A method and apparatus for establishing a three-dimensional sandstone probability volume

Abstract

This invention proposes a method and apparatus for establishing a three-dimensional sandstone probabilistic volume. The method includes: integrating a well seismic database; constructing a three-dimensional structural framework model; converting the time-domain seismic data volume into a depth-domain seismic data volume; extracting target seismic attributes according to preset rules; sampling the target seismic attributes into the three-dimensional structural framework model; using a preset neural network model to perform multi-attribute clustering and fusion of the seismic attribute values ​​of the three-dimensional structural framework model to establish a sedimentary unit model; calculating the variation function between sandstone content data for each partition of the sedimentary unit model; and using the Kriging algorithm for interpolation to predict the sandstone content at unknown points, thereby obtaining a three-dimensional sandstone probabilistic volume model. This invention establishes a multi-type, multi-attribute fused three-dimensional sandstone probabilistic volume. By assigning probability values ​​to each grid, it constrains the value type of reservoir facies and the value magnitude of reservoir parameters, providing reliable trend constraints for geological modeling of low-well-control and highly heterogeneous reservoirs.

Description

Technical Field

[0001] This invention belongs to the field of sedimentary reservoir evaluation in petroleum geological exploration, and particularly relates to a method and apparatus for establishing a three-dimensional sandstone probability volume. Background Technology

[0002] In recent years, relying on computer technology and in-depth research in mathematical geology, three-dimensional geological modeling technology has developed rapidly. Three-dimensional geological modeling technology is a digital model that uses computer technology to characterize the spatial distribution patterns and variation characteristics of geological parameters, based on research results in seismology, well logging, geology, and dynamics, and applying geostatistics, spatial analysis, and prediction methods.

[0003] Commonly used modeling algorithms include deterministic and stochastic algorithms. Objective-based and pixel-based stochastic algorithms have been widely applied in this field, providing fundamental conditions for 3D reservoir characterization. Pixel-based sequential indicator modeling methods mainly utilize the variogram between two points to reflect the heterogeneity of the reservoir structure. This method is suitable for densely networked well areas during the oil and gas field development stage. However, for areas with fewer wells and strong heterogeneity in the exploration and evaluation stage, the global variogram is insufficient to characterize channel variations. Objective-based modeling methods can incorporate prior geological knowledge into the simulation, maximizing the reproduction of geologists' understanding. This method has been widely used.

[0004] Currently, reservoir modeling primarily employs facies boundary constraints and single-trend constraints to constrain the simulation process, which improves the rationality of the modeling results to some extent. However, for areas with limited drilling data, inadequate control of the variation function within the boundary and insufficient consideration of single trends prevent this method from accurately depicting the spatial distribution of reservoirs. Furthermore, multi-condition constraint modeling methods that use weighted average transformations of multiple trends have been widely applied; however, this method struggles to determine the weights of each attribute on each grid, making the multi-trend integration process difficult. Summary of the Invention

[0005] To address the problems encountered by traditional modeling constraint methods under conditions of few wells and strong heterogeneity, this invention proposes a method and apparatus for establishing a three-dimensional sandstone probabilistic volume.

[0006] A method for establishing a three-dimensional sandstone probabilistic volume includes the following steps:

[0007] An integrated well-seismic database is provided, the data of which includes at least well information, geological research results data, seismic results data, and well logging interpretation results data for the study area. The seismic results data include at least two depth domain interpretation fault data, two depth domain interpretation plane data, two time domain interpretation plane data, and one time domain seismic data volume.

[0008] A three-dimensional structural framework model was constructed using the earthquake data.

[0009] The time-domain seismic data volume is converted into a depth-domain seismic data volume, and the well seismic database is updated.

[0010] In the updated well seismic database, target seismic attributes are extracted according to preset rules, and these target seismic attributes can reflect reservoir changes.

[0011] The target seismic attributes are sampled into the three-dimensional structural frame model, and seismic attribute values ​​are assigned to the three-dimensional structural frame model.

[0012] A sedimentary unit model is established by using a pre-defined neural network model to perform multi-attribute clustering and fusion of seismic attribute values ​​from a three-dimensional structural framework model.

[0013] The variation function between sandstone content data is calculated for each partition of the sedimentary unit model to estimate the sandstone content data at unsampled locations in the sedimentary unit model.

[0014] Based on the known sandstone content distribution data in the variation function and well logging interpretation data, the Kriging algorithm is used to interpolate on the three-dimensional structural framework grid to predict the sandstone content at unknown points, thus obtaining a three-dimensional sandstone probabilistic volume model.

[0015] Furthermore, a three-dimensional structural framework model is constructed using the aforementioned seismic data, including:

[0016] The three-dimensional structural framework model is constructed from the two depth domain interpretation layers and the two depth domain interpretation faults.

[0017] Furthermore, the time-domain seismic data volume is converted into depth-domain seismic data, including:

[0018] The time-domain seismic data volume is transformed into a depth-domain seismic data volume using a preset velocity model, wherein the preset velocity model is:

[0019]

[0020] Where V1 and V2 represent two time-domain interpretation layers, Z represents depth, and Z1 and Z2 are two depth-domain interpretation layers. interp This is a depth-domain seismic data volume transformed from a time-domain seismic data volume.

[0021] Furthermore, in the updated well seismic database, target seismic attributes are extracted according to preset rules, including:

[0022] Extract data on seismic attributes related to lithofacies, porosity, and permeability;

[0023] After normalizing and unifying the dimensions of the earthquake attribute data, a spatial intersection diagram of earthquake attributes, porosity, permeability, and lithofacies is drawn.

[0024] Based on the spatial intersection diagram, seismic properties used to distinguish between sandstone and mudstone facies were selected;

[0025] Among the selected seismic attributes, the attribute with the highest similarity value between its spatial variation trend and the sand body distribution map and oil-bearing area map in geological research data was chosen as the target seismic attribute.

