Coal reservoir intelligent inversion method and system for self-adaptive rock physical modeling, and medium

Abstract

The invention relates to a coal measure reservoir intelligent inversion method and system for self-adaptive rock physical modeling and a medium The coal measure reservoir intelligent inversion method for self-adaptive rock physical modeling comprises the following steps: self-adaptive rock physical modeling: calculating priority parameters of a rock physical model according to well logging interpretation data, calculating longitudinal and transverse wave speed parameters and density parameters based on the rock physical model corresponding to the highest priority parameter; performing artificial intelligence seismic inversion; obtaining angle elastic impedance data, forming a training set of a neural network by using the angle elastic impedance data and the corresponding longitudinal and transverse wave speed parameters and density parameters, and training a full-connection deep neural network model for reservoir parameter prediction based on the training data set; and obtaining a regression prediction result based on the density of the pre-stack seismic data and a longitudinal and transverse wave velocity ratio parameter based on the elastic impedance data generated based on the pre-stack seismic data. According to the invention, the elastic parameter inversion of the density and the longitudinal-transverse wave velocity ratio is realized.

Description

Technical Field

[0001] This invention relates to the field of coal-bearing reservoir inversion. In particular, it relates to an intelligent inversion method, system, and medium for coal-bearing reservoirs using adaptive rock physics modeling. Background Technology

[0002] Coal-bearing reservoirs are characterized by strong heterogeneity and the development of superimposed gas-bearing systems, rendering conventional seismic prediction techniques based on elastic parameter inversion unsuitable and severely restricting the improvement of coal-bearing gas extraction capacity. The advantages of deep learning in automatically extracting features offer significant potential for effective prediction of complex coal-bearing reservoirs, potentially overcoming the technical bottlenecks in coal-bearing reservoir prediction and achieving high-quality prediction results.

[0003] There is an urgent need for a method to automatically extract features to invert complex coal-bearing reservoirs, in order to address the inapplicability of conventional seismic prediction techniques based on elastic parameter inversion. Summary of the Invention

[0004] This invention provides an intelligent inversion method, system, and medium for adaptive rock physics modeling of coal-bearing reservoirs, to address the inapplicability of conventional seismic prediction techniques based on elastic parameter inversion.

[0005] To achieve the above objectives, in a first aspect, the present invention relates to an intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling, wherein the adaptive rock physics modeling involves calculating the priority parameters of various rock physics models in a set of candidate rock physics models for each logging depth point of the logging interpretation data, and calculating the P-wave and S-wave velocity parameters and density parameters based on the rock physics model corresponding to the highest priority parameter.

[0006] Artificial intelligence seismic inversion: The angular elastic impedance obtained from the rock physics model and the corresponding calculated P-wave and S-wave velocity parameters and density parameters form a training set for a neural network. Based on the training dataset, a fully connected deep neural network model for reservoir parameter prediction is trained. The elastic impedance data generated from pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results of density and P-wave and S-wave velocity ratio parameters based on pre-stack seismic data are obtained.

[0007] To achieve the above objectives, in a second aspect, the present invention relates to an intelligent inversion system for coal-bearing reservoirs based on adaptive rock physics modeling, comprising: an adaptive rock physics modeling model for calculating priority parameters of various rock physics models in a set of candidate rock physics models for each logging depth point of logging interpretation data, wherein the P-wave and S-wave velocity parameters and density parameters are calculated based on the rock physics model corresponding to the highest priority parameter;

[0008] Artificial intelligence seismic inversion model: Using the P-wave and S-wave velocity parameters and density parameters obtained from the rock physics modeling, angular elastic impedance data is obtained. The angular elastic impedance data and its corresponding P-wave and S-wave velocity parameters and density parameters are used to form a training set for a neural network. Based on the training dataset, a fully connected deep neural network model for reservoir parameter prediction is trained. The elastic impedance data generated based on pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results based on the density and the P-wave and S-wave velocity ratio parameters of the pre-stack seismic data are obtained.

