This study explores the application of quantum neural networks (QNN) and variational quantum regression (VQR) in seismic inversion, leveraging quantum computing’s advantages in superposition, entanglement, and interference. Implemented using variational quantum circuits (VQC) and tested on the Qiskit simulator, QNN captures complex seismic data relationships, while VQR focuses on regression tasks. Results demonstrate their potential in geophysical applications but highlight challenges in model generalization and scalability. As quantum computing advances beyond the NISQ era, quantum intelligence is expected to provide new insights for efficient seismic inversion.

In response to problems such as river channel superposition and intersection, thin sand bodies, and the inability of seismic responses and seismic attributes to directly reflect reservoir characteristics in fluvial reservoirs, this study proposes a fine description and prediction method for fluvial reservoirs based on knowledge graphs. Firstly, by constructing a 3D geological forward model based on different seismic and geological conditions, including sand body thickness, wavelet frequency, noise, and phase variations, the study reveals the seismic response characteristics and patterns of fluvial reservoirs. Then, using the correlation analysis method, it selects key sensitive attributes. Finally, a knowledge graph library of fluvial reservoir attributes is established, covering geological entities, seismic response patterns, and association networks of sensitive attributes, and a structured semantic reasoning framework is formed. The application of actual data shows that the proposed method reduces the average relative error of sand body thickness prediction for the validation wells from 13.90% to 2.91%, and the constructed sensitive attribute knowledge graphs provide a new paradigm for intelligent quantitative prediction of complex fluvial reservoirs.

The Tarim Basin hosts numerous fault-controlled fracture-cavity reservoirs, which represent a distinctive type of reservoir with substantial hydrocarbon potential. However, due to their deep burial depth, conventional methods often fail to accurately characterize these reservoirs, falling short of the precision required for effective exploration and development. To address this challenge, this study proposes a novel method for identifying the core zones of fracture-cavity reservoirs. The proposed approach begins by verifying the advantages of angle-domain seismic data as a superior data source for delineating fracture-cavity systems. A deep learning model incorporating both spatial and channel attention mechanisms is then employed to extract the contours of the fracture-cavity bodies. Through waveform separation processing, a refined dataset devoid of stratigraphic interference is obtained. Subsequently, the first eigenvalue of the structure tensor is extracted to enhance the energy response associated with the fracture-cavity features. Finally, energy tracking is applied to identify the core zones, resulting in a comprehensive characterization of the core–belt architecture of the fracture-cavity system. Field applications demonstrate that the proposed method significantly outperforms conventional techniques in accurately identifying the core zones of fracture-cavity reservoirs. This approach provides an effective and reliable tool that can be widely applied in other regions with similar geological conditions.

A convolutional neural network-bidirectional long short-term memory-physics-informed neural network (CNN-BiLSTM-PINN) architecture is proposed that integrates physical and numerical constraints for seismic elastic parameter inversion. Traditional seismic inversion methods often face challenges of insufficient accuracy or low computational efficiency when dealing with complex geologic conditions. To address these issues, two network architectures - convolutional neural network-physics-informed neural network (CNN-PINN) and CNN-BiLSTM-PINN - were constructed, and a hybrid strategy combining model-driven and data-driven approaches was designed. Systematic experiments using the Marmousi2 model compared the inversion performance across different network architectures and driving strategies. Results indicated that under the hybrid driving mode (model 0.3 + numerical 0.7), CNN-BiLSTM-PINN provided clearer boundaries, better structural continuity, and improved resolution for the Lam & eacute; parameter (lambda) and shear modulus (mu), whereas CNN-PINN demonstrated statistically significant advantages in density (p) inversion (p < 0.001). Practical applications in the CO2 flooding demonstration area of X oilfield further demonstrated that the method accurately delineated the CO2 flow range and aligned well with well logging data. The study also found that although the CNN-BiLSTM architecture showed higher training efficiency than CNN in supervised learning, its computational complexity increased significantly within the PINN framework, primarily due to the interaction between physical constraint calculations and the recursive structure of the BiLSTM. These findings offer a novel paradigm for seismic elastic parameter inversion, enhancing inversion accuracy and stability while preserving physical reasonability, with significant implications for oil and gas exploration and carbon sequestration monitoring.

