Method and apparatus for reservoir geologic modeling based on full-space variogram
By constructing and training a full-space variation function model, the problem of parameter determination in reservoir geological modeling was solved, the accuracy and predictive performance of the model were improved, and accurate modeling of complex reservoirs was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-02-22
- Publication Date
- 2026-03-31
AI Technical Summary
Existing geostatistical reservoir modeling methods struggle to determine the relevant parameters of the variogram when there are significant differences in river flow direction and complex reservoir deposition, resulting in large deviations and low accuracy in prediction results.
By acquiring hard data on reservoir attribute interpretation from single wells and soft data on reservoir attribute trends between wells, an initial full-space variation function model is constructed. Cross-validation is then used to train the target full-space variation function model until a preset accuracy is achieved, which is then used to construct the target reservoir geological model.
It improves the accuracy and generalization ability of reservoir geological models, and realizes accurate modeling of the spatial distribution of continuous and discrete reservoir attributes.
Smart Images
Figure CN117743841B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of reservoir geological modeling and artificial intelligence technology, specifically to a method and apparatus for reservoir geological modeling based on full-space variogram. Background Technology
[0002] Two-point geostatistical algorithms generate experimental variograms based on the semivariogram of spatial reservoir properties between two points. These experimental variograms can then be used to determine theoretical variograms, which are crucial tools for measuring the spatial correlation of reservoir properties. Using the theoretical variograms and known single-well reservoir data, reservoir properties at inter-well locations can be determined. Commonly used two-point geostatistical reservoir modeling methods include simple kriging, ordinary kriging, generalized kriging, indicator kriging, co-kriging, sequential Gaussian simulation, and co-sequential Gaussian simulation.
[0003] Two-point geostatistical algorithms have good predictive capabilities for isotropic and anisotropic strata with dominant orientations. Reservoir geological modeling based on two-point geostatistics requires manual setting of parameters such as principal variation direction, tolerance angle, and bandwidth. It has good predictive performance for channel deposits with a single flow direction, but when there are large differences in channel flow direction and complex reservoir deposits, it is difficult to determine the parameters related to the variation function in two-point geostatistics, resulting in significant deviations in the prediction results. Summary of the Invention
[0004] The purpose of this application is to provide a method and apparatus for reservoir geological modeling based on the full space variation function, so as to solve the problem of low prediction accuracy of reservoir geological models in the prior art.
[0005] To achieve the above objectives, the first aspect of this application provides a method for reservoir geological modeling based on a full-space variation function, the method comprising:
[0006] Acquire hard data for interpreting reservoir attributes in a single well and soft data for inter-well reservoir attribute trends;
[0007] The hard data for interpreting reservoir attributes of a single well and the soft data for the trend of reservoir attributes between wells are divided into grids using well location coordinates to obtain well data;
[0008] The well data is divided into training and testing datasets;
[0009] Construct an initial full-space variation function model;
[0010] The initial full-space variation function model is trained using cross-validation on the training dataset to obtain the target full-space variation function model.
[0011] Determine the accuracy of the target full-space variation function model based on the test dataset;
[0012] If the accuracy is less than or equal to the preset accuracy, continue training the target full-space variation function model until the accuracy is greater than the preset accuracy.
[0013] If the accuracy is greater than the preset accuracy, the target full-space variation function model is used to construct the target reservoir geological model.
[0014] In this embodiment, the initial full-space variation function model satisfies formula (1):
[0015] (1)
[0016] in, The point-to-half variance of the well data. For point-to-point quantity, For any pair of points, Represents a point-to-parameter vector ( ), Indicates the angle of a point relative to a direction. Indicates the distance between points. This represents the dimension of the point-to-parameter vector. Indicates model parameters.
[0017] In this embodiment of the application, the method further includes:
[0018] The preset data table is constructed based on well data and the corresponding actual point pair vectors, actual point pair orientation angles, actual point pair distances, and actual point pair variances.
