The research object of oil and gas exploration is gradually shifting toward complex oil and gas reservoirs. The periphery of the Banghu Syncline in the Qianjiang Sag is a typical inland salt lake deposit. The complex thin sand and mudstone interbedded reservoir structure in the Qian-3 section of the Banghu Syncline area requires high-precision and high-resolution exploration technology to support actual production. In view of this, the study conducts forward modeling analysis and complex lithology classification based on well logging data. First, it analyzes and calculates the lithological data based on the original logging data, analyzes the lithological characteristics of sandstones containing different fluids (water-bearing sandstones, oil-bearing sandstones, and dry-layer sandstones), establishes different wedge forward models based on the convolution theory, and investigates the seismic response characteristics of different lithological combinations. Then, lithology curves are reconstructed using the K-means algorithm with known logging lithology data, and density attributes are used to correct natural gamma values for lithology classification. Finally, a geological model of the $ \mathrm{E}{\mathrm{q}}_{3}^{4} $ oil formation is designed to study the effects of changes in reservoir thickness and fluid content on amplitude.Model analysis and the example demonstrate K-means algorithm can effectively divide salt rock, sandstone, and gypsum mudstone with a prediction accuracy of 90.4%, and the forward model based on high-resolution logging information is consistent with actual geological characteristics. Therefore, it is feasible to analyze the reflection characteristics of salt rock and mudstone interbedded layers and thin sandstone layers by the forward model established through lithology curve reconstruction with logging data.

Faults identification of seismic data plays an important role in oil and gas exploration. At present, the application of machine learning and deep learning techniques has enhanced faults identification precision and efficacy. However, the fracture prediction outcomes remain challenging to meet the production needs. Therefore, a Transformer-based seismic fault identification method, namely 3D SwinTrans-U-Net, is proposed. The network consists of Swin Transformer module and convolution module. Among the aforementioned modules, the Swin Transformer module employs the attention mechanism of the Transformer to extract global information, and transforms global attention computing into window attention computing, resulting in less computational complexity compared to the Transformer. The convolution module, with the property of inductive bias, avoids the Swin Transformer's defect of weak inductive bias. Finally, the U-Net structure is utilized to combine the Swin Transformer and the convolutional layer. Thus, the structure can achieve deep and shallow information fusion, relevant feature extraction, as well as full learning of global and local dependency information. All these improve computational efficiency while ensuring fault identification accuracy, and enable end-to-end seismic fault deep learning. Synthetic and field seismic data tests have proven that the 3D SwinTrans-U-Net network can further improve the accuracy of fault identification.

With the development of offshore oil and gas exploration and development, the application of dual-sensor ocean bottom receivers is becoming more and more extensive. As a key technology of dual-sensor seismic data processing, the separation of the up-going and down-going wavefield determines data processing qua- lity. In response to the failure of conventional methods to complete separation of up-going and down-going wavefields (up-going wavefield separated include some of the down-going wavefield, and the down-going wavefields separated also include some of the up-going wavefield), this paper proposes a separation method of up-going and down-going wavefields for dual-sensor seismic data based on direct wave calibration. First, the spatial weighted function of the dual-sensor seismic data is calculated by using the dual-sensor direct waves in the frequency-space domain. Then the calibration filter operator of the dual-sensor seismic data is directly computed by using the dual-sensor seismic data with and without direct waves in the time-space domain. Finally, the dual-sensor seismic data is calibrated, and up-going and down-going wavefield are separated. The data examples demonstrate the effectiveness and practicality of the proposed method. The separated up-going wavefield data not only eliminates the interference of the ghost multiples but also improves the signal-to-noise ratio and resolution of seismic data. The method provides high-fidelity up-going and down-going wavefield data for subsequent joint deconvolution and migration imaging processing.

As the random noise in seismic data is highly irregular, conventional denoising methods often fail to achieve satisfactory results, thereby hindering subsequent interpretation and analysis of seismic data. Therefor, a seismic signal denoising method based on the diffusion model is proposed. The forward diffusion process of this method involves adding a certain degree of noise to the seismic data, transforming it into noisy seismic data with a large amount of isotropic Gaussian noise. Then, the trained diffusion model is usedtoreconstruct the noisy data and improve its signal-to-noise ratio. The prediction network component is an improved U-Net network, which incorporates attention modules and ResNet modules. These modules elevate the network's attention on important regions and mitigate gradient disappearance in deep networks. Both theoretical and practical data have verified the effectiveness of the proposed method. In terms of denoising effect, it significantly surpasses traditional denoising methods such as FX filtering and SVD, and it also outperforms classic deep lear- ning networks like CNN and GAN. This approach effectively preserves valid signals, thereby significantly enhancing seismic dataquality.

Rendering traditional modeling methods are inadequate for fluvial reservoir characterization due to the reservoir's rapid lateral variations and strong heterogeneity. To address this, we initially employ steerable pyramid technology to decompose and reconstruct seismic data, enhancing the geological laws crucial for sedimentary facies characterization. Subsequently, we apply Bayesian-sequential Gaussian seismic-constraint modeling to the reservoir, establishing constraints between logging data and seismic attributes to enhance the model's vertical resolution. Modeling practices in the fluvial reservoir of the Sha-2 member in Shengli Oilfield demonstrate that, after the steerable pyramid process, seismic data, lateral characterization improves, with effective sand body boundary identification. Compared to traditional Sequential Gaussian methods, the Bayesian-sequential Gaussian approach achieves higher vertical resolution, with an 86% match between sand body thickness and the actual value, making it more effective for sand body identification. This method provides guidance for remaining oil recovery.

