Comparative Study of Surrogate Modelling Techniques Applied to Three Different Chemical Processes

Abstract In this paper, a comparative study of surrogate modelling techniques applied to chemical processes of different complexity is presented. The surrogate modelling techniques considered in this work are support vector regressions (SVR), Kriging and artificial neural networks (ANN). The surrogates were obtained by fitting to process data obtained from rigorous flowsheeting simulations previously developed in Aspen Plus v10. The processing schemes to be surrogated were (in order of decreasing complexity):1) the separation of aromatics-aliphatics mixtures by liquid-liquid extraction using ionic liquids as novel and more sustainable solvents, 2) the toluene hydrodealkylation process and 3) a simple distillation of organic solvents. Besides, in the first process (aromatics-aliphatics separation), advanced predictive thermodynamic models based on quantum chemical calculations (COSMO-SAC) were considered, allowing for the prediction of fluid phase equilibria properties of mixtures containing ionic liquids. In addition, the robustness of the surrogate techniques was assessed by adding a random noise contribution to the variables sampled. Thus, the paper is organized as follows: in section 1 an introduction to the surrogate modelling techniques is presented along with a short description of the chemical processes modelled in Aspen Plus. Afterwards, section 2 includes the methodology detailing the computational approach presented. The main results obtained in the study are presented in section 3 and the final the conclusions summarized in section 4.