Haibo Cheng

Inverse modeling for subsurface flow represents a fundamental scientific challenge in hydrogeology and geotechnical engineering, which seeks to reconstruct critical hydrogeological parameters from sparse observational constraints. The marked spatial heterogeneity of subsurface formations, combined with the prohibitively high costs of data acquisition, renders parameter inversion, especially with very sparse supervision, inherently ill-posed and susceptible to non-uniqueness and instability. Numerical simulation-based iterative inversion methods are computationally expensive and time-consuming. Purely data-driven approaches require extensive labeled data, whereas the existing physics-informed methods lack an explicit architecture-level information transfer channel between parameter and response fields. Under sparse supervision, this prevents hydraulic head observations from effectively constraining hydraulic conductivity identification, resulting in weak parameter identifiability. In this work, we propose a physics-coupled and message-transferred inverse modeling method for transient subsurface flow problems with very sparse supervision. Specifically, the static parameter field estimated by the inversion network is explicitly incorporated into the dynamic response prediction network, and the static inversion and dynamic prediction networks are physics-coupled by the governing equations in parallel. This method enables accurate hydraulic conductivity inversion under extremely limited supervision. Experiments on multiple parameter fields, label scales, and noise levels demonstrate accurate and stable inversion performance under very sparse supervision, with ensemble-based uncertainty analysis, further confirming the reliability of the proposed method.

In this article, the fault diagnosis problem is investigated for rod pumping systems (RPSs) with its application to oil well systems. To address the problem, a new automatic fault diagnosis method by using a federated adversarially robust multitask temporal fusion model (FARMTF) is developed. Specifically, the framework incorporates a conditional generative adversarial network (cGAN) to generate synthetic samples for minority fault types, thereby mitigating the impact of class imbalance. Then, the temporal convolutional network (TCN) is integrated into a multitask learning framework to identify long-term dependencies in operational signals, with a cascaded mechanism facilitating the collaborative modeling of tasks. Moreover, in order to further address the challenges of data privacy and ownership arising from the geographically distributed and sensitive nature of oilfield data, we introduce a federated learning (FL) framework into the fault diagnosis model. Finally, the effectiveness of the proposed method is validated through experiments on historical data from Daqing oil field.

Fault diagnosis is essential for ensuring the reliability and safety of rod pumping systems (RPSs). Deep learning offers a promising approach to intelligent fault diagnosis of rod pumping systems. However, it remains challenging to improve the accuracy in fault scenarios with subtle variations and to enhance generalization across oil wells under identical operating conditions. To tackle these issues, we propose a knowledge-enhanced contrastive Bayesian temporal convolutional network (KEC-BTCN) for fault diagnosis including fault detection and the fault prediction tasks of RPSs. Technically, prior knowledge is incorporated into the model in the form of textual descriptions, which provide contextual information for fault diagnosis. During training, time-series data, including image, text, and structured tabular data, are segmented into overlapping subsequences to generate positive and negative sample pairs, helping the network obtain representations with enhanced discriminability. Furthermore, a dual feedback mechanism is designed with a multi-task loss function that integrates distribution and contrastive losses to improve the generalization performance of fault diagnosis. Finally, extensive experiments confirm that the proposed KEC-BTCN consistently achieves accurate fault diagnosis performance and generalization capability across various oil well scenarios, in both fault detection and prediction tasks.

Subsurface flow is the core of reservoir engineering. Research on subsurface flow problems can enhance our understanding of the development status of oilfields, thus enabling the prediction of the distribution of residual oil and the formulation of production plans. Machine learning-based methods have been widely studied in developing data-driven models to solve subsurface flow problem. However, a large amount of labeled data are required to build highly accurate models. It is applicable to embed domain-knowledge into data-driven machine learning methods to reduce data requirements. In this study, we propose a residual-enhanced physics-guided machine learning method with hard constraints (RHC-PGML) to predict reservoir pressure. Specifically, a physics-guided machine learning with hard constraints strategy is proposed, which makes the prediction results strictly satisfy the prior domain knowledge to improve the prediction accuracy of the model. In addition, we integrate residual learning in RHC-PGML model to compensate for systematic errors caused by the inability of the embedded physical mechanism to perfectly describe the complex seepage process and further improve the prediction accuracy. The proposed method is verified by a seepage problem in a heterogeneous reservoir model. The results show that the RHC-PGML method can obtain reliable prediction results in the case of sparse and limited data.

Integrating physical mechanisms with data-driven methods overcomes the limitations of purely data-driven artificial intelligence and purely mechanism-based models. Purely data-driven approaches suffer from poor interpretability and weak generalization under sparse data, while purely physics-based models are computationally expensive and struggle with complex nonlinearities. This work highlights advances in the physics-constrained, data-driven dual paradigm across petroleum engineering: mechanism–artificial intelligence fusion via Bayesian networks provides traceable hydrocarbon spatial distribution predictions; knowledge–data-driven modelling ensures geological realism; and collaborative physics–data fault diagnosis enhances well monitoring under noise. These advances demonstrate that deep fusion of domain knowledge, physical laws, and multi-source data is essential for creating interpretable, reliable, and efficient intelligent systems for complex subsurface resource development. Document Type: Perspective Cited as: Hui, G., Wang, M., Cheng, H. Advances in physics-constrained and data-driven dual paradigm for artificial intelligence in oil and gas. Advances in Geo-Energy Research, 2026, 20(3): 201-204. https://doi.org/10.46690/ager.2026.06.01

Sucker-rod pumping systems are the most widely applied artificial lift equipment in the oil and gas industry. Accurate and intelligent working condition recognition of pumping systems imposes major impacts on oilfield production benefits and efficiency. The shape of dynamometer card reflects the working conditions of sucker-rod pumping systems, and different conditions can be indicated by their typical card characteristics. In traditional identification methods, however, features are manually extracted based on specialist experience and domain knowledge. In this paper, an automatic fault diagnosis method is proposed to recognize the working conditions of sucker-rod pumping systems with massive dynamometer card data collected by sensors. Firstly, AlexNet-based transfer learning is adopted to automatically extract representative features from various dynamometer cards. Secondly, with the extracted features, error-correcting output codes model-based SVM is designed to identify the working conditions and improve the fault diagnosis accuracy and efficiency. The proposed AlexNet-SVM algorithm is validated against a real dataset from an oilfield. The results reveal that the proposed method reduces the need for human labor and improves the recognition accuracy.

In petroleum production system, interwell connectivity evaluation is a significant process to understand reservoir properties comprehensively, determine water injection rate scientifically, and enhance oil recovery effectively for oil and gas field. In this paper, a novel long short-term memory (LSTM) neural network based global sensitivity analysis (GSA) method is proposed to analyse injector-producer relationship. LSTM neural network is employed to build up the mapping relationship between production wells and surrounding injection wells using the massive historical injection and production fluctuation data of a synthetic reservoir model. Next, the extended Fourier amplitude sensitivity test (EFAST) based GSA approach is utilized to evaluate interwell connectivity on the basis of the generated LSTM model. Finally, the presented LSTM based EFAST sensitivity analysis method is applied to a benchmark test and a synthetic reservoir model. Experimental results show that the proposed technique is an efficient method for estimating interwell connectivity.