[0019] In this embodiment of the application, training an initial full-space variability model using cross-validation on the training dataset to obtain a target full-space variability model includes:
[0020] Divide the training dataset into a predetermined number of sub-training datasets;
[0021] Training groups are determined based on a preset number of sub-training datasets. For each training group, any sub-training dataset is used as validation data, and the other sub-training datasets are used as training data.
[0022] For any training set, the training data is input into the initial full-space variation function model to obtain the half variance of the predicted points of the training data;
[0023] Determine the actual point-to-half variance of the training data based on the preset data table;
[0024] The mean square error is calculated by combining the half variance of the predicted points and the half variance of the actual points in the training data to obtain the training loss.
[0025] If the training loss is less than the preset loss value, the validation loss is determined based on the validation data, and the next set of training is performed until all sets of training are completed.
[0026] In this embodiment of the application, the training loss satisfies formula (2):
[0027] (2)
[0028] in, For training loss, For training data, The variance of the predicted points for the training data is halved. The actual half-variance of the training data. For point-to-point quantity.
[0029] In this embodiment of the application, the verification loss satisfies formula (3):
[0030] (3)
[0031] in, To verify the loss, To verify the data, To verify the half-variance of the predicted points in the data, To verify the actual half-variance of the data, For point-to-point quantity.
[0032] In this embodiment of the application, training an initial full-space variation function model using cross-validation on the training dataset to obtain a target full-space variation function model further includes:
[0033] If the training loss is greater than the preset loss value, the initial full-space variation function model is optimized to obtain the optimized parameters.
[0034] In this embodiment of the application, the optimized parameters satisfy formula (4):
[0035] (4)
[0036] in, For the optimized parameters, These are the parameters before optimization. For learning rate, To obtain the gradient of the parameters before optimization, For training data, The variance of the predicted points for the training data is halved. The actual half-variance of the training data. For point-to-point quantity.
[0037] A second aspect of this application provides an apparatus for reservoir geological modeling based on full-space variation functions, comprising:
[0038] The memory is configured to store instructions; and
[0039] The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the above-described method for reservoir geological modeling based on the full-space variation function.
[0040] A third aspect of this application provides a machine-readable storage medium storing instructions for causing a machine to perform the above-described method for reservoir geological modeling based on the full-space variation function.
[0041] The above technical solution obtains well data based on hard data of single-well reservoir attribute interpretation and soft data of inter-well reservoir attribute trends. This well data is then divided into training and testing datasets. The training dataset is first used to train an initial full-space variogram model using cross-validation to obtain a target full-space variogram model. The accuracy of the target full-space variogram model is then judged based on the testing dataset. If the accuracy is less than or equal to a preset accuracy, the target full-space variogram model is trained further until the accuracy exceeds the preset accuracy. If the accuracy exceeds the preset accuracy, the target full-space variogram model is used to construct the target reservoir geological model. This application establishes a full-space variogram model based on hard data of single-well reservoir attribute interpretation and soft data of inter-well reservoir attribute trends, enabling reservoir geological modeling of continuous and discrete reservoir attribute spatial distributions. The introduction of a conditional information posterior mechanism effectively improves the generalization ability and predictive performance of the reservoir geological model, thereby increasing its accuracy.
[0042] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description
[0043] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:
[0044] Figure 1 A flowchart illustrating a method for reservoir geological modeling based on a full-space variation function according to an embodiment of this application is shown schematically.
[0045] Figure 2 The diagram illustrates the construction of a training dataset and a test dataset according to an embodiment of this application.
[0046] Figure 3A schematic diagram illustrating a full-space variation function according to an embodiment of this application is shown.
[0047] Figure 4 The flowchart illustrates a method for reservoir geological modeling based on a full-space variation function according to a specific embodiment of this application.
[0048] Figure 5 The illustration schematically shows an image reconstruction result of a target reservoir geological model according to an embodiment of this application;
[0049] Figure 6 The diagram schematically illustrates a structural block diagram of an apparatus for reservoir geological modeling based on a full-space variation function according to an embodiment of this application. Detailed Implementation
[0050] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0051] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.
