Geophysical Prospecting最新文献

Enhancing seismic data resolution is a crucial step for geological interpretation and imaging. Deep learning-driven resolution enhancement primarily depends on sophisticated network architectures and extensive datasets. A lightweight seismic super-resolution model based on contrastive learning and knowledge distillation is proposed. Knowledge distillation is implemented by training a compact student network to mimic a powerful teacher model, thereby reducing reliance on extensive datasets and complex architectures. Contrastive learning is leveraged to align the bottleneck features encoded from the teacher network with the ones from the student network across different noisy inputs. The student network's total loss comprises a supervised loss with ground-truth labels, a distillation loss with the teacher's pseudo-labels and a feature-matching loss derived from the bottleneck features of both networks. The comparative experiments were conducted on four field datasets and 3200 pairs of slices extracted from 800 pairs of synthetic three-dimensional seismic cubes. Experimental results demonstrate that the proposed model achieves similar to or better performance than the comparison models for noise suppression and weak signal recovery, even with only $��6.8�\%�$ parameters and $��37.5�\%�$ training data compared to the reference model.

The solution of magnetotelluric equations is used to determine the apparent resistivity and to model the electromagnetic field's behaviour within the Earth. In this paper, we extend the scaled boundary finite element method (SBFEM) to compute the solutions of magnetotelluric equations. The salient features of the proposed framework are that internal features and boundaries are captured through a quadtree decomposition. The SBFEM handles the resulting hanging nodes as a part of local refinement efficiently without needing additional constraints or shape functions. Further, we employ patterns to speed up the computations of the essential matrices without compromising accuracy. The results from the present approach are compared with other approaches, and it is seen that the SBFEM framework is not only efficient but also accurate. The efficacy and robustness are demonstrated with a few examples.

Multiples are generally considered coherent noise in conventional seismic migration. If they are not appropriately separated or eliminated from the observed reflection data, it can result in significant artefacts of the migration image, which will adversely affect subsequent structural interpretation and reservoir description. Inspired by the state-of-the-art model-driven deep learning methods, we have proposed a self-supervised deep-learning approach for multiple elimination. The proposed method contains two parts. The first one is that we use the conventional multiple prediction approach to predict the initial surface-related multiple reflections. The second part is to build a multiple-elimination model based on a deep neural network. In the deep neural network, the input is set as the predicted initial surface-related multiples from a conventional method and the label is set as the observed reflection data with primaries and multiples. Therefore, the proposed approach is a kind of self-supervised deep-learning multiple-elimination model. The deep neural network component of our proposed approach can be interpreted as a corrector in conventional methods that performs amplitude and phase correction on predicted multiples. Moreover, we combine the advantages of L₁ and L₂ loss functions and introduce the attention mechanism to improve the inversion efficiency and accuracy of self-supervised deep networks. Through some experiments with synthesized and field data, we demonstrate that the proposed self-supervised deep-learning approach can effectively and efficiently eliminate multiples from the observed data. It excels in both accuracy and efficiency compared to traditional method.

We propose a novel physics-guided deep learning framework for geophysical inversion that integrates Langevin Monte Carlo (LMC) sampling to quantify uncertainties in model parameters. A statistical sampling strategy is employed to enhance computational efficiency by reducing the number of required samples while preserving diversity and informativeness. The training data for the supervised learning networks are iteratively expanded with outputs from a stochastic sampler and their corresponding observed responses, ensuring representative coverage of the model space. The Jensen–Shannon divergence is adopted as the loss function for training the network model, in which the Gaussian assumption is applied to enable analytical computation. The developed workflow is evaluated on reservoir porosity inversion, where it successfully reconstructs porosity patterns in the subsurface, yielding results that closely match the reference model. Compared to traditional LMC algorithm applied to the entire data cube, the proposed approach attains substantial computational efficiency by leveraging an active learning strategy that identifies and utilizes a limited yet representative subset of the observations. The results demonstrate the effectiveness of the proposed method, highlighting its potential for application to a wide range of geophysical inverse problems.

