Research On Intelligent Network Access In Mobile Edge Computing

With the rapid development of emerging Internet of Things(IoT)mobile applications such as autonomous driving,smart city and industrial automation,IoT mobile devices such as intelligent vehicles,smart phones,and industrial robots have placed higher demands on the computing and storage capabilities of the network.Mobile edge computing(MEC)can provide real-time and reliable computing task processing for IoT devices by pushing the computing and storage capabilities of cloud computing to the edge network close to mobile devices,which has been regarded as one of the key technologies of the fifth generation mobile communication(5G)network.Supported by MEC networks,this thesis considers the differentiated application requirements of IoT mobile devices with low latency,low energy consumption and high reliability,and conducts research on intelligent network access in MEC.By leveraging theories such as stochastic optimization and reinforcement learning to solve the problems of network association,task offloading and resource allocation in scenarios composed of multiple edge servers,it designs intelligent network access strategies that match the requirements of differentiated applications in IoT.This can reduce the latency and energy consumption of IoT mobile devices,and improve the reliability of computing task processing.The main research contents and contributions are as follows:(1)Considering that the network handover and task migration introduced by the mobility of vehicles lead to the deterioration of computing task delay performance,this thesis conducts the research on mobility-aware low-latency network association.By deriving the location transition probability for highways,two-dimensional streets and real-data scenarios of moving vehicles,a low-latency network association method based on Markov decision process(MDP)is proposed.Considering the uncertain state transition probability caused by driving route adjustment in practice,a robust low-latency network association algorithm based on the min-max theory is proposed.The simulation results show that compared with the nearest distance strategy and the cost minimization strategy in T single slots,the proposed algorithm can reduce the average delay of 36.95%and 8.56%,and has stable low-latency performance with the increasing uncertainty of the transition probability.(2)Considering that the mobility of IoT devices and the dynamics of the network environment(such as channel quality,the computing processing capability,etc.)have an impact on the delay performance of computing task,and the network association between devices has asynchronous requirements,this thesis develops mobility-aware low-latency asynchronous dynamic network association.By modeling the device mobility and the dynamics of network environment as semi-Markov states and random events,a two-stage online dynamic network association algorithm based on Lyapunov optimization and MDP method is proposed for the single mobile device scenario.Considering the asynchronous challenge of network association decisions among multiple devices,a low-latency asynchronous dynamic network association algorithm based on nonlinear optimization is proposed.The simulation results show that compared with the best channel strategy and the MDP-based strategy,the proposed algorithm can reduce the average task delay of 26.74%and 13.26%,respectively.(3)Considering that the network access methods based on fixed parameter optimization can not minimize the long-term energy consumption of devices in the dynamic MEC network,and it is difficult to ensure the network stability,this thesis conducts the energy-efficient joint optimization of network association and resource allocation.By taking into account the spatio-temporal dynamic characteristics of channel quality,computing power,and connection capacity between multiple edge servers,an energy-efficient network association and task offloading method based on single-timescale optimization is proposed.The simulation results show that compared with the nearest distance strategy and the myopic optimization strategy,the proposed method can reduce the energy consumption of 86.79%and 66.70%,respectively.Considering the limited wireless resources and computing resources in the MEC networks,in order to alleviate the high energy consumption caused by frequent network associations,an online network association and resource allocation method based on two-timescale optimization is proposed.The simulation results show that the proposed two-timescale optimization algorithm can achieve lower energy consumption than the single-timescale optimization algorithm with the increasing handover cost.(4)Considering that the network access methods with the minimizing objective of the average delay of device or the total delay of system can not meet the requirements of highly reliable computing processing of devices in smart manufacturing,this thesis presents the interactive high-reliability joint optimization of network association and resource allocation.By taking into account the interactive characteristics of mobile devices,an interactive delay model and high-reliability network access problem are formulated.Using edge servers as intelligent agents,the problem of highly reliable network access is transformed into a decentralized partially observable MDP(Dec-POMDP)problem.Based on multi-agent deep deterministic policy gradient(MADDPG)theory,a distributed network association and resource allocation algorithm is proposed.The simulation results show that the proposed algorithm can improve the timely reliable completion rate of 44.76%compared with the deep deterministic policy gradient(DDPG)-based strategy.