Research On User Mobility Service Migration Based On Deep Reinforcement Learning

With the rapid development of 5G technology and the Internet of things(Io T),cloud computing has been proposed to deal with a large number of remote data of users.However,high-performance applications will produce high transmission delay in the long-distance interaction with the cloud center;more and more data is concentrated in the cloud data center,and the cloud server will cause resource shortage and lead to the decline of service quality.As a result,the Mobile Edge Computing(MEC)appears in the field of vision of researchers.MEC will deploy edge servers with computing power to provide services to users in base stations closer to users than the cloud center to solve the above problems.However,MEC still faces many challenges,especially the problem of user mobility,that is,with the mobility of users,when users approach or exceed the coverage of the edge server,it will lead to the decline of user quality of service,and even lead to service interruption.Therefore,in order to make better use of the advantages of MEC,we will focus on the research of service migration decision algorithm.Firstly,this paper focuses on the problem of service migration based on deep Q-networks in mobile edge computing.In order to ensure that the edge server always provides continuous and reliable services to the user during the mobile process,this paper proposes a deep Qbased network based on the introduction of a Software Defined Network(SDN)to centrally acquire and coordinate global information.The service migration decision algorithm(Deep Q-Learning Network Based Service Migration,DQNSM),through the adaptive adjustment of the agent to the current MEC environment,and the problem is modeled as a Markov decision process,so as to achieve the minimum time joint goals for delay and energy consumption.In this paper,multiple groups of experiments are completed and compared for different numbers of users and edge nodes.The experimental results show that compared with the traditional RL algorithm and the baseline algorithm,the proposed algorithm has better performance in terms of delay and energy consumption optimization.performance.Secondly,this paper focuses on the service migration decision problem based on deep recurrent Q-networks in MEC.Considering that it is unrealistic to obtain global information in a real environment,without considering the introduction of SDN controllers to obtain global information,assuming that each edge server can only obtain the user information of its current service.The problem is modeled as a partially observable Markov decision process.While trying to join the problem of computing offloading,it is necessary to study how mobile users can ensure stable and continuous service quality in an environment where some information is visible.Therefore,this paper proposes a service decision algorithm based on Deep Recurrent Q-Network(Deep Recurrent Q-Learning Network Based Service Migration,DRQNSM),with the goal of minimizing delay and system energy consumption,ensures reliable,stable and continuous services during user mobility.This paper conducts a series of experiments considering the number of users and the influence of the clock frequency of the edge server.It is found that the proposed algorithm has better performance than the classic RL algorithm in terms of reducing delay and energy consumption.Finally,summarize the research content proposed in this paper,reflect on the unresolved problems and existing deficiencies,and look forward to further research and discussion in the field of service migration of mobile edge computing in the future.