Multi-source data body fused coal reservoir fracture prediction model construction method

Abstract

The invention relates to a multi-source data volume fused coal reservoir fracture prediction model construction method, which comprises the following steps of: acquiring a coal reservoir coal sample of a well in a certain research area, and adopting a higher fracture strength value in core fracture strength and imaging logging information for a fracture surface strength non-overlapping part obtained by artificially describing the core fracture strength and the imaging logging information; generating a coal sample crack strength curve; establishing a fractal dimension curve; carrying out full-diameter CT (Computed Tomography) scanning on the coal sample, and carrying out quantitative calculation on the single-layer fracture rate to obtain a single-layer fracture rate curve; fusing the obtained three types of curves, and establishing a comprehensive indication curve; constructing a fault distance attribute body; extracting a seismic data body, simulating the correlation between the seismic data body and the comprehensive indication curve, and selecting a sensitive seismic body; constructing a crack comprehensive discrimination function body; and a coal reservoir fracture prediction model is constructed, and coal reservoir fracture prediction is carried out. According to the method, coring data, conventional logging, imaging logging, CT scanning and earthquake crack prediction data are integrated, and the prediction precision of the coal reservoir crack is improved.

Description

Technical field:

[0001] The invention relates to the technical field of coalbed methane development, and in particular to a method for constructing a coal reservoir fracture prediction model by fusing multi-source data. Background technology:

[0002] Fractures are the main seepage channels of coalbed methane reservoirs. Accurate identification and evaluation of fractures is one of the important contents in the exploration and development process. Fracture reservoirs are heterogeneous and anisotropic, and conventional logging has limited vertical resolution and cannot provide high-resolution stratigraphic information. At the same time, due to the high cost and limited data of core and imaging logging, it is difficult to objectively and truly characterize the distribution of fractures in a region by relying solely on core and imaging logging data. At present, seismic data used to evaluate the spatial distribution of fractures is more, but it has multiple solutions and uncertain results, which may lead to misjudgment. Therefore, finding a suitable method to accurately predict the spatial distribution of fractures in coal reservoirs is crucial to the development of coalbed methane reservoirs.

[0003] Patent document CN113534247A discloses a method and device for quantitatively predicting fractures based on post-stack seismic data. It only predicts fractures from a seismic perspective, has poor prediction accuracy for some small-scale fractures, and the prediction model itself has uncertainty. Summary of the invention:

[0004] The purpose of the present invention is to provide a method for constructing a coal reservoir fracture model by fusing multi-source data volumes, which is used to solve the problem in the prior art that the prediction accuracy cannot be ensured when predicting coal reservoir fractures.

[0005] The technical solution adopted by the present invention to solve the technical problem is: the method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes comprises the following steps:

[0006] Step 1: Obtain a coal sample from a coal reservoir in a well in a certain study area, record the number of fractures on the end face of the coal sample, fracture inclination, fracture length, fracture filling degree and fracture properties, calculate the artificial description of the core fracture strength, and also generate the fracture surface strength using the imaging logging data of the well section. The non-overlapping part of the artificial description of the core fracture strength and the fracture surface strength obtained by the imaging logging data is taken as the higher fracture strength value of the two to generate the coal sample fracture strength curve;

[0007] Step 2: Considering the sensitivity of conventional logging to fractures in the formation, AC, DEN, GR, RD, RS, and SP curves are used, and the Hurst index is introduced by R / S analysis method to establish a fractal dimension curve;

[0008] Step 3: Perform full-diameter CT scanning on the coal sample to obtain core fracture image slices, perform fracture identification and segmentation, perform quantitative calculation of single-layer fracture rate on the segmented fracture model, and obtain a single-layer fracture rate curve;

[0009] Step 4: Divide the coal sample crack intensity curve, fractal dimension curve and single-layer crack rate curve at equal intervals, normalize the curve values, use the entropy method to merge the three types of curves, and establish a comprehensive indicator curve;