[0052] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.
[0053] Figure 1 An illustration shows an embodiment according to this application. Figure 1 A flowchart illustrating a method for reservoir geological modeling based on a full-space variation function according to an embodiment of this application is shown schematically. Figure 1As shown in the embodiments of this application, a method for reservoir geological modeling based on a full-space variation function is provided, which may include:
[0054] Step 101: Obtain hard data for interpreting reservoir attributes in a single well and soft data for the trend of reservoir attributes between wells.
[0055] In this embodiment, hard data refers to the actual, objectively existing data of reservoir attributes in a single well, which can be obtained by coupling conventional well logging interpretation and core description, such as sedimentary facies data. Soft data refers to the trend data of reservoirs between wells, such as seismic data, which can be obtained through seismic data interpretation and seismic data inversion. The hard data of single-well reservoir attribute interpretation and the soft data of reservoir attribute trends between wells together constitute the initial well data.
[0056] Step 102: Divide the hard data of single-well reservoir attribute interpretation and the soft data of inter-well reservoir attribute trend into grids using well location coordinates to obtain well data.
[0057] In this embodiment, after acquiring the hard data for interpreting reservoir attributes in a single well and the soft data for reservoir attribute trends between wells, the processor needs to fuse these two datasets. First, the processor needs to construct a reservoir grid based on the study area. It then divides the hard data D for interpreting reservoir attributes in a single well and the soft data d for reservoir attribute trends between wells using well coordinates, thereby establishing a spatial data model. This makes the data in this embodiment more comprehensive, thereby improving the accuracy of the target reservoir geological model.
[0058] Step 103: Divide the well data into training dataset and test dataset.
[0059] In this embodiment, the processor can divide well data into a training dataset and a test dataset. The training dataset is used to train the target full-space variability function model, and the test dataset is used to determine the accuracy of the target full-space variability function model, thereby determining the target reservoir geological model. Figure 2 The diagram illustrates the construction of a training dataset and a test dataset according to an embodiment of this application. Figure 2 As shown, in one example, the processor can randomly select R (e.g., 80%) of the well data as the training dataset D1, and the data from 1 to R as the test dataset, denoted as D2. Furthermore, the training dataset D1 can be divided into training data and validation data. The processor divides the training dataset D1 into K mutually exclusive subsets of equal size for K-fold cross-validation, taking K-1 subsets as training data each time, denoted as D... 11 The remaining data is used as verification data, denoted as D. 12 .
[0060] Step 104: Construct the initial full-space variation function model.
[0061] In this embodiment, the initial full-space variogram model F constructed by the processor establishes a mapping relationship between the input point-pair orientation angles and point-pair distances and the output of the initial full-space variogram model based on the Gaussian kernel density estimation function. Existing technologies only determine the point-pair semi-variance based on the point-pair distances of the data, while the inputs to the initial full-space variogram model F in this embodiment are the point-pair orientation angles and point-pair distances. The point-pair orientation angles can be in two-dimensional, three-dimensional, or even multi-dimensional space. Therefore, the initial full-space variogram model in this embodiment can be used in multiple dimensions, thus having a wider range of applications.
[0062] Step 105: Train the initial full-space variation function model using cross-validation on the training dataset to obtain the target full-space variation function model.
