基于近红外光谱技术成品汽油分类方法的研究

在研究成品汽油的分类方法过程中,首先采用判别式聚类分析方法比较了700~1 100和1 100~1 700 nm两个波段范围判别模型的准确性,然后在识别模型准确性较高的波段(1 100~1 700 nm)采用主成分分析法(PCA)结合自组织竞争神经网络方法,对90#,93#和97#成品汽油建立定性识别模型。在建立定性模型前先用PCA法对原始数据进行主成分压缩。主成分分析结果表明,前3个主成分的累积可信度已达97%,取前3个主成分的32个波长点的吸光度作为神经网络的输入,建立三层自组织竞争神经网络模型。神经网络模型的学习参数为0.01,网络训练迭代次数为500。结果表明,基于主成分分析结合自组织竞争神经网络方法建立的近红外光谱鉴别成品汽油的模型鉴别准确率高、方法可行。

Study on the gasoline classification methods based on near infrared spectroscopy

The purpose of the present paper is to study the classification methods of gasoline. First, two classific models are compared using discriminant cluster analysis method in 700-1100 nm and 1100-1700 nm spectral region. The sample is 90#, 93# and 97# gasoline. The results show that the model in 1100-1700 nm spectral region is veracious. And then a new model has been educed based on principal component analysis (PCA) and self-organizing competitive neural networks in order to classify 90#, 93# and 97# gasoline. The spectral data were condensed by PCA method before modeling, and three principal components were chosen because their cumulative credibility had reached 97%. A three-layer self-organizing competitive neural network model was established. Thirty-two wavelengths' absorbance is the concentrated spectral data by PCA method, and served as inputs of the self-organizing competitive neural network. The learning parameter is set as 0.01 and the training iteration is taken as 500. The conclusion is that it is feasible to apply near infrared spectroscopy to discriminate the gasoline products as the PCA and self-organizing competitive neural networks method is used. Also the PCA and self-organizing competitive neural networks method is better than the discriminant cluster analysis method.