[0010] Step 5: Use seismic data to directly identify medium- and large-scale fractures in the profile through the sudden change of reflection event axis, the offset of wave group, the offset of wave system, and the sudden change of reflection layer occurrence. Use Petrel to perform deterministic modeling on these medium- and large-scale fractures, and use the Geopmetrical modeling module to construct the fault distance attribute body;

[0011] Step 6: Obtain the seismic body and its interpretation data in the study area, interpret the attributes of the seismic body, use relevant algorithms to extract the seismic data body, which includes chaos body, variance body, coherence body, structural curvature body, instantaneous amplitude, dip deviation, and ant body attributes, and use the Co-Krigng method to simulate the correlation between the above seismic data body and the comprehensive indication curve to select sensitive seismic bodies;

[0012] Step 7: Using the selected sensitive earthquake body and the fault distance attribute body, an artificial neural network is used to construct a comprehensive fracture discrimination function body;

[0013] Step 8: Based on the constructed comprehensive fracture discriminant function body, the FractureNetwork module in Petrel is used to construct a coal reservoir fracture prediction model to predict coal reservoir fractures.

[0014] Step one of the above scheme is specifically as follows: according to the actual situation of the coring samples, samples are taken in order at intervals of about 0.2m. In the coal sample section, the core is first returned to its original position, the mud on the surface of part of the sample is brushed off to expose the fresh surface, and the number of cracks on the end face of the coal sample, the crack inclination, the crack length, the degree of crack filling and the crack properties are recorded to obtain an artificial description of the core crack strength and generate an artificial description of the core crack strength curve. In addition, the imaging logging data of the well section is used to generate the fracture surface strength using CIFlog software, and a fracture surface strength curve is generated. The two curves are placed together, and the non-overlapping part of the two curves uses the higher fracture strength value of the two, and they are connected to generate a coal sample fracture strength curve.

[0015] In the above scheme, the artificial description of the core fracture strength and the fracture surface strength generated by CIFlog are both based on the formula f = N / (π*r 2 )(1)

[0016] Where, f is the core strength of artificial fractures, bars / m2 ; N is the number of cracks within the core cross section, r is the core radius, m.

[0017] In step 2 of the above scheme, the R / S analysis method is used to introduce the Hurst index, and the method for establishing the fractal dimension curve is as follows:

[0018] There is a unique conventional logging value for different logging depths, Z = Z(1), Z(2), ... Z(n), the logging sequence is recorded as Z(i) and Z(j), i, j = 1, 2, ... n, the full range R(n) and standard deviation S(n) of the logging sequence are expressed by the following expressions:

[0019]

[0020] When the logging sequence satisfies statistical self-similarity, there exists

[0021]

[0022] lg[R(n) / S(n)]=lgC+H·lgn (5)

[0023] D=2-H(6)

[0024] Where Z is the logging sequence; R is the full range of the logging sequence; S is the full standard deviation of the logging sequence; n is the number of logging sequences; u is the number of scales increasing from 0 to n; C is a constant; h is the delay distance; H is the Hurst exponent; and D is the fractal dimension.

[0025] Quantitative calculation method of single-layer crack rate in step 3 of the above scheme:

[0026] Since the X-ray CT scanning imaging analysis of the internal structure of coal samples is achieved through the difference in relative density, the unfilled microcracks in the coal samples belong to the part with the lowest density, and the microcracks appear in the CT image as a dark gray color that is much darker than the grayscale values ​​of the mirror coal and bright coal. The watershed algorithm is used in Avizo to identify the grayscale boundaries between the mirror coal, bright coal and microcracks, and the volume of 500,000 pixels is used as the threshold for microcrack segmentation. The microcracks in the denoised and sharpened samples are segmented and extracted. After the crack extraction is completed, the volume fraction command in Avizo is used to calculate the ratio of the extracted crack voxels to the set background voxels to obtain the crack rate of the sample.

