The high-dimensional 3-D seismic data are inevitably contaminated by random noise due to environmental interference and acquisition system limitations. While many deep learning methods have been proposed for random noise attenuation, they often suffer from various limitations. For instance, some convolutional neural network (CNN)-based methods tend to be computationally inefficient when processing 3-D seismic volumes, while the implicit neural representation (INR)-based methods frequently fail to recover complex structural details due to the spectral bias. To address these challenges, this study introduces an efficient nonlocal tensor function representation (NLTFR) for unsupervised 3-D seismic denoising. Specifically, we cluster similar cubes within 3-D seismic data into nonlocal similar groups. To efficiently and effectively exploit the nonlocal self-similarity inherent in the seismic data, we employ the tensor function factorization parameterized by INRs to model these clustered nonlocal similar groups. Furthermore, we propose a partially shared transfer learning strategy for NLTFR to accelerate the model convergence when applied to large-scale seismic datasets. Multiple-frequency sinusoidal functions are incorporated into the INRs to mitigate the spectral bias, thereby enhancing the model's representation ability. In addition, a weighted total variation and a second-order 3-D total variation are introduced to improve the denoising performance and robustness of the NLTFR model. Comprehensive experiments conducted on synthetic and field 3-D seismic datasets demonstrate that the proposed NLTFR achieves a superior balance between efficiency and effectiveness compared with conventional deep learning methods. Overall, our method attains an average performance gain of 1.5 dB while significantly reducing execution time relative to traditional deep learning methods for 3-D seismic volumes, underscoring its practical efficiency.

Seismic inversion is crucial for reservoir characterization. Conventional inversion methods are frequently limited by low resolution and the requirement for low-frequency models. Although deep learning has shown promising performance in seismic parameter inversion, convolution-based networks mainly rely on local receptive fields and cannot explicitly capture the long-range dependencies along seismic traces, which limits their ability to recover the overall subsurface parameter trends. To alleviate this problem, we propose a Mamba-based seismic inversion network, termed DMamUNet. Built on the ResUNet++ backbone, DMamUNet introduces Mamba modules into the encoder–decoder framework to model long-sequence contextual information and global parameter variation trends, while deformable convolution blocks are used to enhance the adaptive representation of local nonstationary features. In addition, a multitrace-to-multitrace (M2M) supervision strategy and a composite loss combining well-log parameter supervision with seismic reconstruction are adopted to further improve feature learning and generalization. Experiments on the Marmousi model, post-stack field data, and pre-stack field data demonstrate that DMamUNet outperforms conventional methods. Specifically, DMamUNet significantly enhances both inversion accuracy and lateral continuity. In the post-stack field data experiment, the proposed DMamUNet achieves a 78.22% reduction in blind well mean squared error (MSE) compared with FCRN, and a 14% MSE reduction relative to TransUNet. Additionally, DMamUNet predictions on pre-stack field data exhibit higher resolution and better spatial continuity, highlighting its superiority for improving reservoir characterization.

Lithology prediction is essential for the characterization and exploration of oil and gas reservoirs. Recent studies have demonstrated that integrating time-frequency analysis with deep learning models can enhance lithology prediction performance. However, existing approaches are often constrained by large parameter volumes and high computational complexity, limiting their applicability to large-scale seismic datasets. Furthermore, lithology samples typically exhibit class imbalance, which leads to insufficient feature learning for minority classes. To address these problems, this paper proposes LightLPNet-CA, a lightweight network for seismic lithology prediction. The method first applies a time-frequency transform to generate time-frequency spectral maps from post-stack seismic traces, which are used as input to the network. Subsequently, a channel attention mechanism is integrated to enhance discriminative feature representation, along with a class-balancing strategy to mitigate sample imbalance. With only about 0.1M parameters, the network reduces computational overhead while maintaining predictive performance through its lightweight architecture. Experimental results show that the proposed method achieves higher lithology identification accuracy, particularly excelling in the prediction of minority classes.

