蜂窝基站中基于迁移强化学习的网络节能方案

Network Energy Conservation Scheme Based on Transfer Reinforcement Learning in Cellular Base Stations

近年来,数据密集型应用在通信网络中呈指数级增长,造成通信网络的负载日益增加,保证网络服务质量需要部署大量的基站,从而消耗了大量的能源。如何进行节能增效,是运营商面临的挑战之一。该文提出了一种基于强化学习(Reinforcement Learning,RL)进行动态区域划分的基站切换方案,以有效提高蜂窝网络的能效。此外,通过迁移学习利用先前估计的流量统计数据,进一步提高能源节约并加快学习过程。通过相关数学分析和仿真结果展示了所提出框架的优越性。与传统的基站切换方案相比,所提出的框架为低到中等负载的蜂窝网络降低了约40%的平均能耗。

In recent years,data-intensive applications have grown exponentially in communication networks,resulting in an increasing workload.In order to ensure network service quality,a large number of base stations need to be deployed,which consumes a lot of energy.How to save energy and increase efficiency is one of the challenges faced by operators.This article proposes a base station switching scheme based on reinforcement learning(RL)for dynamic region partitioning to effectively improve the energy efficiency of cellular networks.In addition,utilizing previously estimated traffic statistics through transfer learning can further improve energy conservation and accelerate the learning process.The superiority of the proposed framework was demonstrated through relevant mathematical analysis and simulation results.Compared with traditional base station switching schemes,the proposed framework reduces average energy consumption by about 40%for cellular networks with low to medium loads.