Seismic impedance inversion is one of the key research subjects in the field of seismic exploration. Its basic goal is to quantitatively predict the wave impedance of underground medium from the seismic data. In recent years, with the rapid advancement of artificial intelligence technology, many scholars have proposed a variety of deep leaning (DL) based seismic impedance inversion methods. These methods embed geophysical information, such as physical laws, empirical formulas, and prior expert knowledge, into deep networks from different perspectives, effectively reducing the multi-solution and enhancing the physical interpretability of the inversion problem. This paper presents a review of these methods, which incorporate geophysical information by three strategies: ① designing network architectures embedded with geophysical knowledge; ② applying data constraints; ③ constructing multi-objective loss functions. The network architectures embedded with geophysical knowledge include: a forward physical model module, a reflection coefficient inversion model module, a seismic data spatiotemporal feature representation module, and a synthetic-to-real domain adaptation module. Data constraints include: generating diverse synthetic samples to train the deep network, and quantifying prior knowledge as inputs to the network. Multi-objective loss functions encompass physics-informed regularization terms, including closed-loop loss, generative adversarial loss, dynamic time warping (DTW) loss, spatial structure loss, and uncertainty loss. These strategies can reduce the multi-solution and enhance the reliability of the inversion problem from different perspectives. Finally, this paper provides two prospects for the DL-based seismic impedance inversion method: ① multi-modal large models with robust comprehension and knowledge reasoning capabilities can be leveraged together with multi-modal data to enhance the generalization of the inversion model; ② the multi-model fitting method can be combined with physics-informed neural network (PINN) to transform the "one-to-many" problem of inversion modeling into a "one-to-one" problem within each seismic facies segment, thereby reducing the multi-solution of the seismic impedance inversion.

The Dongfang 1−1 diapir area in the Yinggehai Basin has superior conditions for hydrocarbon accumulation. Several shallow gas fields have been proven in this area using high-resolution towed streamer seismic data. However, the widespread seismic fuzzy zones in the medium–deep layers exhibit low imaging resolution, which cannot be addressed even through rounds of reprocessing, thus seriously restricting the hydrocarbon exploration process in the central depression zone. After identifying the genetic mechanisms of the fuzzy zones, this paper proposes a set of seismic acquisition and P-wave processing technology system based on ocean bottom node (OBN), and innovatively designs an "three highs and one low" (high signal-to-noise ratio, high fold, high quality, and low cost) OBN geometry-variable acquisition scheme to break through the interferences of shallow gas shielding and diapiric fractures. Moreover, the key technologies such as dual-sensor summation in the wavelet domain, full-waveform inversion (FWI) in viscoelastic media and Kirchhoff prestack depth Q-migration are combined to significantly improve the accuracy of OBN P-wave imaging. Applications show that the newly acquired seismic data are significantly improved in the signal-to-noise ratio (SNR) compared with the old data, thus enabling more accurate and reliable velocity modeling. The proposed technology system allows for the first realization of clear imaging of diapir cores in greatly reduced fuzzy areas, providing high-quality data for the hydrocarbon exploration in the medium–deep layers of the Dongfang 1−1 diapir area.

The discontinuous Galerkin finite-element method (DGFEM), which is a high-order finite element method adapting to complex surface conditions, has attracted extensive attention. Based on triangular unstructured meshes and local Lax-Friedrichs flux, the matrix forms of DGFEM calculation using elastic, viscoelastic, and poroelastic wave equations are established, and the general calculation format for single wave field components is developed, which improves the scalability of DGFEM programming. Based on this format, the procedure to construct a universal CUDA kernel is developed, which can be easily extended for more complex media and 3D cases, and the CPU+GPU parallel computing framework of 2D DGFEM is established. The results of a theoretical model and a complex mountain model reveal that the general calculation format and CUDA kernels constructed in this paper can accurately simulate P-waves, S-waves, and slow P-waves described by using acoustic, elastic, viscoelastic, and poroelastic wave equations. Compared to single-core CPU simulation, the speedup ratio of 2D DGFEM elastic-wave GPU calculation is about 100 on the average. Meanwhile, the simulation time for elastic, viscoelastic, and poroelastic waves is approximately 1.7, 2.3, and 3.0 times that of acoustic wave simulation, respectively. This result can be used to guide multi-process load balancing in the simulation of complex coupled media.

