Shale reservoir classification evaluation method based on machine learning and optimal fractal attribute

Abstract

The invention discloses a shale reservoir classification evaluation method based on machine learning and optimal fractal attributes, and relates to the technical field of shale reservoir evaluation. Comprising the following steps: calculating fractal attribute indexes of a T2 spectrum, obtaining a new clustering label, obtaining optimal fractal attribute values, obtaining a final classification label, applying a final classification basis to all optimal fractal attribute values of a target well, and obtaining a new classification result; multi-fractal calculation is carried out on T2 spectrum original data obtained through nuclear magnetic logging, effective porosity data obtained through combination of corresponding well section closed coring data is taken as the basis, after effective porosity of different degrees is classified, systematic machine learning training is carried out on fractal attributes corresponding to each category, and fractal fractal data are obtained. And a set of classification interval of fractal attribute values mainly based on the fractal attribute with the maximum influence degree is obtained and is applied to all fractal attribute values of the target interval, so that the shale reservoir can be accurately classified and evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of shale reservoir evaluation, and in particular to a shale reservoir classification evaluation method based on machine learning and optimal fractal attributes. Background Art

[0002] Shale reservoirs refer to rock layers that are mainly composed of fine-grained sediments (such as clay, mudstone and siltstone). They are rich in organic matter and can generate and store natural gas or oil. The classification and evaluation of shale reservoirs is a systematic analysis and classification based on factors such as the geological characteristics, physical properties and gas content of shale gas reservoirs. This process aims to identify and distinguish different types of shale reservoirs, and can provide a scientific basis for the exploration and development of shale gas, so as to make more effective resource assessment and development decisions.

[0003] Nuclear magnetic resonance logging technology has unique advantages in the evaluation of complex reservoirs. It can be used to evaluate the saturation of the study area. Nuclear magnetic resonance logging measures the relaxation signals of hydrogen atoms in the formation and can provide pore and fluid information that is independent of the lithology of the formation. A set of data obtained by measuring the distribution of the transverse relaxation time (T2) of the fluid in the rock pores is called the T2 spectrum. This data reflects the detailed information of the rock pore structure, including pore size, pore fluid type (oil, gas, water) and their saturation. The T2 spectrum is often used to identify and evaluate the pore structure of the reservoir, including pore size distribution, pore fluid type and saturation. By analyzing the T2 spectrum, different fluids such as oil, gas and water can be distinguished because they exhibit different relaxation characteristics on the T2 spectrum. Therefore, nuclear magnetic resonance logging technology plays an important role in reservoir classification and evaluation.

[0004] Since the T2 spectrum distribution obtained by nuclear magnetic resonance logging is closely related to the size and distribution of rock pores, and is closely related to lithology, pore fluid properties, wettability, and formation water salinity, the positions of oil peaks in the T2 spectra of different wells may be different, and the T2 cutoff values ​​may not be the same. For low permeability or tight reservoirs, the presence of (light) oil will cause the nuclear magnetic resonance T2 spectrum distribution to become wider, so it is difficult for some wells to separate bound fluid from movable fluid on the T2 spectrum, resulting in a certain deviation between the saturation calculated by nuclear magnetic resonance logging and the actual situation, ultimately affecting the accuracy of reservoir evaluation and classification. Therefore, the present invention proposes a shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes to solve the problems existing in the prior art. Summary of the invention

[0005] In response to the above problems, the purpose of the present invention is to propose a shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes, so as to solve the problem that the saturation calculated by nuclear magnetic resonance logging in the existing shale reservoir classification and evaluation method is different from the actual situation, resulting in the inability to accurately classify and evaluate shale reservoirs.

