基于强化学习的异构CPU环境网络资源部署方法  

Network Resource Deployment Method in Heterogeneous CPU Environment Based on Reinforcement Learning

作　　者：王秋琳 梁懿 刘迪 董昌朝 董晓祺 WANG Qiu-lin;LIANG Yi;LIU Di;DONG Chang-chao;DONH Xiao-qi(Fujian Yirong Information Technology Co.,Ltd.,Fuzhou 350003 China)

机构地区：[1]福建亿榕信息技术有限公司,福建福州350003

出　　处：《自动化技术与应用》2024年第11期140-144,173,共6页Techniques of Automation and Applications

基　　金：国家电网公司科技项目(5419-202034209A-0-0-00)。

摘　　要：针对现有方法进行异构CPU网络资源部署后存在的任务迟延长、资源处理效率低等问题,提出基于强化学习的异构CPU环境网络资源统一部署方法。先对Q-Learning强化学习及深度Q学习网络DQN的特点进行分析,通过更新深度神经网络各层权值、进行强化学习训练以及设置目标网络等方式更新Q值,构建异构多核处理器CPU资源调度模型,将其用来调度网络资源,并利用DQN网络确定最佳资源部署策略,实现网络资源统一部署。实验结果表明:该方法可实现异构多核CPU负载均衡,且该方法部署后的任务迟延少、执行时间短,资源处理效率突出,具有一定应用价值。Aiming at the problems of task delay and low resource processing efficiency after the existing methods deploy heterogeneous CPU network resources,a unified resource deployment method based on reinforcement learning is proposed to ensure resource load balance and improve resource processing efficiency.Firstly,the characteristics of Q-learning reinforcement learning and Deep Q learning Network(DQN)are analyzed.By updating the weights of each layer of the deep neural network,carrying out reinforcement learning training and setting the target network,the Q value is updated,and the heterogeneous multi-core processor(CPU)resource scheduling model is constructed,which is used to schedule network resources,and the dqn network is used to determine the best resource deployment strategy to realize the unified deployment of network resources.The experimental results show that this method can achieve heterogeneous multi-core CPU load balancing,and the method has less task delay,short execution time and outstanding resource processing efficiency after deployment,so it has certain application value.

关 键 词：异构CPU 深度Q学习网络 神经网络 网络资源部署 网络权值 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN929.5[自动化与计算机技术—控制科学与工程]