Bayesian optimization (BO) is a strategic and iterative method for identifying optimal solutions in problems where evaluations are costly. Owing to its efficiency, BO has been successfully applied to various fields, including hyperparameter tuning, design of experiments, and process optimization. Adsorption processes, widely used in gas separation and carbon capture, require precise tuning of multiple operating parameters, including adsorption and desorption conditions, step times, and feed conditions, to achieve optimal performance in terms of energy consumption, productivity, or product purity. Dynamic simulations of adsorption systems, such as those used in pressure swing adsorption (PSA) or temperature swing adsorption (TSA), are particularly resource-intensive due to frequent changes in boundary conditions, and the need to reach cyclic steady-state, making optimization especially challenging. BO could be an effective solution for determining the optimal operating parameters in adsorption system. However, as the number of decision variables increases, the efficiency of BO may suffer from the curse of dimensionality, leading to increased number of required iterations and computational costs. We propose an efficient BO approach to accelerate convergence by leveraging prior knowledge derived from thermodynamic data. In this study, BO is applied to optimize PSA processes by systematically tuning operating parameters while incorporating prior information from isotherm data and shortcut models. Although shortcut models cannot provide fully accurate predictions due to their inherent equilibrium assumptions, such prior knowledge significantly enhances the efficiency of Bayesian inference, dramatically reducing the number of BO iterations to reach the optimal solution.
