Research On Resource Optimization Of Multi-user Cooperation Mobile Edge Computing

With the development of Internet of Things applications,the number of mobile terminal users is increasing day by day.In order to analyze and process the big data of the Internet of Things,mobile edge computing(MEC)provides a convenient and efficient communication and computing platform.However,in the face of the communication connections of a large number of user devices,there are problems with insufficient computing power and data privacy protection in distributed mobile edge computing systems.In order to solve these challenges,multi-user collaborative computing is assisted by using federated learning technology.The integration of mobile edge computing and federated learning is a research hotspot in today's communications field.Therefore,this thesis builds a multi-user collaborative MEC system based on federated learning.By analyzing the impact of data heterogeneity,device heterogeneity,and network heterogeneity in network applications,it studies optimization algorithms to improve the computing performance of the application system.Firstly,we investigate the problem of data heterogeneity in the homogeneous terminal of the MEC network.This thesis analyzes the impact of the independent and uneven distribution of data on different terminal devices on the calculation time and performance of the system,and uses federated learning for distributed collaborative computing.Through modeling and analysis,combined with the power-law distribution model for devices' data resource management,this thesis designs an optimization algorithm based on dynamic samples,sets the data threshold as a baseline strategy.According to the size of local resources,devices use different strategies for local training model to reduce the system training deviation caused by unbalanced data on the parallel running homogeneous terminal devices,and optimize the overall convergence rate and computing performance of the system.Secondly,we focus on the problem of device heterogeneity in heterogeneous terminals in the MEC network.Considering the heterogeneity of computing,storage and communication capabilities of different terminal devices,this thesis analyzes the "dropping" phenomenon of heterogeneous terminals in collaborative training,and the network congestion problem when massive devices access at the same time.Aiming at the heterogeneity of devices in distributed system,this thesis designs a joint optimization algorithm of asynchronous communication.By setting two asynchronous threads of scheduling and updating on the central server,the asynchronous parallel scheduling terminal device executes the local training model.Compared with the traditional synchronous optimization algorithm,this algorithm can effectively reduce the loss rate of heterogeneous terminals and avoid network congestion.The system fault tolerance mechanism is improved to ensure the scalability,efficiency and flexibility of the system.Thirdly,we study the heterogeneity of the multi-user collaborative MEC network.Considering that multiple end users are distributed in different network environments,the impact of heterogeneous networks on system performance and the uncertainty of network bandwidth are analyzed.The training time of terminal devices in system iteration is analyzed by modeling,and the system cost is minimized by scheduling computing resources.In this thesis,a joint optimization algorithm strategy based on deep reinforcement learning is designed to make the system self adaptively train the model intelligently,reduce the unnecessary waiting idle time in equipment training,ensure the system training to achieve the expected accuracy,reduce the total time cost of the system,maximize the utilization of system resources,and realize the intelligent management of system computing resources.The quality of service of the mobile edge computing network system with multi-user collaborative training is optimized.This thesis mainly studies heterogeneous edge network computing with multi-user collaboration.Through the heterogeneity of the network system,it solves the problems of insufficient system computing power and data privacy security protection,and improves resource utilization efficiency and computing performance.