Jiulong Wang

Permeability is a fundamental property governing fluid transport in porous media and is critical to enhanced oil recovery. Nevertheless, existing methods remain limited by experimental uncertainty, computational cost, and insufficient interpretability. This study proposed an enhanced permeability prediction equation. The equation was developed by combining machine learning with theoretical analysis. The proposed equation outperformed the calibrated Kozeny–Carman (K–C) and baseline models and achieved performance comparable to that of XGBoost. It therefore provides a concise and physically interpretable description of permeability within the adopted pore network modeling framework.

Relative permeability is a critical parameter governing multiphase fluid flow through porous media, significantly impacting recovery efficiency and CO₂ sequestration potential in geological reservoirs. Accurately evaluating relative permeability in heterogeneous reservoirs remains challenging due to spatially variable porosity and permeability distributions. This study presents a novel intelligent prediction approach for evaluating water-CO₂ relative permeability in heterogeneous porous media by integrating fluid properties, heterogeneity characteristics, and relative permeability measurements from uniform porous media. We established a comprehensive training dataset through systematic micromodel experiments that captured various heterogeneity patterns and fluid conditions. Using this dataset, we developed an Artificial Neural Network (ANN) model that achieved exceptional accuracy with a Mean Squared Error below 0.0025. The model was then applied to predict relative permeability in heterogeneous reservoirs using site-specific relative permeability data obtained from core experiments as input parameters. To validate our approach, we incorporated the predicted relative permeability values into Computer Modelling Group (CMG) reservoir simulations of CO₂ sequestration in saline aquifers. The simulation results demonstrated strong agreement with published literature, confirming the model’s predictive capability. This work provides a practical, efficient, and reliable methodology for predicting relative permeability in heterogeneous reservoirs, addressing a significant challenge in reservoir characterization and flow modeling.

The cornerstone of the global economy, oil and gas reservoir development, faces numerous challenges such as resource depletion, operational inefficiencies, safety concerns, and environmental impacts. In recent years, the integration of artificial intelligence (AI), particularly artificial general intelligence (AGI), has gained significant attention for its potential to address these challenges. This review explores the current state of AGI applications in the oil and gas sector, focusing on key areas such as data analysis, optimized decision and knowledge management, etc. AGIs, leveraging vast datasets and advanced retrieval-augmented generation (RAG) capabilities, have demonstrated remarkable success in automating data-driven decision-making processes, enhancing predictive analytics, and optimizing operational workflows. In exploration, AGIs assist in interpreting seismic data and geophysical surveys, providing insights into subsurface reservoirs with higher accuracy. During production, AGIs enable real-time analysis of operational data, predicting equipment failures, optimizing drilling parameters, and increasing production efficiency. Despite the promising applications, several challenges remain, including data quality, model interpretability, and the need for high-performance computing resources. This paper also discusses the future prospects of AGI in oil and gas reservoir development, highlighting the potential for multi-modal AI systems, which combine textual, numerical, and visual data to further enhance decision-making processes. In conclusion, AGIs have the potential to revolutionize oil and gas reservoir development by driving automation, enhancing operational efficiency, and improving safety. However, overcoming existing technical and organizational challenges will be essential for realizing the full potential of AI in this sector.

Carbon capture and storage (CCS) is the most promising method of curbing atmospheric carbon dioxide levels from 2020 to 2050. Accurate predictions of geology and sealing capabilities play a key role in the safe execution of CCS projects. However, popular forecasting methods often oversimplify the process and fail to guide actual CCS projects in the right direction. This study takes a specific block in Shenhua, China as an example. The relative permeability of CO₂ and brine is measured experimentally, and a multi-field coupling CO₂ storage prediction model is constructed, focusing on analyzing the sealing ability of the block from the perspective of injection modes. The results show that when injected at a constant speed, the average formation pressure and wellbore pressure are positively correlated with the CO₂ injection rate and time; when the injection rate is 0.5 kg/s for 50 years, the average formation pressure increases by 38% and the wellbore pressure increases by 68%. For different injection modes, the average formation pressures of various injection methods are similar during injection. Among them, the pressure increases around the well in the decreasing injection mode is the smallest. The CO₂ concentration around the wellbore is the largest, and the CO₂ diffusion range continues to expand with injection time. In summary, formation pressure increases with the increase in injection rate and injection time, and the decreasing injection mode has the least impact on the increase in formation pressure. The CO₂ concentration is the largest around the well, and the CO₂ concentration gradually decreases. The conclusion helps determine the geological carrying capacity of injection volumes and provides insights into the selection of more appropriate injection modes. Accurate predictions of CO₂ storage capacity are critical to ensuring project safety and monitoring potentially hazardous sites based on reservoir characteristics.