[0026] Furthermore, a pre-defined neural network model is used to perform multi-attribute clustering and fusion of the seismic attribute values ​​of the three-dimensional structural framework model to establish a sedimentary unit model, including:

[0027] The seismic attribute values ​​of the three-dimensional structural framework model are input into a preset neural network model for multi-attribute clustering and fusion.

[0028] Adjust the parameters of the preset neural network model until the loss function converges to the preset level or reaches the predetermined number of training iterations;

[0029] The differences between the output sedimentary unit model and the well information and geological research data are compared. When the degree of conformity between the output sedimentary unit model and the well information and geological research data reaches a preset value, the adjustment is stopped, and the determined sedimentary unit model is obtained.

[0030] Furthermore, the preset neural network model includes an encoder and a decoder, wherein the encoder is represented as:

[0031] h1 = Relu(W1x + b1)

[0032] z = W2h1 + b2

[0033] Where x is the input data, z is the low-dimensional mapping of the input data, W1 is the first layer weight matrix of the encoder, b1 is the first bias, W2 is the second layer weight matrix of the encoder, and b2 is the second bias.

[0034] The loss function of this neural network model is:

[0035]

[0036] Where N is the number of samples, xi is the original input of the i-th sample, and xi′ is the corresponding reconstructed output.

[0037] Furthermore, the step of calculating the variogram between sandstone content data for each partition of the sedimentary unit model, and estimating sandstone content data at unsampled locations in the sedimentary unit model, includes:

[0038] Collect sandstone content data for each partition of the sedimentary unit model;

[0039] Calculate the semi-variance between each pair of sandstone content data;

[0040] Group the semi-mutated values ​​according to distance bandwidth;

[0041] Calculate the mean or variation of the semivariogram within each distance bandwidth to form different lag values ​​of the variation function;

[0042] Model fitting of the variation function;

[0043] Spatial interpolation was performed using the fitted variogram to estimate the sandstone content data at unsampled locations in the sedimentary unit model.

[0044] Furthermore, it also includes visualization processing of the obtained three-dimensional sandstone probabilistic volume model.

[0045] On the other hand, the present invention also discloses a device for establishing a three-dimensional sandstone probability volume, including an integrated well seismic database module, a three-dimensional structural framework model construction module, a data volume conversion module, a target seismic attribute extraction module, a sampling module, a sedimentary unit model establishment module, a variation function module, and a three-dimensional sandstone probability volume model generation module.

[0046] Based on the above technical solution, the beneficial effects of the present invention compared with the prior art are as follows:

[0047] To address the problems encountered by traditional modeling constraint methods under conditions of few wells and strong heterogeneity, this invention first establishes a well-seismic database, extracting seismic attributes such as frequency, amplitude, phase, and wave impedance. These attributes are then geologically interpreted, and seismic attributes reflecting reservoir changes are selected based on the reservoir's seismic response characteristics. Next, unsupervised artificial neural network technology is used for multi-attribute clustering and fusion, combined with prior geological knowledge, to establish a sedimentary unit model. Finally, geostatistical techniques based on the Kriging algorithm are used to establish a three-dimensional sandstone probabilistic volume. This invention establishes a multi-type, multi-attribute fused three-dimensional sandstone probabilistic volume. By assigning probability values ​​to each grid, it constrains the value type of reservoir facies and the magnitude of reservoir parameters. This invention provides reliable trend constraints for geological modeling of reservoirs with low well control and strong heterogeneity, and compared to traditional methods, it more accurately and reliably reproduces the actual reservoir geological characteristics. Attached Figure Description

[0048] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating a method for establishing a three-dimensional sandstone probability volume in Embodiment 1 of the present invention.

[0050] Figure 2 This is a flowchart illustrating step S140 in Embodiment 1 of the present invention;

[0051] Figure 3 This is a schematic diagram illustrating the correlation between the seismic properties of the envelope and porosity in Embodiment 1 of the present invention.

[0052] Figure 4 This is a flowchart illustrating step S160 in Embodiment 1 of the present invention;

[0053] Figure 5 This is a schematic diagram of the output deposition unit model in Embodiment 1 of the present invention;

[0054] Figure 6 This is a schematic diagram of a three-dimensional sandstone probabilistic volume model in Embodiment 1 of the present invention;

[0055] Figure 7 This is a schematic diagram of the lithofacies model established using the three-dimensional sandstone probabilistic volume model constraint of the present invention in Embodiment 1 of the present invention;

[0056] Figure 8 This is a schematic diagram of the porosity model established using the three-dimensional sandstone probabilistic volume model of the present invention in Embodiment 1 of the present invention;

[0057] Figure 9 This is a schematic diagram of the permeability model established using the three-dimensional sandstone probability volume model constraint of the present invention in Embodiment 1 of the present invention;

[0058] Figure 10 This is a schematic diagram of the experimental results of applying the three-dimensional sandstone probabilistic volume model of the present invention to the modeling of a reservoir in an oil field in Embodiment 1 of the present invention;

[0059] Figure 11 This is a schematic diagram of a device for establishing a three-dimensional sandstone probability volume in Embodiment 2 of the present invention. Detailed Implementation

[0060] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0061] A method for establishing a three-dimensional sandstone probabilistic volume, such as Figure 1 As shown, steps 110-180 are included:

[0062] Step 110: Integrate the well-seismic database, wherein the data in the well-seismic database includes at least well information, geological research results data, seismic results data, and well logging interpretation results data for the study area.

[0063] The well information includes at least the well location and well trajectory; the geological research data includes at least a sand body plane contour map (hereinafter referred to as: sand body distribution map) and an oil-bearing area distribution map (hereinafter referred to as: oil-bearing area map); the seismic related data includes at least two depth domain interpretation fault data, two depth domain interpretation bedding plane data, two time domain interpretation bedding plane data, and one time domain seismic data volume; the well logging interpretation data includes at least a comprehensive interpretation well logging curve (porosity, permeability, etc.) and single well lithofacies (sandstone, mudstone) data.

