Longitudinal and transverse wave prediction method based on logging rock physics and mineral component data

Abstract

The invention relates to a longitudinal and transverse wave prediction method based on logging rock physics and mineral component data. The method comprises the following steps: S1, carrying out logging data standardization processing by combining regional geology and oil and gas testing data; s2, performing correlation analysis on logging parameters, mineral components and longitudinal and transverse wave velocities in combination with a standardized processing result in the step S1, and preferably selecting sensitive parameters; s3, a K-Means multi-parameter clustering analysis method is adopted, a deep learning method based on an attention mechanism and a bidirectional long and short term memory network is combined with the sensitive parameters obtained in the step S2, and a longitudinal and transverse wave speed prediction model is established. According to the method, rock mineral component quality control parameters are introduced, intersection fitting purely relying on logging parameters is avoided, and prediction and correction can be carried out on a research area where longitudinal and transverse wave measurement is not carried out or longitudinal and transverse wave data distortion is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas seismic exploration and development, and more specifically to a method for predicting longitudinal and transverse waves based on well logging rock physics and mineral composition data. Background Art

[0002] In shale gas exploration and development, geophysical technology plays an irreplaceable and important role. The relationship between P-wave and S-wave velocities is an important basis for determining reservoir lithology, physical properties and fluid identification. It is also a bridge for conducting joint well-seismic research and a link that expands the information of rock physics points to multiple dimensions of lines, surfaces and volumes. However, in actual exploration and production work, there is a lack of P-wave and S-wave velocity information due to various reasons. For this reason, experts and scholars at home and abroad have done a lot of analysis, testing and research in this regard.

[0003] Xu and White (1995) established the Xu-White model based on the Kuster-Toksoz theory and the DEM equivalent model, describing the pore geometry of sandstone and mudstone by comparing the P-wave velocity and S-wave velocity. Zhang Guangzhi et al. (2012) predicted the S-wave velocity in carbonate rocks based on the improved Xu-White model. Zhang Lianqun et al. (2016) obtained a direct understanding of the regularity of the P-wave and S-wave velocities of different lithologies such as shale, sandstone, and dolomite through experimental core analysis and testing. Zhang Zuhao et al. (2017) established a rock physics model by combining the self-consistent approximate model with the Biot theory, and proposed a velocity prediction method based on the SC-Gassmann model. Zhang Kefei et al. (2019) combined multiple equivalent models and the BK equation to estimate the S-wave velocity by constructing a rock physics model suitable for shale. Xing Wenjun et al. (2021) predicted the S-wave velocity through experimental rock physics. Liu Lanfeng et al. (2022) expanded the assumptions of Cao et al. for shear wave prediction based on the effective elastic modulus calculation method proposed by Sabrima, and further predicted shear waves by calculating pore weakness. Qiao Hanqing (2023) proposed a shear wave prediction method for organic-rich shale based on the Xu-White model with variable pore aspect ratio, and calculated the shear wave velocity at a well in the Jiaoshiba area of ​​the Sichuan Basin.

[0004] Shale, as an important unconventional oil and gas reservoir, is a multi-mineral highly heterogeneous natural composite composed of different clay minerals, organic matter, and minerals such as quartz and calcite with different contents. The properties of shale gas reservoirs are more complex than those of conventional reservoirs. Currently, most rock physics models used to estimate shear wave velocity are only applicable to isotropic rocks, while traditional methods directly use linear fitting between longitudinal and shear wave velocities to predict shear wave velocity, which has low accuracy and is affected by statistical data blocks, limiting its promotion in other blocks. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a P-wave and S-wave prediction method based on logging rock physics and mineral composition data, introduce rock mineral composition quality control parameters, get rid of the simple reliance on logging parameter intersection fitting, and carry out prediction and correction for research areas where P-wave and S-wave measurements have not been carried out or where P-wave and S-wave data are distorted.

[0006] The technical solution adopted by the present invention to solve the technical problem is: construct a P-wave and S-wave prediction method based on well logging rock physics and mineral composition data, including the following steps:

[0007] S1. Standardize the logging data by combining regional geological and oil and gas testing data;

[0008] S2, combining the standardized processing results of step S1 to carry out correlation analysis between logging parameters, mineral components and P- and S-wave velocities, and optimize sensitive parameters;

[0009] S3. Using the K-Means multi-parameter clustering analysis method, a deep learning method based on the attention mechanism and the bidirectional long short-term memory network is used to establish a P-wave and S-wave velocity prediction model in combination with the sensitive parameters obtained in step S2.

