Research On Intelligent Resource Management Based On Reinforcement Learning In Mobile Edge Network

The rapid development of the Internet of Things(IoT)and the Internet has led to the rapid growth of smart mobile devices and data traffic.At the same time,a large number of time-delay sensitive and computationintensive applications have emerged,such as high-definition video,driverless car,telemedicine and face recognition.The massive computing and low latency requirements brought by new applications are constantly challenging the traditional cloud computing network architecture.In order to relieve the backhaul link bandwidth pressure and reduce service response delay,Mobile Edge Computing(MEC)is proposed.MEC uses a distributed network architecture to sink the computing and processing capabilities of cloud servers to the edge of the network.MEC provides real-time computing and localization processing services for end users,reducing network latency and improving quality of service.However,the emergence of MEC makes the original centralized network resource management scheme no longer applicable.Therefore,this dissertation studies the intelligent resource allocation and deployment scheme under MEC network architecture with the help of the highdimensional perception ability and dynamic decision-making ability of deep reinforcement learning and multi-agent deep reinforcement learning.This dissertation studies the two key issues in MEC:computation offloading and edge caching.The main research contents are as follows:In this dissertation,a joint optimization scheme of computation offloading and resource allocation based on multi-agent deep reinforcement learning in heterogeneous cellular networks is studied.Firstly,this dissertation proposes a cloud-edge-end collaborative network architecture which is used to define the system model,computing tasks can be processed locally or unloaded to macro-cell base stations and small-cell base stations.Secondly,this dissertation formulates the joint computation offloading and resource allocation problem,and uses the MADDPG algorithm to solve the problem,so as to minimize the system energy consumption.Finally,the MADDPG-based strategy is compared with the whole unloading strategy and the DDPG-based strategy.Simulation results show that the proposed scheme based on MADDPG algorithm performs better than the two control schemes in reducing system energy consumption and task completion delay.This dissertation studies the base station cooperative cache scheme in heterogeneous cellular network.Considering the problem of low cache performance due to the limited cache space of base stations at the edge of mobile network,this dissertation increases the cache space of base stations at the edge of network by means of communication cooperation between base stations.Then,an average request response delay minimization problem is constructed,and a base station cooperative edge caching scheme based on DDPG algorithm is proposed.Simulation results show that compared with the three traditional cache replacement schemes,the proposed scheme can significantly reduce the average request response delay and guarantee higher request hit ratio.