Using Artificial Neural Network Modeling for Libyan PVT Properties

PVT properties of reservoir fluid are very important in petroleum engineering calculations, therefore the accuracy of the calculations depends on the exactness of PVT properties. Ideally these properties are determined from laboratory analysis of the samples. In some cases PVT data are not available or reliable. At these occasions, empirical correlations are used which are developed for PVT properties estimations. Accuracy of the correlations depends on similarity of fluid properties and fluid that used for developing correlations, thus results of the predictions may not be accurate for new samples. Hence Artificial Neural Network (ANN) has been applied as new technique for PVT properties estimation for Libyan crude oils.

The neural-network models were developed using back-propagation with momentum for error minimization to obtain the most accurate PVT models. This study aims to evaluate some PVT properties to estimate Gas solubility (RS), oil formation volume factor (FVF) below bubble point, dead oil viscosity, oil density (Po) and bubble point pressure (Pb) for Libyan crudes using 1000 data points for about 250 PVT report samples that I have collected exclusively from Libyan oil fields, mainly Sirte, Ghadames and Murzuq basins.

Existing correlations are applied to Libyan data set and error analysis is performed based on a comparison of the predicted value with the original experimental value. Best correlation has been identified for each PVT parameter. Correlations performance will be compared with intelligent model (Artificial Neural Network).

The results show that artificial neural networks, once successfully trained, are excellent reliable predictive tools for estimating Libyan crudes oil PVT properties better than available correlations.