Journal of Applied Geophysics最新文献

This study evaluates the performance of multiple machine learning (ML) algorithms and electrical resistivity (ER) arrays for inversion with comparison to a conventional Gauss-Newton numerical inversion method. Four different ML models and four arrays were used for the estimation of only six variables for locating and characterizing hypothetical subsurface targets. The combination of dipole-dipole with Multilayer Perceptron Neural Network (MLP-NN) had the highest accuracy. Evaluation showed that both MLP-NN and Gauss-Newton methods performed well for estimating the matrix resistivity while target resistivity accuracy was lower, and MLP-NN produced sharper contrast at target boundaries for the field and hypothetical data. Both methods exhibited comparable target characterization performance, whereas MLP-NN had increased accuracy compared to Gauss-Newton in prediction of target width and height, which was attributed to numerical smoothing present in the Gauss-Newton approach. MLP-NN was also applied to a field dataset acquired at U.S. DOE Hanford site.

Road underground defect detection plays a crucial role in assessing transportation infrastructure. Ground penetrating radar (GPR) serves as a widely used geophysical tool for this purpose. However, the traditional manual interpretation of GPR images heavily relies on the experience of the practitioner, leading to inefficiency and inaccuracies. To tackle these challenges, this paper proposes an automatic detection method for underground defects of roads based on an improved YOLOv5s model. First, the dense connection structure is integrated in the C3 module of the backbone to form the Dense-C3 module to enhance the capability of feature extraction. Subsequently, a convolutional block attention module (CBAM) is incorporated after each Dense-C3 module to refine features and enhance efficiency. Furthermore, the focal loss function is employed for the confidence loss to mitigate the impact of sample imbalance on detection performance. Experimental results demonstrate that the proposed model achieves a mean average precision (mAP) of 96.4% for synthetic data and 91.9% for real data, outperforming seven other models. The detection speed of the proposed model for real data reaches 51 frames per second, meeting the real-time detection requirements of road underground defects.

The Nilüfer district experienced the most recent urbanization among the central districts of Bursa in South Marmara region with the completion of rapid construction. Since 358 BCE, many destructive earthquakes were reported on the branches of the North Anatolian Fault (NAF) which caused extensive damage to buildings and loss of life near Bursa city. Besides some studies conducted to define the soil behavior in the vicinity of Bursa, a seismic hazard study in Nilüfer is still lacking. We, therefore, carried out a microzonation study including the following steps. First, an earthquake hazard analysis was conducted and the peak ground acceleration (PGA) values were determined for an expected earthquake. In the next step, MASW (Multi-Channel Analysis of Surface Wave) measurements conducted at 54 points in 28 neighbourhoods of Nilüfer district were evaluated. Soil mechanical parameters were determined at 11 boreholes to assess the liquefaction potential. It was found that almost half of the study area suffers from low damage considering only the vulnerability index (Kg) index, which depends on the site effect. Therefore, in addition to the Kg values, we created a microzonation map using the results of soil liquefaction, settlement, changes in groundwater level, and the average values of spectral acceleration. The study area is classified by four damage levels changing from low to high. Using only the Kg index could not quantify the potential damage level in the study area, thus we showed that the districts of Altınşehir, Hippodrome, Ürünlü and Alaaddinbey, Ertuğrul, 29 Ekim, 23 Nisan, Ahmetyesevi and Minareliçavuş were identified at potentially high-risk damage zones. The results of this study clearly showed that considering the Kg index, which depends only on the local site effect, may lead to inadequate damage values.