In conventional seismic data processing, velocity analysis relies heavily on manually picking the velocity spectrum. However, this method is time-consuming, labor-intensive, and restricts the efficiency and accuracy of large-scale 3D seismic data processing. To address this issue, this paper propose an automatic velocity spectrum picking method using a dual-path convolutional neural network (DCNN). Firstly, this paper utilize a convolutional neural network combined with an attention mechanism as the main network, extract the features of energy clusters from velocity spectrum data, and realize the automatic picking of velocity. Secondly, this paper train a main neural network to integrate the information of velocity and the hidden representation of the uncorrected CMP gather input by another convolutional neural network (auxiliary network) through feature fusion and feature transformation before outputting the time-velocity sequence, reconstructing the corrected CMP gather. Finally, the process of CMP gather dynamic correction is simulated through the auxiliary network, and the accuracy of speed picking is optimized using dynamic correction. Model and real data tests demonstrate that, after incorporating dynamic correction information through the auxiliary neural network, the proposed velocity spectrum picking method achieves a higher accuracy than a single CNN in velocity picking.

To address the shortcomings of traditional lithology identification methods in missing logging curves handling, accuracy, and model interpretability, this paper proposes an interpretable lithology identification method based on a MSCNN-GRU neural network to complement logging curves, along with Optuna hyper parameter optimization for the XGBoost model. Firstly, to tackle the issue of lost and distorted logging curves in specific layer segments, the paper introduces a curve reconstruction method based on the combination of a multi-scale convolutional neural network (MSCNN) and a gated recurrent unit (GRU) neural network, which provides an accurate data basis for subsequent lithology identification. Secondly, the wavelet packet adaptive thresholding method is employed for denoising and normalizing the data, so as to mitigate the impact of noise on lithology identification. Next, the Optuna framework is utilized to determine the hyperparameters of the XGBoost algorithm, and an efficient lithology identification model is established. Finally, the Shapley additive explanations (SHAP) interpretability method is used for the attribution analysis of the XGBoost model, which reveals the contribution of different features to lithology identification and enhances the interpretability of the model. Experimental results demonstrate that the proposed Optuna-XGBoost model achieves a comprehensive lithology identification accuracy of 79.91%, outperforming traditional methods such as support vector machine (SVM), naive Bayes, and random forest, by margins of 24.89%, 12.45%, and 6.33%, respectively. The proposed lithology identification method based on the SHAP interpretability of the Optuna-XGBoost model has enhanced accuracy and interpretability, aligning well with the practical requirements of lithology identification in production scenarios.

The total organic carbon (TOC) content is an important evaluation index for shale gas exploration and development. Logging data can efficiently assess TOC, but it cannot be used for TOC prediction in inter-well areas. The TOC-sensitive factors extracted from seismic data can achieve three-dimensional (3D) prediction. However, due to the thin thickness and strong heterogeneity of shale reservoirs, it is difficult to achieve the required resolution by relying solely on seismic data. Therefore, it is necessary to comprehensively use multiple data sources to improve the accuracy of TOC assessment. For this purpose, a high-precision quantitative prediction method for shale TOC based on a convolutional neural network (CNN) is proposed. Firstly, the correlation analysis between the measured TOC data of the core from drilling and multiple logging characteristic curves is conducted on the Longmaxi Formation shale in southern Sichuan, and the most representative and sensitive features are selected. Secondly, based on the identified sensitive parameters, a CNN prediction model is constructed. The measured TOC samples and the training samples constructed by sensitive logging parameters are divided into datasets at a ratio of 7:3 for model training and validation. Finally, the high-resolution sensitive parameter inversion results obtained by simulation of seismic waveform indication are used as the feature input for 3D TOC content prediction. The sensitive parameters are rearranged, reorganized, and then input into the CNN model to achieve 3D TOC content prediction. The research results show that CNN has more advantages than multiple regression and back propagation (BP) neural networks in fitting the nonlinear relationship between TOC content and sensitive parameters. The average absolute error and root mean square error are both less than 0.6% between the predicted TOC data and the measured values from drilling. The prediction results are consistent with the actual situation. This method has high accuracy and obvious advantages in 3D TOC content prediction of thin shale reservoirs.

In onshore exploration, surface waves are one of the common interference, and the amplitude-preserving effect of surface wave suppression has a significant impact on subsequent data processing. The traditional surface wave suppression method is difficult to effectively suppress the surface wave while protecting the reflected wave. Based on the characteristics of surface waves, which have low speed, low frequency, and high amplitude, this paper proposes a time-frequency domain amplitude-preserving surface wave suppression method based on Gabor transform. Firstly, the seismic trace is divided into several segments using Gabor transform. Then, surface waves are identified in the seismic trace segments based on the difference in the amplitude energy between surface waves and effective waves, achieving a fine separation of surface waves and effective waves in the time domain. Finally, bandpass filtering is applied to the seismic trace segments containing surface waves in the frequency domain, and surface waves are suppressed based on the difference in frequency between surface waves and effective waves. The seismic trace after suppressing surface waves is obtained using Gabor inverse transform. Model tests and actual data processing show that this method effectively suppresses surface waves while protecting the information of effective waves to the greatest extent, improving the the amplitude preservation of surface wave processing.

