Application of Orthogonal Experimental Design to Optimized Selection of Modeling Factors in NIR Spectral Analysis

Orthogonal experimental design (OED) was applied to the optimization of selection of parameters for modeling in NIR spectroscopy by taking determination of moisture in amine fuel as example. The method of division of sample sets, proportion of the calibration set, pretreatment of spectra and selection of spectral band were taken as the factors in the OED. Calibration models were established separately, and standard prediction deviations (SEP) of the valication set obtained by the model were taken as the results of OED. Through analysis of the results of OED, optimum modeling parameters were determined, and optimal adjustments were made further around the most influential factor, in order to improve the steadiness of the prediction models and the accuracy of predicted results.