Skip to Content
You are currently on the new version of our website. Access the old version .
MDPIMDPI

Advanced

Applied SciencesApplied Sciences
PDF
  • Article
  • Open Access

18 December 2025

22 Pages

Fracture Prediction Based on a Complex Lithology Fracture Facies Model: A Case Study from the Linxing Area, Ordos Basin

,
,
,
,
,
and
1
College of Resources and Environment, Yangtze University, Wuhan 430100, China
2
Jiangsu Oilfield, Yangzhou 225009, China
*
Author to whom correspondence should be addressed.
Appl. Sci.2025, 15(24), 13277;https://doi.org/10.3390/app152413277
This article belongs to the Special Issue Technologies and Methods for Exploitation of Geological Resources, 2nd Edition
Version Notes
Order Reprints

Abstract

In the Ordos Basin, the lengths of cores are disproportionate to image logging data (1:9) and fracture research is difficult because of their complex lithology and fracture patterns. Based on the characteristics of conventional logging and cores, this paper describes the color, shape, geophysical characteristics and geological features of the basin to establish an image recognition template and to identify nine distinct lithologies. The genesis, type, occurrence, opening mode, cutting depth, host lithology, density and tectonic stress of the fractures are used to define four types of fracture facies (bedding fracture facies, N100° tectonic fracture facies, N10° tectonic fracture facies and coal fracture facies) and to build four models. The comprehensive coherence among the neural network results, curvatures, ant bodies, lithologies, and thicknesses was used to predict the type of different fracture facies. The results show that the fracture prediction model fully reflects the genesis of the cracks and influencing factors and provides insights into optimal areas for future exploration and development.

1. Introduction

With the growing demand for oil and gas resources and the increasing challenges and difficulty associated with the exploration and development of conventional oil and gas resources, the proportion of the supply composed of unconventional oil and gas resources, such as coalbed methane, tight sandstone oil and gas and shale oil and gas, are accounting for a steadily rising share. Unconventional gas exploration and the development of coal measure strata have attracted increasing attention from domestic and foreign scholars [1,2,3,4,5,6]. The Linxing area is in the central part of the Hedong coal field, which is located in the eastern part of the Ordos Basin. Multiple wells have been drilled into a tight sandstone gas reservoir in the Permian coal-bearing strata, and reservoir tests in the Taiyuan Formation and the Lower Shihezi Formation have demonstrated the presence of industrial gas [7]. For tight sandstone reservoirs, the development of fractures has greatly affected reservoir development and deployment programs and results. Relevant research is not currently available on the fractures in this area. Therefore, strengthening fracture research can contribute to productivity breakthroughs and effectively reduce the exploration and development costs while oil prices are low.
The Linxing area features complex lithology. Following the methodology of “composition–structure–genesis”, the lithology is divided into nine categories: mudstone, silty mudstone, carbonaceous mudstone, muddy siltstone, siltstone, fine sandstone, medium sandstone, coarse sandstone and coal. Core observations show that significant differences occur in the developmental characteristics of these lithologies. In view of this situation, this study uses numerous image logging data to accurately identify lithologies from image logs. In this study, abundant imaging logging data are used to carry out imaging–lithology fine identification. By establishing corresponding identification templates, detailed fracture development characteristics of different lithologies are obtained. Finally, a fracture prediction model that captures the genesis and development patterns of fractures is established.

2. Study Area and Data

2.1. Study Area

The study area is located within the northern Linxing area, at the western margin of the Jinxi flexural fold belt and the eastern Yishan slope in the northeastern Ordos Basin. (Figure 1). The target layer is the Carboniferous Taiyuan Formation, as well as the Permian Shanxi, Upper Shihezi and Lower Shihezi formations (Figure 2). The Taiyuan Formation and the Shanxi Formation were deposited in delta plain sedimentary environments, which hosted coal and carbonaceous mudstone that are now the main gas-source rocks in the study area. The Shihezi Formation was deposited in a fluvial sedimentary environment, and it serves as the main reservoir. The porosity of the reservoir ranges from 3.0% to 10.0%, with an average value of 6.07%. The permeability is between 0.001 mD and 0.4 mD, with an average value of 0.093 mD. Therefore, the reservoir is classified as a tight sandstone reservoir. At present, only two wells in the study area have high natural productivity. Most wells produce industrial gas through fracturing, although the fracturing results are different for each well, with certain wells showing no fracturing effect. This observation, as made from cores, shows that natural fractures are not well-developed in the wells; therefore, the wells only show high productivity after fracturing. Therefore, research on fractures in this area can provide new insights for gas exploration.
Figure 1. Location map of the research area in the Linxing area of the Ordos Basin.
Figure 2. Lithologic histogram of the study area.

2.2. Data and Method

There are 20 wells in this area (Figure 1). The core length is 218 m. The average coring length of each well is 10 m. The observed cores reveal a complex lithology consisting of nine distinct rock types: mudstone, silty mudstone, carbonaceous mudstone, muddy siltstone, siltstone, fine sandstone, medium sandstone, coarse sandstone, and coal. Based on the varying compositions and textures of these rocks, a complex pattern of fracture development is observed. The fracture characteristics of different lithologies are significantly different, and they need to be counted separately to carry out fracture prediction research. The statistics of fracture parameters often need a lot of data as the basis, but from coring data, the coring length of some lithologies is relatively short, such as coarse sandstone with only 8 m (7 fractures observed in the core), carbonaceous mudstone with only 5 m (12 fractures observed in the core), and coal with only 3 m (15 fractures observed in the core), which makes it difficult to obtain the statistical characteristics of fractures.
Fortunately, all 20 wells in the study area have imaging logging data, with a total length of 1790 m, averaging 89 m per well. As we all know, imaging logs provide an effective means to investigate fractures. Compared with the core, its data are usually more continuous and abundant. If we can analyze the response characteristics of different lithologies on imaging logging, and establish an image–lithology identification template, by using imaging logging to identify fractures and lithology at the same time, we can obtain more fracture parameters’ characteristics of different lithologies. This approach addresses the limitation of insufficient core length for certain lithologies, thereby facilitating statistical analysis and subsequent research.
The following are the research methods and steps:
(i)
By comparing core samples with well logs, we analyzed the image log characteristics of nine lithologies and created a lithology identification template. Applying this template to the 1790 m of image log data from the study area allowed us to identify all nine lithology types.
(ii)
By identifying fractures on imaging logging and combining with the nine types of lithologies in (i), we can obtain the fracture parameters of different lithologies, such as fracture occurrence, density and so on (Table 1). Through comprehensive analysis and classification of these parameters, four fracture-phase classification schemes reflecting the causes and development characteristics of fractures are obtained.
Table 1. Statistical table of fracture characteristics in the study area.
(iii)
We sought to establish a geological model containing nine lithologies and transform the 3D lithologic model into the 3D model of fracture facies by the classification of fracture facies. At the same time, the fracture density parameters of different fracture facies can be obtained by a seismic multi-parameter neural network algorithm. Finally, a three-dimensional fracture distribution model can be obtained. The model can reflect the fracture density and parameter characteristics of different fracture facies in 3D space.

