Experimental investigations and developing multilayer neural network models for prediction of CO₂ solubility in aqueous MDEA/PZ and MEA/MDEA/PZ blends

Abstract

In this research, a new set of experimental data for CO 2 solubility in aqueous blended amine solvents were investigated experimentally over the CO 2 partial pressure range from 8 to 100 kPa at 40°C and were compared with the benchmark aqueous 30 wt.% MEA solution. This work developed two multilayer neural network models named models A and B, for predicting the CO 2 solubility in various aqueous blended amine solvents including 36 wt.% MDEA + 17 wt.% PZ, 24 wt.% MDEA + 26 wt.% PZ, and 6 wt.% MEA + 25 wt.% MDEA + 17 wt.% PZ. Models A and B were developed by using Levenberg-Marquardt back propagation algorithm with 427 and 301 of reliable experimental data sets gathered from the published data, respectively. The results indicate that the high accuracy prediction of the CO 2 solubility in Methyldiethanolamine/Piperazine (MDEA/PZ) blends could be obtained by the network developed by Tan-sigmoid transfer function with two hidden layers consist of eight and four neurons, while the network developed by Tan-sigmoid transfer function with three hidden layers consist of 20, 10, and five neurons provided the highest accuracy for predicting the CO 2 solubility in MEA/MDEA/PZ blends comparing to other model structures. The comparison results show that the neural network modeling provided more closer predictions to the experimental results than the simulator and other thermodynamic models when predicting the CO 2 equilibrium solubility in blended amine solvents.

Novel Machine Learning Model Correlating CO2Equilibrium Solubility in Three Tertiary Amines

“…Accordingly, this work compared several network structures to improve the prediction performance. Based on the published work, , it was found that ANN might predict the CO 2 equilibrium solubility in single and blended amine solvents. However, the input variables should be properly adjusted to simplify the network based on the type of solvents.…”

“…Furthermore, it is known that ANN models require sufficient input data for training, so insufficient or excess input data potentially cause over- or underpredictions. In this case, most ANN models from previous work were developed based on sufficient experimental data obtained from different works. , For novel single and blended solvents, missing data are a common issue. If the novel solvent blends are blended by several conventional single solvents, the ANN model can be trained by using the single solvents directly .…”

Study of Methane Solubility Calculation Based on Modified Henry’s Law and BP Neural Network

“…Deng and Guo [13] developed an artificial neural network model to predict the products of CH4 bi-reforming using CO2 and steam, demonstrating its accuracy with correlation coefficients over 0.995 across various operational conditions. Li et al [14] developed two neural network models to predict CO2 solubility in aqueous blended amine solvents, using extensive experimental data and a backpropagation learning algorithm. The models demonstrated high accuracy, outperforming traditional thermodynamic models, and were particularly effective for complex blended amine systems like MDEA/PZ and MEA/MDEA/PZ.…”