Base Oil Process Modelling Using Machine Learning

Muhamad Amir Mohd Fadzil, Haslinda Zabiri, Adi Aizat Razali, Jamali Basar and Mohammad Syamzari Rafeen
Additional contact information
Muhamad Amir Mohd Fadzil: Group Research & Technology, PETRONAS, Kawasan Institusi Bangi, Kajang 43000, Selangor, Malaysia
Haslinda Zabiri: CO2RES, Chemical Engineering Department, Institute of Contaminant Management, University Teknologi PETRONAS, Seri Iskandar 32610, Perak Darul Ridzuan, Malaysia
Adi Aizat Razali: Group Research & Technology, PETRONAS, Kawasan Institusi Bangi, Kajang 43000, Selangor, Malaysia
Jamali Basar: Group Research & Technology, PETRONAS, Kawasan Institusi Bangi, Kajang 43000, Selangor, Malaysia
Mohammad Syamzari Rafeen: Group Research & Technology, PETRONAS, Kawasan Institusi Bangi, Kajang 43000, Selangor, Malaysia

Energies, 2021, vol. 14, issue 20, 1-25

Abstract: The quality of feedstock used in base oil processing depends on the source of the crude oil. Moreover, the refinery is fed with various blends of crude oil to meet the demand of the refining products. These circumstances have caused changes of quality of the feedstock for the base oil production. Often the feedstock properties deviate from the original properties measured during the process design phase. To recalculate and remodel using first principal approaches requires significant costs due to the detailed material characterizations and several pilot-plant runs requirements. To perform all material characterization and pilot plant runs every time the refinery receives a different blend of crude oil will simply multiply the costs. Due to economic reasons, only selected lab characterizations are performed, and the base oil processing plant is operated reactively based on the feedback of the lab analysis of the base oil product. However, this reactive method leads to loss in production for several hours because of the residence time as well as time required to perform the lab analysis. Hence in this paper, an alternative method is studied to minimize the production loss by reacting proactively utilizing machine learning algorithms. Support Vector Regression (SVR), Decision Tree Regression (DTR), Random Forest Regression (RFR) and Extreme Gradient Boosting (XGBoost) models are developed and studied using historical data of the plant to predict the base oil product kinematic viscosity and viscosity index based on the feedstock qualities and the process operating conditions. The XGBoost model shows the most optimal and consistent performance during validation and a 6.5 months plant testing period. Subsequent deployment at our plant facility and product recovery analysis have shown that the prediction model has facilitated in reducing the production recovery period during product transition by 40%.

Keywords: machine learning; base oil; SVR; random forest; decision tree; XGBoost (search for similar items in EconPapers)
JEL-codes: Q Q0 Q4 Q40 Q41 Q42 Q43 Q47 Q48 Q49 (search for similar items in EconPapers)
Date: 2021
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
https://www.mdpi.com/1996-1073/14/20/6527/pdf (application/pdf)
https://www.mdpi.com/1996-1073/14/20/6527/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jeners:v:14:y:2021:i:20:p:6527-:d:653945

Access Statistics for this article

Energies is currently edited by Ms. Agatha Cao

More articles in Energies from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().