[0006] In order to achieve the purpose of the present invention, the present invention is implemented by the following technical scheme: a shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes, comprising the following steps:

[0007] Step 1: Use nuclear magnetic resonance logging technology to obtain the original data of the target well depth T2 spectrum and perform parallel analysis and processing, then calculate the relevant fractal attribute values ​​of the analyzed and processed data to obtain the fractal attribute index of each group of T2 spectra;

[0008] Step 2: Obtain the movable porosity data of the target well and divide it into four categories as the original classification labels, then summarize the fractal attribute indicators and the original classification labels into a data table, and then use the data table as the input data of the K-means clustering model, and use the machine learning method to cluster the fractal attribute values ​​to obtain the corresponding new cluster labels;

[0009] Step 3: Use the random forest method to analyze the importance of fractal attribute indicators, and then use the fractal attribute with the highest importance in the analysis results as the final classification standard, that is, the optimal fractal attribute value;

[0010] Step 4: According to the correspondence between the original classification labels and the new clustering labels, the proportion of each type in each clustering label and the proportion of each type in all clustering labels are calculated respectively, and the mapping relationship between the clustering labels and the original classification labels is obtained through comprehensive analysis. The final label of each group of data is obtained based on the mapping relationship;

[0011] Step 5: Based on the final labels and the optimal fractal attribute values, the intervals of the optimal fractal attribute values ​​of the four types of labels are obtained, and this interval is used as the final classification basis and applied to all the optimal fractal attribute values ​​of the target well to obtain a new shale reservoir classification evaluation result based mainly on the optimal fractal attribute.

[0012] A further improvement is that in step 1, the specific step of analyzing and processing the original data of the target well depth T2 spectrum is: dimensionally processing the original T2 spectrum data through a dimensionless formula so that all the data converges within the interval [0,1].

[0013] A further improvement is that in step 1, the relevant fractal attribute values ​​of the analyzed and processed data are calculated using a fractal dimension calculation method, and the fractal attribute indicators of the T2 spectrum include the singular intensity, offset, and maximum and minimum offset values ​​of the T2 spectrum.

[0014] A further improvement is that in the step 2, the movable porosity data is obtained from the closed coring data of the target well section, and the movable porosity data is divided into four categories from large to small.

[0015] A further improvement is that in step 2, the summarized data table is preprocessed before being input into the K-means clustering model. The specific steps are: first, data cleaning is used to remove missing values ​​and outliers in the data table, then correlation analysis and dimensionality reduction are performed on the data table, and then the data table is standardized in a normalized manner.

[0016] A further improvement is that in step three, the importance of the fractal attribute values ​​is analyzed by the random forest method to obtain the influence ranking of the three fractal attribute values ​​on the clustering process, the influence values ​​of the three fractal attributes and the histogram.

[0017] Further improvement lies in: in the step four, the two-way maximization method is used to calculate the proportion of each type in each cluster label and the proportion of each type in all cluster labels. The specific steps are: based on the analysis of the differences and connections between cluster labels and original classification labels, the two-way maximization method is used to obtain the proportion of each original label sample in each cluster label, as well as the proportion of each original label type in all cluster labels. A comprehensive analysis is performed to obtain the mapping relationship between cluster labels and original classification labels. According to the mapping relationship, the clustered data is reclassified to obtain the final classification label.

[0018] A further improvement is that the mapping relationship is: {clustering label 0: original label 1, clustering label 1: original label 2, clustering label 2: original label 4, clustering label 3: original label 3}.

[0019] The beneficial effects of the present invention are as follows: the present invention performs multiple fractal calculations on the original T2 spectrum data obtained by nuclear magnetic resonance logging to obtain indicators such as the singular spectrum function and spectrum offset of the T2 spectrum data, and uses the effective porosity data obtained from the closed coring data of the corresponding well section as a basis, and uses the Kmeans clustering method and machine learning methods such as random forests to classify effective porosities of different degrees, and then performs systematic machine learning training on the fractal attributes corresponding to each category, and finally obtains a set of classification intervals of fractal attribute values ​​based mainly on the fractal attributes with the greatest influence, and applies them to all fractal attribute values ​​of the target layer section, to obtain a new set of classification results, which can accurately classify and evaluate shale reservoirs, and are more detailed and more accurate than traditional classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a flow chart of the shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes of the present invention;

[0022] Figure 2 Schematic diagram of clustering results of three fractal attribute values ​​in Embodiment 3 of the present invention;

[0023] Figure 3 Schematic diagram of the correspondence between cluster labels and original classification labels in Embodiment 3 of the present invention;

[0024] Figure 4 is a schematic diagram of the importance ranking result in the third embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of the final classification label obtained by the mapping relationship in the third embodiment of the present invention;

[0026] Figure 6 It is a schematic diagram comparing the original classification result in the third embodiment of the present invention and the classification result of the method of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Embodiment 1

