Quantitative prediction method for connectivity of fluvial facies sand body based on characteristic factor master control

Abstract

The application discloses a river facies sand body connectivity quantitative prediction method based on a characteristic factor master control, and belongs to the field of oil and gas field exploration and development and geophysical technology. The method is based on the data of drilled wells in a target area, carries out sand body connectivity analysis, and screens out typical sand bodies with reliable connectivity. Then, the horizontal gradient and waveform change index attributes of the screened sand bodies are extracted to construct a master control characteristic factor. Meanwhile, a plurality of seismic attributes reflecting sand body connectivity are extracted, and high sensitivity attributes are reserved as auxiliary characteristic factors through optimization processing. Finally, based on a depth forest algorithm, a mapping model between a multi-factor combination and sand body connectivity is established through machine learning, and the corresponding factor combination of a to-be-predicted sand body is input into the model, so that the quantitative determination of the connectivity can be realized. The application has high accuracy and timeliness in processing the river facies sand body connectivity prediction problem, and provides a new idea and method for the sand body connectivity research under similar complex geological conditions.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas field exploration and development and geophysical technology, and in particular relates to a quantitative prediction method for fluvial sand body connectivity based on characteristic factors. Background Technology

[0002] Fluvial sand bodies are an important type of oil and gas reservoir, accounting for as much as 42.6% of my country's clastic rock oil reserves. Identifying the discontinuous boundaries of this type of sand body has always been a challenge and a hot topic in geophysical research for oil and gas exploration. Fluvial sand bodies typically exhibit wide distribution, rapid lateral variation, and multiple vertically superimposed sedimentary characteristics, leading to complex internal connectivity relationships and making it difficult to accurately characterize the connectivity structure. Furthermore, due to the limitations of seismic data resolution, these sand bodies often exhibit "seemingly connected but not truly connected" phase axis characteristics on seismic profiles, easily causing misjudgments of connectivity, thus affecting drilling success rates and development effects.

[0003] Current conventional methods for predicting sand body connectivity include seismic attribute analysis (such as amplitude, frequency, and waveform) and high-resolution processing and inversion of seismic data. However, in practical applications, seismic attribute analysis methods are subject to significant human intervention and multiple solutions. While seismic data frequency upscaling and inversion techniques can improve vertical resolution to some extent, they often come at the cost of signal-to-noise ratio and may even introduce spurious phase axes, affecting the reliability of interpretation. Furthermore, existing methods are mostly limited to qualitative judgment of sand body connectivity (i.e., simply classifying them as "connected" or "disconnected"), lacking the ability to quantitatively predict connectivity probabilities, and thus failing to meet the needs of detailed evaluation of complex superimposed sand bodies. Summary of the Invention

[0004] The problem this invention aims to solve is to provide a quantitative prediction method for fluvial sand body connectivity based on feature factors. This method uses existing well data in the study area as a basis to screen typical sand bodies with reliable connectivity. First, it extracts two sensitive attributes, lateral gradient and waveform change index, to construct the main control feature factors. At the same time, it extracts a variety of seismic attributes for sand body connectivity identification. Through optimization, it retains highly sensitive attributes as auxiliary factors. Finally, it uses deep forest algorithm to perform machine learning and establish a mapping model between multi-factor combination and sand body connectivity to achieve quantitative prediction of sand body connectivity.

[0005] To solve the above-mentioned technical problems, the technical solution adopted in this invention is: a quantitative prediction method for fluvial sand body connectivity based on characteristic factors, which predicts sand body connectivity based on drilled well data and seismic data, including the following steps: S1: Based on the exploration and development data of drilled sand bodies in the target area, conduct sand body connectivity analysis, screen out typical sand bodies with reliable connectivity, and divide the typical sand bodies into connected regions and non-connected regions; S2: Extract the lateral gradient and waveform change index attributes of the selected typical sand bodies to construct the main control feature factors; S3: Extract seismic attributes reflecting sand body connectivity, and select highly sensitive attributes as auxiliary feature factors; S4: Using the connected regions and the disconnected regions as training targets, and the main control feature factor and the auxiliary feature factor to form a feature factor combination as input vector, a deep forest algorithm is used to construct a mapping model between the feature factor combination and the sand body connectivity. S5: Input the combination of the characteristic factors of the sand body to be predicted into the mapping model, calculate the connectivity probability of each sampling point on the sand body plane, and realize quantitative prediction of connectivity.

