Shear wave velocity prediction method, device and equipment

Abstract

The application provides a shear wave velocity prediction method, device and equipment, the method comprises the following steps: for each known well, determining a plurality of target logging parameters from all logging parameters according to the first data set of the known well, the influence degree of the target logging parameter on the shear wave velocity of the known well is greater than that of other logging parameters; according to the second data set of the known well, a decision tree model of the known well is constructed; input the to-be-predicted data set of the to-be-predicted well into the decision tree model of the known well, and obtain the initial shear wave velocity output by the decision tree model, the to-be-predicted data set comprises the parameter value of each target logging parameter of the to-be-predicted well at each preset depth; according to the initial shear wave velocity output by the known well in each decision tree model, the target shear wave velocity of the to-be-predicted well is generated. The method of the application effectively improves the accuracy and prediction efficiency of the shear wave velocity prediction.

Description

Technical Field

[0001] This application relates to the field of oil and gas exploration, and in particular to a method, apparatus and equipment for predicting shear wave velocity. Background Technology

[0002] In oil and gas exploration and development, P-wave and S-wave velocities provide crucial subsurface information, with S-wave velocity being a key parameter in rock physics research. It is beneficial for identifying pore fluids and reservoir properties, playing a vital role in pre-stack inversion and amplitude variation with offset (AVO) property analysis. Combined P-wave and S-wave velocity inversion helps reduce reservoir ambiguity. Therefore, obtaining S-wave velocity data is crucial for geophysical exploration.

[0003] Existing methods for obtaining shear wave velocity are typically prediction-based, primarily utilizing data-driven machine learning to predict shear wave velocity data. Specifically, this involves first collecting shear wave velocity data from multiple existing wells, along with corresponding logging parameters (such as sonic transit time, density, and natural gamma), for data preprocessing. Then, machine learning is used to learn and process the nonlinear mapping relationships within the data, completing model training to predict the shear wave velocity of the well to be predicted.

[0004] However, the shear wave velocities obtained by existing prediction methods cannot guarantee both prediction accuracy and efficiency. Summary of the Invention

[0005] This application provides a method, apparatus, and device for predicting shear wave velocity, which solves the problem that the shear wave velocity obtained by the prediction method in the prior art is difficult to guarantee prediction accuracy and efficiency.

[0006] In a first aspect, this application provides a method for predicting shear wave velocity, including:

[0007] For each known well, based on the first dataset of the known well, multiple target logging parameters are determined from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The influence of the target logging parameters on the shear wave velocity of the known well is greater than the influence of other logging parameters on the shear wave velocity of the known well.

[0008] Based on the second dataset of the known wells, a decision tree model of the known wells is constructed. The second dataset includes the shear wave velocity of the known wells at each preset depth and the parameter values ​​of each target logging parameter.

[0009] The dataset to be predicted for the well to be predicted is input into the decision tree model of the known well to obtain the initial shear wave velocity output by the decision tree model. The dataset to be predicted includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth.

[0010] The target shear wave velocity of the well to be predicted is generated based on the initial shear wave velocity of the well known in each decision tree model.

[0011] Secondly, this application provides a shear wave velocity prediction device, comprising:

[0012] The determination module is used to determine multiple target logging parameters for each known well based on a first dataset of the known well from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The influence of the target logging parameters on the shear wave velocity of the known well is greater than the influence of other logging parameters on the shear wave velocity of the known well.

[0013] The construction module is used to construct a decision tree model of the known well based on a second dataset of the known well, wherein the second dataset includes the shear wave velocity of the known well at each preset depth and the parameter values ​​of each target logging parameter;

[0014] The input module is used to input the dataset to be predicted of the well to the decision tree model of the known well, and obtain the initial shear wave velocity output by the decision tree model. The dataset to be predicted includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth.

[0015] The generation module is used to generate the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by the known wells in each decision tree model.

[0016] Optionally, the determining module is specifically used for:

[0017] For each logging parameter, a third dataset corresponding to the logging parameter is extracted from the first dataset of the known well. The third dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of the logging parameter.

[0018] For each parameter value of the logging parameter, the third dataset corresponding to the logging parameter is split according to the parameter value to generate a first subset dataset and a second subset dataset.

[0019] Based on the first subset, the second subset, and the third dataset corresponding to the logging parameters, calculate the decrease in mean square error (MSE) of the logging parameters before and after splitting according to the parameter values;

[0020] Calculate the cumulative decrease in MSE of the logging parameters based on the decrease in MSE corresponding to each parameter value of the logging parameters;

[0021] Based on the cumulative decrease in MSE for each logging parameter, multiple target logging parameters are determined from all logging parameters.

[0022] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0023] The memory stores the instructions that the computer executes;

[0024] The processor executes computer execution instructions stored in memory to implement the transverse wave velocity prediction method as described in the first aspect and various possible implementations of the first aspect above.

[0025] The shear wave velocity prediction method, apparatus, and equipment provided in this application determine multiple target logging parameters from all logging parameters for each known well, based on a first dataset of the known well. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The target logging parameters have a greater influence on the shear wave velocity of the known well than other logging parameters. A decision tree model of the known well is constructed based on a second dataset of the known well. The dataset to be predicted from the well to be predicted is input into the decision tree model of the known well to obtain the initial shear wave velocity output by the decision tree model. Based on the initial shear wave velocity output by the known well in each decision tree model, the target shear wave velocity of the well to be predicted is generated. This method utilizes the characteristic that the logging parameters of each known well differ due to different geological environments. Starting from independently selecting the target logging parameters for each known well, a decision tree model is established for each known well, ultimately generating the decision tree model. When predicting the shear wave velocity of a well, this model can more accurately capture the characteristics of each known well, thereby reducing prediction errors caused by using shared logging parameters from multiple wells. Simultaneously, it improves the model's applicability and robustness in different environments, and enhances its prediction efficiency. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0027] Figure 1 A flowchart illustrating the shear wave velocity prediction method provided in this application embodiment. Figure 1 ;

[0028] Figure 2 A schematic diagram showing the coordinates of each known well in the park and its corresponding distance weight, provided for embodiments of this application;

[0029] Figure 3 A cross-plot of predicted and actual values ​​of four different machine learning methods for a well to be predicted, provided in an embodiment of this application.

[0030] Figure 4 A curve comparing the actual P-wave and S-wave velocity ratios with those predicted by different methods, provided in the embodiments of this application;

[0031] Figure 5(a) is a comparison of the shear wave velocity prediction curves of different prediction methods provided in the embodiments of this application under low-pass filtering;

[0032] Figure 5(b) is a histogram comparing the prediction errors of different prediction methods provided in the embodiments of this application under low-pass filtering;

[0033] Figure 6 A flowchart illustrating the shear wave velocity prediction method provided in this application embodiment. Figure 2 ;

[0034] Figure 7(a) shows the importance scores of the well logging parameters of well A1 in the known well.

[0035] Figure 7(b) shows the ranking of importance scores for well logging parameters of well A2.