[0064] Step 120: Construct a three-dimensional structural framework model using the earthquake data.

[0065] In subsequent steps, to analyze various data types, including well information, seismic data, and well logging interpretation data, a three-dimensional structural framework mesh model needs to be established to accommodate these data types. In practical applications, the three-dimensional structural framework model is constructed from the two depth domain interpretation layers and two depth domain interpretation faults, for example, to present the top and bottom interfaces of an oil layer and faults in a certain area in three dimensions.

[0066] Step 130: Convert the time-domain seismic data volume into a depth-domain seismic data volume and update the well seismic database.

[0067] Since the propagation of seismic waves in strata is inevitably affected by stratum properties (lithology, physical properties, fluid properties, etc.), differences in underground reservoir properties lead to differences in reservoir wave impedance, which in turn causes seismic responses. Seismic attributes such as frequency, amplitude, phase, and wave impedance may reflect geological characteristics such as lithology, lithofacies, stratum continuity, and porosity. Based on this principle, it is necessary to analyze the relationship between seismic attributes and lithofacies, porosity, and permeability, and, in conjunction with geological research data of the study area, select multiple seismic attributes that can reflect reservoir properties. However, in practical applications, time-domain seismic data volumes are time-domain data, while well information and geological research data are depth-domain data. Therefore, it is necessary to convert the time-domain seismic data volume into a depth-domain seismic data volume, and then extract depth-domain seismic attributes to facilitate the analysis of the relationship between depth-domain seismic attributes and lithofacies, porosity, and permeability.

[0068] To convert time coordinates in seismic data into depth coordinates, the inventors established a velocity model using two time-domain interpretation layer data and two depth-domain interpretation layer data to calculate the velocities of the two layers separately, and then used linear interpolation to fill in the velocity variation between the two layers.

[0069] The preset velocity model is:

[0070]

[0071] Where V1 and V2 represent two time-domain interpretation layers, Z represents depth, and Z1 and Z2 are two depth-domain interpretation layers. interp This is a depth-domain seismic data volume transformed from a time-domain seismic data volume.

[0072] Step 140: Extract target seismic attributes from the updated well seismic database according to preset rules. The target seismic attributes can reflect reservoir changes.

[0073] Combination Figure 2 As shown, this step includes sub-steps S141-S144:

[0074] Step S141: Extract data on seismic attributes related to lithofacies, porosity, and permeability.

[0075] In some embodiments, seismic properties related to lithofacies and porosity and permeability are sweet spots, instantaneous frequency, root mean square amplitude, envelope, inversion, etc.

[0076] Step S142: After normalizing and unifying the dimensions of the earthquake attribute data, draw a spatial intersection diagram of earthquake attributes, porosity, permeability, and lithofacies.

[0077] Step S143: Select seismic properties for distinguishing sandstone facies and mudstone facies based on the spatial intersection diagram.

[0078] In some embodiments, such as Figure 3 As shown, taking the correlation between envelope seismic properties and porosity as an example, the envelope seismic property values ​​for sandstone range from 0.3 to 0.6, while those for mudstone range from 0.18 to 0.22 and 0.63 to 0.86. Therefore, envelope seismic property values ​​of 0.3 to 0.6 may correspond to sandstone facies, and different value ranges can be color-coded for easy comparison with geological research maps.

[0079] Step S144: Select the attribute with the highest similarity value between the spatial variation trend and the sand body distribution map and oil-bearing area map of the geological research data from the selected seismic attributes as the target seismic attribute.

[0080] In practical applications, similarity calculation methods can be used to compare the characteristics of seismic attributes and geological data. Example calculation steps:

[0081] Step S1441: For each seismic attribute, calculate its similarity value to the sand body distribution map using the Pearson correlation coefficient (r).

[0082]

[0083] Where Xi is the seismic attribute and Yi is the sand body content.

[0084] Step S1442, similarly, for each seismic attribute, calculate its similarity value with the oil-bearing area map, using the same similarity calculation method.

[0085] In step S1443, a weighted average is used to combine the similarity values ​​calculated in steps S1441 and S1442 to obtain a comprehensive similarity score.

[0086] By comparing the overall similarity scores of all seismic attributes, the seismic attribute with the highest similarity score was selected as the target seismic attribute. This attribute is most similar to the sand body distribution map and the oil-bearing area map in terms of spatial variation trend.

[0087] In some embodiments, the obtained target seismic attributes are sweet spot, root mean square amplitude, inversion, and instantaneous frequency.

[0088] Step 150: Sample the target seismic attributes into the three-dimensional structural frame model and assign seismic attribute values ​​to the three-dimensional structural frame model.

[0089] In some embodiments, seismic attribute data volumes of sweet spot, root mean square amplitude, inversion, and instantaneous frequency are extracted into a three-dimensional structural frame mesh model. This involves sampling the seismic attribute data volumes into the three-dimensional structural frame mesh and assigning seismic attribute values ​​to the mesh. The seismic attributes assigned to the three-dimensional structural frame mesh can effectively reflect changes in lithofacies in space.

[0090] Step 160: Use a preset neural network model to perform multi-attribute clustering and fusion on the seismic attribute values ​​of the three-dimensional structural framework model to establish a sedimentary unit model.

[0091] like Figure 4 As shown, this step specifically includes sub-steps S161-S163, specifically:

[0092] Step S161: Input the seismic attribute values ​​of the three-dimensional structural frame model into the preset neural network model to perform multi-attribute clustering fusion.

[0093] This neural network model has some advantages in multi-attribute clustering fusion:

[0094] 1. Automatic Feature Extraction: Autoencoders can automatically extract features by learning the inherent representation of data. This is very useful for multi-attribute clustering fusion because it helps the system identify relevant features in the data, thereby enabling better clustering.

