Research On Stable And Scalable Offloading In Mobile Edge Computing Based On Deep Reinfocement Learning

In recent years,new computation-intensive applications represented by virtual reality,augmented reality and networked autopilot continue to emerge,which puts forward higher requirements for the computing and communication capabilities of user equipment and wireless networks.Mobile edge computing(MEC),as a new computing paradigm,offloads the computing tasks of user devices to the MEC server,which not only effectively makes up for the lack of computing resources and battery capacity of user devices,but also meets the low latency requirements of applications.However,the existing computation offloading scheme is usually designed and optimized for specific types of applications.When faced with new applications with different and dynamic internal task dependencies,the existing scheme cannot be used without adjustment,and the scalability is poor.In addition,the offloading scheme based on deep reinforcement learning(DRL)cannot guarantee the stability of the application service quality in the training and deployment phase because it needs to randomly explore the trial and error.Therefore,this paper focuses on the scalability of the edge offloading scheme caused by the diversity and dynamic change of the internal task dependencies of mobile applications and the stability of the application quality of service caused by the DRL random exploration and error,respectively for single-user and multi-user computation offloading scenarios.In view of the scalability problem caused by the variety and dynamic change of the internal task dependencies of the application,this paper proposes a scalable computing offloading scheme based on the graph neural network(GNN).First,considering the change of wireless network channel,this paper proposes a fine-grained task-level offload scheduling mechanism;Secondly,this paper designs an expandable preprocessing network based on graph neural network to deal with dynamically changing task dependent structure and wireless network channel information;Finally,the policy network is trained based on the policy gradient algorithm,and the optimal offloading scheduling strategy is obtained.Simulation results show that the proposed scheme has significant effectiveness and advantages under different application traffic densities.Compared with the benchmark scheme,the proposed scheme can reduce the application completion time by at least 172%and the user equipment energy consumption by at least 45%.Aiming at the stability of application quality of service caused by DRL random exploration trial and error,this paper proposes a stable computing offloading scheme based on security reinforcement learning.In particular,the offloading scheme based on traditional reinforcement learning cannot guarantee the stability of application quality of service due to the need for random exploration and trial and error,and cannot provide stable delay guarantee for mobile application computing in the training and deployment stage.This paper proposes a stable computation offloading scheme based on safe reinforcement learning(SRL)by introducing the extended computation offloading scheme based on the above GNN.This scheme can minimize the energy consumption of user equipment under the condition of ensuring the stability of application service quality by optimizing the offload scheduling and power control.The simulation results show that compared with the benchmark scheme,the proposed scheme can effectively improve the stability of the quality of service in the training phase and deployment phase under different user equipment numbers,and has significant advantages in reducing the energy consumption of user equipment.