Modeling and optimization of CO2 capture in spray columns via artificial neural networks and response surface methodology
Parisa Tabarzadi, Ahad Ghaemi
Abstract
This study focuses on the optimization of CO2 capture in the spray columns, a critical process for mitigating greenhouse gas emissions and addressing climate change. The experimental data in the literature was used to optimize the mass transfer coefficient using Artificial Neural Networks (ANNs) and Response Surface Methodology (RSM) to enhance the efficiency of CO2 removal in the spray columns. A multi-layer feedforward ANN and a Radial Basis Function (RBF) model were developed and trained using 269 experimental data points, followed by testing with 89 data points. The models utilized inlet loading and pressure of CO2, liquid and gas flow rates, and Monoethanolamine (MEA) concentration as input parameters to predict outlet loading and pressure. The results demonstrated that a RBF model with a spread of 1.4 yielded a performance of 0.0006, resulting in a coefficient of correlation R2 of 0.999 for both responses with trainlm as an activation function. The Mean Square Error (MSE) for outlet loading was 0.0009, while for outlet pressure it was 0.0003 over 50 epochs. The results of the Multi-Layer Perceptron (MLP) and RBF models were compared in this study. In the RSM approach, we employed a quadratic model, yielding R2 values of 0.980 for outlet pressure and 0.997 for outlet loading, respectively. This research provides valuable insights into the potential of ANNs and RSM in optimizing carbon capture processes in spray columns.