基于多元线性回归分析的汽油辛烷值损失预测建模

Prediction Modeling of Gasoline Octane Loss Based on Multiple Linear Regression Analysis

针对催化裂化汽油精制装置所采集的325个数据样本,运用R型聚类法得到建模的25个主要变量,并通过数据挖掘技术建立了基于多元线性回归分析的汽油辛烷值(RON)损失的预测模型.首先,利用3σ准则去除异常值等方法对原始数据样本进行预处理.其次,根据统聚类法中的R型聚类法,利用相关系数法和最大系数法分别确定指标变量之间的相似性度量并选取具有代表性的变量,得到降维后的25个主要变量.再次,以辛烷值损失为因变量,基于多元线性回归分析和十折交叉验证法,得到了辛烷值损失预测模型.最后,对回归模型和回归系数分别进行了假设检验,验证了所构建的辛烷值损失预测模型的合理性.

25 main variables of modeling are obtained by the R-type clustering method on basis of the 325 data samples collected by the catalytic cracking gasoline refining unit.The predictive model of gasoline octane number(RON) loss based on multiple linear regression analysis is established through data mining technology.Firstly,the original data samples are pre-processed through methods such as 3σ criterion to remove outliers.Secondly,according to the R-type clustering method,one of the unified clustering methods,the similarity measurements among the indicator variables are respectively set by the correlation coefficient method and the representative variables are selected by the maximum coefficient method.After that,25 main variables with dimensionality reduction are obtained.Taking the octane number loss as the dependent variable,the octane number loss prediction model is obtained based on the multiple linear regression analysis and the ten-fold cross-validation method.Finally,hypothesis tests are performed on the regression model and regression coefficients respectively to verify the rationality of the gasoline octane loss prediction model.