Machine learning to assess CO2 adsorption by biomass waste

Biomass Waste Derived Porous Carbon (BWDPC) is widely used for its ability to adsorb carbon dioxide (CO2) in large-scale industrial operations, making it a leading solution for combating air pollution and climate change issues. However, factors such as temperature, pressure, and surface area can influence its performance in adsorbingcarbon dioxide. To maximize CO2 adsorption, it is vital to determine the relationships among these variables. In this paper, we use several preprocessing techniques and various machine learning algorithms, such as Gradient Boosting Regressor, Convolutional Neural Networks, Multi-Layer Perceptron, and Long Short-Term Memory, to explore the efficacy of these algorithms in predicting CO2 capture capacities. We augment our datasets by generating new features and in turn our ML models achieve a better performance on the augmented datasets compared to the original dataset. Our models achieved r2 score of 90.7 % on the training set and 85.73 % on the test set of the augmented datasets. Furthermore, we were able to determine that the ratio of carbon to pressure as well as temperature, and aspects tied to the physical conditions of the adsorbent material emerge as the most influential factors in CO2 adsorption. A python implementation of all our experiments is publicly available in Github https://github.com/mmaheri/CO2_Capturing.git.