改进S/B算法的近红外光谱模型转移
|
摘要
针对模型转移中S/B算法对于非线性问题的局限性,在传统S/B算法进行线性拟合、偏最小二乘法求参数的基础上加以改进,提出了引入变量的高次幂、使用Lagrange插值法与Newton插值法求待定系数和插值多项式来解决两组数据的非线性问题。为了验证改进算法的有效性,先对主机样品建模并分别预测主机和子机样品,然后通过实验数据和评价指标,筛选出最佳函数关系进行子机模型校正,并分别用改进的S/B算法和传统的S/B算法对子机未知样本进行预测。实验结果表明:直接用主机原模型对子机预测的值与真实值差距较大,利用改进的S/B算法(H-S/B)比传统的S/B算法预测值更接近真实值。改进的S/B算法提高了预测值的准确性,解决了传统S/B算法的非线性问题,实现了更好的模型转移效果,增强了网络化模型应用的通用性。
In view of the limitations of S/B algorithm for nonlinear problems in calibration model transfer,based on the traditional S/B algorithm that uses linear fitting and partial least squares method for parameters,this paper improved it by introducing high power of variable and using Lagrange and Newton interpolation method to seek undetermined coefficients and the interpolation polynomial to solve the nonlinear problem of the two sets of data.In order to verify the validity of the improved algorithm,this paper built a model for the master machine firstly and predicted the master and slave machine samples respectively,and then through the experiment data and the valuation index,it selected the best function relation to correct the model and finally predicted the unknown samples of the slave machine with the improved S/B algorithm and the traditional S/B algorithm.Experimental results showed that the gap is larger between reference value and the predicted value with master model directly,the predicted value with improved S/B algorithm was closer to the reference value than the traditional S/B algorithm.The improved S/B algorithm enhanced the accuracy of the prediction and solved the nonlinear problem of the traditional S/B algorithm.The algorithm based on Lagrange and Newton interpolation achieved better effect of model transfer and enhanced the generality of application in network modeling.
In view of the limitations of S/B algorithm for nonlinear problems in calibration model transfer,based on the traditional S/B algorithm that uses linear fitting and partial least squares method for parameters,this paper improved it by introducing high power of variable and using Lagrange and Newton interpolation method to seek undetermined coefficients and the interpolation polynomial to solve the nonlinear problem of the two sets of data.In order to verify the validity of the improved algorithm,this paper built a model for the master machine firstly and predicted the master and slave machine samples respectively,and then through the experiment data and the valuation index,it selected the best function relation to correct the model and finally predicted the unknown samples of the slave machine with the improved S/B algorithm and the traditional S/B algorithm.Experimental results showed that the gap is larger between reference value and the predicted value with master model directly,the predicted value with improved S/B algorithm was closer to the reference value than the traditional S/B algorithm.The improved S/B algorithm enhanced the accuracy of the prediction and solved the nonlinear problem of the traditional S/B algorithm.The algorithm based on Lagrange and Newton interpolation achieved better effect of model transfer and enhanced the generality of application in network modeling.
引文
[1]WANG Jia-jun,ZHE Wei,LIU Yan,et al(王家俊,者为,刘言,等).Acta Tabacaria Sinica(中国烟草学报),2014,6(1):1.
[2]ZHANG Xiao-yu,LI Qing-bo,ZHANG Guang-jun(张晓羽,李庆波,张广军).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2014,34(5):1429.
[3]ZHANG Xue-bo,FENG Yan-chun,HU Chang-qin(张学博,冯艳春,胡昌勤).Chinese Journal of Pharmaceutical Analysis(药物分析杂志).2009,29(8):1390.
[4]CHEN Bin,WANG Hao(陈斌,王豪).Infrared Technology(红外技术),2006,28(4):245.
[5]LIU Xian,DONG Su-xiao,HAN Lu-jia,et al(刘贤,董苏晓,韩鲁佳,等).Transations of the Chinese Society of Agricultural Machinery(农业机械学报),2009,40(5):153.
[6]Zhang Lei,Tian Fengchun,Kadri Chaibou,et al.Sensors and Actuators,2011,79(8):899.
[7]Liu X,Huang C J,Han L J.Energy and Fuels,2015,29(10):6450.
[8]Yahaya O K M,Matjafri M Z,Aziz A A,et al.Journal of Instrumentation,2015,10(5):TO5002.
[9]Watari Masahiro,Ozaki Yukihiro.Applied Spectroscopy,2004,58(10):1210.
[10]JIANG Er-xiong,ZHAO Feng-guang,SU Yang-feng(蒋尔熊,赵风光,苏仰锋).The Numerical Approximation(数值逼近).Shanghai:Fudan University Press(上海:复旦大学出版社),2012.26.
[2]ZHANG Xiao-yu,LI Qing-bo,ZHANG Guang-jun(张晓羽,李庆波,张广军).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2014,34(5):1429.
[3]ZHANG Xue-bo,FENG Yan-chun,HU Chang-qin(张学博,冯艳春,胡昌勤).Chinese Journal of Pharmaceutical Analysis(药物分析杂志).2009,29(8):1390.
[4]CHEN Bin,WANG Hao(陈斌,王豪).Infrared Technology(红外技术),2006,28(4):245.
[5]LIU Xian,DONG Su-xiao,HAN Lu-jia,et al(刘贤,董苏晓,韩鲁佳,等).Transations of the Chinese Society of Agricultural Machinery(农业机械学报),2009,40(5):153.
[6]Zhang Lei,Tian Fengchun,Kadri Chaibou,et al.Sensors and Actuators,2011,79(8):899.
[7]Liu X,Huang C J,Han L J.Energy and Fuels,2015,29(10):6450.
[8]Yahaya O K M,Matjafri M Z,Aziz A A,et al.Journal of Instrumentation,2015,10(5):TO5002.
[9]Watari Masahiro,Ozaki Yukihiro.Applied Spectroscopy,2004,58(10):1210.
[10]JIANG Er-xiong,ZHAO Feng-guang,SU Yang-feng(蒋尔熊,赵风光,苏仰锋).The Numerical Approximation(数值逼近).Shanghai:Fudan University Press(上海:复旦大学出版社),2012.26.