[0006] Furthermore, in S1, the typical sand body screening criteria are as follows: In the multiple wells that encountered sand bodies, there were contradictions in the oil-water relationship, ensuring that there were unconnected zones inside the sand bodies that served to divide them; Regions exhibiting gradient changes or banded anomalies in the planar properties of sand bodies are used as a priori knowledge to divide typical sand bodies into connected and disconnected regions.

[0007] Furthermore, in S2, the lateral gradient attribute reflects the lateral discontinuity of the typical sand body, and the lateral gradient calculation formula is as follows: In the formula, G represents the lateral gradient attribute, and λ2 and λ3 represent the gradient changes of the sand body in the two horizontal directions, respectively.

[0008] Furthermore, in S2, the waveform change index reflects the waveform change of the typical sand body seismic reflection, and the formula for calculating the waveform change index is as follows: In the formula, C represents the waveform change exponential attribute, SK represents the trough skewness, S represents the trough area, β is the dimensionless normalization constant, and β = sample average amplitude / sampling interval.

[0009] Furthermore, the trough skewness reflects the waveform asymmetry, and the formula for calculating the trough skewness is as follows: In the formula, m represents the third central moment. X represents the standard deviation, N represents the sample size, and X represents the standard deviation. i This represents the amplitude value at each sampling point on the seismic waveform. This represents the average amplitude.

[0010] Furthermore, the trough area reflects the change in waveform shape, and the formula for calculating the trough area is as follows: In the formula, T i T i+1 A represents the location of each sampling point at different times. i A i+1 This represents the amplitude value at the corresponding moment.

[0011] Furthermore, the formula for constructing the feature factors is as follows: Furthermore, step S3 includes the following steps: S31: Use Pearson correlation analysis to remove redundant attributes; S32: Conduct test training on the divided sand body samples, use SHAP values ​​to evaluate the sensitivity of various attributes in the test set to connectivity, and retain the required number of highly sensitive attributes as the auxiliary feature factors.

[0012] Furthermore, in S4, the connected regions of the river facies sand bodies are significantly larger than the disconnected regions, resulting in an excessively high proportion of connected samples. Before model training, the samples need to be undersampled to control the ratio of connected to disconnected samples within a preset threshold of 1:1 to 2:1, so as to avoid the model being biased towards the dominant category.

[0013] Furthermore, in S5, the prediction result is determined by the average value of the classification results output by the last basic classifier. The connectivity probability of each sampling point is accurately calculated, and the average value of the classification results is used as the gray connectivity probability attribute to quantitatively predict the connectivity of the sand body.

[0014] The advantages and positive effects of this invention are: This invention systematically screens sand bodies within a target area to obtain typical sand bodies with reliable connectivity. Then, it extracts lateral gradient attributes and waveform change index attributes to construct principal control feature factors. Simultaneously, it extracts and optimizes multiple seismic attributes to obtain auxiliary feature factors. Finally, based on the deep forest algorithm, it establishes a mapping model between the combination of feature factors and the connectivity of sand bodies through machine learning, achieving quantitative prediction of sand body connectivity. In practical applications, this invention demonstrates high accuracy and timeliness in predicting the connectivity of fluvial sand bodies, providing new ideas and methods for studying sand body connectivity under similar complex geological conditions. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.

[0016] Figure 2This is a schematic diagram of typical sand body connectivity region division in an embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram of the preferred process for connectivity-sensitive attributes according to an embodiment of the present invention.

[0018] Figure 4 This is a diagram showing the prediction results of sand body connectivity in an embodiment of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, the quantitative prediction method for riverine sand body connectivity based on characteristic factors includes the following steps.

