Decoding the CO2 adsorption of nitrogen-doped carbon under variable temperature and pressure conditions: A machine learning guideline

Efficient capture and utilization of CO2 is the key to achieving carbon peak and neutrality. This study used the radial basis function neural network and extreme gradient boosting regression to build the regression models about CO2 adsorption uptake based on variable temperature and pressure swing adsorption dataset. Besides, the structure of the models was optimized by the particle swarm optimization algorithm. Results revealed that the extreme gradient boosting regression models showed better performance (Test R2 = 0.80–0.97). The relative importance diagram canulated by Shapley additive explanations showed that the pore structure was highly correlated with CO2 uptake. Furthermore, the partial dependence plot found that the carbon material adsorbent, whose micropore volume was greater than 0.6 cm3/g and N-5 content was between 3.5 and 4.5 wt%, had better adsorption performance, and the CO2 uptake of more than 3 mmol/g could be achieved. What's more, these models were integrated into an interactive web page by the Gradio library. This study offered a new idea about the preparation of high-performance nitrogen-doped carbon adsorbents under various adsorption conditions.