Modeling hydrocarbon surface tension using MLP and RBF neural networks and evolutionary optimization algorithms

Surface tension (ST) is a significant physical property which impacts the efficiency of many industrial operations such as separation processes, surface adsorption, and enhanced oil recovery techniques in petroleum industry. Various empirical models are available for determination of hydrocarbon ST, but due to limitations of these methods, more practical and novel models are still needed for accurate calculation of hydrocarbons ST. In this article, multilayer perceptron and radial basis function neural network models were applied for determination of ST. Different evolutionary algorithms including back propagation (BP), particle swarm optimization (PSO) and imperialist competitive algorithm (ICA) were implemented for models optimization. To develop the proposed models, 982 experimental data points for 67 types of hydrocarbon compounds were used. Different statistical parameters were utilized for verification of the performance of the developed models. It was found that all models developed in the present article have higher efficiency and accuracy in prediction of hydrocarbons ST in comparison with empirical models. Results revealed that the average absolute relative error for MLP-PSO and MLP-ICA models are 0.87% and 0.80%, respectively, while for RBF network is only 0.76%. It was concluded that RBF network is more accurate for predicting hydrocarbons ST than other models developed in this article.