Research On Computation Offloading And Resource Allocation Scheme Of MEC Network Based On Deep Reinforcement Learning

The requirements for powerful computing capability,high capacity,low latency and low energy consumption of emerging services,pose severe challenges to network performance.As an effective distributed computing pattern,multi-access edge computing(MEC)can provide offloading services in proximity to users by deploying resources at the edge of network,which effectively reduce delay and improve the quality of service(QoS).However,with the increasing of user equipment,there are a large number of users competing for limited resources,so it is necessary to design a reasonable computation offloading and resource allocation strategy to further improve the performance of MEC system.Reinforcement learning is naturally suitable for automatic control and decision-making problems in stochastic dynamic environment,while deep reinforcement learning integrates the advantages of deep learning and reinforcement learning,making it especially suitable for solving problems related to MEC computation offloading.Based on deep reinforcement learning,this paper studies the joint optimization of computation offloading and resource allocation strategy in MEC system from different perspectives.The main contents are summarized as follows:(1)In view of the high energy consumption problem of the base stations,the joint optimization of offloading strategy and resource allocation in MEC enabled ultra-dense networks(UDNs)is studied under the green energy supply mode.The optimization goal is designed to minimize the computation cost of all users,which is defined by the weighted sum of time delay and energy consumption to meet the task requirements of different kind of users.To solve this problem,a centralized solution based on deep reinforcement learning was proposed,which divided the original problem into two stages.The simulation results show that the scheme has good performance in computation cost and timeout rate,and can potentially realize load balancing.(2)In view of the coexistence of heterogeneous services,the joint optimization of offloading strategy,caching strategy and resource allocation is formed by integrating the edge cache to maximize the system revenue.At the same time,considering the complex dynamic scene,a distributed optimization scheme based on multi-agent reinforcement learning is proposed to meet the real-time requirements of MEC system by reduce the complexity of solution.The simulation results show that the proposed scheme can effectively improve the task success rate and reduce the time delay while ensuring the fairness of heterogeneous users as much as possible.