3. Establishment of an Imaging—Lithology Identification Template

The features of imaging logs are mainly identified by color changes and geometry; they can be used in lithology, sedimentary structure, paleocurrent, structural dip analysis, in situ stress analysis and fracture research [8,9,10,11,12,13]. Because imaging logging uses color to reflect resistivity value [14], the most common method for identifying lithologies based on image logging data is a color change in the image and the use of core and conventional logs for qualitative identification [15,16]. Some scholars have used it to identify lithofacies assemblages and sedimentary trends, and achieved some results [9,17,18]. Li Chaoliu quantitatively transformed imaging logging color-code values to lithologic particle sizes [19] and divided the lithology of clastic rocks into the following three types based on their particle size: mud, silt and sand. This method of lithologic calculation has quantitative characteristics but presents limitations for distinguishing lithological types. Therefore, for complex lithological areas, such as those with carbonaceous mudstone, coal and other special lithologies, this method is not applicable. Lai Jin classifies the image into nine categories according to the texture combination features of the image [20]. Wang Man proposed statistical methods for analyzing image textures to identify igneous lithologies and obtained positive results [21]. Some scholars have applied it to identify carbonate rocks [16,22]. BinAbadat recognized not only stromatolites but also the growth pattern of stromatolites in imaging logging according to their morphological characteristics [23]. Zhou Zhenglong proposed imaging identification methods that identified five types of sedimentary facies in the Ordos Basin [18]. In retrospect of these studies, few comprehensive applications of imaging logging for lithologic classification and few examples are used to identify complex lithologies. You Zheng [24] used morphological classification as the main identifier and geological significance as the foundation to describe the image color, morphology, geophysical characteristics and geological features of an area (Color: According to the color of the image, it can be divided into bright, light, dark and miscellaneous; Form: According to the shape of the image, it can be divided into blocks, strips, lines, grooves, spots and disorders; Geophysical characteristics: according to the geophysical information of the image, such as high-resistivity layer, low-resistivity layer and inhomogeneous layer; Geological characteristics: According to the geological information contained in the imaging, it can be classified into grain size, layer, scour surface, bedding, fault and hole. In the process of expression, it is not necessary to describe all the features but instead focus on describing typical features). This comprehensive classification scheme proposed by Youzheng is the scheme adopted in this paper.
Nine types of lithologies were identified according to the cores from the study area. Certain lithologies are similar in their imaged morphological characteristics. For example, mudstone and silty mudstones with horizontal bedding are shown as having banding patterns [16,25,26] and alternating displays of high resistivity and low resistivity in dark and light colors [16], but silty mudstones or siltstones have thicker, more continuous sinusoidal curves and lower GR (Natural Gamma Ray Log Curve) values [22]. To improve the accuracy of lithology identification, this paper uses a combination of conventional logging and considers the morphology, color, geophysical characteristics and geologic genesis of the image logs to identify and classify a lithological model according to the “color—morphology—geophysical characteristics—geological significance” scheme. The lithologies in the study area can be divided into two main types. The first type is mainly distinguished by the geological genesis and color change, and the particle sizes of seven lithology types (mudstone, silty mudstone, muddy siltstone, siltstone, fine sandstone, medium sandstone and coarse sandstone) gradually change from fine to coarse; the imaging features are also characterized by a gradual increase in granular spots, and the GR curve values gradually decrease. The second lithology type is mainly distinguished by its color and image morphology and is used to identify the coal and carbonaceous mudstone lithologies, which have a large number of carbonaceous chips, and their image morphologies are mainly a porphyritic pattern. The characteristics of coal are obvious: it has a low density and has an image morphology with a massive pattern. Detailed characteristics of the lithologies are presented as follows (Table 2).
Table 2. Identification of complex lithologies in the Linxing area of the Ordos Basin based on imaging.
(1)
Mudstone: Mudstone is usually formed in a static water environment. The horizontal bedding is well-developed, the resistivity is generally low, and the GR values are generally high. The horizontal bedding in the image shows better layering and has good correspondence with the core.
(2)
Muddy siltstone and silty mudstone: These two lithologies are composed of silt and clay in different proportions and their sedimentary environment and sediment particle sizes are similar; therefore, effectively identifying them based on the core alone is difficult. However, the differences in clay minerals in the image logs are easier to identify. For example, silty mudstone contains more mud, and dark bands are more common than bright bands. Alternatively, muddy siltstone contains more sand, and bright bands are slightly more common than dark bands. Because the depositional environments of the two lithologies are similar, with both deposited in water with low energy, the main sedimentary structure that develops is horizontal bedding. Therefore, layered stripes are common throughout the image logs. The GR value of muddy siltstone is slightly lower than that of silty mudstone; thus, these two lithologies can be further distinguished by the GR values on conventional logs.
(3)
Sandstone (siltstone, fine sandstone, medium sandstone and coarse sandstone): The GR values of these four lithologies decrease in sequence, and spots of different colors are observed on the image logs. Because the particle sizes increase from fine to coarse, the colors of the spots change from dark to bright, with coarse sandstone the brightest and siltstone the darkest. Sandstones of different particle sizes are more easily distinguished via their color characteristics. Finer rock particles, including fine sandstone, tend to form abundant horizontal bedding, while coarse sandstone typically forms cross-bedding.
(4)
Carbonaceous mudstone: the organic carbon content of carbonaceous mudstone is generally 10–30%, and the organic carbon content in coal-bearing strata is between that of mudstone and coal. The GR value is higher than that of mudstone, and the density is lower than that of mudstone. Because of the carbon content, a large number of black spots can be found on the image logs. These spots are obvious, and the resistivity is low.
(5)
Coal: Coal occurs in the Shanxi Formation and Taiyuan Formation. The values of GR, density, interval transit time, neutron, and resistivity are obviously low; therefore, coal is easy to identify using conventional logs. The resistivity of coal in this area is mainly high, and it is displayed as obvious bright white bands on the image logs and is significantly different from surrounding rocks with dark horizontal bedding. A thin, dark massive interlayer and fractures are all developed.
By comparing the imaging characteristics (color, morphology, geophysical characteristics and geologic features) of the different lithologies, we combine the observations with conventional logs to reduce the possibility of multisolutions and improve the accuracy of lithology identification. Using this process, we established an imaging identification model of the nine lithologies in the Linxing area (Table 2).