[0010] According to the above scheme, in step S1, the standardization of logging data is carried out in combination with regional geological and oil and gas testing data, which specifically includes the following steps:

[0011] S101. Carry out borehole environment correction based on regional geological and lithological logging data to reduce the impact of non-geological factors on logging data.

[0012] S102. Combined with the oil and gas test data, determine the standard wells and standard layers in the study area, carry out standardized processing of logging data, and provide reference standards for subsequent result comparison and accuracy analysis

[0013] According to the above scheme, in step S2, the correlation analysis between logging parameters, mineral components and P- and S-wave velocities is carried out, and the sensitivity parameters are optimized. The Pearson analysis method is used to carry out the correlation analysis for the main target layers in the study area. The specific contents are as follows:

[0014] The Pearson correlation analysis method is used to conduct a correlation analysis on the P-wave velocity of the main target layer with the logging parameters and mineral components, and the correlation coefficient R between the logging parameters, mineral components and the P-wave velocity is obtained. The parameters with correlation coefficient R close to 1 or -1 are selected as sensitivity parameters. Pearson correlation analysis is a commonly used statistical method to measure the linear relationship between two variables. Its principle is to calculate the quotient of the covariance and standard deviation between two variables to obtain a value between -1 and 1, which represents the strength of the linear correlation between the two variables. Assuming the independent variable X and the dependent variable Y, the steps to calculate the correlation between the two are as follows:

[0015] 1) Calculate the covariance between variables X and Y:

[0016] Cov(X,Y)=E[(XE(X))(YE(Y))] (1)

[0017] Where: X is the logging parameters (natural gamma, resistivity, acoustic time difference, density, etc.) and mineral component parameters (clay, quartz, calcite and carbonate content, etc.), Y is the measured P-wave velocity Vp or the measured S-wave velocity Vs, Cov(X,Y) is the covariance of X and Y, E(X) and E(Y) are the expected values ​​of variables X and Y, respectively.

[0018] The expected values ​​E(X) and E(Y) of variables X and Y are calculated as follows:

[0019] The distribution law of discrete random variable X is:

[0020] P{x i =p i i=1,2,3…n (2)

[0021] If the following series converges absolutely: Then the expected value E(X) of variable X is:

[0022] Where i is the depth sampling point of logging parameters and mineral component parameters.

[0023] E(Y) can be calculated in the same way as above.

[0024] 2) Calculate the standard deviation of variables X and Y:

[0025] First calculate the variance Var(X), Var(Y) of variables X and Y:

[0026] Var(X)=E[(XE(X)) 2 ],Var(Y)=E[(YE(Y)) 2 ] (5)

[0027] The standard deviations of variables X and Y are:

[0028]

[0029] 3) Calculate Pearson's coefficient R:

[0030]

[0031] The Pearson coefficient R is used to judge the sensitivity of the above X variables (logging parameters, mineral composition parameters) to the Y variables (P- and S-wave velocities).

[0032] When R>0, it means that variables X and Y are positively correlated, and when R<0, it means that variables X and Y are negatively correlated. The closer |R| is to 1, the better the correlation between variables X and Y, and the more sensitive it is. The closer |R| is to 0, the worse the correlation between variables X and Y.

[0033] According to the above scheme, in step S3, the preferred sensitivity parameters are used as independent variables, and the longitudinal and transverse wave velocities are used as dependent variables, and K-Means multi-parameter cluster analysis is performed, and a deep learning method based on the attention mechanism and the bidirectional long short-term memory network is combined to establish a longitudinal and transverse wave velocity prediction model, wherein the multi-parameter cluster analysis method is as follows:

[0034] 6) Initialization: That is, from the above sensitive parameter data set D = {x1, x2, ..., x m} randomly select K={u1,u2,…,u k} cluster points as the initial centroids.

[0035] 7) Assign clusters: Calculate the distance from each sensitive parameter data point to all centroids (usually using Euclidean distance). i and the center of mass u j , the Euclidean distance calculation formula is:

[0036]

[0037] Assign each data point to the cluster corresponding to the nearest centroid, that is:

[0038]

[0039] In the formula, x i is the data point of the above sensitive parameters (logging parameters, mineral composition parameters), u j The clustering points in the sensitive parameter set are the centroids.

[0040] 8) Update the centroid: For each sensitive parameter cluster (C i ), calculate the average value of all data points in the cluster and use the average value as the new centroid. The new centroid calculation formula is:

[0041]

[0042] 9) Iteration: Repeat the steps of assigning sensitive parameter clusters and updating the centroid until the centroid position no longer changes or the maximum number of iterations is reached.

[0043] 10) Output cluster partition C = {C1, C2, ..., C k}.