[0095] 2. Dimensionality Reduction and Reconstruction: The encoder maps high-dimensional data to a low-dimensional space (z), reducing the data's dimensionality. In the reconstruction stage, the decoder maps the low-dimensional representation back to high-dimensional data, thus preserving the data's essential information. This dimensionality reduction and reconstruction process reduces redundant information and helps improve clustering results.

[0096] 3. Nonlinear modeling: Fully connected layers in the encoder and decoder can learn nonlinear relationships, which is very useful for dealing with complex data distributions and nonlinear relationships between features.

[0097] The pre-defined neural network model includes an encoder and a decoder. The encoder consists of fully connected layers and can be represented as follows:

[0098] h1 = Relu(W1x + b1)

[0099] z = W2h1 + b2

[0100] Where x is the input data, z is the low-dimensional mapping of the input data, W1 is the first layer weight matrix of the encoder, b1 is the first bias, W2 is the second layer weight matrix of the encoder, and b2 is the second bias.

[0101] These parameters are the weight matrix and bias terms of the neural network, which are learned through training the neural network model. During training, the model adjusts these parameters based on the input data (seismic attribute values) and the target data (data identical to the input data used for reconstruction) to enable the model to best reconstruct the input data.

[0102] The decoder has the opposite structure to the encoder and can use fully connected layers. The decoder decodes the low-dimensional representation z into the reconstructed feature x'. Assuming the first layer weight matrix of the decoder is W3 with a bias of b3, and the second layer weight matrix is ​​W4 with a bias of b4, the output of the decoder can be expressed as:

[0103] h2 = Relu(W3z + b3)

[0104] x'=W4h2+b4

[0105] The loss function of this neural network model is as follows, which measures the error between the reconstructed output and the original input:

[0106]

[0107] Where N is the number of samples, xi is the original input of the i-th sample, and xi′ is the corresponding reconstructed output.

[0108] Step S162: Adjust the parameters of the preset neural network model until the loss function converges to the preset level or reaches the predetermined number of training iterations.

[0109] Adjusting neural network parameters is accomplished by training the neural network model. The main goal of neural network training is to minimize the loss function, which measures the difference between the model's output and the actual target; that is, to minimize the difference between the input data and the reconstructed data.

[0110] Initialize parameters: First, the parameters of the neural network need to be initialized, including the weight matrices (W1, W2) and bias terms (b1, b2). Typically, the parameters are initialized to small random values ​​to ensure the network has sufficient randomness to begin training.

[0111] Forward Propagation: Forward propagation is performed on each training sample. This means that the input data (seismic attribute values ​​x) is passed to the encoder part (fully connected layer) to obtain the encoded representation z. Then, z is passed to the decoder part to reconstruct the data. This will generate the model's output, i.e., the reconstructed data.

[0112] Calculate the loss: Use a loss function to calculate the difference between the model output and the actual target data. In this case, the loss function is typically the mean squared error between the input data and the reconstructed data.

[0113] Backpropagation: Using the calculated loss, backpropagation is performed to compute the gradient of the parameters in order to minimize the loss function.

[0114] Parameter update: Gradient descent is used to update the parameters (W1, b1, W2, b2). Typically, multiple forward propagation, loss calculation, and backpropagation are performed on each training sample, and then the parameters are updated according to the gradient descent rule.

[0115] Iterative training: Repeat the forward propagation, loss calculation, back propagation and parameter update steps until the loss converges to a satisfactory level or the predetermined number of training iterations is reached.

[0116] Step S163: Compare the output sedimentary unit model with the well information and geological research data. When the consistency between the output sedimentary unit model and the well information and geological research data reaches a preset value, stop the adjustment and obtain the determined sedimentary unit model.

[0117] In some embodiments, the seismic attribute data of sweet spot, root mean square amplitude, inversion, and instantaneous frequency determined in step S150 are input into the preset neural network model in step S161, the parameters are adjusted, and the differences between the output sedimentary unit model and the well information and geological research results are compared until the consistency between the sedimentary unit model and the well information and geological research results reaches a preset value, at which point training stops.

[0118] The inventor's sedimentary unit model exhibits a high degree of similarity (>80%) in its distribution within low-dimensional space to geological research data, indicating that the output reflects geological understanding to a certain extent. Therefore, in this embodiment, after 1000 iterations of training, with the error limited to 10% and a preset value set to 80%, the output is a sedimentary unit model divided into 5 partitions. Figure 5 As shown, in the output sedimentary unit model, Zone I corresponds to areas with high sandstone development and high oil production, Zone II corresponds to areas with lower sandstone development and lower oil production than Zone I, and Zones III, IV, and V correspond to areas with more mudstone facies. In these three zones, the mudstone facies gradually decreases. The sedimentary unit model of this invention integrates microfacies characteristics of various seismic attributes, highlighting the spatial distribution of sedimentary facies.

[0119] Step 170: Calculate the variation function between sandstone content data for each partition of the sedimentary unit model, and estimate the sandstone content data at unsampled locations in the sedimentary unit model.

[0120] The variability function between sandstone content data is calculated for each partition of the sedimentary unit model. This function describes how the variability between data points changes with increasing distance. It is commonly used for spatial interpolation of geoscientific data to understand the spatial variability of the data.

[0121] Sandstone content data needs to be collected. Each data point is assigned to a 3D spatial model. For each pair of data points, the distance between them is calculated. If the spatial orientation between data points affects variability, the orientation angle can be calculated. For each pair of data points, the semivariogram is calculated. The semivariogram represents the spatial correlation or difference between data points. The semivariograms are grouped according to distance bandwidth. Distance bandwidth is a user-defined parameter indicating that semivariograms should be grouped into different groups within different distance ranges. The mean or variogram of the semivariograms within each distance bandwidth is calculated. These values ​​constitute the different lags of the variogram function. A model is fitted to the variogram function, using a semivariogram model (which can be an exponential model, Gaussian model, spherical model, etc.) to describe the change of semivariograms with distance. The model fit can show the characteristics of spatial variability, such as the degree and range of variation. Using the fitted variogram function, spatial interpolation is performed to estimate the sandstone content data at unsampled locations in the sedimentary unit model.

