Comprehensive Investigation of Operating Parameters for Enhanced CO₂ Capture Using CaO Sorbent and Machine Learning

Abstract

The addition of sorbent to capture CO2 in steam methane reforming has become a method to reduce CO2 emissions in conventional hydrogen production. The aim of this paper is to propose a fast method for predicting the capture performance of sorbents by applying the eXtreme Gradient Boosting (XGBoost) method. First, the effects of inlet temperature, velocity, CO2 mass fraction, and initial material inventory height on the sorbent CaO capture efficiency and resulting flow regime are systematically analyzed, while an extensive database is constructed. Second, a comparative assessment is conducted to determine the relative significance of the four parameters in influencing the capture efficiency. Finally, the XGBoost model is trained and deployed to enhance the computational accuracy of capture efficiency predictions. The results emphasize that temperature has the most significant effect on capture efficiency. Increasing the temperature and decreasing the gas velocity helped to increase the capture efficiency. The initial material inventory height of 0.3 m proved to be favorable for CO2 capture compared to 0.1 and 0.2 m, especially under gas velocity conditions of 0.3 m/s. In addition, the XGBoost model can be used to quickly predict the CO2 capture efficiency of fluidized beds by using less training data and obtaining the prediction result with a high accuracy of R2 = 0.9699 in 2 s. This model has potential applications for the prediction of the sorbent capture efficiency.