| 英文摘要 | As a new and high technique, infrared spectral analysis technology (Near infrared, middle infrared and far infrared) develops rapidly. It can not only measure and analyze various components constituting complex mixtures simultaneously, but also reflect the overall absorption properties of mixture correctly. Due to the advantages of no sample preparation, high efficiency and fast time analysis, good repeatability and reproducibility of analysis result and real-time online detection, infrared spectral analysis technology is widely used in many fields such as agriculture, biology, food, medicine, chemical industry. For some simple mixture, traditional and linear multivariate calibration methods can solve the quantitative analysis problem of the infrared spectrum perfectly. However, for most complex mixture, the relationship between spectral variables and concentration or property is nonlinear, especially when the content of the sample range is larger, the nonlinearity is likely to be more prominent. In addition, some physical (optical scattering, etc.) and chemical (intermolecular interaction, hydrogen bonding effect, etc.) factors can also cause nonlinearity. Therefore, in the quantitative analysis field of infrared spectrum, how to establish nonlinear calibration models with high precision, strong robustness and good interpretability is still a problem and worth of studying. In this thesis, based on the nonlinearity source in the analysis system, several different nonlinear calibration models are proposed. They contains (1) For binary solution of hydroxyl compound with strong intermolecular interaction causing nonlinearity, polynomial based least squares (LSP) and polynomial based total least squares (TLSP) are proposed to capture the nonlinear relationship between absorbance and concentration, respectively. LSP is based on assuming that only absorbance noise exists; while TLSP takes both absorbance noise and concentration noise into consideration.Furthermore, according to global and local solving strategy, Limited-memory Broyden-Fletcher-Goldfarb-Shanno (LBFGS) and Levenberg-Marquardt (LM) optimization algorithm are combined with TLSP respectively; and then two different TLSP versions (termed as TLSP-LBFGS and TLSP-LM) are generated. The optimum polynomial order of each nonlienar model (LSP, TLSP-LBFGS and TLSP-LM) is determined by Leave-one-out cross-validation method. Comparison and analyses of the four models are made from two aspects: absorban... |
修改评论