摘要
近年来,国家电网积极推动泛在电力物联网的建设,以实现电力系统的万物互联与优化管理。其中,射频识别技术(RFID)作为泛在电力物联网建设的核心技术,凭其价格低廉、无需电源、非视距通信、远距离通信等特点,被广泛应用于电力仓储物资管理、电力巡检等应用场景。为了盘点电力设备仓库中的物品,需要快速识别粘贴在物品上的标签,然而由于仓库中存在大量标签,在通信过程中容易产生标签信号冲突。针对当前商用RFID系统普遍采用的符合EPCC1G2标准的动态帧时隙ALOHA协议,提出了一种新型的基于Q-learning与神经网络的帧长优化算法(记作QN-learning)。通过将动态帧长选择问题转化为马尔可夫决策过程(MDP),即观察到的状态为不同种类时隙的个数(空时隙数、单时隙数、冲突时隙数),执行的动作为设置合理的帧长,从而利用Q-learning与神经网络来自主学习帧长选择策略,基于学习到的策略可以指导系统根据最新观察选择能够实现全局最优的帧长。仿真实验结果表明,基于QN-learning算法在动态调整帧长方面表现优异,能够实现标签的有效识别,在保障高吞吐率的同时控制阅读器的询问次数,减少数据传输量。
In recent years,the State Grid has actively promoted the construction of ubiquitous power Internet of things,so as to realize the interconnection and optimized management of things in the power system.Specifically,radio frequency identification(RFID)is the core technology for the construction of ubiquitous power Internet of things.Due to the advantages such as low-cost,battery-less,non-line-of-sight communication and long-distance communication,RFID has been widely used in the power equipment management,the power inspection,and other applications.To inventory the items in the power equipment warehouse,the ID collection requires the fast tag identification.However,there are usually a large number of tags in the warehouse,and the signals from different tags will easily conflict with each other.Considering the dynamic framed ALOHA protocol conforming to EPC C1G2 standards in commodity RFID systems,this paper proposes a frame size adjustment algorithm based on Q-learning and neural network(denoted as QN-learning).The problem of adjusting the frame size can be modeled as the Markov decision process(MDP),the observed states are the number of different kinds of slots,i.e.,empty slot,single slot and collision slot,and the actions correspond to the selected frame sizes.Therefore,the neural network-based Q-learning,named as QN-learning,is preferred to learn how to adjust the frame size adaptively.Referring to the learned strategy,the agent is able to select the global-optimal frame size with the latest observation.Simulation results show that the proposed QN-learning-based method performs well in terms of the frame size adjustment.Particularly,the QNlearning-based method can identify tags fast with high throughput and few query rounds,and it reduces the data transmission as well.
作者
何金栋
卜艳玲
石聪聪
谢磊
HE Jindong;BU Yanling;SHI Congcong;XIE Lei(State Grid Fujian Electric Power Research Institute,Fuzhou 350007,China;State Key Laboratory for Novel Software Technology,Nanjing University,Nanjing 210023,China;State Grid Key Laboratory of Information&Network Security,Global Energy Interconnection Research Institute Co.,Ltd.,Nanjing 210023,China)
出处
《计算机科学与探索》
CSCD
北大核心
2021年第7期1227-1236,共10页
Journal of Frontiers of Computer Science and Technology
基金
国家自然科学基金(61872174)。
