基于深度强化学习的协作通信中继选择  

Collaborative Communication Relay Selection Based on Deep Reinforcement Learning

作　　者：胡文杰 钟良骥[3] HU Wenjie;ZHONG Liangji(School of Computer Science and Technology,Huazhong University of Science and Technology,Wuhan 430074,China;School of Information Engineering,Xianning Vocational and Technical College,Xianning 437100,China;School of Computer,Hubei University of Science and Technology,Xianning 437100,China)

机构地区：[1]华中科技大学计算机科学与技术学院,武汉430074 [2]咸宁职业技术学院信息工程学院,湖北咸宁437100 [3]湖北科技学院计算机学院,湖北咸宁437100

出　　处：《电讯技术》2020年第12期1425-1431,共7页Telecommunication Engineering

基　　金：湖北省教育厅科学研究项目(B2018487);咸宁市科技局科技项目(201854);咸宁职业技术学院校级科研项目(2017yjd011)。

摘　　要：协作通信是无线传感器网络(Wireless Sensor Network,WSN)实现数据可靠传输的关键技术,而协作通信技术的关键在于中继方案的选择。为此,提出了一种基于深度强化学习的协作通信中继选择算法(Deep Q-Learning Based Relay Selection Scheme,DQ-RSS)。首先,将WSN中具有中继选择的协作通信过程建模为马尔科夫决策过程,并采用Q学习在未知网络模型的情况下获取最佳中继选择策略;其次,针对高维状态空间下Q学习收敛时间长的问题,采用DQN(Deep-Q-Net)算法来加速Q学习的收敛。对比仿真实验结果表明,DQ-RSS在中断概率、系统容量和能耗方面均优于现有的中继选择方案,且能够有效节省收敛时间。Cooperative communication is the key technology for wireless sensor network(WSN)to achieve reliable data transmission,and the key to cooperative communication technology is the choice of relay scheme.Therefore,a cooperative communication relay selection algorithm based on deep reinforcement learning called DQ-RSS is proposed.First,the cooperative communication process with relay selection in WSN is modeled as a Markov decision process,and Q-Learning is used to obtain the best relay selection strategy in the case of unknown network models.Secondly,for the problem of long Q-Learning convergence time in high-dimensional state space,Deep-Q-Net(DQN)algorithm is used to accelerate the convergence of Q-Learning.Comparative simulation experiments show that DQ-RSS is superior to existing relay selection schemes in terms of outage probability,system capacity and energy consumption,and can effectively save convergence time.

关 键 词：无线传感器网络 协作通信 中继选择 Q学习 DQN算法 

分 类 号：TN929.5[电子电信—通信与信息系统]