4. Fracture Identification and Fracture Facies Classification

Natural fracture, as an important flow channel of fluid, will affect the permeability and fluidity of the reservoir [11,27,28,29]. Fractures not only play an important role in oil and gas accumulation [30] but also play an important role in oil and gas production [31]. Imaging logging can be used to interpret the position and state of fractures, such as type and direction, and fracture parameters such as fracture pore and aperture [8,15,27,32].

4.1. Identification of Fractures

Core observations of fractures represent the most important and effective method of studying fractures underground and are the basis for fracture logging and computer simulations. This method also provides a direct test of the accuracy of fracture logging and computer simulations. The characteristics of fractures, such as spacing, density, occurrence, filling and mechanical properties, can be directly obtained from cores. Fracture openings are obtained from imaging logging. Different types of tectonic fractures and bedding fractures develop in different lithologies of the target layer; however, because of the limited amount of core data and the short core lengths through the target layer, accurately classifying the fracture characteristics is difficult. To use the abundantly available image logs to obtain the fracture characteristics from different lithologies, a method of identifying fractures from the image logs is required.
The fractures in the study area are mainly classified as high-angle tectonic fractures, which are formed when the rock has been subjected to stress that is greater than its strength. The curve on image logs is characterized by the high amplitude of the sine curve (Figure 3a). Bedding fractures are formed when the layer is split along sedimentary bedding, and they usually form in groups; the bedding fracture usually develops in the fine-grained lithology (mudstone, muddy siltstone and silty mudstone) with more argillaceous horizontal bedding. We can know the characteristics of this kind of lithology from the imaging map in Table 1. At the same time, the bedding fracture is mostly low-angle. From these two aspects, we can identify the bedding fracture (Figure 3b). Coal fractures form when the stress on the coal is greater than its strength. Because of the brittleness of coal, a high density of fractures is observed; therefore, the main characteristics that are displayed on the image logs are messy dark lines (Figure 3c).
Figure 3. Identification of fractures on the imaging logs of the Linxing area of the Ordos Basin: (a) high-angle tectonic fracture. (b) Low-angle tectonic fracture. (c) Coal seam fractures.

4.2. Classification of Fracture Facies

The fracture facies represent a combination of fracture systems that control the fluid flow within the reservoir, and they are commonly described by a combination of fracture occurrence, length, density and opening [33]. Domestic scholars have performed relevant research [34] in the Kuqa Depression of the Dabei area. The fracture facies can be divided into four types: netted fractures, high-angle cross fractures, low-angle cross fractures, and nearly horizontal fractures. However, this classification method only considers the dip angle, and the evidence for classification is relatively simple and insufficient for predicting fracture development and distribution. This paper establishes a fracture facies classification method that reflects the genesis and distribution of fractures based on their regional tectonic background and complex lithologic characteristics using parameters such as their occurrence, opening, cutting depth and density.
(1) Genesis and occurrence of fractures
Since the Paleozoic era, the Ordos Basin has undergone multiple tectonic events, each generating a distinct tectonic stress field. The superposition of these varied stresses through time has ultimately produced the modern tectonic framework [35]. By analyzing the occurrence of faults in the study area, two groups of faults with strikes of N53° and N159° are mainly developed, and they are oriented to the northeast and northwest (Figure 4a,b). In addition, the structural fractures in the area also mainly develop two groups of N10° and N100°, and they are oriented in approximately north–south and east–west directions (Figure 4c). The research method is to inject water-soluble substances into a well, take water samples from wells around the injection well, and analyze the concentration of water-soluble substances in the water samples. It is found that the fractures near the fault are not all parallel to the fault, and the angles between the fractures and the fault mostly range from 30° to 60° [36]. Zhang Yunfeng carried out a physical simulation experiment on the formation of associated fractures near faults. The associated fractures mainly developed on both sides of the fault plane, and the angle between these associated fractures and faults was approximately 45° [37]. The strikes of the main two groups of structural fractures in this area are 43° and 59, respectively, with the strikes of the two groups of faults (Figure 4b,c). These results are consistent with previous research, in that the development of fractures in the study area is directly controlled by the development of faults, and the strike characteristics reflect its tectonic genesis.
Figure 4. Characteristics of the faults and fractures in the study area: (a) plane distribution of faults. (b) Occurrence of faults. (c) Occurrence of fractures. (d) Average dip angle of tectonic fractures. (e) Average dip angle of coal fractures. (f) Occurrence of coal fractures.
Based on their dip angles, the fractures have been divided into four types [38]: horizontal fractures (0–15°), low-angle cross fractures (15–45°), high-angle cross fractures (45–75°) and vertical fractures (75–90°). Based on the statistical results, fractures in the target layer dip at 65°, and most fractures are high-angle cross fractures. Among them, the average dip angle of fractures in the Upper Shihezi Formation is 61°, the average dip angle in the Lower Shihezi Formation is 64°, and the average dip angle in the Shanxi Formation is 68°. Similarly, the average dip angle is 71° in the Taiyuan Formation (Figure 4d).
Generally, the dip angles of coal fractures are higher (Figure 4e), and their orientations are more chaotic compared with other fractures (Figure 4f). Therefore, coal fractures are analyzed separately in the classification of the fracture facies and the establishment of the fracture facies model.
(2) Opening of fractures
Many classification modes are available at the fracture scale [38]. The opening, density and cutting depth of fractures can be observed directly in cores. According to the statistical results, the opening of bedding fractures is below 0.1 mm, which is smaller than that of tectonic fractures and belongs to the microfracture category (Figure 5a). Certain bedding fractures are closed in the study area and must be opened by external forces (hydraulic fracturing). Most openings of tectonic fractures are large. The distribution of fracture openings is 53.6% in the 1 to 10 mm range, 39.3% in the 0.1 to 1 mm range, and 7.1% in the 0.1 mm range (Figure 5b). Therefore, according to the opening, the tectonic fractures were divided into two types: medium fractures and small fractures.
Figure 5. Statistical results of opening and cutting depth of the bedding fractures and tectonic fractures: (a) opening interval of bedding fractures (mm). (b) Opening interval of tectonic fractures (mm). (c) Cutting depth interval of tectonic fractures (cm).
(3) Cutting depth and density of fractures
The longest observed cutting depth (the length of the longitudinal extension of the fracture) of a tectonic fracture was 1 m (Figure 5c). The fracture lengths from image logs mostly range from 10 to 100 cm, which is consistent with the core observations and reflects medium and small tectonic fractures.
According to the tectonic stress, the statistical results for the cutting depth characteristics, which were divided by the strike, showed a good correlation between the fracture density and cutting depth. The fracture cutting depth is relatively small when the density is high, and vice versa (Figure 6). There are two main parameters for calculating fracture density: sampling interval and window length. Because the fracture prediction in this paper is based on seismic multiattribute calculation, in order to compare the calculated density with the predicted fracture density, the parameters used here are a sampling interval of 1 m (determining the number of data points of the curve) and window length of 6 m (macroscopic variation trend of the curve). Domestic researchers have performed simulation experiments and found that the strain energy released by the fracture offsets the surface energy needed to increase the fracture surface area [37]. The results of this study verify these findings. When two fractures undergo the same strain, the surface area of shallow fractures that develop in a high density is similar to that of deep fractures that develop in a low density.
Figure 6. Density, opening and cutting depth characteristics of fractures: (a) density and opening characteristics of bedding fracture facies. (b) Density, opening and cutting depth characteristics of the N10° tectonic fracture facies. (c) Density, opening and cutting depth characteristics of the N100° tectonic fracture facies.
(4) Classification of fracture facies
Through the research described above, we consider five factors: occurrence, density, opening, cutting depth and lithology of fractures. The occurrence and lithology reflect the genesis of the fractures; therefore, the fracture facies in the study area could be divided into four types: bedding fracture facies, N100° tectonic fracture facies, N10° tectonic fracture facies and coal fracture facies. According to Figure 7, in lithological sections where fractures occur in high density, because fracture development causes serious damage to rocks, the fracture has a larger opening. Because of the accumulated tectonic strain that was released by the surface area through the fractures, the cutting depth was usually shallow where the fractures were well-developed (Table 3). Therefore, based on the development level and fracture genesis classification, the fractures were divided into three types: well-developed fracture facies, poorly developed fracture facies and undeveloped fracture facies. These classifications are used for the fracture facies modeling presented later in the document (Figure 7).
Figure 7. Classification of fracture facies.
Table 3. Development characteristics of the bedding fractures, N10° tectonic fracture facies and N100° tectonic fracture facies.

