Application of artificial neural network and general machine learning modelling on CO ₂ adsorption in moisture equilibrated South African high and medium rank coals

This study investigates the effect of moisture on CO 2 adsorption in South African coals using both experimental and machine learning approaches. Three coal samples (SL, TN, and EM) with varying ranks ( R o V mr : 3.49%, 1.26%, and 0.64%, respectively) were collected from different regions of South Africa. The samples were moisture‐equilibrated at different levels (0–4.4 wt.%) and tested using a high‐pressure volumetric adsorption system (HPVAS) at 35°C and up to 90 bar. Adsorption isotherms were modelled using Langmuir, Freundlich, and Temkin models. The Langmuir model consistently provided the best fit, with high R 2 values of at least 0.97 across all moisture levels in all the coals. Results showed a significant drop in CO 2 uptake in the TN coal sample with increasing moisture, up to 77% at the highest moisture level (4.45 wt.%), confirming that water competes with CO 2 for sorption sites and blocks access to micropores. To model the experimental data, artificial neural networks (ANNs) and other machine learning methods, including Gaussian process regression, support vector machines, and decision tree‐based models, were trained using coal properties, moisture content, and pressure as inputs. The binary decision trees ensemble model achieved the best performance with R 2 above 0.99, and the lowest error metrics of 0.0174 for MSE, 0.1319 for RMSE, and 0.0728 for MARE, proving the ability of the model to accurately predict CO 2 adsorption under varying moisture conditions. The findings from this combined experimental and modelling approach supports future assessments of deep coal seams for long‐term storage.