Fault recognition is an essential step in seismic data interpretation. The development of deep learning has effectively improved the efficiency and accuracy of automatic fault recognition. However, in automatic fault recognition, it is still challenging to accurately capture subtle structures of faults and effectively resist noise interference. Thus, in this study, we propose a high-resolution fault recognition network model, AHRFaultSegNet, based on the HRNet network and decoupled self-attention mechanisms. The decoupling of self-attention mechanisms combines spatial attention and channel attention, replacing parallel convolution layers in HRNet. This reduces the computational amount of traditional self-attention mechanisms while allowing the model to calculate the relevance of input feature on a global scale, thus more accurately modeling non-local features. In decoupled self-attention, the residual connection is employed to preserve the original feature, speeding up model training and better maintaining detailed information. Experimental results demonstrate that the proposed network model outperforms other common automatic fault recognition network models in performance evaluation indexes such as Dice, Fmeasure, IoU, Precision, and Recall. Through fault recognition experiments on synthetic seismic data and actual seismic data, this method is proven to be effective in subtle fault structure identification and robust in noise resistance.

Horizon tracking is a very important step in seismic data processing and interpretation. The existing automatic horizon tracking technology often has a poor tracking effect when encountering faults. Therefore, a pure data-driven multihorizon tracking method using dynamic time warping (DTW) is proposed based on the waveform similarity between adjacent seismic channels. Firstly, the characteristic values of the seismic channels are extracted and the seismic channels are divided according to the characteristics of troughs, peaks, or zero crossings so that the tracked horizon positions strictly follow the above characteristics. In this way, a high accuracy of the tracking results can be ensured. Secondly, the DTW method is used to calculate the characteristic value sequences divided into two adjacent seismic channels to obtain similar paths. Finally, all the similar characteristic value point pairs, extracted from the similar paths, are divided according to the reference horizon to obtain each set of horizons. The results of model data calculation and actual data tests show that the proposed method can quickly track the positions of each set of horizons in the target area, and it effectively overcomes the hindering effect of faults on horizon tracking. It has a certain generalizability to seismic data in different geological conditions and has a certain application value.

The gravity flow sedimentary sand bodies of in rife lake basins has various genetic types. Rapid facies transition, relatively small sedimentary scale, complex formation mechanism and distribution pattern, determines the complex seismic response characteristics of this type of reservoir. In this paper, the quality of existing 3D seismic data is analyzed, and the seismic data body suitable for target processing is selected for research. The polarity of seismic data of the target layer is analyzed by means of seismic response characteristics of oil shale in the fourth member of the Eocene Shahejie Formation and the multi-well calibration of positive and negative polar wavelets, which lays a foundation for target processing. According to the development characteristics of turbidite sand body and its geophysical response in the middle third member of the Eocene Shahejie Formation of Niuzhuang oilfield, the theoretical and practical forward models of different sand body combinations are designed. By selecting different frequency wavelets to complete the forward experiment, the reasonable parameters are determined to complete the three-parameter wavelet high-resolution target processing method, which shows remarkable results. The following conclusions are obtained: ①the high frequency components of the processed seismic data are obviously broadened compared with the old data, the compounded wave decomposition is realized, the characteristics of the wave group are clear, and the synthetic record calibration also shows that the processed data is obviously better than the original pure wave data; ②After processing, the continuity of the data wave group is improved, and the energy relationship is clear, which is conducive to the identification of turbidite sand and other lithology; ③The processed data is conducive to calibration. Through multi-well calibration, the fine structure model and well constraint model are established to complete seismic inversion, and the results are in good agreement with the actual drilling. On the basis of detailed stratigraphic division and correlation as well as geological research, the formation slice is adopted to find out the vertical development and plane distribution of the 5~8 major oil-bearing sand bodies of the middle third member of the Eocene Shahejie Formation, and expand the sand body boundary in the meantime. This research provides a new idea for the precise characterization of gravity-flow sandstone reservoirs in rife lake basins.

Velocity modeling is an important part of prestack depth migration, which usually requires lateral extrapolation of the formation velocity of observation points under layer constraints. However, in the early stage of velocity modeling, there is a lack of framework information such as the seismic interpretation layer. Therefore, a method for establishing a velocity model with Siamese networks based on curvature spectrum lateral similarity and improved cyclic structure is proposed in the article. Siamese network is currently a commonly used deep learning based object recognition and tracking network, which can quickly compare the similarity of target images without the need for manual labeling. The curvature spectrum can be seen as a two-dimensional image that reflects the characteristics and velocity information of the formation. Velocity modeling as a lateral feature similarity analogy problem can automatically obtain the framework and velocity update information of the formation by analogy with the curvature spectrum. Firstly, the prestack depth migrated gathers are converted into curvature spectra; Secondly, the curvature spectrum images are determined to be searched and its corresponding target tracking object, and the similarity coefficients between the current tracking target and the objective tracking target are calculated; Then, the reference curvature spectrum image and the current tracked object are updated based on the similarity coefficient; Finally, with all tracking objects traversed, a velocity model is established based on the layer velocity and depth of each tracked object. Theoretical models and actual data experimental results show that this method can quickly generate velocity models that are congruent with geological structures and stratigraphic characteristics without interpretation data.

