基于VDE-PSO的汽油调合设计公式挖掘  

Formula mining in gasoline blending design based on VDE-PSO

作　　者：张运陶[1] 郑伟 白春艳 刘金迪 高世博[1] 

机构地区：[1]西华师范大学应用化学研究所,四川南充637002 [2]南充炼油化工总厂,四川南充637000

出　　处：《计算机与应用化学》2010年第5期627-630,共4页Computers and Applied Chemistry

基　　金：四川省企业信息化项目计划(2005-199-16)

摘　　要：提出一种将多变量非线性问题线性化的VDE-PSO-MLR的建模方法。该方法基于变量扩维、微粒群优化等手段选择扩维变量,在此基础上再建立拟线性的多元回归方程;并通过对所建立的各回归方程及其回归系数的显著性检验结果确定最佳回归模型。将该方法用于某炼油厂的汽油调合设计公式的挖掘,研究表明,与直接用自变量建立的线性回归方程以及二次回归方程相比,只有该方法建立的最佳模型方程和方程变量同时通过显著性检验。最后将最佳模型用于生产数据预测,计算调合汽油辛烷值测定值与预测值误差绝对值AE最大为0.185,符合AE≤0.3的要求。We proposed a modeling method to translate the multivariable nonlinear problem to linear problem which was called VDE-PSO-MLR./n this method a quasilinear multiple equation was established based on Variable Dimension Expansion and Particle Swarm Optimization. The optimal regression model was determined by the significant test of the regression equation and the corresponding regression coefficients. This method was used to mine formulae of gasoline blending design in a petroleum refinery. The results indicted that compared with the linear regression equation set by independent variable directly and the quadratic regression equation robustness of the model set by VDE-PSO-MLR was the best. Within the three models, only the best model set by VDE-PSO-MLR passed the significant test of regression equation and the corresponding regression coefficients. Then the best model was used in the prediction of production data. All the predicted results met the demand that absolute error （AE） ≤0.3 and the maximum of AE was 0.185 only.

关 键 词：变量扩维 微粒群 多元线性回归 汽油调合 公式挖掘 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TE6[自动化与计算机技术—控制科学与工程]