Research On Mobile Edge Network Offloading Based On Deep Learning

In the research of Mobile Edge Computing(MEC),the computing offloading of tasks has always been a key issue.Perfect computing offloading decisions and corresponding optimal resource allocation schemes can greatly improve the service performance and user experience of MEC.In addition,in the process of solving the optimal resource allocation problem,how to effectively improve the computing efficiency to meet the MEC low latency and other performance also has important research significance.As a powerful analysis tool that can process massive data,deep learning has sufficient research capabilities.feasibility.In addition,deep reinforcement learning(DRL),which combines deep learning and reinforcement learning,has both the powerful analytical capabilities of deep learning and the complex exploration and interaction capabilities of reinforcement learning,which can be used to solve the problem of offloading optimization in complex environments.This paper mainly solves the problems related to the application of deep learning and deep reinforcement learning in the field of MEC computing offloading,and solves the computing offloading and subsequent resource allocation of user task requests in a single MEC unit environment.The specific research work is summarized as follows:First,for the task request generated by the mobile terminal equipment,a collaborative system is constructed to realize the offloading decision and resource allocation as efficiently as possible.This paper mainly considers combining deep neural networks with existing iterative optimization algorithms to build a joint system to solve the problem model.At the same time,in the process of learning and training the neural network,considering the random fluctuation of the channel coefficient during the data transmission of the network system,a sub-regional training mode is proposed to ensure the timeliness of the system.The simulation results show that the prediction of some important iterative factors in the optimization algorithm through the neural network greatly reduces the number of iterations of the original algorithm and effectively improves the efficiency of the algorithm.Secondly,for the task data that has been requested to be offloaded and executed,a resource allocation scheme problem model is constructed at the edge node,and the optimization goal is to minimize the total cost of the task in the offloaded execution of the MEC server.Considering the insufficient computing resources of edge servers in the MEC network system,a heterogeneous MEC problem model in the 5G scenario is established.Deployed on the system is a single mobile edge application,and the goal of the application is to reasonably allocate its resources to ensure that more offload tasks can be processed in the MEC system as soon as possible,and to ensure the lowest total cost of task execution.By establishing a Markov decision process,this paper introduces the concept of task execution priority,and uses deep reinforcement learning to solve this problem.In addition,in the specific solution,it is proposed to add a filter layer at the end of the Q-network to filter the state-action values corresponding to invalid actions.The simulation results show that deep reinforcement learning can effectively solve the problem model proposed in this paper.