Application Of Ridge Regression And Its Improved Algorithm In Infrared Spectral Data

Infrared Spectra model is an important part of chemometrics.According to the different purposes of infrared spectroscopy modeling,it can be divided into quantitative analysis and qualitative analysis.In the process of analysis,infrared spectrum data is used to establish prediction models,from which the best prediction can be selected.The portable infrared analyzers provide convenience for the measurement of samples.However,in actual problems,it faces with few samples but many variables.It is difficult to meet the modeling requirements by using a least-squares method.In order to make up for the deficiencies of this modeling method,the ridge regression algorithm is applied in this thesis,which is a method to solve multicollinearity problem.It indirectly solves the multicollinearity problem by restricting the length coefficient.Therefore,in recent years,the modeling of ridge regression method has been widely studied and applied in many fields.Ridge regression and improved ridge regression algorithm are applied in infrared spectrum data and the linear model is discussed in this thesis.The main contents are as follows:This thesis introduces the background and significance of the topic,analyzes some infrared spectrum data modeling methods and basic principles,describes the ridge regression method,analyzes the nature of the ridge regression estimation and the selection of the ridgeparameters(k value).Verifying the feasibility of this algorithm in infrared spectrum modeling by using citrus spectral data.This thesis mainly studies the improvement of ridge regression algorithm.The core part of the improved algorithm is to add prior signal and solve the problem of sample deficiency and multicollinearity.Four groups of experiments data are used to compare the improved ridge regression algorithm with the partial least square method(comparing the root mean square error of the prediction set).Experiments adopt glucose,fructose,and sucrose as experimental sample,and data are collected in a constant temperature environment.From analysis of the two algorithm results,the improved ridge regression method is found to be more accurate than the partial least squares method in modeling the infrared spectrum.Increasing the number of samples in the calibration set(training set),it can improve the prediction accuracy.At the same time,the robustness of the improved regression model established in the complex system(non-constant temperature and concentration changes)is also discussed.The experimental results of the two groups show that the improved ridge regression algorithm is more robust than the partial least square method in the application of infrared spectroscopy.