[0063] In this embodiment, cross-validation involves dividing the training dataset into a predetermined number of sub-training datasets. One portion is used as training data, and the other as validation data. The target full-space variability function model is trained using the training data, and then validated using the validation data. This process is repeated cyclically. This improves the model's generalization ability. The target full-space variability function model is a full-space variability function model obtained through training on the training dataset and is not equivalent to the final reservoir geological model. In this embodiment, each data point in the training dataset includes input well data and the corresponding half-variable variance of the actual points. In one example, the processor can divide the training dataset into K mutually exclusive sub-training datasets of equal size, thereby determining K groups of sub-training datasets. For each group of sub-training datasets, one training dataset is used as validation data, and the other sub-training datasets are used as training data. Therefore, the number of cross-validations is K, and the validation data is different each time. For each training set, inputting the training data into the initial full-space variability function model yields the half-variable variance of the predicted points corresponding to the training data. Then, the training loss is determined based on the half-variance of predicted points and the half-variance of actual points. If the loss is less than the preset loss value, it means that the training results of this group meet the preset conditions. The validation loss can then be determined using validation data, and the training of the next group can begin, until all groups have been trained. Training the initial full-space variation function model using cross-validation can improve the generalization ability of the target full-space variation function model.
[0064] Step 106: Determine the accuracy of the target full-space variation function model based on the test dataset;
[0065] Step 107: If the accuracy is less than or equal to the preset accuracy, continue training the target full-space variation function model until the accuracy is greater than the preset accuracy.
[0066] Step 108: If the accuracy is greater than the preset accuracy, use the target full-space variation function model for the construction of the target reservoir geological model.
[0067] In this embodiment, the test dataset includes multiple test data sets. After the processor obtains the target full-space variation function model, it also needs to evaluate the modeling effect of the target full-space variation function model based on the test dataset. In one example, the test dataset can be input into the target full-space variation function model to obtain the half-variance of the predicted points of the test dataset. Based on the half-variance of the predicted points and the half-variance of the actual points of the test data, it is determined whether the current target full-space variation function model predicts accurately, and the accuracy of the current target full-space variation function model is determined. Wherein, accuracy = N true / N D2 N true The number of correctly predicted values for the test data; N D2 This represents the total number of test data.
[0068] In this embodiment, the preset accuracy rate is used to determine whether the target full-space variation function model needs further training. If the accuracy rate is greater than the preset accuracy rate, it indicates that the current target full-space variation function model can be used for the final target reservoir geological model construction and further training is unnecessary. However, if the accuracy rate is less than or equal to the preset accuracy rate, it indicates that the current reservoir geological model needs further optimization. Therefore, it is necessary to return to step 103 and re-model until the accuracy rate of the target full-space variation function model is greater than the preset accuracy rate, thus completing the establishment of the target reservoir geological model.
[0069] This application's embodiments obtain well data based on hard data of single-well reservoir attribute interpretation and soft data of inter-well reservoir attribute trends. The well data is then divided into training and testing datasets. First, an initial full-space variogram model is trained on the training dataset using cross-validation to obtain a target full-space variogram model. Then, the accuracy of the target full-space variogram model is judged based on the testing dataset. If the accuracy is less than or equal to a preset accuracy, the target full-space variogram model continues to be trained until the accuracy exceeds the preset accuracy. If the accuracy exceeds the preset accuracy, the target full-space variogram model is used to construct the target reservoir geological model. This application's embodiments establish a full-space variogram model based on hard data of single-well reservoir attribute interpretation and soft data of inter-well reservoir attribute trends, realizing reservoir geological modeling of continuous and discrete reservoir attribute spatial distributions. The introduction of a conditional information posterior mechanism effectively improves the generalization ability and predictive performance of the reservoir geological model, thereby increasing its accuracy.
[0070] In this embodiment of the application, the method may further include:
[0071] The preset data table is constructed based on well data and the corresponding actual point pair vectors, actual point pair orientation angles, actual point pair distances, and actual point pair variances.
[0072] Specifically, the preset data table is a table consisting of well data and the corresponding actual values. Here, the actual point-to-vector is the vector difference between the two points, the actual point-to-direction angle is the angle between the vector differences of the two points, the actual point-to-distance is the Euclidean distance between the two points, and the actual point-to-semivariance is half the variance of the two points. By constructing the preset data table, it is easy to compare it with the predicted values output by the model, thereby determining the accuracy of the model.