[0009] To achieve the above objectives, in a third aspect, the present invention also relates to a computer-readable storage medium storing instructions that, when executed, perform the above-described intelligent inversion method for adaptive rock physics modeling of coal-bearing reservoirs.

[0010] The present invention relates to an intelligent inversion method, system, and medium for adaptive rock physics modeling of coal-bearing reservoirs, which has the following advantages compared with the prior art:

[0011] 1. Based on well logging data and 3D pre-stack time migration data, a smart seismic inversion technology for coal-bearing reservoirs based on adaptive rock physics modeling is proposed. By introducing adaptive rock physics modeling and artificial intelligence deep learning technology, the technology can effectively describe the heterogeneity of the reservoir and make high-precision predictions of the reservoir's P-wave and S-wave velocities and densities.

[0012] On the cross section, conventional inversion results have low resolution and cannot accurately identify differences in elastic parameters within the coal seam. Intelligent inversion results have better resolution. On the plane, intelligent inversion has higher lateral resolution, which is conducive to conducting more detailed research on the characteristics of coal-bearing strata and reservoirs.

[0013] 3. Artificial intelligence seismic inversion effectively solves the problem of distinguishing sand and mud layers, which cannot be solved by conventional inversion, and also provides a good display of thin coal seams, which can be used for subsequent thin coal seam identification work.

[0014] 4. By combining fully connected deep neural networks with elastic impedance inversion, a bridge is built between well logging parameters and seismic parameters using elastic impedance, ultimately achieving elastic parameter inversion such as density and P-wave / S-wave velocity ratio. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an intelligent inversion method for coal-bearing reservoirs using adaptive rock physics modeling, as described in Example 1.

[0016] Figure 2 This is a schematic diagram of the adaptive rock physics modeling process in Example 1 of the intelligent inversion method for coal-bearing reservoirs using adaptive rock physics modeling, as described in Example 1.

[0017] Figure 3 This is a schematic diagram illustrating the reservoir types and basic rock physics model types in Example 1 of an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs.

[0018] Figure 4 This figure shows the prediction results of well logging data from the rock physics model in Example 1 of the intelligent inversion method for coal-bearing reservoirs using adaptive rock physics modeling, as described in Example 1. In the figure, the "Depth" column represents the formation depth in meters. In the "Combined Trajectory" column, the left column represents the stratigraphic division, with B1-B7 as stratigraphic identifiers corresponding to 7 sets of strata. The right column represents the stratigraphic lithology, with different patterns representing different lithologies. The "Density" column represents the density parameter, with the vertical axis representing the formation depth in meters and the horizontal axis representing the density in kilograms per cubic meter. The "P-wave Velocity" column represents the P-wave velocity parameter, with the vertical axis representing the formation depth in meters and the horizontal axis representing the P-wave velocity in meters per second. The "S-wave Velocity" column represents the S-wave velocity parameter, with the vertical axis representing the formation depth in meters and the horizontal axis representing the S-wave velocity in meters per second.

[0019] Figure 5 This is a fully connected deep neural network for elastic parameter prediction, which is Example 1 of the intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling in Example 1.

[0020] Figure 6 This is an AI seismic inversion technology process for an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs, as described in Example 1.

[0021] Figure 7 This is a comparison of the conventional inversion (middle) and intelligent inversion (right) results based on pre-stack data (left) for Example 1 of an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs. The horizontal axis of both left and right figures represents the seismic signal observation point number, and the vertical axis represents the seismic wave double travel time in milliseconds.

[0022] Figure 8 This is a planar attribute comparison diagram of conventional inversion (left) and intelligent inversion (right) of Example 1 of an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs in Example 1; the horizontal and vertical axes of both diagrams are the x and y coordinates of geodetic space, in meters.

[0023] Figure 9 This is a schematic diagram of the layer flattening technique in Example 1 of the intelligent inversion method for coal-bearing reservoirs using adaptive rock physics modeling in Embodiment 1; the horizontal axis represents the seismic signal observation point number, and the vertical axis represents the double travel time of seismic waves in milliseconds.

