A supervised model based on a Backpropagation Neural Network for simultaneous estimation of four interdependent properties: compressibility factor (Z), viscosity (μ), density (ρ) and gas formation volume factor (Bg) is proposed.
Abstract
Accurate prediction of natural gas thermophysical properties is essential for applications in production and transportation engineering, including reservoir simulation and flow modeling. Although machine learning (ML) techniques have been widely used, most studies focus on the estimation of these properties, with limited integration into practical applications. In this study, we propose a supervised model based on a Backpropagation Neural Network for simultaneous estimation of four interdependent properties: compressibility factor (Z), viscosity (μ), density (ρ) and gas formation volume factor (Bg). The multi-output model was trained on 58,165 data points generated from thermodynamic correlations, using pressure, temperature, composition (mole fractions of N2, CO2 and H2S), and gas specific gravity as inputs. The results yielded RMSE values of 5.56 × 10−4, 3.24 × 10−5, 3.01 × 10−2, and 6.33 × 10−4 for Z, μ, ρ and Bg, respectively, with R2 coefficients close to unity. The model’s applicability was evaluated by integrating the Z-factor into pressure drop calculations in pipelines using the Cullender and Smith method, resulting in a mean percentage error of 3.78%, close to the traditional method (3.83%). The results indicate that the model is an efficient and consistent alternative, highlighting the potential for integrating ML with classical hydraulic models.
Accurate prediction of dew point pressure (Pd) is critical for managing gas condensate reservoirs, as liquid dropout near the wellbore creates condensate banking that reduces permeability and well productivity. This study proposes a data-driven approach using a Random Forest (RF) machine learning algorithm to predict Pd based on reservoir temperature and fluid composition. Utilizing a comprehensive dataset of 375 records, the model incorporates 13 predictors, including hydrocarbon fractions (C1 through C7+), non-hydrocarbons (N2, CO2, H2S), and heavy fraction properties (MC7+, γC7+).
The proposed RF model exhibits highly competitive accuracy, outperforming most traditional empirical correlations by achieving a superior overall data variance capture with a Coefficient of Determination (R2) of 0.8790 and an Average Absolute Percent Relative Error (APE) of 8.17%. While the Ahmadi-Elsharkawy model shows a marginally lower APE (7.90%), the RF framework avoids complex genetic programming equations and delivers superior global consistency across the entire pressure envelope. By capturing complex, non-linear thermodynamic interactions, the RF model provides a robust, fast, and cost-effective alternative to expensive laboratory PVT tests and complex equations of state, optimizing fluid characterization and production system design.
Alejandro Osorio Pozo, Lucio A. Perez, Karim Botan et al.· Romanian Journal of Petroleu...· 0 citations
Solar air heaters are attractive for low-carbon thermal applications, but their performance is constrained by weak convective heat transfer in the near-wall region. In this work, a single-pass rectangular solar air heater duct equipped with spherical turbulators is investigated numerically and then accelerated using machine-learning surrogate models for rapid prediction of thermal–hydraulic responses. Five turbulator arrangements (V-, M-, W-shaped, inclined, and arc-shaped) were evaluated at three discrete spacing levels (pitch ratio P/D = 3, 6, 9) with a constant sphere diameter of 25 mm over Re = 3500–23 500, consistent throughout all the simulations. The computational fluid dynamics model employed a constant heat flux of 1000 W/m² and standard pressure–velocity coupling/discretisation practices, and was validated against established previous work. A leakage-safe machine-learning dataset (675 samples, 36 variables) was constructed from computational fluid dynamics outputs and physics-informed engineered features; models were trained using geometry-grouped cross-validation to ensure generalisation across turbulator arrangements and spacing levels. Among candidate regressors, histogram gradient boosting provided the best Nu surrogate (OOF RMSE = 3.299, R² = 0.9908, MAPE = 2.44%), while ridge regression yielded the most accurate friction factor surrogate (OOF RMSE = 4.70×10⁻⁴, R² = 0.9886, MAPE = 1.12%). The combined computational fluid dynamics and machine-learning framework enables fast, reliable evaluation of solar air heater thermo-hydraulic performance for configuration screening and optimisation within the validated operating envelope.
