Research On Task Migration Edge Computing Offloading Strategy Based On Reinforcement Learning

Along with the growth of network digital technology,the use of Smart Mobile Devices(SMDs)has become widespread.At the same time,unmanned driving,artificial intelligence and other technologies are becoming increasingly mature,but the battery capacity,processing power and other aspects of smart equipment can not keep with the growing demands of people to perform computation-intensive applications.In order to reduce the burden of SMDs and prolong its service life,researchers have proposed Mobile Cloud Computing(MCC)and Mobile Edge Computing(MEC)to solve the problems.MEC enables devices to finish the task offloading at the edge of the network node,greatly reducing the communication bandwidth and computing delay and has high security.However,the computing resources and storage space of edge cloud servers are limited,how to make reasonable and efficient offloading decision is very important.Simultaneously,through recent years,it is a major trend to consider the mobility of users in mobile edge computing research.This paper studies the offloading strategy of mobile edge system,considering task dependency and user mobility.The major work and contributions of this thesis are as follows:1.This thesis summarizes the mobile edge computing offloading strategy framework and the application of artificial intelligence technology in edge computing,introduces the main research direction of edge computing and analyzes the impact of user mobility and task dependency on computing offloading decision.2.For single user,this paper considers the dependency between tasks and the impact of migration cost on offloading decision and propose a single user Computing Offloading based on Q-learning(COQL)algorithm.The task offloading process is described as Markov Decision Process(MDP).The effects of energy consumption and time delay on offloading decision are considered evenly.A loss function is constructed to measure user quality and an optimization problem to minimize the sum of loss functions is proposed.For the sake of solving the thorny problem,we decompose the problem into two subproblems.To find the optimum path to minimize the loss of the system.The simulation results show that the proposed scheme can obviously reduce system total energy consumption and time delay.3.Aiming at the problem of multi-user cooperative mobile edge computing offloading,a COQL algorithm based on user classification is presented to solve the problem.Users are divided into two types from the perspective of mobility:low mobility users and medium-high mobility users.Considering that for users with low mobility,the proposed solution will increase the avoidable energy consumption of the system.An interrupt mechanism is put forward for low mobility users,namely the user terminal sends interrupt instruction to stop the current task execution status and recalculate.If user still makes the offloading decision,it will not consider the user’s data migration,thus will be more accord with the reality communication scenarios.The simulation analysis shows that the COQL algorithm based on user classification can obviously improve the system performance.