As seabed topography is closely related to the ocean gravity field, utilizing gravity data for seabed topography inversion has become the mainstream method. Although conventional deep neural network (DNN) methods have great potential in bathymetric prediction, they can neither evaluate the uncertainty of the prediction process nor the impact of uncertainty on prediction results, which limits their practical application value. To address this problem, a Bayesian deep neural network (BDNN) method is proposed for bathymetric prediction and uncertainty quantification. This method introduces Monte Carlo dropout variational inference into the architecture of a conventional DNN. Thus, the model achieves uncertainty quantification of prediction results with only a small amount of network structure changes. In addition, the captured uncertainty is fed back into the network training process to constrain the model parameters and calibrate the bathymetric prediction results. The experimental results show that the proposed BDNN model provides more reliable and accurate bathymetric prediction results than the conventional DNN and seabed topography inversion models. Moreover, the uncertainty results quantified by the model have a significant spatial correlation with the seabed topography, providing high confidence in the prediction results and reducing the risk in interpretation of seabed topography, thus proving the potential of BDNN for accurate bathymetric prediction from gravity data.

Longitudinal and converted shear wave seismic data depict unique and abundant seismic response characteristics of gas reservoirs and help elucidate the multicomponent seismic attributes required for predicting gas-bearing distributions. Although machine learning (ML) methods, including deep learning, show potential for predicting gas-bearing reservoirs, most conventional ML methods combine longitudinal and converted shear wave seismic attributes as inputs for learning; however, this approach cannot fully extract the unique characteristics of gas reservoirs. Additionally, the prediction results are deterministic, with no room for uncertainty quantification. To address this issue, we developed a parallel linked Bayesian neural network (PLBNN) model for multicomponent seismic gas-bearing prediction and uncertainty quantification. First, we used unsupervised ML methods to optimize longitudinal and converted shear wave seismic attributes, reduce redundant information, and extract feature data. Then, we separately input the obtained feature data of longitudinal and converted shear waves into the feature-extraction layer of the PLBNN to extract high-dimensional features of the target gas reservoir. The extracted features were then fused through the feature-fusion module for gas reservoir prediction and uncertainty quantification. Finally, we input the uncertainty quantified using the Bayesian approximation method of Monte Carlo dropout into the network framework for model parameter optimization to further improve model prediction performance and reduce uncertainty. This method exhibited excellent predictive ability under strong noise conditions when using synthetic data. When using actual data, the model was deemed suitable for gas reservoir prediction in unexplored areas. Compared with conventional deep neural network and Bayesian neural network models, the proposed model better utilizes the characteristics of multicomponent seismic data to predict gas-bearing distributions with higher accuracy and lower uncertainty.

Multicomponent seismic data contain abundant reservoir information and have significant advantages for reservoir prediction. However, due to limited exploration in some areas, the available data samples are limited. Hence, it is important to develop new methods that can identify the complex relationship between oil-gas-bearing property and the multicomponent seismic response with limited samples. The least-squares support vector machine (LSSVM) model is suitable for solving small-sample and optimization problems. However, the parameter optimization of the LSSVM kernel function affects the prediction results. Therefore, a hybrid artificial intelligence model based on adaptive particle swarm optimization (APSO) and LSSVM was developed for multicomponent seismic reservoir prediction. Different kernel functions of the LSSVM model were analyzed to determine which one had the best predictive performance, and this initial model was used for optimization. The problem of the PSO falling into the local minimum was alleviated by adaptively adjusting the inertia weight and velocity coefficients. The kernel function parameters of the LSSVM model were globally searched via APSO, and the APSO-LSSVM model for reservoir distribution prediction was obtained. Finally, the gas-bearing probability distribution in the area with few wells was predicted using the multicomponent seismic data and the APSO-LSSVM model. The results showed that APSO had a faster convergence speed in parameter optimization than PSO. Additionally, the prediction results of the proposed hybrid artificial intelligence model (APSO-LSSVM) were more accurate than those of the LSSVM alone. This model achieved good prediction results with a small number of samples, providing a new method for gas reservoir prediction in areas with insufficient exploration degree.

Fully utilizing multicomponent seismic data to predict gas-bearing distributions has great potential, though it remains challenging. One difficulty in gas reservoir prediction using multicomponent seismic attribute data (MSAD) is the dimensional curse caused by excessive data. Additionally, the relationship between MSAD and gas reservoirs is complex and unclear; new methods are required to explore this relationship. Accordingly, this study proposed a multicomponent seismic gas-bearing distribution prediction method that combines unsupervised and deep learning. First, the dimensions of MSAD are reduced using principal component analysis (PCA) and fast independent component analysis, jointly called PICA, to highlight the sensitive characteristics of the gas reservoir and reduce redundant information. Subsequently, the attention mechanism (AM) is combined with a convolutional neural network (CNN) to construct the AMCNN. AM can adaptively allocate weights to each feature channel, improve the gas reservoir characteristic mining ability of the CNN model, and thus improve its gas-bearing prediction accuracy. Finally, the proposed method was applied to synthetic and real data to predict gas-bearing distributions. The results show that the developed model can effectively predict gas-bearing distributions. By comparing the PICA-optimized MSAD with that optimized only using PCA, PICA was found to better highlight the gas reservoir characteristics. Compared with the CNN and conventional machine learning models, the constructed AMCNN has a better prediction ability for gas-bearing distribution. This study provides a new approach and improves the accuracy of gas-bearing prediction using deep learning methods from data optimization and model construction.

Gas reservoir identification using seismic data has become a major focus of geophysical exploration. This study presents a gas reservoir identification and structural evaluation method using artificial neural networks (ANNs) and the Viterbi algorithm to improve processing efficiency and evaluate gas reservoir structural control. Initial identification was conducted using deep neural networks (DNNs). Composite seismic attributes sensitive to the multicomponent seismic response characteristics of gas reservoirs were obtained. Subsequently, a model expansion dataset and network hyperparameter optimization strategy were employed to assess the optimal DNN model for ReLU activation with nine hidden layers (3–5–7–7–7–9–9–11–11–11–1). The training model was run with the three composite attributes as input to predict the gas-bearing probability distribution. Considering the importance of evaluating geological structural characteristics, an automatic horizon tracking method using the Viterbi algorithm was proposed to evaluate the structural factors of gas reservoirs. Finally, the ANN-based gas reservoir identification results were comprehensively evaluated based on structural characteristics, thus, reducing the uncertainty, or multiple solutions, predicted by mathematical methods. This scheme was successfully applied to assess synthetic and real data, demonstrating the consistency between the predicted gas reservoir areas and the true situation. The effective implementation of this scheme improves processing efficiency and provides a new way to shorten the exploration cycle of a gas reservoir.