Domain self-adaptive PLS regression model modeling method

Abstract

The invention discloses a domain self-adaptive PLS regression model modeling method. According to the method, an original domain spectrum centralization matrix is constructed by adopting near infraredspectrum data obtained from an original domain; a target domain spectrum centralization matrix is constructed by adopting near infrared spectrum data obtained from a target domain; the mean value difference of the spectrums of the original domain and the target domain is eliminated; based on the original domain spectrum centralization matrix and the target domain spectrum centralization matrix, atransfer matrix is mapped to a kernel matrix space, so that an optimal projection direction can be found from the transfer matrix; an optimal projection matrix is determined; a final PLS regression model is constructed based on the optimal projection matrix; and projection scores between the different domains and the non-independence of domain labels can be weakened. A domain adaptive algorithm is adopted to eliminate the difference of near-spectral data acquired in the different domains; the concentration information of target domain samples does not need to be acquired, so that a modeling process is simplified; and the constructed PLS regression model has good prediction precision for the near-infrared spectral data of the target domain.

Description

technical field [0001] The invention relates to a PLS regression model modeling method, in particular to a domain adaptive PLS regression model modeling method. Background technique [0002] Near-infrared spectroscopy is a simple, fast and reliable detection technique. It comprehensively uses the research results of multiple disciplines such as spectral technology, computer technology, and pattern recognition. With its unique advantages, it has been increasingly widely used in many fields, and has gradually been generally accepted by the public and officially recognized. Near-infrared spectral analysis is an indirect analysis method, which often needs to construct a regression model that reflects the relationship between near-infrared spectral data and the attributes of samples to be analyzed. Among them, the partial least squares (PLS) regression model is the most commonly used multiple regression model. PLS can eliminate the noise information in the spectral matrix and c...

AI Technical Summary

Problems solved by technology

However, with the complexity of the application scenarios of near-infrared spectroscopy, it is often encountered that the detection conditions or the instrument itself change, such as the temperature / humidity change of the sample inspection, the change of the sample shape, the aging of the instrument, and the replacement of accessories. Absorbance differences and wavelength shifts often occur in the near-infrared spectral data of standard samples, which leads to the partial least squares (PLS) regression model constructed based on the original domain (source domain, corresponding to the near-infrared spectral data collected under condition 1) data. The prediction results of the target domain (target domain, corresponding to the near-infrared spectral data collected under condition 2) data have large deviations

Images