Application of radial basis function neural networks in bubble point oil formation volume factor prediction for petroleum systems

- Radial basis function (RBF) network was used for modeling bubble point oil FVF.
- A large data bank of experimental data sets was used to develop the model.
- Comparative studies were done between RBF model and several empirical correlations.
- Results indicated the superiority of the developed model to empirical correlations.
- Leverage approach confirmed the developed RBF model as being statistically valid.

This paper presents a powerful and comprehensive predictive model based on radial basis function (RBF) neural networks to predict the bubble point oil formation volume factor (FVF), which is one of the most important pressure-volume-temperature properties of crude oils. For this purpose, a large reliable data bank covering a wide range of various crude oil samples was used, with the data collected from the open literature. The performance of the proposed model for the prediction of the bubble point oil FVF was evaluated, using statistical and graphical error analyses, against a number of well-known predictive empirical correlations. The results indicated that, the developed RBF model is able to provide a strong agreement between the predicted values and corresponding experimental data, with an average absolute percent relative error and a coefficient of determination of 1.4562% and 0.9887, respectively, making it more accurate and reliable than the published empirical correlations. In addition, the leverage approach showed that the developed model was statistically acceptable and valid, and only six data points may be considered as probable outliers.