边缘计算动态资源分配在通信网络时延优化的研究  

Edge Computing Dynamic Resource Allocation in Communication Network Delay Optimization

作　　者：贺承启 HE Chengqi(Guiyang Sangli Electronic Industry Co.,Ltd.,Guiyang 550000,China)

机构地区：[1]贵阳桑力电子实业有限公司,贵州贵阳550000

出　　处：《通信电源技术》2024年第6期179-181,共3页Telecom Power Technology

摘　　要：针对边缘计算动态资源分配的通信网络时延优化问题,提出一种基于深度强化学习的算法。该算法建立一个综合考虑边缘节点的异构性、地理分散性、能量因果性等约束条件,以及时延、能耗、费用等多目标优化指标的数学模型,充分反映边缘计算网络的复杂性和多样性。该算法使用多层感知器作为策略函数,通过策略梯度方法和随机梯度下降法进行动态资源分配决策,实现通信网络时延的最小化。仿真实验结果表明,该算法在不同场景和参数下均能有效降低通信网络时延,提高资源利用效率和系统性能。与固定资源分配算法、贪心资源分配算法以及基于遗传算法的资源分配算法相比,具有明显的优势,尤其是在处理边缘计算网络的动态性和不确定性方面。基于深度强化学习的动态资源分配算法为边缘计算资源管理问题提供一种新的解决方案,也为通信网络时延优化提供有力的支持。A deep reinforcement learning-based algorithm is proposed for the communication network delay optimization problem with dynamic resource allocation for edge computing.The algorithm establishes a mathematical model that comprehensively considers the constraints of heterogeneity,geographic dispersion,and energy causality of edge nodes,as well as multi-objective optimization metrics such as delay,energy consumption,and cost,to fully reflect the complexity and diversity of edge computing networks.The algorithm uses a multilayer perceptron as a policy function,and makes dynamic resource allocation decisions through the policy gradient method and stochastic gradient descent method to minimize the communication network delay.Simulation experimental results show that the algorithm can effectively reduce the communication network delay and improve the resource utilization efficiency and system performance under different scenarios and parameters,and has obvious advantages over fixed resource allocation algorithms,greedy resource allocation algorithms,and resource allocation algorithms based on genetic algorithms,especially in dealing with the dynamics and uncertainty of edge computing networks.The dynamic resource allocation algorithm based on deep reinforcement learning provides a new solution to the edge computing resource management problem,and also provides strong support for communication network delay optimization.

关 键 词：边缘计算 动态资源分配 通信网络时延优化 深度强化学习 多目标优化 

分 类 号：TN929.5[电子电信—通信与信息系统]