| With the continuous innovation in the application of the Internet of Things,autonomous driving,industrial Internet of Things and other rapid development,intelligent terminals change with each passing day.Intelligent applications have developed from the traditional simple sensor data detection and transmission to the current multimedia context information awareness.To solve the problem of insufficient processing power and limited battery capacity of mobile terminals,and to meet the quality of service for computation-intensive applications,the MEC architecture has been proposed.In MEC,storage and computing resources are deployed to the edge of the network,so mobile terminals can access computing and storage services,resulting in a low latency,high reliability,high bandwidth solution.And the use of D2 D communication technology to jointly utilize the loaded resources of MEC and idle mobile device resources in the network can benefit the development of computation-intensive applications,which is a research direction with great potential.In this thesis,in the MEC system assisted by D2 D,the energy consumption in the MEC system is further optimized.And the MEC service deployment scheme is also designed to further optimize the delay in the MEC system.The specific research contents are as follows:(1)A MEC deployment scheme based on node importance is designed.According to the distribution characteristics of network environment and user request,service node selection and service deployment are carried out based on the theory of complex network and genetic algorithm.Firstly,the nodes are sorted according to the network environment using the node importance evaluation index of complex network theory,and the service nodes are selected.Then according to the user's request distribution,use genetic algorithm to deploy the service.The simulation results show that the proposed scheme is advanced.On the one hand,the simulation scheme based on service node migration can approximate the performance of the proposed scheme,but the performance of this scheme depends on the number of node migration,the complexity is too high,and it is not suitable for the situation of large network scale,which indicates that the proposed scheme is optimal to a certain extent.On the other hand,when the service node is determined,the service deployment of MEC server is a complex 0-1 problem,and the genetic algorithm with low algorithm complexity can get a better solution.The most important is that the simulation results fully demonstrate the good performance of the proposed scheme compared with other related schemes.(2)A deep reinforcement learning based MEC computation offloading scheme is designed.In a D2D-MEC system,the mobility of the user's equipment has a significant impact on computing offloading.To consider the effect of user mobility on computational offloading,this thesis define a long-term cost minimization problem affected by mobility,and then decompose the problem to obtain a Markov decision process involving explicit cost minimization.In order to solve this optimization problem,this thesis designs a computational offloading scheme for dynamic sensing environment information.And the proposed scheme uses deep reinforcement learning method to make offloading decisions based on computational task states and D2 D contact states.Simulation results show that the proposed scheme can make full use of idle user equipment's computing resources and save the total energy consumption of user equipment in continuous time. |
PDF Full Text Request