A near-infrared quantitative model construction method combining qualitative and quantitative

Abstract

The invention provides a near-infrared quantitative model construction method that combines qualitative and quantitative methods, which includes the following steps: obtaining actual samples of the modeling calibration set and detecting their basic chemical components; scanning the spectrum corresponding to the calibration sample and eliminating abnormal samples; Qualitative spectral projection; classify the projection data; use the near-infrared spectra and chemical values ​​of each category as a verification set, use the modeling set to predict the verification set, and find the prediction error; randomly select near-infrared wavelength points; solve for each generated wavelength The overall calibration set error corresponding to the point; determine the near-infrared wavelength selection point and the characteristic information of the near-infrared spectrum based on the minimum overall calibration set error; re-establish a regression model for the calibration set spectrum and chemical values; detect the chemical values ​​of the verification sample and The corresponding spectrum is obtained and the regression model is quantitatively evaluated. Since the present invention performs qualitative projection analysis on the correction set spectrum, it is adaptable to changes in the spectrum and can keep the prediction of the model stable.

Description

technical field [0001] The invention relates to a method for constructing a near-infrared quantitative model combining qualitative and quantitative methods, and belongs to the field of near-infrared detection and analysis. Background technique [0002] Near-infrared spectroscopy is an electromagnetic wave belonging to the 850-2500nm band. Different types of spectra reflect molecular information at different sample levels. Because near-infrared spectroscopy contains a large amount of material information and has advantages such as rapid detection, near-infrared spectroscopy has achieved great success in online production quality monitoring. Wide range of applications; the traditional method of building a near-infrared quantitative model is: obtain the near-infrared spectral information of the sample, use the flow analyzer to detect the basic chemical value information, randomly divide the spectrum into a calibration level and a verification set, and eliminate abnormal samples....

AI Technical Summary

Problems solved by technology

[0002] Near-infrared spectroscopy is an electromagnetic wave belonging to the 850-2500nm band. Different types of spectra reflect molecular information at different sample levels. Because near-infrared spectroscopy contains a large amount of material information and has advantages such as rapid detection, near-infrared spectroscopy has achieved great success in online production quality monitoring. Wide range of applications; the traditional method of building a near-infrared quantitative model is: obtain the near-infrared spectral information of the sample, use the flow analyzer to detect the basic chemical value information, randomly divide the spectrum into a calibration level and a verification set, and eliminate abnormal samples. Band selection or wavelength point selection selects the wavelength of the spectrum to form a modeling interval, and then uses pls (partial least squares) to establish a relationship between the spectrum and the chemical value after screening the wavelength interval and verify the verification sample; the above methods exist as follows (1) Due to the many factors affecting the near-infrared spectrum, it is easily affected by the external environment, temperature, and sample type; in actual use, when the external spectrum changes and the calibration sample is inconsistent, it is easy to cause the model The external prediction error is too large; (2) For the construction of near-infrared models of complex sample types, in the traditional modeling method, there will be heavy calibration and light verification, so that in the actual quality monitoring process, when the actual sampling sample The chemical value and the corresponding spectrum change slightly, and the traditional modeling method will not have the consistency of the actual sample change trend

Images