[0029] See also Figure 1 This embodiment provides a shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes, comprising the following steps:

[0030] Step 1: Parallel analysis and processing of the original data of the target well depth T2 spectrum, and calculation of the fractal attribute index of the T2 spectrum

[0031] First, the original data of T2 spectrum of the target well depth is obtained by using nuclear magnetic resonance logging technology, and then analyzed and processed in parallel. The original data of T2 spectrum is dimensionless processed by dimensionless formula, so that all data converge within the interval of [0,1] to eliminate the influence of dimension on data and improve the performance of subsequent model processing. Then, the relevant fractal attribute values ​​of the analyzed data are calculated by the calculation method of fractal dimension, and the fractal attribute indicators of each group of T2 spectrum are obtained, including the singular intensity, offset and maximum and minimum offset values ​​of T2 spectrum.

[0032] Step 2: Use machine learning methods to cluster fractal attribute values ​​and obtain new cluster labels

[0033] The movable porosity data is obtained from the closed coring data of the target well, and the movable porosity data of the target well is obtained and divided into four categories from large to small as the original classification labels, and then the fractal attribute index calculated in step 1 and the original classification labels are summarized into a data table, and then the summarized data table is used as the input data of the K-means clustering model, and the fractal attribute values ​​are clustered using a machine learning method to obtain corresponding new clustering labels. Analogously, the movable porosity is divided into four categories from large to small in the closed coring data. In this embodiment, the clustering labels are also divided into four categories, which is convenient for finding the corresponding relationship in the next step. After the clustering is completed, four clustering results with obvious differences are obtained;

[0034] In this embodiment, the summarized data table is first preprocessed before being input into the K-means clustering model. The specific steps are: first, the missing values ​​in the data table are removed by data cleaning (the missing values ​​in the data table are checked, and filling or deletion is selected according to the specific situation) and outliers (outliers are identified and processed to prevent them from having an adverse effect on the clustering results), and then the data table is subjected to correlation analysis (the correlation between each feature and the fractal attribute value is analyzed, and the most relevant feature is selected) and dimensionality reduction processing (if the feature dimension is high, the principal component analysis method is used for dimensionality reduction to reduce the amount of calculation and improve the model performance), and then the data table is standardized by normalization (the data is normalized so that each feature is in the same dimension to avoid some features having too much impact on the clustering results due to dimensional differences) to ensure the quality of the data and the effectiveness of the model;

[0035] Step 3: Analyze the importance of the fractal attribute values, and take the fractal attribute with the highest importance as the optimal fractal attribute value

[0036] Use the random forest method to analyze the importance of the fractal attribute values ​​obtained in step 1, obtain the ranking of the influence of the three fractal attribute values ​​on the clustering process, the influence values ​​of the three fractal attributes and the histogram, and then use the fractal attribute with the highest importance in the analysis results as the final classification standard, that is, the optimal fractal attribute value, as the main basis for the subsequent final classification;

[0037] Step 4: Get the final classification label based on the mapping relationship between the cluster label and the original classification label

[0038] According to the correspondence between the original classification labels and the obtained new clustering labels in step 2, and based on the analysis of the differences and connections between the clustering labels and the original classification labels, the bidirectional maximization method is used to obtain the proportion of each original label sample in each clustering label, as well as the proportion of each original label type in all clustering labels. The mapping relationship between the clustering labels and the original classification labels is obtained through comprehensive analysis. According to the mapping relationship, the clustered data is reclassified to obtain the final classification label, i.e., the final label;

[0039] Step 5: Apply the final classification basis to all the optimal fractal attribute values ​​of the target well to obtain a new classification result

[0040] According to the final labels obtained by mapping in step 4 and the optimal fractal attribute values ​​obtained in step 3, the intervals of the optimal fractal attribute values ​​of the four types of labels are obtained, and this is used as the final classification basis and applied to all the optimal fractal attribute values ​​of the target well. Finally, a new shale reservoir classification and evaluation result based on the optimal fractal attribute is obtained, realizing the shale reservoir classification and evaluation based on machine learning and optimal fractal attributes.

