CAJ | 학술논문

采用拓扑结构描述符中的分子电性距离矢量(MEDV)，对催化裂化(FCC)汽油中48种硫化物在PONA柱上的气相色谱保留指数值(RI)建立多元线性回归模型和神经网络BP模型，并进行模型对比。结果表明，MEDV能很好分辨FCC汽油中不同硫化物以及同种硫化物异构体，由此建立的定量结构–保留相关关系的多元线性回归(MLR)模型和神经网络BP模型都具有较好的稳定性和良好预测能力，而非线性BP模型优于MLR模型的预测能力。
채용탁복결구묘술부중적분자전성거리시량(MEDV)，대최화열화(FCC)기유중48충류화물재PONA주상적기상색보보류지수치(RI)건립다원선성회귀모형화신경망락BP모형，병진행모형대비。결과표명，MEDV능흔호분변FCC기유중불동류화물이급동충류화물이구체，유차건립적정량결구–보류상관관계적다원선성회귀(MLR)모형화신경망락BP모형도구유교호적은정성화량호예측능력，이비선성BP모형우우MLR모형적예측능력。
Molecular electronegativity distance vector (MEDV) based on topological structure was used to establish multiple linear regression (MLR) model and back propagation (BP) neural network model about gas chromatographic retention index value of 48 kinds of sulfides in fluid catalytic cracking (FCC) gasoline on the PONA columns. Furthermore, these models were compared. The results showed that MEDV could well distinguish different types of sulfides and sulfide isomers in FCC gasoline,so MLR model and BP neural network model with quantitative structure-retention relationship had strong stability and good predictive ability. But the predictive ability of BP model was superior to MLR’s.