In seismic survey, seismic data are often polluted by serious random noise, which has a negative impact on subsequent processing and interpretation. The existing noise suppression methods of seismic data cannot effectively separate signals and noise, leading to the problems such as loss of image details and the introduction of false textures. In order to improve the quality of seismic images, a method for suppressing random noise of seismic data based on multi-scale information perception generative adversarial network (MSIP GAN) is proposed. Firstly, a multi-scale information perception generative network is designed to remove random noise from seismic data. The denoising network is based on the traditional deep convolutional neural network(CNN), and combines parallel multi-scale module, multi-channel information fusion module and consistency regularization module to improve the accuracy of texture structure and retain more details. Secondly, the label data of the discriminative network and the images of the generative network are constructed to assist in training the generative network. Finally, a composite loss function is designed to guide the generative network, thus improving the ability of the generative network to recover image details. The experiments on actual data from the Daqing Oilfield demonstrate that the proposed method achieves significantly enhanced noise suppression effectiveness compared to existing mainstream models and industrial software when removing noise of varying intensities. Furthermore, the experiments on the data from the offshore F3 block in the Netherlands reveal that this method exhibits a robust generalization capability.

The marine magnetotelluric (MT) method utilizes the natural MT signal received by the ocean-bottom electromagnetic (OBEM) stations, and calculates the apparent resistivities and phases based on the impedance tensor estimation to investigating the sub-oceanic conductivity structures.The OBEM often generate noises that appear as irregular high-frequency pulses with particular frequency characteristics. Consequently, the impedance tensors of the affected frequencies deviate from the true values, which makes it difficult to interpret the data correctly. To mitigate the influence of the impulse noise on marine MT signals, an impulse noise suppression method based on frequency-domain Hampel identification is proposed. Firstly, the low-frequency energy of marine MT data is reduced by first-order differential pre-whitening to improve the noise recognition rate of the Hampel identifier. Then, the frequencies affected by the impulse noise in the four components of MT data are identified using the Hampel identifier in the frequency domain. Finally, considering the stability of MT impedance estimation, the noise frequencies in the four components are merged and universally amplitude-attenuated to suppress the impulse noise, so that the apparent resistivity and phase distortion can be corrected. Based on the periodic difference between MT signal and impulse noise, the autocorrelation function which can be used for measured data is introduced to evaluate the efficacy of the proposed method in suppressing impulse noise. Moreover, the proposed method is tested using the synthetic and measured MT data. The results show that the proposed method can effectively suppress the impulse noise in the marine MT data, reduce the periodic peaks caused by the noise in terms of the autocorrelation function curve, correct the apparent resistivity and phase distortion induced by the noise, and thereby enhances the quality of MT data .

Surface wave suppression directly influences the imaging quality and interpretation accuracy of seismic data. In view of the limited identification accuracy and generalization of intelligent noise suppression based on a completely data-driven deep learning architecture, a physically constrained surface wave suppression technique is proposed. To enhance noise recognition, a UNET architecture with dual-channel input, jointly from the time-space domain and corresponding frequency-wavenumber domain before denoising, is constructed to establish the mapping relationship between input data and output noise data in the time-space domain. Based on the characteristics of surface waves as regular noises, structural similarity regularization operators are introduced into the loss function to further enhance the network's ability to recognize noises. To address different physical characteristics of surface waves in different work areas, noise frequency and apparent velocity distributions are used as the constraints for further processing of output noise data to obtain higher-precision noise predictions, which will be subtracted from input data using an adaptive subtraction algorithm to obtain final denoised data. The testing on several field data sets shows superior denoising accuracy and generalization capability of the proposed algorithm.

First-arrival picking is a crucial step in seismic data processing, as its accuracy directly affects velocity model building and static correction. Although conventional CNN-based deep learning methods have achieved remarkable results in first-arrival picking, their performance degrades in a survey with complex surface conditions, e.g. loess tableland, due to the weak energy of first arrivals and strong noises. To address this issue, we propose a deep learning-based first-arrival picking method that integrates multi-data fusion with an adaptive weighted hybrid loss function. To enhance the robustness of the method, seismic, offset, and elevation data are integrated to construct a multi-data fusion model. To enhance the accuracy of first-arrival picking, an adaptive weighting strategy is employed to optimize the combination of multiple loss functions and construct an adaptive weighted hybrid loss function, which effectively constrains the model training process. The tests on three field seismic datasets demonstrate that our method outperforms conventional methods, e.g. STA/LTA and deep semantic segmentation, in picking accuracy and noise robustness in the geologic complexity scenarios with weak first arrivals and strong noises. These results validate the effectiveness and robustness of the proposed method.