[0027]

[0028] Where F represents the single-layer fracture rate in the core section, S f Represents the grayscale image area of ​​a single layer of cracks, m 2 ; r is the core radius, m, the higher the F value, the higher the proportion of fractures.

[0029] The method of establishing a comprehensive indication curve using the entropy method in step 4 of step 4 of the above scheme is as follows:

[0030] (1) Sample data normalization

[0031] Normalize the data of fracture intensity curve, fractal dimension curve and core single layer fracture rate curve;

[0032] (2) Normalized data volume translation

[0033] In the entropy weight method, when x mn When =0, the entropy value is meaningless, and all normalized data are shifted to the right by 0.0001 units;

[0034] (3) Entropy weight calculation

[0035] After data preprocessing is completed, the entropy value and weight of each curve parameter are calculated, and each curve value is multiplied by its weight, and then normalized to obtain a comprehensive indicator curve;

[0036] The entropy calculation formula is:

[0037]

[0038] Q=α·K f +β·K D +λ·K F (11)

[0039] In the formula, Q b is the information entropy value of each ion, ε b is the weight, m ​​is the number of samples, n is the number of indicators, α is the fracture intensity weight, β is the fractal dimension weight, λ is the core fracture rate weight, K f is the crack strength value, K D is the fractal dimension value, K F is the core fracture rate value.

[0040] Step 5 of the above scheme is as follows: according to the deterministic modeling of the faults in the study area, the seismic data is used to directly identify medium- and large-scale fractures in the profile through the sudden changes of reflection phase axes, the offset of wave groups and wave systems, and the sudden changes of reflection layer attitude, and the size of the fault is analyzed. The deterministic modeling of this part is carried out using Petrel, and the scope of the fault influence is delineated according to experience, the seismic fracture attribute characteristics are cut, and the geometrical modeling module is used to construct the fault distance attribute body.

[0041] The specific method of step 7 in the above scheme is: the artificial neural network adopts supervised machine learning, the training supervision points are pre-set, the training supervision points include crack sample points and non-crack sample points, the learning samples include sensitive data bodies, comprehensive indicator curves and crack training sets, and the artificial neural network assigns values ​​to the training supervision points, extrapolates the probability of crack development in the three-dimensional direction, and finally constructs a comprehensive discriminant.

[0042] Beneficial effects:

[0043] 1. The present invention integrates coring data, conventional logging, imaging logging, CT scanning and seismic prediction of fracture data, and adopts mathematical methods to analyze and integrate them, so as to solve or partially solve the problem that the prediction accuracy cannot be guaranteed in the existing technology when predicting coal reservoir fractures.

[0044] 2. The present invention utilizes artificial coring description and imaging logging to extract fracture strength, uses R / S analysis to identify conventional logging fractals, uses CT scanning to extract single-layer fracture rate, uses entropy method to fuse comprehensive curves, and finally combines sensitive seismic bodies and fault distance attribute bodies with artificial neural networks. With the help of Petrel, a fracture model is constructed, which can effectively fuse multiple types of data bodies, improve the prediction accuracy of fractures, and provide a basis for actual production.

[0045] 3. The present invention analyzes from the perspective of seismic and well logging, selects seismic attribute bodies, screens out sensitive attribute bodies, and has the characteristics of accurate prediction through comprehensive judgment of artificial, conventional well logging, and CT scanning for the predicted fracture area.

