation proposes a task offloading selection algorithm based on DQN deep reinforcement learning,and further optimizes the neural network architecture for sampling training of the network model.Finally,several experiments demonstrate the superiority of the proposed algorithm compared with the baseline algorithm in UAV group offloading scenario.3.We propose a joint optimization strategy for task offloading in UAV-assisted scenarios.Combined with the offloading strategy of mobile terminal task and UAV task,this dissertation constructs the offloading scenario model of device terminal task with the assistance of multiple UAV s,and gives the information of each node and the execution actions under the system model.At the same time,deep reinforcement learningbased task offloading strategies were deployed on the mobile user side and the UAV side to minimize the energy consumption of mobile devices.Finally,experiments show that the task offloading joint algorithm designed in this dissertation has high time-delay sensitivity and low resource consumption,which can realize the optimization of time-delay sensitive tasks and effectively expand the application of deep reinforcement learning in the field of edge computing.. |
PDF Full Text Request