[0036] Figure 7(c) shows the ranking of importance scores for well A3 logging parameters of known wells;

[0037] Figure 8 A linear correlation heatmap between each logging parameter of the well to be predicted is provided for embodiments of this application;

[0038] Figure 9 A schematic diagram of the interface for the shear wave velocity prediction method provided in the embodiments of this application;

[0039] Figure 10 This is a schematic diagram of the structure of the shear wave velocity prediction device provided in the embodiments of this application;

[0040] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0041] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0042] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0043] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0044] In oil and gas exploration and development, P-wave and S-wave velocities provide crucial subsurface information. S-wave velocity is beneficial for identifying pore fluids and reservoir properties, playing a vital role in pre-stack inversion and AVO attribute analysis. Combined P-wave and S-wave velocity inversion helps reduce reservoir ambiguity. S-wave velocity can be obtained through laboratory core analysis or dipole sonic logging. While these methods provide high-quality data, they are costly and have low measurement efficiency under complex geological conditions. Furthermore, for older wells that have already been drilled, the conditions for these specialized logging methods are often unavailable, leading to a lack of S-wave velocity data and limiting the ability to perform detailed reservoir characterization and analysis. Therefore, predictive methods for obtaining S-wave velocity data are widely used by researchers and engineers.

[0045] Currently, methods for predicting shear wave velocity typically rely on data-driven machine learning. Specifically, this involves first collecting shear wave velocity data from multiple wells, along with corresponding logging parameters (such as sonic transit time, density, and natural gamma), and then preprocessing the data (including denoising, missing value completion, and standardization) to ensure data quality and consistency. Machine learning is then used to learn and process the nonlinear mapping relationships within the data, completing model training, and applying it to logging data from wells in unmeasured areas to predict shear wave velocity.

[0046] However, most existing machine learning methods typically utilize multiple logging parameters and corresponding shear wave velocities from multiple known wells, selecting common target logging parameters for each well. This approach ignores the differences between individual wells, easily introducing errors during modeling. For example, the geological environment of different wells can lead to significant variations in logging parameters, and using common target logging parameters prevents the model from accurately capturing the characteristics of each well, thus affecting the accuracy of shear wave velocity prediction. Secondly, this method also struggles to guarantee improved prediction efficiency. Because it fails to fully utilize the unique data features of each well, the model may require more adjustments and optimizations in practical applications, increasing computational costs and time consumption.

[0047] Based on this, this application proposes a method, apparatus, and equipment for predicting shear wave velocity. Due to differences in the geological environment of known wells, the logging parameters of each known well may vary significantly. Target logging parameters can be selectively chosen for different known wells. Specifically, using a single known well as a unit, target logging parameters are selected based on its own logging parameters and corresponding shear wave velocities. Furthermore, a decision tree model is generated using the target logging parameters for each known well. When a prediction is needed for a well to be predicted, the decision tree model outputs the predicted shear wave velocity for each known well, thus obtaining the target shear wave velocity for the well to be predicted. This method, by independently selecting and analyzing target logging parameters based on the unique logging parameters of each known well and establishing a decision tree model, allows the decision tree model to more accurately capture the target logging parameters for each well, thereby improving prediction accuracy. Furthermore, selecting target logging parameters on a unit basis (one well at a time) makes the decision tree model more concise and targeted, reducing computational load and improving the operational efficiency of the decision tree model.

[0048] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0049] Figure 1 A flowchart illustrating the shear wave velocity prediction method provided in this application embodiment. Figure 1 .like Figure 1 As shown, the method includes:

[0050] S101. For each known well, based on the first dataset of the known well, determine multiple target logging parameters from all logging parameters;

[0051] The first dataset includes the shear wave velocity of a known well at each preset depth and the parameter value of each logging parameter. The target logging parameter has a greater influence on the shear wave velocity of the known well than other logging parameters.

[0052] Among them, the known wells are those in which the shear wave velocity at each preset depth and the parameter values ​​of each logging parameter have been obtained within a certain area.

[0053] For example, suppose the influence of the input G logging parameters on the shear wave velocity is sorted from high to low, and the top G logging parameters with the highest influence are selected as the target logging parameters for constructing the final decision tree model. This completes the selection of the target logging parameters. After screening, it was transformed into

[0054] Understandably, for each known well, a first dataset is constructed, containing the shear wave velocity and the parameter values ​​of each logging parameter at each preset depth. This dataset is then used to analyze the influence of all logging parameters on the shear wave velocity. Several logging parameters with high influence are selected as target logging parameters to construct a decision tree model. This process eliminates redundant and irrelevant logging parameters, reduces the impact of noise on the decision tree model, makes it more stable, and enhances its robustness and generalization ability. Furthermore, it enables the decision tree model to more accurately capture the variation patterns of shear wave velocity, reducing prediction errors and laying the foundation for improved prediction accuracy.

[0055] S102. Based on the second dataset of known wells, construct a decision tree model for the known wells;

[0056] The second dataset includes the shear wave velocity of known wells at each preset depth and the parameter values ​​of each target logging parameter.

[0057] Among them, the decision tree model is a weak classifier that can output the predicted initial shear wave velocity of the well to be predicted based on the prediction dataset of the well to be predicted. The prediction dataset includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth.

[0058] Understandably, by determining the parameter values ​​of each target logging parameter of a known well and the shear wave velocity at each preset depth, a second dataset can be established. Then, based on this dataset, a decision tree model can be built for each known well, thereby constructing a complete decision tree model of the known wells, which can be used for subsequent shear wave velocity prediction of the well to be predicted.

[0059] The decision tree model is an inverted tree structure, including a root node and leaf nodes. The root node contains the complete dataset, i.e., the second dataset. Leaf nodes are the splitting nodes of the decision tree, i.e., the nodes that split the dataset. In decision tree regression, the tree is constructed recursively using a greedy algorithm to divide the dataset into increasingly smaller subsets at the splitting points. A greedy algorithm is an algorithm that chooses the optimal option at each step based on the current state. Its goal is to eventually reach the global optimum through local optima.

[0060] In the decision tree splitting process for a known well, a logging parameter and its corresponding value are selected as the root node. The dataset is then split into two similar datasets to generate the next node. The next node can continue to split until a stopping condition is met (such as the maximum depth of the tree, the minimum number of samples in the node, etc.). The decision tree calculates the average target shear wave velocity at each leaf node as the final prediction result for that region.

[0061] S103. Input the dataset of the well to be predicted into the decision tree model of the known well to obtain the initial shear wave velocity output by the decision tree model.

[0062] The dataset to be predicted includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth.

[0063] The initial shear wave velocity refers to the shear wave velocity of the well to be predicted, which is output by each known well in the prediction model after the dataset of the well to be predicted is input into the decision tree model of the known wells.

[0064] Understandably, the dataset of wells to be predicted is input into a decision tree model composed of known wells for prediction. The decision tree models of all known wells are constructed into a random forest model, which can integrate the prediction results of multiple known wells, reducing the overfitting problem that may exist in the decision tree model of a single known well, thereby improving the accuracy of shear wave velocity prediction.

[0065] S104. Based on the initial shear wave velocity of the well output known in each decision tree model, generate the target shear wave velocity of the well to be predicted.

[0066] The target shear wave velocity refers to the final shear wave velocity of the well to be predicted, generated by processing the initial shear wave velocity output from each known well.

