Research On Computation Offloading Strategy In Mobile Edge Computing

As an important technology with broad prospects in 5G network,mobile edge computing could deploy computing and storage resources closer to terminal equipment,solving problems of terminal equipment such as insufficient computing and storage capacity.It is currently a research hotspot in the field of mobile edge computing to implement computing offloading schemes with high performance according to different optimization metrics such as task computation delay for tasks and energy consumption of terminal equipment.Deep reinforcement learning has both the ability of deep learning on dealing with perception problems and the advantage of reinforcement learning in decision-making,which is especially suitable for solving the problem of dynamic computing offloading in mobile edge computing scenarios.In this thesis,the problem of dynamic task offloading and resource allocation in multi-user scenarios is studied based on deep reinforcement learning algorithms.The main contents of the thesis include:(1)Considering the problem of partial computation offloading in the scenario of multiple terminal users with a single MEC server,this thesis constructs a network communication and task calculation model and an objective function whose total cost is the weighted sum of the total execution delay of all tasks and the total energy consumption of all terminal equipment.Based on this,a partial task offloading and resource allocation scheme based on deep deterministic policy gradient algorithm is proposed.The simulation results show that the proposed scheme has better performance in reducing the total cost of the system than other benchmark strategies in terms of the computing ability of MEC server,the number of terminal users and the amount of computing resources required for a task.(2)To target the problem of full computation offloading in the scenario of multiple terminal users and multiple MEC servers,this thesis designs the MEC system model and problem formulization model in this scenario,in which the optimization goal is to reduce the total system overhead of task computation delay and mobile device energy consumption.These models solve the complete task offloading and resource allocation under each time slot via the proximal policy optimization algorithm.In terms of the computing ability of MEC server,the number of mobile users,and the number of MEC servers,etc.,the simulation experimental results show the proposed models could obtain higher long-term total reward of the system than other baseline offloading algorithms,leading to the improvement of the quality of user experience.In conclusion,this thesis proposes two dynamic computation offloading schemes based on the deep deterministic policy gradient and the proximal policy optimization algorithm,which are designed for two application scenarios,i.e.,multi-user with single-server and multi-user with multi-server respectively.Our schemes can effectively reduce the weighted cost of the total time delay of task processing and energy consumption of mobile devices.