[0021] S1: Based on the exploration and development data of drilled sand bodies within the target area, conduct sand body connectivity analysis, screen out typical sand bodies with reliable connectivity, and divide these typical sand bodies into connected and disconnected regions. Specifically, the screened typical sand bodies all meet the following criteria: In the multiple wells that encountered sand bodies, there were contradictions in the oil-water relationship, ensuring that there were unconnected zones inside the sand bodies that served to divide them; The sand body should have areas with gradient changes or strip-shaped anomalies in its planar properties. Based on this prior knowledge, it can be divided into connected and disconnected regions.

[0022] S2: Extract the lateral gradient and waveform change index attributes of the typical sand bodies selected in S1, and construct the main control feature factors.

[0023] Specifically, the controlling characteristic factor is composed of the lateral gradient attribute and the waveform change index. The lateral gradient attribute mainly reflects the lateral discontinuity of the sand body. In three-dimensional space, the sand body gradient attribute includes three directions: λ1 is the vertical gradient, mainly reflecting the internal changes of the sand body; λ2 and λ3 are the lateral gradients, mainly reflecting the lateral changes of the sand body. Conventional gradient properties include gradients in three directions. However, the large value of λ1 weakens the responses of λ2 and λ3. Lateral gradient properties, by eliminating the influence of large vertical gradient values ​​(λ1), better highlight the lateral discontinuities of the sand body than conventional gradient properties. The formula for calculating lateral gradient properties is as follows: In the formula, G represents the lateral gradient attribute, and λ2 and λ3 represent the gradient changes of the sand body in the two horizontal directions, respectively.

[0024] The waveform variation index primarily reflects the waveform changes in seismic reflections from sand bodies. In positive polarity seismic data, the wave impedance of fluvial sandstone mudstone is often greater than that of sandstone, thus the top surface of sandstone corresponds to a wave trough. For a single sand body, the sand body exhibits a single-axis wave trough with an approximate skewness and a large trough area. In areas of superimposed sand bodies, a complex wave is often observed, where the skewness of the trough increases and the trough area decreases. Therefore, by calculating the skewness and area of ​​the trough in superimposed sand body areas separately and then combining them, the waveform differences in superimposed sand body areas are highlighted through quotient calculation, thereby achieving a more detailed characterization of the waveform structure. The formula for calculating the waveform variation index is as follows: The trough skewness SK, which reflects the waveform asymmetry, is calculated using the following formula: In the formula, m represents the third central moment. X represents the standard deviation, N represents the sample size, and X represents the standard deviation. i This represents the amplitude value at each sampling point on the seismic waveform. This represents the average amplitude. If the waveform is symmetrical, SK≈0. In disconnected zones of sand bodies, the waveform is usually asymmetrical, and |SK| increases.

[0025] The trough area S, which reflects the change in waveform shape, is calculated using the following formula: In the formula, T i T i+1 A represents the location of each sampling point at different times. i A i+1 This represents the amplitude value at the corresponding moment. In the disconnected zones of the sand body, the waveform typically narrows, and S decreases.

[0026] Furthermore, considering that the dimension of the trough skewness SK is the square of the amplitude, and the dimension of the trough area S is time × amplitude, in order to eliminate the influence of the dimensions on feature factor fusion and subsequent machine learning modeling, this invention introduces a normalization constant β, as shown in the following formula: In the formula, The average amplitude of the samples within the representative work area, To determine the sampling interval, the same β value is used for all sand bodies within the same target area during actual calculations to ensure feature comparability.

[0027] The waveform change index is calculated to highlight changes in waveform structure. The formula is as follows: In the formula, C represents the waveform change exponential property, SK represents the trough skewness, S represents the trough area, and β is a dimensionless normalization constant.

[0028] After obtaining the lateral gradient properties and waveform change index, the two are fused through multiplication to construct a feature factor, which has a higher sensitivity to sand body connectivity.

[0029] S3: Extract as many seismic attributes as possible that can reflect the connectivity of the sand bodies selected in S1, and optimize them, retaining the required number of highly sensitive attributes as auxiliary feature factors. Specifically, the optimization of seismic attributes includes the following two steps: The first step is to use Pearson correlation analysis to remove redundant attributes; The second step involves testing and training the sand body samples that have been divided in S1. The SHAP value is used to evaluate the sensitivity of various attributes in the test set to connectivity, and the required number of highly sensitive attributes are retained as auxiliary feature factors.

