近红外光谱数据建模方法的研究进展

应用化学计量学对近红外光谱数据进行建模是近红外光谱分析中的难点和关键。随着研究的深入，现有分析建模的改进算法和新方法的应用不断出现。本文详细阐述了数据预处理、定性和定量建模优化方法的研究进展。在降噪消噪方法中，小波(WT)变换是最为常用和有效的工具，应用非常广泛。波长优化选择方法是现在研究的重点之一，主要有遗传算法(GA)、无信息变量消除方法(UVE)和连续投影算法(SPA)等。在模型优化算法中，主要是在偏最小二乘法(PLS)、人工神经网络(ANN)和支持向量机(SVM)的基础上提出多种改进算法，能够更加有效地优化模型。但每种算法各具优点的同时也存在着一定的局限性，对于不同类型待测物的数学模型，优化的方法也有所不同，所以在实际应用中将多种数学挖掘方法结合，相互取长补短，将成为今后研究趋势。 

Progress of Near-infrared Spectral Data Modeling Method

It is the key step of the Near Infrared Spectral analysis that the chemometrics is applied to the near infrared spectral modeling. With the deep-going research, much more researchers pay attention to the improving of existing algorithms and the application of new methods to the NIR analysis. The data preprocessing, qualitative and quantitative research and modeling optimization methods have been analyzed. In De-noising methods, Wavelet Transform (WT) is the most commonly used and effective tool. Now, wavelength selection method is one of the key study, the main including Genetic Algorithm (GA)，Uninformative Variable Elimination (UVE), Successive Projections Algorithm (SPA) and so on. In terms of modeling, a variety of algorithms has been improved based on Partial Least Squares (PLS), Artificial Neural Networks (ANN) and Support Vector Machine (SVM), which can be optimized the predictive models effectively. However, each algorithm have the advantages at same time with some limitations, also the optimization methods of modeling are different for different types of analytes. So it will be the future research trends that a variety of mathematical methods have been combined and mutual complemented in the practical application.