The shear wave (S-wave) velocity plays a crucial role in interpreting the lithology in seismic data, identifying fluids and predicting reservoirs. However, S-wave velocity is often unavailable due to the high cost of measurement and technical constraints. Conventional methods exhibit limitations that potentially impact the accuracy or efficiency on predicting S-wave velocity. Moreover, these methods always ignore the uncertainty quantification associated with the predicted results. This paper proposes a sparse Gaussian process regression (SGPR) method to predict the S-wave velocity in tight sandstone reservoirs. SGPR is a highly efficient regression technique that is based on the Gaussian process regression (GPR) method. In the SGPR method, inducing inputs are introduced to approximate the kernel matrix to decrease the computational complexity. A sparse set of inducing inputs and kernel hyperparameters are optimized through minimizing the Kullback-Leibler (KL) divergence between the exact posterior distribution and the approximate one. In this study, we select several types of logging data, which include porosity, water saturation, shale content, lithology and P-wave velocity, as the inputs for the SGPR method to predict S-wave velocity. To validate its effectiveness, we use the SGPR method to predict S-wave velocity in tight sandstone and compare the results with those from the GPR method, the bidirectional long short-term memory (BiLSTM) method and the Xu-White model. Additionally, we conduct cross-validation to demonstrate the robustness of the SGPR method. Our findings indicate that the SGPR method presents better performance and significant advantages about the accuracy and efficiency. Moreover, the SGPR method offers uncertainty quantification for the predicted S-wave velocity.

Accurate prediction of S-wave velocity from well logs is essential for understanding subsurface geological formations and hydrocarbon reservoirs. Machine learning techniques, including clustering and regression, have emerged as effective methods for indirectly estimating S-wave logs and other rock properties. In this study, we employed clustering algorithms to identify similarities among well log datasets, encompassing depth, sonic, porosity, neutron, and apparent density, facilitating the discovery of correlations among various wells. These identified correlations served as a foundation for predicting S-wave values using a novel semi-supervised approach. Our approach combined clustering, specifically k-means clustering, with different types of regressors, including Least Squares Regression (LSR), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP). Our results demonstrate the superior performance of this integrated approach compared to traditional regression methods. We validated our methodology using various parametric and non-parametric regression techniques, showcasing its effectiveness not only on wells within the training region but also on wells outside the study area. We achieved a significant improvement in the $R^2$ score metric, ranging from 2.22% to 6.51%, and a reduction in Mean Square Error (MSE) of at least 31% when compared to predictions made without clustering. This study underscores the potential of machine learning techniques for accurate prediction of S-wave velocity and other rock properties, thereby enhancing our comprehension of subsurface geology and hydrocarbon reservoirs.

Distributed Acoustic Sensing (DAS) is an effective exploration technology for acquiring Vertical Seismic Profile (VSP) data due to its characteristics of high-density collection and strong environmental adaptability. However, DAS-VSP is susceptible to various noises that distribute non-uniformly in both t-x and frequency domains. Existing denoising methods generally adopt single feature-extraction mechanisms (e.g. local convolutional operation or long-distance attention calculation), which are not sufficient for non-uniform feature extraction. Therefore, leveraging the advantages of Convolution (Conv) and Transformer, we propose a high-low level feature fusion model for DAS signal recovery. This model comprises three modules: low-level feature extraction (LFE), high-level feature extraction (HFE), and signal recovery (SR). First, LFE utilizes a Conv layer to extract the basic features, including energy, attributes, and fuzzy contours. The Conv utilizes small kernels to fitter the effective signal feature and introduce spatial information for the following layers. Second, HFE is the core module of the network to extract rich high-level features, such as sharper waveform features and high-dimension representation features. HFE consists of the Swin-Transformer blocks and the Conv blocks. The Swin-Transformer blocks utilize cross-window attention to extract the features between the windows and shift the window to continue recognizing the global features. Then, the Conv blocks further filter and enhance the high-attention features. The cross-use of these two blocks realizes the extract-enhance-extract-enhance process. Finally, the SR module employs a residual connection to create a direct mapping to add the low-level features to the last layer, achieving the fusion of the low-level and high-level features. Through the fusion, more complete and detailed features can be used to improve the accuracy of the recovering weak signals. The design of our model can combine long-distance and local detailed information to extract rich high-low level features, facilitating the recognition of weak signals and non-uniform noise in complex geological structures.

In this paper, we investigate the full-waveform inversion (FWI) for recovering three media parameters of the poroelastic wave equations as training a neural network. We recast the poroelastic wave simulation in the time domain by the staggered-grid schemes into a process of recurrent neural networks (RNNs). Furthermore, the parameters of RNNs coincide with the inverted parameters in FWI. The algorithm of FWI with a stochastic gradient optimizer named Adam is proposed. The gradients of the objective function with respect to the media parameters are computed by the automatic differentiation. FWI is implemented numerically for three media parameters, i.e., solid density, Lamé parameter of of saturated matrix and shear modulus of dry porous matrix. The numerical computations with two designed models show the good imaging ability of the described method in this paper. It can be applied to invert more media parameters of the poroelastic wave equations.

