Near-infrared spectral variable selection method based on Monte Carlo variable combination cluster

Abstract

The invention relates to a near-infrared spectral variable selection method based on a Monte Carlo variable combination cluster, and belongs to the field of analytical chemistry and spectroscopy. The method comprises the following steps that firstly, calibration set samples are sampled randomly by the Monte Carlo sampling method; secondly, feature variables of each sample subset are selected by the variable combination cluster analysis method, and the feature variables of all the sample subsets are preserved to obtain a new variable space; then feature variables of the new variable space are further selected by using the variable combination cluster analysis method. The method not only samples the variable space by the binary matrix sampling method, but also realizes the sampling of a sample space by the Monte Carlo sampling method, and avoids the influence of the change of sample sets on the variable selection.

Description

technical field [0001] The invention belongs to the field of analytical chemistry and spectroscopy, in particular to a near-infrared spectrum variable selection method based on Monte Carlo variable combination clusters [0002] technical background [0003] The spectral frequency band of near-infrared is 780nm~2500nm. The source of spectral information comes from the frequency multiplication and combined frequency absorption of hydrogen-containing groups in organic matter. Near-infrared spectral analysis technology can be widely used in the field of qualitative and quantitative analysis of substances. Therefore, this project The technology has been hailed as "a technology with the potential to improve global agricultural analysis capabilities". Since the near-infrared spectrum has hundreds of spectral bands, when the instrument collects these bands, in addition to the information of the sample itself, it also contains a lot of external information, such as noise, sample backg...

AI Technical Summary

Problems solved by technology

[0005] Although the above methods are widely used in the near-infrared field, in scientific research practice, because the number of collected samples is generally not too large, there will be a situation where there are fewer samples and more variables, and there will be a large number of uninformative variables and variables. Interference variables are sandwiched, so the above variable selection methods are not only difficult to achieve all variable combinations, but also affected by a large number of uninformative variables and interference variables

It is also because the number of samples collected for modeling is not enough to fully express the overall information, so the parameters such as variable importance obtained by the above-mentioned several variable selection methods that only sample the variable space will have great uncertainty The fluctuation of the sample will inevitably have a certain impact on the importance distribution of the variables, which will affect the predictive performance of the model

Images