Research And Implementation Of Computation Offloading Strategy Optimization For Multi-Access Edge Computing Vehicular Networks

With the development of 5G and Internet of Vehicles(IoV),many IoV applications such as autonomous driving,real-time road conditions,virtual reality and so on have emerged,which require efficient processing of large-scale data to improve the driving safety of vehicles and the quality of experience of users.However,the limited computing and storage resources of vehicle equipment are not enough to support the delay-sensitive and computing-intensive applications.Meanwhile,there are hidden dangers of network fluctuation and transmission interruption in long-distance transmission of data caused by cloud computing,which may affect service quality.Multi-access Edge Computing can effectively solve this problem.By deploying small data centers at the edge of the network,MEC can provide users with computing services in the place closest to the terminal equipment,reducing the communication dependence on public network and improving the distributed processing capacity of system resources.It is an appropriate solution for the IoV scenario.Vehicular tasks can be offloaded to MEC servers to solve the problems of insufficient computing capacity of vehicular hardware devices and improve the stability of service.However,the computing resources of MEC are also limited.If system resources are not reasonably allocated and the offloading strategies are not suitable,the increase in computing efficiency brought by MEC will be greatly reduced.Therefore,in this paper,we studied the optimization of offloading strategies for MEC in the Internet of Vehicles scenario,and discussed the problem of whether users offload the tasks,where to offload them,and how to allocate system resources under different scenarios.Firstly,we proposed a Multi-access Edge Computing network architecture oriented to the Internet of Vehicles and regarded the resource competition between vehicles as a complete information dynamic non-cooperative game model.In order to get the Nash Equilibrium solution of the game model,we designed a distributed computation offloading strategy optimization algorithm.The advantage of DGT is simple and efficient.Compared with the other two offloading schemes which are ALC and AEC,the performance of DGT is better.Next,on the basis of VMEC architecture,we regarded cloud server as new computing resources,and proposed an edge-cloud hybrid Multi-access Edge Computing network architecture for the Internet of Vehicles.In this scenario,the communication and computation models are more complex.Thus,we combined two kinds of wireless access technologies which are OFDMA and NOMA to model the communication process.In order to solve the optimization problem of computation offloading strategy in ECH-VMEC architecture,we modelled the offloading process as the MDP and proposed a deep reinforcement learning-based algorithm to jointly optimize the strategies of computation offloading and resource allocation.Experiments showed that the algorithm had the best performance compared with the other computation offloading schemes.In order to verify the feasibility of JCORA algorithm and provide technical reference for IoV service,we designed and implemented an MEC task scheduling management system based on microservices.The system can serve two kind of user roles.One is the system administrator,responsible for the management of MEC nodes such as instance addition,image deployment,and node performance monitoring.The other is vehicle users.Ordinary users can upload the information of tasks to system and generate computation offloading strategy by invoking the algorithm.We have given the detailed design and test results of the system,which proved that the system had certain practical value.