Relationship Between The Spectral Difference Coefficient And The Model Precision Of Quantitative Spectral Analysis

Spectral overlapping can be a big factor affecting the analysis accuracy of the near-infrared spectroscopy quantitative analysis,in order to improve the analysis precision,in this paper,the influence mechanism of the degree of the spectral multicollinearity between an analyte and the interferences on affecting the model precision is explored,besides,a new wavelength selection method is proposed to reduce the negative effects of the spectral multicollinearity between an analyte and the interferences on the model precision.The research work in this paper mainly includes the following parts.First,the theoretical basis of the quantitative spectrometric analysis technique,the general steps of the quantitative spectrometric analysis method and the common chemometrics methods were reviewed in this paper.Then,the relationship of the degree of the spectral multicollinearity between the analyte and the interferences to the analysis precision was explored.Firstly,spectral difference coefficient was proposed as the index that measured the degree of the spectral multicollinearity between the analyte and the interferences,then through simulations,the effectiveness and feasibility of the spectral difference coefficient were verified and the relation between it and analysis precision was investigated.The results showed that effects of the interferences on analysis precision related exclusively to spectral multicollinearity between them and the analyte,i.e.,the spectral difference coefficient,and not to their number or shape of their spectra.The RMSEP(Root Mean Square Error of Prediction)was decreased by half when the spectral difference coefficient doubled.Moreover,the spectral difference coefficient was proved to be independent of the noise intensity,sensitivity of the analyte and the number of modeling wavelengths on affecting the analysis precision,which was more evident for the universal validity of the relation and offered the possibility of combining these factors together to improve the analysis precision by weighing their respective effects on the analysis precision.Thus,based on the three influence factors,the wavelength importance index was defined and a new wavelength selection method based on the index was proposed.This wavelength selection method was verified to be valid by actual quantitative analysis experiment of the ethanol in the ethanol–water solution.In this paper,based on the advantages of the net analyte signal,a new wavelengthselection method called the net analyte signal(NAS)based forward variable selection method(NAS-FVS)was proposed.It is a revision of the wavelength selection method based on the wavelength importance index.Compared with the old method,the new method is perfectly general as it can be used without knowing the pure spectrum of the analyte.The ability of the new method to select wavelength regions was illustrated with three real samples.The result showed that the method of NAS-FVS can provide higher prediction precision and own good compression ability.