摘要
为实现直馏减压馏分油(VGO)中噻吩硫化物组成分布的快速分析,收集160个具有代表性的VGO样本,测定其常规物性及其含有噻吩硫化物的组成信息,构造160组数据集,并将其随机划分为训练集和测试集。以VGO的常规物性为输入特征,采用随机森林回归算法(RFR)分别构建预测VGO中苯并噻吩、二苯并噻吩、萘苯并噻吩以及总噻吩质量分数的模型。利用训练集样本的袋外估计,进行模型超参数的寻优。结果表明,模型对VGO中3种噻吩硫化物和总噻吩质量分数的预测标准偏差(RMSEP)分别为0.268%、0.131%、0.111%、0.385%,说明模型的预测值和实测值接近,具有较高的准确度和较强的泛化能力。
In order to realize the rapid analysis of the composition distribution of thiophene sulfides in straight-run VGO,bulk properties and compositions of thiophene sulfides of the 160 representative VGO samples were obtained to construct the database,which were randomly divided to the train set and test set.VGO’s conventional bulk properties were used as the input features,based on the random forest regression algorithm,the models were established respectively for predicting the mass fraction of benzothiophenes,dibenzothiophenes,napathabenzothiophenes and total thiophenes in VGO.The hyper-parameters of the model were selected by the out-of-bag estimates of train set samples,and the models were trained in the train set samples.For the test set samples,the prediction standard deviations of each model are0.268%,0.131%,0.111%,0.385%,respectively,which are close to the measured ones,indicating the prediction models have high accuracy and strong generalization ability.
作者
任小甜
褚小立
田松柏
REN Xiaotian;CHU Xiaoli;TIAN Songbai(Research Institute of Petroleum Processing,SINOPEC,Beijing 100083,China)
出处
《石油学报（石油加工）》
EI
CAS
CSCD
北大核心
2020年第5期995-1002,共8页
Acta Petrolei Sinica(Petroleum Processing Section)
基金
国家重点研发计划项目(2017YFB0306501)基金资助。
关键词
减压馏分
噻吩硫化物
组成分布
预测模型
随机森林算法
vacuum gas oil
thiophene sulfide
composition distribution
prediction model
random forest regression algorithm