The SAⅠFormation of the Saertu oil layer in the western slope area of Daqing Oilfield is mainly composed of meandering channel sand deposits. The low accuracy of identifying the boundary of the point bar sand bodies and the internal abandoned channels restricts the development plan of horizontal wells. Therefore, based on high-density seismic data, we adopt the migration processing technology based on dynamic Q compensation to significantly improve the imaging accuracy of the channel sand bodies, clarifying the wave group characteristics of sand body point bars on meandering channels and the internal abandoned channels on seismic sections.To further highlight the internal heterogeneity characteristics of meandering channel sand bodies on the plane, we utilize a wedge shaped seismic forward model to establish the maximum thickness of sand bodies that can be recognized by seismic data volumes at different frequencies. RGB three-color fusion technology is used on three S-transform-based single frequency bodies selected according to the value range of sandstone thickness distribution in the area to accurately predict the sand bodies with different thickness levels. At the same time, a response model of core facies-logging facies-seismic faciesis established for the boundary of the point bar sand bo-dies and the internal abandoned river channels, clarifying their oil-bearing properties.The research results show that the migration processing technology based on dynamic Q compensation has improved the recognition accuracy of channel sand bodies on seismic sections.The frequency divider RGB attribute fusion technology based on S-transform can accurately predict abandoned channels, point bar boundaries, and the heterogeneity therein on the plane. The established response model of core facies-logging facies-seismic facies can accurately predict the oil-bearing properties of different types of sand bodies. The application research results guide the deployment of 304 well locations. To be specific, the drilling success rate of 188 vertical wells reaches 98.0%, the actual drilling length of 116 horizontal wells reaches 387.8 m, and the average encounter rate of oil-bearing sandstones is 93.4%.

Electrical anisotropy typically exhibited by subsurface media increases the complexity of borehole-to-surface electromagnetic (BSEM) responses. Therefore, forward modeling for anisotropic media models and summarizing their characteristics are crucial for accurately interpreting EM exploration data. Currently, precise modeling for complex three-dimensional geological structures is one of the challenges in forward modeling. We employ a high-precision modeling method based on octree grids. By subdividing hexahedral elements into eight smaller elements to locally refine the grids, this method ensures the accuracy of modeling with fewer cells, thereby reducing computational costs. Additionally, to improve the accuracy of numerical solutions, we use higher-order shape functions during calculations. The three-dimensional anisotropic forward modeling method for borehole-to-surface EM surveys is achieved by a combination of octree grids and higher-order shape functions. The comparison of numerical solutions with analytical solutions of anisotropic layered models validates the correctness of the algorithm. We utilize octree grids to achieve precise modeling of complex anisotropic oil and gas models and set different horizontal and vertical resistivities for anomalous bodies. The results indicate that vertical resistivity has a greater influence on the response than horizontal resistivity. These findings provide practical guidance for field applications and lay the groundwork for further three-dimensional inversion of anisotropic media.

Gas hydrates, widely distributed along the South Shetland margin (offshore Antarctic Peninsula), have become a hot research topic nationwide and worldwide due to their significant role in addressing resources, geohazards, and global climate change issues. This paper reviews and summarizes the geophysical research on gas hydrate reservoirs in the South Shetland margin from the aspects of seismic identification characteristics of gas hydrate reservoirs, geological structures related to gas hydrates, quantitative evaluation of gas hydrate reservoirs, and the relationship between gas hydrates and climate change. Some ideas and understandings are obtained to provide useful references for future exploration and development of gas hydrates and research on climate change.Finally, prospects for geophysical research on regional gas hydrate reservoirs are presented. It is proposed that comprehensively utilizing the longitudinal and transverse wave information of underwater OBS data for 3D characterization of gas hydrate reservoirs and further researching the response mechanism and feedback effects of gas hydrates on global warming are important development directions in the future.

The absorption attenuation and dispersion of seismic waves are both related to frequency. Therefore, it is theoretically and physically reasonable to study the viscoacoustic reverse time migration in the frequency domain. Similar to that in the time domain, the method in the frequency domain will inevitably cause the amplification of high-frequency noise while compensating for the absorption attenuation, thus lowering the stability and imaging accuracy of the migration algorithm due to the instable compensation. To solve this problem, we use the Kolsky-Futterman model to derive a viscoacoustic wave equation in the frequency domain, and then achieve a stabilized viscoacoustic reverse time migration. First, we derive the above equation, where the amplitude attenuation is decoupled with the phase velocity dispersion, based on the Kolsky-Futterman model. Then, based on the decoupling characteristics of the equation, we construct a stabilized absorption attenuation compensation operator by utilizing the ratio of the dispersion-only and viscoacoustic wavefields. Finally, we use the stabilized operator to perform absorption attenuation compensation on both source forward wavefields and receiver backward wavefields, and further apply the cross-correlation imaging condition to perform the viscoacoustic migration imaging on the underground structure. The experiments and data applications show that the proposed method can effectively suppress the amplification of high-frequency noise while compensating for the absorption, ensuring the stability, imaging accuracy, and quality of the migration algorithm.

