Research On Task Offloading Algorithm Based On DDPG In Mobile Edge Computing

With the development of the 5th generation communication technology(5G),new applications such as autonomous driving,augmented reality and virtual reality put forward high requirements for the performance of the equipment.It is difficult for user equipment to meet the computing needs of these applications only by expanding hardware.Therefore,the industry proposes mobile edge computing(MEC)as a solution to solve the above problems.In MEC,the user equipment can offload tasks to the server located at the edge of the mobile network for execution,so as to reduce the task processing delay and the energy consumption of the user equipment.However,the resources of the edge server are limited and can’t meet the offloading needs of all tasks.In order to optimize the performance of MEC task offloading,this paper studies the offloading decision and resource allocation in the resource constrained MEC environment,and gives a specific scheme combined with the deep deterministic policy gradient(DDPG).The work in this paper can be summarized as follows:(1)Aiming at the MEC scenario of multi-user equipment and single edge server,a DDPG based task offloading decision and resource allocation algorithm called OADDPG is proposed.In this environment,considering the resource constraints of the edge server and the tasks can be processed,the system is modeled with the weighted total cost composed of user equipment energy consumption and task processing delay and the success rate of task execution as the optimization goal.We transform the established MEC model into a Markov decision process,then design reasonable state space,action space and reward function according to the optimization objectives.OADDPG can make offloading decisions and allocate edge server computing resources for tasks with changing attributes.In the training process,it can continuously improve the accuracy of the strategy by updating the network,so as to maximize the cumulative reward and obtain the lowest total cost and the highest success rate.Experimental results show that OADDPG algorithm can effectively reduce the total cost of user equipment and improve the success rate of task execution compared with other three baseline algorithms.(2)Aiming at the MEC scenario with mobility of user equipment,a task offloading decision and resource allocation algorithm called PREDDPG based on improved DDPG is proposed.In this environment,considering the impact of the different location of user equipment in each time slot on the information transmission rate,a dynamic MEC task offloading model is constructed to minimize the weighted total cost composed of user equipment energy consumption and task processing delay,and maximize the success rate of task execution.In order to further improve the performance of the algorithm,the DDPG algorithm is improved by using the prioritized experience replay(PER).Finally,the proposed PERDDPG algorithm is used to make offloading decisions and computing resource allocation schemes for more dynamic MEC scenarios.Simulation results show that PERDDPG algorithm can accelerate the convergence speed and improve the cumulative reward,effectively reduce the weighted total cost of mobile devices and improve the success rate of task execution.To sum up,the two task offloading decisions and resource allocation algorithms proposed in different MEC scenarios effectively reduce the total cost of user equipment and improve the success rate of task execution,which have a certain contribution to the research of task offloading strategy in MEC.