Data-driven optimization of CO2 adsorption on KOH-activated hydrochars using hybrid RSM–ANN framework

The development of sustainable carbon adsorbents derived from biomass is driven by the need to address effective CO2 capture, which remains a significant challenge due to the intricate, nonlinear interactions between material properties and operating conditions in adsorption processes. In this study, a hybrid framework integrating response surface methodology (RSM) and artificial neural networks (ANNs) was developed to optimize the CO2 adsorption performance of KOH-activated hydrochars. A comprehensive database of experimental data related to five process variables, including SBET, temperature, pressure, activation temperature, and the KOH/precursor mass ratio, has been compiled based on recent studies in the literature. The RSM model achieved an R2 value of 0.986, indicating a good fit between predicted and experimental CO2 adsorption values. Analysis of variance indicated that pressure, temperature, and the KOH ratio are the most significant factors, revealing pronounced synergies among these variables. ANN modeling significantly enhanced predictive accuracy, with a two-layer multilayer perceptron (trainlm) achieving a mean squared error (MSE) of 4.5 × 10–4 and an R2 value of 0.998. In contrast, the radial basis function with 26 neurons and a spread of 2 achieved an MSE of 2.8 × 10–4 and an R2 of 0.999. The hybrid ANN–RSM model outperformed standalone methods, providing a robust, interpretable tool for optimizing advanced hydrochar adsorbents.