Khoshraftar Zohreh
School of Chemical, Petroleum and Gas Engineering, Iran University of Science and Technology, P.O. Box: 16765-163, Tehran, Iran.
Sci Rep. 2025 Jan 13;15(1):1800. doi: 10.1038/s41598-025-86144-2.
In this study, Response Surface Methodology (RSM) and Artificial Neural Networks (ANN) were developed to estimate the equilibrium solubility and partial pressure of CO in blended aqueous solutions of diisopropanolamine (DIPA) and 2-amino-2-methylpropanol (AMP). In this study, several key parameters were analyzed to understand the behavior of the aqueous DIPA/AMP system for CO capture. Including DIPA (9-21 wt%), AMP (9-21 wt%), temperature (323.15-358.15 K), pressure (2.140-332 kPa) and CO solubility (0.0531-0.8796 mol/mole). The results of the RSM analysis for CO solubility indicate that the model demonstrates a strong fit, as evidenced by a Pred-R² of 0.9601, an adjusted R² of 0.9481, and a highly significant F-value of 80.22. The high predicted R² of 0.9601 and 0.9292 values for CO solubility and CO partial pressure indicate that the predictor variables can explain a substantial amount of the variability in the response variable. The multilayer perceptron (MLP) architecture demonstrated strong correlation capabilities, featuring one hidden layer with 10 and 5 neurons, respectively. Its topology was structured as 4-10-1 for predicting CO solubility and 4-5-1 for predicting CO partial pressure. The accuracy of the predictions was notably high, with coefficients of determination of 0.99581 for CO solubility and 0.99839 for CO partial pressure, achieved using the Levenberg-Marquardt algorithm. Upon further analysis, it was concluded that the MLP model exhibited the lowest error rates, with mean square errors of 0.00009085 for CO solubility and 0.00316632 for CO partial pressure. The findings emphasized that the MLP model not only outperformed the RSM model in accuracy but also demonstrated greater adaptability in handling the intricate variables associated with CO solubility and partial pressure in capture technologies.
在本研究中，开发了响应面法（RSM）和人工神经网络（ANN）来估算二异丙醇胺（DIPA）和2-氨基-2-甲基丙醇（AMP）混合水溶液中CO的平衡溶解度和分压。在本研究中，分析了几个关键参数，以了解用于CO捕集的DIPA/AMP水溶液体系的行为。包括DIPA（9-21 wt%）、AMP（9-21 wt%）、温度（323.15-358.15 K）、压力（2.140-332 kPa）和CO溶解度（0.0531-0.8796 mol/mole）。CO溶解度的RSM分析结果表明，该模型拟合良好，预测决定系数（Pred-R²）为0.9601，调整后决定系数（adjusted R²）为0.9481，F值为80.22，具有高度显著性。CO溶解度和CO分压的预测决定系数分别高达0.9601和0.9292，表明预测变量可以解释响应变量中相当大的变异性。多层感知器（MLP）架构显示出很强的相关能力，其分别具有一个包含10个和5个神经元的隐藏层。其拓扑结构为用于预测CO溶解度的4-10-1和用于预测CO分压的4-5-1。使用Levenberg-Marquardt算法，预测的准确性非常高，CO溶解度的决定系数为0.99581，CO分压的决定系数为0.99839。进一步分析得出结论，MLP模型的错误率最低，CO溶解度的均方误差为0.00009085，CO分压的均方误差为0.00316632。研究结果强调，MLP模型不仅在准确性上优于RSM模型，而且在处理捕集技术中与CO溶解度和分压相关的复杂变量方面表现出更大的适应性。
