Prediction and experimental verification of CO₂ adsorption on Ni/DOBDC using a genetic algorithm-back-propagation neural network model

A model combined a back-propagation neural network (BPNN) with a genetic algorithm (GA) based on experimental data as training samples was established to predict the CO₂ adsorption capacity for metal organic frameworks (MOFs) of Ni/DOBDC. The random function of the conventional BPNN model was modified by the GA-BPNN model for optimizing the initial weights and bias nodes. The amounts of adsorbed CO₂ and corresponding isosteric heat of adsorption on Ni/DOBDC were synchronously studied within a wide temperature range (25-145°C) and pressure range (0-3.5 MPa). The predicted results of the proposed GA-BPNN model and those of theoretical models and a BPNN model were compared with the experimental data. The proposed model provided a more accurate prediction than those of the theoretical models and BPNN model. In particular, the theoretical models were invalid in the low-pressure range (0-0.1 MPa).