mohammadjavad ebrahimkhani

Recently, optimization methods have been extensively applied in phase equilibrium calculations. Among these methods, Particle Swarm Optimization (PSO) can be used to calculate the interaction parameters of activity coefficient models in equilibrium systems. In this study, based on the particle swarm optimization, the interaction parameters of 5  activity coefficient models (2-suffix Margules, 3-suffix Margules, Wilson, NRTL and UNIQUAC) have been calculated for 20 ternary extraction systems (water + carboxylic acids + organic solvents) including 126 tie-lines. The values of binary interaction parameters of these models along with the root mean square deviations (RMSD) are reported. The mean values of RMSD of the systems in the order of the mentioned models have been calculated 0.0294, 0.0041, 0.0114, 0.0016 and 0.0034, respectively. The results show that all models except 2-suffix Margules model have relatively good accuracy. By comparing the RMSD values in literature and the values determined by PSO for NRTL and UNIQUAC models, the RMSD values of the models improved from 0.0124 and 0.0181 to 0.0016 and 0.0034, respectively.

One of the main processes in the refining industries of the oil industry is the extraction of aromatic hydrocarbons from aliphatic hydrocarbons. Accordingly, accurate prediction of the phase behavior of these systems can improve liquid-liquid extraction.  In this study, the phase thermodynamic behavior of the ternary system of aliphatic and aromatic hydrocarbons with ionic liquids is predicted by the Adaptive Neuro-Fuzzy Inference System (ANFIS) and the Multilayer Perceptron (MLP)neural network. The model inputs were considered in modeling the liquid-liquid extraction system, the molar ratio of aliphatic, aromatic, and ionic compounds in the feed, as well as the molecular mass of the ions and the temperature of the extraction system, and the model output was the molar ratio. Aliphatic and aromatic compounds in the alkane-rich phase and molar ratio of aromatic compounds and ionic liquids in the iron-rich phase were considered. The design parameters of these neural networks, including the number of neurons and the clustering radius of the MLP and ANFIS networks, were optimized by the genetic algorithm evolution method (GA) in order to improve their prediction accuracy. Comparison of prediction accuracy of ANFIS and MLP networks with experimental data based on statistical parameters R2, RMSD, and MAD for ANFIS model was calculated 0.9999, 0.0190, and 0.0129 respectively and for MLP neural network model was 0.996, 0.0204, and 0.0127 respectively. Also, a comparison was made between the prediction accuracy of ANFIS, MLP networks and the NRTL thermodynamic model for two different liquid-liquid extraction systems, their RMSD for the two extraction systems were 0.0093, 0.0110, and 0.0113, respectively. The results of statistical parameters show that these networks have relatively good accuracy in predicting the thermodynamic behavior of liquid-liquid equilibrium and are an effective method.