[0030] S4: The connected and disconnected regions of the sand body in S1 are used as training targets. The main control feature factors in S2 and the auxiliary feature factors in S3 are combined into multiple factor combinations as input vectors. The deep forest algorithm is used to construct a mapping model between each factor combination and the connectivity of the sand body.

[0031] Specifically, in riverine facies sand bodies, the connected regions are significantly larger than the disconnected regions, resulting in an excessively high proportion of connected samples. Before model training, undersampling is necessary to control the ratio of connected to disconnected samples within a preset threshold (e.g., between 1:1 and 2:1) to prevent the model from favoring the dominant class. The deep forest algorithm comprises two parts: multi-granularity scanning and cascaded forest. The principle of multi-granularity scanning is as follows: Given an m-dimensional sequence feature vector as input. If a vector window of size n and step size 1 is used to scan X, the final result will be... We generate n-dimensional feature vectors. We then use these generated feature vectors to train a random forest and a fully random forest, respectively. Assuming the original data has k categories of labels, both forests will generate... We have k-dimensional class vectors. Flattening and concatenating these vectors, we can finally obtain a vector with dimension k. The feature vectors are obtained, thus completing a single sliding window scan. Subsequently, the size of the sliding window is modified to obtain the corresponding probability vectors. The probability vectors from different sliding windows are then merged, and the merged result serves as the training dataset for the cascaded forest.

[0032] The principle of cascaded forests is as follows: Cascaded forests consist of multiple layers of learners. Each layer comprises a 1:1 ratio of ordinary random forests and fully random forests. For example, a single-layer ensemble learner might consist of two fully random forests and two ordinary random forests. In the first layer, the training input samples have a dimension of n. The four random forests are trained using multi-fold cross-validation, outputting four probability vectors of dimension k, corresponding to the probabilities of k predicted classifications. In subsequent layers (second to Nth layers), the 4k-dimensional predictions from the previous layer are merged with the initial n-dimensional samples to form a single ensemble learner. The model uses k-dimensional vectors as new training samples for training. Each layer updates its training samples based on the results of the previous layer until a stopping condition is met. There are typically two stopping conditions: one is that the model reaches a preset number of layers; the other is that training stops when the accuracy decreases to a given threshold after two consecutive rounds, based on the changes in accuracy between layers. After the stopping condition is met, the last layer of the model outputs four sets of k-dimensional prediction probability vectors. The average of the k-class probabilities of these four vectors is used to generate the prediction probabilities for the k categories. This completes the entire training of the Deep Forest algorithm.

[0033] S5: For the sand body to be predicted within the target area, extract the corresponding attributes to establish a combination of feature factors, input them into the model established in S4, and calculate the connectivity probability of each sampling point on the sand body plane to achieve quantitative prediction of connectivity. Specifically, the prediction result is determined by the average value of the classification results output by the last basic classifier, thereby accurately calculating the connectivity probability of each sampling point. This value can be used as the connectivity probability attribute to quantitatively determine the connectivity of the sand body.

[0034] The present invention will now be described in detail with reference to specific embodiments: This invention integrates traditional attribute analysis techniques with artificial intelligence methods, and innovatively proposes a quantitative prediction method for sand body connectivity based on characteristic factors, addressing the problem of inaccurate prediction of riverine sand body connectivity. The process is as follows: Figure 1 As shown, starting from the drilled data in the target area, typical sand bodies with reliable connectivity are selected. Based on exploration and development experience and changes in planar properties, these sand bodies are divided into connected and disconnected regions. Then, the top and bottom interfaces of the selected sand bodies are meticulously tracked, and lateral gradient properties and waveform change indices are extracted to construct the main control feature factors. Simultaneously, multiple seismic attributes reflecting sand body connectivity are extracted, and highly sensitive attributes are retained as auxiliary feature factors through optimization. Finally, based on the deep forest algorithm, the combination of multiple factors is used as the input feature vector, and sand body connectivity is used as the desired output. A mapping model between the two is constructed using the deep forest algorithm. By inputting the corresponding factor combination of the sand body to be predicted into this model, the quantitative determination of sand body connectivity can be achieved.