Physics-informed neural networks (PINNs) enable unsupervised inversion by integrating seismic forward modeling equations directly into neural network loss functions, operating on angle-domain seismic data. The inverse PINN (iPINN) extends this framework by treating the central frequency of the Ricker wavelet as a trainable parameter, thereby enhancing the adaptability to spectral mismatch and improving lateral continuity in inversion results. However, iPINN do not provide uncertainty quantification for their predictions. To address this limitation, an inverse Bayesian PINN (iBPINN) is proposed, incorporating Flipout-based Bayesian convolutional and fully connected layers into the iPINN architecture. Through the variational inference, iBPINN enables the probabilistic modeling of both network weights and wavelet frequency, yielding not only predicted physical parameters but also posterior standard deviation (std) maps for direct visualization and quantitative assessment of predictive uncertainty. Numerical experiments on the Marmousi2 synthetic model show that PINN achieves rapid convergence for large-scale stratigraphic structures, while iPINN provides more consistent inversion results for deeper layers and weakly sensitive parameters via dynamic frequency correction. iBPINN maintains comparable inversion consistency and, crucially, enables intuitive mapping of predictive uncertainty through std analysis, particularly in fault zones, high-impedance contrasts, and low-signal-to-noise ratios (SNR) regions. Field application to seismic data from a CO2 injection site further demonstrates that the uncertainty maps provided by iBPINN offer additional insights into fluid substitution and reservoir connectivity. This methodological progression-from PINN ("statistics + logic") to iPINN ("logic + adaptability"), and ultimately to iBPINN ("logic + reasoning")-mirrors the broader evolution of artificial intelligence (AI) in geophysical inversion: from statistical models to logic-embedded frameworks, and finally to reasoning-enhanced architectures. The proposed iBPINN framework, thus, offers a unified, interpretable, and trustworthy approach for next-generation seismic inversion in complex geological environments.

We analyze the differences before and after CO2 flooding in the Gao89 area of the Shengli Oilfield using nonrepeated field time-lapse seismic data and geologic modeling. This study also demonstrates the feasibility of monitoring dynamic CO2 plume changes through reservoir parameter differences inverted from baseline and monitoring data sets. To ensure data consistency, we introduce a "four-step" matched-filter processing technique. First, sparse sampling is used to align the monitoring survey with the baseline seismic data. Second, temporal interpolation is applied to adjust the CO2 injection target cubes - each with different two-way traveltimes - into stratigraphic cubes of equal time length. Third, amplitude consistency matching is performed using the top surface amplitude as a reference, ensuring amplitude alignment between the two seismic data sets. Finally, the difference data between the baseline and monitoring seismic data sets, after amplitude matching, are embedded into the baseline seismic data to restore the actual formation structure. The embedded data allow for the extraction of key seismic attributes that capture changes before and after CO2 injection. The study finds that time thickness and average area properties effectively delineate the extent and internal characteristics of CO2 plume changes, showing strong agreement with well tracer data. In addition, validation using prestack seismic azimuth attributes at a well confirms the consistency between prestack and poststack seismic dominant channel identification.

Seismic attribute analysis technology has been widely used in the prediction of fluvial reservoir sand body, but the traditional seismic attribute fusion technology based on linear model has low prediction accuracy and limited application range. This study focused on the non-linear fitting between seismic attributes and reservoir thickness, and used a variety of machine learning technologies to predict the fluvidal reservoir in Chengdao area of Dongying Sag (China).The channel sand body in Chengdao area is deep buried, thin in thickness, fast in velocity and affected by gray matter, so it is difficult to predict, which greatly restricts the oil and gas exploration in this area. In this study, on the basis of fine well earthquake calibration, several seismic attributes such as amplitude, frequency, phase, waveform and correlation are extracted and correlation analysis is done to remove redundant attributes. Then model training and parameter set optimization are carried out, thickness prediction is carried out with verification set, and vertical resolution is improved by logging reconstruction and waveform indication inversion. The results show that compared with the conventional support vector machine and back propagation neural network, the prediction accuracy of echo state network optimized by Sparrow algorithm is greatly improved. Based on the comprehensive prediction method of fluvial reservoir, three large channels developed in the lower part of Chengdao area and several small channels developed in the upper part of Chengdao area are effectively described. The research method can be used for reference to the similar complicated river facies prediction.

