一种深度强化学习的C-RAN动态资源分配方法  

C-RAN Dynamic Resource Allocation Method for DRL

作　　者：张永棠 ZHANG Yong-tang(School of Computer,Guangdong Neusoft Institute,Foshan 528225,China;Institute of Cooperative Sensing and Advanced Computing Technology,Nanchang Technology Institute,Nanchang 330003,China)

机构地区：[1]广东东软学院计算机学院,广东佛山528225 [2]南昌工程学院江西省协同感知与先进计算技术研究所,南昌330003

出　　处：《小型微型计算机系统》2021年第1期132-136,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61663029)资助;广东省高校重点平台与特色创新项目(2017KTSCX200)资助。

摘　　要：移动边缘计算(MEC)技术已成为云无线接入网(C-RAN)提供近距离服务的一个很有前途的例子,从而减少了服务延迟,节约了能源消耗.本文考虑一个多用户MEC系统,解决了计算卸载策略和资源分配策略问题.我们将延迟总成本和能耗作为优化目标,在一个动态的环境中获得一个最优的策略.提出了一个基于深度强化学习的优化框架来解决资源分配问题,利用深度神经网络(DNN)对批评者的价值函数进行估计,从当前状态直接提取信息,不需要获取准确的信道状态.从而降低了优化目标的状态空间复杂度.参与者使用另一个DNN来表示参数随机策略,并在批评者的帮助下改进策略.仿真结果表明,与其它方案相比,该方案显著降低了总功耗.Mobile edge computing(MEC)technology has emerged as a promising example of Cloud radio access networks(C-RAN)providing near distance services,thereby reducing service latency and reducing energy consumption.In this paper,a multi-user MEC system is considered to solve the problem of computing unload policy and resource allocation policy.We took deferred total cost and energy consumption as optimization objectives to obtain an optimal strategy in a dynamic environment.An optimization framework based on Deep reinforcement learning(DRL)is proposed to solve the problem of resource allocation.Deep neural network(DNN)is used to estimate the value function of critics,and the information is directly extracted from the current state without the need to obtain accurate channel state.Thus,the state space complexity of the optimization target is reduced.The participants used another DNN to represent the parametric random strategy and improved the strategy with the help of the critics.The simulation results show that compared with other schemes,this scheme significantly reduces the total power consumption.

关 键 词：云无线接入网 移动边缘计算 深度神经网络 深度强化学习 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]