汽油组分辛烷值预测改进  

Improved Pridiction of Octane Number of Gasoline

作　　者：李为民[1] 邬国英[1] 单玉华[1] 李亮[1] 文戈 

机构地区：[1]江苏石油化工学院化学工程系

出　　处：《江苏石油化工学院学报》1998年第3期39-42,共4页Journal of Jiangsu Institute of Petrochemical Technology

基　　金：中国石油化工总公司资助

摘　　要：基于人工神经网络（ＢＰ）方法，用毛细管色谱法对南京炼油厂２９个汽油馏分样本进行全馏分分析，从色谱图中取出７个正构烷烃峰作为特征峰，共构成１５个虚拟组分，用于关联预报汽油组分的辛烷值，定性工作量大为减少。所建立的人工神经网络拓扑结构为１５－７－２，学习速率为０２，动量因子取０９，人工神经网络训练次数为４０００次，训练集样本１９个，检验集样本１０个，其辛烷值预测最大绝对误差为０２８，平均误差为０１２２，比常用的线性回归数学模型法更能准确地预报辛烷值。The analysis of 29 gasoline samples was carried out by a glass capillary column.7 normal alkyl peaks were picked out from the gas chromatgrphyspectroscopy.Thus 15 pesudo-components were composed.The octane amount of gasoline was predicted using a Back-Propagation Artificial Neural Network approach.The maximum error was 0 28.And the average error was 0 122.It was shown that ANN approach can predict octane amount of gasoline with quite better accuracy than linear regression method.

关 键 词：人工神经网络 汽油 辛烷值 成分 预测 

分 类 号：TE626.21[石油与天然气工程—油气加工工程]