Conventional vertical seismic profiling (VSP) wavefield separation techniques often struggle to achieve high-accuracy results in practical applications due to three key factors: uncertainties in first-break estimation, the non-stationary characteristics of time-variant wavelets, and the influence of formation dip angles on wave propagation patterns. Most existing methods, including traditional filtering-based techniques (e.g., median filtering) and deep learning-based wavefield separation approaches, heavily rely on high-quality velocity models or training datasets, which limits their applicability and separation accuracy.Therefore, this paper proposes a physical prior-guided VSP wavefield separation method. Specifically, we choose physical prior information to guide the training of the deep learning model. We utilize the Curvelet transform and gradient structure tensor (GST) to extract physical prior information from VSP data. Combined with a dual-convolutional autoencoder model, this method effectively separates upgoing and downgoing waves. Experimental results on both synthetic and real data demonstrate that the proposed method outperforms traditional separation techniques, effectively reduces pseudo-axis effects, prevents energy leakage, and significantly improves the accuracy and robustness of wavefield separation.

The interpretability and reliability of pre-stack seismic waveform classification are often hindered by the entanglement of various geological factors in pre-stack seismic waveforms, along with their intrinsic uncertainties. To address the problem, we propose a physical knowledge-guided disentangled representation network for a specific interpretation task of pre-stack seismic waveform classification. Unlike physics-driven deep learning methods that incorporate physical laws directly into network structures or loss functions, our approach uses a physical knowledge constraint network to guide deep feature learning based on physical attributes. This enables the model to capture more interpretable and separable features. First the concept of group supervision and disentangled representation theory is applied to separate specific physical attributes within the pre-stack seismic waveform. Then, physical knowledge about the seismic data is incorporated to impose additional constraints during the disentangling of deep features. Finally, a clustering technique is used to cluster disentangled deep features and generate the clustering result, which can assist geological experts in inferring sedimentary environments and the distribution of reservoir. Testing on synthetic data demonstrates that our method effectively separates and reconstructs geological elements within seismic data, accurately classifies different types of reflection patterns, and shows strong robustness against noise. Furthermore, the application of actual data shows that the classification results of pre-stack seismic waveform using the proposed method have better interpretability.

Ultra-deep seismic interpretation requires the full utilization of the low-frequency components of seismic signals. Compared to time-frequency analysis (TFA) methods such as wavelet transform and S-transform (ST), the W-transform (WT) method can improve the time resolution of the low-frequency band of the time-frequency (TF) spectra and highlight the low-frequency components. However, there is a nondifferentiable problem in the standard deviation function of WT, which causes the peak energy of time spectra to split at the dominant frequency. To address this problem, a new double-parameter WT (DWT) method, including time-varying dominant frequency, is proposed to overcome the defects of original WT. The parameter set of the transform is optimized using a TF distribution concentration measurement criterion and Rényi entropy, with the optimal parameter set being selected adaptively. Test results of synthetic data and field data show that the DWT eliminates the problem of spectral energy splitting at the dominant frequency. In addition, the focusing performance of different frequency components in the TF spectra is superior. The low-frequency band of the TF spectra exhibits higher time resolution, providing an effective means to highlight the low-frequency information of ultra-deep seismic signals.

Seismic inversion accuracy significantly affects the quality of reservoir modeling. Given the limited samples of well logging, prior information such as geological data and stratigraphic characteristics is crucial for enhancing the accuracy and reliability of inversion results. The inability to fully exploit prior information in the data to compensate for data deficiencies may impair network performance. Moreover, the traditional artificial neural networks (ANNs) cannot fully exploit prior information in the data to compensate for data deficiencies, which may impair network performance and further leads to the network’s limited ability to effectively assimilate external prior knowledge. Therefore, a prestack seismic inversion driven by prior information neural network (PINN) and statistical characteristic is proposed to integrate various geological prior information. Initially, lithological analysis is conducted on the well logging data, leading to the classification of the strata into a series of sub-layers based on lithology. After obtaining the statistical characteristics of thickness and elastic parameters for each sub-layer, a geostatistical algorithm is employed to generate numerous pseudo-wells that conform to actual sedimentary laws. Subsequently, the PINN is pre-trained using substantial samples, encompassing both pseudo and real well logging data, to integrate geological structural prior information into its network parameters. Lastly, PINN served as a specially designed prior constraint for the inversion network, allowing its geological information to effectively guide the inversion process via backpropagation. This inversion method is applied to both synthetic and field examples, and in comparison to conventional inversion algorithms, it demonstrated superior accuracy in blind well verification.

