Dynamic local multi-model consensus modeling for near infrared calibration of inorganic substances

Abstract

For inorganic substances such as potassium and chlorine in tobacco that do not directly absorb near infrared radiation energy, and need to be quantified indirectly with the interaction of organic groups that influence the spectrum. These factors result in the undesirable performance of the conventional partial least squares regression model. For this reason, this paper presented a dynamic local multi-model consensus modeling method for the quantitative analysis of inorganic components of tobacco. The method is conducted by dividing the content value of the analyte into multiple intervals and then establish sub-models in each local interval, while the content values of the target sample spectra are dynamically determined according to the proposed decision strategy. The method is applied to the modeling of inorganic matter of potassium and chlorine in tobacco. A calibration set of 210 samples and a test set of 70 samples were used and the results of the test set showed that the reference values of potassium and chlorine are highly correlated with the predicted values, the correlation coefficients are 0.987 and 0.977, respectively, and the average predicted errors from 9.02 and 13.23% decrease to 7.33 and 7.61% compared with the single PLS modeling approach, which effectively improves the quantitative performance and satisfies the demands of the enterprises.