[0122] Step 180: Based on the known sandstone content distribution data in the variation function and well logging interpretation data, the Kriging algorithm is used to interpolate on the three-dimensional structural framework grid to predict the sandstone content at unknown points, thus obtaining a three-dimensional sandstone probabilistic volume model.

[0123] Using the Kriging algorithm, interpolation is performed on a 3D structural framework grid based on the known sandstone content distribution data from the variogram and well logging interpretation data to predict the sandstone content at unknown points. For each grid cell, the probability distribution of the Kriging prediction values ​​for its surrounding points is calculated. The established 3D sandstone probabilistic volume model is visualized, using color coding to represent different probability regions, providing an intuitive understanding of the uncertainty in predicting the sandstone content at unknown points.

[0124] For each unknown point in the 3D model grid cell, the Kriging interpolation method is used for prediction. Kriging is a geostatistical method used to estimate values ​​at unknown locations, taking into account the weights and distances of known points. In this step, for each unknown point, the sandstone content is estimated using the Kriging predictions from its surrounding known points.

[0125] 1. Calculate the Kriging Estimate:

[0126] For each unknown point P_i, calculate the distance d_ij between it and the surrounding known points P_j.

[0127] The kriging estimate Z_i of the unknown point P_i is calculated using the observed value Z_j of the known point P_j and the semi-variogram function γ(d_ij):

[0128] Z_i = ∑[λ_j*Z_j], where j represents all known points and λ_j is the weight coefficient.

[0129] The weighting coefficients λ_j can be determined by optimization methods (such as least squares) to minimize the mean square error between the Kriging estimate and the known point observations.

[0130] 2. Calculate the covariance:

[0131] The covariance C_ij between the unknown point P_i and the surrounding known points P_j is calculated using the semi-variogram γ(d_ij):

[0132] C_ij=Cov(Z_i,Z_j)=σ^2-(d_ij), where σ^2 is the global variance.

[0133] 3. Calculate the variance:

[0134] The variance Var_i of the unknown point P_i is calculated using the following formula:

[0135] Var_i = σ^2 - ∑[λ_j*C_ij], where j represents all known points and λ_j represents the weight coefficients.

[0136] 4. Calculate the probability distribution:

[0137] Using the Kriging estimate Z_i and variance Var_i, a probability distribution of sandstone content can be constructed. It is assumed that the sandstone content follows a normal distribution.

[0138] Mean μ_i = Z_i

[0139] Varianceσ^2_i=Var_i

[0140] The normal distribution of sandstone content at the unknown point P_i is obtained, where the mean is the Kriging estimate and the variance is the estimated variance.

[0141] 5. Visualize the probability distribution:

[0142] Using the obtained mean and variance, visualization tools can be employed to display the probability distribution. A common approach is to use color maps or contour maps to represent different probability regions, where the density of colors or lines indicates the level of probability.

[0143] a. Choose a visualization tool or software, such as a scientific computing tool (e.g., Python's Matplotlib library).

[0144] b. Create a color code for each unknown point, where the color is chosen based on the probability distribution of sandstone content at that point. Typically, cool colors can be used to represent low probability, and warm colors to represent high probability.

[0145] c. Plot these color-coded points or regions on a 3D construction framework mesh to represent areas of different probabilities. Contour lines, gradient color maps, or 3D plots can be used to present this information for a clearer understanding of the distribution of uncertainty.

[0146] Finally, the visualization results are analyzed to understand the spatial variation of the probability distribution of sandstone content. This identifies regions with high confidence levels in sandstone content estimates and regions that may have significant uncertainties.

[0147] The established 3D sandstone probabilistic volume integrates multiple types of seismic attributes and lithofacies data, with each grid having a probability value ranging from 0 to 1. This 3D sandstone probabilistic volume can be used as a trend constraint in lithofacies, porosity, and permeability modeling, improving the accuracy of reservoir modeling. Figure 6 The diagram illustrates the three-dimensional sandstone probabilistic volume model obtained in this embodiment of the invention. Regions with a probability greater than 0.5 represent high sandstone regions, while regions with a probability less than 0.5 represent low sandstone and mudstone facies. The established three-dimensional probabilistic volume model can constrain the types of rock facies values ​​and the values ​​of porosity and permeability.

[0148] To address the problems encountered by traditional modeling constraint methods under conditions of few wells and strong heterogeneity, this invention first establishes a well-seismic database, extracting seismic attributes such as frequency, amplitude, phase, and wave impedance. These attributes are then geologically interpreted, and seismic attributes reflecting reservoir changes are selected based on the reservoir's seismic response characteristics. Next, unsupervised artificial neural network technology is used for multi-attribute clustering and fusion, combined with prior geological knowledge, to establish a sedimentary unit model. Finally, geostatistical techniques based on the Kriging algorithm are used to establish a three-dimensional sandstone probabilistic volume. This invention establishes a multi-type, multi-attribute fused three-dimensional sandstone probabilistic volume. By assigning probability values ​​to each grid, it constrains the value type of reservoir facies and the magnitude of reservoir parameters. This invention provides reliable trend constraints for geological modeling of reservoirs with low well control and strong heterogeneity, and compared to traditional methods, it more accurately and reliably reproduces the actual reservoir geological characteristics.

[0149] Specifically, such as Figure 7-9 As shown, a sequential simulation algorithm was selected, and sandstone probabilistic volume constraints were applied to establish lithofacies, porosity, and permeability models. The trends of the lithofacies model and the porosity and permeability models maintained good consistency with the three-dimensional sandstone probabilistic volume, reflecting the effectiveness of the three-dimensional sandstone probabilistic constraint in controlling the models. Therefore, the three-dimensional sandstone probabilistic volume obtained in this embodiment of the invention can be directly applied to lithofacies, porosity, and permeability modeling as a trend constraint, taking into account the uncertainties of reservoir prediction and significantly improving the accuracy of reservoir modeling.