Unlike horizontal layered sedimentary reservoirs, fracture-cavity bodies are primarily characterized by vertical strike-slip faulting and dissolution, leading to distinct seismic wave imaging features at different incidence angles. This paper proposes a novel approach for fracture-cavity bodies identification by leveraging angle-domain seismic information, with a focus on the utilization of large-angle data By systematically analyzing seismic data partitioned into small-angle (0-6 degrees), medium-angle (7-26 degrees), and large-angle (27-36 degrees) domains, this work broadens the framework of deep learning-based identification for fracture-cavity systems and enriches seismic interpretation methodologies for complex geological bodies. First, forward modeling and comparative analysis of real seismic data validate that large-angle seismic data exhibit superior capability in resolving fracture-cavity structures. Subsequently, deep learning models are employed to extract seismic signatures of these reservoirs. Field applications demonstrate that conventional methods exhibit limited accuracy due to noise interference, whereas large-angle seismic data provide superior resolution in characterizing fracture-cavity geometries, offering more comprehensive spatial representations than traditional migration datasets. In summary, this research establishes a new strategy for identifying fracture-cavity bodies using large-angle seismic information, providing a transferable reference for analogous geological settings.

Seismic inversion methods based on traditional frameworks face fundamental challenges in accurately characterizing spatial structure and quantifying model reliability. Here, we introduce a next-generation approach, systematically evolving from a convolutional physics-informed neural network (PINN), to a Bayesian PINN (BPINN) with uncertainty modeling, and culminating in a Bayesian Physics-Informed Vision Transformer (BPI-ViT) architecture that enables structure-level uncertainty quantification. Consistent evaluation is performed on the Marmousi2 benchmark and validated on field-scale CO₂ EOR monitoring data. The BPI-ViT framework integrates structure-aware self-attention and Bayesian inference, effecting a transition from pixel-level optimization to structural collaboration. Empirical results reveal that BPI-ViT outperforms previous methods in target horizon recovery, fault and anomaly detection, spatial continuity, and uncertainty quantification. This study establishes a structural-intelligent paradigm, advancing seismic inversion beyond error minimization towards structure-aware, reliable, and cognitively informed modeling, and provides a foundation for future multi-physics and complex geological inversion applications.

The recognition of fractured-vuggy core is a pivotal task in the characterization of fractured-vuggy reservoirs, where pores, fractures, cavities, fluids, and faults often overlap intricately. Sole reliance on seismic facies cannot provide an accurate and effective interpretation. Traditional methods such as coherence and curvature are no longer sufficient to meet the precision requirements of modern exploration and development. In this paper, we propose a novel methodology for characterizing fractured-vuggy systems in the complex reservoirs of the Y area in the Tarim Basin. Initially, through seismic interpretation, the top and bottom layers of the reservoir were determined. The target layer was extracted from the field seismic data to form an equal-depth body. Spatial wavenumber domain filtering was then applied to remove the stratigraphic background from the equal-depth body. Subsequently, the first eigenvalue of the gradient structure tensor (GST) of the fractured-vuggy body was extracted. Threshold and amplitude control were used to trace the energy ridge and identify the core of the fractured-vuggy body. Finally, the identification results were integrated and embedded into the field seismic data, yielding the final fractured-vuggy characterization volume. The actual application of the Y5 well area shows that the identified results are in good agreement with well logs, and it is an effective method to describe fractured-vuggy body. In addition, the application and analysis in the Y2 well area predict three favorable target points. In general, the proposed method for fractured-vuggy body characterization proves to be effective in complex fractured-vuggy development areas and can be promoted and applied as a new approach.

We analyze the differences before and after CO 2 flooding in the Gao89 area of the Shengli Oilfield using non-repeated field time-lapse seismic data and geological modeling. This study also demonstrates the feasibility of monitoring dynamic CO 2 plume changes through reservoir parameter differences inverted from baseline and monitoring datasets. To ensure data consistency, we introduce a “four-step” matched-filter processing technique. First, sparse sampling is used to align the monitoring survey with the baseline seismic data. Second, temporal interpolation is applied to adjust the CO 2 injection target cubes—each with different two-way travel times—into stratigraphic cubes of equal time length. Third, amplitude consistency matching is performed using the top surface amplitude as a reference, ensuring amplitude alignment between the two seismic datasets. Finally, the difference data between the baseline and monitoring seismic datasets, after amplitude matching, is embedded into the baseline seismic data to restore the actual formation structure. The embedded data allows for the extraction of key seismic attributes that capture changes before and after CO 2 injection. The study finds that time thickness and average area properties effectively delineate the extent and internal characteristics of CO 2 plume changes, showing strong agreement with well tracer data. Additionally, validation using pre-stack seismic azimuth attributes at a well confirms the consistency between pre-stack and post-stack seismic dominant channel identification.

