Artificial Intelligence in Geosciences最新文献

The polarity of first P-wave arrivals plays a significant role in the effective determination of focal mechanisms specially for smaller earthquakes. Manual estimation of polarities is not only time-consuming but also prone to human errors. This warrants a need for an automated algorithm for first motion polarity determination. We present a deep learning model - PolarCAP that uses an autoencoder architecture to identify first-motion polarities of earth-quake waveforms. PolarCAP is trained in a supervised fashion using more than 130,000 labelled traces from the Italian seismic dataset (INSTANCE) and is cross-validated on 22,000 traces to choose the most optimal set of hyperparameters. We obtain an accuracy of 0.98 on a completely unseen test dataset of almost 33,000 traces. Furthermore, we check the model generalizability by testing it on the datasets provided by previous works and show that our model achieves a higher recall on both positive and negative polarities.

Machine learning provides a powerful alternative data-driven approach to accomplish many petrophysical tasks from subsurface data. It can assimilate information from large and rich data bases and infer relations, rules, and knowledge hidden in the data. When the physics behind data becomes extremely complex, inexplicit, or even unclear/unknown, machine learning approaches have the advantage of being more flexible with wider applicability over conventional physics-based interpretation models. Moreover, machine learning can be utilized to assist many labor-intensive human interpretation tasks such as bad data identification, facies classification, and geo-features segmentation out of imagery data.

However, the validity of the outcome from machine learning largely depends on the quantity, quality, representativeness, and relevance of the feeding data including accurate labels. To achieve the best performance, it requires significant effort in data preparation, feature engineering, algorithm selection, architecture design hyperparameter tuning, and regularization. In addition, it needs to overcome technical issues such as imbalanced population, overfitting, and underfitting.

In this paper, advantages, limitations, and conditions of using machine learning to solve petrophysics challenges are discussed. The capability of machine learning algorithms in accomplishing different challenging tasks can only be achieved by overcoming its own limitations. Machine learning, if properly utilized, can become a powerful disruptive tool for assisting a series of critical petrophysics tasks.

Land suitability analysis (LSA) is an evaluation method that measures the degree to which land is suitable for certain land use. The primary aims of this study are to identify potentially viable agricultural land in the Gangarampur subdivision (West Bengal) using Multiple Criteria Decision Making (MCDM) and machine learning procedures and to evaluate the efficacy of the employed methodologies. The Analytic Hierarchy Process (AHP) model was used to assign relative weights to the fifteen various criteria in this suitability analysis, and then the Fuzzy Complex Proportional Assessment (FCOPRAS) model was applied using the AHP's normalised pairwise comparison matrix, whereas the Waikato Environment for Knowledge Analysis (Weka) Software was used to apply machine learning algorithms to the field data. The Random Forest (RF) model, on the other hand, is a better fit for the locational study of soil potential. According to the RF findings, areas of 14.67 per cent (15368.46 ha) are excellent (ZONE V) for growing crops, approximately 22.30 per cent (23367.9 ha) are highly suitable (ZONE IV), and 23.63 per cent (24762.12 ha) are moderately suitable (ZONE III) for cultivation, respectively. The numbers for FCOPRAS are roughly 15.39% (16130.52 ha), 22.54% (23620.65 ha), and 19.79% (20733.26 ha). The Receiver Operating Characteristic (ROC) curve and accuracy measurements of the results indicate the high accuracy of the applied models, with Random Forest and FCOPRAS being the most popular and effective techniques. This study will make an important contribution to evaluations of soil fertility and site suitability. This will help local government officials, academics, and farmers scientifically use the land.

