A neural network approach to predict activity coefficients

Abstract

Artificial neural networks (ANNs) and a group-contribution approach were used to develop an algorithm to predict activity coefficients for binary solutions. The Levenberg-Marquardt algorithm was used to train the ANN and to predict the parameters of the Margules equation. The ANN was trained using phase-equilibrium database from DECHEMA. The selected systems include alcohols, phenols, aldehydes, ketones, and ethers. The trim mean based on 20% data elimination was selected as the best representation of the Margules-equation parameters. The algorithm was validated with 121 VLE systems and results show that the ANN provides a relative improvement over the UNIFAC method.Des réseaux de neurones artificiels (RNA) et une méthode de contribution de groupe ontété utilisés pour créer un algorithme permettant de prédire les coefficients d'activité des solutions binaires. L'algorithme de Levenberg-Marquardt aété utilisé pour entraîner le RNA et pour prédire les paramètres de l'équation de Margules. Le RNA aété entraîné en utilisant la base de données du diagramme d'équilibre de DECHEMA. Les systèmes sélectionnés comprennent les alcools, les phénols, les acétaldéhydes, les cétones et leséthers. La moyenne basée sur l'élimination de 20% des données aété sélectionnée commeétant celle représentant le mieux les paramètres de l'équation de Margules. L'algorithme aété validé avec 121 systèmes d'ELV et les résultats indiquent que le RNA conduità une amélioration relative par rapport au modèle UNIFAC.

Current State of Atmospheric Aerosol Thermodynamics and Mass Transfer Modeling: A Review

“…A modified Pankow [182] type mass partitioning between the gas and aerosol phase that accounts for the Kelvin effect can be used for relatively volatile compounds but for LVOC compounds a surface condensation scheme (e.g., [274]) should be adopted. A neural-network trained two-parameter or four-parameter Margules equation [168,392] can be used to obtain more accurate binary organic activity coefficients. The Margules equation is based on the observation that log(a i ) is approximately a linear function of the mole fraction of constituent i in the liquid phase.…”

“…Machine learning could be applied to determine the optimal reduced aerosol composition that is required. Ramírez-Beltrán et al [392] trained a neural network combined with a group-contribution approach to obtain parameters for the four-parameter Margules equation. The activity coefficient predictions were slightly better than those of the UNIFAC model.…”

Designing and Testing a Chemical Demulsifier Dosage Controller in a Crude Oil Desalting Plant: An Artificial Intelligence‐Based Network Approach

“…This technique was selected because historical data for this system is available and because it has the ability to adapt itself to sudden and drastic (nonlinear) changes that may occur during the training phase of the neural network. Applications of neural networks can be found in different fields of science and engineering [17][18][19][20].…”

Predictive Modeling of Henry’s Law Constant in Chemical Structures Using LSSVM and ANFIS Algorithms

“…In the literature, there are some approaches to predict H of organic compounds in water based on chemical structure directly. Additionally, a number of indirect approaches for prediction of H based on vapor-liquid equilibrium data including activity coefficient, however their applications for prediction of the H are not exactly assessed [5,6]. Consequently in this paper, we focus on those approaches which can predict the H directly.…”