神经网络与结构编码法预测直馏汽油色谱保留指数  

Prediction of Gas Chromatography Retention Indices of Straight Run Gasoline by Using Neural Networks and Structure Coding

作　　者：刘伟[1] 刘赞 王玲玲 

机构地区：[1]郑州大学生物工程系,河南郑州450052 [2]河南省环境监测中心站,河南郑州450004

出　　处：《郑州大学学报（工学版）》2004年第3期26-28,共3页Journal of Zhengzhou University（Engineering Science）

基　　金：河南省重点科技攻关项目(0223031800)

摘　　要：对直馏汽油中的单体烃的分子结构进行了数字编码,并采用误差反向传播神经网络算法构造了直馏汽油中单体烃的气相色谱保留指数与其分子结构的非线性相关模型,神经网络结构为3层,隐含层节点为7个,有15个输入,对应单体烃的15位数字编码,1个输出,对应气相色谱保留指数.预测结果表明,由误差反传算法所得的相关系数和标准偏差均优于多元线性回归方法.The molecular structures of the hydrocarbons of straight run gasoline are numerically coded. The nonlinear models of relationships between the chromatography retention indices of the hydrocarbons and their molecular structures are constructed by using error back-propagation neural network algorithm and their chromatography retention indices are predicted. The three-layer BPN which contains only one hidden layer, comprising fifteen input nodes, one output nodes and seven hidden nodes,is employed. The molecular structures and the chromatography retention indices are used as input and output, respectively. The results show that the correlation coefficient and the standard derivation obtained by means of error back-propagation algorithm are better than those obtained by using multi-linear regression.

关 键 词：神经网络 结构编码法 直馏汽油 气相色谱保留指数 BP学习算法 

分 类 号：TE626.21[石油与天然气工程—油气加工工程] O657.71[理学—分析化学]