Hybrid heterogeneous transfer learning for kinetics models

Read full text Summarise with VIRAT
Authors: Youba Abed, Julien Jacques, Victor Costa, Denis Guillaume, Edouard Chappon, Benoît Celse
Cited by: 0
References: 4

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Hydrocracking units use hydrotreatment reactors to remove nitrogen impurities from heavy feedstocks. The energy transition is driving the evolution of refining processes toward renewable feedstocks, creating a need for predictive models based on limited experimental data. Although kinetic models (ordinary differential equations, or ODEs) offer interpretability and respect physical trends, they are difficult to recalibrate for these new chemistries such as tire feedstocks due to limited experimental data and incomplete chemical knowledge. Heterogeneous transfer learning (HTL) neural networks have emerged to address data scarcity, but these data-driven models often fail to extrapolate beyond the regions for which they were trained. This work overcomes this limitation by integrating HTL within the ODE itself.

Fields of Study & Subjects

Model Reduction and Neural Networks Domain Adaptation and Few-Shot Learning Neural Networks and Applications

Publication details

Journal / source
HAL
Publication date
10 May 2026
Language
EN