[0067] In one possible implementation, the initial shear wave velocity can be weighted to generate the final shear wave velocity of the well to be predicted. However, commonly used weighting methods fail to consider the spatial similarity between two wells and cannot properly assign appropriate weight values ​​to different initial shear wave velocities. This may lead to prediction errors between the well to be predicted and known wells far away due to differences in geological sedimentary patterns. Since the spatial similarity between two wells is related to the distance between them, the closer the wells are, the more similar their geological sedimentary patterns and logging parameter responses, and the higher the spatial similarity; conversely, the farther the distance, the lower the correlation of geological sedimentary patterns and the lower the spatial similarity. Therefore, further consideration of distance can be incorporated. When generating the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by known wells in each decision tree model, the distance between each known well and the well to be predicted can also be considered as a weight value.

[0068] Figure 2 This is a schematic diagram showing the coordinates of each known well within the park and its corresponding distance weight, provided for embodiments of this application. Figure 2 In the diagram, (a) indicates the location between different wells. There are three known wells A1, A2, and A3 in the park, and well A4 to be predicted. The distances of wells A1, A2, and A3 to well A4 are arranged from closest to furthest. Figure 2 (b) shows the distance weighting curves for each known well. Based on the principle that closer distances indicate higher spatial similarity, it can be seen that the closest well, A1, has the largest weight, followed by A2, and then A3. When predicting the shear wave velocity of well A4, wells A1, A2, and A3 are assigned distance weights of 0.55, 0.27, and 0.18, respectively. Well A1 has the greatest influence on the prediction results for well A4.

[0069] In one feasible approach, the location information of each known well and the location information of the well to be predicted can be obtained first. Then, based on the initial shear wave velocity output by the decision tree model of the known wells, the location information of the known wells, and the location information of the well to be predicted, the target shear wave velocity of the well to be predicted is generated.

[0070] Location information refers to the geographical coordinates corresponding to different wells. This location information can be obtained through geographic information system databases, global positioning system data, and map data, etc. In practical applications, the acquisition of location information can be determined according to the actual situation, and this application embodiment does not impose specific limitations on this.

[0071] Understandably, considering that the distance between each known well and the well to be predicted has a certain impact on the target shear wave velocity of the well to be predicted, an inverse distance-weighted ensemble is performed on the initial shear wave velocity output by each known well in the decision tree model. Therefore, the location information of each known well and the location information of the well to be predicted are first obtained, and then weighted according to the initial shear wave velocity output by the decision tree model. Different weight values ​​are assigned to the initial shear wave velocities output by different known wells based on their distance, thus obtaining the target shear wave velocity of the well to be predicted. This method effectively reduces the high error impact of distant known wells on the prediction of the target shear wave velocity of the well to be predicted, providing higher accuracy prediction results.

[0072] In one feasible approach, the specific calculation formula for generating the target shear wave velocity of the well to be predicted is as follows, based on the initial shear wave velocity output by the known wells in each decision tree model, the location information of the known wells, and the location information of the well to be predicted:

[0073]

[0074] Among them, y pred The target shear wave velocity of the well to be predicted. Let be the initial shear wave velocity output by the decision tree model for well i, n be the number of known wells, b be the weighted power exponent, x be the x-coordinate of the well to be predicted, and y be the y-coordinate of the well to be predicted. i Given the x-coordinate of well i, y i The ordinate of well i is known. The abscissa and ordinate of the well to be predicted are determined based on the location information of the well to be predicted. The abscissa and ordinate of well i are known based on the location information of well i.

[0075] The weighted power exponent is a form of weighted average, a type of average defined by different power values ​​b. By choosing different power values ​​b, it is possible to adapt to different types of data and application scenarios, thus providing a flexible tool for complex data analysis. This application does not impose specific limitations on the selection of the weighted power exponent b.

[0076] For example, suppose there are n known wells W1, W2…W n And the well W to be predicted p In the process of predicting W p Before weighting the predicted shear wave velocities of different initial values, it is necessary to calculate the predicted W. p With known wells W1, W2…W n The distance between them. That is, using the well W to be predicted. p The coordinates of a point (x, y), and n known wells W1, W2…W n The coordinates of the point (x) i y i), i = 1, 2, 3…n. For each known well W1, W2…W n With the well to be predicted W p There will be a reciprocal of the distance as a weight value, based on the weight values ​​of different known wells and the initial shear wave velocity output by different known wells in the decision tree model. The products are summed to obtain the final target shear wave velocity y of the well to be predicted. pred .

[0077] Understandably, by selecting target logging parameters for each known well and establishing a decision tree model for each well based on these parameters, the prediction error caused by low-impact logging parameters is effectively reduced. Furthermore, by utilizing the initial shear wave velocities output from different known wells in the decision tree model, an inverse distance-weighted ensemble strategy is employed. This strategy uses the distance information between different known wells and the well to be predicted as weights, distributing them to the initial shear wave velocity results output from the corresponding wells. Ultimately, the target shear wave velocity of the well to be predicted is determined, achieving high-precision shear wave velocity prediction by combining the similarity of location information between different wells. This ensures that known wells closer to the well to be predicted have a greater weight in the prediction results, thereby improving the model's prediction accuracy and reliability.

[0078] Figure 3 This is a cross-plot of the predicted and actual values ​​of four different machine learning methods for predicting the well to be predicted, provided in an embodiment of this application. The diamonds represent extreme gradient boosting (XGBoost), the squares represent random forest (RF), the triangles represent support vector regression (SVR), and the circles represent geologically constrained random forest (GCRF). It can be seen that the predicted values ​​of all four methods have a high correlation with the actual values, with a high coefficient of determination (Rsquared, R...). 2 The mean absolute error (MAE) values ​​were all above 0.93. However, the GCRF method used in this application performed best, with the smallest MAE value, the most concentrated data point distribution, and the smallest prediction error. Therefore, this model has the highest prediction accuracy and good stability.

[0079] Figure 4 A comparison graph showing the actual P-wave and S-wave velocity ratios provided in the embodiments of this application, and the P-wave and S-wave velocity ratios predicted by different methods. Among them, Figure 4 In the figure, A is a comparison curve of the actual P-wave and S-wave velocity ratio and the P-wave and S-wave velocity ratio predicted by the XGBoost method; Figure 4B in the figure represents the comparison curve between the actual P-wave and S-wave velocity ratio and the P-wave and S-wave velocity ratio predicted by the RF method. Figure 4 C in the figure represents the comparison curve between the actual P-wave and S-wave velocity ratio and the P-wave and S-wave velocity ratio predicted by the SVR method. Figure 4 In the figure, D represents a comparison curve between the actual P-wave and S-wave velocity ratio and the P-wave and S-wave velocity ratio predicted by the GCRF method. From... Figure 4 As can be seen from the data, at a depth of 1536m, the R0 of the actual P-wave and S-wave velocity ratio obtained by the XGBoost method is significantly different from the predicted P-wave and S-wave velocity ratio. 2 The smallest value and the largest MAE value indicate a poor fit between the actual and predicted P- and S-wave velocities, resulting in low prediction accuracy. In contrast, the R-values ​​of the actual and predicted P- and S-wave velocities obtained by traditional RF and SVR methods are significantly lower. 2 The moderate MAE values ​​indicate that both models outperform the extreme gradient boosting method in predicting the P-wave and S-wave velocity ratios. Furthermore, the R-values ​​of the actual P-wave and S-wave velocity ratios obtained using the method described in this application are significantly lower than those of the predicted P-wave and S-wave velocity ratios. 2 The maximum and minimum MAE values ​​indicate that the GCRF model, which is constructed by selecting the target logging parameters for each well and establishing a decision tree model, has a good fit between the predicted P-wave and S-wave velocity ratios and the actual P-wave and S-wave velocity ratios, with a small prediction error and higher prediction accuracy.

