基于复杂因果关系模型的特征选择新方法及其在蜡油加氢工业装置上的应用  

New Method for Feature Selection Based on Complex Causality Model andIts Application in the Industrial Unit of Wax Oil Hydrogenation

作　　者：田旺 秦康 吴昊 李明丰 谢煜 张璠玢 梁家林 TIAN Wang;QIN Kang;WU Hao;LI Mingfeng;XIE Yu;ZHANG Fanbin;LIANG Jialin(SINOPEC Research Institute of Petroleum Processing,Co.,Ltd.,Beijing 100083,China)

机构地区：[1]中石化石油化工科学研究院有限公司,北京100083

出　　处：《石油学报（石油加工）》2024年第3期580-590,共11页Acta Petrolei Sinica(Petroleum Processing Section)

基　　金：中国石油化工股份有限公司科研项目基金(323063)资助。

摘　　要：复杂因果关系模型的特征选择方法是当前人工智能领域的难点问题,开发新的特征选择方法对于因果关系模型的发展意义重大。以蜡油加氢工业装置为研究对象,针对模型涉及输入原料性质、操作参数等自变量特征和输出因变量特征精制蜡油硫含量之间复杂的因果关系,综合考虑输入变量两两之间的关系和输入变量与输出变量之间的关系筛选出合适的特征。首先,分别求出自变量特征之间和自变量与因变量之间的相关关系;其次,设计阈值和判别函数,将自变量、因变量结合起来,综合考虑两类变量之间的关系,最终将蜡油原料馏程中的初馏点筛选出来并剔除。相比传统的相关性特征选择方法,新筛选特征模型预测精制蜡油硫质量分数的平均绝对误差(MAE)减少62.97μg/g,平均相对误差(MRE)减少2.02百分点,决定系数R 2增加0.395,充分说明了新特征筛选方法对复杂因果关系模型的有效性。The feature selection method for complex causal relationship models is a challenging issue in the current field of artificial intelligence,and to develop new feature selection methods is of great significance for the development of causal relationship models.This study targets the wax oil hydrogenation industrial device,and focuses on the complex causal relationship between the features of input independent variables including raw material properties,operation parameters and output dependent variables of the model and the sulfur content in refined wax oil.Suitable features are selected by comprehensively considering the relationship between input variables and that between input and output variables.The first step is to calculate the correlation between the features of the independent variable and between the independent variable and the dependent variable.Then it is suggested to design thresholds and discriminant functions,combine independent and dependent variables,comprehensively consider the relationship between the two types of variables,and ultimately screen and remove the initial boiling point in the wax oil distillation range.Compared with traditional correlation feature selection methods,the average absolute error(MAE)and the average relative error(MRE)in prediction of sulfur mass fraction for refined wax oil by the new screening feature model are reduced by 62.97μg/g and 2.02 percentage points respectively,and the determination coefficient R 2 is increased by 0.395,fully demonstrating the effectiveness of the new feature selection method for complex causal relationship models.

关 键 词：蜡油加氢 硫含量 因果关系 相关关系 特征选择 

分 类 号：TE624[石油与天然气工程—油气加工工程]