基于特征提取的烃类气体电子鼻检测方法  

Electronic nose detection for hydrocarbon gas based on feature extraction

作　　者：翁小辉 栾祥宇 陈冬雪 张书军 肖英奎[1] 常志勇[1,2,5] WENG Xiao-hui;LUAN Xiang-yu;CHEN Dong-xue;ZHANG Shu-jun;XIAO Ying-kui;CHANG Zhi-yong(College of Biological and Agricultural Engineering,Jilin University,Changchun 130022,China;Key Laboratory of Bionic Engineering,Ministry of Education,Jilin University,Changchun 130022,China;School of Mechanical and Aerospace Engineering,Jilin University,Changchun 130022,China;School of Computing and Technology,The University of Gloucestershire,The Park,Cheltenham GL502RH,UK;National-Local Joint Engineering Laboratory of In-situ Conversion,Drilling and Exploitation Technology for Oil Shale,Jilin University,Changchun 130021,China)

机构地区：[1]吉林大学生物与农业工程学院,长春130022 [2]吉林大学工程仿生教育部重点实验室,长春130022 [3]吉林大学机械与航空航天工程学院,长春130022 [4]格罗斯特郡大学计算机与技术学院,公园区,切尔滕纳姆GL502RH,英国 [5]吉林大学油页岩地下原位转化与钻采技术国家地方联合工程实验室,长春130021

出　　处：《吉林大学学报（工学版）》2020年第6期2306-2312,共7页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(51875245);吉林省科技发展计划项目(20190302040GX,20190303118SF);吉林省产业技术研究与开发专项项目(2019C039-5);长春市科技计划项目(18DY007);吉林省教育厅“十三五”科学技术项目(JJKH20190186KJ,JJKH20190197KJ,JJKH20200969KJ).

摘　　要：针对目前油气探测中的气测录井技术无法进行现场快速检测的问题,提出将电子鼻用于油气探测中烃类气体的快速检测。将特征提取方法(最大值、平均值、傅里叶变换、高斯变换、指数函数曲线拟合、正弦函数曲线拟合)用于电子鼻数据预处理,分析表明:最大值与原始数据的相关性最高,指数函数曲线拟合与原始数据的相关性最低。主成分分析(PCA)和机器学习分类结果表明:主成分分析无法识别不同组分的烃类气体;选择最大值为特征值时,机器学习分类效果最好,逻辑回归、K最近邻、CatBoost、GBDT、Bagging识别率可达1;不同机器学习算法中支持向量机整体分类效果最好,所有特征提取方法的平均识别率达到了0.989。In order to solve the problem that the gas logging technology in oil and gas exploration can not be used in the field of fast detection,the electronic nose is introduced in the fast detection of hydrocarbon gas in oil and gas exploration.The feature extraction methods,such as maximum value,average value,Fourier transform,Gaussian transform,exponential curve fitting,sine curve fitting,are used to preprocess the data of the electronic nose.The analysis shows that the correlation between the maximum value and the original data is the highest,and the parameter of exponential curve fitting is the lowest.The results of principal component analysis and machine learning classification show that principal component analysis can not identify hydrocarbon gases of different components.When the maximum value is selected as the feature,the effect of machine learning classification is the best,and the recognition rate of logical regression,K nearest neighbor,CatBoost,GBDT and Bagging can reach 1.Compared with different machine learning algorithms,SVM has the best overall classification effect,and the average recognition rate of all feature extraction methods is 0.989.

关 键 词：精密仪器及机械 电子鼻 油气探测 特征提取 模式识别 

分 类 号：TH763[机械工程—仪器科学与技术]