首页 | 官方网站   微博 | 高级检索  
     

基于深度双Q网络的多用户蜂窝网络功率分配算法研究
引用本文:王伟.基于深度双Q网络的多用户蜂窝网络功率分配算法研究[J].计算机应用研究,2021,38(5):1498-1502.
作者姓名:王伟
作者单位:辽宁工程技术大学基础教学部,辽宁葫芦岛125105;辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105
基金项目:国家自然科学基金青年基金资助项目(61701211,61806186);福建省智能物流产业技术研究院建设项目(2018H2001);机器人与系统国家重点实验室(HIT)项目(SKLRS-2019-KF-15);辽宁省教育厅科学技术研究项目(LJCL008);泉州市科技计资助划项目(2019C112,2019C011R,2019STS08);新兴产业集成化检验检测服务平台研发与应用项目(2018YFB1403303)。
摘    要:针对现有蜂窝网络功率分配算法存在泛化能力弱、效率低等问题进行了研究,提出基于深度双Q网络(deep double Q network,DDQN)的功率分配算法。采用双神经网络结构,解决强化学习过程中易出现的维度灾难及值函数过估计问题;对状态信息进行设计并输入神经网络,输出智能体的动作行为,并设计奖赏函数反馈给神经网络,使智能体可以有效地自主学习,多次迭代得到最优的功率分配策略。仿真结果表明,所提的模型可获得的平均速率为1.89,平均运行时间为0.0013 s,在不同用户密度及小区数量下均可达到最高的平均速率,验证了算法的有效性,为蜂窝网络资源分配问题提供了新的思路。

关 键 词:蜂窝网络  干扰多址信道  功率分配  深度强化学习  深度双Q网络
收稿时间:2020/7/27 0:00:00
修稿时间:2021/4/12 0:00:00

Multi-user cellular network power allocation based on deep double-Q network
Wang Wei,Yin Shuangshuang.Multi-user cellular network power allocation based on deep double-Q network[J].Application Research of Computers,2021,38(5):1498-1502.
Authors:Wang Wei  Yin Shuangshuang
Affiliation:(Foundation Dept.,Liaoning Technical University,Huludao Liaoning 125105,China;School of Electronic&Information Engineering,Liaoning Technical University,Huludao Liaoning 125105,China)
Abstract:In view of the weak generalization ability and low efficiency of the existing cellular network power allocation algorithm,this paper proposed the power allocation algorithm based on the deep double Q network.It adopted the dual neural network structure to solve the problem of dimensional disasters and value function overestimation that were easy to occur in the reinforcement learning process.It designed the state information and inputted it to the neural network,outputted the action behavior of the agent,and designed the reward function to feed back to the neural network to make the agent can effectively learn independently and obtain the optimal power allocation strategy through multiple iterations.The simulation results show that the average speed that the proposed model can obtain is 1.89,and the average running time is 0.0013 s.The highest average speed can be reached under different user densities and cell numbers,which verifies the effectiveness of the algorithm.The problem of network resource allocation provides new ideas.
Keywords:cellular network  interference multiple access channel  power allocation  deep reinforcement learning  deep double Q network
本文献已被 维普 万方数据 等数据库收录!
点击此处可从《计算机应用研究》浏览原始摘要信息
点击此处可从《计算机应用研究》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号