Interpretable Machine Learning for Accelerating Reverse Design and Optimizing CO2 Methanation Catalysts with High Activity at Low Temperatures

CO2 methanation represents a promising technological pathway for achieving efficient carbon dioxide resource utilization and mitigation of greenhouse gas emissions. However, the development of CO2 methanation catalysts with high activity at low temperatures (<250 °C) remains a formidable challenge. To address the time-consuming and costly nature of traditional catalyst development methods, this study proposes an interpretable machine learning (ML)-assisted reverse design framework for CO2 methanation catalysts. This framework integrates the advantages of the ML, interpretability analysis, and multiobjective optimization methods to elucidate the intricate interplay among catalyst compositions, preparation conditions, reaction parameters, and catalyst activity. A data set containing 2777 data points is established to construct various ML models. After fine-tuning the key hyperparameters of the four models, a comprehensive comparison is conducted to evaluate their predictive performance. The light gradient boosting machine (LGBM) model demonstrates superior predictive accuracy, attributed to its minimal toot mean squared error of less than 0.31 and the highest R2 value surpassing 0.90. An interpretable analysis is conducted to ascertain the most significant features and their impact on the outputs of the optimal LGBM model using postvalidation interpretation methods. It indicates that appropriately reducing active component content, first support content, calcination temperature, and inert gas content are favorable for the reaction. Finally, the LGBM model is coupled with the NGSA-III algorithm to maximize the CO2 conversion ratio and CH4 selectivity in CO2 methanation reactions. Three Ru- and three Ni-based new CO2 methanation catalysts have been successfully predicted and are recommended to have high activity at low temperatures. In particular, the optimized Ru–Ba/Cr2O3–SrO catalysts have a high CO2 conversion ratio higher than 97.04% and a CH4 selectivity higher than 72.22% at low temperatures.