[0046] 4. The present invention utilizes the description of artificial fractures in gas well coring and combines it with array acoustic logging to construct a fracture intensity curve; considering the sensitivity of conventional logging to fractures in the formation, the R / S analysis method is used to introduce the Hurst index and establish a fractal dimension curve; based on full-diameter CT scanning, the grayscale image area of ​​a single-layer fracture is effectively identified, and a single-layer fracture rate curve is established; the numerical values ​​of each curve are normalized, and the entropy method is used to fuse three types of curves to establish a comprehensive indicator curve; using seismic interpretation data, deterministic modeling is adopted for medium and large-scale fractures, and a fault distance data body is established; Co-Krigng simulation is used to extract multiple types of sensitive seismic attribute bodies, and a comprehensive discriminant function body is constructed with the help of artificial neural networks and comprehensive indicator curves, a coal reservoir fracture prediction model is established, and multiple types of data bodies are comprehensively considered to improve the prediction accuracy of coal reservoir fractures. Description of the drawings:

[0047] Figure 1 is a flow chart of the present invention;

[0048] Figure 2 It is a fault fracture data volume map of the present invention;

[0049] Figure 3This is a diagram of the calculation process of the single-layer crack rate of full-diameter CT scanning in the present invention;

[0050] Figure 4 This is a coal reservoir fracture prediction model diagram in the present invention;

[0051] Figure 5 Graph of the method used to generate the crack intensity curve in step 1. Specific implementation method:

[0052] The present invention will be further described below in conjunction with the accompanying drawings:

[0053] Combination Figure 1-Figure 5 As shown, the method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes comprises:

[0054] Step 1: According to the actual situation of the coring samples, samples are taken in order at intervals of about 0.2m. In the cross section of the cored coal sample, the core is first returned to its original position, the mud on the surface of some samples is brushed to expose the fresh surface, and the number of cracks on the end face of the coal sample, the crack inclination, length, degree of crack filling and crack properties and other parameters are recorded to calculate the crack strength. Some sections of the well have imaging logging data. Combined with CIFlog software, the crack surface strength is generated. The overlapping part of the two uses the higher crack strength value to generate the coal sample fracture strength curve. That is, the fracture strength value is obtained and the fracture strength curve is generated through artificial core description and imaging logging interpretation respectively, and the two curves are placed together. The non-overlapping part of the two curves uses the higher fracture strength value of the two, and they are connected to generate the final fracture strength curve. Figure 5 As shown, the final crack strength curve is the curve formed by the thick solid line.

[0055] The formulas for manually describing the core fracture strength and the fracture surface strength generated by CIFlog are

[0056] f=N / (π*r 2 )(1)

[0057] Where, f is the core strength of artificial fractures, bars / m 2 ; N is the number of cracks within the core cross section, r is the core radius, m.

[0058] Step 2: Considering the sensitivity of conventional logging to fractures in the formation, AC, DEN, GR, RD, RS, and SP curves are mainly used. The Hurst index is introduced by R / S analysis method to establish the fractal dimension curve.

[0059] There is a unique conventional logging value for different logging depths, Z = Z(1), Z(2), ... Z(n). The sequence can be recorded as Z(i) and Z(j), i, j = 1, 2, ... n. The full range R(n) and standard deviation S(n) of the sequence are expressed by the following expressions:

[0060]

[0061] When the sequence satisfies statistical self-similarity, there exists

[0062]

[0063] lg[R(n) / S(n)]=lgC+H·lgn (5)

[0064] D=2-H(6)

[0065] Where Z is the logging sequence; R is the full range of the logging sequence; S is the full standard deviation of the logging sequence; n is the number of logging sequences; u is the number of scales increasing from 0 to n; C is a constant; h is the delay distance; H is the Hurst exponent; and D is the fractal dimension.

[0066] For all the logging data used, starting from the third sampling point, the R(n) / S(n) of each group of logging data points is obtained using equations (2) to (5). When the horizontal and vertical axes are set to logarithmic scales, the linear fit between R(n) / S(n) and N scattered points is very high, and its slope is the index of the logging sequence. Its fractal dimension D can be calculated by equation (6).

[0067] Step 3: Perform a full-diameter CT scan on the core of the coal sample to obtain core fracture image slices, use image processing methods such as threshold segmentation to identify and segment fractures, and perform quantitative calculation of the single-layer fracture rate of the segmented fracture model.

