The isolated fracture-vug systems controlled by small-scale strike-slip faults within ultra-deep carbonate rocks of the Tarim Basin exhibit significant exploration potential. The study employs a novel training set incorporating innovative fault labels to train a U-Net-structured CNN model, enabling effective identification of small-scale strike-slip faults through seismic data interpretation. Based on the CNN faults, we analyze the distribution patterns of small-scale strike-slip faults. The small-scale strike-slip faults can be categorized into NNW-trending and NE-trending groups with strike lengths ranging 200–5000 m. The development intensity of small-scale strike-slip faults in the Lower Yingshan Member notably exceeds that in the Upper Member. The Lower and Upper Yingshan members are two distinct mechanical layers with contrasting brittleness characteristics, separated by a low-brittleness layer. The superior brittleness of the Lower Yingshan Member enhances the development intensity of small-scale strike-slip faults compared to the upper member, while the low-brittleness layer exerts restrictive effects on vertical fault propagation. Fracture-vug systems formed by interactions of two or more small-scale strike-slip faults demonstrate larger sizes than those controlled by individual faults. All fracture-vug system sizes show positive correlations with the vertical extents of associated small-scale strike-slip faults, particularly intersection and approaching fracture-vug systems exhibit accelerated size increases proportional to the vertical extents.

Natural fractures controlled by faults in ultradeep carbonate strata play substantial roles as both fluid migration channels and storage spaces. However, characterizing the heterogeneous distribution of underground fractures within the complex three-dimensional geometry of strike-slip fault zones remains challenging. This study investigates the characteristics of natural fractures controlled by strike-slip faults in the fractured Middle and Lower Ordovician reservoirs of the central and northern Tarim Basin, China. Seismics, cores, and image logs were integrated to quantitatively analyze the intensity and dip angle of natural fractures and findings were verified using published sandbox simulations. The carbonate reservoir contains three main types of natural fractures: tectonic fractures, abnormal high-pressure-related fractures, and stylolites. Strike-slip faults control the distribution and characteristics of tectonic fractures across various scales. Generally, both fracture intensity and porosity exhibit a decreasing trend as the distance from the main fault surface increases. Compared with those in non-stepover zones along a strike-slip fault, natural fractures and faults in stepover zones are more developed along the fault strike, with significantly greater development intensity in central stepover regions than that at its two ends. Furthermore, strike-slip faults influence the dip angles of both natural fractures and secondary faults. The proportion of medium-to-low-dip angle fractures and faults in the stepover zone is greater than that in the non-stepover zone. Additionally, the proportion of medium-to low-dip angle fractures and faults in the middle of the stepover is greater than that at both ends. Therefore, strike-slip fault structures control the dip angle of natural fracture and the heterogeneity of secondary fault and fracture intensity. The linking damage zone in the stepover contains a larger volume of fractured rocks, making it a promising petroleum exploration target. The development of stepovers and the orientation of present-day in-situ stress substantially influence the productivity of fractured reservoirs controlled by strike-slip faults. The analysis in this study reveals that reservoir productivity increases as the angle between the strike-slip fault segment and the maximum horizontal principal stress decreases. This study provides valuable insights for quantitatively evaluating fracture heterogeneity in fractured reservoirs and establishing optimized selection criteria for favorable targets in fault-related fractured reservoirs.

Prolonged waterflooding leads to the development of waterflood-induced fractures in tight sandstone reservoirs. Clarifying the dynamic responses and evolutionary characteristics of these fractures holds great geological significance for the emplacement of dense well patterns and the tapping of residual oil potential of tight sandstone reservoirs. Integrating data from core observations, logs, oil production, pressure-buildup well tests, and water injection profiles, we explore the dynamic responses and distributions of waterflood-induced fractures across different development stages within the tight sandstone reservoirs in the 8^th member of the Yanchang Formation (also referred to as the Chang 8 Member) in well block L, Jiyuan oilfield, Ordos Basin. The results indicate that waterflood-induced fractures in the tight sandstone reservoirs of the Chang 8 Member within well block L originate from the propagation of natural fractures, and the natural fractures exhibit a preferential opening direction of NEE-SWW and NE-SW, followed by NW-SE. The water injection profiles of injection wells tend to exhibit small water absorption thickness but high water absorption capacity due to the formation of waterflood-induced fractures. Concurrently, the production performance curves of wells display a spurt or stepped upward trend in water cut, while pressure-buildup well tests reveal open double- logarithmic derivative curves that trend upward at a slope of 1/2. In the case where waterflood-induced fractures occur between production and injection wells, the production well test-interpreted formation pressure exceeds that in wells without waterflood-induced fractures and even far surpasses the initial formation pressure. In the initial development stage of tight sandstone reservoirs in the Chang 8 Member within well block L, waterflood-induced fractures in the reservoirs are primarily found in east-central, northeastern, and southeastern parts of the well block, where natural fractures are well developed. Waterflooding causes changes in the reservoir stress and thereby the opening pressure of natural fractures decreases. As a result, in the middle development stage, waterflood-induced fractures striking NW-SW come into being in the southern and north-central parts of well block L, accompanied by the small-scale propagation of pre-existing waterflood-induced fractures. In the late development stage, further waterflooding triggers the opening of natural fractures in different orientations around injection wells, leading to the formation of small-scale waterflood-induced fractures. This further exacerbates the fracture-induced waterlogging of production wells.