[0024] Figure 10This is a schematic diagram comparing the P-wave impedance results of conventional inversion (left) and intelligent inversion (right) in Example 1 of an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs. The horizontal axis of both the left and right graphs represents the seismic signal observation point number, and the vertical axis represents the double travel time of the seismic waves, in milliseconds.

[0025] Figure 11 This is a schematic diagram of the intelligent inversion results of shear wave impedance (left) and density (right) in Example 1 of an adaptive rock physics modeling intelligent inversion method for coal-bearing reservoirs in Example 1. The horizontal axis of both the left and right graphs represents the seismic signal observation point number, and the vertical axis represents the double travel time of the seismic wave in milliseconds.

[0026] Figure 12 This is a schematic diagram of the structure of an intelligent inversion system for coal-bearing reservoirs with adaptive rock physics modeling, as described in Embodiment 2 of the present invention. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention and not the entire structure.

[0028] Example 1

[0029] A smart inversion method for coal-bearing reservoirs based on adaptive rock physics modeling; please refer to [link / reference]. Figure 1-11 As shown, the present invention provides an intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling, comprising the following steps: S101 to S102.

[0030] S101 Adaptive Rock Physics Modeling: For each logging depth point in the well logging interpretation data, the priority parameters of various rock physics models in the set of candidate rock physics models are calculated, and the P-wave and S-wave velocity parameters and density parameters are calculated based on the rock physics model corresponding to the highest priority parameter.

[0031] S102 AI-based seismic inversion: Using P-wave and S-wave velocity parameters and density parameters obtained from rock physics modeling, angular elastic impedance data is obtained. The angular elastic impedance data and its corresponding P-wave and S-wave velocity parameters and density parameters are used to form a training set for a neural network. Based on the training dataset, a fully connected deep neural network model for reservoir parameter prediction is trained. The elastic impedance data generated from pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results of density and P-wave / S-wave velocity ratio parameters based on pre-stack seismic data are obtained.

[0032] Density and P-wave / S-wave velocity ratio are crucial elastic parameters for reservoir characterization and fluid identification, which can be obtained through elastic impedance retrieval based on pre-stack seismic data. However, due to the limited incident angle range and poor data quality of seismic data, the accuracy of density retrieval using elastic impedance is low, and the inversion stability of the P-wave / S-wave velocity ratio at its velocity peak is also relatively low. This step combines a fully connected deep neural network with elastic impedance retrieval, using elastic impedance to build a bridge between logging parameters and seismic parameters, ultimately achieving the inversion of elastic parameters such as density and P-wave / S-wave velocity ratio.

[0033] In this embodiment, S101 further includes:

[0034] S111 sets the overall error objective function for logging P-wave and S-wave velocities and densities. Based on the P-wave and S-wave velocity parameters, density parameters, and the overall error objective function for logging P-wave and S-wave velocities and densities, the calculation error of logging P-wave and S-wave velocities and densities calculated by the rock physics optimization model corresponding to the optimization parameters relative to the logging interpretation data is determined.

[0035] In this embodiment, the overall error objective function for logging P-wave and S-wave velocities and densities is set as follows:

[0036] Determine if there is shear wave data. If there is shear wave data, the objective function is: L s =(V pi -V′ pi ) 2 +(V si -V′ si ) 2 +(Den i -Den′ i ) 2 If there is no shear wave data, then the objective function is: L s =(V pi -V′ pi ) 2 +(Den i -Den′ i ) 2 Where Vpi is the measured P-wave velocity from the well logging data, and V′ is... pi The calculated P-wave velocity data, Vsi is the measured S-wave velocity from the well logging data, and V′ is the calculated P-wave velocity data. si Deni represents the calculated shear wave velocity data; Deni represents the measured density from the logging data; Den′ represents the calculated density data; and i represents the logging sample number.

[0037] If S112 determines that the calculation error remains within the preset error threshold, then the rock physics model will be screened.

[0038] In some embodiments, S101 further includes S113: if the calculation error is higher than a preset error threshold, the default parameters of the highest priority rock physics model are optimized, and the above calculation steps are repeated to recalculate new P-wave and S-wave velocity parameters and density parameters until the rock physics model matching is lower than the threshold, and a rock physics model matching the logging data at that depth point is obtained, wherein the error threshold is set to 1% of the sum of squares of P-wave velocity and density.

