Integration of Multiple Optimization Algorithms with Machine Learning: Predicting Nitrogen Adsorption Volume of Coal Rocks

The nitrogen adsorption capacity is of great significance for the characterization of porous materials, carbon sequestration, and coalbed methane development. Essentially, this is a typical research scenario of solid-gas interface interaction in the broad field of interface science. Accurate prediction of adsorption volume had proven challenging due to the highly nonlinear relationship between coal rocks complex characteristics and adsorption behavior. This study investigated coal rocks of varying particle sizes. Based on low-temperature liquid nitrogen adsorption experimental data, the pore structure characteristics were systematically analyzed. Random Forest (RF) and Support Vector Machine (SVM) models were subsequently employed, in combination with four optimization algorithms (Sparrow Search Algorithm (SSA), Snake Optimization (SO), Chameleon Swarm Algorithm (CSA), and Northern Goshawk Optimization (NGO)), to predict nitrogen adsorption volume. The results demonstrated that optimized models, particularly those enhanced by NGO and CSA, outperformed alternative approaches. Compared with SVM, the RF model achieved superior prediction accuracy. Among optimization algorithms, the RF model enhanced by NGO/CSA exhibited the best performance, attaining a coefficient of determination (<i>R</i>²) value of 0.9804. Regarding validation set performance, the optimized RF model showed substantial improvements, with a mean absolute error (MAE) of 0.1119, a mean absolute percentage error (MAPE) of 0.0545, a mean squared error (MSE) of 0.0389, and a root mean squared error (RMSE) of 0.1972. The <i>R</i>² value of the optimal model on the validation set was improved by nearly 10% compared with the original RF model. Coal rock size constitutes a key predictive factor for nitrogen adsorption volume. 120 to 200 mesh exhibits the most significant influence. This research highlights the importance of particle size as a predictive factor and validates the effectiveness of integrating optimization algorithms with machine learning models for predicting nitrogen adsorption capacity in coal rocks. It also provides a novel quantitative method for the study of the interaction between solid and gas interfaces in interface science.