边缘计算中工业任务卸载调度与资源分配算法研究

Research on Task Offloading Scheduling and Resource Allocation Algorithms in Edge Computing for Industrial Applications

移动边缘计算(MEC)将具有计算和存储等功能的服务器部署在网络边缘以满足某些对时延要求苛刻的任务。针对工业场景中任务处理实时性的要求,本文考虑多用户有向无环图(DAG)任务的卸载决策、通信带宽和计算资源在线分配问题,通过构建马尔可夫决策过程模型,并采用基于强化学习离散动作空间的DQN和连续动作空间的TD3网络协同优化DAG节点任务二进制卸载决策和带宽计算资源的分配,以最大化长期实时任务卸载成功率。仿真结果表明,本文采取DQN+TD3算法的实时任务卸载成功率最高,验证了算法有效性。

Mobile edge computing(MEC)deploys servers with capabilities functions such as computation and storage at the edge of the network to satisfy certain tasks with demanding latency requirements.In order to meet the requirements of realtime task processing in industrial scenarios,we consider the offloading decisions of multi-user Directed Acyclic Graph(DAG)tasks,communication bandwidth and online allocation of computational resources.By constructing the Markov decision process model,this paper adopted DQN based on reinforcement discrete action space collaborating with TD3 based on continuous action space network to optimize the binary task offloading decision of DAG nodes and bandwidth computational resources allocation,aiming to maximize the success rate of long-term real-time task offloading.The simulation results show that the DQN+TD3 algorithm adopted in this paper has the highest success rate of real-time task offloading,which verifies the effectiveness of the algorithm.