论文部分内容阅读
烷基苯精馏分离是石油化工重芳烃加工的基本方法,各种烷基苯的热物性智能数据库对重芳烃加工过程优化控制有实用价值。本文研究了烷基苯系化合物若干热物性与化合物结构间的关系。采用新近提出的、特别适合于小样本多变量训练集的支持向量回归(support vector regression,SVR)算法总结了烷基苯系化合物已知物性的实验数据,建立了预报烷基苯系化合物若干物性的数学模型。47个烷基苯系化合物正常沸点、沸点汽化热、临界温度、临界压力和临界体积的SVR留一法(learing-one-out,LOO)预测的平均相对误差值(mean relative error,MRE)分别为0.370%,1.655%,0.791%,2.069%, 0.933%。结果表明,支持向量回归算法预测结果优于人工神经网络(ANN)和偏最小二乘(PLS)算法。
Alkylbenzene rectification separation is the basic method of petrochemical heavy aromatic processing. The thermophysical intelligent database of various alkylbenzenes has practical value for the optimization control of heavy aromatic hydrocarbon processing. In this paper, the relationship between some thermophysical properties of alkylbenzene compounds and the structure of compounds was studied. Based on the newly proposed support vector regression (SVR) algorithm, which is particularly suitable for the multivariable training set of small samples, the experimental data of the known physical properties of alkylbenzene compounds are summarized, and the prediction of some physical properties of alkylbenzene compounds Mathematical model. The mean relative error (MRE) of predicting learing-one-out (LOO) of 47 alkylbenzenes, normal boiling point, boiling point heat of vaporization, critical temperature, critical pressure and critical volume 0.370%, 1.655%, 0.791%, 2.069%, 0.933%. The results show that the prediction of support vector regression algorithm is superior to artificial neural network (ANN) and partial least squares (PLS) algorithm.