[0150] The inventors also conducted experiments using embodiments of the present invention in reservoir modeling at an oilfield in a certain region. The experimental results are as follows: Figure 10 Table 1 shows that, based on the reservoir parameter prediction results, three sweet spots were identified. In Category I, three productive wells were selected, indicating well-developed oil layers, and the actual drilling results showed 100% agreement with the model. In Category II, one evaluation well was selected, indicating well-developed reservoirs but poorly developed oil layers, and the actual drilling results showed 90% agreement with the model. Compared with traditional methods, the geological model constructed by the three-dimensional probabilistic volume constraint method of this invention more accurately and reliably reproduces the actual reservoir geological characteristics. This invention can be applied and promoted in the reservoir evaluation stage of oil and gas exploration and has reference value for three-dimensional geological modeling of other similar mineral reservoirs.

[0151] Table 1

[0152]

[0153]

[0154] In summary, the application of sandstone probabilistic volumes as constraints in 3D reservoir modeling allows for the use of 3D sandstone probabilistic volumes as a 3D trend in reservoir facies and parameter modeling. This trend-driven sequential simulation method and sequential Gaussian indicator algorithm directly control the probability of reservoir facies and parameters appearing in the grid. This approach can be applied to lithofacies, porosity, and permeability modeling, reducing the randomness and uncertainty of reservoir prediction, improving the reliability of model predictions, and ensuring the final model meets the requirements of the probabilistic volume trend distribution. This invention is suitable for oil and gas blocks with limited drilling data, primarily seismic data, strong reservoir heterogeneity, and in the exploration and evaluation stage. It also provides valuable insights for geological modeling of other mineral reservoirs.

[0155] Example 2

[0156] In other embodiments, the present invention also discloses a device for establishing a three-dimensional sandstone probabilistic volume model, including an integrated well seismic database module 10, a three-dimensional structural framework model construction module 20, a data volume conversion module 30, a target seismic attribute extraction module 40, a sampling module 50, a sedimentary unit model establishment module 60, a variation function module 70, and a three-dimensional sandstone probabilistic volume model generation module 80, wherein:

[0157] The integrated well seismic database module 10 is used to integrate a well seismic database. The data in the well seismic database includes at least well information of the study area, geological research results data, seismic results data, and well logging interpretation results data. The seismic results data includes at least two depth domain interpretation fault data, two depth domain interpretation plane data, two time domain interpretation plane data, and one time domain seismic data volume.

[0158] The three-dimensional structural framework model construction module 20 is used to construct a three-dimensional structural framework model using the earthquake results data.

[0159] In subsequent steps, to analyze various data types, including well information, seismic data, and well logging interpretation data, a three-dimensional structural framework mesh model needs to be established to accommodate these data types. In practical applications, the three-dimensional structural framework model is constructed from the two depth domain interpretation layers and two depth domain interpretation faults, for example, to present the top and bottom interfaces of an oil layer and faults in a certain area in three dimensions.

[0160] The data volume conversion module 30 is used to convert the time-domain seismic data volume into a depth-domain seismic data volume and update the well seismic database.

[0161] Since the propagation of seismic waves in strata is inevitably affected by stratum properties (lithology, physical properties, fluid properties, etc.), differences in underground reservoir properties lead to differences in reservoir wave impedance, which in turn causes seismic responses. Seismic attributes such as frequency, amplitude, phase, and wave impedance may reflect geological characteristics such as lithology, lithofacies, stratum continuity, and porosity. Based on this principle, it is necessary to analyze the relationship between seismic attributes and lithofacies, porosity, and permeability, and, in conjunction with geological research data of the study area, select multiple seismic attributes that can reflect reservoir properties. However, in practical applications, time-domain seismic data volumes are time-domain data, while well information and geological research data are depth-domain data. Therefore, it is necessary to convert the time-domain seismic data volume into a depth-domain seismic data volume, and then extract depth-domain seismic attributes to facilitate the analysis of the relationship between depth-domain seismic attributes and lithofacies, porosity, and permeability.

[0162] The target seismic attribute extraction module 40 is used to extract target seismic attributes from the updated well seismic database according to preset rules. The target seismic attributes can reflect reservoir changes.

[0163] Specifically, the first step is to extract seismic attribute data related to lithofacies, porosity, and permeability. In some embodiments, seismic attributes related to lithofacies, porosity, and permeability include sweet spots, instantaneous frequency, root-mean-square amplitude, envelope, and inversion. Then, the seismic attribute data needs to be normalized and dimensionally standardized before spatial cross-plots of seismic attributes, porosity, permeability, and lithofacies are plotted. Next, seismic attributes used to distinguish between sandstone and mudstone facies are selected based on these spatial cross-plots. Finally, from the selected seismic attributes, the attribute with the highest similarity value between its spatial variation trend and the sand body distribution map and oil-bearing area map from geological research data is chosen as the target seismic attribute.

[0164] The sampling module 50 is used to sample the target seismic attributes into the three-dimensional structural frame model and assign seismic attribute values ​​to the three-dimensional structural frame model.

[0165] In some embodiments, seismic attribute data volumes of sweet spot, root mean square amplitude, inversion, and instantaneous frequency are extracted into a three-dimensional structural frame mesh model. This involves sampling the seismic attribute data volumes into the three-dimensional structural frame mesh and assigning seismic attribute values ​​to the mesh. The seismic attributes assigned to the three-dimensional structural frame mesh can effectively reflect changes in lithofacies in space.

[0166] The sedimentary unit model establishment module 60 is used to perform multi-attribute clustering and fusion of the seismic attribute values ​​of the three-dimensional structural framework model using a preset neural network model to establish a sedimentary unit model.

