A two-position spectral modeling method to increase the robustness of NIR analysis model

In NIR quantitative analysis, the robustness of the model determines the value in its application. When the application and modeling conditions show substantial variation, the model becomes less robust, and the prediction ability of the model is significantly reduced. Based on the “multi-dimensional, multi-mode and multi-position” strategy of “M + N” theory, the two-position spectral modeling method for the NIR quantitative analysis of non-scattering or weakly scattering solutions is proposed. The external influence of measurement conditions on the robustness of the model is illustrated by significantly changing the position of the light source. The two-position spectral modeling method is used to suppress these effects, thereby maintaining a high prediction accuracy of the model. In this paper, a verification experiment was designed. Three different intensities were set for the light source. Spectra transmitted from 35 sample solutions containing the mixture of intralipid and ink, were measured at two positions. Models built with the NIR transmission spectra measured at two positions were compared with the models built with those measured at a single position. The experimental results showed that the RMSEP of two-position spectral modeling method was declined by 76% compared to that of the traditional method, which proved that the models established with a two-position spectra were more robust towards the intensity changes of incident light.