There are abundant oil and gas resources in deep waters, however, the exploration and development cost is huge. High seismic data imaging quality is essential. Marine seismic data mainly consists of primary and multiple. Traditional seismic imaging only relies on the contribution of primary reflected waves, and the high energy multiples cannot be effectively used, which limits the further imaging quality improvement of marine seismic data. In order to use full wavefield information (primary wave, surface multiple wave and interlayer multiple wave) of marine seismic data, a full wavefield migration method for deep water seismic data has been developed and successfully applied in deep water data. Firstly, the one-way wave continuation operator considering the transmission effect was illustrated, and it was extended to the full wavefield forward modeling through one-way wave closed-loop continuation, achieving the simulation of multiples under the condition of smooth velocity models. Secondly, the full wavefield migration imaging objective function including primary and multiple was constructed, its gradient was solved, and a step size optimization strategy was proposed to ensure the rapid and stable convergence of the inversion calculation. Finally, tests were conducted on the Sigsbee2B model and actual data from a deep water exploration area, and the imaging effect was significantly improved. This study shows that the information from multiples is relatively important, and the comprehensive utilization of full wavefield information to improve imaging accuracy is an inevitable direction for marine seismic data processing, which has important research significance and application value for deep water oil and gas exploraton.

Amplitude variation with angle (AVA) prestack seismic inversion plays a critical role in oil and gas exploration and mineral resource assessment. Recently, deep learning methods, particularly convolutional neural networks (CNNs), have been widely adopted for seismic inversion. However, many of these methods, especially supervised learning, struggle with poor generalization and noise resistance. Seismic data contains rich texture information that can be used as prior to constrain the convolutional kernels of the network. Gabor functions have long been used for seismic data representation, and learnable Gabor filters improve upon this by dynamically extracting latent seismic data information via adaptively updating Gabor filter parameters. In this letter, we propose a multitask AVA inversion method using learnable Gabor filters within a 2-D multitask attention U-Net. We equip the network's first layer with learnable Gabor filters for latent seismic data feature extraction to enhance both generalization and noise resistance. An adaptive weight update method (AWUM) is employed to balance multitask learning efficiency and generalization performance. By creating a training dataset that combines synthetic and field seismic data with corresponding labels, we integrate field samples into the network training. Experiments for both synthetic and field datasets demonstrate that the proposed method exhibits superior generalization and stability compared to several existing approaches.

Seismic imaging techniques are essential to geophysical exploration and subsurface characterization, with full wave-equation depth migration (FWDM) emerging as a highly effective method for imaging complex geological structures. However, optimizing FWDM requires a deep understanding of the factors influencing its performance to achieve accurate and detailed subsurface images. To address this need, this study presents a systematic approach to refining FWDM by first providing a comprehensive review of the extrapolation equations and imaging mechanisms that form its foundation. We then introduce a series of numerical tests designed to identify and analyze key factors impacting imaging quality, including migration velocity, wave propagation effect, and numerical stability. Based on these findings, we optimize parameters and apply them to two challenging models, allowing us to assess the improvements in imaging clarity and accuracy. This research not only highlights critical factors affecting FWDM but also demonstrates how targeted optimizations can significantly enhance its effectiveness for advanced subsurface imaging applications.

Fault identification is vital for geological structure analysis and the optimization of oil–gas extraction. Deep neural networks, especially U-Net and its variants, are widely used for seismic fault interpretation. However, when applied to 3D seismic data volume, these models typically require substantial computation resources and memory consumption. For one reason, they do not take into consideration the obvious differences in characteristics of seismic data in space and time dimensions; therefore, they require a huge number of parameters to capture inherent information for seismic fault detection. This paper presents a lightweight 3D seismic fault interpretation network based on a spatial–temporal asymmetric convolution set (STA-Fault3D) to mitigate the aforementioned issue. STA-Fault3D uses the spatial–temporal asymmetric convolution set to construct a lightweight network and take into consideration seismic data dimension discrepancies. Multi-scale feature fusion operation and an enhanced-training workflow are adopted to improve the performance of the network on field data. Compared with the classic model, FaultSeg3D, it demonstrates improved performance on fault detection continuity with only 12.33% of the parameters and 18.57% of the computational quantity. Compared with the state-of-the-art (SOTA) lightweight network, Fault3DNnet, it reduces parameters by 10% and computational quantity by 4.2% for marginally improved detection results.