Fault detection is one of the core tasks in fault interpretation, which is of great significance to the study of underground oil and gas transport and reservoir distribution. In recent years, with the development of artificial intelligence (AI), especially artificial neural network technology, many new intelligent fault detection methods have emerged. They often have fixed input sizes. When detecting faults in field seismic data, it is necessary to split the data into patches that fit the input size and then combine the detection results. Thus, the information used for fault detection is derived from a single data patch at most, without considering the relations between patches. We propose a Transformer-based network that considers relations between patches and incorporates a larger range of information for 3-D fault detection. Our network employs an encoder-decoder architecture. We construct the encoder using Transformers, which are more effective at extracting global features compared to convolutional neural networks. Moreover, we propose a feature fusion block based on Transformers, which is able to introduce the relations between adjacent data patches, allowing our network to utilize information beyond a single patch and extend to multiple patches. Compared to some existing studies, our network can cover a larger perspective. We apply our method on several datasets, including synthetic and field data, and our method performs well and shows improvements in noise resistance, fault continuity, and the ability to handle complex fault situations.

Fault surface extraction is a crucial step in fault interpretation, which aids in the analysis of subsurface oil and gas migration and reservoir distribution. One challenge of fault surface extraction is the ability to deal with complex fault situations. To enhance such ability, we transfer the fault attribute data into a multirelationship graph and employ the provided graph clustering method to obtain the distribution of each entire fault by considering the constraints of multiple interrelationships between faults. First, we develop a method for transforming fault attribute data into graph data, in which we can identify and modify problematic fault nodes, determine the edge relationships, and compute probability values that indicate the likelihood of fault nodes belonging to the same fault. Second, we propose a graph clustering method for multirelationship graphs, which can provide a systematic analysis of fault distribution by considering multiple relationships between fault nodes and obtaining the distribution of each entire fault. Finally, based on the spatial distribution of each fault, we extract all the fault surfaces in the data. The consideration of multiple relationships provides our method with the ability to figure out the distribution of faults in some complex situations, such as X-shaped faults and Y-shaped faults with similar fault orientations while avoiding producing erroneous results. We apply our method to several data and the results illustrate the effectiveness of our method. We also compare our method with an existing method, and our method outperforms the compared method by both qualitative and quantitative evaluations.

Applying artificial intelligence (AI) to data from Industrial Internet of Things (IIoT) devices is a novel direction in geological studies. However, privacy and security concerns hinder the sharing of data, thus affecting the performance of current AI-based approaches. In this article, we propose a novel data management style to address the privacy and security issues in joint hydrocarbon explorations. Federated learning can facilitate the analysis of multiple datasets without the need to share them, protecting private information of different companies in a virtual joint venture. We use the inference of petroleum reservoirs in karst stratigraphy as a case study. A federated learning-based enterprise data management framework is proposed to virtually integrate the information from different organizations. Our key contributions are summarized as follows. 1) A method for karst identification and inference is proposed, which uses neural networks to recognize the size of petroleum reservoirs in different karst areas. 2) A federated learning algorithm is applied to virtually aggregate data samples from different companies. 3) The performance of the new privacy-preserving integration model is compared with those of the individual/local deep learning models. Our results show that the proposed approach can substantially improve the accuracy of petroleum reservoir explorations.

The transfer learning between the source and target domain has already achieved significant success in machine learning areas. However, the existing methods can not achieve satisfactory result when solving the two distant domains transfer learning problem. In the worst case, it could lead to the negative transfer. In this paper, we propose a novel framework called transitive transfer sparse coding (TTSC) to solve the two distant domains transfer learning problem. On the one hand, as an extension of the sparse coding, the TTSC framework constructs a robust and high-level dictionary across three different domains and simultaneously obtains three good feature sparse representations. On the other hand, TTSC utilizes the intermediate domain as a strong bridge to transfer valuable knowledge between the source domain and target domain. Empirical studies validated that the TTSC framework significantly could outperform state-of-the-art methods.