The surface condition of loess mountainous landforms is complex, the disturbance is strong, and the surface medium is heterogeneous. At the same time, due to the absorption attenuation of the surface thick loess, some problems have emerged in seismic data such as low signal-to-noise ratio, low main frequency, and narrow frequency band. After pre-stack fidelity noise attenuation processing, there are still random noise and scattering interference in the stack data after the pre-stack time migration, which restrict the precision accuracy of thin reservoirs.The proposed method in this paper breaks through the processing methods commonly employed in the industry to improve the signal-to-noise ratio and resolution by four-dimensional de-noising in the F-XY domain and post-stack zero-phase de-convolution methods. In this paper, the method of Cadzow filtering in the FX domain is employed to realize the synchronous suppression of scattering interference and random noise of 3D post-stack data. On this basis, continuous wavelet transform is adopted to improve the resolution.Various quality control methods are applied in each processing link to ensure the reasonable selection of parameters, including noise profile, F-K spectrum, amplitude spectrum, time slicing, frequency division scanning, and well seismic calibration. The processing methods have good fidelity and can provide a good data basis for subsequent inversion and attribute analysis.

With the deepening of exploration, the a priori information available for regularization inversion becomes increasingly rich and accurate. A model constraint method is proposed based on the principle of mutual differences in the physical properties of inversion units, after an analysis of methods that introduce a priori rock physical property constraint, including parameter transformation functions, the fuzzy C-means (FCM) clustering model constraint and the alternating direction optimization of multipliers. The physical properties are introduced via the imposition of constraints on the values assigned to discrete inversion units, given that the essence of the model constraint based on the principle of mutual differences among inversion units is the finite number of geological units and that each discrete inversion unit can only belong to one lithology. A unified objective function is constructed through the combination of the aforementioned introduction of a priori physical properties with classical regularization inversion and is then optimized through the Gauss-Newton method. The mechanisms and mathematics essence underlying the introduction of rock physical properties through four distinct schemes are examined, and a strategy of employing multiple schemes to improve the inversion effect is proposed. Finally, the inversion effect of combining L₁ norm minimum structure model constraints and the introduction strategy of a priori physical property constraints is compared with that of combining L₂ norm minimum structure model constraints and the same strategy. The inversion results of the model and field data all demonstrate that leveraging a priori rock physical property information can effectively improve the inversion effect.

The Radon transform is a commonly used algorithm in seismic data processing, which achieves seismic data interpolation, multiple attenuation, and wave field separation through specific path summation. However, due to the limitations of acquired data and algorithms, resolution in the transform domain is relatively low. Improving the resolution of the Radon domain has always been a research hotspot. The most frequently used method is the reweighted algorithm with many times of iteration, namely updating weights iteratively to focus the weighted value on the curvature of the seismic data, but it is difficult for the algorithm to focus it on the actual curvature position. Therefore, this paper proposes an optimized new method to improve the resolution of the transform domain, which involves calculating a weighting matrix within the dominant frequency band of the seismic data and focusing the weighted value on the actual seismic curvature position. Firstly, the dominant frequency of the seismic data is calculated, and a range around this dominant frequency is taken as the constrained frequency band. A lower-frequency constraint strategy is applied to the seismic data within the dominant frequency band, and the weighting matrix from low to high frequency is iteratively calculated. The final weighting matrix is obtained from the last weighting matrix within the dominant frequency band and is then applied to the calculations of all other frequencies. The data in the dominant frequency band have a high signal-to-noise ratio and strong amplitude, which thus indicates the proposed method is more stable, and the weighting matrix obtained through it can significantly enhance the resolution of the transform domain. In addition, compared with other iterative methods, it avoids the iterative process of each frequency's weighting matrix, with a higher computational efficiency. Tests on synthetic and real data have proven the effectiveness and advantages of the proposed method in multiple attenuation.

The two dimensional forward modeling of magnetotelluric (MT) fields based on the Gauss-Lobatto-Chebyshev (GLC) basis function spectral element method is put forward to improve the accuracy and efficiency of the numerical simulation of magnetotelluric fields. Under the guidance of this method, we derive the 2D MT boundary value problem and then transform it into the integral weak form through Galerkin weighted residual method. At last, we discretize the global problem using GLC interpolation basis functions and obtain MT fields after solving the large-scale sparse linear equations system using the Pardiso solver, successfully numerically simulating it. To improve the computational efficiency of the numerical simulation, we adopt a variable density regular grid generation technique. This technique, by using finer grids in electrically complex areas and coarser grids in electrically homogeneous areas, can reduce computational time. In addition, parallel processing of multiple frequencies is achieved using OpenMP programming. Numerical simulation results of a one-dimensional layered media model validate the correctness and accuracy of the proposed algorithm, showing a higher accuracy of the GLC polynomial spectral element method compared with GLL. Forward modeling based on the GLC polynomial spectral element method, finite difference method, and triangular mesh finite element method is conducted for the COMMEMI 2D-1 model and the terrain model, and the comparison results show that the GLC method has a higher accuracy and less grid dependence.