[0073] In this embodiment, the initial full-space variation function model can satisfy formula (1):
[0074] (1)
[0075] in, The point-to-half variance of the well data. For point-to-point quantity, For any pair of points, Represents a point-to-parameter vector ( ), Represents the direction angle of a point (in two-dimensional space: ; Three-dimensional space situation: ), Indicates the distance between points. The dimension of the point-to-parameter vector ( =2 or 3), Indicates model parameters, .
[0076] Specifically, the initial full-space variogram model F constructed by the processor establishes a mapping relationship between the input point-pair orientation angles and point-pair distances and the output of the initial full-space variogram model based on the Gaussian kernel density estimation function. Existing technologies only determine the point-pair semi-variance based on the point-pair distances of the data, while the inputs to the initial full-space variogram model F in this embodiment are the point-pair orientation angles and point-pair distances, where the point-pair orientation angles can be in two-dimensional space, three-dimensional space, or even multi-dimensional space. Figure 3 A schematic diagram illustrating a full-space variation function graph according to an embodiment of this application is shown. Figure 3 As shown, the full-space variogram is a graph of the initial full-space variogram model F. The full-space variogram can reflect the functional relationship between the point-to-point orientation angle and the point-to-point distance and the point-to-point semi-variance, and can be applied to two-dimensional or three-dimensional applications. Therefore, the initial full-space variogram model of this embodiment can be used in multiple dimensions, thus having a wider range of applications.
[0077] In this embodiment of the application, step 105, training an initial full-space variation function model using cross-validation on the training dataset to obtain a target full-space variation function model, may include:
[0078] Divide the training dataset into a predetermined number of sub-training datasets;
[0079] Training groups are determined based on a preset number of sub-training datasets. For each training group, any sub-training dataset is used as validation data, and the other sub-training datasets are used as training data.
[0080] For any training set, the training data is input into the initial full-space variation function model to obtain the half variance of the predicted points of the training data;
[0081] Determine the actual point-to-half variance of the training data based on the preset data table;
[0082] The mean square error is calculated by combining the half variance of the predicted points and the half variance of the actual points in the training data to obtain the training loss.
[0083] If the training loss is less than the preset loss value, the validation loss is determined based on the validation data, and the next set of training is performed until all sets of training are completed.
[0084] Specifically, cross-validation involves dividing the training dataset into a predetermined number of sub-training datasets. One part is used as training data, and the other part as validation data. The target full-space variation function model is trained using the training data, and then validated using the validation data. This process is repeated cyclically. This improves the model's generalization ability. The target full-space variation function model is a reservoir geological model obtained by training the training dataset. In this embodiment, each data point in the training dataset includes input well data and the corresponding actual point variance. The processor can divide the training dataset into K mutually exclusive sub-training datasets of equal size, thus determining K groups of sub-training datasets. For each group of sub-training datasets, one training dataset is used as validation data, and the other sub-training datasets are used as training data. Therefore, the number of cross-validations is K, and the validation data is different each time. For each training set, the training data is input into the initial full-space variation function model to obtain the predicted point variance corresponding to the training data. Then, the training loss is determined based on the predicted point variance and the actual point variance. The training loss is set when it is less than a predetermined loss value. If the training results of a given set meet the preset conditions, the validation loss can be determined using validation data, and the training of the next set can begin, continuing until all sets have been trained. Training the initial full-space variogram model using cross-validation can improve the generalization ability of the target full-space variogram model.
[0085] In this embodiment of the application, the training loss can satisfy formula (2):
[0086] (2)
[0087] in, For training loss, For training data, For training data The half variance of the predicted points, which represents the training data. The initial full-space variation function model values corresponding to the point-to-point orientation angle and point-to-point distance. For training data The actual point variance of half For point-to-point pairs. If Then, the current initial full-space variation function model also needs to be optimized. If the current group is not properly trained, the validation loss is calculated so that the training of the next group can begin.
