基于深度强化学习的IRS辅助无线传感网充电策略研究  

Research on charging strategy of IRS-assisted wireless sensor networks based on deep reinforcement learning

作　　者：白义倩 刘韬[1] 杨晋 张亮[1] BAI Yiqian;LIU Tao;YANG Jin;ZHANG Liang(School of Computer Science&Engineering,Southwest Minzu University,Chengdu 610200,China)

机构地区：[1]西南民族大学计算机科学与工程学院,四川成都610200

出　　处：《传感器与微系统》2025年第1期11-16,共6页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(62171390);西南民族大学研究生创新型科研项目(YCZD2024001)

摘　　要：无线可充电传感器网络(WRSNs)广泛应用于许多领域,然而,传感器节点的电池容量有限阻碍了其发展。借助无线能量传输技术,引入智能反射面(IRS)对传感器节点进行无线充电已成为延长WRSNs寿命的一项有前景的技术。但在大规模WRSNs环境中,传统强化学习方法会遇到维度灾难而导致学习效率低下,于是提出了一种无模型深度强化学习(DRL)的WRSNs能量传输方案。首先,研究如何对IRS的相位偏移进行优化,来补充传感器能量供应的问题,使得目标节点处的接收功率最大化;基于上述相移优化,在WRSNs中以IRS作为智能体,结合传感器实时状态,设计基于DRL的高效充电算法,采用不同的双网络结构解决传统DRL算法会存在的过估计问题,克服大规模网络下强化学习计算量过大和学习效率下降的问题。仿真结果表明:该算法能够达到显著降低节点失效率并最大程度地延长WRSNs生命周期的目的。Wireless rechargeable sensor networks(WRSNs)are widely used in many fields,however,the limited battery capacity of sensor nodes hinders their development.The introduction of intelligent reflective surface(IRS)for wireless charging of sensor nodes with the help of wireless energy transfer technology has become a promising technique to extend the lifetime of WRSNs.However,in large-scale WRSNs environments,traditional reinforcement learning methods may encounter the problem of dimensional catastrophe,resulting in inefficient learning,so a model-free deep reinforcement learning(DRL)scheme for WRSNs energy transmission is proposed.Firstly,study how to optimize the phase shift of the IRS to supplement the sensor energy supply,so as to maximize the received power at the target node;based on the above phase shift optimization,use the IRS as an intelligent body in WRSNs,combined with the real-time state of the sensors,design an efficient charging algorithm based on DRL,and use different dual-network structure to solve the over-estimation problem that would exist in traditional DRL algorithms,and overcome the problem of excessive computation and reduced learning efficiency of reinforcement learning under large-scale network.Simulation results show that the algorithm can significantly reduce node failure rate and extend the life cycle of WRSNs to an extreme.

关 键 词：无线传感器网络 智能反射面 相位偏移优化 深度强化学习 双网络结构 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]