The oilfield's further fine development is significantly impacted by the interlayer of calcareous sandstone. Projecting the lateral distribution of subterranean calcareous sandstone is crucial for determining sequence boundary division, reservoir quality, and even CO₂ storage. Research on the sensitive characteristics of calcareous sandstone is still lacking. This study computes the percentage of lithologic difference and performs an intersection analysis of rock physical properties. It is found that Young's impedance, Poisson's ratio, and S-wave modulus have pleasurable sensitivity to distinguish calcareous sandstone. On the basis of this, a new sensitive factor for calcareous sandstone was built. The traditional approximate YPD reflection coefficient equation is only applicable to the weak contrast interface, and the accuracy is limited. This difficulty is solved in this paper by deriving a new equation for the nonlinear reflection coefficient. The equation is expressed by Young's modulus, Poisson's ratio, S-wave modulus, and density. Finally, the broadband nonlinear inversion method is adopted to provide a reasonable low-frequency model for the inversion of parameters. This allows for the realization of a stable inversion of parameters. The simultaneous broadband nonlinear inversion of Young's modulus, Poisson's ratio, and S-wave modulus provides a novel approach for calcareous sandstone prediction. We tested the accuracy and rationality of the method with both synthetic and field data examples.

The present study attempts to explore the efficacy of self-organizing maps (SOMs) in understanding the pattern of seismic reflections and analyze their implications for revealing the subsurface tectono-depositional environment prevailing within the Oligocene-Miocene intervals of the Upper Assam foreland basin, NE India. A series of seismic attributes including geometrical, spectral, amplitude, and GLCM-textures are extracted using high-resolution three-dimensional seismic data acquired from the upper shelf of the basin. These attributes are amalgamated into two different cases to compute the SOM models with an aim to highlight the subsurface structures and reveal sedimentary deposits engulfed within these structures. It is observed that the model SOM Case 1 highlights subsurface fault networks that structurally control the Oligocene-Miocene intervals. However, the model SOM Case 2 not only hints at the presence of these structures but also illuminates different patterns of seismic reflections and geomorphic features associated with sediment entrapped within the fault-bounded structures. Through this research, we envisage that for the SOMs to be optimal, geologically meaningful sets of seismic attributes should be used as an input such that attributes assisting seismic interpreters could successfully identify relations or patterns within the data. The method presented in this study can be applied to similar geologic settings to aid subsurface interpretation.

Marine controlled-source electromagnetic (MCSEM) inversion plays a crucial role in hydrocarbon exploration and pre-drill reservoir evaluation. Deep learning techniques have been widely used in geophysical inversions. Although they work on theoretical data well, their performance on survey data needs to be improved. Since no constraint of physical laws is applied in the training phase, the trained neural network often exhibits large errors when extended to new datasets with different distributions from the train set. To solve this problem, we add a differentiable marine EM forward operator at the end of the neural network that maps the network-predicted results back to the response data. We incorporate a data error term to the loss function and the gradient of data error with respect to model parameters in the gradient back-propagation process so that we can successfully introduce the physical law constraints into the network training process. Experiments on synthetic data validate the effectiveness of our Physics-driven Deep Neural Network (PhyDNN) inversions. It performs significantly better than the conventional DNN as it can recover the model accurately while maintaining data fitting. Tests on theoretical data with different noise levels further demonstrate the superiority of our PhyDNN, which can achieve stable inversions under high noise levels. Moreover, we use the t-distributed stochastic neighbor embedding (t-SNE) algorithm to analyze the similarity between the train sets and real data. The results show that the real data falls within the data distribution of the train sets, ensuring the credibility of the inversion results. Finally, we use PhyDNN to invert an EM survey dataset acquired over a deep-sea sedimentary basin. The inversion results match well Occam's inversions, indicating that our physics-driven network has enhanced the data adaptability and overcome the limitation of conventional DNN in handling new data.