5. Predictive Model of Fracture Facies

5.1. Establishing a Fracture Facies Prediction Model Combined with a Geological Model

Lithology is the main factor that influences the occurrence of fractures. To establish a fracture facies prediction model that reflects the developmental characteristics of the 4 fracture facies types described above, a reasonable geological model must be established to reflect the complex distribution of lithologies in the study area. Different modeling methods have been developed by domestic scholars [39]. A reliable geologic model was difficult to obtain with only two types of lithologies (sandstone and mudstone) in the simulation. Accordingly, the trending constraint was suitable for representing the complex lithology of the study area. The modeling results were most ideal when the seismic wave impedance could effectively distinguish the different lithologies. However, the high density and hardness of the dense sandstone reservoir resulted in a small difference between the geophysical characteristics [40,41,42,43,44,45]; therefore, effectively predicting the lithology is difficult. Specifically, the lithology of the study area was complex, and almost all longitudinal wave impedances overlapped with one other, which led to difficulties in refining the research data.
To reasonably characterize the nine lithologies in three-dimensional space and reduce the uncertainty in the predicted results, this paper used a combination of geological modeling and seismic forward modeling. Scholars have performed related research and generated good results [39]. The variogram function can provide spatial structure information of lithology. Firstly, nine variograms of different lithology are counted out (Table 4). The lithology model is established by a sequential indicator simulation algorithm. Then, the facies-controlled modeling method is adopted, that is, under the control of the lithology model, sequential Gaussian simulation algorithm is used to model wave impedance. Sequential indicator simulation algorithm is the most widely used for discrete data, and sequential Gauss algorithm is the most widely used algorithm for continuous property [46]. Assuming that the actual underground wave impedance is known (the reflection coefficient can be calculated), the seismic data (a) can be obtained by convoluting the wavelet with the corresponding reflection coefficient. The wave impedance is simulated by stochastic simulation technology, and the reflection coefficient model is obtained. The seismic data (b) can be obtained by convolution with the same wavelet. The closer the stochastic impedance model is to the actual impedance, the more consistent the seismic data (a) and (b) are. The more similar the two impedance models are, that is, the more reliable the simulation results are, the more consistent the seismic data (a) and (b) are theoretically; on the contrary, the greater difference the between a and b, the more unreliable the simulation results are [39]. Therefore, the simulation results and seismic wavelet were added to the seismic forward model, and the seismic forward model data were compared with the seismic data. The parameters were iteratively revised in cases of considerable differences to minimize the differences. During this process, the influence of subjective factors was avoided to obtain the most objective prediction results.
Table 4. Variogram of the different lithologies in different groups.
Figure 8a shows the initial simulation of the lithology model that uses the facies-controlled impedance simulation results to compare the seismic forward model results with the original seismic data. Figure 9a shows a significant difference between the seismic forward profile and original seismic profile. Continuous iterative calculations were performed for the different results shown in Figure 9a until the differences were minimized. This process was achieved using a Markov chain–Monte Carlo algorithm [47]. Figure 8b shows the lithology model after 32 iterative calculations. Figure 9b shows a comparison between the seismic forward model and the original seismic model. After performing iterative calculations, the correlation coefficient between the two models was above 99%. The simulation results were consistent with realistic geological conditions, which confirmed that the lithology model is reliable.
Figure 8. Markov chain–Monte Carlo lithology simulation: (a) initial simulation results. (b) Revised simulation results.
Figure 9. Comparison between the seismic forward model results and actual seismic data (red represents the forward model results, and black represents the actual seismic data): (a) comparison of the initial forward model results and actual seismic data. (b) Comparison of the revised forward model results and actual seismic data.
According to the results for the different fracture facies, a model that contained the nine lithologies in the study area could be transformed. For the four types of fracture facies, the macroregion where the fractures developed could be qualitatively obtained. In addition, certain characteristics of the opening, cutting depth, etc., were also obtained (Figure 10). As for the distribution of microfractures, such as the facture density distribution in certain types of fracture facies, further work regarding fracture prediction is needed.
Figure 10. Four types of fracture facies models in the Taiyuan Formation: (a) bedding fracture facies model. (b) N100° tectonic fracture facies model. (c) N10° tectonic fracture facies model. (d) Coal fracture facies model.

