Determination of critical properties and acentric factors of petroleum fractions using artificial neural networks

Various correlations are available that can determine the critical properties and acentric factors of petroleum fractions. The available methods may have low accuracy in determining these properties for heavy petroleum fractions and may require further verification because, during the development of the original predictive methods, the data describing the critical properties and acentric factors of heavy hydrocarbons and petroleum fractions were not available. In this work, after a quick review of the most common correlations reported in the literature, an alternative method based on the artificial neural network (ANN) technique is proposed to predict the critical temperatures, critical pressures, critical volumes, and acentric factors of petroleum fractions, especially heavy fractions, from their specific gravity and the average normal boiling-point temperature values. Among the different neural networks reported in the literature, the feed-forward neural network method with a modified Levenberg−Marquardt optimization algorithm is used. The model is trained and tested using the data recommended in the literature for critical properties and acentric factors of C1−C45 petroleum fractions. Independent data (not used in training and developing the model) are used to validate and examine the reliability of this tool. The predictions of this model are found in acceptable agreement with the data recommended in the literature, demonstrating the reliability of the ANN technique used in this work.