V. S. Bisht, Desh Bandhu Singh, Pushpendra Kumar et al.· Archives of Thermodynamics· 0 citations
This paper proposes an artificial neural network model, MLP 17-37-5, to predict the share of five natural gas components (methane, ethane, propane, nitrogen, and carbon dioxide) in the natural gas mixture in a pipeline network, depending on selected calendar and weather factors. Natural gas composition variability in the pipeline network results from supplying it with gas of varying composition and from different suppliers. Using statistical analysis of 35,064 actual measurement data sets, factors (modelinput data) significantly affecting gas composition variability were selected. Models differing in structure (the MLP 17-37-5 model or five MLP 17-37-1 models) and the number of neurons in the hidden layer (from 20 to 230 neurons) were trained using sets ranging from 8,760 to 35,064 actual data points obtained using the chromatographic method. The quality of the models was assessed based on the correlation coefficient, while the quality of the forecasts was assessed based on the nRMSE error of the forecasts obtained for a new data set of 8,760 data points. The MLP 17-37-5 model was shown to predict natural gas composition with an average error of nRMSE = 0.321. The article proposes a ready-made tool, which has no equivalent in the literature, allowing for a significant reduction in the number of chromatographic tests performed to determine the composition of natural gas. This is a completely new approach to studying changes in the composition of natural gas over time, which in the proposed forecasting model depend on selected factors (not analyzed in chromatographic studies).
J. Szoplik, Paulina Muchel· Chemical and Process Enginee...· 0 citations
Reliable estimation of the minimum miscibility pressure (MMP) is a critical requirement for the successful design of gas injection processes in enhanced oil recovery, as miscibility strongly controls displacement efficiency and recovery performance. Direct laboratory measurements of MMP are expensive and time-intensive, while traditional empirical correlations often fail to account for the complex interactions between fluid composition and reservoir conditions, leading to limited predictive reliability. To overcome these challenges, an interpretable data-driven framework based on a stacking ensemble learning strategy is developed for MMP prediction. The framework combines Random Forest, Gradient Boosting, Support Vector Regression, and XGBoost models and incorporates polynomial feature expansion together with recursive feature elimination to explicitly capture nonlinear and interaction effects among gas composition, oil composition, and thermodynamic variables. Model performance was assessed using an independent test dataset and evaluated using the root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). The optimized ensemble model demonstrates a substantial improvement over individual base learners, achieving an RMSE of 2.21, an MAE of 1.317, and an R² of 0.912. Beyond predictive accuracy, model interpretability was enhanced through SHapley Additive exPlanations and partial dependence analysis. The results reveal that the interaction between reservoir temperature and the molecular weight of heavy oil fractions exerts the strongest influence on MMP, followed by key compositional interactions involving nitrogen and carbon dioxide. These trends are physically consistent with established miscibility and phase-behavior principles. The main contribution of this work lies in the integration of an optimized stacking ensemble with interaction-focused feature engineering and explainable artificial intelligence techniques. This combination enables accurate, transparent, and physically meaningful MMP predictions, advancing beyond previous studies that emphasize either accuracy or interpretability alone. The proposed framework offers a practical and reliable tool for supporting gas injection design and decision-making in diverse reservoir systems.