[0167] The inventors hope to use an unsupervised artificial neural network to fuse four seismic attributes—sweet spot, root mean square amplitude, inversion, and instantaneous frequency earthquakes—to output a sedimentary unit model in order to discover potential relationships between the attributes.

[0168] The inventors designed a neural network model, choosing an autoencoder architecture as the neural network, including encoder and decoder parts. An encoder consisting of fully connected layers was designed to map input features to a low-dimensional representation.

[0169] Let the input feature be x, and the encoder maps the input feature to a low-dimensional representation z. Assuming the first layer weight matrix of the encoder is W1 with bias b1, and the second layer weight matrix is ​​W2 with bias b2, then the encoder output can be expressed as:

[0170] h1 = Relu(W1x + b1)

[0171] z = W2h1 + b2

[0172] The decoder is designed to decode the low-dimensional representation into the original features. The decoder's structure is the reverse of the encoder's and can use fully connected layers. The decoder decodes the low-dimensional representation z into the reconstructed feature x'. Assuming the first layer weight matrix of the decoder is W3 with a bias of b3, and the second layer weight matrix is ​​W4 with a bias of b4, the decoder output can be expressed as:

[0173] h2 = Relu(W3z + b3)

[0174] x'=W4h2+b4

[0175] Network Training: To train the network, a loss function is defined to measure the error between the reconstructed output and the original input.

[0176]

[0177] Where N is the number of samples, xi is the original input of the i-th sample, and xi′ is the corresponding reconstructed output.

[0178] The main goal of neural network training is to minimize the loss function, which measures the difference between the model's output and the actual target, i.e., to minimize the difference between the input data and the reconstructed data.

[0179] After training, the differences between the output sedimentary unit model and the well information and geological research data are compared. When the degree of conformity between the output sedimentary unit model and the well information and geological research data reaches a preset value, the adjustment is stopped, and the determined sedimentary unit model is obtained.

[0180] The variation function module 70 is used to calculate the variation function between sandstone content data for each partition of the sedimentary unit model, and to estimate the sandstone content data at unsampled locations in the sedimentary unit model.

[0181] The variability function between sandstone content data is calculated for each partition of the sedimentary unit model. This function describes how the variability between data points changes with increasing distance. It is commonly used for spatial interpolation of geoscientific data to understand the spatial variability of the data.

[0182] Data on sandstone content needs to be collected. Each data point is assigned to a 3D spatial model. For each pair of data points, the distance between them is calculated. If the spatial orientation between data points affects variability, the orientation angle can be calculated. For each pair of data points, the semivariogram is calculated. The semivariogram represents the spatial correlation or difference between data points. The semivariograms are grouped according to distance bandwidth. Distance bandwidth is a user-defined parameter indicating that semivariograms should be grouped into different groups within different distance ranges. The mean or variogram of the semivariograms within each distance bandwidth is calculated. These values ​​constitute the different lags of the variogram. A model is fitted to the variogram, using a semivariogram model (which can be an exponential model, Gaussian model, spherical model, etc.) to describe how the semivariogram changes with distance. The model fit can show the characteristics of spatial variability, such as the degree and range of variation. Using the fitted variogram, spatial interpolation is performed to estimate the values ​​at unsampled locations.

[0183] The 3D sandstone probabilistic volume model generation module 80 is used to predict the sandstone content of unknown points by interpolating on the 3D structural framework grid using the Kriging algorithm based on the known sandstone content distribution data in the variation function and well logging interpretation data, and thus obtain the 3D sandstone probabilistic volume model.

[0184] The Kriging algorithm is used to predict the sandstone content at unknown points by interpolating data on a 3D structural framework grid based on the known sandstone content distribution data from the variogram and well logging interpretation results. For each grid cell, the probability distribution of the Kriging prediction values ​​for its surrounding points is calculated. The established 3D sandstone probability model is visualized, using color coding to represent different probability regions, to intuitively understand the uncertainty in the sandstone content prediction at unknown points.

[0185] For each unknown point in the 3D model grid cell, the Kriging interpolation method is used for prediction. Kriging is a geostatistical method used to estimate values ​​at unknown locations, taking into account the weights and distances of known points. In this step, for each unknown point, the sandstone content is estimated using the Kriging predictions from its surrounding known points.

[0186] The three-dimensional sandstone probabilistic volume established in this invention integrates multiple types of seismic attributes and lithofacies data, with each grid having a probability value ranging from 0 to 1. This three-dimensional sandstone probabilistic volume can be used as a trend constraint in lithofacies, porosity, and permeability modeling, improving the accuracy of reservoir modeling.

[0187] For more detailed information on the operation of the device for establishing a three-dimensional sandstone probabilistic volume model according to the present invention, please refer to the embodiments, which will not be repeated here.

[0188] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for characterizing prior information of physical property parameters.

[0189] The present invention also provides a computer-readable storage medium storing a computer program that performs the above-described method for characterizing a priori information of gaseous property parameters.

[0190] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.

[0191] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term “comprising” as used in the specification or claims is interpreted in a manner similar to the term “including,” just as “including,” is interpreted as a conjunction in the claims. Additionally, the use of any term “or” in the specification of the claims is intended to mean “non-exclusive or.”

Claims

1. A method for establishing a three-dimensional sandstone probabilistic volume model, characterized in that, Includes the following steps: An integrated well-seismic database is provided, the data of which includes at least well information, geological research results data, seismic results data, and well logging interpretation results data for the study area. The seismic results data include at least two depth domain interpretation fault data, two depth domain interpretation plane data, two time domain interpretation plane data, and one time domain seismic data volume. A three-dimensional structural framework model was constructed using the earthquake data. The time-domain seismic data volume is converted into a depth-domain seismic data volume, and the well seismic database is updated. In the updated well seismic database, target seismic attributes are extracted according to preset rules, and these target seismic attributes can reflect reservoir changes. The target seismic attributes are sampled into the three-dimensional structural frame model, and seismic attribute values ​​are assigned to the three-dimensional structural frame model. A sedimentary unit model is established by using a pre-defined neural network model to perform multi-attribute clustering and fusion of seismic attribute values ​​from a three-dimensional structural framework model. The variation function between sandstone content data is calculated for each partition of the sedimentary unit model to estimate the sandstone content data at unsampled locations in the sedimentary unit model. Based on the known sandstone content distribution data in the variation function and well logging interpretation data, the Kriging algorithm is used to interpolate on the three-dimensional structural framework grid to predict the sandstone content at unknown points, thus obtaining a three-dimensional sandstone probabilistic volume model.