[0080] Furthermore, Figure 5(a) is a comparison of the shear wave velocity prediction curves of different prediction methods provided in the embodiments of this application under low-pass filtering. The black line represents the low-pass filtered result of the measured shear wave velocity, the gray triangle represents the low-pass filtered result of the shear wave prediction obtained by the Logistic Regression (LR) method, and the gray square line represents the low-pass filtered result of the shear wave prediction obtained by the GCRF model. Figure 5(b) is a comparison of the histogram of the shear wave velocity prediction errors of different prediction methods provided in the embodiments of this application under low-pass filtering. In the diagram, the white squares represent the histogram of the shear wave velocity prediction error under low-pass filtering obtained by the LR method, while the gray squares represent the histogram of the shear wave velocity prediction error under low-pass filtering obtained by the LR method. It can be seen that the shear wave velocity prediction curve obtained by the gray square-shaped line, representing the GCRF model, has a better fit with the actual shear wave velocity curve. Furthermore, the distribution of the shear wave velocity prediction error histogram obtained by the GCRF model under low-pass filtering is more stable. This indicates that the method in this application selects target logging parameters for each known well based on its unique logging parameters, establishes a decision tree model for each well, and finally constructs a complete random forest shear wave velocity prediction model. At the same time, considering the distance and spatial similarity between different wells, the distance is taken into account, and the prediction results of the decision tree model for each known well are weighted, thereby improving the model's prediction accuracy, precision, and stability.

[0081] This application provides a method for predicting shear wave velocity. For each known well, based on a first dataset of the known well, multiple target logging parameters are determined from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The target logging parameters have a greater influence on the shear wave velocity of the known well than other logging parameters. A decision tree model of the known well is constructed based on a second dataset of the known well. The dataset to be predicted of the well to be predicted is input into the decision tree model of the known well to obtain the initial shear wave velocity output by the decision tree model. The dataset to be predicted includes the parameter value of each target logging parameter of the well to be predicted at each preset depth. Based on the initial shear wave velocity output by the known well in each decision tree model, the target shear wave velocity of the well to be predicted is generated. This method leverages the unique characteristics of logging parameters for each known well due to varying geological environments. It starts by independently selecting target logging parameters for each known well and then uses these parameters to build a complete decision tree model. Based on the initial shear wave velocities output from different known wells in this prediction model, an inverse distance-weighted integration strategy is employed. This strategy uses the distance information between different known wells and the well to be predicted as weights, distributing them to the initial shear wave velocity results of the corresponding known wells. Ultimately, the target shear wave velocity of the well to be predicted is determined. This approach ensures more accurate and reliable shear wave velocity predictions even under complex geological conditions and significant differences in logging data between wells, significantly improving the efficiency and accuracy of oil and gas exploration.

[0082] In one feasible approach, the specific process of constructing a decision tree model for known wells based on a second dataset of known wells can be achieved through the following steps:

[0083] Step a: Determine the logging parameters to be processed from the target logging parameters corresponding to the known wells;

[0084] Among them, the logging parameters to be processed refer to the logging parameters that need to be split in the subsequent process.

[0085] It is understandable that once the target logging parameters corresponding to a known well are determined, the subsequent logging parameters to be split and processed are then determined from them.

[0086] Step b: Extract the fourth dataset corresponding to the logging parameters to be processed from the second dataset of known wells;

[0087] The fourth dataset includes the shear wave velocity of known wells at each preset depth and the parameter values ​​of the logging parameters to be processed.

[0088] Understandably, after determining the logging parameters to be processed based on the target logging parameters from the known well, the shear wave velocity and the parameter values ​​of the logging parameters to be processed at each preset depth are extracted from the second dataset.

[0089] For example, Table 1 is a second dataset for a known well, which includes the shear wave velocity at each preset depth and the parameter values ​​of the target logging parameters. If the density parameter is determined to be the logging parameter to be processed from the target logging parameters corresponding to the known well, then the fourth dataset corresponding to the logging parameter to be processed, extracted from the second dataset, is shown in Table 2, which includes the shear wave velocity at each preset depth of the known well and the parameter values ​​of the logging parameters to be processed.

[0090] Table 1. Second dataset for a known well.

[0091] depth density Natural Gamma transverse wave velocity 1600 2.1 30 3.5 1650 2.8 32 4.0 1700 2.5 28 4.5 1750 3.0 29 5.0

[0092] Table 2. Fourth dataset of a known well.

[0093] depth density transverse wave velocity 1600 2.1 3.5 1650 2.8 4.0 1700 2.5 4.5 1750 3.0 5.0

[0094] Step c: Based on the fourth dataset, determine the target parameter value from the parameter values ​​of the logging parameters to be processed. The sum of squared errors (SSE) of the third and fourth subsets generated after splitting the fourth dataset based on the target parameter value is less than the total SSE of the third and fourth subsets generated after splitting the fourth dataset based on other parameter values.

[0095] Here, the target parameter value refers to the value used to split the fourth dataset. The third and fourth subsets refer to two similar datasets formed after splitting the fourth dataset according to the target parameter value.

[0096] For example, assuming that a density parameter of 2.5 is selected as the target parameter value from the fourth dataset of known wells in Table 2, then the third subset generated after splitting the fourth dataset according to this value is [1600, 2.1, 3.5] and [1700, 2.5, 4.5], and the fourth subset is [1650, 2.8, 4.0].

[0097] The sum of squares (SSE) was calculated based on the third and fourth subsets [1750, 3.0, 5.0]. Furthermore, the SSE generated by splitting with a density parameter of 2.5 as the target parameter value was smaller than the total SSE of the third and fourth subsets generated by splitting with density parameters of 2.1, 2.8, and 3.0.

[0098] Understandably, by selecting the parameter value corresponding to the minimum total SSE as the target parameter value, it can be ensured that the generated shear wave velocity prediction error is minimized within the region split by the current target parameter value. This means that each split reduces the error in the sense of a local optimum. Furthermore, each split divides the data more precisely, allowing the model to better capture the characteristics of the parameters, thereby improving the model's prediction accuracy.

[0099] In one possible approach, the target parameter value is determined from the parameter values ​​of the logging parameters to be processed based on the fourth dataset, including:

[0100] First, for each parameter value of the logging parameters to be processed, the fourth dataset is split according to the parameter value to generate a third subset and a fourth subset. Then, the total SSE of the third subset and the fourth subset is calculated. Finally, the parameter value with the smallest total SSE is determined as the target parameter value.

[0101] Optionally, the specific formula for splitting the fourth dataset based on the target parameter value is as follows:

[0102] Q1(j,a)={X|X (j) ≤a},Q2(j,a)={X|X (j) >a}

[0103]

[0104] Where a is the target parameter value; j is the logging parameter to be processed; Q1 is the third subset generated after splitting the fourth dataset according to the target parameter value; Q2 is the fourth subset generated after splitting the fourth dataset according to the target parameter value; The transverse wave velocity is the output based on the third subset Q1 or the output based on the fourth subset Q2; N m X represents the number of shear wave velocities contained in the Q1 or Q2 dataset; X is the shear wave velocity.