Coherent noise suppression in common reflection point (CRP) gather is a crucial task in seismic data processing. The valid signals in CRP gathers are horizontally aligned which can be represented effectively by low-rank approaches but not the coherent noise. Leveraging this characteristic, we propose a neural network low-rank approximation method for CRP gather coherent noise suppression. Specifically, to increase the self-similarity between adjacent traces in CRP gather, we first use plane-wave structural prediction operator to flatten seismic events within a local neighboring window. Subsequently, a fully connected neural network with low-rank regularization is used to approximate the locally flattened seismic data. Moreover, transfer learning strategy is used to improve the efficiency of multiple seismic gather processing. On two field CRP gathers, the proposed method achieves the lowest local similarity (LS) values of 0.0328 and 0.0629, compared with both state-of-the-art and traditional methods. The denoising results on synthetic and field data demonstrate the effectiveness of the proposed method both in attenuating coherent noise and protecting valid signals.

Carbonate bioreef formations serve as crucial hydrocarbon reservoirs, and their accurate identification bears significant implications for oil and gas exploration. Moreover, the precise and refined delineation of prograding body structures aids in the comprehensive analysis of stratigraphic geologic configurations. We develop the knowledge graph and geologic strata interpolation constraints (KGGSICs) model for the intricate identification of carbonate bioreefs and prograding body structures. Furthermore, we assess our KGGSIC-Unet architecture on the Dengying Formation Sections 3-4 carbonate bioreefs and prograding bodies in the Moxi area of the Sichuan Basin. Experimental results indicate that the KGGSIC enhances the predictive performance of the U-Net and realizes the precise and refined segmentation of carbonate bioreefs and prograding body structures. In addition, through a meticulous geologic study of the area, we synthesize the 2D profile identification results to achieve the precise and refined identification of carbonate bioreefs and prograding bodies.

The strong impedance interfaces of the sea surface and seabed cause significant multiples and ghost wavefield in marine seismic data. The finite-difference method of the two-way wave equation widely used in seismic data modeling cannot separate multiples and ghost wavefield of different orders. That cannot perfectly combine with the migration and parameters inversion process. Traditional methods limit the popularization and application of multiples suppression and migration methods. In this study, the marine seismic wavefield modeling method based on a close-loop one-way propagation operator was derived and the forward modeling equation of ghost wavefield was established. Firstly, one-way propagation operator is derived, and the propagation model of traditional primary reflections is given. Secondly, propagation models of surface- and internal-related multiples are obtained by loading the last round of reflected data at different depths and using closed-loop iterative calculations. The proposed method can achieve multiples simulations and separation of different orders. Then, the ghost waves propagation model is derived by placing the source below the water surface and changing the loading order of seismic sources. Surface-, internal-related multiples, and ghost wavefield of the lenticle model have been accurately simulated and analyzed. The Marmousi model was implemented to compare the accuracy of proposed mothed and finite-difference method, in which the synthetic data of different methods show similar results. The proposed method can reasonably and effectively obtain multiples and ghost wavefield of desired orders, which can be well combined with subsequent migration and inversion processes to improve imaging illumination and resolution.

Seismic inversion is crucial for oil and gas exploration. Recently, the development of deep learning (DL) has provided new means for the continuous improvement of seismic inversion. However, field seismic data exhibits nonstationarity and multiscale features due to wavelet attenuation and dominant frequency variations. Although the atrous spatial pyramid pooling (ASPP) module concatenated of dilated convolution greatly facilitates the multiscale feature learning ability of the network, the adjustment of the dilation rate, which is a hyperparameter, requires ample ablation experiments and is a tedious and time-consuming process. To alleviate these issues, the stacked multiple deformable convolutions (DConv) layers are employed as the feature extraction module to adaptively capture the multiscale correspondence between seismic data and elastic parameters in multitask seismic inversion. The sampling grids of DConv can automatically be modulated by adding a learnable offset. Thus, DConv can provide a flexible and effective receptive field, which is conducive to aggregating pivotal information of seismic data and improving the inversion accuracy. To further enhance the reliability and stability, the proposed method incorporates a multitrace to single-trace (M2S) strategy and the closed-loop Wasserstein generative adversarial network with gradient penalty (WGAN-GP) framework. Experiments show that the application of DConv to the inversion of P-wave velocity ( Vp ) and density ( rho ) yields superior transverse continuity and vertical resolution. Compared with ASPP, the MSE of the predicted profiles and the true models in the synthetic Marmousi2 experiment is degraded by 36% and 32%, respectively, and the MSE of the reconstructed seismic data and the real data is reduced by an order of magnitude for the field test.