The Xingmeng orogenic belt is located in the eastern section of the Central Asian orogenic belt, which is one of the key areas to study the formation and evolution of the Central Asian orogenic belt. At present, there is a huge controversy over the closure time of the Paleo-Asian Ocean in the Xingmeng orogenic belt. One of the reasons is that the genetic tectonic setting of the Carboniferous volcanic rocks is not clear. Due to the diversity of volcanic rock geochemical characteristics and its related interpretations, there are two different views on the tectonic setting of Carboniferous volcanic rocks in the Xingmeng orogenic belt: island arc and continental rift. In recent years, it is one of the important development directions in the application of geological big data technology to analyze geochemical data based on machine learning methods and further infer the tectonic background of basalt. This paper systematically collects Carboniferous basic rock data from Dongwuqi area of Inner Mongolia, Keyouzhongqi area of Inner Mongolia and Beishan area in the southern section of the Central Asian Orogenic Belt. Random forest algorithm is used for training sets of major elements and trace elements in global island arc basalt and rift basalt, and then the trained model is used to predict the tectonic setting of the Carboniferous magmatic rock samples in the Xingmeng orogenic belt. The prediction results shows that the island arc probability of most of the research samples is between 0.65 and 1, which indicates that the island arc tectonic setting is more credible. In this paper, it is concluded that magmatism in the Beishan area of the southern part of the Central Asian Orogenic belt in the Early Carboniferous may have formed in the heyday of subduction, while the Xingmeng orogenic belt in the Late Carboniferous may have been in the late subduction stage to the collision or even the early rifting stage. This temporal and spatial evolution shows that the subduction of the Paleo-Asian Ocean is different from west to east. Therefore, the research results of this paper show that the subduction of the Xingmeng orogenic belt in the Carboniferous has not ended yet.

Seismic data interpolation, especially irregularly sampled data interpolation, is a critical task for seismic processing and subsequent interpretation. Recently, with the development of machine learning and deep learning, convolutional neural networks (CNNs) are applied for interpolating irregularly sampled seismic data. CNN based approaches can address the apparent defects of traditional interpolation methods, such as the low computational efficiency and the difficulty on parameters selection. However, current CNN based methods only consider the temporal and spatial features of irregularly sampled seismic data, which fail to consider the frequency features of seismic data, i.e., the multi-scale features. To overcome these drawbacks, we propose a wavelet-based convolutional block attention deep learning (W-CBADL) network for irregularly sampled seismic data reconstruction. We firstly introduce the discrete wavelet transform (DWT) and the inverse wavelet transform (IWT) to the commonly used U-Net by considering the multi-scale features of irregularly sampled seismic data. Moreover, we propose to adopt the convolutional block attention module (CBAM) to precisely restore sampled seismic traces, which could apply the attention to both channel and spatial dimensions. Finally, we adopt the proposed W-CBADL model to synthetic and pre-stack field data to evaluate its validity and effectiveness. The results demonstrate that the proposed W-CBADL model could reconstruct irregularly sampled seismic data more effectively and more efficiently than the state-of-the-art contrastive CNN based models.

Machine learning (ML) to predict lithofacies from sparse suites of well-log data is difficult in laterally and vertically heterogeneous reservoir formations in oil and gas fields. Meandering, braided fluviatile depositional environments tend to form clastic sequences with laterally discontinuous layers due to the continuous shifting of relatively narrow sandstone channels. Three cored wellbores drilled through such a reservoir in a large oil field, with just four recorded well logs available, are used to classify four lithofacies using ML models. To augment the well-log data, six derivative and volatility attributes were calculated from the recorded gamma ray and density logs, providing sixteen log features for the ML models to select from. A novel, multiple-optimizer feature selection technique was developed to identify high-performing feature combinations with which seven ML models were used to predict lithofacies assisted by multi-k-fold cross validation. Feature combinations with just seven to nine selected log features achieved overall ML lithofacies accuracy of 0.87 for two wells used for training and validation. When the trained ML models were applied to a third well for testing, lithofacies ML prediction accuracy declined to 0.65 for the best performing extreme gradient boosting model with seven features. However, an accuracy of ∼0.76 was achieved by that model in predicting the presence of the pay bearing sandstone and siltstone lithofacies in the test well. A model using only the four recorded well logs was only able to predict the pay-bearing lithofacies with ∼0.6 accuracy. Annotated confusion matrices and feature importance analysis provide additional insight to ML model performance and identify the log attributes that are most influential in enhancing lithofacies prediction.

