Explainable machine learning to reveal the parameter effect on Mg-based silicate mineral carbonation and screening its desired conditions

Mg-based silicates present a viable approach for mineral carbonation aimed at mitigating CO₂ emissions. Numerous experimental studies have been conducted on this topic, resulting in a substantial amount of data. The main purpose of this study is to investigate the influence of various carbonation operating parameters (features) on CO₂ sequestration efficiency through a machine learning (ML) approach. This study method began with the development of a dataset consolidating data points from a wide range of previously published studies. In this study, two regression models were employed to evaluate the effects of the different carbonation conditions on CO₂ sequestration efficiency: Random Forest Regression (RFR) and Extreme Gradient Boosting Regression (XGBR). Based on SHapley Additive exPlanations (SHAP) analysis, the influence of features was consistent with existing theoretical understanding. This study's primary finding is that iron content, along with the molar ratios of Fe/Si and Fe/Mg, positively influences CO₂ sequestration efficiency. Furthermore, the developed ML model identified an optimal carbonation condition (CO₂ sequestration efficiency exceeding 65 %) for olivine at the following parameters: temperature of 167°C, pCO₂ of 116.92 atm, stirring speed of 864 rpm, concentrations of NaHCO₃ and NaCl at 1.64 M and 0.95 M, respectively, solid percentage of 40.55 %, particle size of 26.45 μm, and reaction duration of 8.76 hours. This study defined the top five features that most significantly affect CO₂ sequestration efficiency from the selected ML model: (1) reaction time, (2) particle size, (3) NaHCO₃ concentration, (4) temperature, and (5) CO₂ partial pressure.