Research On Computation Offloading Strategy In Mobile Edge Computing Based On Multi-Agent Deep Reinforcement Learning

With the rapid development of the fifth generation mobile communication technology and the Internet of Things,the number of intelligent terminals is growing explosively.More and more computing intensive applications have caused high network load.The demand for low delay response makes the traditional mobile cloud computing(MCC)no longer applicable.Mobile edge computing(MEC)can effectively deal with the challenges of limited resources of mobile users by transferring computing tasks to the edge server near mobile users,providing low latency services while solving the problems of insufficient computing and storage capacity of mobile users.However,due to the limitation of fixed location,traditional MEC can not be deployed anytime and anywhere,and can not meet the needs of emergency scenarios and hot spots.In addition,the rapid development of the Internet of vehicles has brought new application scenarios to the traditional MEC.The mobility of vehicles and untrusted environment have also brought new challenges to vehicles edge computing(VEC).The thesis studies the above problems.The main research contents and innovations are as follows:1.In view of the problem that the traditional MEC can not be deployed anytime and anywhere due to the limitation of fixed location,the thesis considers a MEC network assisted by unmanned aerial vehicle(UAV)to provide computing offloading service for mobile users with random movement and task arrival.Firstly,the offloading problem in UAV assisted MEC network is modeled in terms of mobility,computing,communication and energy consumption.Then,the long-term constrained optimization problem is decoupled into time slot optimization problem by Lyapunov optimization.Finally,a computing task offloading scheme based on Multi-Agent reinforcement learning algorithm MADDPG is proposed.Simulation results show that the proposed scheme can significantly reduce the processing time of mobile users’ computing tasks,and can still perform well in the case of random movement of users and limited energy of UAV.2.Aiming at the untrusted environment in the vehicle edge computing scenario,the thesis proposes a trusted multi-vehicle task offloading framework in the vehicle edge computing scenario.Firstly,the thesis designs a vehicle edge computing network with multiple vehicles and multiple road side units(RSU),in which the vehicle moves according to the track,and the best target of task offloading is selected according to the state of surrounding RSU and its own state during the movement.Then,based on the network,the thesis constructs the offloading decision,delay,energy consumption and reputation model,with the optimization goal of reducing vehicles delay,energy consumption and migration rate.Finally,the thesis uses the trusted multi-vehicle offloading scheme based on Multi-Agent reinforcement learning algorithm MAPPO to solve this problem.Simulation results show that the proposed scheme effectively reduces the delay,energy consumption and migration rate of vehicles in the process of task offloading,and has obvious advantages in untrusted environment.