Identification of reservoir types in deep carbonates has always been a great challenge due to complex logging responses caused by the heterogeneous scale and distribution of storage spaces. Traditional cross-plot analysis and empirical formula methods for identifying reservoir types using geophysical logging data have high uncertainty and low efficiency, which cannot accurately reflect the nonlinear relationship between reservoir types and logging data. Recently, the kernel Fisher discriminant analysis (KFD), a kernel-based machine learning technique, attracts attention in many fields because of its strong nonlinear processing ability. However, the overall performance of KFD model may be limited as a single kernel function cannot simultaneously extrapolate and interpolate well, especially for highly complex data cases. To address this issue, in this study, a mixed kernel Fisher discriminant analysis (MKFD) model was established and applied to identify reservoir types of the deep Sinian carbonates in central Sichuan Basin, China. The MKFD model was trained and tested with 453 datasets from 7 coring wells, utilizing GR, CAL, DEN, AC, CNL and RT logs as input variables. The particle swarm optimization (PSO) was adopted for hyper-parameter optimization of MKFD model. To evaluate the model performance, prediction results of MKFD were compared with those of basic-kernel based KFD, RF and SVM models. Subsequently, the built MKFD model was applied in a blind well test, and a variable importance analysis was conducted. The comparison and blind test results demonstrated that MKFD outperformed traditional KFD, RF and SVM in the identification of reservoir types, which provided higher accuracy and stronger generalization. The MKFD can therefore be a reliable method for identifying reservoir types of deep carbonates.

How to fit a properly nonlinear classification model from conventional well logs to lithofacies is a key problem for machine learning methods. Kernel methods (e.g., KFD, SVM, MSVM) are effective attempts to solve this issue due to abilities of handling nonlinear features by kernel functions. Deep mining of log features indicating lithofacies still needs to be improved for kernel methods. Hence, this work employs deep neural networks to enhance the kernel principal component analysis (KPCA) method and proposes a deep kernel method (DKM) for lithofacies identification using well logs. DKM includes a feature extractor and a classifier. The feature extractor consists of a series of KPCA models arranged according to residual network structure. A gradient-free optimization method is introduced to automatically optimize parameters and structure in DKM, which can avoid complex tuning of parameters in models. To test the validation of the proposed DKM for lithofacies identification, an open-sourced dataset with seven conventional logs (GR, CAL, AC, DEN, CNL, LLD, and LLS) and lithofacies labels from the Daniudi Gas Field in China is used. There are eight lithofacies, namely clastic rocks (pebbly, coarse, medium, and fine sandstone, siltstone, mudstone), coal, and carbonate rocks. The comparisons between DKM and three commonly used kernel methods (KFD, SVM, MSVM) show that (1) DKM (85.7%) outperforms SVM (77%), KFD (79.5%), and MSVM (82.8%) in accuracy of lithofacies identification; (2) DKM is about twice faster than the multi-kernel method (MSVM) with good accuracy. The blind well test in Well D13 indicates that compared with the other three methods DKM improves about 24% in accuracy, 35% in precision, 41% in recall, and 40% in F1 score, respectively. In general, DKM is an effective method for complex lithofacies identification. This work also discussed the optimal structure and classifier for DKM. Experimental results show that (m₁,m₂,0) is the optimal model structure and linear SVM is the optimal classifier. (m₁,m₂,0) means there are m₁ KPCAs, and then m₂ residual units. A workflow to determine an optimal classifier in DKM for lithofacies identification is proposed, too.

Typically, relationship between well logs and lithofacies is complex, which leads to low accuracy of lithofacies identification. Machine learning (ML) methods are often applied to identify lithofacies using logs labelled by rock cores. However, these methods have accuracy limits to some extent. To further improve their accuracies, practical and novel ensemble learning strategy and principles are proposed in this work, which allows geologists not familiar with ML to establish a good ML lithofacies identification model and help geologists familiar with ML further improve accuracy of lithofacies identification. The ensemble learning strategy combines ML methods as sub-classifiers to generate a comprehensive lithofacies identification model, which aims to reduce the variance errors in prediction. Each sub-classifier is trained by randomly sampled labelled data with random features. The novelty of this work lies in the ensemble principles making sub-classifiers just overfitting by algorithm parameter setting and sub-dataset sampling. The principles can help reduce the bias errors in the prediction. Two issues are discussed, videlicet (1) whether only a relatively simple single-classifier method can be as sub-classifiers and how to select proper ML methods as sub-classifiers; (2) whether different kinds of ML methods can be combined as sub-classifiers. If yes, how to determine a proper combination. In order to test the effectiveness of the ensemble strategy and principles for lithofacies identification, different kinds of machine learning algorithms are selected as sub-classifiers, including regular classifiers (LDA, NB, KNN, ID3 tree and CART), kernel method (SVM), and ensemble learning algorithms (RF, AdaBoost, XGBoost and LightGBM). In this work, the experiments used a published dataset of lithofacies from Daniudi gas field (DGF) in Ordes Basin, China. Based on a series of comparisons between ML algorithms and their corresponding ensemble models using the ensemble strategy and principles, conclusions are drawn: (1) not only decision tree but also other single-classifiers and ensemble-learning-classifiers can be used as sub-classifiers of homogeneous ensemble learning and the ensemble can improve the accuracy of the original classifiers; (2) the ensemble principles for the introduced homogeneous and heterogeneous ensemble strategy are effective in promoting ML in lithofacies identification; (3) in practice, heterogeneous ensemble is more suitable for building a more powerful lithofacies identification model, though it is complex.