Search

Conventional galaxy mass estimation methods suffer from model assumptions and degeneracies. Machine learning (ML), which reduces the reliance on such assumptions, can be used to determine how well present-day observations can yield predictions for the distributions of stellar and dark matter. In this work, we use a general sample of galaxies from the TNG100 simulation to investigate the ability of multibranch convolutional neural network (CNN) based ML methods to predict the central (i.e. within 1−2 effective radii) stellar and total masses, and the stellar mass-to-light ratio (M */ L). These models take galaxy images and spatially resolved mean velocity and velocity dispersion maps as inputs. Such CNN-based models can, in general, break the degeneracy between baryonic and dark matter in the sense that the model can make reliable predictions on the individual contributions of each component. For example, with r -band images and two galaxy kinematic maps as inputs, our model predicting M */ L has a prediction uncertainty of 0.04 dex. Moreover, to investigate which (global) features significantly contribute to the correct predictions of the properties above, we utilize a gradient-boosting machine. We find that galaxy luminosity dominates the prediction of all masses in the central regions, with stellar velocity dispersion coming next. We also investigate the main contributing features when predicting stellar and dark matter mass fractions (f *, f DM) and the dark matter mass M DM, and discuss the underlying astrophysics. [ABSTRACT FROM AUTHOR]

SUMMARY: Due to the strong heterogeneity of the reservoir, it is difficult to extend the classical reservoir parameter prediction model to the new work area. Traditional geological methods rely on experience, while machine learning algorithms lack consideration of reservoir heterogeneity. This makes it impossible for these methods to achieve good predictive performance in new work areas. In this paper, we discuss designing a multilayer deep transfer learning network (MDTL). The network enables accurate prediction of reservoir parameters in new areas based on core and logging data from mature areas. For the architectural design, we establish a network structure that realizes data interaction in different work areas. For network parameter optimization, we design a multilayer maximum mean discrepancy loss function. It guides network training to learn feature knowledge about data in different work areas. For logging data processing, we explore a support vector machine method to remove abnormal noise in logging data. We apply MDTL to predict shale gas porosity and total organic carbon content in southern Sichuan. Compared to state-of-the-art reservoir parameter prediction models, MDTL achieves the best performance in new well areas. The network integrates the idea of multilayer transfer learning, reduces the influence of reservoir heterogeneity and effectively ensures the prediction accuracy. [ABSTRACT FROM AUTHOR]

Classification and evaluation of reservoirs conforming to geological characteristics are the proposed premises of the high‐quality development of unconsolidated sandstone heavy oil reservoirs. The P oilfield in Bohai Bay is one of the largest offshore oilfields in China, but the reservoirs are highly heterogeneous, have complex pore structures, and change fluid properties, which make the classification of reservoirs very difficult. At present, the most commonly used reservoir classification methods are based on the results of logging interpretation, which cannot eliminate the constraints of linear relationships such as empirical formulas. Machine learning has significant advantages for identifying reservoir classification problems that are complex and nonlinear with geological characteristics. First, this study fully considered the pore structure and fluid properties and selected four representative evaluation parameters. An improved K‐means method was used to establish a reservoir classification evaluation system. Characteristics, such as the sedimentary microfacies, pore structure and oilfield production performance, of different types of reservoirs were then analysed. Finally, six logging curves were selected as input results, reservoir classification was used as an output result, and XGBoost was used to carry out reservoir classification evaluation for the entire area. The accuracy of XGBoost was 85.35%. Furthermore, XGBoost was compared with three other machine learning methods to confirm its reliability. The results showed that the reservoirs in the study area can be divided into three types. Sandstone thickness and pore structure were the most important parameters that affected reservoir quality. Among various machine learning algorithms, XGBoost can accurately and quickly achieve a quantitative evaluation of reservoir quality. [ABSTRACT FROM AUTHOR]