5.2. Multiattribute Comprehensive Prediction of Fracture Density

To investigate the model’s predictive ability for fracture spatial distribution, we utilized the current seismic data, incorporating both pre-stack and post-stack attributes. The use of these two technologies necessitates different data inputs due to the multifactorial origin of fractures; relying on a single attribute is insufficient for characterizing their development, a conclusion affirmed by existing literature [45,48,49,50,51]. To increase the accuracy of fracture predictions, domestic scholars have used neural networks to predict multiattribute fractures, and positive results have been achieved [52,53,54,55]. Previous work has only used seismic attributes and did not consider the characteristics related to the genesis and influencing factors of fractures in the process of predicting fractures by neural networks. To improve the accuracy of fracture predictions, this paper uses neural networks to predict the distributions of fracture facies with different genesis and development characteristics.
The seismic data used for the study area were post-processed data. We extracted nearly 100 types of attributes, such as the amplitude and instantaneous and spectrum attributes, from the seismic data volume. In this paper, three types of seismic attributes (coherence, curvature and ant body), which showed good correlations with fracture development, were optimized according to the actual conditions of the study area (Figure 11). Figure 11 shows that coherence and curvature could describe the details more accurately, whereas the ant body could describe the fractures more accurately. Previous work confirmed that the thickness of each lithology is inversely proportional to the fracture density [56,57], and this finding was applied to the study area (Figure 12). Thinner lithologies corresponded to greater fracture development. The dosage points indicate the consistency of the different lithologies (Figure 13). In addition, the average single-layer thickness attribute is considered without distinguishing the lithology (Figure 14). The average single-layer thickness of the same lithology was considered with three other types of seismic attributes for the fracture predictions through neural networks.
Figure 11. Seismic parameters of the fractures in the Taiyuan Formation: (a) coherence. (b) Curvature. (c) Ant body.
Figure 12. Fracture development for each lithology with different thicknesses. (a) Fracture development of different mudstone thickness. (b) Fracture development of different silty mudstone thickness. (c) Fracture development of different muddy siltstone thickness. (d) Fracture development of different siltstone thickness. (e) Fracture development of different fine sandstone thickness. (f) Fracture development of different medium sandstone thickness. (g) Fracture development of different coarse sandstone thickness. (h) Fracture development of different carbonaceous mudstone thickness. (i) Fracture development of different coal thickness.
Figure 13. Regression chart of each lithology with different thicknesses and fracture densities: (a) different lithology fracture density-layer thickness data projection map. (b) Fracture density average-layer thickness histogram.
Figure 14. Single-layer thickness chart of the lithologies in the Taiyuan Formation.
Initially, the four parameters mentioned above were normalized, and the input and output values were limited to [0,1] to improve the convergence of network calculations. According to the results presented above, the back propagation (BP) algorithm was adopted to synthetically predict the multiattributes of the four fracture types. The BP algorithm is a hierarchical neural network with two or more stages. The basic process includes reverse transmission through errors in the network output and adjusting and revising the connection weights and thresholds constantly to diminish the error function along the gradient direction until a minimum is reached [58].
Figure 15a,b shows a comparison of the N10° and N100° fracture density prediction results with the imaging results. From the imaging logs and fracture density curve, the N10 ° fracture facies showed greater development than the N100° fracture facies. A comparison of the imaging results with the multiattribute prediction indicates that the predicted fracture density is different from the image fracture density, although the overall trend is consistent. The prediction results reflect the fracture distribution and indicate the fracture density development in different fracture facies. Figure 16a,b shows the plane distribution of the results from the multiattribute prediction and shows the different development of the N10° and N100° fracture facies. The fracture density is higher in the well-developed fracture facies. Figure 15c,d shows the prediction of the bedding fracture facies and coal fracture facies, and the results are relatively similar to the imaging results and present an error within 10%.
Figure 15. Comparison between four groups of fracture curves: (a) panel shows the N10° fracture facies. (b) panel shows the N100° fracture facies. (c) panel shows bedding fractures. (d) panel shows coal fractures.
Figure 16. Plane graph of four fractures: (a) panel shows the N10° fracture facies. (b) panel shows the N100° fracture facies. (c) panel shows bedding fractures. (d) panel shows coal fractures.
Figure 16c,d illustrates the distribution of bedding fracture density across different fracture facies. These fractures are most developed within the well-developed fracture facies. Prediction results indicate that fractures within the coal seams are highly developed and exhibit more extensive planar distribution compared to other facies. These predictions for fracture facies of different genetic origins show good agreement with imaging log interpretations, demonstrating their robust predictive capability.