[0039] Using the measured data and the results calculated by this method, the objective function value, i.e. the sum of squared errors, is calculated. If the objective function value is less than or equal to the threshold, the requirement is met; if the objective function value is greater than the threshold, the requirement is not met.

[0040] In some embodiments, if the error of the highest priority rock physics model still exceeds a threshold after optimizing its default parameters to the default limit, a lower priority rock physics model is selected and the steps are repeated.

[0041] S112-S113, until the model matching falls below a threshold, obtain a rock physical model that matches the logging data at that depth point. If none of the candidate rock physical models match, return to optimizing the logging parameters to obtain logging interpretation data. Specifically, in this method, optimizing logging parameters to obtain logging interpretation data takes logging data and logging interpretation results (interpretation parameters) as input. This method does not include specific logging and logging interpretation techniques. When the parameters cannot be optimized to reach the set threshold using this method, it is generally considered that there may be a problem with the logging data or logging interpretation results. In this case, logging interpretation personnel need to verify the logging data and logging interpretation results, and in most cases, the logging data needs to be reinterpreted.

[0042] In this embodiment, before S101, S103-S104 are also included:

[0043] S103 interprets the actual logging data to obtain logging interpretation data, which includes density, porosity, water saturation, and logging mineral data;

[0044] S104 performs regularization processing on the logging mineral data in the logging interpretation data, expanding the logging mineral data at each depth point to include all minerals in the work area. Among them, the value of minerals that do not exist at the corresponding depth point is set to 0, and the data is processed into standard logging mineral data that includes all mineral types in the work area.

[0045] The well logging mineral data includes quartz content, clay content, calcite content, dolomite content, and gypsum content.

[0046] In S101, for each logging depth point in the well logging interpretation data, priority parameters for various rock physics models in the set of candidate rock physics models are calculated, including S113-S114:

[0047] S113 calculates priority parameters for various rock physics models at each logging depth point based on the regularized well logging mineral data, and ranks the rock physics models according to the priority parameters. The calculation method for the priority parameters of the rock physics model at any depth point is as follows: Suppose the rock physics model has m variables that require input, the penalty value βj for the j-th variable matching value is 2, and the penalty value βj for the non-matching value is 0.5. Where Ad is the priority parameter.

[0048] Based on a preset priority calculation criterion, S114 performs principal component analysis on the minerals at each logging depth point in the set of candidate rock physics models to obtain the priority parameter of each rock physics model. The priority parameter calculation criterion is as follows: if the time logging data does not provide the required data, the rock physics model corresponding to the set of rock physics models is removed from the set, and the remaining models are the candidate rock physics models. Each rock physics model first compares the required input data. If the required input data does not exist in the logging data, the priority parameter Ad value is 0.

[0049] In S101, the P-wave and S-wave velocity parameters are calculated based on the rock physics model corresponding to the highest priority parameter. Specifically, the rock matrix modulus is calculated based on the default mineral modulus of the highest priority rock physics model. Then, according to the default pore type and pore aspect ratio, the P-wave and S-wave velocity parameters and density parameters under the logging fluid state are obtained through the Gassmann formula.

[0050] In this embodiment, the rock matrix modulus is calculated based on the default mineral modulus of the highest priority rock physics model, including: calculating the rock matrix modulus using the Voigt-Reuss-Hill average. Where Mm is the equivalent elastic modulus of the carbonate rock framework, Mk is the modulus of the k-th component of the carbonate rock mineral composition, fk is the volume component of the k-th component of the carbonate rock mineral composition, Mv is the rock modulus calculated using the Voigt upper bound method, and MR is the rock modulus calculated using the Reuss lower bound method. In this embodiment, the rock matrix modulus is calculated based on the default mineral modulus of the highest priority rock physics model. The calculation also includes: if multiple data points with similar mineral components exist, the mixed fluid bulk modulus is calculated using the Brine formula; if the data points are isolated, the fluid bulk modulus is calculated using the Patchy formula.