As an effective method to improve the vertical resolution of seismic data, the deconvolution technique has been being a research hotspot in the field of seismic exploration. The traditional seismic deconvolution method adopts a trace-by-trace sparse inversion strategy to reconstruct the subsurface reflectivity series, which ignores the spatial relationship between adjacent seismic traces, resulting in poor spatial continuity of the deconvolution result. Although the multichannel seismic deconvolution method based on total variation (TV) constraint can mitigate the spatial discontinuity caused by the trace-by-trace inversion, it is only applicable to block structures because it does not consider the geological structural information. To overcome this problem, a multichannel seismic deconvolution algorithm with adaptively directional TV (ADTV) constraint is proposed in this paper. The approach constructs the ADTV regularization based on the local geological structure direction, unlike traditional TV regularization that only consider horizontal and vertical directions. Therefore, ADTV can better protect weak reflection signals, more clearly depict the spatial distribution of the subsurface geological bodies, and guarantee the spatial continuity of the deconvolution result in the case of complex geological structures. Both synthetic and field data processing results show that the proposed method can provide deconvolution results with high resolution and strong spatial continuity.

In OBN seismic data acquisition, the coupling between the geophones and the seafloor often leads to the fact that the z-component of the land inspection receives shear wave noise. Shear wave noise seriously interferes with the subsequent processing, imaging, interpretation, and reservoir monitoring of OBN data. Besides, the frequency band and curvature of the shear wave noise are very similar to the effective signal, which is difficult to suppress through conventional methods. Therefore, an OBN shear wave noise suppression method based on parabolic Radon transform and unsteady adaptive subtraction is proposed. Firstly, this method transforms the P-component and z-component data of the OBN data into the τ-q domain through the parabolic Radon transform method and utilizes the difference in propagation speed between the shear wave noise and the reflected signal to separate shear wave noise and effective wave as much as possible in the τ-q domain. Secondly, the shear wave noise is extracted by removing the reflected signal component in the P-component from the z-component data containing the shear wave noise through the high flexibility of the unsteady adaptive subtraction algorithm. Finally, the extracted shear wave noise is transformed back to the spatio-temporal domain and removed from the original z-component data. The reflected signal in the recorded data is preserved as much as possible while realizing the elimination of noise.Processing results of actual data demonstrate that this method can effectively suppress the shear wave noise in OBN data and improve the signal-to-noise ratio of the seismic data.

Seismic data reflectivity inversion is a critical step to connect reservoir parameters and seismic data, which remains a research hotspot. At present, reflectivity inversion is mostly in the form of the sparse-spike deconvolution based on the L₁-norm constraint. In recent years, the emergence of the odd-even decomposition algorithm has weakened the inter-wavelet tuning effect, which makes the spectral inversion based on the L₁-norm constraint receive further applications. The sparse constraint ability is related to the accuracy of the inversion reflectivity remaining to be solved. Because of the insufficient sparsity constraint ability of the L₁-norm and the Lp norm, this paper introduce the transformed L₁ (TL₁) sparse constraint, which is conducive to the obtainment of a more accurate inversion reflectivity. Meanwhile, given that the fitting ability of the large reflectivity needs to be enhanced, this paper propose the reweighted TL₁ (RTL₁) norm to further enhance the sparse constraint ability. Parameter tests show that the reconstruction ability of the reweighted norm is better than that of the non-reweighted norm, which proves the effectiveness of the RTL₁ norm in sparse reconstruction. Models and field data processing demonstrate that the RTL₁ norm is more effective in enhancing the accuracy of reflectivity inversion in spectral inversion compared to conventional sparse constraints.

Quality factor Q is an important parameter to characterize seismic wave attenuation, which is widely used in reservoir fracture development characteristics and oil & gas bearing identification. However, under the constraints of different media models such as isotropic, anisotropic, and viscoelastic models, the obtained Q values have different reservoir characterization capabilities. To accurately describe the anisotropy and fracture development characteristics of fracture type, fracture-porosity and porosity-fracture reservoirs, we propose a new method for reservoir fracture prediction based on the HTI medium medium model and seismic wave attenuation anisotropy theory in the OVT (offset vector tile) domain. Firstly, we use generalized S-transform (GST) to perform high-precision time-frequency analysis of OVT pre-stack gather. Then, we calculate the single frequency T-K energy of seismic wave by Teager-Kaiser energy operator. After that, we calculate the relative attenuation within the sampling wavelength distance to obtain the attenuation Q-value for ellipse fitting. Besides, we utilize ellipse oblateness to characterize attenuation anisotropy strength and to predict fracture development density. This method not only gives full play to the advantage that OVT pre-stack gather contains more reservoir fluid and fracture information than post-stack data, but also links attenuation attributes with reservoir fracture characterization, providing a new approach for fracture prediction. In the application of XC gas field in Western Sichuan Depression, the method has achieved the fracture identification effect with high consistency between core analysis, logging interpretation and seismic prediction.

Fluid identification is critical for reservoir prediction. However, in isotropic media, the Russell fluid factor is mainly obtained through the inversion of the approximate formula based on the Zoeppritz equation or the indirect calculation of elastic parameters. However, the inversion based on the approximate formula is prone to disturbance by assumptions such as a small angle of incidence and a low contrast, which affects the inversion accuracy. Therefore, this paper proposes the reservoir fluid identification method based on the exact Zoeppritz equation. Firstly, the exact Zoeppritz equation is rewritten to include the Russell fluid factor, Poisson's ratio, and density. Then, the objective function for the simultaneous inversion of the fluid factor, Poisson's ratio, and density is constructed under the Bayesian inversion framework. The iteratively reweighted least square (IRLS) method is used to solve the inversion objective function nonlinearly. The test results of synthetic data and actual work area data show that the proposed method can effectively identify fluids, which indicates that this method can provide new research ideas on reservoir fluid prediction and reservoir fluid calibration.

