随机森林-相关向量机算法结合中红外光谱法快速测定在用柴油机油酸值
|
摘要
研究了随机森林-相关向量机(RF-RVM)算法与ATR中红外光谱法相结合测定柴油机油样品的酸值的方法。选择了某型车辆发动机在不同情况下所用的柴油机油样品48个,并按GB/T 4945-2002标准所述方法对48个样品的酸值进行测定,测得其酸值在0.59～1.71mg(KOH)·g~(-1)之间。运用SG平滑结合一阶微分对原始光谱图进行预处理,采用RF算法和相关系数法选择了特征波数,并进行RVM建模分析。RF算法选取6个特征波数,不仅大大减少了输入变量数,还使RVM算法建模效果更佳。模型验证集的相关系数为0.922 5,验证集标准偏差为0.091 7。结果表明,RF-RVM算法与红外光谱法相结合能实现在用油酸值的快速而准确的测定。
Combination of ATR middle IRS with algorithm of random forests-relevance vector machine(RFRVM)was studied for determination of acid value of diesel engine oil.48 samples of oil for diesel of vehicles in use under various circumstances were selected in this work,and their acid values were pre-determined by the method given in the standard of GB/T 4945-2002,giving results in the range of 0.59-1.71 mg(KOH)·g~(-1).The original IR spectra were preprocessed by SG smoothing in combination with first-order differential calculus,characteristic wave numbers were selected by applying RF algorithm and method of correlation coefficient,and modeling analysis was made by RVM.Six characteristic wave numbers were selected in RF algorithm,which not only making a remarkable reduction of number of input of variables,but also giving better modeling effect by the RVM algorithm.Correlation coefficient for the model verification set was found to be 0.922 5,with the standard deviation SEP of 0.091 7.It was shown that rapid and accurate determination of acid value of oil in use could be realized by the combination of IRS with algorithm of RF-RVM.
Combination of ATR middle IRS with algorithm of random forests-relevance vector machine(RFRVM)was studied for determination of acid value of diesel engine oil.48 samples of oil for diesel of vehicles in use under various circumstances were selected in this work,and their acid values were pre-determined by the method given in the standard of GB/T 4945-2002,giving results in the range of 0.59-1.71 mg(KOH)·g~(-1).The original IR spectra were preprocessed by SG smoothing in combination with first-order differential calculus,characteristic wave numbers were selected by applying RF algorithm and method of correlation coefficient,and modeling analysis was made by RVM.Six characteristic wave numbers were selected in RF algorithm,which not only making a remarkable reduction of number of input of variables,but also giving better modeling effect by the RVM algorithm.Correlation coefficient for the model verification set was found to be 0.922 5,with the standard deviation SEP of 0.091 7.It was shown that rapid and accurate determination of acid value of oil in use could be realized by the combination of IRS with algorithm of RF-RVM.
引文
[1]杨其明,严新平,贺石中.油液监测分析现场实用技术[M].北京:机械工业出版社,2006.
[2]GB/T 7607-2010柴油机油换油指标[S].
[3]GB/T 4945-2002石油产品和润滑剂酸值和碱值测定法[S].
[4]GB/T 7304-2014石油产品酸值的测定电位滴定法[S].
[5]严衍禄,赵龙莲,韩东海,等.近红外光谱分析基础与应用[M].北京:中国轻工业出版社,2005.
[6]王全才.随机森林特征选择[D].大连:大连理工大学,2011.
[7]BREIMAN L.Random forests[J].Machine Learning,2001,45:5-32.
[8]秦玉华,丁香乾,宫会丽.高维特征选择在近红外光谱分类中的应用[J].红外与激光,2013,42(5):1355-1359.
[9]BISHOP C M.Pattern Recognition and Machine Learning[M].[s.l.]:Springer,2007.
[10]赵榈.相关向量机优化方法研究[D].南宁:广西大学,2012.
[11]陆婉珍.现代近红外光谱分析技术[M].北京:中国石化出版社,2007.
[12]KALOUSIS A,PRADOS J,HILARIO M.Stability of feature selection algorithms:A study on high-dimensional spaces[J].Knowledge and Information Systems,2007,12(1):95-116.
[13]宋源,梁雪春,张然.基于统计特性随机森林算法的特征选择[J].计算机应用,2015,35(5):1459-1461.
[14]宋源,梁雪春,张然.基于统计特性随机森林算法的特征选择[J].计算机应用,2015,35(5):1459-1461.
[15]孙雪莲,舒清态,欧光龙,等.基于随机森林回归模型的思茅松人工林生物量遥感估测[J].林业资源管理,2015,2(1):71-76.
[16]曾甜玲,温志渝,温中泉.相关向量机在污水硝氮检测中的应用[J].光谱学与光谱分析,2013,33(4):1048-1051.
[17]褚小立,袁洪福,陆婉珍.近红外分析中光谱预处理及波长选择方法进展与应用[J].化学进展,2004,16(4):528-542.
[2]GB/T 7607-2010柴油机油换油指标[S].
[3]GB/T 4945-2002石油产品和润滑剂酸值和碱值测定法[S].
[4]GB/T 7304-2014石油产品酸值的测定电位滴定法[S].
[5]严衍禄,赵龙莲,韩东海,等.近红外光谱分析基础与应用[M].北京:中国轻工业出版社,2005.
[6]王全才.随机森林特征选择[D].大连:大连理工大学,2011.
[7]BREIMAN L.Random forests[J].Machine Learning,2001,45:5-32.
[8]秦玉华,丁香乾,宫会丽.高维特征选择在近红外光谱分类中的应用[J].红外与激光,2013,42(5):1355-1359.
[9]BISHOP C M.Pattern Recognition and Machine Learning[M].[s.l.]:Springer,2007.
[10]赵榈.相关向量机优化方法研究[D].南宁:广西大学,2012.
[11]陆婉珍.现代近红外光谱分析技术[M].北京:中国石化出版社,2007.
[12]KALOUSIS A,PRADOS J,HILARIO M.Stability of feature selection algorithms:A study on high-dimensional spaces[J].Knowledge and Information Systems,2007,12(1):95-116.
[13]宋源,梁雪春,张然.基于统计特性随机森林算法的特征选择[J].计算机应用,2015,35(5):1459-1461.
[14]宋源,梁雪春,张然.基于统计特性随机森林算法的特征选择[J].计算机应用,2015,35(5):1459-1461.
[15]孙雪莲,舒清态,欧光龙,等.基于随机森林回归模型的思茅松人工林生物量遥感估测[J].林业资源管理,2015,2(1):71-76.
[16]曾甜玲,温志渝,温中泉.相关向量机在污水硝氮检测中的应用[J].光谱学与光谱分析,2013,33(4):1048-1051.
[17]褚小立,袁洪福,陆婉珍.近红外分析中光谱预处理及波长选择方法进展与应用[J].化学进展,2004,16(4):528-542.