[0035] Specifically, it includes the following steps: Typical Sand Body Screening: A systematic review and screening of drilled sand bodies within the work area was conducted. Selected sand bodies met the following conditions: There were contradictions in the oil-water relationship between the wells encountering the sand body, ensuring the existence of discontinuous zones within the sand body that act as dividing lines. Simultaneously, the sand should exhibit gradient changes or banded anomalies in its planar properties. Based on this prior knowledge, it can be divided into connected and disconnected regions. Figure 2 For example, during the development of the Nm13 sand body, the injection and production of well D17 and well I32H showed no connection, while the injection and production of well I32H and well D23 showed connection. Furthermore, there are attribute change bands within the sand body. Based on these change bands, the sand body can be segmented into connected and disconnected regions, which can then be used as training samples for subsequent machine learning.

[0036] Feature factor construction: The top and bottom interfaces of typical sand bodies are precisely tracked, and the lateral gradient attribute and waveform change index attribute are extracted to construct the main control feature factors. Then, multiple types of seismic attributes that can reflect the connectivity of sand bodies are extracted to obtain auxiliary feature factors. Figure 3 The process of selecting the feature factors is demonstrated. First, the correlation coefficients between the factors are calculated using Pearson correlation analysis. Here, factors with a correlation coefficient greater than or equal to 0.7 (strong correlation) are selected and retained. Next, the model is tested, and the influence of each factor on sand body connectivity is calculated using SHAP values. Here, factors with an influence greater than 10% are retained. After two steps of screening, six types of feature factors are finally retained for subsequent connectivity prediction.

[0037] Model Training: After screening, 31 sand bodies with 6 feature factors were selected. Undersampling was applied to ensure a 1:1 ratio of connected to disconnected samples, resulting in a total of 178,000 sample points used in model training. Regarding model construction... Figure 4 The parameters selected for model training and the training results are shown. The sliding window size used in multi-granularity scanning is [3, 5]. Each layer of the cascaded forest uses two completely random forests and two ordinary random forests. The number of decision trees in each random forest is 100, the maximum tree depth is 10, and the maximum number of cascaded layers is 10. The sample points are divided into training and test sets in a 5:1 ratio to verify the model accuracy. The final training results are shown in Table 1.

[0038] Table 1: Model Training Results Sand body connectivity prediction: Taking sand body No. 3 in well A1 as an example, this sand body has already encountered the oil-water interface in well A1. Well A2, which is located in a low position and will be deployed, can also encounter this sand body. In order to determine the exploration potential of well A2, it is necessary to determine the connectivity between the two wells. Figure 4As can be seen, the inversion profile of the two wells showed a seemingly connected but not truly connected phenomenon, and conventional amplitude attributes could not effectively determine their connectivity. By extracting feature factors from the sand body and inputting them into the model, machine learning results could calculate the connectivity probability of each sampling point in the plane, dividing the sand body into connected regions (greater than 75%), ambiguous regions (35%–75%), and disconnected regions (less than 35%) based on connectivity probability. The two wells were separated by one disconnected region and one ambiguous region, indicating that the sand body was not connected, and well A2 still had significant exploration potential. Subsequently, the well successfully encountered a 9m oil layer, confirming the reliability of the model.

[0039] In summary, this invention systematically screens fluvial sand bodies in the Shijiu Tuo Uplift A oilfield of the Bozhong Depression to obtain typical sand bodies with reliable connectivity. Then, it extracts lateral gradient attributes and waveform change index attributes to construct principal control feature factors. Simultaneously, it extracts and optimizes multiple seismic attributes to obtain auxiliary feature factors. Finally, based on the deep forest algorithm, it establishes a mapping model between the combination of feature factors and the connectivity of sand bodies through machine learning, achieving quantitative prediction of sand body connectivity. Practical application demonstrates that this method has high accuracy and timeliness in dealing with the problem of fluvial sand body connectivity prediction, providing new ideas and methods for sand body connectivity research under similar complex geological conditions.

