Assessment of Thermal Mode-Based Kinetic Models via Stratified Cross-Validation and TPE Optimization: Application to the Hydrogenation of Furfural Using Synthetic Data

Kinetic models developed under isothermal and adiabatic modes enable process optimization in terms of safety, product selectivity, and energy efficiency. However, their development is often constrained by experimental efforts, and assessing their predictive validity remains challenging. This work introduces a stratified cross-validation framework for evaluating kinetic models across different ratios of isothermal and adiabatic runs. The methodology was applied to the hydrogenation of furfural to 2-methylfuran using synthetic data generated by a Langmuir–Hinshelwood model under ideal batch conditions without transport limitations, with added noise to simulate experimental uncertainty. A hybrid optimization strategy combining the tree-structured Parzen estimator (TPE) with a derivative-free local method improved robustness to convergence. Model-based design of experiments (MBDO) was applied to select informative runs and enhance parameter identifiability. The results show that stratified cross-validation can effectively distinguish between kinetic models under different thermal modes. The ratio of thermal data modes significantly influences model performance, with a 3:1 isothermal-to-adiabatic ratio yielding the most consistent validation results when data are limited.