Pore structure impacts the capability of seepage pattern of subsurface fluid mineral and mineral exploitation efficiency. Because of the strong heterogeneity in carbonate reservoirs, the pore structure is nonlinearly varying and complex in reservoirs. It is necessary to establish a method for pore structure type (PST) prediction. Machine learning provides an efficient prediction method by finding the relationship between core test data and well‐logging data. In this paper, a reservoir identification method based on Gradient Boost Decision Trees (GBDT) is proposed for the pore structure characteristics of carbonate reservoirs integrating core test and logging data. Core testing data are utilized to form the training set for PST prediction. Then, we adopt the mutual information method to optimize the logging data as the input of the machine learning model. GBDT hyperparameters are optimized using the learning curve to make it have optimal performance. Finally, we compare the predicted result of GBDT with K‐Nearest Neighbor and support vector machines, demonstrating the superior performance of GBDT. The results indicate that the four PS are different in some properties, such as permeability, pore radius and shape of pore and throat. 20%, 10 and 0.15 are recommended as the optimal parameter values for Maximum feature, iteration number and Sub‐sample number to improve GBDT accuracy in carbonate reservoir PST prediction. Evaluation indicators of machine learning, including the Kappa coefficient, Heming distance, Jaccard similarity coefficient and Confusion matrices, indicate that GBDT has better performance. This research proves the high accuracy and applicability of machine learning technology in the carbonate reservoir and also provides a meaningful reference for the identification of the carbonate reservoirs in the central area of Iraq. [ABSTRACT FROM AUTHOR]

Owing to a lengthy oil‐bearing interval, strong anisotropism, and significant difference in the fluid properties of the sandstone oil reservoir in P Oilfield, it is quite challenging to accurately the productivity of the oil well at the initial stage. In this study, a deep neural network model is established, based on a gradient boosting algorithm, XGBoost, to forecast the initial productivity of oil wells, followed by an evaluation of the main controlling factors of productivity. One hundred oil wells in the study area were divided into training and verification groups. With a specific productivity index of an oil well with a stable period of approximately 6 months at the initial production stage as the target data, and geological, engineering, and oil reservoir parameters as input data, hyper‐parameters for adjustment and optimization were selected, and a deep‐learning‐based unconsolidated sandstone productivity forecast model was established to forecast the initial productivity of oil wells in the target area. The mean square root error of the forecast result was <0.15, which is highly consistent with actual productivity. Finally, by adopting the XGBoost algorithm, the weight ranking of the controlling factors of productivity was clarified as follows: microscopic pore structure parameter > crude oil viscosity > median grain size > lithology index > well completion method > flow zone indicator. Machine learning has the advantages of effective forecasting of oil well productivity and the main controlling factors using multiple dimensions and big data. [ABSTRACT FROM AUTHOR]

Shale gas is a clean and low-carbon natural gas resource. It mainly exists in both adsorbed and free states in pores and fractures. To accurately estimate the in-situ adsorption gas content (AGC), which is helpful in resource evaluation and development planning, methane isothermal adsorption data and geological parameters have been collected, such as total organic carbon (TOC) content, thermal maturity (R o), siliceous mineral content (V QF), total clay content (V TC), water content (V WC), and temperature (T). Using machine learning (ML) methods, the in-situ AGC estimation models were constructed and optimized. Various geological factors affecting methane adsorption were evaluated, and an application was conducted in the Wufeng-Formation shale. The results reveal that the four ML models have higher accuracy in predicting Langmuir volume (V L) and Langmuir pressure (P L) than empirical formulas and linear regression models. Among the four ML models, the Random Forest Regression (RFR) and eXtreme Gradient Boosting Regression (XGBR) models perform the best, with R2 higher than 0.85. TOC and T are the main factors affecting methane adsorption, followed by R o and V QF , while the importance of V TC and V WC is relatively low. According to different combinations of geological parameters, there are three schemes for ML model construction. Among them, scheme 1 based on all six geological parameters has the highest accuracy and is most beneficial to predicting AGC. Gradually reducing V WC , V TC , and V QF results in a slight decrease in accuracy, with R2 decreasing by at most about 6%, scheme 2 is suitable for rougher estimation of AGC. Further removal of T and R o results in a significant decrease in accuracy, with R2 decreasing by up to 50% and MRE exceeding 30%, rendering scheme 3 unavailable for AGC prediction. The AGC of Wufeng-Longmaxi shale is successfully predicted based on XGBR model, with AGC mainly in 1.0 m3/ton-4.0 m3/ton. Overall, the ML models based on multiple geological parameters can simulate the real reservoir environment and achieve rapid and accurate estimation of in-situ AGC. • Construction of high accuracy shale adsorption gas content estimation machine learning models. • Revelation of key geological parameters affecting methane adsorption capacity. • Proposition of three prediction model construction schemes with different geological parameters combination. [ABSTRACT FROM AUTHOR]