Kinetic modelling: Regression and validation stages, a compulsory tandem for kinetic model assessment

Abstract

The development of robust and reliable kinetic models is vital to build safe, eco‐friendly, and cost‐competitive chemical processes. Establishing kinetic models for complex chemical systems such as biomass valorization is cumbersome because the kinetic modeller must test different models and fit several experimental observables (or concentrations). Usually, in chemical reaction engineering, kinetic model assessment is based solely on the regression stage outputs. The implementation of a validation stage can aid in choosing the most reliable kinetic models, essentially in the case of complex chemical systems. We studied the solvolysis of 5‐hydroxymethylfurfural (5‐HMF) to butyl levulinate (BL) as a model reaction constituting several consecutive and parallel reaction steps. From an existing kinetic model, we created 60 synthetic runs in batch conditions. In the first part, we tested four different models with 5 degrees of noise, and we carried out the modelling on the 60 synthetic runs. In the second part, two types of holdout methods were evaluated. In the last part, cross‐validation, namely the k‐fold method, was used. We found that the 10‐fold method allowed more efficient selection results even when the noise level was high. Besides, k‐fold allows for not scarifying experimental runs and selecting the most reliable model.

Combining Isothermal and Adiabatic Mode Experiments for Kinetic Constant Estimation: Application to the Hydrogenation of 5-(Hydroxymethyl)furfural (5-HMF)

“…By complex model, we mean a system with multiphases and several reactions. To develop kinetic models for such a system, one needs to know the concentration evolution of species. − Using adiabatic conditions in reaction calorimeters requires no withdrawal of samples to avoid interference with temperature, limiting the kinetic description of all reaction steps. On the other hand, one could imagine combining some experiments in isothermal conditions, including offline analysis, and some in adiabatic conditions, including online reaction temperature.…”