Machine Learning Approaches for Compressibility Factor Prediction at High- and Low-Pressure Ranges

An accurate value of compressibility factor, also called Z-factor or deviation factor, is essential for petroleum engineering, especially for reservoir simulation and material balance calculations. Standing and Katz (S-K) charts were published as an industry reference to calculate Z-factor since 1942. After that, many direct and indirect methods were implemented that can fit the Standing and Katz charts and extend the approach for high-range of pressure and temperature data. The problem arises from the difficulty of having accurate, direct, and quick correlations that can fit with low- and high-range pressure data. The objective of this study is to apply different machine learning (ML) techniques such as artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) to predict the gas compressibility factor within high and low-pressure ranges with high accuracy. Additionally, an empirical equation was extracted based on the ANN model to convert the black box model into a white box. It was concluded that the ANN-based model outperformed the other models with a correlation coefficient (R) of 0.99 and an average absolute percentage error (AAPE) of 0.159. The proposed correlation was validated using new experimental high-pressure range data for Z-factor and compared with existing correlations in the literature showing a lower error, which indicates that the proposed correlation can be used for low and high ranges of pressure and temperature and can be evaluated without having special programs to run the ANN model.