[0041] Embodiment 2

[0042] The data used in this example comes from the Jimusar shale reservoir. The specific steps for classifying and evaluating the shale reservoir are as follows:

[0043] S1. Arrange and select the nuclear magnetic resonance logging data and closed coring data of the target well, calculate the fractal dimension of the nuclear magnetic resonance T2 spectrum data, and obtain the singular intensity, offset, and maximum and minimum offset values ​​of the T2 spectrum;

[0044] S2. Combine the movable porosity data obtained from the closed coring and conduct a joint analysis. The movable porosity of the closed coring is divided into four categories from large to small according to the preset standard as the original classification labels;

[0045] For the fractal attribute values ​​obtained from the T2 spectrum, they are first dimensionally processed so that all data converge within the interval of [0,1], eliminating the impact of the dimension on the data and improving the performance of subsequent model processing. Then, the three main fractal attribute values ​​of the T2 spectrum are calculated using the multifractal calculation method. The three fractal attribute values ​​are clustered using the Kmeans method. Analogously, the movable porosity in the closed coring data is divided into four categories from large to small according to the preset standard. Here, the cluster labels are also divided into four categories to facilitate the next step of finding the corresponding relationship. After the clustering is completed, four distinct clustering results are obtained.

[0046] S3. Use the random forest method to rank the influence of the three fractal attribute values ​​on the clustering process, obtain the influence values ​​and histograms of the three fractal attributes, and record the fractal attribute with the greatest influence as the optimal fractal attribute as the main basis for the subsequent final classification. After the entire processing process is completed, the number of original classified samples corresponding to each clustering label is obtained, for example, how many samples with the original label of 1 are there under the label of clustering 0, how many samples with the original label of 2 are there, and so on;

[0047] S4. According to the difference and connection between cluster labels and original classification labels, the bidirectional maximization method is used to obtain the proportion of each original label sample in each cluster label, as well as the proportion of each original label type in all cluster labels. The mapping relationship between cluster labels and original classification labels is obtained by comprehensive analysis: {cluster label 0: original label 1, cluster label 1: original label 2, cluster label 2: original label 4, cluster label 3: original label 3};

[0048] According to the above mapping relationship, the clustered data is reclassified to obtain the final classification label;

[0049] S5. According to the optimal fractal attribute value corresponding to the final classification label, the interval of the optimal fractal attribute value under each final label is obtained, and this is used as the final classification standard of the entire classification model;

[0050] Since the data used for model training is only the part corresponding to the movable porosity of closed coring, rather than all T2 data, this embodiment applies the final classification standard obtained above to the optimal fractal attribute values ​​obtained from all T2 data, and divides all the layers with measured T2 data into four new categories. Compared with the traditional reservoir classification and evaluation method, the interpretation based on machine learning and optimal multi-fractal attributes is more detailed and the results obtained are more reasonable.

[0051] Embodiment 3

[0052] See also Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 The specific steps of the shale reservoir classification evaluation in this embodiment are as follows:

[0053] A1. First, the obtained T2 raw data is organized into a standard data file, and then dimensionless processing is performed on it using a Python script with a dimensionless formula, and the obtained results are organized into a standard data format for fractal calculation;

[0054] A2. Import the sorted data into Matlab script to calculate the fractal properties and obtain the corresponding three fractal property values ​​of Δα, B and Δf. Meanwhile, prepare the movable porosity data obtained from the closed coring data and divide it into four categories from large to small according to certain standards as the original category labels;

[0055] A3. The T2 fractal attribute values ​​and original classification labels of the movable porosity corresponding to the depth are aggregated together as the raw data for the subsequent machine learning method. The aggregated data is imported into the Python script for Kmeans clustering operation. While obtaining the clustering results, the random forest method is used to rank the importance of the fractal attribute values. The ranking results are as follows: Figure 4 As shown in the figure, after running the script, we can finally get the distribution of the original label samples corresponding to each cluster label and the importance ranking of the three fractal attributes. The clustering results of the three fractal attribute values ​​are shown in Figure 2 As shown;

[0056] A4. The fractal attribute value with the greatest importance is taken as the optimal fractal attribute and the main indicator for subsequent classification. The distribution of the original label samples corresponding to the cluster labels is subjected to a two-way maximization analysis. That is, the proportion of samples with different original labels under each cluster label and the proportion of different cluster labels corresponding to each original sample label are calculated respectively. All proportion distributions are comprehensively considered to select the most reasonable one-to-one correspondence as the mapping relationship between the cluster label and the final classification label. According to the mapping relationship, the clustering results are reclassified and the reclassified results are taken as the final classification results. The corresponding relationship between the cluster label and the original classification label is as follows: Figure 3 As shown, the final classification label obtained according to the mapping relationship is as follows Figure 5 As shown;

