Predicting saturates of sour vacuum gas oil using artificial neural networks and genetic algorithms

Accurate predictions of chemical composition by physical properties of sour vaccum gas oil (VGO) fractions are important for the refinery. In this paper, a feed-forward type network based on genetic algorithm (GA), was developed and used for predicting saturates of sour vacuum gas oil. The number of neurons in the hidden layer, the momentum and the learning rates were determined by using the genetic algorithm. The five physical properties of sour VGO, namely, average boiling point, density at 20 °C, molecular weight, kinematic viscosity at 100 °C and refractive index at 70 °C were considered as input variables of the ANN and the saturates of sour VGO was used as output variable. The study shows that genetic algorithm could find the optimal networks architecture and parameters of the back-propagation algorithm. Further, the artificial neural network models based on genetic algorithm are tested and the results indicate that the adopted model is very suitable for the forecasting of saturates of sour VGO. Compared with other forecasting models, it can be found that this model can improve prediction accuracy.