2. The method for establishing as described in claim 1, characterized in that, A three-dimensional structural framework model is constructed using the aforementioned seismic data, including: The three-dimensional structural framework model is constructed from the two depth domain interpretation layers and the two depth domain interpretation faults.

3. The method for establishing as described in claim 1, characterized in that, The process of converting the time-domain seismic data volume into depth-domain seismic data includes: The time-domain seismic data volume is transformed into a depth-domain seismic data volume using a preset velocity model, wherein the preset velocity model is: Where V1 and V2 represent two time-domain interpretation layers, Z represents depth, and Z1 and Z2 are two depth-domain interpretation layers. interp This is a depth-domain seismic data volume transformed from a time-domain seismic data volume.

4. The method for establishing as described in claim 1, characterized in that, In the updated well seismic database, target seismic attributes are extracted according to preset rules, including: Extract data on seismic attributes related to lithofacies, porosity, and permeability; After normalizing and unifying the dimensions of the earthquake attribute data, a spatial intersection diagram of earthquake attributes, porosity, permeability, and lithofacies is drawn. Based on the spatial intersection diagram, seismic properties used to distinguish between sandstone and mudstone facies were selected; Among the selected seismic attributes, the attribute with the highest similarity value between its spatial variation trend and the sand body distribution map and oil-bearing area map in geological research data was chosen as the target seismic attribute.

5. The method for establishing as described in claim 1, characterized in that, A pre-defined neural network model is used to perform multi-attribute clustering and fusion of seismic attribute values ​​from a three-dimensional structural framework model to establish a sedimentary unit model, including: The seismic attribute values ​​of the three-dimensional structural framework model are input into a preset neural network model for multi-attribute clustering and fusion. Adjust the parameters of the preset neural network model until the loss function converges to the preset level or reaches the predetermined number of training iterations; The differences between the output sedimentary unit model and the well information and geological research data are compared. When the degree of conformity between the output sedimentary unit model and the well information and geological research data reaches a preset value, the adjustment is stopped, and the determined sedimentary unit model is obtained.

6. The method for establishing as described in claim 5, characterized in that, The preset neural network model includes an encoder and a decoder, wherein the encoder is represented as: h1 = Relu(W1x + b1) z = W2h1 + b2 Where x is the input data, z is the low-dimensional mapping of the input data, W1 is the first layer weight matrix of the encoder, b1 is the first bias, W2 is the second layer weight matrix of the encoder, and b2 is the second bias. The loss function of this neural network model is: Where N is the number of samples, xi is the original input of the i-th sample, and xi′ is the corresponding reconstructed output.

7. The method for establishing as described in claim 1, characterized in that, The calculation of the variation function between sandstone content data for each partition of the sedimentary unit model, and the estimation of sandstone content data at unsampled locations in the sedimentary unit model, includes: Collect sandstone content data for each partition of the sedimentary unit model; Calculate the semi-variance between each pair of sandstone content data; Group the semi-mutated values ​​according to distance bandwidth; Calculate the mean or variation of the semivariogram within each distance bandwidth to form different lag values ​​of the variation function; Model fitting of the variation function; Spatial interpolation was performed using the fitted variogram to estimate the sandstone content data at unsampled locations in the sedimentary unit model.

8. The method for establishing as described in claim 1, characterized in that, It also includes visualization processing of the obtained three-dimensional sandstone probabilistic volume model.

9. A device for establishing a three-dimensional sandstone probabilistic volume model, characterized in that, It includes an integrated well seismic database module, a 3D structural framework model construction module, a data volume conversion module, a target seismic attribute extraction module, a sampling module, a sedimentary unit model establishment module, a variation function module, and a 3D sandstone probabilistic volume model generation module, among which: The integrated well seismic database module is used to integrate a well seismic database. The data in the well seismic database includes at least well information of the study area, geological research results data, seismic results data, and well logging interpretation results data. The seismic results data includes at least two depth domain interpretation fault data, two depth domain interpretation plane data, two time domain interpretation plane data, and one time domain seismic data volume. The three-dimensional structural framework model construction module is used to construct a three-dimensional structural framework model using the seismic data. The data volume conversion module is used to convert the time-domain seismic data volume into a depth-domain seismic data volume and update the well seismic database. The target seismic attribute extraction module is used to extract target seismic attributes from the updated well seismic database according to preset rules. The target seismic attributes can reflect reservoir changes. The sampling module is used to sample the target seismic attributes into the three-dimensional structural frame model and assign seismic attribute values ​​to the three-dimensional structural frame model; The sedimentary unit model building module is used to perform multi-attribute clustering and fusion of the seismic attribute values ​​of the three-dimensional structural framework model using a preset neural network model to build a sedimentary unit model. The variation function module is used to calculate the variation function between sandstone content data for each partition of the sedimentary unit model, and to estimate the sandstone content data at unsampled locations in the sedimentary unit model. The three-dimensional sandstone probabilistic volume model generation module is used to predict the sandstone content at unknown points by interpolating on a three-dimensional structural framework grid using the kriging algorithm based on the known sandstone content distribution data in the variation function and well logging interpretation data, thereby obtaining a three-dimensional sandstone probabilistic volume model.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements any of the methods described in claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that performs the method of any one of claims 1 to 8.

Images