[0105] The specific formula for determining the target parameter value 'a' from the values ​​of the logging parameters to be processed is as follows:

[0106]

[0107] Where min is the minimum value; C1 is the shear wave velocity output according to the third subset Q1; C2 is the shear wave velocity output according to the fourth subset Q2; SSE generated based on the third subset dataset; This is an SSE generated based on the fourth subset of the dataset.

[0108] Continuing with the previous example, suppose we select density parameters of 2.1, 2.5, 2.8, and 3.0 as parameter values ​​from the fourth dataset of known wells in Table 2. Then, based on these four different parameter values, we split the fourth dataset into a third and fourth subset. Taking 2.1 as an example, the split third subset is [1600, 2.1, 3.5], and the fourth subset is [1700, 2.5, 4.5], [1650, 2.8, 4.0], and [1750, 3.0, 5.0]. Next, we calculate the total SSE of the third and fourth subsets. Similarly, assuming that in this example, the total SSE of the third and fourth subsets obtained after splitting with 2.5 as the parameter value is greater than the total SSE of the third and fourth subsets obtained after splitting with 2.1, 2.8, and 3.0 as parameter values, then 2.5 is selected as the target parameter value for the density logging parameter.

[0109] Understandably, by selecting the parameter value corresponding to the minimum total SSE of the third and fourth subsets generated after splitting as the target parameter value, the accuracy of the predicted shear wave velocity generated within the target parameter value region is guaranteed.

[0110] Step d: Based on the target parameter values, split the second dataset of known wells into two new second datasets;

[0111] The new second dataset consists of two similar datasets formed by splitting the second dataset of known wells according to the target parameter values.

[0112] Continuing with the previous example, if we split the second dataset of known wells in Table 2 based on a density parameter of 2.5 as the target parameter value, we will obtain two new second datasets as shown in Tables 3 and 4.

[0113] Table 3. New Second Dataset for a Known Well (1)

[0114] depth density Natural Gamma transverse wave velocity 1600 2.1 30 3.5 1700 2.5 28 4.5

[0115] Table 4. New Second Dataset 2 for a Known Well

[0116] depth density Natural Gamma transverse wave velocity 1650 2.8 32 4.0 1750 3.0 29 5.0

[0117] Step e: Determine the logging parameters to be processed as processed logging parameters;

[0118] Among them, the processed logging parameters refer to the logging parameters that have been split and calculated to select the corresponding target parameter values.

[0119] Understandably, identifying the logging parameters to be processed as already processed logging parameters allows for further processing of other target logging parameters. This approach enables the decision tree model to process logging parameters more quickly, avoids redundant processing of logging parameters, and improves the efficiency of model operation.

[0120] Repeat steps a through e until all target logging parameters have been processed.

[0121] In this process, by repeatedly executing steps a to e, the target parameter values ​​of all target logging parameters are determined using recursive logic. This means that each leaf node of the decision tree model constructed for each known well is determined. Finally, the second dataset is divided into multiple new second datasets, which allows for the further construction of decision tree models corresponding to the known wells.

[0122] Optionally, assume the given training dataset is D = {(X1,Y1),(X2,Y2),…,(X... N ,Y N )}, that is, the first dataset. For each known well at each preset depth, G represents the parameter values ​​of all logging parameters, where G is the number of logging parameters. i = 1, 2, ..., N, where N is the sample size of the known wells, and Y... i This corresponds to the transverse wave velocity value. When performing a regression task, the first dataset space before the decision tree splits is defined as Q. As the decision tree grows, the space Q is continuously divided into M split dataset spaces Q1, Q2, ..., Q' using classification and regression tree algorithms. m The predicted shear wave velocity corresponding to the split dataset space is defined as C. m The formula corresponding to the final decision tree model is:

[0123]

[0124] Where f(X) is the decision tree model formed by the known wells; M is the number of datasets after the known wells are split according to the target parameter values; C m This represents the shear wave velocity output from the dataset after splitting based on the target parameter values.

[0125] Furthermore, in step c above, the purpose of calculating the target parameter value for each target logging parameter is to minimize the sum of squares error of the predicted shear wave velocity values ​​corresponding to each dataset space obtained by splitting the dataset based on the target logging parameters, thereby ensuring the prediction accuracy of the prediction model. The specific formula for the sum of squares error is:

[0126] L=∑(Y i -f(X i )) 2 i = 1, 2, ..., N

[0127] Where L is the sum of squares error of the predicted shear wave velocity values ​​corresponding to the dataset space obtained by splitting the target logging parameters. Q is taken when the sum of squares error L is minimized. m All transverse wave velocities Y in the corresponding split dataset space i The average value is used as the predicted transverse wave velocity value of the spatial output of the split dataset.

[0128] It should be understood that a complete random forest model for each known well is generated by constructing a decision tree model corresponding to the known wells. In existing technologies, the number of decision tree models built in a random forest model is usually set according to the network parameters. However, in this embodiment, the number of decision tree models is consistent with the number of known wells. This ensures that each known well corresponds to a decision tree model. This avoids the uncertainty and error introduced by random sampling in random forests, ensuring the stability and accuracy of the prediction model.

[0129] Understandably, since the second dataset includes different target logging parameters for known wells, and each target logging parameter contains different parameter values, the recursive logic is used to determine the target parameter value for each target logging parameter one by one, which effectively improves the model running efficiency. At the same time, the selection process of target parameter values ​​also ensures the prediction accuracy of the decision tree model for each known well, thereby further ensuring the prediction accuracy of the random forest model composed of decision trees from multiple known wells.

[0130] Figure 6 A flowchart illustrating the shear wave velocity prediction method provided in this application embodiment. Figure 2 This embodiment, based on the above embodiments, provides a detailed explanation of the process for determining target logging parameters. For example... Figure 6 As shown, the method includes:

[0131] S601. For each logging parameter, extract the third dataset corresponding to the logging parameter from the first dataset of known wells;

[0132] The third dataset includes the shear wave velocity and logging parameter values ​​of known wells at each preset depth.

[0133] For example, Table 5 shows the third dataset corresponding to the natural gamma logging parameters extracted from the first dataset of known wells.

[0134] Table 5 shows the third dataset corresponding to the natural gamma logging parameters of a known well.

[0135] depth Natural Gamma transverse wave velocity 1600 30 3.5 1650 32 4.0 1700 28 4.5 1750 29 5.0

[0136] Understandably, the third dataset corresponding to the logging parameters is extracted from the first dataset of known wells to provide a data basis for the subsequent calculation of the cumulative decrease in the MSE value of the logging parameters.

[0137] S602. For each parameter value of the logging parameter, split the third dataset corresponding to the logging parameter according to the parameter value to generate a first subset dataset and a second subset dataset.

[0138] The first and second subsets refer to two similar datasets formed by splitting the third dataset according to the target value of each logging parameter.

[0139] It is understandable that the specific process of splitting the third dataset corresponding to the logging parameters according to the parameter values ​​in S202 to generate the first and second sub-datasets is similar to the splitting process in step c above, and this application will not elaborate on it.