6. Conclusions

(1)
Core observations and conventional logging analyses of the abundant image logs of the study area were conducted to describe the morphology, color, geophysical features and geological characteristics to establish imaging identification templates for the 9 observed lithologies. These templates provide data that are deficient in conventional logging and reflect the geological characteristics of the different lithologies, such as their bedding and particle size. Ultimately, the results improved the identification accuracy of the complex lithologies and provided a foundation for future research of fracture facies.
(2)
Fractures were identified by integrating characteristics observed from both image logs and core samples. The genetic types, occurrence, opening, cutting depth, lithology and density of the fractures were considered in the fracture facies research. The fractures in the study area were divided into 4 facies types: bedding fracture facies, N100° tectonic fracture facies, N10° tectonic fracture facies and coal fracture facies. The different genetic fractures were further divided into the following three types according to the degree of fracture development: well-developed fracture facies, poorly developed fracture facies and undeveloped fracture facies.
(3)
For different fracture facies types, seismic attributes such as coherence, curvature, ant body, and thickness were analyzed. The neural network algorithm method indicated that the prediction results for multiattribute fracture densities in the different fracture facies were consistent with the image logs. The plane distribution of the fracture density effectively reflects the fracture facies’ characteristics, and the method mentioned in this paper is reliable. Fractures are one of the key factors that influence the development of “sweet points” within tight sandstone reservoirs. The predictive results for the fractures could effectively indicate the location of a “sweet point”. The fracture prediction method based on the complex lithologic fracture facies model is important for target optimization in tight gas exploration and productivity breakthroughs in tight sandstone gas wells.

Author Contributions

Conceptualization, Y.Z. and Y.H.; methodology, Y.H. and W.H.; validation, Z.W. and Z.R.; forma analysis, Z.Z. and Z.R.; investigation, X.C.; writing—original draft preparation, Y.Z.; writing—review and editing, Y.Z.; project administration, Y.Z.; funding acquisition, Y.H. All authors have read and agreed to the published version of the manuscript.

Funding

The research in this paper has been funded by the following projects: 2025 Autonomous Region Natural Science Foundation Project, Research on Meandering River Delta Modeling Method Based on Conditional Generative Adversarial Networks (202501A1344).

Institutional Review Board Statement

Not applicable.

Data Availability Statement

Data are available from the authors upon request.

Conflicts of Interest

Xiaoming Chen, the author, was employed by the Second Oil Production Plant of SINOPEC Jiangsu Oilfield. The rest of the authors claim that this study is in no business or financial relationship that could be interpreted as a potential conflict of interest.