[0068] The single-layer crack rate, extraction and calculation process is as follows:

[0069] Since the X-ray CT scanning imaging analysis of the internal structure of the sample is mainly achieved through its relative density difference, and the unfilled microcracks in the coal are the lowest density part, they appear in the CT image as a dark gray color that is much darker than the grayscale value of the mirror coal and bright coal. In Avizo, the 1500 photos generated by the scan are reconstructed, and the grayscale boundaries between the mirror coal, bright coal and the cracks are identified using the watershed algorithm. The volume of 500,000 pixels is used as the threshold for microcrack segmentation, and the microcracks in the denoised and sharpened samples are segmented and extracted. After the crack extraction is completed, the volumefraction command in Avizo can be used to calculate the ratio of the crack voxels extracted from the 1500 photos to the set background voxels to obtain the crack rate of the sample.

[0070]

[0071] In the formula, F represents the single-layer fracture rate in the core section, S fRepresents the grayscale image area of ​​a single layer of cracks, m 2 ; r is the core radius, m, the higher the F value, the higher the proportion of fractures.

[0072] Step 4: Divide each curve into equal intervals, normalize the curve values, use the entropy method to fuse the three types of curves, and establish a comprehensive indicator curve.

[0073] The specific measures for entropy method and comprehensive indicator curve calculation are as follows:

[0074] (1) Sample data normalization

[0075] The data of the fracture intensity curve, fractal dimension curve, and core single-layer fracture rate curve are normalized so that all indicators in the sample are in the same dimension to eliminate the analysis errors caused by large numerical differences between the indicators.

[0076] (2) Normalized data volume translation

[0077] In the entropy weight method, when x mn When =0, the entropy value is meaningless. Based on this, all normalized data are shifted to the right by 0.0001 units.

[0078] (3) Entropy weight calculation

[0079] After data preprocessing is completed, the entropy value and weight of each curve parameter can be calculated, and the obtained curve value is multiplied by its obtained weight, and then normalized to obtain a comprehensive indicator curve.

[0080] The entropy calculation formula is:

[0081]

[0082] Q=α·K f +β·K D +λ·K F (11)

[0083] In the formula, Q b is the information entropy value of each ion, ε b is the weight, m ​​is the number of samples, n is the number of indicators, α, β, λ are the weight of fracture intensity, fractal dimension, and core fracture rate, K f , K D , K F It is the fracture intensity value, fractal dimension value and core fracture rate value.

[0084] Step 5: From the perspective of seismic interpretation, seismic data is used to directly identify medium- and large-scale fractures in the profile through the sudden change of reflection phase axis, the offset of wave group, the offset of wave system, and the sudden change of reflection layer. Petrel is used to perform deterministic modeling on these medium- and large-scale fractures, and the Geopmetrical modeling module is used to construct the fault distance attribute body. This data body is mainly used to ensure the development of small fractures near medium- and large faults, so that the final constructed model conforms to the actual geological characteristics.

[0085] That is, the fault distance attribute body is constructed based on the deterministic modeling of the faults in the study area, the fault size is analyzed, the possible impact range of the fault is delineated according to experience, the earthquake crack attribute characteristics are cut, and the fault distance attribute body is constructed.

[0086] Step 6: Interpret the attributes of the seismic body, use relevant algorithms to extract the chaotic body, variance body, coherence body, structural curvature body, instantaneous amplitude, dip deviation, and ant body attributes, use the Co-Krigng method to simulate the correlation between the above seismic data body and the comprehensive indication curve, and select sensitive seismic bodies to proceed to step 7;

[0087] Step 7: Select sensitive earthquake bodies and use artificial neural networks in combination with fault distance attribute bodies to construct a comprehensive fracture discriminant function body; the artificial neural network specifically uses supervised machine learning, that is, the training supervision points are set in advance, the training supervision points include fracture sample points and non-fracture sample points, and the learning samples include the sensitive data body acquired in the early stage, the fracture development body (comprehensive indicator curve construction) and the fracture training set. The artificial neural network assigns values ​​to the supervision points and extrapolates the fracture development probability in the three-dimensional direction to finally construct a comprehensive discriminant body.

