Research On Task Offloading Model And Strategy In Mobile Edge Environment

With the continuous development and maturity of 5th Generation Mobile Communication Technology(5G),the explosive data traffic generated by multimedia services such as interactive live broadcasting and epidemic prevention and control not only poses challenges to the cache and computing capacity of the Internet of Things(Io T)devices but also has a profound impact on the design and evolution of network architectures.Therefore,Mobile Edge Computing(MEC),as the key to achieving efficient data caching and preprocessing,and improving data access and response,is of great significance for improving user quality of service(Qo S).Due to the limited resources on the edge side,when tasks are highly concurrent,it is necessary to use task offloading to allocate tasks that cannot be processed on the local device to other idle edge nodes(EN).In the face of the high heterogeneity and high time variability of edge resources,how to reasonably allocate computing and network resources in the process of task offloading and determine the best task collaboration strategy has become a major problem.Therefore,this paper will use the Deep Reinforcement Learning(DRL)algorithm to study the task offloading model and strategy in the MEC environment,aiming to strengthen the cooperation between ENs and reduce the overhead of the task processing.The main work of this paper is as follows:(1)Considering the limited and unbalanced resources on the edge side,this paper will study the multi-task offloading optimization problem of a single base station in the MEC environment to minimize the task processing consumption.In the process of task processing,this paper designs two task offloading strategies.In addition,in order to achieve the balanced optimization of delay and energy consumption during task offloading,this paper deploys Double Deep Q Network(DDQN)intelligent agent to indicate the processing of the task.Then,to reduce the transmission cost of a large amount of training data in model training,this paper uses Federated Learning(FL)to train agents in a distributed manner,which not only solves the problem of resource imbalance but also ensures data security.The final simulation experiments show that FL+DDQN can not only achieve the goal of minimizing the task processing overhead,but also has advantages in improving the average utility of the system.(2)Considering the heterogeneity of resources in the MEC scenario,the number of tasks allocated to the edge side,the uncertainty of computing processing time,etc.,this paper designs a cross-base station heterogeneous collaborative edge computing task offloading architecture to strengthen the collaboration between ENs at a different level.On this basis,four task offloading strategies are designed,and multiple parallel Asynchronous Advantage Actor-Critic(A3C)agents are deployed at each node in the scene.The agent automatically learns the offloading strategy based on the observation data to determine the best resource allocation scheme.The final simulation results show that the scheme proposed in this paper is effective in improving the processing capability of the edge side task,reducing the system overhead,maximizing the long-term utility of the system,and improving the successful completion rate of the task.There are 19 figures,5 tables and 100 references.