Non-linear modeling method based on near infrared spectroscopy and system thereof

Abstract

The invention provides a non-linear modeling method based on near infrared spectroscopy and a system thereof. The method comprises the following steps: spectrogram of each experimental sample is acquired and converted into near infrared spectroscopic data; a part of each near infrared spectroscopic data is picked out at random as a calibration set, and a part is picked out as a validation set; thecalibration set and the validation set undergo principal component analysis to obtain a spectral feature space; in the spectral feature space, samples in the calibration set that are the most similarto the samples of the validation set are picked out as a calibration subset through a Mahalanobis distance method; and principal component number is extracted from the calibration subset as an inputlayer of a BP neural network to establish a regression model. By the above method, the problem of multiple correlations among various factors can be solved, and much noise and some useless informationare also avoided. Variable dimension is reduced, and based on nonlinear mapping ability and adaptive learning ability of the BP neural network, predictive stability and precision of the model are raised.

Description

technical field [0001] The invention relates to the field of nonlinear modeling, in particular to a near-infrared spectrum-based nonlinear modeling method and system. Background technique [0002] Near-infrared spectroscopy is mainly the absorption of intramolecular frequency doubling and combined frequency. The spectral intensity is weak, the band is wide, and the overlap is serious. The analysis of near-infrared spectroscopy often needs to be combined with chemometric methods. Commonly used chemometric methods include multiple linear regression, principal component regression, partial least squares and other regression methods. These regression analyzes are based on the linear relationship between the spectral response and the corresponding chemical value. Ability to correct for linearity. [0003] However, there is not a strict linear relationship between spectral response and chemical value, especially for samples with complex compositions. In regression analysis, the ...

AI Technical Summary

Problems solved by technology

However, a large number of samples not only brings interference between a large number of irrelevant factors, but also there are large differences between different samples, especially for nonlinear spectral responses, which seriously affect the applicability of the model and the accuracy of prediction

Images