Inspired by the electromagnetic perturbation method, we regard the influence of fractures as the perturbation of electromagnetic fields of undisturbed formations. Based on the solution of the Helmholtz equation in radial layered media, we propose an analytical solution of induction logging electromagnetic fields applicable to any fracture occurrence, any well diameter, and any mud resistivity after studying the analytical solution of the electromagnetic field of infinitesimal antenna vibrator outside the column. Relying on this, we take the HDIL 1515 EA/MA array induction logging tool commonly used in China as an example to calculate the induced electromotive forces of receiver coils in four kinds of fractured formation models and draw the (theoretical charts of) apparent conductivity curves. The research shows that there is not necessarily a spike in the array induced apparent conductivity curve in the fracture development formation. At the same time, under the condition of oil-based mud, the apparent conductivity curves of the seven sub-arrays in the formations where microfractures are densely developed show amplitude differences, and they can be used to identify the microfracture development section for the different characteristics compared with mud invasion formations.

Intelligent seismic velocity inversion is currently a hot and challenging topic in seismic exploration research. Nevertheless, the complex structure of deep learning networks demands significant computing power from hardware devices, which restricts the application of the model in scenarios with large data volumes and high timeliness requirements. To address these practical issues, in this paper, the U-Net is improved based on the concepts of feature engineering and model lightweighting, and the inversion networks U-Net vG for GPU and U-Net vC for CPU are proposed. Firstly, the characteristics of the velocity inversion network are analyzed to deduce the lightweighting principles of convolutional neural networks. Subsequently, lightweight processing is conducted on the multi-scale module, attention gate module, and feature extraction module to obtain a lightweight convolutional neural network for velocity modeling, which reduces the network volume while maintaining prediction accuracy. Data test results demonstrate that the training process of the proposed network has lower requirements for high-performance hardware resources, and that the network enables efficient velocity inversion, possesses higher seismic velocity inversion accuracy, and exhibits superior noise resistance. It provides a new idea for solving the computing power bottleneck problem in seismic data inversion.

Horizon tracking is basic and important work in seismic data interpretation, but the accuracy of conventional methods makes it difficult to meet the requirements of actual production. Therefore, this paper proposes a method for high-precision multi-set horizon tracking based on U²-Net. Firstly, a method is designed to produce filling labels, which traverse each pixel of seismic data to determine the location of each pixel and divide a horizon region for it. The method can automatically search for the adjacent horizons for the fault-crossing horizons, thus realizing the production of filling labels for the seismic reflection horizons and unconformities under the complex conditions of partial horizons and faults. Then, a U²-Net model is used to train the F3 data and the M-zone seismic data by utilizing the filling labels. Compared with the U-Net+PPM model, the U²-Net model has higher prediction accuracy, better stability, stronger generalization, and shorter training time. In addition, its accuracy and Mean Intersection over Union both exceed 95% in predicting the seismic reflection horizons in complex areas. The proposed method can be better adapted to the horizon tracking of seismic data with low signal-to-noise ratio.

The compressed sensing (CS) seismic exploration technique can increase the trace density and improve the image quality at the same cost. The key challenge of the technique is the missing seismic data reconstruction. The reconstruction method based on sparse domain constrained inversion has the advantage of high reconstruction accuracy. However, the low computational efficiency limits its application in the industry. To solve the issue, this paper proposes a fast iterative thresholding 3D seismic data reconstruction method based on curvelet transform by optimizing the conventional reconstruction method based on sparse domain constrained inversion and applied it to industrial data processing. 3D curvelet transform has an excellent sparse signal characterization ability, and the computation redundancy is high. In actual processing, seismic data segmentation is adopted, followed by parallel computation of 3D curvelet transform coefficient, which improves computational efficiency. In addition, a fast iterative thresholding algorithm is designed to improve the convergence speed and the reconstruction accuracy in comparison with the conventional method. Finally, amploying for the CS acquisition data, the proposed method is validated to be effective in data reconstruction and is applicable to data reconstruction in industrial production.

With increasingly complex oil and gas exploration targets, seismic exploration faces challenges such as low signal-to-noise ratio (SNR), low resolution, and difficulties in velocity modeling and imaging of seismic data. Conventional seismic data processing and interpretation methods have certain limitations in accuracy or efficiency when applied to massive seismic data. The artificial intelligence (AI)-based seismic data processing and interpretation methods can effectively improve accuracy and efficiency. To this end, this paper provides an overview of supervised, semi-supervised, and unsupervised deep learning techniques, and summarizes the applications of deep learning in data processing such as first break picking, SNR improvement, data reconstruction, velocity spectrum interpretation, migration, and resolution enhancement. Meanwhile, it discusses the applications of deep learning in identifying geological bodies such as faults, seismic facies, river channels, and salt domes, as well as in wave impedance inversion, AVO inversion, full waveform inversion, lithology identification, reservoir parameter prediction, and fluid identification. The production of training sets, optimization of neural networks, training strategies, and large models are discussed, with an outlook on the development trend of AI-based processing and interpretation methods for seismic data provided. It is pointed out that the generalization of networks should be continuously increased and large models suitable for seismic exploration should be studied.

