Search

Automatic waveform classification and arrival picking methods are widely studied to reduce or replace the manual works. Machine learning based methods, especially neural networks, and clustering based methods have shown great potentials in previous studies. However, most of the existing methods are sensitive to noise. The convolution neural networks (CNNs), developed from the traditional neural networks, have been successfully applied in many different fields, but are rarely studied in seismic waveform classification. In this paper, we propose a novel antinoise CNN architecture for waveform classification and also propose to combine k‐means clustering (KC) with CNN classification to pick arrivals (CNN‐KC). Seismic data are sampled to 1‐D vectors using a specific time window. Using the trained CNN classifier, these 1‐D vectors are classified into two categories: waveform and nonwaveform. With the constraint of the first waveform, CNN‐KC can pick the arrival more accurately. We also apply the proposed methods to the synthetic microseismic data with different noise levels and the actual field microseismic data to test their robustness. CNNs perform much better than the traditional multilayer perceptron on the waveform classification of the noisy microseismic data. Based on the analysis of the CNN internal architecture, we also conclude that the main reason that CNN is insensitive to noise is the convolution and pooling layers which behave like smooth operation in some ways. The final results show that the CNN and CNN‐KC are effective and robust methods for waveform classification and arrival picking. Key Points: We introduce the concept of waveform classification, by which we can automatically pick the arrivals and potentially reject noiseWe propose a novel CNN architecture for waveform classification and also combine k‐means clustering with CNN to pick arrivalsCNNs perform much better than the traditional multilayer perceptron (MLP) on the waveform classification of the noisy microseismic data [ABSTRACT FROM AUTHOR]

Seismic prediction has been a huge challenge because of the great uncertainties contained in the seismic data. Deep learning (DL) has been successfully applied in many fields and brought revolutionary changes, such as computer vision and natural language processing. The traditional artificial neural networks have been studied to improve the accuracy and resolution of seismic prediction for years, but not DL. In this paper, we develop a new architecture for seismic reservoir characterization based on the DL technique. We apply the convolutional neural network (CNN), which is a DL framework, to predict lithology and have achieved better results compared with traditional methods. We also propose to use continuous wavelet transforms (CWTs) to get a time–frequency spectrum for neural networks. CWTs help to make full use of the frequency content of the post-stack seismic data. According to the difference in the convolution layers and the organization of the input data, we propose four DL architectures for seismic lithology prediction, namely the deep neural networks (DNNs), the CNNs, the CWT-DNNs and the CWT-CNNs. All of these four architectures are applied in the case study. The final results on blind wells, profile and horizontal slice show that CWT-CNN models have the best performance on post-stack seismic lithology prediction. CWT maps contain more information about thinner layers and convolution layers are better at feature extraction from CWT maps. The CWT-CNN model has higher accuracy and resolution, especially on medium and thin layer prediction. [ABSTRACT FROM AUTHOR]

Explicit structural inference is one key point to improve the accuracy of scene parsing. Meanwhile, adversarial training method is able to reinforce spatial contiguity in output segmentations. To take both advantages of the structural learning and adversarial training simultaneously, we propose a novel deep learning network architecture called Structural Inference Embedded Adversarial Networks (SIEANs) for pixel-wise scene labeling. The generator of our SIEANs, a novel designed scene parsing network, makes full use of convolutional neural networks and long short-term memory networks to learn the global contextual information of objects in four different directions from RGB-(D) images, which is able to describe the (three-dimensional) spatial distributions of objects in a more comprehensive and accurate way. To further improve the performance, we explore the adversarial training method to optimize the generator along with a discriminator, which can not only detect and correct higher-order inconsistencies between the predicted segmentations and corresponding ground truths, but also exploit full advantages of the generator by fine-tuning its parameters so as to obtain higher consistencies. The experimental results demonstrate that our proposed SIEANs is able to achieve a better performance on PASCAL VOC 2012, SIFT FLOW, PASCAL Person-Part, Cityscapes, Stanford Background, NYUDv2, and SUN-RGBD datasets compared to the most of state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Deeply buried paleocaves exist widely in the world and act as an important type of hydrocarbon reservoirs. Three-dimensional (3-D) seismic data are commonly used to detect deeply buried paleocaves, but regular interpretation methods can hardly characterize their shape and uncertainty. We propose a novel seismic interpretation method based on Bayesian deep learning to solve this problem. The proposed Bayesian encoder–decoder network employs convolution layers, residual connections, especially dropout as the approximation of Bayesian inference. The Bayesian encoder–decoder network is trained on the synthetic 3-D seismic data and paleocave models as inputs and labels respectively. The synthetic paleocave models integrate geological knowledge such as the size and shape of paleocaves acquired from outcrops and drilled wells. The synthetic seismic data are generated based on the synthetic paleocaves and the wavelet extracted from the field seismic data. The trained Bayesian encoder–decoder model is finally tested on the blind synthetic data and the field seismic data from the Tarim Basin. Results show that Bayesian encoder–decoder can characterize the shape of paleocaves more accurately than the regular encoder–decoder and seismic attribute methods. Bayesian encoder–decoder also estimates uncertainty of the identified paleocaves which helps us evaluate the reliability of the result. Paleocave edges and small paleocaves of the results always show great uncertainty. The proposed approach based on Bayesian deep learning has great potential in similar scenarios of seismic reservoir characterization. • Bayesian encoder–decoder is proposed for seismic characterization of paleocaves with uncertainty quantification. • A synthetic dataset generation workflow is proposed for deep learning model training. • Deep learning models are trained on the synthetic data and predict satisfying results on the field data. • The uncertainty of prediction is estimated to help us evaluate the reliability of the result. [ABSTRACT FROM AUTHOR]

Social networks are ubiquitous in contemporary society, where one's opinions and behaviors are easily influenced by those around them. The effectiveness of social influence analysis has become significant for numerous application fields such as online recommendation, advertising, and viral marketing. Graph-based deep learning has achieved significant success in influence prediction. However, the existing methods ignore the privacy issue and are incapable of cross-organizational collaboration. Inspired by the recent success of federated learning in privacy protection and breaking the data barrier. In this paper, a novel federated learning framework, FedInf, is proposed to tackle the problem of social influence prediction. Specially, to further protect user privacy and achieve secure aggregation for multiple local models, we introduce differential privacy into the local parameters, adding artificial noise before model aggregation. In order to trade-off utility and privacy, we first freeze the embedding layers to reduce the number of upload parameters, and then project model parameters into low-dimensional space. Therefore, less noise is required to provide the same level of privacy protection. Extensive experiments on four real-world datasets demonstrate that the proposed FedInf is effective while providing privacy protection. [ABSTRACT FROM AUTHOR]