Traditional seismic inversion frameworks struggle to preserve spatial structure and to quantify model reliability. We present a next-generation pathway that progresses from a convolutional Physics-Informed Neural Network (PINN) to a Bayesian PINN (BPINN) with uncertainty modeling, and culminates in a Bayesian Physics-Informed Vision Transformer (BPI-ViT) that enables structure-level uncertainty quantification. In our formulation, PINN “training data” are equation-domain samples used to minimize physical residuals—supporting physics-driven, data-agnostic generalization—while BPI-ViT integrates multi-layer self-attention and Bayesian inference to transition from pixel-level optimization to structure-aware collaboration. Consistent evaluation on the Marmousi2 benchmark and validation on field-scale CO₂ EOR monitoring data show that BPI-ViT outperforms prior methods in target-horizon recovery, fault and anomaly detection, spatial continuity, and uncertainty quantification, while maintaining physical consistency. These results establish a structural-intelligent paradigm that moves seismic inversion beyond error minimization toward structure-aware, reliable, and cognitively informed modeling, and provide a foundation for future multi-physics and complex-geology applications.

High-precision sand thickness data are fundamentally important for optimizing exploration strategies in petroleum geology. In the Chengbei work area of the Jiyang Depression, the stratigraphic channels are chaotically developed, with channels of varying sizes in different strata overlapping, intersecting, and exhibiting narrow widths. The actual well-seismic relationship is poor. Therefore, individual seismic attributes in this area exhibit extremely low correlation with channel sandstone thickness. Conventional attributes such as root mean square amplitude show no distinct channel characteristics, necessitating the integration of multiple seismic attributes for effective prediction. Moreover, the high multicollinearity among seismic attributes introduces significant interference in prediction results. Therefore, this study integrates the Pearson correlation coefficient and variance inflation factor (VIF) to optimize seismic attribute selection, effectively eliminating redundant attributes and those with low correlation. To further enhance prediction accuracy and address the significant bias inherent in single-model predictions, this study introduces the ensemble learning XGBoost model, which integrates predictions from multiple weak learners to improve the precision of sandstone thickness estimate. The Newton-Raphson-based optimization algorithm was employed to fine-tune the XGBoost parameters. Results from test wells demonstrate a remarkable improvement in prediction accuracy, achieving reliable sandstone thickness estimation despite poor well-seismic correlations. This research provides valuable insights and offers a widely applicable methodology for predicting the thickness of complex channel sand bodies.

Based on 2011 and 2021 non-repeated time-lapse data in G89 well block of Shengli oilfield, a four-step processing method is proposed. The first step is survey consistence. Bysparse sampling, it makes the monitoring survey consistent with the base exploration. The second step is generating tw-o stratal cubes. Because seismic travel time is different for the injection layer, the object cubes of two data must be int-erpolated, and make them having the same samples. The th-ird step is amplitude matching processing. Using the top su-rface amplitude as a reference, the surface amplitude of twodata is implemented coherently. The fourth step is embedd-ing and predicting. After matching processing, embed the difference attribute into the basic exploration data, and ext-ract the sensitive seismic attributes of CO2 flooding. Throu-gh analysis and comparison, the thickness and average area of peak can better reflect the plume of CO2 flooding.

With the increasing application of deep learning in geoscience, In this paper, a new method of fault-karst reservoir identification based on ResU-Next model is proposed on the basis of previous research. Traditional learning and prediction of karst caves and faults are conducted separately, resulting in increased learning costs and reduced efficiency. In this research, fault and karst cave labels are combined for simultaneous training. This approach enables the network to directly obtain prediction results for both karst caves and faults during the prediction process, significantly enhancing efficiency while ensuring accuracy. Model testing reveals that compared to conventional U-Net network, the proposed method demonstrates superior learning performance with higher accuracy in capturing intricate details. It serves as a valuable reference for fault solution identification using deep learning techniques. When applied to actual seismic data analysis, this method effectively indicates the distribution relationship between faults and karst caves, reduces manual interpretation costs, and provides guiding suggestions for oil and gas exploration.

