Research On Near-infrared Temperature Correction Modelling Method

The technique of near-inf-rared(NIR)spectroscopy is widely used in various modern industrial processing for the reason that the fastness,non-destructiveness and low cost However,the NIR spectra are sensitive to temperature variation,and the accuracy of the NIR model is reduced very seriously,which limits the practical application of NIR techniques.For considering the influence of' temperature on NIR calibration model and according to the research of NIR temperature correction scheme all above the world,this paper is devoted to investigating the NIR modelling method with temperature correction to improve the robustness and accuracy of the model in the case of temperature fluctuation.The theoretical results are applied to the measurement of bisphenol-A viscosity to verify the validity and applicability of the proposed approaches.The main research and contributions are as follow:(1)Theoretical analysis and verification of prediction variances and confidence limits for global temperature correction modelling approaches.The prediction accuracy of traditional global temperature correction model of Partial Least Squares(PLS)is analyzed theoretically according to the prediction variance,which verified the effectiveness of this method.Furthermore,with the combination of the confidence interval,a verification method for a single sample based on predictive confidence interval is proposed.In practical applications,the effect of temperature on the spectral shape of the samples and the trends of the first and second principal components with temperature are analyzed by taking the measurement of bisphenol-A viscosity.In addition,the prediction accuracy of model with or without temperature correction is compared and analyzed.(2)NIR modeling with temperature correction based on MLPCR.For the consideration of the effect that caused by the temperature fluctuation on NIR quantitative calibration model,the multilevel principal component regression(MLPCR)temperature correction modelling scheme is proposed in this paper.At first,the multilevel simultaneous component analysis(MLSCA)method is used to decompose the NIR spectrum into two parts,named as"between-part" and "within-part",in order to split the characteristics caused by temperature variation and concentration respectively.Then the quantitative relationship between within-part of spectra that eliminates the temperature influence,and concentration data is established by principal component regression(PCR).With an application in the viscosity measurement of bisphenol-A,the proposed MLPCR-based temperature correction modelling approach is made a comparison with the PLS standard temperature calibration modelling method,and it is obviously found that the proposed MLPCR scheme overcomes the effect of temperature fluctuation on NIR analytical model.(3)MLLASSO-based NIR temperature correction modelling.Based on the multilevel temperature correction framework and the regularization method of Least Absolute Shrinkage and Selection Operator(LASSO),a multilevel LASSO with temperature correction modeling method is proposed.More than eliminating the influence of temperature change,the characteristic bands related to the concentration information were selected by MLLASSO,which maintained a better balance between the robustness and accuracy of NIR calibration model to the temperature variation and improved the efficiency of calculation.The experimental results carried out on the viscosity measurement of bisphenol-A using MLLASSO algorithm are compared with those of the PLS standard temperature model and the MLPCR model.It turns out that the MLLASSO temperature correction algorithm shows the superiority in robustness and accuracy under the condition of temperature fluctuation.This paper is focusing on correcting the influence of temperature fluctuation on NIR quantitative calibration model from the aspects of the single-stage PLS global temperature correction,multilevel full-spectrum temperature correction,and multilevel wavelength selection temperature correction modelling schemes respectively.The effectiveness of PLS global temperature correction modeling is verified by theoretical analysis.Then the MLPCR and MLLASSO temperature correction modeling strategies are proposed,and a comparison is made in the experimental results of bisphenol-A viscosity measurement.Above all,the temperature correction is necessary to the NIR quantitative calibration model.Furthermore,the single-stage PLS global temperature correction modelling method has high requirements for sample representation of calibration set.Besides,both MLPCR and MLLASSO temperature correction methods can effectively compensate for temperature influence,and MLLASSO shows better robustness to temperature fluctuation.