基于半监督学习的称重衡器传感器干扰源定位  

Interference source localization of weighing scale sensor based on semi supervised learning

作　　者：何开宇 李鹏飞 江浩 石岩 HE Kaiyu;LI Pengfei;JIANG Hao;SHI Yan(Henan Institute of Metrology,Zhengzhou 450008,China)

机构地区：[1]河南省计量测试科学研究院,河南郑州450008

出　　处：《电子设计工程》2025年第7期168-171,176,共5页Electronic Design Engineering

摘　　要：称重衡器传感器的光电耦合器限制了上下级电路的隔离效果,难以根据传感器传输范围划分干扰源影响范围,干扰源定位误差较大。因此,提出基于半监督学习的称重衡器传感器干扰源定位方法。考虑正常传感器节点与被干扰节点的位置分布,设置多个虚拟边界节点,通过拟合划分干扰源的范围。建立基于CS(客户端/服务器)框架的唤醒机制,构建干扰区域传感器观测方程,量化描述干扰信号对传感器输出信号的影响,获取干扰范围内传感器的输出信号。应用MLP神经网络充当基分类器,构造半监督学习网络模型。运用网格搜索方法,求解干扰源估计位置方程,输出干扰源的具体坐标,实现称重衡器传感器的干扰源定位。实验结果表明,应用所提方法得到的干扰源的位置坐标分别为(5.6,9.1)、(17.9,16.4)。当信噪比处于0~30 dB时,干扰源定位结果的均方根误差保持在0.3 m以下,实现了对干扰源位置的准确描述。The photoelectric coupler of the weighing scale sensor limits the isolation effect of the upper and lower circuits,making it difficult to divide the interference source influence range according to the transmission range of the sensor,and the interference source positioning error is relatively large.Therefore,a semi supervised learning based method for locating interference sources in weighing scale sensors is proposed.Consider the position distribution of normal sensor nodes and disturbed nodes,set multiple virtual boundary nodes,and divide the range of interference sources through fitting.Establish a wake-up mechanism based on the CS(Client/Server)framework,construct an observation equation for sensors in the interference area,quantitatively describe the impact of interference signals on sensor output signals,and obtain the output signals of sensors within the interference range.Apply MLP neural network as the base classifier,construct a semi supervised learning network model,use grid search method to solve the interference source estimation position equation,output the specific coordinates of the interference source,and achieve the interference source localization of the weighing scale sensor.The experimental results show that the position coordinates of the interference source obtained by the proposed method are(5.6,9.1)and(17.9,16.4),respectively.When the signal-to-noise ratio is between 0~30 dB,the root mean square error of the interference source localization result remains below 0.3 m,achieving accurate description of the interference source location.

关 键 词：称重衡器传感器 半监督学习 唤醒机制 干扰源范围 MLP神经网络 

分 类 号：TN918.91[电子电信—通信与信息系统]