Karst caves exhibit distinctive "string-of-beads" reflection configurations on seismic profiles, with their spatial distribution governed by intricate fracture networks and thus forming complex fracture-cave systems. Conventional methods, constrained by ambiguities in reservoir architecture and limited sample availability, face challenges in achieving accurate delineation. This study proposes a knowledge graph-guided intelligent identification technique based on coupled fracture-cave modeling, which innovatively integrates geological prior knowledge with deep learning through encoding geological topological relationships into adjacency matrix constraints. The methodology establishes a multi-task learning framework by synergistically combining forward modeling-derived label data volumes with expert-annotated data volumes. The approach employs knowledge graphs to characterize connectivity relationships between fractures and karst caves and designs geologically interpretable loss functions to dynamically adjust model optimization trajectories. Application in the Ordovician Lianglitage Formation of the Tarim Basin demonstrates substantial reduction in manual interpretation workload and significant enhancement in boundary delineation precision for fracture-cave systems. This methodology presents an innovative solution integrating knowledge-driven and data-driven approaches for prediction of strongly heterogeneous carbonate reservoirs.

Horizon tracking is a key step in seismic data interpretation.It is typically performed manually by interpreters in a human-computer interaction manner, which results in low efficiency. Convolutional neural network （CNN） can establish a nonlinear mapping relationship between seismic data and training labels to achieve horizon tracking. However, since it is difficult to obtain manually interpreted results, models trained merely with a few labels tend to have relatively poor generalization capability. Therefore, a semi-supervised horizon tracking method based on a convolutional neural network is proposed to transform horizon tracking into image segmentation between horizons and faults. First, the unlabeled data is trained by the autoencoder. Then a small amount of labeled data is used for supervised learning after part of the parameters are transferred to the supervised learning network. Finally, the seismic data of the whole working area is predicted, and the edge of the segmentation result is extracted as the horizon tracking result. The test results of both synthetic data and the real data show that compared with the supervised learning horizon tracking method, the proposed method pre-sents less error segmentation and smaller errors between the horizon extracted from the segmentation edge and the artificial horizon interpretation results, and thus has better generalization capability.

This paper addresses the challenge of shallow subsurface geological stratification by investigating the feasibility of joint inversion using direct current (DC) resistivity data and Rayleigh wave data. The study focuses on applying the quantum-behaved particle swarm optimization (QPSO) algorithm enhanced with centroid opposition-based learning (COBL) and chaos search (CS), named the COBL-CS-QPSO algorithm, which integrates multiple optimization strategies, in one-dimensional joint inversion of the two methods. The joint inversion approach enables the extraction of layer thickness information from resistivity data, thereby overcoming the limitations of Rayleigh wave inversion in accurately resolving layer thickness. The incorporation of multi-strategy algorithms mitigates the risk of solutions becoming trapped in local optima during the search process and improves the efficiency of random searches under uncertain conditions. In theoretical model settings, various scenarios are examined, including cases with and without noise, as well as with known and unknown model layer numbers. The inversion is performed over a broad search range, yielding favorable results. Subsequently, the joint inversion algorithm is applied to actual data. The results demonstrate that the joint inversion of the DC resistivity and Rayleigh wave with the COBL-CS-QPSO algorithm produces more accurate outcomes than single-method inversions under field conditions lacking borehole data or detailed subsurface stratification. Furthermore, a comparison with the adaptive particle swarm optimization (APSO) algorithm highlights the advantages of the improved algorithm in inversion performance.

With the continuous breakthrough of exploration and development technology, the Qiangtang Basin has been called an important potential area of oil and gas resources in China. The Qiangtang Basin is located in the northern part of the Qinghai-Xizang Plateau. Complex geological structural factors, strong wind interference, unique perennial distribution of plateau permafrost, alpine and anoxic environments have brought many challenges to seismic exploration. The non-uniform distribution of high-speed permafrost can shield seismic wave energy, distort seismic wave travel time, and reduce the signal-to-noise ratio and consistency of seismic data. How to correctly understand the propagation mechanism of seismic waves in permafrost has become a key issue in seismic exploration in a plateau environment. Firstly, in response to the problem of permafrost in the Qiangtang Basin, this study measures the rock physical elastic parameters of permafrost in the field by using rock physical measuring equipment special for plateaus and establishes a complex structure model with permafrost by combining the field understanding and the actual two-dimensional seismic reflection profile. Secondly, the seismic wave field characteristics of permafrost are studied by finite difference elastic wave forward modeling. It is found that permafrost makes the surface wave more developed, which influences seismic wave travel time. When the direct wave field has a "cap" feature, permafrost shields the seismic reflection amplitude, resulting in a weaker seismic reflection amplitude. Finally, by comparing the reverse time migration profiles of the model with permafrost and the model without permafrost, it is found that the seismic reflection imaging energy below permafrost is weaker. The research results of seismic propagation characteristics and mechanism of permafrost provide theoretical support for permafrost identification, field seismic data acquisition, permafrost energy compensation, and migration imaging in Qiangtang Basin.