Modeling the hydrogen solubility in methanol, ethanol, 1-propanol and 1-butanol

Abstract

Modeling hydrogen solubility in primary normal alcohols (methanol, ethanol, 1-propanol and 1-butanol) has been studied in this article. Equations of state (EOS), simple correlations and Artificial Neural Networks (ANN) have been compared to find the best modeling technique. Utilizing an equation of state requires an iterative calculation procedure and optimized interaction parameters. Iterative calculation is not suitable when time is important and optimized interaction parameters are not always available. In addition, selection of proper equation of state and mixing rules are serious problems. Simple correlations can be applied to avoid iterative calculations but they have limited flexibility. Artificial Neural Network is an alternative to traditional techniques. Neural networks are flexible and after training, they are very fast. 2-3-1 networks have been used to model hydrogen–alcohol systems and negligible errors indicate reliability of this method. However, high performance of 2-3-1 neural networks is limited to systems which networks have been trained for. In addition, number of adjustable parameters in 2-3-1 networks is a great disadvantage. Number of carbon atoms has been used to train one network for all studied systems. 3-4-1 network has been trained and tested, and Average Relative Deviation (ARD) has been calculated 5% and 3% in training and testing stages, respectively. Beside excellent accuracy, 3-4-1 network has less adjustable parameters and can provide good estimations for similar systems.