Task Offloading And Resource Allocation Algorithm Based On Actor-Critic Framework In Mobile Edge Computing

As an emerging computing paradigm,Mobile Edge Computing(MEC)provides new ideas for resource-constrained devices to handle compute-intensive and delay-sensitive tasks,which deploys servers in the network at the edge to bring cloud functions closer to user devices.However,the time-varying feature of computation and communication resources in MEC and the randomness of user devices make it difficult to generate an optimal computation offloading policy.Furthermore,most of the existing studies consider multi-user single-edge server scenarios,which suffer from the deficiencies of single model and low practicality in describing complex realistic edge networks.To solve the above problems,this paper designs computation offloading and resource allocation strategies under a three-layer network architecture consisting of cloud-edge-user for multi-user multi-edge server scenario.The main research contents are summarized as follows:(1)In the multi-user multi-edge server scenario,this paper studies binary task offloading decision and the problem of resource allocation.A cost function for computation is constructed based on factors such as computing resources,task execution delay,and energy consumption.Then,a multi-objective joint optimization model,which is a mixed-integer nonlinear programming problem with NP-hard properties and difficult to solve directly,is formulated under relevant constraints.On the basis of this,the HAS-DDPG algorithm is proposed.Specifically,the proposed algorithm transforms the objective function into a two-layer optimization problem,i.e.,the upper layer utilizes the S-ACS algorithm that fuses prioritization technology and ant colony algorithm,to generate computational offloading decisions,then based on the upper layer decisions,the lower layer uses the ODGR algorithm that combines the deep deterministic policy gradient algorithm with prioritized experience replay technology to achieve optimal resource allocation.Finally,the experimental results show that the HAS-DDPG algorithm performs better in binary offloading compared with other baseline algorithms.(2)The problem of partial offloading decision for compute-intensive applications is more complex in the edge computing scenario with multiple competitive edge servers.In this problem,it is characterized by the large amount of input data for computational tasks and the frequent dependencies between multiple tasks in the application.Therefore,this paper firstly describes the dependencies using directed acyclic graph and solves the scheduling order between tasks based on the latest execution time of the application.Then,a task computation cost function for the user device is established to construct the model for multi-objective joint optimization under the relevant constraints.A Markov decision process is built on this model to translate the constraints into system states and continuously explore.Finally,the optimal unloading decision and resource allocation problem under the minimization cost function are solved by using the MP-SAC algorithm fused with the prioritized experience replay technique and the Soft ActorCritic algorithm.The simulation experiments show that the MP-SAC algorithm has better stability and solution efficiency compared with other baseline algorithms.