支持向量机算法应用于生物活性混合体系的定量分析及重元素光谱能级分类

【作者】 张丹；

【导师】 曹晓卫；

【作者基本信息】 上海师范大学 ， 分析化学， 2009， 硕士

【摘要】 化学计量学(chemometrics)研究目的在于优化化学量测过程,并从化学量测数据中最大限度地获取有用的化学信息。支持向量机(support vector machine,SVM)方法,是一种基于结构风险最小化的新兴化学计量学方法。SVM算法可以在很大程度上避免误差反向传播(back propagation neural networks,BPN)使用过程中存在的“过学习(over-fitting)”问题;通过选用不同的核函数可以寻找出空间最优平面,以期避免信息的丢失,取得更为可靠、更为准确的结果。SVM方法正逐步应用于包括多元分辨与校正分析、模式分类等研究领域中,也有望在数据处理和分析任务愈来愈繁重的现代分析科学中发挥它的积极作用。本论文主要应用SVM算法对多元混合体系的定量分析以及光谱的化学模式分类这两个方面进行了研究,具体内容如下:对于多元混合体系的定量分析,常需要花费大量的时间和精力在多组分的预分离,而借助化学计量学手段则可较简单的实现复杂多组分的同时直接测定。我们将支持向量机方法分别应用于处理多种混合氨基酸体系的拉曼光谱、儿茶酚胺类物质混合体系的微分脉冲伏安图谱的定量分析研究。研究表明,支持向量机方法能更好地从混合体系的量测数据中提取信息以实现定量分析目的,较传统的BPN方法,其分析结果更为精确。原子光谱的电子组态通常是根据谱线的能级、强度、同位素位移、塞曼效应等测量数据进行确定,或者应用量子理论计算来指认。但由于原子光谱的复杂性,仍有部分高激发态的原子光谱所属的电子组态难于确定。因此,尝试采用支持向量机方法来对UⅡ等重元素原子光谱的分类研究,对于完善原子光谱数据信息具有重要的意义。计算结果表明,相对于传统化学模式识别方法,支持向量机能够更为全面和准确地预报了未知能级的组态归属。更多还原

【Abstract】 Chemometrics was designed to optimize the process of chemical measurements and get useful chemical information from the data of chemical measurements.Support vector machine (SVM) has solid theoretical foundation and can deal with small dataset, nonlinear optimization, high-dimensional feature space, local minimization and other realistic problems. Along with the development of SVM, some derived algorithms have been put forward and the application of SVM has gradually been the hot point for researchers in the world. Today, SVM has been successfully applied in face recognition, voice identification, handwritten digit recognition, text classification, risk assessment, protein structure recognition, gene recognition and other pattern recognition domains and achieves equivalent or superior results compared to those obtained by some other methods. It is very exciting that their capability to generalize input-output mapping from a limited set of training examples is great. In this paper, we use SVM to solve the problems of determining mixture and to classify the spectrums of heavy metal atom:Using SVM to determing mixtures, such as amino acid, catecholamines(CATs) by informations from the mix spectrograms of DPV and Raman without pre-separation. Study shows that SVM can well deal with such mixture, relative to BPN , it gains more accuracy information.Using SVM to classify the unknown energy levels of heavy metal-U II, which can not be classified by experiment. Although some people have tried to use traditional chemometric techniques to predict the unknown energy levels, there still have some samples which can not be predicted. So we use SVM to deal with such heavy metal to gain the energy levels. The results show that SVM predict more accuracy and completely than traditional methods of chemometrics.更多还原