[0051] In this embodiment, the rock physics model set includes at least the DEM differential equivalent model, the SCA self-consistent model, the KT inclusion model, the Xu-White model, the Xu-Payne model, the KG model, the Hudson fractured medium model, the anisotropic SCA-DEM model, the cemented sandstone model, and the uncemented sandstone model.

[0052] Each model has its most applicable range for mineral composition and porosity. There's an assumption that the more requirements a model has, the higher its accuracy but the lower its generalizability. Therefore, the concept of a model penalty β is introduced. Suppose the model has m input variables, the penalty for a matching value of the j-th variable is βj = 2, and the penalty for a non-matching value is βj = 0.5. Where Ad is the priority parameter.

[0053] In one example, assume Model 1 is applicable to Type 1 sandstone-mudstone reservoirs, with specific parameters ranging from 40-100% quartz + feldspar, 0-60% clay, and 5-20% porosity. Model 2 is applicable to Type 2 sandstone-mudstone reservoirs, with specific parameters of 60-100% quartz + feldspar, 0-40% clay, and 20-40% porosity. Model 3 is applicable to carbonate rocks, with specific parameters of 50%-100% limestone + dolomite and 0-30% porosity. For a sandstone-mudstone reservoir with 15% porosity interbedded with 6% porosity limestone, the priority parameter Ad for Model 1 is calculated to be 8, for Model 2 to be 2, and for Model 3 to be 1. Therefore, the priority order is Model 1 - Model 2 - Model 3. For carbonate rock interlayers, the priority parameter Ad = 0.5 for sandstone and mudstone strata model 1, Ad = 0.25 for model 2, and Ad = 4 for model 3. Therefore, the priority order is model 3 - model 1 - model 2.

[0054] Adaptive rock physics modeling calculates priority parameters for various rock physics models at each logging depth. The highest priority model calculates the overall error objective function of P-wave and S-wave velocity parameters, density parameters, and logging P-wave and S-wave velocities and densities, keeping the error within a threshold. This process effectively filters rock physics models, making it particularly suitable for work areas with rapid vertical variations. By employing the most appropriate rock physics model for different depths, the accuracy of rock physics model description and prediction is improved. Based on the established adaptive rock physics model, high-precision prediction of logging curves can be achieved, and ultimately, a predicted S-wave velocity curve is provided.

[0055] In Example 1, Figure 2 A schematic diagram of the adaptive rock physics modeling process. Figure 3The document provides a description of reservoir types and rock physics models, totaling over 30 models and more than 1000 combinations, requiring over 8000 lines of code. The final calculation results are as follows: Figure 4 As shown in the figure, the lithological calculation results and well logging interpretation results have a high degree of agreement, and the density and P-wave velocity calculation results also have a high degree of agreement with the well logging curves.

[0056] In this embodiment, S102 specifically includes S121-S125:

[0057] S121 divides all logging data into training and testing sets. The logging data includes three angle elastic impedance data calculated using a rock physics model, and the corresponding P-wave and S-wave velocity parameters and density parameters calculated by the rock physics model to form a neural network training set. S101 The elastic impedance (EI) obtained by logging interpretation data modeling is a function of P-wave velocity VP, S-wave velocity VS, density ρ and angle θ. The elastic impedance EI of three angles, small (3°), medium (8°) and large (13°), can be calculated based on the P-wave velocity VP, S-wave velocity VS and density ρ of the actual logging data. This application calculates the P-wave velocity and S-wave velocity and density based on the P-wave velocity VP, S-wave velocity VS and density ρ obtained by an adaptive rock physics model, and calculates the elastic impedance of the three angles based on the given three angles.

[0058] S122 performs parameter standardization on the training set and retains the standardized training data, and then uses the same parameter standardization process to standardize the test set to obtain standardized test data.

[0059] S123 inputs standardized training data into a deep neural network algorithm to train a deep neural network model for reservoir parameter prediction.