[0088] Step 8: Finally, based on the constructed comprehensive fracture discriminant function body, the Fracturenetwork module in Petrel is used to construct a coal reservoir fracture prediction model (i.e., a coal reservoir fracture discrete network model); and the coal reservoir fracture prediction model is used to predict coal reservoir fractures.

Claims

1. A method for constructing a coal reservoir fracture prediction model by fusion of multi-source data, characterized in that The steps include: Step 1: Obtain a coal sample from a coal reservoir in a well in a certain study area, record the number of fractures on the end face of the coal sample, fracture inclination, fracture length, fracture filling degree and fracture properties, calculate the artificial description of the core fracture strength, and also generate the fracture surface strength using the imaging logging data of the well section. The non-overlapping part of the artificial description of the core fracture strength and the fracture surface strength obtained by the imaging logging data is taken as the higher fracture strength value of the two to generate the coal sample fracture strength curve; Step 2: Considering the sensitivity of conventional logging to fractures in the formation, AC, DEN, GR, RD, RS, and SP curves are used, and the Hurst index is introduced by R / S analysis method to establish a fractal dimension curve; Step 3: Perform full-diameter CT scanning on the coal sample to obtain core fracture image slices, perform fracture identification and segmentation, perform quantitative calculation of single-layer fracture rate on the segmented fracture model, and obtain a single-layer fracture rate curve; Step 4: Divide the coal sample crack intensity curve, fractal dimension curve and single-layer crack rate curve at equal intervals, normalize the curve values, use the entropy method to merge the three types of curves, and establish a comprehensive indicator curve; Step 5: Use seismic data to directly identify medium- and large-scale fractures in the profile through the sudden change of reflection event axis, the offset of wave group, the offset of wave system, and the sudden change of reflection layer occurrence. Use Petrel to perform deterministic modeling on these medium- and large-scale fractures, and use the Geopmetrical modeling module to construct the fault distance attribute body; Step 6: Obtain the seismic body and its interpretation data in the study area, interpret the attributes of the seismic body, use relevant algorithms to extract the seismic data body, which includes chaos body, variance body, coherence body, structural curvature body, instantaneous amplitude, dip deviation, and ant body attributes, and use the Co-Krigng method to simulate the correlation between the above seismic data body and the comprehensive indication curve to select sensitive seismic bodies; Step 7: Using the selected sensitive earthquake body and the fault distance attribute body, an artificial neural network is used to construct a comprehensive fracture discrimination function body; Step 8: Based on the constructed comprehensive fracture discriminant function body, the FractureNetwork module in Petrel is used to construct a coal reservoir fracture prediction model to predict coal reservoir fractures.

2. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 1, characterized in that: The step one is specifically as follows: according to the actual situation of the coring samples, the samples are sampled in order at intervals of about 0.2m. At the cross section of the coal sample, the core is first returned to its original position, the mud on the surface of part of the sample is brushed off to expose the fresh surface, and the number of cracks on the end face of the coal sample, the crack inclination, the crack length, the degree of crack filling and the crack properties are recorded to obtain an artificial description of the core crack strength and generate an artificial description of the core crack strength curve. In addition, the imaging logging data of the well section is used to generate the fracture surface strength using CIFlog software, and a fracture surface strength curve is generated. The two curves are placed together, and the non-overlapping part of the two curves uses the higher fracture strength value of the two, and they are connected to generate a coal sample fracture strength curve.

3. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 2, characterized in that: The formulas for manually describing the core fracture strength and the fracture surface strength generated by CIFlog are: f=N / (π*r 2 (1) Where, f is the core strength of artificial fractures, bars / m 2 ; N is the number of cracks within the core cross section, r is the core radius, m.

4. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 3, characterized in that: In the step 2, the R / S analysis method is used to introduce the Hurst index, and the method for establishing the fractal dimension curve is as follows: There is a unique conventional logging value for different logging depths, Z = Z(1), Z(2), ... Z(n), the logging sequence is recorded as Z(i) and Z(j), i, j = 1, 2, ... n, the full range R(n) and standard deviation S(n) of the logging sequence are expressed by the following expressions: When the logging sequence satisfies statistical self-similarity, there exists: lg[R(n) / S(n)]=lgC+H·lgn (5) D=2-H(6) Where Z is the logging sequence; R is the full range of the logging sequence; S is the full standard deviation of the logging sequence; n is the number of logging sequences; u is the number of scales increasing from 0 to n; C is a constant; h is the delay distance; H is the Hurst exponent; and D is the fractal dimension.

5. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 4, characterized in that: The quantitative calculation method of the single layer crack rate in step 3 is: Since the X-ray CT scanning imaging analysis of the internal structure of the sample is achieved through its relative density difference, the unfilled microcracks in the coal sample belong to the lowest density part, and the microcracks appear in the CT image as a dark gray color that is much darker than the grayscale value of the mirror coal and bright coal; the 1500 photos generated by the scan are reconstructed in Avizo, and the grayscale boundaries between the mirror coal, bright coal and microcracks are identified using the watershed algorithm. The volume of 500,000 pixels is used as the threshold for microcrack segmentation, and the microcracks in the denoised and sharpened samples are segmented and extracted. After the crack extraction is completed, the volume fraction command in Avizo is used to calculate the ratio of the crack voxels extracted from the 1500 photos to the set background voxels to obtain the crack rate of the sample: Where F represents the single-layer fracture rate in the core section, S f Represents the grayscale image area of ​​a single layer of cracks, m 2 ; r is the core radius, m, the higher the F value, the higher the proportion of fractures.

6. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 5, characterized in that: The method for establishing a comprehensive indication curve using the entropy value method in step 4 of the step 4 is as follows: (1) Sample data normalization Normalize the data of coal sample fracture intensity curve, fractal dimension curve, and core single layer fracture rate curve; (2) Normalized data volume translation In the entropy weight method, when x mn When =0, the entropy value is meaningless, and all normalized data are shifted to the right by 0.0001 units; (3) Entropy weight calculation After data preprocessing is completed, the entropy value and weight of each curve parameter are calculated, and each curve value is multiplied by its weight, and then normalized to obtain a comprehensive indicator curve; The entropy calculation formula is: Q=α·K f +β·K D +λ·K F (11) In the formula, Q b is the information entropy value of each ion, ε b is the weight, m ​​is the number of samples, n is the number of indicators, α is the fracture intensity weight, β is the fractal dimension weight, λ is the core fracture rate weight, K f is the crack strength value, K D is the fractal dimension value, K F is the core fracture rate value.

7. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 6, characterized in that: The step 5 is specifically as follows: according to the deterministic modeling of the faults in the study area, the seismic data is used to directly identify the medium and large scale fractures in the profile through the sudden change of the reflection phase axis, the offset of the wave group and wave system, and the sudden change of the reflection layer, the fault size is analyzed, and the medium and large scale fractures are deterministically modeled by Petrel. The scope of the fault influence is delineated according to experience, the seismic fracture attribute characteristics are cut, and the geometrical modeling module is used to construct the fault distance attribute body.

8. The method for constructing a coal reservoir fracture prediction model by fusion of multi-source data volumes according to claim 7, characterized in that: The specific method of step 7 is: the artificial neural network adopts supervised machine learning, the training supervision points are pre-set, the training supervision points include crack sample points and non-crack sample points, the learning samples include sensitive data bodies, comprehensive indicator curves and crack training sets, the artificial neural network assigns values ​​to the training supervision points, and extrapolates the probability of crack development in the three-dimensional direction, and finally constructs a comprehensive discriminant.

Images