Velocity modeling is a key challenge in geophysical exploration. Although artificial intelligence methods enable efficient model building, they often lack domain-specific guidance and fail to consider the structural similarity between the migration image and velocity model, resulting in inaccurate geological structure interpretation. To address this limitation, we propose a Dix-guided intelligent velocity modeling method, which uses the structural similarity between imaging profiles and interval velocities as well as the Dix relation to optimize the neural network-based modeling process. Specifically, common midpoint(CMP) gathers and velocity spectra are used as inputs to extract features that reflect relative velocity magnitude and variation trends. Then, the stacked profile is introduced to guide the network to convert there features into interval velocities (hidden layers), which are structurally consistent with the profile. This process is constrained by root-mean-square(RMS) velocity lables and a newly designed Dix activation function. Experimental results demonstrate that the Dix activation function effectively guides the network to incorporate geological structures into velocity modeling, enabling it to learn simple morphological structural mappings and enhancing the model's generalization capability. The predicted velocity model accurately delineates stratigraphic boundaries and exhibits strong lateral continuity.

The shale oil resources of the Qingshankou Formation in the northern Songliao Basin exhibit substantial potential, highlighting extensive opportunities for exploration and development. However, the Gulong area is controlled by factors such as bedding density, fractures, and formation pressure, showing pronounced characteristics of strong anisotropy and attenuation in geophysical properties. Such factors lead to a complex seismic response mechanism in Gulong shale, leaving the response characteristics somewhat ambiguous. To tackle the viscoelastic anisotropy characteristics of the Gulong shale formation, a multi-faceted fusion approach integrating well logging data, seismic stratigraphy, structural features, and core experiments was utilized. This method refined the initial model through meticulous stratification, enabling the development of a high-precision and sophisticated viscoelastic anisotropy model on a meter scale for the Gulong shale area. The wave equation grounded on the constitutive relationships of a viscoelastic vertically transverse isotropy (VTI) medium model was developed. The staggered-grid finite difference algorithm was employed for the numerical simulation of the seismic wavefield in the Gulong shale formation. Findings reveal that under explosive source or shear-wave source conditions, the seismic amplitude of the Gulong shale is markedly influenced by the viscoelastic anisotropy. The seismic response outcomes under the strong anisotropy and attenuation characteristics in the Gulong area underscore the importance of incorporating Q-compensated anisotropic imaging technology for achieving high-precision seismic imaging in the region.

A large amount of statistical data has demonstrated a close correlation between porosity and permeability. However, traditional permeability prediction methods based on porosity and permeability empirical formulas exhibit significant errors and fail to meet the requirements of predicting the permeability of complex reservoirs. The advantage of artificial intelligence can be leveraged to explore the hidden relationship of data, and the multi-task learning and sharing mechanism can effectively alleviate the overfitting problem of single-task learning under the few-shot condition. In this paper, a seismic prediction method of reservoir permeability based on multi-task learning was proposed. The method employed post-stack seismic data and P-wave impedance as network inputs, with well-log porosity and permeability serving as labeled data of the network. Through network training, an optimal network model was established by integrating near-well seismic and well-log data, realizing the simultaneous prediction of reservoir porosity and permeability parameters between wells. Application results from the tight gas reservoir in the Shaximiao Formation of Jinqiu Gas Field, Sichuan Basin demonstrated high consistency between predicted permeability parameters of Sand Body No. 8 and actual drilling data, along with superior vertical and horizontal resolution. This validated the method’s effectiveness for spatial permeability prediction using seismic data.

For ultra-deep carbonate reservoirs with strong heterogeneity and complex hydrocarbon charge in Shunbei field, this study makes seismic response analysis and reservoir-characteristics-constrained AVO inversion. Using log, core, and seismic data, three reflection types and AVO characteristics of reservoirs in Shunbei field are investigated. Then, forward modeling is conducted on reservoirs containing different fluids, demonstrating that gas-bearing reservoirs induce significant changes in seismic amplitude and AVO response. This provides the theoretical basis for AVO inversion-based reservoir prediction in Shunbei field. Next, the stochastic model capturing the reflectivity variation of underground reservoirs is established, and the reservoir-characteristics-constrained AVO equation is derived, enabling the prestack inversion to obtain fluid-sensitive parameters. Application of this proposed technique in Shunbei field demonstrates the AVO inversion results in agreement with actual drilling results, proving it reliable for identifying ultra-deep carbonate reservoirs and characterizing hydrocarbon accumulations in Shunbei field.

