Artificial intelligence and response surface methodology to predict CO2 capture using piperazine‐modified activated alumina

Abstract

AbstractCarbon dioxide (CO2) is a major greenhouse gas that causes global warming. In this study, piperazine‐modified activated alumina was used for CO2 capture process. The number of 626 experimental data were obtained at different operating conditions to develop the artificial neural networks (ANNs) and response surface methodology (RSM), employed to identify the behavior and performance of the CO2 capture process. Three independent factors, including time (t: 0–1800 s), temperature (T: 20–80°C), and pressure (P: 2.067–10.088 bar), were used as the input variables and CO2 adsorption capacity was used as the response. Among 243 different structures, the multilayer perceptron (MLP) model was optimized with Bayesian regularization backpropagation algorithm and two hidden layers with 17 and 16 neurons. The radial basis function (RBF) model was used with 367 different structures. The optimized structure of the RBF model was obtained with 100 neurons and the spread of 0.5. For optimal networks of MLP and RBF, the best mean square error (MSE) was observed at 1.72e‐04 and 8.28e‐05, and the best coefficient of determination (R2) was obtained 0.998 and 0.999, respectively. Moreover, the response surface methodology (RSM) was used to model the process. The best MSE and R2 values were obtained 0.00454 and 0.9462, respectively. The results showed that ANNs produced more accurate predictions than RSM.