[0140] S603. Based on the first subset, the second subset, and the third dataset corresponding to the logging parameters, calculate the decrease in mean squared error (MSE) of the logging parameters before and after splitting according to the parameter values.

[0141] MSE is a commonly used metric for evaluating model prediction performance. It is primarily assessed by calculating the average of the squared errors. The MSE reduction value refers to the difference between the MSE of the logging parameters before splitting based on the parameter values ​​and the MSE of the logging parameters after splitting based on the parameter values.

[0142] It is understandable that by using the first subset of data to calculate the MSE of the first subset, using the second subset of data to calculate the MSE of the second subset of data, and using the third dataset corresponding to the logging parameters to calculate the MSE before splitting according to the parameter values, and then combining these three MSE values, the decrease in MSE of the logging parameters before and after splitting according to the parameter values ​​can be calculated.

[0143] In one feasible approach, the specific method for calculating the decrease in the mean square error (MSE) of the logging parameters before and after splitting based on the parameter values ​​can be achieved as follows:

[0144] First, based on the third dataset corresponding to the logging parameters, determine the first MSE of the logging parameters before splitting according to the parameter values. Then, based on the first and second subsets, determine the second MSE of the logging parameters after splitting according to the parameter values. Finally, the difference between the first and second MSEs is determined as the decrease in MSE before and after splitting according to the parameter values.

[0145] Optionally, the specific formula for calculating MSE is as follows:

[0146]

[0147] Where k is the number of known well shear wave velocities in different datasets; r i represents the shear wave velocity of the known wells in the dataset; r^ represents the average shear wave velocity of all known wells in the dataset.

[0148] Optionally, the specific formula for calculating the decrease in MSE is as follows:

[0149]

[0150] Where ΔI(j,s) is the decrease in MSE generated by splitting the logging parameter j; I(s) is the first MSE; k s k represents the number of shear wave velocities in the third dataset. sleft k represents the number of shear wave velocities in the first subset of data. sright I(sleft) represents the number of shear wave velocities in the second subset; I(sright) represents the MSE calculated based on the second subset; I(sright) represents the MSE calculated based on the second subset.

[0151] For example, suppose we select a natural gamma of 28 from Table 5 as the parameter value for splitting, resulting in the first subset [1700, 28, 4.5] and [1750, 29, 5.0], and the second subset [1650, 32, 4.0] and [1600, 30, 3.5]. Then, r^ in the first MSE is the average of all shear wave data in the third dataset, 4.25, where k is 4, and r... i The values ​​are 3.5, 4.0, 4.5, and 5.0 respectively. The first MSE can be obtained by calculating using the MSE formula. Similarly, the MSE generated from the first subset, the MSE generated from the second subset, and the corresponding second MSE can be calculated. Finally, by calculating the difference between the first and second MSE, the decrease in MSE before and after the parameter value 28 split can be determined when natural gamma is used as the logging parameter.

[0152] Understandably, by calculating and combining the mean square error (MSE) of the logging parameter before and after the parameter value split, the decrease in the MSE of the logging parameter before and after the parameter value split can be determined. This can provide a basis for obtaining the complete MSE decrease of the logging parameter and further quantify the influence of the logging parameter on the shear wave velocity.

[0153] S604. Calculate the cumulative decrease in MSE of the logging parameters based on the decrease in MSE corresponding to each parameter value.

[0154] The cumulative decrease in MSE refers to the value obtained by summing the decreases in MSE corresponding to each logging parameter value.

[0155] Optionally, the specific formula for calculating the cumulative decrease in MSE of well logging parameters is as follows:

[0156]

[0157] Among them, I j This represents the cumulative decrease in MSE for logging parameter j.

[0158] It is understandable that the cumulative decrease in MSE corresponding to each logging parameter can be calculated by summing up the decreases in MSE for each parameter value.

[0159] S605. Based on the cumulative decrease in MSE for each logging parameter, determine multiple target logging parameters from all logging parameters.

[0160] Understandably, the larger the cumulative decrease in MSE of a logging parameter, the greater its contribution to reducing the error in the model's predicted shear wave velocity, thus making it a suitable target logging parameter. Therefore, it is necessary to sort the calculated cumulative decrease in MSE for each logging parameter to further determine multiple target logging parameters.

[0161] In one feasible approach, the specific method for selecting multiple target logging parameters is as follows:

[0162] Sort all logging parameters in descending order of cumulative MSE decrease, and determine the first preset number of logging parameters as target logging parameters.

[0163] The preset quantity is a specific number that is pre-set or specified based on the actual situation before the model is built. This preset quantity is a positive integer greater than or equal to 1.

[0164] Optionally, to more easily select logging parameters that have a high impact on shear wave velocity, the cumulative decrease in MSE corresponding to all logging parameters can be normalized to obtain the importance score of each logging parameter, and the sum of the importance scores of all logging parameters is 1.

[0165] Optionally, the formula for normalizing the cumulative decrease in MSE for different logging parameters is as follows:

[0166]

[0167] Among them, NormalizedI j Importance score for each logging parameter, ∑ G IG This represents the sum of the cumulative decreases in MSE for G logging parameters. The target logging parameters are selected by prioritizing a predetermined number of g (g = 1, 2, 3, ..., g < G) logging parameters based on their importance scores. The parameter values ​​corresponding to these parameters are used as input to the decision tree model for the known well, thus completing the selection of the target logging parameters.

[0168] Figure 7(a) shows the importance score ranking of logging parameters in well A1, Figure 7(b) shows the importance score ranking of logging parameters in well A2, and Figure 7(c) shows the importance score ranking of logging parameters in well A3. It can be seen that in the importance ranking of logging parameters in well A1, the order is: P-wave velocity (Vp) > Spontaneous potential (SP) > Density (DEN) > Compensated neutron log (CNL) > Shale content (SH). In well A2, the importance ranking is Vp > CNL > SP > SH > DEN. In well A3, the importance ranking is Vp > CNL > SH > SP > DEN. These rankings represent the degree of influence of each logging parameter on the model prediction results, decreasing from high to low. Therefore, for well A1, P-wave velocity and spontaneous potential should be prioritized when constructing the model, while for wells A2 and A3, more attention should be paid to P-wave velocity and neutron logging parameters. By ranking the importance scores of logging parameters for wells A1, A2, and A3, the influence of each parameter on the model's prediction results is clearly demonstrated. This provides a strong basis for parameter selection for shear wave velocity prediction and helps improve the performance of the prediction model.

[0169] Correspondingly, Figure 8 A linear correlation heatmap between each logging parameter of the well to be predicted, provided in an embodiment of this application. The linear correlation heatmap is a visualization tool used to display the linear correlation between multiple variables. By using colors to represent the magnitude and direction of the correlation coefficient, the interrelationship between variables can be observed intuitively. The correlation coefficient ranges from -1 to 1. Figure 8 As shown in the figure, A4 is the well to be predicted. The linear correlation between each logging parameter in well A4 and shear wave velocity is ranked as follows: Vp > CNL > DEN > SP > SH. This indicates that Vp has the strongest linear correlation with shear wave velocity, CNL has a strong linear correlation, DEN has a moderate linear correlation, and SP and SH have weak linear correlations. Therefore, for well A4, the decision tree model of well A1 will output a more accurate shear wave velocity prediction result.

