基于SVM的近红外光谱定性分析及其应用-模式识别与智能系统专业论文.docxVIP

浙江大学硕士学位论文 浙江大学硕士学位论文 III Abstract Near-infrared(NIR)spectroscopy technology is a novel chemistry analysis method．It has some obvious advantages such as unpolluted，nondestructive， fast—speed，high efficiency,low cost as well as easily applicable in on．1ine analysis． NIR spectroscopy has been successfully applied in the fields of food engineering， pharmacy,chemical engineering，petroleum industry etc．This thesis focused on the qualitative analysis techniques of NIR spectroscopy．The main contributions of this thesis are as follows： 1．The principles of NIR spectroscopy technology is introduced，and then the applications of pattem recognition technology and quantitative calibration in NIR spectroscopy technology are presented． 2．Support vector machine(SVMl classifier iS succefully applied in fast recognition of gasoline brands．Compared with other classifiers such as KNN (K。nearest neighbour method)，SIMCA(soft independent modeling of class analogy)． experimental results show that the SVM classifier has lower percentage of misclassification and more steady recognition results，which Can be widely used in most classification problems in various fields． 3．To improve the accuracy of near-infrared spectral quantitative analysis，a new hybrid PLS algorithm is proposed．An unknown sample is first classified by an SVM classifier，and then several similar training samples of the same type are selected to build the calibration model and predict the property of the unknown sample．To avoid the negative influence of classification failure，the new hybrid PLS includes a local PLS model based on the same—type training samples and a local PLS model based on the total training samples．It determines the property value of the unknown sample by comparing the outputs of the two models． For a set of gasoline samples，the hybrid PLS achieves high prediction accuracy of research octane number． 4．Based on the above research results，the software of an independently developed gasoline octane number NIR an