マグネシアベース吸着材への二酸化炭素吸着を予測する解釈可能なアンサンブル機械学習モデル

Accurately predicting the carbon dioxide (CO₂) adsorption capacity of magnesium oxide (MgO)-based sorbents is crucial for advancing carbon capture technologies. This study develops an interpretable machine learning framework to address this challenge. Fourteen candidate models were evaluated, with six top-performing ones selected for Bayesian hyperparameter optimization using the Optuna framework. A novel weighted ensemble model was then constructed by integrating the entropy weight method and the Technique for Order Preference by Similarity to Ideal Solution. The optimized ensemble model demonstrated outstanding performance on an independent test set, achieving an R² of 0.9903 with MSE, RMSE, and MAE of 0.361, 0.6008, and 0.2544 respectively, surpassing most individual models. SHapley Additive exPlanations analysis identified material morphology as the most significant feature (43.8% contribution), particularly granular morphology. Process parameters including time (23.3%) and temperature (14.1%) were the next major contributors, while pore structure variables collectively contributed 15.3%. The analysis further revealed complex non-linear dependencies and interaction effects, with the combination of medium temperature (200-400°C) and adsorption time (100-250 min) proving optimal. Additionally, a web application was deployed to provide real-time prediction capability. This work provides both a reliable predictive tool and valuable mechanistic insights for designing high-performance MgO-based CO₂ adsorbents.