[0170] Furthermore, assume there are q known wells and one well W to be tested within the work area. p Each training well W1, W2, ..., W q As a separate dataset, the total dataset is now H = {D1, D2, ..., D...} q}. Given wells W1, W2, ..., W... q The dataset is modeled separately. The separate modeling process involves building a separate decision tree model f(x) for each known well, generating a decision tree model for each known well. When the well W to be tested is input into the decision tree model... p After obtaining the corresponding target logging parameters, the decision tree model for each known well will obtain the well W to be tested. p Corresponding prediction results

[0171] Understandably, sorting the cumulative decrease in MSE for each logging parameter from highest to lowest and selecting a certain number of logging parameters can reduce the error caused by low-impact logging parameters during the modeling process. Given the potential differences in logging data distribution between known wells and the well to be predicted, this method can better adapt to these differences, thus predicting the shear wave velocity of the well to be predicted more accurately. Especially under complex geological conditions and when there are significant differences in logging data between wells, this method can significantly improve the model's prediction accuracy.

[0172] This application provides a method, apparatus, and device for shear wave velocity. First, based on a first dataset, each logging parameter value is split to obtain different similar subsets. Then, based on the dataset before splitting and the different similar subsets after splitting, the cumulative MSE reduction value corresponding to each logging parameter value is calculated. Finally, the logging parameters are sorted according to the cumulative MSE reduction values ​​for each logging parameter value, and a certain number of logging parameters are selected as target logging parameters. This can provide logging parameters with more significant impact on the decision tree model established for each known well. Furthermore, the strategy of independently establishing a decision tree model for each known well effectively captures subtle differences between each known well, avoiding modeling errors caused by differences in logging parameter distribution. This improves the model's prediction accuracy and better reflects actual geological conditions.

[0173] Figure 9 This is a schematic diagram of the interface for the shear wave velocity prediction method provided in the embodiments of this application; as shown below. Figure 9 As shown, in Figure 9In the spatial constraint process, black dots represent the locations of multiple known wells, and gray dots represent the locations of wells to be predicted. First, target logging parameters are selected for each known well, and the input logging parameters are ranked by the importance score of the decision tree attribute. The top two parameters with the highest scores are selected as inputs to the decision tree model for each known well, establishing a single known well model. During the prediction process, the logging parameters of the well to be predicted are input into the decision tree model of each known well, obtaining their respective prediction results. Subsequently, the distance between each known well and the well to be predicted is calculated, and an inverse distance weighting strategy is applied, using the reciprocal of the distance as a weight to distribute the prediction results of each known well's decision tree model. This method effectively utilizes the spatial correlation between wells, reduces the influence of distant known wells on the prediction results, and improves prediction accuracy. Finally, the weighted prediction results are summed to obtain the final shear wave velocity prediction result for the well to be predicted. Compared with traditional machine learning shear wave velocity prediction algorithms, this application has higher accuracy and reliability. By selecting target logging parameters and a distance weighting method for each known well, it can more accurately reflect the subsurface geological conditions and establish a more accurate shear wave velocity prediction model.

[0174] Figure 10 This is a schematic diagram of the structure of the shear wave velocity prediction device provided in the embodiments of this application; as shown. Figure 10 As shown, the shear wave velocity prediction device 100 includes:

[0175] The determination module 1001 is used to determine multiple target logging parameters for each known well based on the first dataset of the known well from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The influence of the target logging parameters on the shear wave velocity of the known well is greater than the influence of other logging parameters on the shear wave velocity of the known well.

[0176] Module 1002 is used to construct a decision tree model of the known well based on a second dataset of known wells. The second dataset includes the shear wave velocity of the known well at each preset depth and the parameter values ​​of each target logging parameter.

[0177] The input module 1003 is used to input the dataset to be predicted of the well to the decision tree model of the known well, and obtain the initial shear wave velocity output by the decision tree model. The dataset to be predicted includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth.

[0178] The generation module 1004 is used to generate the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by the known wells in each decision tree model.

[0179] In one possible implementation, module 1001 is specifically used for:

[0180] For each logging parameter, a third dataset corresponding to the logging parameter is extracted from the first dataset of known wells. The third dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of the logging parameter.

[0181] For each parameter value of the logging parameter, the third dataset corresponding to the logging parameter is split according to the parameter value to generate a first subset dataset and a second subset dataset.

[0182] Based on the first subset, the second subset, and the third dataset corresponding to the logging parameters, calculate the decrease in MSE of the logging parameters before and after splitting according to the parameter values;

[0183] Calculate the cumulative decrease in MSE of the logging parameters based on the decrease in MSE corresponding to each parameter value.

[0184] Based on the cumulative decrease in MSE for each logging parameter, multiple target logging parameters are determined from all logging parameters.

[0185] In one possible implementation, the determining module 1001 is specifically used to sort all logging parameters in descending order of cumulative MSE decrease value, and determine the first preset number of logging parameters as target logging parameters.

[0186] In one possible implementation, module 1001 is specifically used for:

[0187] Based on the third dataset corresponding to the logging parameters, determine the first MSE of the logging parameters before splitting according to the parameter values;

[0188] Based on the first and second subsets, determine the second MSE of the logging parameters after splitting according to the parameter values;

[0189] The difference between the first MSE and the second MSE is determined as the decrease in MSE before and after the logging parameters are split according to the parameter values.

[0190] In one possible implementation, the shear wave velocity prediction device further includes a processing module, specifically used for:

[0191] Step a: Determine the logging parameters to be processed from the target logging parameters corresponding to the known wells;

[0192] Step b: Extract the fourth dataset corresponding to the logging parameters to be processed from the second dataset of the known well. The fourth dataset includes the shear wave velocity of the known well at each preset depth and the parameter values ​​of the logging parameters to be processed.

[0193] Step c: Based on the fourth dataset, determine the target parameter value from the parameter values ​​of the logging parameters to be processed. The total SSE of the third and fourth subsets generated after splitting the fourth dataset based on the target parameter value is less than the total SSE of the third and fourth subsets generated after splitting the fourth dataset based on other parameter values.

[0194] Step d: Based on the target parameter values, split the second dataset of known wells into two new second datasets;

[0195] Step e: Determine the logging parameters to be processed as processed logging parameters;

[0196] Repeat steps a through e until all target logging parameters have been processed.

[0197] In one possible implementation, module 1001 is specifically used for:

[0198] For each parameter value of the logging parameters to be processed, the fourth dataset is split according to the parameter value, resulting in a third subset dataset and a fourth subset dataset.

[0199] Calculate the total SSE for the third and fourth subsets based on the third and fourth subsets.

[0200] The parameter value that minimizes the total SSE is determined as the target parameter value.

[0201] In one possible implementation, the generation module 1004 is specifically used for:

[0202] Obtain the location information of each known well and the location information of the well to be predicted;

[0203] Based on the initial shear wave velocity output by the known wells in each decision tree model, the location information of the known wells, and the location information of the well to be predicted, the target shear wave velocity of the well to be predicted is generated.