References

  1. Keles, C.; Vasilikou, F.; Ripepi, N.; Agioutantis, Z.; Karmis, M. Sensitivity analysis of reservoir conditions and gas production mechanism in deep coal seams in Buchanan County, Virginia. Simul. Model. Pract. Theory 2019, 94, 31–42. [Google Scholar] [CrossRef] [Scilit]
  2. Zhao, Q.; Yang, S.; Qian, W.; Sun, Q.; Jiang, X.; Zang, H. China’s unconventional natural gas development: Status and prospects. Environ. Impact Assess. 2020, 42, 34–37. [Google Scholar]
  3. Zhao, W.; Jia, A.; Wei, Y.; Wang, J.; Zhu, H. Progress and development prospects of shale gas exploration and development in China. China Pet. Explor. 2020, 25, 31–44. [Google Scholar]
  4. Zou, C.; Zu, R.; Wu, S.; Yang, Z.; Tao, S.; Yuan, X.; Hou, L.; Yang, H.; Xu, C.; Li, D.; et al. Types, characteristics, genesis and prospects of conventional and unconventional hydrocarbon accumulations: Taking tight oil and tight gas in China as an instance. Acta Pet. Sin. 2012, 33, 173–187. [Google Scholar]
  5. Lu, Y.; Zhang, H.; Zhou, Z.; Ge, Z.; Chen, C.; Hou, Y.; Ye, M. Current Status and Effective Suggestions for Efficient Exploitation of Coalbed Methane in China: A Review. Energy Fuels 2021, 35, 9102–9123. [Google Scholar] [CrossRef] [Scilit]
  6. Wang, N.; Pei, L.; Lei, D.; Zeng, B. Distribution and development status of unconventional natural gas resources in China. Pet. Geol. Recover. Effic. 2015, 1, 26–31. [Google Scholar]
  7. Xie, Y.; Meng, S.; Gao, L.; Sun, X.; Duan, C.; Wang, H. Assessments on potential resources of deep coalbed methane and compact sandstone gas in Linxing Area. Coal Sci. Technol. 2015, 43, 21–28. [Google Scholar]
  8. Xu, J.; Zhang, B.; Qin, Y.; Cao, G.; Zhang, H. Method for calculating the fracture porosity of tight-fracture reservoirs. Geophysics 2016, 81, 57–70. [Google Scholar] [CrossRef] [Scilit]
  9. Folkestad, A.; Veselovsky, Z.; Roberts, P. Utilising borehole image logs to interpret delta to estuarine system: A case study of the subsurface Lower Jurassic Cook Formation in the Norwegian northern North Sea. Mar. Petrol. Geol. 2012, 29, 255–275. [Google Scholar] [CrossRef] [Scilit]
  10. Ameen, M.S. Fracture and in-situ stress patterns and impact on performance in the khuff structural prospects, eastern offshore Saudi Arabia. Mar. Petrol. Geol. 2014, 50, 166–184. [Google Scholar] [CrossRef] [Scilit]
  11. Kosari, E.; Ghareh-Cheloo, S.; Kadkhodaie-Ilkhchi, A.; Bahroudi, A. Fracture characterization by fusion of geophysical and geomechanical data: A case study from the Asmari reservoir, the Central Zagros fold-thrust belt. J. Geophys. Eng. 2015, 12, 130–143. [Google Scholar] [CrossRef] [Scilit]
  12. Wilson, T.H.; Smith, V.; Brown, A. Developing a model discrete fracture network, drilling, and enhanced oil recovery strategy in an unconventional naturally fractured reservoir using integrated field, image log, and three-dimensional seismic data. AAPG Bull. 2015, 99, 735–762. [Google Scholar] [CrossRef] [Scilit]
  13. Brekke, H.; MacEachern, J.A.; Roenitz, T.; Dashtgard, S.E. The use of microresistivity image logs for facies interpretations: An example in point-bar deposits of the McMurray Formation, Alberta, Canada. AAPG Bull. 2017, 101, 655–682. [Google Scholar] [CrossRef] [Scilit]
  14. Moreau, J.; Joubert, J.B. Glacial sedimentology interpretation from borehole image log: Example from the Late Ordovician deposits, Murzuq Basin (Libya). Interpretation 2016, 4, B1–B16. [Google Scholar] [CrossRef] [Scilit]
  15. Khoshbakht, F.; Azizzadeh, M.; Memarian, H.; Nourozi, G.H.; Moallemi, S.A. Comparison of electrical image log with core in a fractured carbonate reservoir. J. Petrol. Sci. Eng. 2012, 86, 289–296. [Google Scholar]
  16. Muniz, M.C.; Bosence, D.W.J. Pre-salt microbialites from the campos basin (off-shore Brazil): Image log facies, facies model and cyclicity in lacustrine carbonates. Geol. Soc. Lond. Spec. Publ. 2015, 418, 221–242. [Google Scholar] [CrossRef] [Scilit]
  17. Ran, Y.; Wang, G.; Zhou, Z.; Lai, J.; Dai, Q.; Chen, J.; Fan, X.; Wang, S. Identification of lithology and lithofacies type and its application to Chang 7 tight oil in Heshui area. Ordos Basin. Chin. Geol. 2016, 43, 1331–1340. [Google Scholar]
  18. Zhou, Z.; Wang, G.; Ran, Y.; Lai, J.; Cui, Y.; Zhao, X. A logging identification method of tight oil reservoir lithology and lithofacies: A case from Chang 7 Member of Triassic Yanchang Formation in Heshui area, Ordos Basin, NW China. Petrol. Explor. Dev. 2016, 43, 61–68. [Google Scholar] [CrossRef] [Scilit]
  19. Li, C.; Zhou, C. A new method to calculate lithological profile with micro-resistivity imaging log. Well Logging Technol. 2008, 32, 45–48. [Google Scholar]
  20. Lai, J.; Wang, G.; Wang, S.; Cao, J.; Li, M.; Pang, X.; Han, C.; Fan, X.; Yang, L.; He, Z.; et al. A review on the applications of image logs in structural analysis and sedimentary characterization. Mar. Pet. Geol. 2018, 95, 139–166. [Google Scholar] [CrossRef] [Scilit]
  21. Wang, M.; Xue, L.; Pan, B. Lithology identification of igneous rock using FMI texture analysis. Well Logging Technol. 2009, 33, 110–114. [Google Scholar]
  22. Wilson, M.E.J.; Lewis, D.; Yogi, O.; Holland, D.; Hombo, L.; Goldberg, A. Development of a Papua New Guinean onshore carbonate reservoir: A comparative borehole image (BHI) and petrographic evaluation. Mar. Petrol. Geol. 2013, 44, 164–195. [Google Scholar] [CrossRef] [Scilit]
  23. BinAbadat, E.; Bu-Hindi, H.; Lehmann, C.; Kumar, A.; AL-Harbi, H.; AL-Ali, A.; Al Katheeri, A. A Workflow to Integrate Core and Image Logs in Order to Enhance the Characterization of Subsurface Facies on Carbonate Reservoirs, Offshore Abu Dhabi. In Proceedings of the Abu Dhabi International Petroleum Exhibition and Conference, SPE, Abu Dhabi, United Arab Emirates, 12–15 November 2018. [Google Scholar]
  24. You, Z.; Du, X.; Hou, H.; Zhou, K.; Zhang, H. Discussion on Imaging Log Interpretation Mode. Well Logging Technol. 2000, 24, 393–399. [Google Scholar]
  25. Lai, J.; Wang, G.; Huang, L.; Li, W.; Ran, Y.; Wang, D.; Zhou, Z.; Chen, J. Brittleness index estimation in a tight shaly sandstone reservoir using well logs. J. Nat. Gas Sci. Eng. 2015, 27, 1536–1545. [Google Scholar] [CrossRef] [Scilit]
  26. Zhang, R.; Wang, J.; Ma, Y.; Chen, G.; Zeng, Q.; Zhou, C. Sedimentary microfacies and palaeogeomorphology as well as their controls on gas accumulation within the deep-buried cretaceous in Kuqa depression, Tarim basin, China. J. Nat. Gas Geosci. 2016, 1, 45–59. [Google Scholar] [CrossRef] [Scilit]
  27. Ameen, M.S.; MacPherson, K.; Al-Marhoon, M.I.; Rahim, Z. Diverse fracture properties and their impact on performance in conventional and tight-gas reservoirs, Saudi Arabia: The Unayzah, South Haradh case study. AAPG Bull. 2012, 96, 459–492. [Google Scholar] [CrossRef] [Scilit]