[0060] S124 examines the performance of the deep neural network model in the test set of well logging data to obtain a deep neural network model whose performance meets the preset conditions.

[0061] S125 extracts elastic impedance data volumes for three corresponding angles from the pre-stack angle partial superposition channel set, where the three angles are 3°, 8° and 13° respectively.

[0062] S126 uses the same standardized parameter processing as the training set to normalize the corresponding seismic elastic impedance data.

[0063] S127 applies a deep neural network model with performance meeting preset conditions to standardized seismic elastic impedance data to obtain regression prediction results based on the density and P-wave / S-wave velocity ratio parameters of the seismic data.

[0064] Among them, such as Figure 5 , 6As shown in Figure 9, in Example 1, the network model for reservoir parameter prediction is a fully connected deep neural network model consisting of 5 hidden layers, with the number of neurons in each hidden layer ranging from 20 to 80. The activation function is the ReLU function, the optimization algorithm is the Adam algorithm, and Dropout regularization is used to prevent overfitting. The elastic impedance at three angles—small (EI_N), medium (EI_M), and large (EI_F)—is used as the input parameters of the neural network, and the output parameters can be the longitudinal wave velocity V. P Shear wave velocity V S and the ratio of density ρ to longitudinal and transverse wave velocities V P / V S Each output parameter corresponds to a network model.

[0065] In Example 1 above, a neural network model was trained based on known well information and seismic data to invert and obtain the P-wave and S-wave impedances and densities. A comparison of the P-wave impedance inversion results obtained by conventional methods and artificial intelligence methods shows that, on the cross-section, the conventional inversion results have lower resolution and cannot accurately identify differences in elastic parameters within the coal seam, while the intelligent inversion results have better resolution. Figure 7 On a planar surface, intelligent inversion has higher lateral resolution, which is beneficial for conducting more detailed studies on the characteristics of coal-bearing strata reservoirs. Figure 8 ).

[0066] While intelligent inversion can yield high-resolution results, the significant difference in target layer height causes spatial aliasing. This study employs layer flattening techniques to overcome this spatial aliasing. Figure 9 Meanwhile, the conventional wave impedance inversion results and the intelligent inversion results were re-compared based on the data from the layer flattening. Figure 10 The results show that artificial intelligence seismic inversion effectively solves the problem of distinguishing sand and mud layers, which cannot be solved by conventional inversion, and also provides a good indication of thin coal seams, which can be used for subsequent thin coal seam identification. Figure 11 The results of intelligent inversion of transverse wave impedance and density are presented.

[0067] Example 2

[0068] An intelligent inversion system for coal-bearing reservoirs based on adaptive rock physics modeling is used for predicting the gas content of coal-bearing reservoirs. It is implemented using electronic hardware with a central processing unit, such as a personal computer, smart terminal, local area network, or server. For implementation details in this example, please refer to [link to relevant documentation]. Figure 12 This includes an adaptive rock physics modeling model 61 and an artificial intelligence seismic inversion model 62.

[0069] Adaptive Rock Physics Modeling Model 61: Used to calculate the priority parameters of various rock physics models in the set of candidate rock physics models for each logging depth point of the logging interpretation data, and calculate the P-wave and S-wave velocity parameters and density parameters based on the rock physics model corresponding to the highest priority parameter.

[0070] Artificial intelligence seismic inversion model 62: It is used to obtain angular elastic impedance data by using the P-wave and S-wave velocity parameters and density parameters obtained from rock physics modeling. The angular elastic impedance data and its corresponding P-wave and S-wave velocity parameters and density parameters are used to form a training set for a neural network. Based on the training dataset, a fully connected deep neural network model for reservoir parameter prediction is trained. The elastic impedance data generated based on pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results of density and P-wave and S-wave velocity ratio parameters based on pre-stack seismic data are obtained.

[0071] The intelligent inversion system for coal-bearing reservoirs based on adaptive rock physics modeling in this embodiment is the same as the intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling described in Embodiment 1 in terms of implementation process, method and effect, and will not be repeated here.

