We first classify various basin effects, as the result of sequential combinations of basic wave phenomena, into wedge effects, lens effects, and surface-wave related effects. We then investigate their contributions to the basin aggravation. Our results show that, in a shallow basin, current 1D-based approaches to assess site effects can significantly underestimate the seismic hazard in its close-to-edge region inside the basin at spectral periods below the fundamental period of the soil column at the deepest portion of the basin. In contrast, for a deep basin, ground motions across the whole-width could be underestimated using 1D approaches.

Seismic inversion and petrophysical inversion are the most common methods used in exploration geophysics to obtain elastic and petrophysical parameters, which are essential for reservoir characterization. However, they are commonly ill-posed problems and both are usually performed independently. Recently, deep learning has been successfully applied to the solution of inverse problems (e.g., seismic inversion and petrophysical inversion) by using large amounts of labeled training data to establish a mapping relationship between the input and the target. On the one hand, the performance of deep learning-based inversion depends heavily on diversity of the training dataset. However, the number of wells in actual production is limited, which greatly limits the application of deep learning-based inversion methods. On the other hand, deep learning-based inversion methods usually calculate elastic and petrophysical parameters independently, which lacks clear physical meaning and leads to large computational errors. To overcome these problems, by considering the spatial variability of elastic and petrophysical parameters from well-log data, a large amount of realistic pseudo-well-log and post-stack seismic data are first generated based on geostatistics to obtain the diversity of data required for network training. Meanwhile, we propose a fusion neural network architecture to build a physically meaningful network to simultaneously estimate acoustic impedance and porosity within the reservoir conditioned to post-stack seismic and limited well data. The fusion neural network consists of two subnetworks based on the spatio-temporal neural network. The method is validated by the application of real data and compared to traditional inversion approaches.

Vehicle-mounted ground-penetrating radar (GPR) technology is an effective means of detecting railway subgrade diseases. However, existing methods of GPR data interpretation largely rely on manual identification, which is not only inefficient but also highly subjective. This paper proposes a semi-supervised deep learning method to identify railway subgrade diseases. This method addresses the sample imbalance problem in the defect dataset by utilizing a data augmentation method based on a generative adversarial network model. An initial network model for disease identification is obtained by training the YOLOv5 network with a small number of existing samples. The intelligently extended samples are then labeled to achieve a balance in the disease samples. The network is trained to improve the recognition accuracy of the intelligent model using a more complete dataset. The experimental results show that the accuracy of the proposed method can reach up to 94.53%, which is 23.85% higher than that of the supervised learning model without an extended dataset. This has strong industrial application value for railway subgrade disease detection as the potential learning ability of the model can be explored to a greater extent, thereby improving the recognition accuracy of subgrade diseases.

Prediction of elastic parameters (e.g., P-, S-wave velocity, and density) from observed seismic data is one of the most common means of reservoir characterization. Recently, deep learning (DL), as a data-driven approach, has been attracting increasing interest in seismic inversion. DL is proven to have the potential to learn complex systems and solve inverse problems efficiently. One of the most key components of DL is the training dataset, and an effective training dataset is a prerequisite for the success of DL-based methods. In seismic inversion, the training dataset needs to be artificially expanded due to the limited number of actual training data pairs. Traditional approaches of using the exact Zoeppritz equation (EZE) or its approximations for training dataset construction have limitations, principally, the single interface assumption and the neglect of wave propagation effects. Alternatively, the analytical solution of the 1D wave equation (i.e., reflectivity method-RM) can simulate the full-wave, including transmission losses and internal multiples, and can be executed in a target-oriented manner. Inspiration from this, we develop a data-driven elastic parameter prediction method based on waveform formulation. The method uses RM to construct training dataset, which both compensates for the inadequate training dataset in data-driven seismic inversion and improves the accuracy of the inversion results. We implement the method in a synthetic model as well as field data. The results are compared with model-driven methods (EZE and RM) and data-driven method based on EZE, and it is shown that the proposed method outperforms these three methods.

Seismic attenuation compensation is a widely used technique for enhancing the resolution of nonstationary seismic data. An accurate and reliable quality factor ( $Q$ ) is a prerequisite for attenuation compensation and an important indicator of oil and gas. $Q$ -factor estimation and attenuation compensation are ill-posed inverse problems, and they are poorly robust when noise is present. Moreover, the process is usually performed in steps, which leads to an accumulation of errors. Deep-learning (DL)-based methods have gained great popularity because of their powerful ability to obtain exact solutions for inverse problems. However, most DL frameworks rely heavily on huge training datasets and lack clear physical meaning. For these reasons, we propose a fusion neural network architecture to build a physically meaningful network to simultaneously implement interval- $Q$ estimation and seismic attenuation compensation. The fusion neural network consists of two subnetworks based on the spatio-temporal neural network (STNN). And, the method ensures the coupling relationship between $Q$ -factor and nonattenuated/attenuated seismic data by building a loss function that contains both the $Q$ loss term and the seismic data loss term. The training dataset is then used to update the subnetworks simultaneously to obtain a fusion network that achieves both interval- $Q$ estimation and seismic attenuation compensation. We demonstrate the effectiveness of the proposed simultaneous interval- $Q$ estimation and seismic attenuation compensation algorithm by applying both synthetic and field data examples.

Seismic acoustic impedance bridges the gap between post-stack seismic reflection data and reservoir parameters such as lithology and porosity, and hence it plays an important role in stratigraphic interpretation. Due to sedimentation and propagation effects, both seismic records and impedance are a type of spatio-temporal data, i.e., there should be coupling between adjacent data points. However, most deep learning inversion methods only consider local shape information, ignore the time-series characteristics of the data, and are demanding on the training data, which makes inversion more difficult and leads to low inversion accuracy. For this reason, we develop a spatio-temporal neural network (STNN) to perform post-stack impedance inversion. The network consists mainly of a convolutional neural network (CNN) block and a recurrent neural network (RNN) block in series. Thus, the proposed method can take full advantage of CNN and RNN to capture the dynamics and correlations of seismic series in the spatio-temporal levels, yielding more continuous and stable results. We use an overthrust model example and an actual data case to test the performances of the STNN and demonstrate its advantages over traditional deep learning (i.e., CNN) based impedance inversion. Through a series of numerical experiments, we find that STNN produces more geologically reliable results, which not only ensure coupling relationships between adjacent points in the vertical direction, but also reveal well the stratigraphic variations in the lateral direction.