摘要
基于上千个柴油样本建立了测定柴油十六烷值的近红外光谱数据库,采用一次性空瓶解决了光谱快速采集的问题,通过向数据库中添加少量样本的方式改进了模型在某石化企业的适用性,通过偏最小二乘法、支持向量机法和最小二乘支持向量机法将不同类型的柴油建立了统一的分析模型,并比较了不同算法建模的准确性。结果表明:使用PLS,SVM,LSSVM算法建立的校正模型对柴油样本十六烷值的预测标准偏差分别为1.6,1.4,1.3,可满足快速评价要求。本研究节约了建模成本,减少了数据库的维护工作量。
Based on more than one thousand diesel samples,the near infrared spectroscopic database for determination of cetane number was established,where a disposable headspace bottle was used for the spectral fast acquisition. By adding a small amount of samples to the database,a suitable model for a petrochemical company was obtained. A robust and uniform calibration model was developed for different types of diesels by partial least squares method,SVM and LSSVM algorithm and by comparing their results. The standard error of prediction of diesel CN using PLS,SVM and LSSVM calibration model were 1. 6,1. 4 and 1. 3,respectively. The work has advantages of high speed,cost saving of modeling and reduction of the database maintenance workload.
出处
《石油炼制与化工》
CAS
CSCD
北大核心
2016年第5期101-107,共7页
Petroleum Processing and Petrochemicals