Research On Task Migration And Offloading Methods Based On Deep Reinforcement Learning In Edge Computing

With the rapid development of Internet of Things(Io T)technology and 5G technology,massive computation-intensive and latency-sensitive tasks are generated,which brings high load and high latency to Mobile Edge Computing(MEC)system.How to migrate and offload tasks to the edge server efficiently,thus improving the quality of user experience(Qo E),is one of the key issues that mobile edge computing needs to address.This thesis combines mobile edge computing with deep reinforcement learning theory to investigate task migration and offloading in mobile edge computing.The main research work of this thesis are as follows:1.In a heterogeneous network scenario with multiple users and multiple MEC servers,this thesis proposes a task co-migration and offloading method based on deep reinforcement learning algorithm named Deep Q-Networks(DQN),which alleviates the system’s high energy consumption and high latency.By taking measures for task co-migration and offloading among MEC servers to minimize the average cost of this system,and combining with deep reinforcement learning theory,the decision of task migration and offloading is constructed as a Markov Decision Process(MDP).In addition,in order to improve the longterm cost estimation of this algorithm,the method further combines the dueling network architecture,Double Deep Q-Networks(DDQN)technology and the prioritized experience replay mechanism.Simulation results show that the proposed method can effectively reduce the energy consumption and latency of MEC server clusters.2.In a heterogeneous network scenario with single user and multiple MEC servers,this thesis proposes a task migration and offloading method based on deep reinforcement learning algorithm named Rainbow DQN,which alleviates the problem of user experience quality declined that caused by user mobility.Firstly,with the goal of maximizing user experience quality,task migration and offloading are performed based on user mobility patterns.By combining the deep reinforcement learning theory,the states,actions,and rewards during task migration and offloading are determined.After the model is learned and trained,the agent can make the optimal migration and offloading decisions based on user mobility patterns.The simulation results verify that the proposed method can effectively improve the quality of user experience.