[0044] The optimization goal of this clustering algorithm is to minimize the sum of the squares of the distances from all sensitive parameter data points to the centroid of the cluster to which they belong. The optimization function can be expressed as:

[0045]

[0046] This function is also called the total square error within the sensitive parameter set, and the smaller the error, the higher the accuracy.

[0047] According to the above scheme, in step S3, the preferred sensitivity parameters are used as independent variables, and the longitudinal and transverse wave velocities are used as dependent variables, and K-Means multi-parameter clustering analysis is performed, and a deep learning method based on the attention mechanism and the bidirectional long short-term memory network is combined to establish a longitudinal and transverse wave velocity prediction model, wherein the deep learning method based on the attention mechanism and the bidirectional long short-term memory network (AT-BLSTM) is as follows:

[0048] 7) Input feature matrix x t Normalization, eliminating the impact of different logging curve dimension differences, x t is a two-dimensional matrix, k is the number of columns, representing the number of input curves:

[0049]

[0050] 8) Obtain the attention distribution of each element through the fully connected layer, and the output dimension is consistent with the number of features:

[0051] z=w·x+b

[0052] (13)In the formula, x represents the input feature matrix, i.e., the sensitive parameter cluster mentioned above, z represents the fully connected layer neurons, i.e., the corresponding longitudinal and transverse wave data clusters, the number of which is consistent with the number of columns in x, w is the weight matrix, and b is the bias term of each neuron.

[0053] 9) Normalize z through the Softmax function to obtain the weights a of different sensitive parameters:

[0054]

[0055] 10) Multiply the weight matrix by the corresponding element of the input feature matrix to get l, and get the output value of the attention layer, which is the input value of the BLSTM layer:

[0056]

[0057] 11) BLSTM layer: This layer uses the output value l of the attention layer as the input value, fully considering the impact of the upper and lower layer data on the P- and S-wave velocities, and obtains BLSTM by connecting two LSTM hidden layers with opposite time series. This method considers using a bidirectional long short-term memory network to mine the deep nonlinear relationship between the feature series and the P- and S-wave velocity labels.

[0058]

[0059] In the formula, T is the sequence length, which is related to the size of the sensitive parameter data cluster mentioned above. Then the output value of the BLSTM neural network is:

[0060]

[0061] 12) Output the predicted shear wave velocity value of the BLSTM layer.

[0062] When training a neural network, the output result h of the output layer is t After denormalization, the data is compared with the label data (i.e., the measured P- and S-wave velocity values), the error is obtained, back-propagated, and the parameters are iteratively updated until the training is completed.

[0063] According to the above scheme, step S4 is also included: using the P-wave and S-wave velocity prediction model established in S3 and comparing and analyzing it with the measured P-wave and S-wave velocities to verify the accuracy of the model. The specific accuracy comparison and analysis is as follows:

[0064]

[0065] Where y is the measured value of P-wave and S-wave velocity, and f is the predicted value output by the model. is the average of the measured values.

[0066] Usually, R 2 The closer it is to 1, the better the explanation of the corresponding variable of the independent variable in the regression analysis, and the higher the model accuracy. Otherwise, repeat the sensitive parameter optimization in step S2 and adjust the cluster centroid in step S3, and optimize the prediction model based on the deep learning method of the attention mechanism and the bidirectional long short-term memory network until the model prediction accuracy meets the actual needs.

[0067] The implementation of the P-wave and S-wave prediction method based on well logging rock physics and mineral composition data of the present invention has the following beneficial effects:

[0068] The present invention makes full use of the limited logging data and mineral composition data in the study area, and based on the sensitivity analysis of static data such as logging rock physical parameters and mineral composition and P- and S-wave velocities, adopts K-Means multi-parameter clustering analysis technology, and combines AT-BLSTM deep learning method to construct a P- and S-wave measurement model, thereby realizing P- and S-wave velocity prediction and correction. The present invention is simple to operate and easy to promote and apply. It has unique advantages in carrying out P- and S-wave velocity prediction and correction in areas where P- and S-wave measurements have not been carried out or where P- and S-wave data are distorted, and establishing a well-seismic joint reservoir prediction measurement version. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0070] Figure 1 This is a flow chart of a method for predicting longitudinal and transverse waves based on well logging rock physics and mineral composition data of the present invention;

[0071] Figure 2 Cross-plot for correlation analysis between P-wave velocity and natural gamma (GR) in the main target layer;

[0072] Figure 3 Cross plot for correlation analysis between P-wave velocity and deep-direction resistivity (Rt) of the main target layer;

[0073] Figure 4 Cross-plot for correlation analysis between P-wave velocity and calcite content in the main target layer;

