The incorporation of microkinetic method for surface reaction calculation is a key development direction in monolith reactor simulations, albeit posing significant computational stability and efficiency challenges. A data-driven multi-scale simulation framework is proposed to bridge the gap between macroscopic numerical model and microscopic reaction solvers. After the generating of precomputed microkinetic dataset, we utilize a tailor-designed preprocessing transformation, AD_log10, to effectively manage the common challenges of large variances and reversed directions in microkinetic rate data. A suite of machine learning regression models is developed as reaction rate solver integrated into the 1D two-phase monolith model. We employ NH3 selective catalytic oxidation on Cu(100) and Cu(111) as probe systems to validate our framework’s efficacy. In the context of on-board simulations under operating conditions, the Extra Trees Regression (with best regression performance) based model exhibits commendable accuracy and efficiency, successfully addressing complex processes previously unmanageable with interpolation-based models. This innovative approach paves the way for advanced monolith reactor modeling and sheds light on multi-scale simulation methodologies in chemical reaction engineering.
Read full abstract