[0040] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A quantitative prediction method for fluvial sand body connectivity based on characteristic factors, which predicts sand body connectivity based on drilled well data and seismic data, characterized by: Includes the following steps, S1: Based on the exploration and development data of drilled sand bodies in the target area, conduct sand body connectivity analysis, screen out typical sand bodies with reliable connectivity, and divide the typical sand bodies into connected regions and non-connected regions; S2: Extract the lateral gradient and waveform change index attributes of the selected typical sand bodies to construct the main control feature factors; S3: Extract seismic attributes reflecting sand body connectivity, and select highly sensitive attributes as auxiliary feature factors; S4: Using the connected regions and the disconnected regions as training targets, and the main control feature factor and the auxiliary feature factor to form a feature factor combination as input vector, a deep forest algorithm is used to construct a mapping model between the feature factor combination and the sand body connectivity. S5: Input the combination of the characteristic factors of the sand body to be predicted into the mapping model, calculate the connectivity probability of each sampling point on the sand body plane, and realize quantitative prediction of connectivity.

2. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control as described in claim 1, characterized in that: In S1, the typical sand body screening criteria are as follows: In the multiple wells that encountered sand bodies, there were contradictions in the oil-water relationship, ensuring that there were unconnected zones inside the sand bodies that served to divide them; Regions exhibiting gradient changes or banded anomalies in the planar properties of sand bodies are used as a priori knowledge to divide typical sand bodies into connected and disconnected regions.

3. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 1 or 2, characterized in that: In S2, the lateral gradient attribute reflects the lateral discontinuity of the typical sand body, and the lateral gradient calculation formula is as follows: In the formula, G represents the lateral gradient attribute, and λ2 and λ3 represent the gradient changes of the sand body in the two horizontal directions, respectively.

4. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control as described in claim 3, characterized in that: In S2, the waveform change index reflects the waveform change of the typical sand body seismic reflection, and the formula for calculating the waveform change index is as follows: In the formula, C represents the waveform change exponential attribute, SK represents the trough skewness, S represents the trough area, β is the dimensionless normalization constant, and β = sample average amplitude / sampling interval.

5. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor mastery as described in claim 4, characterized in that: The trough skewness reflects the waveform asymmetry, and the formula for calculating the trough skewness is as follows: In the formula, m represents the third-order central moment. X represents the standard deviation, N represents the sample size, and X represents the standard deviation. i This represents the amplitude value at each sampling point on the seismic waveform. This represents the average amplitude.

6. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 4, characterized in that: The trough area reflects the change in waveform shape, and the formula for calculating the trough area is as follows: In the formula, T i T i+1 A represents the location of each sampling point at different times. i A i+1 This represents the amplitude value at the corresponding moment.

7. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 4, characterized in that: The formula for constructing the feature factors is as follows: 。 8. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 1 or 2, characterized in that: S3 includes the following steps: S31: Use Pearson correlation analysis to remove redundant attributes; S32: Conduct test training on the divided sand body samples, use SHAP values ​​to evaluate the sensitivity of various attributes in the test set to connectivity, and retain the required number of highly sensitive attributes as the auxiliary feature factors.

9. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 1 or 2, characterized in that: In S4, the connected regions of the river facies sand bodies are significantly larger than the disconnected regions, resulting in an excessively high proportion of connected samples. Before model training, the samples need to be undersampled to control the ratio of connected to disconnected samples within a preset threshold of 1:1 to 2:1, in order to avoid the model being biased towards the dominant category.

10. The quantitative prediction method for fluvial facies sand body connectivity based on feature factor control according to claim 1 or 2, characterized in that: In S5, the prediction result is determined by the average value of the classification results output by the last basic classifier. The connectivity probability of each sampling point is accurately calculated, and the average value of the classification results is used as the connectivity probability attribute to quantitatively predict the connectivity of the sand body.

Images