Research On Resource Management Strategy Of Internet Of Vehicles Based On Mobile Edge Computing

With the rapid development of vehicle networking technology,many resourceintensive applications continue to emerge,and these applications have an explosive growth in demand for computing resources,posing a huge challenge to the limited computing resources of vehicles themselves,making the shortage of computing resources in vehicle networking increasingly severe.In this context,mobile edge computing,as an emerging technology,pushes data storage and task processing to the edge of mobile communication networks to realize the integration of information transmission,data storage and computing,which significantly improves the support capability of mobile communication systems for computation-intensive and time-delay sensitive applications.Therefore,edge computing and its intelligent technology become an effective technical solution to the above problems.However,the limitation of multi-dimensional resources such as computing,caching and transmission,combined with the trend of large-scale device connection in the future network and the performance requirements of business services,make the integration and reasonable allocation of limited multi-dimensional resources become particularly important.In order to give full play to the role of edge computing in vehicle networking applications,it is necessary to rely on efficient multidimensional resource management strategies,and solve the challenges of effective task unloading,dynamic resource management,distributed intelligent learning,and so on.In view of the above challenges,based on the relevant theories and methods of resource allocation,this study gradually and deeply studies the resource management strategy of vehicle networking from three perspectives: wireless resource management of vehicle networking,multi-source computing resource management,task unloading combined with queuing process and joint optimization of resource allocation.The aim of the research is to provide an efficient and reliable resource management scheme for the Internet of vehicles under the background of edge computing.Specific content and innovation points are summarized as follows:(1)Research on radio resource management schemes based on spectrum multiplexing to provide reliable communication basis for edge computing of vehicle networking.In order to give full play to the advantages of spectrum multiplexing in Cellular Vehicle to Everything(C-V2X),the research objective of this paper is to design radio resource management schemes to ensure the reliability of communication between vehicle users and improve the capacity of internet of vehicles.Firstly,this paper formulates a radio resource optimization problem of C-V2 X based on spectrum multiplexing.In order to effectively solve the problem of interference term caused by spectrum sharing in this non-convex problem,a new decoupling method is proposed in this paper,which not only realizes the decoupling of channel allocation and power control,but also provides a compact lower bound of system capacity for the original problem.Based on the proposed decoupling method,this paper designs a hybrid centralized-distributed radio resource management scheme for Mode 3 of C-V2 X.In this scheme,the base station uses the graph matching theory to model the centralized channel allocation and realizes efficient channel allocation and sharing.According to the channel allocation results,vehicle users use the deep deterministic policy gradient algorithm to learn continuous power control strategies in a distributed way to further reduce inter-channel interference.This scheme not only significantly increases the system capacity,but also has low computation complexity.In order to further improve the independent decision-making ability of vehicles and reduce the communication cost of central control,this paper further designs a new distributed radio resource management scheme for Mode 4 of C-V2 X.This scheme innovatively uses the cascade network structure of reinforcement learning to realize the joint and iterative optimization of discrete channel allocation and continuous power control.The experimental results show that the scheme not only endows the vehicle users with independent decisionmaking ability of distributed radio resource management,but also significantly reduces the communication cost.(2)Research on computation resource management strategy under collaborative computing,aiming to make full use of the computation resources of vehicle users and edge nodes,improve the resource utilization,and realize the diverse goal of different equipments.In view of the differentiated needs under the context of collaborative computing of C-V2 X,this paper formulates a bi-level optimization problem of "end-edge collaboration" with mixed incentives,aiming to improve the service efficiency of the upper edge servers,while optimizing the energy consumption and delay performance of the lower vehicle users.In order to effectively solve this problem,this paper innovatively proposes a multi-agent bilevel reinforcement learning framework.In this framework,as the leader,the upper agent learns resource allocation strategy;while,as the follower,the lower agent makes the best response to the leader’s strategy for task offloading.In order to theoretically analyze the properties of the bi-level strategy and the relationship between the "end-edge" entities under mixed incentives,the innovation of this paper is to model the bi-level optimization problem as Stackelberg game.Therefore,the bi-level architecture can effectively simulate the interaction process of Stackelberg game,and realize the cooperation and strategy optimization among the agents through the iterative update of the upper and lower levels,and finally converge to the Stackelberg equilibrium.Finally,the convergence of the proposed learning algorithm is proved to ensure the stability and reliability of the bi-level framework.In addition,the experimental results show that compared with other learning algorithms and Nash equilibrium strategies,this framework is superior in balancing the performance of upper and lower nodes,and can better meet the differentiated needs of various sources in the internet of vehicles.(3)Research on the joint optimization of task offloading and resource allocation based on queuing process to realize online distributed optimization of resource management.In order to solve the problem of random task arrival and time-varying channel state in edge computing,this paper formulates a joint stochastic optimization problem of task offloading and resource allocation considering queuing process,which aims to improve the computing efficiency and guarantee the long-term stability of task queues.In order to achieve this goal,this paper proposes an architecture that integrates Lyapunov optimization method and ActorCritic learning framework.Under this framework,Lyapunov optimization transforms a stochastic problem into a centered online optimization problem,while Actor-Critic learning framework provides a distributed solver for each vehicle user.On this basis,the relationship between the average queue length and the average energy consumption is analyzed in detail,which provides a reliable theoretical guarantee for the performance of the framework.Based on this framework,according to different problem solving methods of Critic module,two low complexity and distributed online schemes are proposed.In the first scheme,the Actor module uses the policy gradient to learn the offloading strategy,and the Critic module decouples the global inequality constraints by dual decompostion.In order to further improve the distributed execution ability of the vehicle users,the Critic module of the second scheme uses the distributed primitive dual algorithm to solve the resource allocation problem.The experimental results show that both online schemes are superior to the current centralized online scheme in terms of average queue length,average energy consumption and execution time.In addition,they also have the characteristics of low complexity and sublinear regret online optimization,which is suitable for online real-time decision making in vehicular environment.