Hybrid Predictive Modeling of Wet Gas PVT Properties for Niger Delta Reservoirs
Akinsete Oluwatoyin Olakunle
*
Department of Petroleum Engineering, University of Ibadan, Ibadan, Nigeria.
Adeleye Emmanuel Oluwatunmininu
Department of Petroleum Engineering, University of Ibadan, Ibadan, Nigeria.
Falade Adetokunbo Ademola
Department of Petroleum Engineering, University of Ibadan, Ibadan, Nigeria and Department of Mining and Petroleum Resources Engineering, Federal Polytechnic Ado-Ekiti, Nigeria.
Rotimi Isaac Ayodele
Department of Mining and Petroleum Resources Engineering, Federal Polytechnic Ado-Ekiti, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Understanding Pressure-Volume-Temperature (PVT) properties has always been a major interest to reservoir engineers for fluid characterization, reserves estimation, recovery and design of facilities to ensure an efficient operating infrastructure either for processing or transportation. Existing PVT correlations often lacks interpretability and struggle to capture the intricate Niger Delta gas reservoirs. To address this gap, this study explores regression algorithms for the development of selected PVT properties of a wet gas reservoir in Niger Delta.
This work integrate both black box and white box modeling techniques to develop a hybrid predictive analytics models which formed the basis for optimal feature selection for the development of a simplified correlation from a blend of two linear models to predict PVT properties of Gas Compressibility factor (z-factor), Gas Formation Volume Factor (Bg) and Gas Viscosity (μg). The modelling architecture employs supervised machine learning regression algorithms that are developed on 1,111 wet gas data points with a 5-fold cross-validation technique. Statistical metrics of evaluation was used to validate performance.
Findings revealed that pseudopressure and gas viscosity are major determinants for gas z-factor, while gas density, gas viscosity, and pseudopressure are crucial for Bg. The hybrid models achieved AARE, R² and RMSE scores of 0.69%, 99.95% and 0.0067 for z-factor, 0.33%, 99.22% and 0.00023 for Bg, and 0.92%, 99.97% and 0.0056 for μg, respectively. Additionally, the developed mathematical correlations yielded R² of 99.68%, 94.67%, 99.45% for z-factor, Bg, and μg, respectively.
Hybrid wet gas Pressure-Volume-Temperature correlations from regression algorithms was developed for the Niger Delta region. The models ensures a contextualized and reliable representation of the region by reducing biases and improving correlation accuracy. The high accuracy and interpretability of the models suggests their utility in effective reservoir management and enhancing predictive modelling capabilities in reservoir engineering.
Keywords: Hybrid predictive model, PVT properties, black box algorithm, wet gas properties, statistical metrics, white box algorithm