Transfer Learning of Hydroprocessing Model from Fossil Feedstocks to Waste Plastic Pyrolysis Oil

Hydroprocessing of waste plastic pyrolysis oil (WPPO) is a promising technology for upgrading low-quality pyrolysis oil in order to send it to a steam cracker. This unit operation is the first step to the chemical plastic recycle. However, developing predictive models for this process is challenging...

Full description

Saved in:
Bibliographic Details
Published inComputer Aided Chemical Engineering Vol. 53; pp. 3115 - 3120
Main Authors Pejpichestakul, Warumporn, Becker, Per Julian, Celse, Benoit
Format Journal Article
LanguageEnglish
Published Elsevier 2024
Subjects
Computer Science
Modeling and Simulation
Transfer Learning
Hydrocracking
Bayesian Statistics
Pyrolysis Oil
Online AccessGet full text

Cover

More Information
Summary:Hydroprocessing of waste plastic pyrolysis oil (WPPO) is a promising technology for upgrading low-quality pyrolysis oil in order to send it to a steam cracker. This unit operation is the first step to the chemical plastic recycle. However, developing predictive models for this process is challenging due to limited data availability. The aim of this paper is to show that the knowledge from fossil feedstocks can be transferred to plastic recycle. This study shows an application of transfer learning application to develop a naphtha density model for WPPO using data from fossil fuels. The Bayesian transfer learning approach effectively transferred knowledge from the source data to the target data. The cross validation at different g-prior was applied to obtain the optimal g value. The transfer model with optimal g value results in an accurate predicted naphtha density on the testing and unseen datasets. It outperforms the model trained solely on the target data while delivering comparable performance on the training dataset. This confirms the robustness and predictive capability of the transfer model.
ISSN:1570-7946