Abstract
The petroleum refining industry is under growing pressure to reduce energy consumption and carbon emissions while maintaining product quality. Catalytic reforming – a key process for producing high-octane motor fuel components, aromatic hydrocarbons, and hydrogen – offers significant optimisation potential through digitalisation. This paper presents a conceptual framework and literature analysis for building a digital twin of a catalytic reforming unit using physics-based mathematical models enhanced with machine learning (ML) and artificial intelligence (AI) techniques. We review existing kinetic and thermodynamic modelling approaches, evaluate their suitability as the physical layer of a digital twin, and identify how data-driven ML models can compensate for catalyst deactivation non-stationarity and feed composition variability. The proposed hybrid framework targets a 5-15% reduction in fuel consumption and a measurable decrease in CO 2 intensity per tonne of reformate, contributing to the broader industrial decarbonisation agenda.