In the last decade, low-quality unconventional oil and gas resources have become the primary source for domestic oil and gas storage and production, and hydraulic fracturing has become a crucial method for modifying unconventional reservoirs. This paper puts forward a framework for predicting hydraulic fracture parameters. It combines eXtreme Gradient Boosting and Bayesian optimization to explore data-driven machine learning techniques in fracture simulation models. Analyzing fracture propagation through mathematical models can be both time-consuming and costly under conventional conditions. In this study, we predicted the physical parameters and three-dimensional morphology of fractures across multiple time series. The physical parameters encompass fracture width, pressure, proppant concentration, and inflow capacity. Our results demonstrate that the fusion model applied can significantly improve fracture morphology prediction accuracy, exceeding 0.95, while simultaneously reducing computation time. This method enhances standard numerical calculation techniques used for predicting hydraulic fracturing while encouraging research on the extraction of unconventional oil and gas resources.

Recent advancements in artificial intelligence (AI) technology have offered new ways to obtain the relative permeability curve that is crucial for subsurface engineering problems. The existing data-driven AI approaches for the relative permeability prediction are only suitable for a specific geometry rock size and lack the cross-size prediction ability. In this paper, we propose a new physics-informed prediction model called the three-dimensional (3D) spatial pyramid pooling convolutional network (3DSPPConvNet) model embedding three important parameters: computed tomography (CT) scanning resolution, interfacial tension, and contact angle distribution. Our model can deal with 3D digital rocks with variable sizes as it is based on a multi-size digital rock dataset containing 1512 couples of relative permeability curves. Their relative permeability curves for training and testing are obtained from pore network modeling (PNM) results with a linear postprocessing process. The results show that the physics-informed 3DSPPConvNet has a high prediction accuracy of 95.3%, with scale information being the most important physics feature accounting for 51%. In addition, compared with PNM, the computational time and memory of the intelligent model are saved by 63% and 80%, respectively.

Intelligent modeling with physics-informed machine learning for petroleum engineering problems

Unconventional gas is a momentous energy source due to its considerable reserves and eco-friendly properties, where relative permeability is a key evaluative parameter of unconventional gas extraction. However, the geo-complexity, multiscale and multieffect of the unconventional gas reservoir challenge the relative permeability evaluation and production enhancement. Here, we establish a gas–water flow model by integrating multifractal theory, covering from nanoscale to macroscale and regarding the effects of slip, gas desorption–diffusion and water film separation, to reliably evaluate the relative permeability evolution during unconventional gas development. Based on our model, we describe the permeability of the unconventional reservoir with an 88% less evaluation error compared to the single fractal Darcy with the literature benchmark. Moreover, we characterize the gas–water relative permeability with a no more than 10% evaluation error based on the experimental data. The slip effect plays the most crucial role in the evaluation precision of relative permeability. We reveal that the permeability of the unconventional gas reservoir is decreased by the increase of generalized fractal dimension which enhances the heterogeneity and tortuosity of pores. We uncover that the slip effect facilitates the relative permeability of gas and water. Besides, the gas desorption–diffusion boosts gas relative permeability while limiting water relative permeability, whereas water film separation enhances water relative permeability but hinders gas relative permeability. This work brings insights into the precise description of multiscale and multieffect gas–water porous flow in unconventional gas development.

Machine learning is becoming prevalent increasingly for reservoir characteristics analysis in the petroleum industry. This investigation proposes an alternative way for evaluating interwell connectivity in oil fields utilizing machine learning. In this study, three-dimensional convolutional neural network (CNN) was utilized to establish a deep learning model, which can invert interwell connectivity combining with dynamic production data. Different from traditional methods that try to construct mathematical formulas to calculate the connectivity among wells basing on physical laws, deep learning model can capture autonomously the changing characteristics of dynamic production data by training continuously and provide a potential to characterize the interwell connectivity accurately without physical model. At the same time, the back propagation (BP) neural network has also been built to analyze the prediction performance, which are compared with CNN. The results demonstrate that CNN has better performance in predicting the connectivity with the overall AARD below 15.35%. Moreover, the connectivity predicted by CNN is closest to the real connectivity factor compared with some traditional methods. The evaluation method on interwell connectivity proposed by this paper provides effective guidance for the secondary development of both conventional and unconventional reservoirs.