[0074] Figure 5 The cross-plot of the correlation analysis between the P-wave velocity and the quartz content in the main target layer;

[0075] Figure 6 The cross plot of the correlation analysis between the P-wave velocity and clay content of the main target layer;

[0076] Figure 7 This is the cross-plot of the correlation analysis between the P-wave velocity and the carbonate content in the main target layer;

[0077] Figure 8 Cross-plot for correlation analysis between shear wave velocity and natural gamma (GR) in the main target layer;

[0078] Fig. 9 Cross plot for correlation analysis between shear wave velocity and deep direction resistivity (Rt) of the main target layer;

[0079] Fig.10 Cross-plot for correlation analysis between shear wave velocity and calcite content in the main target layer;

[0080] Fig.11 The cross-plot for the correlation analysis between shear wave velocity and quartz content in the main target layer;

[0081] Fig.12 Cross-plot for correlation analysis between shear wave velocity and clay content in the main target layer;

[0082] Fig.13 Cross-plot for correlation analysis between shear wave velocity and carbonate content in the main target layer;

[0083] Fig.14This is a comparative analysis diagram of the P-wave velocity (Vpp) of the main target layer predicted based on the model and the measured P-wave velocity (Vp);

[0084] Fig.15 This is a comparative analysis diagram of the shear wave velocity (Vsp) of the main target layer predicted based on the model and the measured shear wave velocity (Vs). DETAILED DESCRIPTION

[0085] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0086] like Figure 1 As shown, the present invention provides a method for predicting P- and S-waves based on well logging rock physics and mineral composition data, comprising the following steps:

[0087] S1. Standardize the logging data by combining regional geological and oil and gas testing data;

[0088] S2, combining the standardized processing results of step S1 to carry out correlation analysis between logging parameters, mineral components and P- and S-wave velocities, and optimize sensitive parameters;

[0089] S3. Using the K-Means multi-parameter clustering analysis method, a deep learning method based on the attention mechanism and the bidirectional long short-term memory network is used to establish a P-wave and S-wave velocity prediction model in combination with the sensitive parameters obtained in step S2.

[0090] Preferably, in step S1, the well logging data standardization process is carried out in combination with the regional geological and oil and gas testing data, specifically including the following steps:

[0091] S101. Carry out borehole environment correction based on regional geological and lithological logging data to reduce the impact of non-geological factors on logging data.

[0092] S102. Combined with the oil and gas test data, determine the standard wells and standard layers in the study area, carry out standardized processing of logging data, and provide reference standards for subsequent result comparison and accuracy analysis

[0093] Preferably, in step S2, a correlation analysis between logging parameters, mineral components and P- and S-wave velocities is carried out, and sensitivity parameters are optimized. A correlation analysis is carried out using the Pearson analysis method for the main target layers in the study area. The specific contents are as follows:

[0094] The Pearson correlation analysis method is used to conduct a correlation analysis on the P-wave velocity of the main target layer with the logging parameters and mineral components, and the correlation coefficient R between the logging parameters, mineral components and the P-wave velocity is obtained. The parameters with correlation coefficient R close to 1 or -1 are selected as sensitivity parameters. Pearson correlation analysis is a commonly used statistical method to measure the linear relationship between two variables. Its principle is to calculate the quotient of the covariance and standard deviation between two variables to obtain a value between -1 and 1, which represents the strength of the linear correlation between the two variables. Assuming the independent variable X and the dependent variable Y, the steps to calculate the correlation between the two are as follows:

[0095] 1) Calculate the covariance between variables X and Y:

[0096] Cov(X,Y)=E[(XE(X))(YE(Y))] (1)

[0097] Where: X is the logging parameters (natural gamma, resistivity, acoustic time difference, density, etc.) and mineral component parameters (clay, quartz, calcite and carbonate content, etc.), Y is the measured P-wave velocity Vp or the measured S-wave velocity Vs, Cov(X,Y) is the covariance of X and Y, E(X) and E(Y) are the expected values ​​of variables X and Y, respectively.

[0098] The expected values ​​E(X) and E(Y) of variables X and Y are calculated as follows:

[0099] The distribution law of discrete random variable X is:

[0100] P{x i =p i i=1,2,3…n (2)

[0101] If the following series converges absolutely:

[0102]

[0103] Then the expected value E(X) of variable X is:

[0104]

[0105] Where i is the depth sampling point of logging parameters and mineral component parameters.

[0106] E(Y) can be calculated in the same way as above.