[0088] In this embodiment of the application, the verification loss can satisfy formula (3):
[0089] (3)
[0090] in, To verify the loss, To verify the data, To verify the half-variance of the predicted points in the data, To verify the actual half-variance of the data, For point-to-point quantity.
[0091] In this embodiment of the application, step 105, training an initial full-space variation function model using cross-validation on the training dataset to obtain a target full-space variation function model, may further include:
[0092] If the training loss is greater than the preset loss value, the initial full-space variation function model is optimized to obtain the optimized parameters.
[0093] Specifically, when the training loss is greater than the preset loss value In this case, it means that the current initial full-space variation function model still needs to be optimized.
[0094] In this embodiment of the application, the optimized parameters can satisfy formula (4):
[0095] (4)
[0096] in, For the optimized parameters, These are the parameters before optimization. For learning rate, To obtain the gradient of the parameters before optimization, For training data, The variance of the predicted points for the training data is halved. The actual half-variance of the training data. For point-to-point quantity.
[0097] Figure 4 The diagram schematically illustrates a flowchart of a reservoir geological modeling method based on a full-space variation function according to a specific embodiment of this application. Figure 4 As shown in one specific embodiment, a method for reservoir geological modeling based on a full-space variation function is provided, which may include:
[0098] S1. Obtain hard data for interpreting reservoir properties in a single well;
[0099] S2. Obtain soft data on reservoir attribute trends between wells;
[0100] S3. Fusion of hard data on single-well reservoir attribute interpretation and soft data on inter-well reservoir attribute trends;
[0101] S4. Divide the well data into training dataset D1 and test dataset D2;
[0102] S5. Train the initial full-space variation function model through cross-validation to obtain the target full-space variation function model;
[0103] S6. Evaluate the accuracy of the target full-space variation function model and determine whether the accuracy meets the conditions. If yes, proceed to S7; otherwise, return to S3.
[0104] S7. Modeling is complete, and the geological model of the target reservoir is obtained.
[0105] S5 uses K-fold difference validation to partition the training dataset, resulting in training data D. 11 and verification data D 12 The actual half-variance of the training data is obtained from the point-pair data table (i.e., the preset data table) corresponding to the training data, thereby determining the training loss of the training data. It is then determined whether the parameters of the current initial full-space variation function model need to be updated. If not, the validation loss is determined based on the validation data, and the next iteration is initiated. Otherwise, the parameters are optimized.
[0106] This application establishes a full-space variation function model of hard data for interpreting reservoir attributes in single wells and soft data for the trend of reservoir attributes between wells, realizing reservoir geological modeling of the spatial distribution of continuous and discrete reservoir attributes. It introduces a posterior mechanism of conditional information to effectively improve the generalization ability and predictive performance of the reservoir geological model, thereby improving the accuracy of the reservoir geological model.
[0107] Figure 5 The illustration schematically shows an image reconstruction result of a target reservoir geological model according to an embodiment of this application. For example... Figure 5 As shown, reservoir geological modeling is performed using data from a specific well as an example. Based on the conditions for creating hard data on single-well reservoir attributes and soft data on inter-well reservoir attribute trends, spatial fusion is performed using x, y, and z spatial well coordinates and corresponding porosity, permeability, oil saturation, or clay content values. The well data is then processed by constructing a target full-space variogram model from the initial full-space variogram model and optimizing its parameters. A verification mechanism for the target full-space variogram model is then established to obtain the target reservoir geological model. Finally, the target reservoir geological model is used for reservoir geological modeling to obtain the generated geological image. Figure 5 As can be seen, the embodiments of this application can generate reservoir geological models well for some conditional well data. Examples demonstrate that the method of the embodiments of this application can establish reservoir geological models relatively accurately.
