Issue Highlights
Abstract
This article applies neural ordinary differential equations (neural ODEs) to learn hydrocracking reaction dynamics directly from data in a five-lump kinetic framework.Synthetic datasets from two models, a 2.5-order slurry hydrocracking system and a first-order Athabasca upgrading system, spanning multiple temperatures and feedstocks are used for training and testing.The models achieve sub-0.5 wt.% RMSE for in-range conditions, while revealing extrapolation limits at high temperatures.Gradient fields extracted from the learned dynamics recover expected reaction orders, Arrhenius-type temperature trends, and dominant pathways, demonstrating that neural ODEs can provide both accurate predictions and interpretable insight for hydrocracking reactor modelling.[1]
Fields of Study & Subjects
Model Reduction and Neural Networks
Petroleum Processing and Analysis
Neural Networks and Applications
Publication details
- Journal / source
- The Canadian Journal of Chemical Engineering
- Volume
- 104
- Issue
- 3
- Pages
- 1099–1099
- Publication date
- 8 February 2026
- Language
- EN