Deep learning methods, especially convolutional neural networks, achieve state-of-the-art performance on seismic impedance inversion. Most of the methods are based on one-dimensional (1-D) convolution, tending to yield lateral discontinuities of impedance on field data applications. To alleviate this problem, we design a network equipped with 2-D convolutions and a coordinate attention (CA) block. The former can take the relationship between adjacent traces into consideration. The latter can capture the positional relationship of the geological structure, both horizontally and vertically. At the same time, we use a hybrid loss combined with an edge operator and mean square error to further improve the stability of the designed network. Comparison experiments on the synthetic SEAM model and field seismic data demonstrate the effectiveness of the adopted components, 2-D convolution, CA, and hybrid loss function in improving the lateral continuity of inverted impedance. For field seismic data, the impedance predicted by the proposed method shows improved lateral continuity and high resolution compared with the 1-D network and constrained sparse spike inversion method using commercial software (InverTrace Plus module in Jason).

为构建高性价比数据库系统,解决油气勘探领域地震资料处理系统更新中数据库迁移问题,提出将Oracle数据库迁移到PostgreSQL数据库的数据迁移解决方案.对地震资料处理数据库数据组织结构进行分析,针对迁移难点提出以地震项目工区为迁移单位和基于模板机制的数据迁移方法.通过设计SQL模板、Perl转换程序及SQL脚本等实现了基于Linux平台的数据库迁移工具.以一个时间偏移地震项目工区为案例进行验证,结果表明,数据迁移完整性与正确性为100%,为油气勘探领域数据库迁移提供了参考.

Seismic facies characterization plays a key role in hydrocarbon exploration and development. The existing unsupervised methods are mostly waveform-based and involve multiple steps. We have developed a method to leverage unsupervised contrastive learning to automatically analyze seismic facies. To obtain a stable result, we use 3D seismic cubes instead of seismic traces or their variants as inputs of networks to improve lateral consistency. In addition, we treat seismic attributes as geologic constraints and feed them into the network along with the seismic cubes. These different seismic and multiattribute cubes from the same position are regarded as positive pairs and the cubes from a different position are treated as negative pairs. A contrastive learning framework is used to maximize the similarities of positive pairs and minimize the similarities of negative pairs. In this way, we can enforce the samples with similar features to get close while pushing the samples with different features to be separated in the space where we make the seismic facies clustering. This contrastive learning framework is a one-stage, end-to-end, and unsupervised fashion without any manual labels. We have determined the effectiveness of this method by using it to a turbidite channel system in the Canterbury Basin, offshore New Zealand. The obtained facies map is continuous, resulting in a stable and reliable classification.

The conductive model of complex shaly sandstones is used to describe the rock-electric characteristics, which is the key to reservoir saturation evaluation. At present, conductive models as a single factor are unable to accurately reflect the conductive property of complex shaly sandstones, which limits the evaluation precision of reservoir saturation. In this paper, by incorporating multiple factors of shale, pore structure, and conductive structure, a novel modified equivalent rock element model (MEREM) is developed to analyze the rock-electric characteristics and calculate the reservoir saturation in complex shaly sandstones. Our studies show that pore structure and shale significantly influence the conductive property of complex shaly sandstones. However, they have the opposite effect and may cancel out each other. Moreover, the conductive model presented here has achieved promising results in interpreting experimental data. Furthermore, the MEREM is extended to oil-bearing shaly sandstones, demonstrating that the rock resistivity at different saturation is sensitive to pore structure and shale. The MEREM is applied to predict the reservoir saturation, and the computed saturation is found to be well-matched with cores. Therefore, the proposed MEREM is good for interpreting rock-electric characteristics and the evaluation of reservoir saturation in complex shaly sandstones.

