基于数据挖掘的汽油精制过程辛烷值损失预测模型

Prediction Model of Octane Number Loss in Gasoline Refining Process Based on Data Minin g

汽油精制过程中造成的辛烷值损失会降低汽油的燃烧效率,如何降低汽油精制过程中辛烷值的损失量是目前相关企业面临的一个重要课题。本文利用我国某石化企业在催化裂化汽油精制过程中积累的数据,建立基于神经网络、测量误差模型以及DC-SIS数据降维方法的两阶段特征筛选模型,选择出对辛烷值影响比较大的因素。设计了一种基于XGBoost和神经网络的辛烷值预测模型,可以实现对不同原材料和不同操作下精制后辛烷值的预测,经验证,模型的均方误差为0.06876,所设计模型在处理辛烷值预测问题时可以达到比较好的预测效果。

The loss of octane number in the process of gasoline ref ining will reduce the combustion eff iciency of gasoline.How to reduce the loss of octane number in the process of gasoline ref ining is an important issue facing related enterprises.This paper uses the data accumulated by a petrochemical enterprise during the ref ining process of catalytic c racking g asoline to e stablish a t wo-stage feature screening model based on neural network,measurement error model and DC-SIS data dimensionality reduction method,and select the one that has a greater impact on the octane number factor.An octane number prediction model based on XGBoost and neural network is designed,which can predict the octane number after ref ining under different raw materials and different operations.After verif ication,the mean square error of the model is 0.06876.A better prediction effect can be achieved in the alkane number prediction problem.