论文标题:近红外光谱分析中若干关键技术的研究
Study on Several Key Techniques of Near-Infrared Spectroscopy Analysis
论文作者 李庆波
论文导师 徐可欣,论文学位 博士,论文专业 测试计量技术及仪器
论文单位 天津大学,点击次数 163,论文页数 122页File Size2720k
2002-12-01论文网 http://www.lw23.com/lunwen_66967472/ 近红外,光谱分析,化学计量学,多变量校正方法
Near-infrared spectroscopy,Spectral analysis,Chemometrics,Multivariate calibration
近红外光谱分析技术具有高效、快速、成本低、无损伤和绿色环保等优点。它不仅可以应用于实验室分析，而且适用于现场快速检测和实时在线分析。但是目前一些关键的技术基础问题尚未得到彻底解决，严重阻碍着近红外光谱分析技术的进一步应用和发展。本论文综合多学科的知识深入研究了从重叠复杂的光谱中提取微弱的化学成份信息的方法，最佳测量条件的设计，建模方法的优化，校正模型稳健性的提高以及测量结果的物理解释等关键技术基础问题。论文的主要研究内容为：1．首次提出并建立了多变量光谱分析中最佳光程长的选择原理。对“多波长最佳光程长组合方法”提高测量精度的机理进行了理论证明和实验验证。研究了光程长的选择对于单变量光谱分析和多变量光谱分析的测量精度的影响。并提出以水为主要背景的样品最佳光程长的简单确定方法。2．首次建立了测量精度与仪器精度、建模方法之间的传递函数，并进行了实验验证。提出了实现预期测量精度所必需的仪器精度的确定方法。研究了建模方法和测量样品成份的复杂性对测量精度和必要仪器精度的影响，提出了提高测量精度的有效途径。3．首次系统研究了随着样品成份复杂性的变化以及光谱数据预处理方法和波长区域的不同，主成分的个数、意义及其可解释性的变化情况。指出多变量校正模型的主成分与多种因素有关，其表征的意义需要根据专业知识对具体问题做具体分析，而不存在完全固定的判断规则。并提出了通过主成分的分析来判断校正模型中是否存在非样品成分信息的方法。对主成分进行分析可以为测量结果的解释和测量方法、测量条件的优化提供物理依据。4．对遗传算法波长优选方法进行了系统的应用性研究，同时对基于遗传算法的常规波长选择方法进行了改进，使得波长优选在提高多变量校正模型预测精度的同时，还提高了模型的稳健性。同时深入研究了根据被测成份净信号误差进行波长选择的方法，并解决了逆校正模型中未知浓度样品的净信号偏灵敏度的计算方法问题，实现了实际光谱分析应用中的波长选择。5．对模型标准化方法进行了系统的应用性研究。分别对几种模型转换方法和提高模型稳健性的数学预处理方法进行了分析和比较，并进行了实验验证。同时研究了不同模型转换方法的选择依据以及转换集样品的选择对模型转换效果的影响。
Near-infrared spectroscopy analysis technique is efficient, rapid, noninvasive, environmental friendly and can be run at low costs. It is not only suitable for laboratory analysis, but also in-field fast measurement and real-time on-line analysis. However, there are some key techniques have not been solved thoroughly, which deter further application and development of near infrared spectroscopy (NIRS) technique. In the dissertation, an in-depth study is carried out by the author by synthesizing multi-disciplinary knowledge on the following topics: the abstraction of weak spectral signal, the optimal pathlength measurement condition, the optimization of regression method, the enhancement of multivariate calibration model robustness and the physical explanation of measurement results are studied.The main research content of the dissertation involves:1. The selection principle of optimal pathlength in multivariate calibration is proposed and established. "The method of multiple optimal pathlength combination" to improve prediction accuracy in multivariable calibration is theoretically verified and experimentally validated for the first time. The influence of pathlength selection on prediction accuracy of univariate calibration and multivariate calibration is investigated. The simple determination method of the optimal pathength of water solution is given. 2. The propagation function between the prediction accuracy, instrumental precision, and regression method is established, which has been verified experimentally. The determination method of essential instrumental precision to realize the anticipated prediction accuracy is given. The influence of regression method and the complexity of the sample components on prediction accuracy and essential instrumental precision are discussed. The efficient means to enhance the prediction accuracy is presented.3. The variation of the number of the principle components, shape, physical meaning and the explainablity with the change of calibration data and the sample complexity is systematically studied for the first time. It is pointed out that the principle components are influenced by many factors. The analysis of meaning for the principle components in multivariate calibration should combine with the professional knowledge about measuring sample. The analysis method is not completely fixed and specific for the different measurement object. We can know if the calibration mode is disturbed by the environmental noise by the principle components analysis, which provide the physical basis for the explanation of measurement result and the optimization of the measurement method and the measurement condition.The application of wavelength selection with genetic algorithms is systematically studied. The routine wavelength selection method based on genetic algorithms is modified, which enhance prediction accuracy and improve the robustness of the calibration model. These are all verified by experiments. The wavelength selection approach by the net analyte signal relative error is studied thoroughly. The4. calculation method of net partial sensitivity of unknown sample spectrum in inverse multivariate calibration is put forward, which enable the wavelength selection in practical near-infrared spectroscopy analysis application.5. The calibration model standardization is systematically studied to enhance the robustness and adaptability. Several mathematical pretreatment methods and model transfer methods are analyzed and compared, and some have been verified by experiments. Meanwhile, the effect of the selection of calibration transfer method and the influence of transfer set samples selection on the calibration model transfer result are also discussed.
|
