应用于区间组合的非连续光谱数据的偏最小二乘方法改进研究

Spectroscopy analysis is a rapid measuring technology for the quantitative analysis of sample components by using the spectral information of samples. Based on the high dimensional spectral data including information and noises, chemometric methods would be effectively employed, simultaneously to eliminate noise and as much as possible to reduce data dimension, further to establish a good mathematical model to predict the composition of sample components. Partial least squares (PLS) is a commonly used spectral analysis method integrating principal component analysis and multiple regression. It can effectively remove noise and reduce data dimension. However, the PLS algorithm is usually used for continuous spectral data. For the analytes with discontinuous characteristics absorption, the PLS models are difficult to provide good prediction results. This project will study the principle of PLS, and mainly achieve the algorithm improvement in the derivation, computation and association process of the score matrix and the loading matrix, so that PLS can be applied to the analysis of the non-continuous spectral data by waveband combination. Meanwhile, the spectral information abstraction methods will be studied, to select out the wavebands with high signal to noise ratio, and to establish the optimized spectroscopy analysis model by using the improved PLS algorithm, and to realize the analysis of the discontinuous spectral data by waveband combination. By computer programming design, an algorithm platform will be built up to enhance the prediction results of spectroscopy analysis.

光谱分析技术是利用样品的光谱信息来对样品成分进行定量分析的一种快速检测技术。基于包含噪音的高维光谱数据，采用有效的化学计量学方法，在消除噪音的同时尽可能地降低光谱数据的维度，进而建立良好的数学模型来预测样品成分含量。偏最小二乘法（PLS）是融合主成分分析和多元回归的一种常用的光谱分析方法。它可以有效地去除噪音、降低数据维度。然而，PLS算法通常是对连续光谱数据建立模型，对于一些特征吸收不连续的分析对象，PLS模型很难提供良好的预测结果。本项目研究PLS算法的基本原理，主要对得分矩阵与载荷矩阵的推导、计算和关联过程做算法改进，使PLS能够适用于区间组合的非连续光谱数据的分析。同时，研究光谱信息优选方法，挑选出具有高信噪比的光谱波段，结合改进的PLS算法，建立优化的光谱分析模型，实现对区间组合的非连续光谱的数据分析，并进行计算机程序设计，建立相应的算法平台，从而达到提高光谱分析预测结果的目的。

偏最小二乘法（PLS）是光谱定量分析的一种核心回归方法，它融合了主成分分析和多元回归，可以有效地去除噪音、降低数据维度。本项目一方面研究PLS 算法的基本原理，针对得分与载荷矩阵的推导、计算和关联过程做算法改进，简化PLS的因变量和自变量交错迭代过程。第二方面，研究光谱信息优选方法，挑选具有高信噪比的光谱波段，结合改进的PLS算法，建立优化的光谱分析模型，实现对区间组合的非连续光谱的数据分析。第三方面，进行计算机程序设计，建立改进PLS的算法平台，从而达到实现光谱快速定量分析的目的。本项目在理论基础、应用基础和实际应用三方面都取得了较丰富的成果，发表国内外核心刊物论文6篇。