Tight sandstone gas reservoir, a significant resource in hydrocarbon exploration, is characterized by low porosity, low permeability, strong heterogeneity, and complex pore structures. The log interpretation and reservoir evaluation are facing challenges. Current pore structure evaluation methods from well logs are limited, new log interpretation methods require investigation. Mercury injection capillary pressure (MICP) experiment serves as an effective means to characterize pore-throat structure and flow capacity, and nuclear magnetic resonance (NMR) can describe the distribution of different pore components. Combining these two methods, the pore structure of tight sandstone reservoir is obtained. Based on core MICP and NMR experiments, six parameters sensitive to pore structure are extracted, relationships between pore structure parameters and NMR component porosity are established, and the pore structure evolution from NMR log interpretation is achieved. For pore structure classification, K-means clustering algorithm, principal component analysis, and dimensionality reduction techniques are investigated from pore structure parameters of log interpretation. The above methods are applied to the Yanghugou Formation in the Weizhou area of the Ordos Basin. The pore structure classification was completed. Some case studies indicate that these methods improve the accuracy of pore structure classification and are suitable for other low-porosity, low-permeability tight sandstone reservoirs.

Complex resistivity and polarizability are important parameters in electromagnetic exploration to describe the electrical properties of underground reservoir rocks. To investigate the variations of complex resistivity and polarizability in fluid-saturated tight sandstone reservoirs, complex resistivity measurement was conducted at different temperatures and pressures using outcrop and core samples from a gas field in southwestern China. Based on the experimental results, we establish an induced polarization (IP) petrophysical model for tight sandstones in the study area and analyzed the variations of complex resistivity and polarizability with porosity and gas saturation. The results show that both complex resistivity and polarizability increase with gas saturation; in the context of equal petrophysical properties, resistivity is more sensitive to gas saturation than polarization; resistivity is more easily affected by reservoir petrophysical properties than polarization. The research results lay the foundation for the joint electromagnetic and seismic prediction of reservoir fluid properties.

Hydraulic fracturing is currently one of the most important techniques for enhancing oilfield production, with the primary objective of transforming shale formations to generate artificial fractures. Understanding the development process, depth, length characteristics, and distribution of these fracturing-induced fractures is crucial for the efficient exploitation of shale formations and the improvement of production capacity. In practical engineering applications, surface-based electromagnetic methods are significantly limited by detection depth and accuracy. To enhance the efficiency and accuracy of inter-well detection, it is feasible to conduct simultaneous observations from both downhole well patterns and surface arrays. The influence of various shale formation parameters in vertical wells on measurement results was investigated. By utilizing the principles of transient electromagnetic (TEM) methods, a grounded electric dipole source was employed to transmit signals. Hydraulic fracturing was carried out at one or multiple locations along the inner wall of the well casing, with surface fracturing equipment used to compress the sandstone around the shale, thus inducing fractures. A saline electrolyte was then injected, creating low-resistivity fractures around the wellbore. This resulted in a vertically distributed geological model with varying conductivity due to the injected fracturing fluid. Geoelectric data was acquired through a surface concentric circular array composed of three loops, with a total of 72 measurement points arranged along the loops, enabling the monitoring of the development of underground fracturing-induced fractures. Based on finite element numerical simulations conducted using COMSOL Multiphysics software, the response characteristics of fractures with different geometries in shale formations were analyzed. The results demonstrate that this method effectively determines the orientation, burial depth, and length of subsurface fractures, thereby providing a theoretical foundation and technical approach for the characterization of fracturing-induced low-resistivity fractures in subsurface formations.

With the development of deep learning, image segmentation models have been increasingly applied to obstacle detection in variable geometry design for 3D seismic acquisition. Most exploration areas cover a large and irregular range, and the remote sensing images used for observation system design are usually large in size and contain a large amount of invalid data. Deep learning methods for obstacle recognition focus on improving accuracy but lack pre-recognition processing of remote sensing images and post-recognition procedures such as obstacle boundary vectorization, which significantly limits their applicability. To overcome these limitations, we propose a "four-step" method for large-scale remote sensing image processing: remote sensing images preprocessing, automatic obstacle recognition, postprocessing, and variable geometry design for obstacle avoidance. This end-to-end workflow enables automatic obstacle localization in large-scale remote sensing images and subsequent variable geometry design. First, an adaptive preprocessing method is formulated to fit the minimum circumscribed rectangle of the valid data region for optimized image tiling. Next, an enhanced UMiT-Net (Softmax) model is proposed to simultaneously extract different types of obstacles. The subsequent vectorization techniques incorporate boundary extraction, stitching, expansion, and coordinate transformation to derive projected coordinates. Finally, an obstacle-avoidance algorithm relocates shot and receiver points to determine optimized shot coordinates for geometry adjustment. Experimental results demonstrate that the proposed method eliminates 50.4% of invalid image tiles in a real-world exploration area, improving computational efficiency by 102%. Obstacle detection achieves high accuracy and completeness, and vectorized boundaries enable rapid obstacle-avoiding geometry design. The end-to-end workflow is both efficient and user-friendly.