Prediction of Saponification Value Based on Parameter Optimization Algorithms Combined with NIR

Abstract

Regression prediction methods of the saponification value of edible oils, which were based on different parameter optimization algorithms combined with near infrared spectroscopy (NIR) were studied. The 42 spectra of soybean, peanut and so on were collected by NIR spectrometer. The raw spectra were pretreated with SNV_DT, MSC, and SNV algorithms. Parameters (C, g) were optimized with SG, PSO, and GA methods, on which SVR models were built to predict the saponification value. The results show that these models could well predict the saponification value, in which the models with SNV_DT were superior. Particularly, in SG-SNV-DT-SVR model, the correlation coefficients of correction set and prediction set reached 99.2538% and 93.6408%, respectively, and the penalty factor C and kernel parameter g of this model are small. Therefore, the generalization ability, steadiness, and prediction ability are optimal. The results show that NIR technology can realize the rapid quantitative analysis of the saponification value of edible oils, which provides a theoretical basis for the development of rapid testing instruments.