[0057] A5. According to the distribution interval of the optimal fractal attribute corresponding to the final classification label, it is applied to all T2 main fractal attributes of the target well to obtain the final classification result of the target well. The final classification result is introduced into the well logging curve. Compared with the original classification result, it is found that the classification result obtained by the classification model is more sufficient and reasonable for explaining the oil test result. The comparison diagram of the original classification result and the classification result of the present invention is shown in FIG. Figure 6 As shown, Figure 6 Track 1 is the original classification result, track 2 is the classification result of the present invention, and track 3 is the oil test interpretation result.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes, characterized in that: The following steps are involved: Step 1: Use nuclear magnetic resonance logging technology to obtain the original data of the target well depth T2 spectrum and perform parallel analysis and processing, then calculate the relevant fractal attribute values ​​of the analyzed and processed data to obtain the fractal attribute index of each group of T2 spectra; Step 2: Obtain the movable porosity data of the target well and divide it into four categories as the original classification labels, then summarize the fractal attribute indicators and the original classification labels into a data table, and then use the data table as the input data of the K-means clustering model, and use the machine learning method to cluster the fractal attribute values ​​to obtain the corresponding new cluster labels; Step 3: Use the random forest method to analyze the importance of fractal attribute indicators, and then use the fractal attribute with the highest importance in the analysis results as the final classification standard, that is, the optimal fractal attribute value; Step 4: According to the correspondence between the original classification labels and the new clustering labels, the proportion of each type in each clustering label and the proportion of each type in all clustering labels are calculated respectively, and the mapping relationship between the clustering labels and the original classification labels is obtained through comprehensive analysis. The final label of each group of data is obtained based on the mapping relationship; Step 5: Based on the final labels and the optimal fractal attribute values, the intervals of the optimal fractal attribute values ​​of the four types of labels are obtained, and this interval is used as the final classification basis and applied to all the optimal fractal attribute values ​​of the target well to obtain a new shale reservoir classification evaluation result based mainly on the optimal fractal attribute.

2. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1, characterized in that: In the step 1, the specific steps of analyzing and processing the original data of the target well depth T2 spectrum are: dimensionally processing the original T2 spectrum data through a dimensionless formula so that all the data converge within the interval [0,1].

3. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1, characterized in that: In the step 1, the relevant fractal attribute values ​​of the analyzed data are calculated using a calculation method of fractal dimension, and the fractal attribute indicators of the T2 spectrum include the singular intensity, offset, and maximum and minimum offset values ​​of the T2 spectrum.

4. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1, characterized in that: In the step 2, the movable porosity data is obtained from the closed coring data of the target well section, and the movable porosity data is divided into four categories from large to small.

5. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1, characterized in that: In the step 2, the summarized data table is preprocessed before being input into the K-means clustering model. The specific steps are: first, the missing values ​​and outliers in the data table are removed by data cleaning, then the data table is subjected to correlation analysis and dimensionality reduction, and then the data table is standardized in a normalized manner.

6. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1, characterized in that: In the step three, the importance of the fractal attribute values ​​is analyzed by the random forest method to obtain the influence ranking of the three fractal attribute values ​​on the clustering process, the influence values ​​of the three fractal attributes and the histogram.

7. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 1 is characterized by: In the step four, the two-way maximization method is used to calculate the proportion of each type in each cluster label and the proportion of each type in all cluster labels. The specific steps are: according to the analysis of the difference and connection between the cluster label and the original classification label, the two-way maximization method is used to obtain the proportion of each original label sample in each cluster label and the proportion of each original label type in all cluster labels, and the mapping relationship between the cluster label and the original classification label is obtained by comprehensive analysis. According to the mapping relationship, the clustered data is reclassified to obtain the final classification label.

8. The shale reservoir classification and evaluation method based on machine learning and optimal fractal attributes according to claim 7 is characterized by: The mapping relationship is: {clustering label 0: original label 1, clustering label 1: original label 2, clustering label 2: original label 4, clustering label 3: original label 3}.

Images