[0072] Example 3

[0073] This invention relates to a computer-readable storage medium storing instructions that, when executed, perform an intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling, as described in Embodiment 1. The execution process and effects are the same as those of the intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling described in Embodiment 1, and will not be repeated here.

[0074] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0075] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling, characterized in that, include: Adaptive rock physics modeling: For each logging depth point in the well logging interpretation data, the priority parameters of various rock physics models in the set of candidate rock physics models are calculated, and the P-wave and S-wave velocity parameters and density parameters are calculated based on the rock physics model corresponding to the highest priority parameter. Artificial intelligence seismic inversion: Using the P-wave and S-wave velocity parameters and the density parameters obtained from the rock physics modeling, angular elastic impedance data is obtained. The angular elastic impedance data and its corresponding P-wave and S-wave velocity parameters and density parameters are used to form a training set for a neural network. A fully connected deep neural network model for reservoir parameter prediction is trained based on the training dataset. The elastic impedance data generated based on pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results based on the density and P-wave and S-wave velocity ratio parameters of the pre-stack seismic data are obtained.

2. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 1, characterized in that, The adaptive rock physics modeling also includes: Set an overall error objective function for logging P-wave and S-wave velocities and densities. Based on the P-wave and S-wave velocity parameters, the density parameters, and the overall error objective function for logging P-wave and S-wave velocities and densities, determine the calculation error of logging P-wave and S-wave velocities and densities calculated by the rock physics optimization model corresponding to the optimization parameters relative to the logging interpretation data. If the calculation error is determined to be within a preset error threshold, the rock physics model is then screened.

3. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 1, characterized in that, Prior to the adaptive rock physics modeling, the following also includes: The well logging interpretation data is obtained by interpreting the actual well logging data, which includes density, porosity, water saturation, and well logging mineral data. The logging mineral data in the logging interpretation data is processed into a regularization process, and the logging mineral data at each depth point is expanded to include all minerals in the work area. The value of minerals that do not exist at the corresponding depth point is set to 0, and the process is processed into standard logging mineral data that includes all mineral types in the work area. The priority parameters for each logging depth point in the well logging interpretation data, calculated for each candidate rock physics model in the set of alternative rock physics models, include: Based on the regularized well logging mineral data, priority parameters for various rock physics models are calculated for each well logging depth point. The priority parameters are then used to rank the rock physics models. The calculation method for the priority parameters of the rock physics model at any depth point is as follows: Assume the rock physics model has m required input variables. The penalty for a matching value of the j-th variable is βj = 2, and the penalty for a non-matching value is βj = 0.

5. Where Ad is the priority parameter; Based on a preset priority calculation criterion, in the set of candidate rock physical models, the minerals at each logging depth point are analyzed for their main components to obtain the priority parameter of each rock physical model. The priority parameter calculation criterion is as follows: if the time logging data does not provide the required data, the rock physical model corresponding to the set of rock physical models is removed from the set, and the remaining models are the candidate rock physical models. Each rock physical model first compares the required input data. If the required input data does not exist in the logging data, the priority parameter Ad value is 0. The calculation of P-wave and S-wave velocity parameters based on the rock physics model corresponding to the highest priority parameter is specifically as follows: the rock matrix modulus is calculated based on the default mineral modulus of the highest priority rock physics model, and then the P-wave and S-wave velocity parameters and density parameters under the logging fluid state are obtained according to the default pore type and pore aspect ratio and the Gassmann formula.

4. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 2, characterized in that, The overall error objective function for setting the logging P-wave and S-wave velocities and densities is specifically as follows: Determine if there is shear wave data. If there is shear wave data, then the objective function is: L s =(V pi -V′ pi ) 2 +(V si -V′ si ) 2 +(Den i -Den′ i ) 2 If there is no shear wave data, then the objective function is: L s =(V pi -V′ pi ) 2 +(Den i -Den′ i ) 2 , where V pi V′ is the measured P-wave velocity from well logging data. pi For the calculated P-wave velocity data, V si V′ is the measured shear wave velocity from well logging data. si The calculated shear wave velocity data; Deni is the density measured in the logging data; Den′ is the calculated density data; and i is the logging sample number. It also includes: if the calculation error is higher than a preset error threshold, the default parameters of the highest priority rock physics model are optimized, and the above calculation steps are repeated to recalculate new P-wave and S-wave velocity parameters and density parameters until the rock physics model matching is lower than the threshold, and a rock physics model matching the logging data at that depth point is obtained, wherein the error threshold is set to 1% of the sum of the squares of the P-wave velocity and density.

5. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 3, characterized in that, The set of rock physics models includes at least the DEM differential equivalent model, the SCA self-consistent model, the KT inclusion model, the Xu-White model, the Xu-Payne model, the KG model, the Hudson fractured medium model, the anisotropic SCA-DEM model, the cemented sandstone model, and the uncemented sandstone model.

6. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 3, characterized in that, The rock matrix modulus is calculated based on the default mineral modulus of the highest priority rock physics model, including: calculating the rock matrix modulus using the Voigt-Reuss-Hill average. Among them, M m M represents the equivalent elastic modulus of the carbonate rock framework. k Let fk be the modulus of the k-th component of the mineral composition of the carbonate rock, fk be the volume component of the k-th component of the mineral composition of the carbonate rock, and Mv be the rock modulus calculated using the Voigt upper bound method. R The rock modulus is calculated using the Reuss lower bound method.

7. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 6, characterized in that, The calculation of the rock matrix modulus based on the default mineral modulus of the highest priority rock physics model also includes: if there are multiple data points with similar mineral components, the mixed fluid bulk modulus is calculated using the Brine formula; if the data points are isolated, the fluid bulk modulus is calculated using the Patchy formula.

8. The intelligent inversion method for coal-bearing reservoirs based on adaptive rock physics modeling according to claim 1, characterized in that: The artificial intelligence seismic inversion steps specifically include: All logging data are divided into training and testing sets. The logging data includes three angle elastic impedance data calculated using the rock physics model and the corresponding P-wave and S-wave velocity parameters and density parameters calculated by the rock physics model, forming a training set for the neural network. The training set is standardized by parameter processing and the standardized training data is retained. Then, the same standardized parameter processing is used to standardize the test set to obtain standardized test data. The standardized training data is input into a deep neural network algorithm to train a deep neural network model for reservoir parameter prediction. The performance of the deep neural network model is tested in the test set of well logging data to obtain a deep neural network model whose performance meets the preset conditions; The elastic impedance data volumes of three corresponding angles are extracted from the pre-stack angle partial superposition channel set, where the three angles are 3°, 8° and 13° respectively; The corresponding seismic elastic impedance data are normalized using the same standardized parameters as those used in the training set. The deep neural network model whose performance meets the preset conditions is applied to the standardized seismic elastic impedance data to obtain regression prediction results based on the density and P-wave / S-wave velocity ratio parameters of the seismic data.

9. An intelligent inversion system for coal-bearing reservoirs based on adaptive rock physics modeling, characterized in that, include: Adaptive rock physics modeling model: used to calculate the priority parameters of various rock physics models in the set of candidate rock physics models for each logging depth point of the logging interpretation data, and to calculate the P-wave and S-wave velocity parameters and density parameters based on the rock physics model corresponding to the highest priority parameter; Artificial intelligence seismic inversion model: Using the P-wave and S-wave velocity parameters and the density parameters obtained from the rock physics modeling, angular elastic impedance data is obtained. The angular elastic impedance data and its corresponding P-wave and S-wave velocity parameters and density parameters are used to form a training set for a neural network. A fully connected deep neural network model for reservoir parameter prediction is trained based on the training dataset. The elastic impedance data generated based on pre-stack seismic data is input into the fully connected deep neural network to perform regression prediction of reservoir parameters, and the regression prediction results based on the density and P-wave and S-wave velocity ratio parameters of the pre-stack seismic data are obtained.

10. A computer-readable storage medium, characterized in that: The storage medium stores instructions that, when executed, perform an adaptive rock physics modeling method for intelligent inversion of coal-bearing reservoirs as described in any one of claims 1-8.

Images