摘要
网络边缘计算多级卸载模型进行卸载决策执行时,决策结果非最优结果,导致边缘计算服务器无法在规定时间内执行完所有的计算任务,对此,研究基于深度强化学习的网络边缘计算多级卸载模型,构建网络边缘计算多级卸载模型,并分析卸载模型执行所有计算任务所耗用的能量和时间;结合强化学习的决策能力和深度学习的感知能力得到强化学习算法,将强化学习算法与实际边缘计算环境相结合,构建马尔科夫决策过程模型,得到调度计算任务的最优决策结果。实验结果表明,当深度Q网络的经验池大小为512、学习速率为0.002时,该模型的收敛性能最佳;该模型可实现边缘计算任务的多级卸载,且在计算任务传输速度和边缘计算服务器中CPU运行速度不同时,仍具有良好适应性。
When the multi-level unloading model of network edge computing executes the unloading decision,the decision result is not the optimal result,which causes the edge computing server to be un⁃able to complete all computing tasks within the specified time.In this regard,we study the multi-level unloading model of network edge computing based on deep reinforcement learning,build a multi-level unloading model of network edge computing,and analyze the energy and time consumed by the unload⁃ing model to perform all computing tasks.The reinforcement learning algorithm is obtained by combin⁃ing the decision-making ability of reinforcement learning and the perception ability of deep learning.Combining the reinforcement learning algorithm with the actual edge computing environment,the Mar⁃kov decision process model is constructed to obtain the optimal decision result of scheduling computing tasks.The experimental results show that the convergence performance of the deep Q network is opti⁃mal when the experience pool size is 512 and the learning rate is 0.002.This model can realize multilevel offloading of edge computing tasks,and has good adaptability when the transmission speed of com⁃puting tasks is different from the CPU running speed in the edge computing server.
作者
刘敏
LIU Min(Information Center,Meizhouwan Vocational Technology College,Putian Fujian 351119,China)
出处
《保山学院学报》
2023年第5期67-74,共8页
JOURNAL OF BAOSHAN UNIVERSITY
基金
2019年度福建省中青年教师科研项目“基于SDN的网络运维管理研究与应用”(项目编号:JAT191930)。
关键词
深度强化学习
无线网络
边缘计算
多级卸载模型
边缘计算服务器
马尔科夫决策
Deep reinforcement learning
Wireless network
Edge computing
Multi level unloading model
Edge computing server
Markov decision