  28. Lyu, W.; Zeng, L.; Liu, Z.; Liu, G.; Zu, K. Fracture responses of conventional logs in tight-oil sandstones: A case study of the Upper Triassic Yanchang Formation in southwest Ordos Basin, China. AAPG Bull. 2016, 100, 1399–1417. [Google Scholar] [CrossRef] [Scilit]
  29. Liu, J.; Ding, W.; Wang, R.; Yang, S.; Yang, H.; Gu, Y. Simulation of paleotectonic stress fields and quantitative prediction of multi-period fractures in shale reservoirs: A case study of the niutitang formation in the lower Cambrian in the Cen’gong block, South China. Mar. Petrol. Geol. 2017, 84, 289–310. [Google Scholar] [CrossRef] [Scilit]
  30. Lyu, W.; Zeng, L.; Zhang, B.; Miao, F.; Lyu, P.; Dong, S. Influence of natural fractures on gas accumulation in the Upper Triassic tight gas sandstones in the northwestern Sichuan Basin, China. Mar. Petrol. Geol. 2017, 83, 60–72. [Google Scholar] [CrossRef] [Scilit]
  31. Khoshbakht, F.; Memarian, H.; Mohammadnia, M. Comparison of Asmari, Pabdeh and Gurpi formation’s fractures, derived from image log. J. Petrol. Sci. Eng. 2009, 67, 65–74. [Google Scholar] [CrossRef] [Scilit]
  32. Ameen, M.S. Fracture modes in the Silurian Qusaiba shale play, northern Saudi Arabia and their geomechanical implications. Mar. Petrol. Geol. 2016, 78, 312–355. [Google Scholar] [CrossRef] [Scilit]
  33. Dai, J.; Zhang, Y.; Wang, Z.; Chen, Y. Study on Artificial Fracture System in Block 9 of Cainan Oilfield. J. Southwest Pet. Univ. 2003, 25, 16–18. [Google Scholar]
  34. Lai, J.; Wang, G.; Zheng, X.; Zhou, L.; Han, C.; Wu, D.; Huang, L.; Luo, G. Quantitative evaluation of tight gas sandstone reservoirs of Bashijiqike formation in Dabei gas field. J. Cent. South Univ. 2015, 46, 2285–2298. [Google Scholar]
  35. Zhang, Y.; Liao, C.; Shi, W.; Hu, B. Neotectonic Evolution of the Peripheral Zones of the Ordos Basin and Geodynamic Setting. Geol. J. China Univ. 2006, 12, 285–297. [Google Scholar]
  36. Yang, Y. The Research of Fractures in the Ha’nan Oilfield. Master’s Thesis, Northwest University, Xi’an, China, 2001. [Google Scholar]
  37. Zhang, Y.; Zhao, X.; Wang, Y.; Liu, P.; Bai, H. Physical Simulation Experiment of Normal Faults Associated Fractures. Sci. Technol. Eng. 2010, 36, 8975–8979. [Google Scholar]
  38. Mu, L. Study on Reservoir Fracture Prediction; Petroleum Industry Press: Beijing, China, 2009. [Google Scholar]
  39. Hu, Y.; Yu, X.; Li, S.; Chen, G. Improving the accuracy of geological model by using seismic forward and inversion techniques. Pet. Explor. Dev. 2014, 41, 190–197. [Google Scholar] [CrossRef] [Scilit]
  40. Si, W.; Di, B.; Wei, J. Seismic responses of tight gas sand of different water saturations: A physical model research. In SEG Technical Program Expanded Abstracts 2013; Society of Exploration Geophysicists: Houston, TX, USA, 2013; pp. 2592–2596. [Google Scholar]
  41. Dong, N.; Liu, Z.; Zhou, X.; Wang, J.; Liu, J.; Xia, H.; She, G. Seismic detection and prediction of tight gas reservoirs in the Ordos Basin. Oil Gas Geol. 2008, 29, 668–675. [Google Scholar]
  42. Ye, T.; Fu, S.; Lyu, Q.; Lai, W. Application of multiwave joint inversion to the prediction of relatively high-quality reservoirs—An example from the prediction of deep tight clastic sandstone reservoirs in the western Sichuan Basin. Oil Gas Geol. 2009, 30, 357–362. [Google Scholar]
  43. Liu, Z. Key techniques of integrated seismic reservoir characterization for tight oil & gas sands. Prog. Geophys. 2014, 29, 182–190. [Google Scholar]
  44. Zhu, C.; Xia, Z.; Wang, C.; Song, G.; Wei, X.; Wang, P.; Wang, H.; Wang, B. Seismic prediction for sweet spot reservoir of tight oil. J. Jiliin Univ. 2015, 45, 602–610. [Google Scholar]
  45. Wang, F.; Cheng, L.; Liu, J.; Xie, Q.; Huang, S.; Huang, Z. Research and application of post-stack seismic attributes in recognizing shale gas reservoir fracture. Coal Geol. Explor. 2015, 43, 113–116. [Google Scholar]
  46. Caers, J. Petroleum Geostatistics; Society of Petroleum Engineers: Richardson, TX, USA, 2005. [Google Scholar]
  47. Bardou, E. Markov Processes and Their Applications; Higher Education Publishing House: Beijing, China, 2019. [Google Scholar]
  48. Chen, H.; Zhang, G. Detection of Natural Fractures from Observed Surface Seismic Data Based on a Linear-Slip Model. Pure Appl. Geophys. 2018, 175, 2769–2784. [Google Scholar] [CrossRef] [Scilit]
  49. Wang, Y.; Chen, X.; Wang, Y.; Gui, Z. Application of multiple post-stack seismic attributes in the prediction of carboniferous fracture in west Hashan. Prog. Geophys. 2014, 29, 1772–1779. [Google Scholar]
  50. Wang, D.; Jia, Y.; Wei, S.; Zheng, W. The integrated poststack fracture prediction of Xusi formation in Xinchang survey. Geophys. Geochem. Explor. 2014, 38, 1038–1044. [Google Scholar]
  51. Qi, Q.; Qu, S.; Li, Z.; Sun, Z.; Chen, Y. Fracture prediction by using pre-stack and post-stack seismic techniques in S2 block, Yemen. Geophys. Prospect. Pet. 2014, 53, 603–608. [Google Scholar]
  52. Zhang, Y. Logging Evaluation Technique for Tight Clastic Reservoir in Area of West Sichuan. Well Logging Technol. 2004, 28, 419–422. [Google Scholar]
  53. Sun, S.; Zhu, Q.; Peng, S. Artificial neural network pattern recognition of fractures in the bedrock reservoir of the Qijia buried hill. Geophys. Geochem. Explor. 2007, 32, 160–163. [Google Scholar]
  54. Zhou, J.; Yan, T.; Tu, H. Application of artificial neural network to identification of drilled strata. Earth Sci.-J. China Univ. Geosci. 2000, 25, 642–645. [Google Scholar]
  55. Xue, Y.; Cheng, L.; Mou, J.; Zhao, W. A new fracture prediction method by combining genetic algorithm with neural network in low-permeability reservoirs. J. Pet. Sci. Eng. 2014, 121, 159–166. [Google Scholar] [CrossRef] [Scilit]
  56. Fan, J.; Qu, X.; Wang, C.; Lei, Q.; Cheng, L.; Yang, Z. Natural fracture distribution and a new method predicting effective fractures in tight oil reservoirs of Ordos Basin. Pet. Explor. Dev. 2016, 43, 740–748. [Google Scholar] [CrossRef] [Scilit]
  57. Liu, G.; Huang, C.; Zhou, X.; Zhang, L.; Pan, Y. Quantitative evaluation of fracture development in Triassic Yanchang Formation, Ordos Basin. Pet. Explor. Dev. 2015, 42, 444–453. [Google Scholar] [CrossRef] [Scilit]
  58. Kajornrit, J.; Chaipornkaew, P. A comparative study of ensemble back-propagation neural network for the regression problems. In Proceedings of the 2017 2nd International Conference on Information Technology (INCIT), Nakhonpathom, Thailand, 2–3 November 2017; pp. 1–6. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Journal Statistics
Multiple requests from the same IP address are counted as one view.
A Multistage Manufacturing Process Path Planning Method Based on AEC-FU Hybrid Decision-Making
Previous
Coastal Environments: Threats to Investment Processes in the Coastal Area
Next