Infrared spectrum model transfer method based on random forest migration learning

Abstract

The present invention discloses an infrared spectrum model transfer method based on random forest migration learning. The method comprises the steps of utilizing the random forest thought to generate a sampling data set scanned by a master instrument into a plurality of sub-data sets by utilizing a Bootstrap method; aiming at each sub-data set, combining a sample data set scanned by a target instrument and utilizing a migration learning algorithm to establish an analysis model on the target instrument; aiming at a to-be-measured sample infrared spectrum acquired on the target instrument, predicting the to-be-measured component content of the to-be-measured sample infrared spectrum according to each established analysis model; calculating the structural distribution similarity between each to-be-measured sample and the samples in the established analysis models to determine the target analysis model weight factors corresponding to each to-be-measured sample; and then utilizing a weighted average method to gather the prediction results to obtain the final to-be-measured component content. The method has the advantages of being strong in robustness and being adaptive, enables the accuracy and the stability of the model transfer to be improved effectively, and can be widely used in the solid phase, liquid phase and gas phase infrared spectrum model transfer field.

Description

technical field [0001] The invention relates to an infrared spectrum model transfer method based on random forest transfer learning, which is applicable to a cross-platform model universal method for different manufacturers and different models of infrared spectrometers. Background technique [0002] Infrared spectroscopic analysis is a new analytical technique, which has been widely used in the fields of agriculture, chemical industry and environmental monitoring due to its advantages of fast, non-destructive and non-polluting. Infrared spectral analysis technology requires that the infrared spectrometer and the qualitative / quantitative analysis model must work in harmony, otherwise the analysis results will be affected. However, in the actual application process, there are usually different manufacturers and models of infrared spectrometers, so that the established analysis model cannot be applied to all infrared spectrometers, and it will cost a lot to build a separate an...

AI Technical Summary

Problems solved by technology

However, this method has the following two disadvantages: (1) In this method, support vector machine (Support Vector Machine, SVM), K-nearest neighbor (K-NearNeighbor, KNN) and partial least square (Partial Least Square, PLS) are used respectively The three methods establish a regression model, and then carry out weighted integration, but the models established by the three methods are all completed under the premise of the same sample distribution, so when the distribution of the sample to be tested is different from the sample distribution used in modeling , the phenomenon of "negative transfer" may occur; in other words, when the generalization performance (robustness) of the original model is poor, the error of the original model will also be transmitted to the target instrument; (2) when the target instrument When the distribution of samples to be tested changes, how to adaptively adjust the weights of each weak target analysis model according to the local structure of the samples to be tested

Images