Due to the influence of multiple factors on the conductive properties of rocks, the Archie's formula, considering only a single factor, makes it difficult to reasonably explain rock-electric characteristics of cracked porous rocks. In order to better describe the conductive mechanism of cracked porous rocks, a generalized multifactor conductivity model was proposed by considering and introducing multiple influencing factors such as the series-parallel structure, conductive matrix, cracks, and fluids, which is conducive to more accurate research on the conductive mechanism of rocks. It should be noted that the developed model is not only applicable to cracked porous rocks but also useful for porous rocks. Through the study and analysis of various influencing factors, it is demonstrated by the simulation results that both the conductive matrix and cracks improve the conductive ability, which are crucial factors resulting in the non-Archie behavior and low-resistivity pay zone, and rock conductivity is more sensitive to the conductive matrix and cracks in tight reservoirs with porosity below 10%. Furthermore, experimental data are available to validate the novel multifactor conductivity model, and the comparison results show its advantages in predicting and explaining the conductive properties of cracked porous rocks.

传统偏移成像方法是建立在波场一次反射的假设条件之上,事实上,完整的地震成像来自于地下全波场(一次波和多次波).为了能够利用全波场信息以提高成像质量,提出了基于单程波算子的全波场最小二乘偏移(FW-LSM)方法.首先,引入地层上、下界面反射系数和背景速度,推导了基于单程波算子闭循环延拓的全波场正演算子,模拟了平滑速度模型情况下的全波场信息;其次,在二范数意义下求解FW-LSM的误差泛函和梯度项表达式,构建基于反演框架的全波场最小二乘偏移方法;最后,针对透镜体与加入水层的Marmousi模型进行测试分析,验证了方法的有效性.研究表明:多轮次反演迭代压制了复杂波场产生的相干假象,多次反射波信息的利用显著改善了成像品质;该方法拓宽了地震成像的手段,尤其在多次波发育的海域地震资料处理中有着重要的作用和潜在的推广价值.

渤海湾盆地南堡凹陷海陆过渡带的地下构造复杂,由于地表及地下黏弹介质吸收,导致地震波能量衰减严重,造成地震信号强度和频带宽度的损失,为了提高成像精度,需要补偿地震波能量在地层中的衰减.反Q滤波和时频分析方法没有考虑地震波的传播路径,然而实际地震波的能量衰减是与传播路径密切相关的.为此,采用基于Kirchhoff叠前深度偏移的地震成像方法,在偏移过程中考虑了沿不同路径传播的地震波能量衰减,从能量归位和衰减补偿两方面提高偏移成像质量.应用Q叠前深度偏移,利用Q层析方法获得了 Q模型,基于Q模型进行Kirchohoff叠前深度偏移成像,有效补偿了振幅、校正了相位畸变,提高了深部数据的信噪比、保真度及构造解释精度,为叠前反演、岩性识别和流体检测提供了可靠依据.

Suppressing multiples from seismic records is necessary to improve imaging quality. Deep neural networks (DNNs) can automatically mine features from data. Once a network is successfully trained, it can process data with extremely high efficiency. In this letter, a generative adversarial network (GAN) framework is proposed to remove surface-related multiples in both synthetic and field datasets, where the generator is U-Net with Markov discriminator. Adding self-attention (SA) blocks to GAN improves processing precision. Improved signal noise ratio (SNR), and accurate reverse time migration (RTM) images implemented by network's outputs of synthetic datasets, jointly support that this network is effective on surface-related multiple suppression. Based on the results from field application, deep learning method in this letter is comparable to conventional adaptive surface-related multiple elimination (SRME) method but time-saving. By constructing an end-to-end workflow for seismic surface-related multiples suppression, small batches dataset can be used to train the network, and large batches of datasets can be processed accurately and efficiently.