基于深度强化学习的多用户蜂窝网络能效优化  

Energy efficiency optimization of multi-user cellular networks based on deep reinforcement learning

作　　者：徐钰龙 李君 李正权[3] 高伟栋 XU Yu-long;LI Jun;LI Zheng-quan;GAO Wei-dong(College of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;College of Electronic and Information Engineering,Wuxi University,Wuxi 214105,China;State Key Laboratory of Network and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]无锡学院电子信息工程学院,江苏无锡214105 [3]北京邮电大学网络与交换技术国家重点实验室,北京100876

出　　处：《计算机工程与设计》2025年第3期734-740,共7页Computer Engineering and Design

基　　金：网络与交换技术全国重点实验室(北京邮电大学)开放课题基金项目(SKLNST-2023-1-13)。

摘　　要：针对多用户蜂窝网络中能量效率的重要性以及传统优化算法的局限性和泛化性能差的问题,提出一种基于深度强化学习的EEO-Dueling DQN算法,旨在满足约束发射功率条件下实现整个网络的能量效率最大化。Dueling DQN采用竞争网络优化神经网络结构解决DQN中出现的高估问题。仿真结果表明,该算法获得的平均能量效率比DQN算法高出65%,在收敛情况和稳定性方面也有较好表现,具有较强泛化能力,可适用于实际中不同通信场景。Considering the significance of energy efficiency in multi-user cellular networks and the limitations and poor generalization performance of traditional optimization algorithms,an EEO-Dueling DQN algorithm based on deep reinforcement learning was proposed to maximize the energy efficiency of the entire network under constrained transmission power conditions.A competitive network was employed in Dueling DQN to optimize the neural network structure,effectively resolving the issue of ove-restimation present in DQN.Simulation results indicate that the proposed algorithm achieves an average energy efficiency that is 65% higher compared to the DQN algorithm.It exhibits favorable performance in terms of convergence and stability,along with strong generalization capabilities,making it suitable for various communication scenarios in practical applications.

关 键 词：多用户 蜂窝网络 深度强化学习 神经网络 竞争网络 能量效率 泛化性能 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]