Prediction of PVT Properties in Crude Oil Using Machine Learning Techniques MLT

Abstract

Abstract A new mathematical model is proposed using machine learning techniques for estimating PVT fluids properties such as bubble pressure and oil formation volume factor as a function of the solution gas-oil ratio, gas specific gravity, oil specific gravity, and temperature. The result obtained with this new approach are compared with previous published correlations. The proposed method for PVT properties estimation consists of two stages: data decorrelation through Principal Component Analysis (PCA) and PVT properties estimation through an Artificial Neural Network (ANN). Data decorrelation is used to reduce redundancy in the data, which decrease the number of neurons and hidden layers needed for an ANN to achieve a high accuracy estimation. In the development of the proposed method there were used 504 points obtained from the literature as follows: 360 for training, 40 for cross-correlation and 104 for testing. The present model was compared with empirical correlations of PVT fluids properties in terms of absolute average percent error, standard deviation, and correlation coefficient; using worldwide experimental PVT data. The results obtained show that the proposed model provides better estimation and higher accuracy than the published empirical correlations. The present model provides predictions of the bubble pressure and formation volume factor with a correlation coefficient approximately of 98%. Trend tests were performed to check the behavior of the predicted values of bubble pressure and formation volume factor for any change in reservoir temperature, solution gas-oil ratio, gas gravity and stock-tank oil gravity. The model was based on artificial neural networks, and developed using 504 published data sets from the Middle East, Malaysia, and North Sea fields. This improvement in PVT calculation accuracy will be of invaluable support for simulations and designs applied in Oil industry.