Masoud Mofarahi

A multi-layer perceptron neural network (MLPNN) model with Levenberg–Marquardt learning algo-rithm were applied to model CO2capture by a novel amine solution called 4-diethylamino-2-butanol(DEAB). The MLPNN model predicted the CO2concentration and temperature profiles along the heightof the packed column as the model output. Inlet feed conditions of the absorber column (flue gas andamine) were selected as the inputs of the MLPNN model. Experimental data about random and structuredpacked columns were extracted from the literature and used to train the MLPNN model. In addition, asystematic procedure, i.e. Taguchi method, was applied to obtain the significant sequence of processparameters affecting CO2removal efficiency and to optimize the variables in the absorber column. Fivelevels of five variables, including lean amine temperature, amine concentration, CO2loading of amine,gas temperature, and amine flow rate, were used for the optimization of the absorber column. Theaverage absolute relative deviations (AARD) between the predicted results and the experimental datasuggested that our MLPNN model could predict CO2concentration and temperature profiles along thepacked column (AARD% = 5.47 and 3.61, respectively). The signal to noise ratio analysis of the Taguchimethod yielded a significant sequence of factors affecting CO2removal efficiency in the packed column(CO2loading > amine flow rate > amine concentration > gas temperature > amine temperature). This studydemonstrated the acceptable accuracy of the MLPNN and Taguchi method in, respectively, the modelingand optimization of CO2capture in amine solutions.