近红外光谱数据的统计推断及在中草药分析中的应用

With the rapid development of science and technology, many high-throughput analysis instruments have been developed rapidly and applied in diverse fields of sciences. Near infrared spectroscopy (NIR) instrument, as a new high-throughput analytical instrument, has become the most widely used analytical tool in traditional Chinese medicine analysis because of its high speed, non-destructive and pollution-free characteristics. The near infrared spectroscopic data is high-dimensional, natural ordered, homologous as well as massive, which makes the traditional statistical methods no longer applicable. .Based on these characteristics，this proposal focuses on developing the corresponding statistical methods to analyze the near infrared spectroscopic data. Firstly, to deal with the problem of high dimension and natural ordered, a new variable selection method is developed and its statistical properties are studied. Secondly, to deal with problem of the homology and heterogeneity, new variable selection method is proposed and its statistical properties are discussed. Thirdly, the multi response model of near infrared spectroscopic data is studied, and the method of detecting the homology of multiple response regression coefficients is also developed. Finally, these methods will be applied to study traditional Chinese medicine. Through these studies, we hope to find more valuable results and more accurate estimates for NIR data, and provide better and more practical statistical tools for the data analysis.

伴随科学技术的高速发展，许多高通量分析仪器从无到有得到了迅速发展，并在不同的科学领域中得到应用。近红外光谱仪作为一种新的高通量分析仪器，凭借其高速、无损、无污染等特点，已成为中草药分析中最为广泛使用的分析工具。近红外光谱数据信息丰富，具有高维性、有序性及同源性等特点，使得传统的统计方法不再适用。.本项目结合这些数据特征来发展相应的统计方法分析近红外光谱数据。首先，针对近红外光谱数据高维性及有序性，拟发展新的变量选择方法，并研究其统计性质。其次，针对近红外光谱数据同源性和异质性，拟发展新的变量选择方法，并证明其统计性质。接着，研究近红外光谱数据的多响应模型，发展探测多响应回归系数同源性的方法。最后研究这些方法在中药分析中的应用。希望通过这些研究发现更有价值的结果和适用于近红外光谱数据的优良估计，同时为应用工作者提供较优且切实可行的统计工具。

伴随科学技术的高速发展，许多高通量分析仪器从无到有得到了迅速发展，并在不同的科学领域中得到应用。近红外光谱仪作为一种新的高通量分析仪器，凭借其高速、无损、无污染等特点，已成为中草药分析中最为广泛使用的分析工具。近红外光谱数据信息丰富，具有高维性、多重共线性、有序性及同源性等特点，使得传统的统计方法不再适用。. 本项目结合这些数据特征发展相应的统计方法分析近红外光谱数据。首先，针对近红外光谱数据高相关性及样本异质性，发展一种新的随机森林算法，可以有效的解决原始随机森林无法处理高相关数据的缺点。其次，针对近红外光谱数据同源性与稀疏性，发展新的变量选择方法用于对近红外光谱数据进行变量选择。接着，为了更全面、更好的了解近红外光谱数据的研究进展、热点及关注的问题，运用社区探索技术对他们进行相应的分析，研究结果表明近红外光谱数据仍然是应用统计学，特别是化学计量学界主要的研究对象之一。通过这些研究发现更有价值的结果和适用于近红外光谱数据的优良估计，同时为应用工作者提供较优且切实可行的统计工具。