[0204] In one possible implementation, the generation module is specifically used for:

[0205] Through formula Generate the target shear wave velocity of the well to be predicted;

[0206] Among them, y pred The target shear wave velocity of the well to be predicted. Let be the initial shear wave velocity output by the decision tree model for well i, n be the number of known wells, b be the weighted power exponent, x be the x-coordinate of the well to be predicted, and y be the y-coordinate of the well to be predicted. i Given the x-coordinate of well i, y iThe ordinate of well i is known. The abscissa and ordinate of the well to be predicted are determined based on the location information of the well to be predicted. The abscissa and ordinate of well i are known based on the location information of well i.

[0207] The shear wave velocity prediction device provided in this application embodiment can be used to execute the shear wave velocity prediction method in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0208] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. Additionally, these modules can be fully or partially integrated together, or implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through the integrated logic circuits in the hardware of the processor element or through software instructions.

[0209] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 11 As shown, the electronic device 110 provided in this embodiment includes at least one processor 1101 and a memory 1102. The device 110 also includes a communication component 1103. The processor 1101, the memory 1102, and the communication component 1103 are connected via a bus 1104.

[0210] In the specific implementation process, at least one processor 1101 executes computer execution instructions stored in memory 1102, causing at least one processor 1101 to execute the above-mentioned transverse wave velocity prediction method.

[0211] The specific implementation process of processor 1101 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0212] The electronic device provided in this application embodiment can be used to execute the shear wave velocity prediction method provided in any of the above method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0213] In the above Figure 11In the illustrated embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0214] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0215] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0216] Finally, it should be noted that other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and alterations may be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for predicting shear wave velocity, characterized in that, include: For each known well, based on the first dataset of the known well, multiple target logging parameters are determined from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The influence of the target logging parameters on the shear wave velocity of the known well is greater than the influence of other logging parameters on the shear wave velocity of the known well. Specifically, for each known well, based on the first dataset of the known well, multiple target logging parameters are determined from all logging parameters, including: For each logging parameter, a third dataset corresponding to the logging parameter is extracted from the first dataset of the known well. The third dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of the logging parameter. For each parameter value of the logging parameter, the third dataset corresponding to the logging parameter is split according to the parameter value to generate a first subset dataset and a second subset dataset. Based on the first subset, the second subset, and the third dataset corresponding to the logging parameters, calculate the decrease in mean square error (MSE) of the logging parameters before and after splitting according to the parameter values; Calculate the cumulative decrease in MSE of the logging parameters based on the decrease in MSE corresponding to each parameter value of the logging parameters; Based on the cumulative decrease in MSE for each logging parameter, multiple target logging parameters are determined from all logging parameters. Based on the second dataset of the known wells, a decision tree model of the known wells is constructed. The second dataset includes the shear wave velocity of the known wells at each preset depth and the parameter values ​​of each target logging parameter. The dataset to be predicted for the well to be predicted is input into the decision tree model of the known well to obtain the initial shear wave velocity output by the decision tree model. The dataset to be predicted includes the parameter values ​​of each target logging parameter of the well to be predicted at each preset depth. Based on the initial shear wave velocity of the wells known in each decision tree model, the target shear wave velocity of the well to be predicted is generated. The step of generating the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by the known well in each decision tree model includes: Obtain the location information of each known well and the location information of the well to be predicted; Based on the initial shear wave velocity output by the known wells in each decision tree model, the location information of the known wells, and the location information of the well to be predicted, the target shear wave velocity of the well to be predicted is generated.

2. The method according to claim 1, characterized in that, The step of determining multiple target logging parameters from all logging parameters based on the cumulative decrease in MSE for each logging parameter includes: All logging parameters are sorted in descending order of cumulative MSE decrease value, and the first preset number of logging parameters are determined as the target logging parameters.

3. The method according to claim 1, characterized in that, The step of calculating the decrease in MSE of the well logging parameters before and after splitting according to the parameter values, based on the first subset, the second subset, and the third dataset corresponding to the well logging parameters, includes: Based on the third dataset corresponding to the logging parameters, determine the first MSE of the logging parameters before splitting according to the parameter values; Based on the first subset and the second subset, determine the second MSE of the logging parameters after splitting according to the parameter values; The difference between the first MSE and the second MSE is determined as the decrease in MSE of the logging parameter before and after splitting according to the parameter value.

4. The method according to any one of claims 1 to 3, characterized in that, The step of constructing a decision tree model for the known wells based on the second dataset of the known wells includes: Step a: Determine the logging parameters to be processed from the target logging parameters corresponding to the known wells; Step b: Extract the fourth dataset corresponding to the logging parameters to be processed from the second dataset of the known well. The fourth dataset includes the shear wave velocity of the known well at each preset depth and the parameter values ​​of the logging parameters to be processed. Step c: Based on the fourth dataset, determine the target parameter value from the parameter values ​​of the logging parameters to be processed. The sum of squares error (SSE) of the third and fourth subsets generated after splitting the fourth dataset based on the target parameter value is less than the total SSE of the third and fourth subsets generated after splitting the fourth dataset based on other parameter values. Step d: Based on the target parameter values, split the second dataset of the known wells into two new second datasets; Step e: Determine the logging parameters to be processed as processed logging parameters; Repeat steps a through e until all target logging parameters have been processed.

5. The method according to claim 4, characterized in that, The step of determining the target parameter value from the parameter values ​​of the logging parameters to be processed based on the fourth dataset includes: For each parameter value of the logging parameters to be processed, the fourth dataset is split according to the parameter value to generate a third subset dataset and a fourth subset dataset. Calculate the total SSE of the third and fourth subsets based on the third and fourth subsets. The parameter value that minimizes the total SSE is determined as the target parameter value.

6. The method according to claim 1, characterized in that, The step of generating the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by the known wells in each decision tree model, the location information of the known wells, and the location information of the well to be predicted includes: Through formula Generate the target shear wave velocity of the well to be predicted; in, The target shear wave velocity of the well to be predicted is... Let be the initial shear wave velocity output by the decision tree model for well i, n be the number of known wells, b be the weighted exponent, x be the x-coordinate of the well to be predicted, and y be the y-coordinate of the well to be predicted. Given the x-coordinate of well i, The ordinate of well i is known, and the abscissa and ordinate of the well to be predicted are determined based on the location information of the well to be predicted. The abscissa and ordinate of well i are known based on the location information of well i.

7. An apparatus for implementing the shear wave velocity prediction method of claim 1, characterized in that, include: The determination module is used to determine multiple target logging parameters for each known well based on a first dataset of the known well from all logging parameters. The first dataset includes the shear wave velocity of the known well at each preset depth and the parameter value of each logging parameter. The influence of the target logging parameters on the shear wave velocity of the known well is greater than the influence of other logging parameters on the shear wave velocity of the known well. The construction module is used to construct a decision tree model of the known well based on a second dataset of the known well, wherein the second dataset includes the shear wave velocity of the known well at each preset depth and the parameter values ​​of each target logging parameter; The input module is used to input the dataset of the well to be predicted into the decision tree model of the known well, and obtain the initial shear wave velocity output by the decision tree model. The generation module is used to generate the target shear wave velocity of the well to be predicted based on the initial shear wave velocity output by the known wells in each decision tree model.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the transverse wave velocity prediction method as described in any one of claims 1 to 6.

Images