摘要
降低汽油中的硫、烯烃含量,同时尽量保持其辛烷值是汽油清洁化的重点。由于炼油工艺过程的操作变量之间具有高度非线性及相互强耦联的关系,因此,本文在提高脱硫率的前提下,降低辛烷值损失的比例。采用3σ准则剔除异常值,用均值进行填充。再用核主成分分析方法,筛选与产品辛烷值和硫含量关联度高的30个变量,考虑降维的结果,赋予不同权重,建立量化评价模型,根据量化得分,确定出影响辛烷值预测的主要变量。最后利用岭回归和随机森林的方法,分别建立辛烷值损失预测模型,并对预测的结果进行了对比分析,发现岭回归方法的判定系数R2可达到0.927,预测精度较好。
Reducing sulfur and olefin content in gasoline and keeping octane number as far as possible is the key to clean gasoline.Due to the highly non-linear and strongly coupled relationship between the operational variables of the refinery process,the percentage of octane loss had been reduced under the premise of improving the desulfurization rate.The outliers had been removed using the 3σcriterion and filled with the mean values.Then the kernel principal component analysis(KPCA)method had been used to screen 30 variables that were highly correlated with the octane and sulfur content of the products.By considering the results of dimensionality reduction,we had assigned different weights to these variables to build a quantitative evaluation model.Subsequently,based on the quantitative scores,the main variables affecting octane prediction had been identified.Finally,the prediction models of octane number loss had been established by ridge regression method and random forest method respectively.Meanwhile,we had conducted a comparative analysis of the predicted results.As a result,it is found that the determination coefficient R~2 of the ridge regression method can reach 0.927,which indicates a better prediction accuracy.
作者
库在强
付爽
熊一凡
KU Zai-qiang;FU Shuang;XIONG Yi-fan(School of Mathematics and Statistics,Huanggang Normal University,Huanggang 438000,Hubei,China)
出处
《黄冈师范学院学报》
2023年第3期38-42,共5页
Journal of Huanggang Normal University
关键词
汽油
辛烷值
特征选择
核主成分分析
岭回归
gasoline
octane number
feature selection
nuclear principal component analysis
ridge regression
