基于传感器阵列和卷积神经网络的毒害气体泄漏预测方法研究  

Study of Toxic Gas Leakage Predicting Method Based on Sensor Array and Convolutional Neural Network

作　　者：汤旭翔 金锦华 TANG Xuxiang;JIN Jinhua(School of Management Engineering and E-Commerce,Zhejiang Gongshang Univesity,Hangzhou Zhejiang 310018,China;Logistics Service Center,Zhejiang Gongshang Univesity,Hangzhou Zhejiang 310018,China)

机构地区：[1]浙江工商大学管理工程与电子商务学院,浙江杭州310018 [2]浙江工商大学后勤服务中心,浙江杭州310018

出　　处：《传感技术学报》2023年第4期662-666,共5页Chinese Journal of Sensors and Actuators

基　　金：浙江省实验室工作研究会重点招标项目(ZB202102);浙江省教育厅数字化改革创新试点项目(2022);浙江工商大学"数字+学科"资助项目(2022);浙江工商大学研究生教育改革项目(YJG2018230)。

摘　　要：研究了一种基于传感器阵列和卷积神经网络的毒害气体泄漏预测方法,在实验室中布设监测传感器阵列,实验模拟检测氯气、硫化氢和挥发性有机物(VOCs)三类毒害气体泄漏,获取实验室安全监测的扁平化信号。将扁平化信号输入非线性特征计算模型(NFC)得到对应的模型输出信号参数(NFCOP)特征值,根据实时传感器的特征值与阈值下边界之间进行比较,构建传感器响应特征值向量,建立训练集和预测集,并采用卷积神经网络CNN对实验室安全情况进行预测,结果表明基于NFCOP特征分析和卷积神经网络方法,优于BP、PLS等预测方法,预测准确率最高可达93%,表明所探索的方法能够实现毒害气体预测的目标。A toxic gas leakage predicting method based on sensor array and convolutional neural network(CNN)is studied.Monitoring sensor array is deployed in the laboratory,and simulation tests using chlorinen,hydrogen sulfide,and volatile organic compounds(VOCs)are conducted to obtain the flat signal of laboratory safety monitoring.The flat signal is processed by using the nonlinear feature calculation(NFC)model for NFC output parameters(NFCOP)calculation.NFCOP values are compared with the down threshold bounda⁃ries to construct the sensor response eigenvalue vector.The training set and prediction set are established to train and test the CNN model for laboratory security prediction.Results show that NFCOP vector combined with CNN model presents better predicting accuracy than BP and PLS methods.The maximal prediction accuracy of the proposed method is 93%,demonstrating that the proposed method achieves the research aim of toxic gas leakage monitoring.

关 键 词：传感器阵列 非线性模型 毒害气体泄漏 实验室安全 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP212.9[自动化与计算机技术—控制科学与工程]