[0108] Figure 6A schematic diagram illustrates the structural block of an apparatus for reservoir geological modeling based on a full-space variation function according to an embodiment of this application. Figure 6 As shown in the figure, this application provides an apparatus for reservoir geological modeling based on full-space variation functions, which may include:
[0109] Memory 610 is configured to store instructions; and
[0110] Processor 620 is configured to retrieve the instructions from memory 610 and, when executing the instructions, to implement the above-described method for reservoir geological modeling based on the full-space variation function.
[0111] Specifically, in this embodiment of the application, the processor 620 can be configured to:
[0112] Acquire hard data for interpreting reservoir attributes in a single well and soft data for inter-well reservoir attribute trends;
[0113] The hard data for interpreting reservoir attributes of a single well and the soft data for the trend of reservoir attributes between wells are divided into grids using well location coordinates to obtain well data;
[0114] The well data is divided into training and testing datasets;
[0115] Construct an initial full-space variation function model;
[0116] The initial full-space variation function model is trained using cross-validation on the training dataset to obtain the target full-space variation function model.
[0117] Determine the accuracy of the target full-space variation function model based on the test dataset;
[0118] If the accuracy is less than or equal to the preset accuracy, continue training the target full-space variation function model until the accuracy is greater than the preset accuracy.
[0119] If the accuracy is greater than the preset accuracy, the target full-space variation function model is used to construct the target reservoir geological model.
[0120] In this embodiment, the initial full-space variation function model satisfies formula (1):
[0121] (1)
[0122] in, The point-to-half variance of the well data. For point-to-point quantity, For any pair of points, Represents a point-to-parameter vector ( ), Indicates the angle of a point relative to a direction. Indicates the distance between points. This represents the dimension of the point-to-parameter vector. Indicates model parameters.
[0123] Furthermore, the processor 620 can also be configured as follows:
[0124] The preset data table is constructed based on well data and the corresponding actual point pair vectors, actual point pair orientation angles, actual point pair distances, and actual point pair variances.
[0125] Furthermore, the processor 620 can also be configured as follows:
[0126] The initial full-space variation function model is trained using cross-validation on the training dataset to obtain the target full-space variation function model, which includes:
[0127] Divide the training dataset into a predetermined number of sub-training datasets;
[0128] Training groups are determined based on a preset number of sub-training datasets. For each training group, any sub-training dataset is used as validation data, and the other sub-training datasets are used as training data.
[0129] For any training set, the training data is input into the initial full-space variation function model to obtain the half variance of the predicted points of the training data;
[0130] Determine the actual point-to-half variance of the training data based on the preset data table;
[0131] The mean square error is calculated by combining the half variance of the predicted points and the half variance of the actual points in the training data to obtain the training loss.
[0132] If the training loss is less than the preset loss value, the validation loss is determined based on the validation data, and the next set of training is performed until all sets of training are completed.
[0133] In this embodiment of the application, the training loss satisfies formula (2):
[0134] (2)
[0135] in, For training loss, For training data, The variance of the predicted points for the training data is halved. The actual half-variance of the training data. For point-to-point quantity.
[0136] In this embodiment of the application, the verification loss satisfies formula (3):
[0137] (3)
[0138] in, To verify the loss, To verify the data, To verify the half-variance of the predicted points in the data, To verify the actual half-variance of the data, For point-to-point quantity.
[0139] Furthermore, the processor 620 can also be configured as follows:
[0140] Training the initial full-space variation function model using cross-validation on the training dataset to obtain the target full-space variation function model also includes:
[0141] If the training loss is greater than the preset loss value, the initial full-space variation function model is optimized to obtain the optimized parameters.
[0142] In this embodiment of the application, the optimized parameters satisfy formula (4):
[0143] (4)
[0144] in, For the optimized parameters, These are the parameters before optimization. For learning rate, To obtain the gradient of the parameters before optimization, For training data, The variance of the predicted points for the training data is halved. The actual half-variance of the training data. For point-to-point quantity.
