Research On Multiple Objective Task Offloading Strategy Based On Deep Reinforcement Learning In Mobile Edge Computing Environment

With rapid development of the mobile Internet,the growing number of mobile users leads to the result that traditional network architecture can't maintain a well service quality.In order to improve edge network service quality,mobile edge computing is proposed to solve above problem.Through setting server at edge network,mobile users can nearby get service that can reduce network latency and improve mobile network service quality.An important problem in mobile edge computing is how to efficiently offload tasks.Most existing work focus on independent task offloading in mobile edge computing.However,these methods are not applicable to workflow task offloading due to the presence of temporal dependencies and data dependencies between workflow tasks.In view of this,this research project focuses on a multi-objective workflow task offloading mechanism in mobile edge computing,with the goal of minimizing the overall completion time of the workflow and the total energy consumption of the user equipment.Specifically,the main contributions of this paper are as follows:(1)In this paper,the multi-workflow offloading problem in the mobile edge computing environment is modeled as a multi-objective optimization problem.The optimization problem comprehensively considers the overall completion time of the workflow and the total energy consumption of the user equipment.By exploring the relationship between the completion time of the workflow and the energy consumption of the user equipment,an effective compromise can be made between the completion time of the workflow and the energy consumption of the user equipment.(2)This paper transforms the constructed optimization problem into a partially observable Markov game model.And a multi-workflow offloading algorithm based on multi-agent deep reinforcement learning is designed and implemented to obtain the optimal strategy of the partially observable Markov game model.The algorithm adopts the way of centralized training and distributed execution.For each agent,train a critic that needs global information and an actor that needs local information.The optimal offloading strategy is obtained by coordinating the activities among multiple agents.(3)This paper evaluates the effectiveness and advancement of the proposed multiobjective workflow task offloading algorithm through a large number of simulation experiments.The experimental results show that the proposed algorithm improves the average reward value by 50% compared with the single-agent deep deterministic policy gradient algorithm.Compared with other advanced offloading algorithms,the proposed algorithm performs best in minimizing the overall completion time of the workflow and the total energy consumption of the user equipment.