2024
Historically, kinetic modelling is used to represent the underlying physics of the system via first principle or empirical derived systems of ODEs. Assumptions and simplifications, however, introduce inductive bias. Data-driven models, however, are plagued with further unique challenges including a lack of interpretability and high data intensity. Hybrid modelling combines mechanistic and data driven information to (ideally) combine the advantages of both. In this work several data-driven and hybrid models were compared with a traditional kinetic model for a simplified hexadecane hydroisomerisation network. It was found that the hybrid techniques were most promising, offering increased accuracy, low data requirements and high flexibility, managing to boast the largest improvement in model accuracy.
