基于Lasso回归和神经网络的辛烷值损失预测模型

Octane Value Loss Prediction Model Based on Lasso Regression and Neural Network

由于目前的炼油工艺过程具有复杂性和设备多样性,它们的操作变量间不仅仅是简单的线性关系,而是具有高度非线性和相互强耦联的关系,而且现阶段的研究中传统数据关联模型的操作变量较少,对过程损失预测的响应不及时,所以效果并不理想。本文根据催化裂化汽油精制装置采集的325个样本数据,通过lasso回归结合spearman相关系数筛选出影响辛烷值缺失的主要操作变量,通过BP神经网络建立了辛烷值损失的预测模型,并验证了预测模型的有效性。

Due to the complexity of the current refining process and the diversity of equipment, their operating variables are not only a simple linear relationship, but have a highly nonlinear and strongly coupled relationship, and the variables in the traditional data correlation model are relatively Less;the mechanism modeling requires higher analysis of raw materials, and the response to process loss prediction is not timely, so the effect is not ideal. In this paper, based on 325 sample data collected by catalytic cracking gasoline refining unit, the main operating variables that affect the missing octane value are screened out by lasso regression combined with Spearman correlation coefficient, and the prediction model of octane value loss is established by BP neural network, and the prediction model validity is verified.