【作者】 冯毅； 方亚平； 李梦龙；

【Author】 Yi Feng Yaping Fang Menglong Li College of Chemistry,State Key Laboratory of Biotherapy,Sichuan University,Chengdu 610064,Sichuan

【机构】 四川大学化学学院国家生物治疗重点实验室；

【摘要】 碳碳双键是有机化合物中常见官能团,而围绕碳碳双键衍生的结构非常多,其中比较常见的是碳碳双键上的几种共轭体系：p-π共轭,π-π共轭,σ-π超共轭。本文结合各种共轭体系与对应红外光谱的特点,设计了如下分类系统：第一层识别共轭体系和非共轭体系, 第二层识别 p-π共轭,π-π共轭和σ-π超共轭三种具体的共轭体系。在第一层分类以全谱数据作为特征向量利用支持向量机(SVM)、人工神经网络(ANN)、K-最邻近法(KNN)分别进行建模。而在第二层分类的时候,为了提高分类的准确性,在用全谱作为特征向量分类的同时,还用遗传算法和偏最小二乘法(GA-PLS)相结合对全谱数据进行特征挑选,以挑选出来的特征值作为特征向量用上述三种方法进行分类。实验结果表明,SVM 的分类效果最好, 而进行特征挑选后的分类效果比利用全谱分类的效果有较大提高。这说明 GA-PLS 是一个较好的特征挑选工具,在不减少信息量的情况下,大量减少了特征变量,从而降低了模型的复杂性,减少了计算量,同时还提高了模型的预测效果,将其与 SVM 相结合可以快速、准确地对三种共轭体系进行识别。更多还原

【Abstract】 C=C double bond is a familiar functional group in organic compounds,and the number of derivational structures from it is very large,which includes p-p conjugation,p-p conjugation and s-p hyperconjugation.In this paper,we have designed a multiple-level classification system according to three conjugational types and their infrared spectra.Firstly,all compounds were classified to conjugation system and non-conjugation system using SVM,ANN and KNN based on full spectrum;Secondly,the conjugation system were classified to p-p conjugation,p-p conjugation and s-p hyperconjugation based on the subset features which selected by GA-PLS. The result indicates that the classfier based on SVM had the best performance,while GA-PLS was a effective tool for feature selection and the model with selected feature was better than the full spectrum.Also it reduced the complexity of the model,lessened the consumption of compution and increased the forecast ability.Then we can conclude that three conjugation systems will be rapidly and accurately identified by SVM based on the GA-PLS.更多还原