Machine learning-based improved pressure-volume-temperature correlations for black oil reservoirs

Pressure-volume-temperature (PVT) properties of crude oil are considered the most important properties in petroleum engineering applications as they are virtually used in every reservoir and production engineering calculation. Determination of these properties in the laboratory is the most accurate way to obtain a representative value, at the same time, it is very expensive. However, in the absence of such facilities, other approaches such as analytical solutions and empirical correlations are used to estimate the PVT properties. This study demonstrates the combined use of two machine learning (ML) technique, viz., functional network (FN) coupled with particle swarm optimization (PSO) in predicting the black oil PVT properties such as bubble point pressure (Pb), oil formation volume factor at Pb, and oil viscosity at Pb. This study also proposes new mathematical models derived from the coupled FN-PSO model to estimate these properties. The use of proposed mathematical models does not need any ML engine for the execution. A total of 760 data points collected from the different sources were preprocessed and utilized to build and train the machine learning models. The data utilized covered a wide range of values that are quite reasonable in petroleum engineering applications. The performances of the developed models were tested against the most used empirical correlations. The results showed that the proposed PVT models outperformed previous models by demonstrating an error of up to 2%. The proposed FN-PSO models were also compared with other ML techniques such as an artificial neural network, support vector regression, and adaptive neuro-fuzzy inference system, and the results showed that proposed FN-PSO models outperformed other ML techniques.