Seismic inversion can effectively establish the connection between seismic data and underground reservoir parameters. Aiming at the current problem of low accuracy of deterministic inversion, a seismic attribute-oriented deterministic inversion method is proposed. The method is similar to most inversion algorithms and consists of two parts: modelling and inversion. The innovation of the model lies in extracting seismic attributes and learning the mapping between the seismic attributes of the well-side and the parameters to be inverted based on the support vector regression (SVR) algorithm. Then, the mapping relationship is used to realize the modelling of the parameters to be inverted in the well-free area. Under the constraints of this model, seismic inversion is implemented through the Markov Chain Monte Carlo (MCMC) approach, yielding inversion results that exhibit strong consistency with the corresponding seismic responses. As the multi-trace structural attributes contain more high-frequency information, the resolution of the parameters to be inverted based on this model is also higher. We continue to conduct inversion tests using post-stack seismic data. The results show that seismic attribute-oriented inversion has a significant advantage in inversion resolution over partially deterministic inversion algorithms (model-based inversion, sparse spike inversion, etc.).

We present deep learning (DL) networks for three-dimensional (3D) joint inversion of active seismic full waveform and passive seismic traveltime data to image reservoirs and their properties and quantify imaging uncertainties. Active seismic full-waveform data can provide high-resolution monitoring images but are collected only intermittently because of their high acquisition cost. In contrast, passive seismic data can be gathered at relatively low cost between regular active surveys, although their imaging quality can be compromised by factors such as low signal-to-noise ratios and limited ray coverage of the target. Although these datasets are routinely acquired together at CO₂ storage sites, their combined inversion within a 3D DL framework has not been previously demonstrated. To our knowledge, this is the first study to address this gap, combining the strength of both data types. For efficient data storage and DL training with large 3D seismic datasets, we use a 3D data matrix in which a random number of passive seismic traveltime data are stored as parabolic envelopes using one-hot encoding and a 3D full-waveform data matrix in which multiple shot gathers are summed. Two network architectures are evaluated: a single-encoder U-Net for single-data type inversion and a dual-encoder U-Net for joint inversion of active and passive seismic data. We also evaluate the single-encoder U-Net for joint inversion by concatenating full-waveform data and traveltime data. We propose a systematic approach for selecting an optimal dropout rate that balances regularization during training and Monte Carlo dropout-based uncertainty quantification during prediction by examining the correlation coefficient between standard deviation and prediction error, along with the training misfit, across a range of dropout rates. 3D DL inversion experiments include five different network configurations, with evaluations under ideal, noisy and dropout-enabled conditions. Both model and data uncertainties are assessed, as well as their combined effects. Across all conditions, the networks consistently predict accurate CO₂ saturation models with low prediction errors, such as a structural similarity index of 0.993 and CO₂ difference of 1.1%. Uncertainty estimates show strong spatial correlation with prediction errors, confirming the effectiveness of the proposed dropout selection approach. The results demonstrate that our DL approach, utilizing compact data representations and appropriate uncertainty quantification, yields accurate subsurface images under various inversion conditions and provides valuable insights into the reliability of predictions.

As demand for large-scale seismic data interpretation tasks increases, machine learning-based horizon autotracking methods have gained attention in the geological and geophysical fields. Although such methods have demonstrated time- and cost-efficiency in large-scale data interpretation, studies on the expansion of interpreted horizons into the reservoir characterization process are relatively limited. Hence, a reservoir characterization process that can incorporate the machine learning-interpreted horizons and their structural uncertainties into the reservoir uncertainty assessment is necessary for an efficient reservoir modelling process. The proposed workflow consists of various modelling processes, including horizon construction where machine learning-interpreted horizons are used instead of manually interpreted horizons, facies modelling and petrophysical modelling. The modelling algorithms are based on stochastic methods: sequential indicator simulation for facies models and Gaussian random function simulation for petrophysical properties. Each modelling process incorporates variables such as variogram parameters, facies ratios and modified porosity values. The results show promising performance in incorporating machine learning-interpreted horizons into the uncertainty quantification process and analysing their impact by capturing the influence of structural uncertainties of horizons in the final reservoir pore volume.