Porosity, as a key parameter to describe the properties of rock reservoirs, is essential for evaluating the permeability and fluid migration performance of underground rocks. In order to overcome the limitations of traditional logging porosity interpretation methods in the face of geological complexity and nonlinear relationships, this study introduces a CNN (convolutional neural network)-transformer model, which aims to improve the accuracy and generalization ability of logging porosity prediction. CNNs have excellent spatial feature capture capabilities. The convolution operation of CNNs can effectively learn the mapping relationship of local features, so as to better capture the local correlation in the well log. Transformer models are able to effectively capture complex sequence relationships between different depths or time points. This enables the model to better integrate information from different depths or times, and improve the porosity prediction accuracy. We trained the model on the well log dataset to ensure that it has good generalization ability. In addition, we comprehensively compare the performance of the CNN-transformer model with other traditional machine learning models to verify its superiority in logging porosity prediction. Through the analysis of experimental results, the CNN-transformer model shows good superiority in the task of logging porosity prediction. The introduction of this model will bring a new perspective to the development of logging technology and provide a more efficient and accurate tool for the field of geoscience.

Lithology identification plays a key role in CO2 sequestration projects, helping to improve the feasibility, safety, and effectiveness of sequestration projects; therefore, accurate prediction of lithology is critical. The mapping relationship between the logging parameters and lithology complex and the logging response is multi-solution, resulting in inaccurate results of traditional logging lithology identification methods. In this study, we propose an enhanced predictive model for lithology, termed the deep residual shrinkage network (DRSN). This network incorporates a residual network, an attention mechanism, and a soft threshold strategy. The introduction of residual blocks in residual networks addresses the gradient vanishing problem inherent in deep neural networks by allowing the learning of residuals through skip connections. The attention mechanism enhances the focus of the model on crucial input elements, thereby improving its capacity to capture key information. Soft thresholding strategies are employed to eliminate noise from the inputs, enhancing the robustness of the model. Six logging parameters (photoelectric index, density, acoustic, gamma, spontaneous potential, and neutron) are chosen as inputs, with lithology serving as the output of the model. For comparison, we introduce the widely used residual network (ResNet), a classical lightweight network (SqueezeNet), and the convolutional neural network (CNN). Testing on three wells in China's Tarim Oilfield demonstrates that the DRSN model accurately locates reservoirs and identifies lithology more precisely. In lithology prediction for the three wells, DRSN achieved accuracy rates of 90.84, 85.51, and 93.70%, respectively. This research offers a novel approach to lithology prediction in the realm of carbon dioxide geological storage.

In recent years, fluid prediction through well logging has assumed a pivotal role in the realm of oil and gas exploration. Seeking to enhance prediction accuracy, this paper introduces an adaptive piecewise flatness-based fast transform (APFFT) algorithm in conjunction with the XGBoost (extreme gradient boosting) method for logging fluid prediction. Initially, the APFFT technology is employed to extract frequency-domain features from the logging data. This algorithm dynamically determines the optimal frequency interval, transforming raw logging curves into frequency domain data. This adaptive process enhances the preservation of frequency domain information reflective of fluid characteristics, simultaneously minimizing the impact of noise and non-fluid compositions. Subsequently, the acquired frequency domain features are utilized as inputs to construct an XGBoost model for fluid prediction. To validate the efficacy of this proposed approach, real logging data were collected, and an extensive experimental evaluation was conducted. The experimental findings underscore the substantial advantages of the APFFT-XGBoost method over traditional machine learning models such as XGBoost, random forest, K-nearest neighbor algorithm, support vector machine, and backpropagation neural network in logging fluid prediction. The proposed method demonstrates the ability to accurately capture fluid features, leading to improved prediction accuracy and stability.

The seismic interpretation and reservoir description of complex fault-block reservoirs pose the following challenges: 1. The structural anomalies are complex, making it difficult to trace the strata, and conventional methods fail to establish accurate well-seismic correlations. 2. Data has a low signal-to-noise ratio, making it difficult to establish precise fault architecture even with three-dimensional visualization techniques like coherence cubes. To address these challenges, the following targeted strategies are proposed: 1. Implement measures such as well log curve similarity analysis, calibration of continuous and complex strata concatenations, and cross-validation of seismic layer velocity with logging conversion velocity to obtain comparatively accurate well-seismic relationships. 2. Utilize confidence interval analysis to enhance the three-dimensional anisotropic diffusion filtering method, thus further enhancing the edge features of the data volume faults. These techniques have been effectively applied in the Dagang Zilaitun Oilfield, validating the efficacy and feasibility of the approach.