Reliable seismic phase identification is often challenging especially in the circumstances of low-magnitude events or poor signal-to-noise ratio. With improved seismometers and better global coverage, a sharp increase in the volume of recorded seismic data has been achieved. This makes handling seismic data rather daunting by using traditional approaches and therefore fuels the need for more robust and reliable methods. In this study, we develop 1D deep Residual Neural Network (ResNet), for tackling the problem of seismic signal detection and phase identification. This method is trained and tested on the dataset recorded by the Southern California Seismic Network. Results demonstrate that the proposed method can achieve robust performance for the detection of seismic signals and the identification of seismic phases. Compared to previously proposed deep learning methods, the introduced framework achieves around 4% improvement in earthquake detection and a slightly better performance in seismic phase identification on the dataset recorded by Southern California Earthquake Data Center. The model generalizability is also tested further on the STanford EArthquake Dataset. In addition, the experimental result on the same subset of the STanford EArthquake Dataset, when masked by different noise levels, demonstrates the model’s robustness in identifying the seismic phases of small magnitude.

Estimation of good velocity models under complex near-surface conditions remains a topic of ongoing research. We propose to predict near-surface velocity profiles from surface-waves transformed to phase velocity-frequency panels in a data-driven manner using deep neural networks. This is a different approach from many recent works that attempt to estimate velocity from directly reflected body waves or guided waves. A secondary objective is to analyze the influence on the prediction accuracy of various commonly employed deep learning practices, such as transfer learning and data augmentations. Through numerical experiments on synthetic data as well as a real geophysical example, we demonstrate that transfer learning as well as data augmentations are helpful when using deep learning for velocity estimation. A third and final objective is to study lack of generalization of deep learning models for out-of-distribution (OOD) data in the context of our problem, and present a novel approach to tackle it. We propose a domain adaptation network for training deep learning models that uses a priori knowledge on the range of velocity values in order to constrain mapping of the output. The final comparison on field data, which was not part of the training data, show the deep neural network predictions compare favorably with a conventional velocity model estimation obtained with a dispersion curve inversion workflow.

Among the biggest challenges we face in utilizing neural networks trained on waveform (i.e., seismic, electromagnetic, or ultrasound) data is its application to real data. The requirement for accurate labels often forces us to train our networks using synthetic data, where labels are readily available. However, synthetic data often fail to capture the reality of the field/real experiment, and we end up with poor performance of the trained neural networks (NNs) at the inference stage. This is because synthetic data lack many of the realistic features embedded in real data, including an accurate waveform source signature, realistic noise, and accurate reflectivity. In other words, the real data set is far from being a sample from the distribution of the synthetic training set. Thus, we describe a novel approach to enhance our supervised neural network (NN) training on synthetic data with real data features (domain adaptation). Specifically, for tasks in which the absolute values of the vertical axis (time or depth) of the input section are not crucial to the prediction, like classification, or can be corrected after the prediction, like velocity model building using a well, we suggest a series of linear operations on the input to the network data so that the training and application data have similar distributions. This is accomplished by applying two operations on the input data to the NN, whether the input is from the synthetic or real data subset domain: (1) The crosscorrelation of the input data section (i.e., shot gather, seismic image, etc.) with a fixed-location reference trace from the input data section. (2) The convolution of the resulting data with the mean (or a random sample) of the autocorrelated sections from the other subset domain. In the training stage, the input data are from the synthetic subset domain and the auto-corrected (we crosscorrelate each trace with itself) sections are from the real subset domain, and the random selection of sections from the real data is implemented at every epoch of the training. In the inference/application stage, the input data are from the real subset domain and the mean of the autocorrelated sections are from the synthetic data subset domain. Example applications on passive seismic data for microseismic event source location determination and on active seismic data for predicting low frequencies are used to demonstrate the power of this approach in improving the applicability of our trained NNs to real data.