Research On MEC Task Migration Strategy Based On Reinforcement Learning

The number of future iot devices is expected to grow exponentially,with global IP traffic expected to grow to 400 exabytes per month by 2024.Mobile Edge Computing(MEC)can effectively solve the problem of massive node data processing of Internet of Things by sinking the Mobile Edge server closer to the Internet of Things devices and providing task Computing and data storage services for nearby devices to reduce network latency.Able to cope with the rapid growth of Internet of Things devices.As the computing resources of cloud server and mobile edge server are limited,how to make full use of their own resources for optimal allocation is an urgent problem to be solved at present.Firstly,Reinforcement Learning(RL)is used to guide the resource allocation of mobile edge server,which can find the best solution faster than traditional model-based technology and release the maximum potential of mobile edge server.Secondly,considering the security of user information,the computing power of federated learning fusion cloud server and mobile edge server is adopted to ensure the security of user information while training the model.Finally,considering the mobility of end users,the reinforcement learning algorithm suitable for dealing with continuous actions can provide the optimal decision for end users efficiently and accurately.The research objective of this paper is to jointly optimize dynamic task transfer and resource allocation based on reinforcement learning in resource-constrained Internet of Things scenarios.The research work of this paper is summarized as follows:(1)To solve the problem of user information security,a task unloading strategy based on user information security is proposed in the scenario of multi-user single cell.Firstly,the single cell system model of multi-user is established.According to the user’s different task unloading strategy,the calculation formula of total energy consumption and total delay of the system is derived,the definition of total cost function is given,and the optimization of total cost function is the optimization goal.Secondly,the multi-level linkage algorithm based on federated reinforcement learning is used to train the decision model and the local weight model.The user converts the information used for training the decision model into model parameters,and replaces the real information with model parameters to train the local weight model.Finally,the simulation results show that the proposed algorithm can significantly reduce the total cost of the system,while protecting the user information security,reducing the user’s own computing pressure,and improving the user’s service quality.At the same time,the research content in this section also provides theoretical support for the multi-cell cell scenario considering user mobility in(2).(2)In view of the problem that task calculation cannot be completed due to mobile edge server service coverage caused by mobile edge users’movement,in the multi-user multi-cell scenario,the conclusion obtained in(1)is applied as the support,and the strategy of joint optimization task migration and resource allocation is proposed.Firstly,a multi-user multi-cell system model is established,and the total cost function of the current system model is defined to optimize the total cost function of the system.Secondly,a task transfer strategy based on DDPG algorithm and a task transfer strategy based on federated reinforcement learning are proposed.Finally,the simulation results show that compared with DDPG algorithm,the multi-level linkage algorithm based on federated reinforcement learning can make full use of the computing resources of MEC server and carry out reasonable task migration for user computing tasks.When computing resources are limited,the task allocation decision reduces the total cost of the system and improves the quality of service for users.