摘要
基于催化原理的传感器在矿井环境工作时,受内在外在因素的干扰,测量的可燃气体浓度值存在误差过大的问题。设计了基于列文伯格-马夸尔特(L-M)训练算法及反向传播(BP)神经网络的传感器无效数据过滤器。通过离线采集传感器响应特性曲线数据的方式构建网络模型,并用Matlab工具对模型进行仿真训练。综合对比分析L-M训练算法、拟牛顿训练算法、自适应线性回归(LR)动量梯度下降训练算法的收敛速度和误差性能。对比结果表明,基于L-M训练算法构建的BP神经网络模型收敛速度更快、误差值更小、效率更高,有利于矿用催化原理传感器无效检测非线性数据的过滤。
Sensors based on the catalytic principle suffer from excessive errors in the measured combustible gas concentration values when working in a mine environment due to the interference of internal and external factors. A sensor invalid data filter based on Levenberg-Marquardt (L-M) training algorithm and back propagation (BP) neural network is designed. The network model is constructed by collecting the sensor response characteristic curve data offline, and the model is simulated and trained with Matlab tools. A comprehensive comparison is made to analyze the convergence speed and error performance of the L-M training algorithm, the proposed Newton training algorithm, and the adaptive linear regression (LR) momentum gradient descent training algorithm. The comparison results show that the BP neural network model constructed based on the L-M training algorithm has faster convergence speed, smaller error value and higher efficiency, which is conducive to the filtering of nonlinear data for ineffective detection of catalytic principle sensors for mining.
作者
王静文
王博文
WANG Jingwen;WANG Bowen(School of Information Engineering,Yellow River Institute of Water Conservancy and Technology,Kaifeng 475000,China;Chongqing Research Institute Co.,Ltd.,China Coal Technology and Industry Group,Chongqing 400050,China)
出处
《自动化仪表》
CAS
2023年第9期101-106,共6页
Process Automation Instrumentation
关键词
催化传感器
反向传播神经网络
拟牛顿训练算法
电桥原理
梯度下降训练算法
Catalytic sensor
Back propagation(BP)neural network
Proposed Newton training algorithm
Bridge principle
Gradient descent training algorithm
作者简介
王静文(1980—),女,学士,讲师,主要从事计算机软件方向的研究,E-mail:good170@163.com;通信作者:王博文,男,硕士,高级工程师,主要从事仪器仪表方向的研究,E-mail:29681717@qq.com。