[0145] This application embodiment establishes a full-space variation function model of single-well reservoir attribute interpretation hard data and inter-well reservoir attribute trend soft data to realize reservoir geological modeling of continuous and discrete reservoir attribute spatial distribution. The introduction of a conditional information posterior mechanism effectively improves the generalization ability and prediction performance of the reservoir geological model, thereby improving the accuracy of the reservoir geological model.
[0146] This application also provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described method for reservoir geological modeling based on the full-space variation function.
[0147] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0148] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0149] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0150] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0151] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0152] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0153] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0154] It should also be noted that 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 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.
[0155] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for reservoir geologic modeling based on a full-space variogram, characterized in that, The method comprises: obtaining single-well reservoir attribute interpretation hard data and interwell reservoir attribute trend soft data; performing grid division on the single-well reservoir attribute interpretation hard data and the interwell reservoir attribute trend soft data through well location coordinates to obtain well data; constructing a preset data table according to the well data and actual point pair vectors, actual point pair directions, actual point pair distances and actual point pair semivariograms corresponding to the well data; dividing the well data into a training data set and a test data set; constructing an initial full-space variogram model; training the initial full-space variogram model through cross-validation of the training data set to obtain a target full-space variogram model; judging an accuracy of the target full-space variogram model according to the test data set; in a case where the accuracy is less than or equal to a preset accuracy, continuing to train the target full-space variogram model until the accuracy is greater than the preset accuracy; in a case where the accuracy is greater than the preset accuracy, determining the target full-space variogram model as a target reservoir geological model; wherein the initial full-space variogram model satisfies formula (1): ；（1） wherein, is the point-to-semivariance of the well data, is the number of point pairs, is an arbitrary point pair, denotes a point pair parameter vector (p), ), denotes a point pair direction angle, denotes a point pair distance, denotes a point pair parameter vector dimension, denotes a model parameter; the training of the initial full-space variogram model through cross-validation of the training data set to obtain a target full-space variogram model comprises: dividing the training data set into a preset number of sub-training data sets; determining training groups based on the preset number of sub-training data sets, wherein for each training group, any sub-training data set is taken as validation data and other sub-training data sets are taken as training data; for any training group, inputting the training data into the initial full-space variogram model to obtain predicted point pair semivariograms of the training data; determining actual point pair semivariograms of the training data according to the preset data table; performing mean square error calculation on the predicted point pair semivariograms and the actual point pair semivariograms of the training data to obtain a training loss; in a case where the training loss is less than a preset loss value, determining a validation loss according to the validation data and performing next group training until training of all groups is completed.
2. The method of claim 1, wherein, the training loss satisfies formula (2): ；（2） wherein, is the training loss, is the training data, is the predicted point-to-half variance of the training data, is the actual point-to-half variance of the training data, is the number of point pairs.
3. The method of claim 1, wherein, the validation loss satisfies formula (3): ；（3） wherein, is the validation loss, is the validation data, is the predicted point-to-half variance of the validation data, is the actual point-to-half variance of the validation data, is the number of point pairs.
4. The method of claim 1, wherein, the training of the initial full-space variogram model through cross-validation of the training data set to obtain a target full-space variogram model further comprises: in a case where the training loss is greater than the preset loss value, optimizing the initial full-space variogram model to obtain optimized parameters.
5. The method of claim 4, wherein, the optimized parameters satisfy formula (4): ；（4） wherein, are optimized parameters, are parameters before optimization, is a learning rate, is gradient calculation on the parameters before optimization, is training data, is a predicted point-to-half variance of the training data, is an actual point-to-half variance of the training data, is a point pair number.
6. An apparatus for reservoir geologic modeling based on a full-space variogram, characterized in that, comprise: a memory configured to store instructions; and a processor configured to call the instructions from the memory and capable of implementing the full-space variogram-based reservoir geological modeling method according to any one of claims 1 to 5 when the instructions are executed.
7. A machine-readable storage medium, characterized in that, The machine-readable storage medium has instructions stored thereon for causing a machine to perform the full-space variogram-based reservoir geological modeling method according to any one of claims 1 to 5.