Predictive model for CO2 absorption and mass transfer process based on machine learning methods

Abstract

The CO₂ mass transfer properties of the absorption process are crucial for optimizing industrial absorption packing columns, affecting the efficiency, economy, and eco-friendliness of CO₂ capture processes. Traditional experimental approaches for studying mass transfer parameters were limited by their resource-intensiveness, time-consuming nature, and substantial costs. This research focused on leveraging machine learning (ML) methodologies, specifically back-propagation neural networks (BPNN), random forests (RF), and support vector machines (SVM), to devise predictive models for the intricate mass transfer parameters involved in the amine-based CO₂ sequestration process. A comprehensive set of operational and physicochemical factors was employed as inputs to predict the total mass-transfer coefficient (K_G) and the gas-phase mass-transfer coefficient (K_g), which are crucial indicators of the CO₂ capture process’s performance. Based on a large amount of experiment data, ML established a reliable SVM model to accurately predict the physical properties and mass transfer distribution inside the tower, and identify the control steps of mass transfer under different conditions. The obtained results can be used to design packing configuration that varies along the tower height, and optimize the absorption parameter and the fluid dynamics design of the tower, including tower capacity, and packing height, the liquid distribution and gas–liquid phase contact, to ensure optimal mass transfer effect. Such theoretical predictions would reduce experimentation, accelerate development and industrial applications, and optimize the economic and environmental operational strategies of carbon capture processes.