基于分子同系物向量表示的石脑油特征提取方法

Naphtha characterization based on a molecular-type homologous series vector representation

根据石脑油馏分的特点,提出一种同系物向量表征方法。该方法将石脑油内部的每一个同系物分子定义为一个状态变量,这些状态变量构成了一个多维的向量空间。因此,任意一种石脑油都对应着该向量空间内的一个点,并且能够由该向量空间内的一组相互独立的石脑油馏分(基础油品)线性表示。在此基础上,提出一种基于非负矩阵分解(non-negative matrix factorization,NMF)方法的基础油品选择方法。该方法将具有较高维数的石脑油样品数据矩阵分解为一个较低维数的特征矩阵及其系数矩阵。在研究实例中,从59组石脑油样本数据中可以提取出21组基础油品,并且由它们还原得到的石脑油模型与样本数据相比,其相对误差不超过原数据的2.5%。

A novel homologous series vector representation method was developed for naphtha in which each homologous molecule of naphtha is defined as a state variable and all these variables are then used to construct a high dimension vector space. Thus, any variation of naphtha as one point in this vector space can be blended linearly by a group of independent naphthas named Basis Oils. These basis oils are obtained using the non-negative matrix factorization （NMF） method with the components data matrix of a huge number of naphtha samples factorized into a characteristic matrix with a lower dimension and its coefficient matrix. In a case study, a naphtha model containing 21 groups of naphtha bases was extracted from 59 groups of naphtha samples with a maximum representation error of less than 2. 5 percent of the original data.