M. Ahmadi· Journal of Petroleum Explora...· 0 citations
Accurate prediction of non-Newtonian nanofluid flow and heat transfer under melting heat conditions is essential for applications in advanced thermal management systems, polymer extrusion and coating processes, metallurgical melting operations, biomedical fluid transport, micro- and nano-scale heat exchangers, and energy systems. Motivated by these practical applications, the current work systematically examines the Williamson-Casson nanofluid flow over a slippery curved expanding surface while simultaneously incorporating the Cattaneo-Christov heat flux model. The flow is affected by melting heat transfer conditions at boundary, activation energy, chemical reactivity and Soret/Dufour effects. The modeled equations have been solved through the bvp4c approach in dimensionless form. The solution obtained from this approach is then used to provide a dataset for Artificial Neural Network approach. It has discovered in this work that velocity profiles are augmented with growth in thermal buoyancy factor and velocity slip parameter, while declining with augmentation in magnetic parameter, ferrohydrodynamic interaction and Weissenberg number. Thermal profiles escalated with augmentation in heat source/sink parameter, heat dissipation factor, unsteadiness factor, radiation factor and Dufour number, while declining with augmentation in melting heat parameter, thermal buoyancy factor and thermal relaxation factor. The analysis of the histograms across the entire dataset by plotting the frequency of errors within discrete intervals confirms the validity and reliability of the model. With growth in local Weissenberg number, Casson factor and curvature parameter there is augmentation in skin friction. For higher Prandtl number and melting heat parameter there is intensification in Nusselt number. With escalation in Schmidt number and Soret number there is augmentation in Sherwood number while higher factor of activation energy causes reduction in it.
E. Algehyne, I. A. A. Manahill, M. Rabih et al.· Discover Nano· 0 citations
The Minimum Miscibility Pressure (MMP) between injected CO2 and reservoir oil is a critical parameter for the successful design and implementation of miscible CO2enhanced oil recovery (EOR) projects. Accurate and rapid determination of MMP is essential for optimizing injection pressure and maximizing sweep efficiency. Traditional experimental methods, such as the slim-tube test, are time-consuming and expensive, while existing empirical correlations often lack generalizability due to reliance on limited data or simplistic fluid characterization. This study proposes a novel, physics-informed approach for predicting CO2–oil MMP by integrating reservoir fluid thermodynamics with advanced machine learning techniques. Dimensional analysis was first used to identify key dimensionless groups that inherently capture the complex interplay of reservoir temperature, oil composition (characterized by molecular weight and mole fractions of C5–C6 and C7⁺ components), and CO2properties. These dimensionless groups serve as physically meaningful input features, significantly reducing the dimensionality of the problem and improving the physical consistency of the model. Subsequently, optimized Neural Architectures, specifically a Bayesian-Optimized Deep Neural Network (BO-DNN) and a Physics-Informed Neural Network (PINN), are developed and trained on a comprehensive dataset of experimental MMP values. The BO-DNN is optimized for hyperparameter selection to maximize predictive accuracy, while the PINN incorporates the relevant phase behavior constraints and equations of state (EoS) as soft constraints in its loss function, enforcing adherence to fundamental thermodynamic principles. To evaluate performance, the proposed Physics-Informed Neural Network (PINN) and Bayesian-Optimized Deep Neural Network (BO-DNN) were benchmarked against industry-standard empirical models, including the Yellig-Metcalfe, Glaso, and Alston et al. correlations. The results demonstrate a significant paradigm shift in predictive accuracy. While the Alston et al. and Glaso correlations exhibited limited reliability with coefficients of determination (R2) of 0.78 and 0.81 respectively, and Mean Absolute Percentage Errors (MAPE) often exceeding 18-22%, the proposed dimensionless framework achieved an R2 above 0.94. Specifically, the optimized neural architectures achieved a MAPE reduction of 15% to 25% compared to the best-performing traditional algorithms, significantly lowering the Root Mean Square Error (RMSE). The PINN, in particular, demonstrated superior predictive stability by enforcing Equation of State (EoS) constraints, preventing the physically inconsistent "drifting" often seen in legacy formulas. By providing a high-fidelity, cost-effective alternative to laboratory experiments, this research offers a resilient tool for real-time field optimization, aligning with the digital transformation goals of the contemporary energy landscape.
M. O. Oyegbile· SPE Nigeria Annual Internati...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