[0107] 2) Calculate the standard deviation of variables X and Y:

[0108] First calculate the variance Var(X), Var(Y) of variables X and Y:

[0109] Var(X)=E[(XE(X)) 2],Var(Y)=E[(YE(Y)) 2 ] (5)

[0110] The standard deviations of variables X and Y are:

[0111]

[0112] 3) Calculate Pearson's coefficient R:

[0113]

[0114] The Pearson coefficient R is used to judge the sensitivity of the above X variables (logging parameters, mineral composition parameters) to the Y variables (P- and S-wave velocities).

[0115] When R>0, it means that variables X and Y are positively correlated, and when R<0, it means that variables X and Y are negatively correlated. The closer |R| is to 1, the better the correlation between variables X and Y, and the more sensitive it is. The closer |R| is to 0, the worse the correlation between variables X and Y.

[0116] Preferably, in step S3, the preferred sensitivity parameters are used as independent variables and the longitudinal and transverse wave velocities are used as dependent variables, and a K-Means multi-parameter cluster analysis is performed, and a deep learning method based on an attention mechanism and a bidirectional long short-term memory network is combined to establish a longitudinal and transverse wave velocity prediction model, wherein the multi-parameter cluster analysis method is as follows:

[0117] 11) Initialization: That is, from the above sensitive parameter data set D = {x1, x2, ..., x m} randomly select K={u1,u2,…,u k} cluster points as the initial centroids.

[0118] 12) Assign clusters: Calculate the distance from each sensitive parameter data point to all centroids (usually using Euclidean distance). i and the center of mass u j , the Euclidean distance calculation formula is:

[0119]

[0120] Assign each data point to the cluster corresponding to the nearest centroid, that is:

[0121]

[0122] In the formula, x i is the data point of the above sensitive parameters (logging parameters, mineral composition parameters), u j The clustering points in the sensitive parameter set are the centroids.

[0123] 13) Update the centroid: For each sensitive parameter cluster (C i ), calculate the average value of all data points in the cluster and use the average value as the new centroid. The new centroid calculation formula is:

[0124]

[0125] 14) Iteration: Repeat the steps of assigning sensitive parameter clusters and updating the centroid until the centroid position no longer changes or the maximum number of iterations is reached.

[0126] 15) Output cluster partition C = {C1, C2, ..., C k}.

[0127] The optimization goal of this clustering algorithm is to minimize the sum of the squares of the distances from all sensitive parameter data points to the centroid of the cluster to which they belong. The optimization function can be expressed as:

[0128]

[0129] This function is also called the total square error within the sensitive parameter set, and the smaller the error, the higher the accuracy.

[0130] Preferably, in step S3, the preferred sensitivity parameters are used as independent variables and the longitudinal and transverse wave velocities are used as dependent variables, and a K-Means multi-parameter clustering analysis is performed, and a deep learning method based on an attention mechanism and a bidirectional long short-term memory network is combined to establish a longitudinal and transverse wave velocity prediction model, wherein the deep learning method based on an attention mechanism and a bidirectional long short-term memory network (AT-BLSTM) is as follows:

[0131] 13) Input feature matrix x t Normalization, eliminating the impact of different logging curve dimension differences, x t is a two-dimensional matrix, k is the number of columns, representing the number of input curves:

[0132]

[0133] 14) Obtain the attention distribution of each element through the fully connected layer, and the output dimension is consistent with the number of features:

[0134] z=w·x+b

[0135] (13)In the formula, x represents the input feature matrix, i.e., the sensitive parameter cluster mentioned above, z represents the fully connected layer neurons, i.e., the corresponding longitudinal and transverse wave data clusters, the number of which is consistent with the number of columns in x, w is the weight matrix, and b is the bias term of each neuron.

[0136] 15) Normalize z through the Softmax function to obtain the weights a of different sensitive parameters:

[0137]

[0138] 16) Multiply the weight matrix by the corresponding element of the input feature matrix to get l, and get the output value of the attention layer, which is the input value of the BLSTM layer:

[0139]

[0140] 17) BLSTM layer: This layer uses the output value l of the attention layer as the input value, fully considering the impact of the upper and lower layer data on the P-wave and S-wave velocities, and obtains BLSTM by connecting two LSTM hidden layers with opposite time series. This method considers using a bidirectional long short-term memory network to mine the deep nonlinear relationship between the feature series and the P-wave and S-wave velocity labels.

[0141]

[0142] In the formula, T is the sequence length, which is related to the size of the sensitive parameter data cluster mentioned above. Then the output value of the BLSTM neural network is:

[0143]

[0144] 18) Output the predicted shear wave velocity value of the BLSTM layer.

[0145] When training a neural network, the output result h of the output layer is t After denormalization, the data is compared with the label data (i.e., the measured P- and S-wave velocity values), the error is obtained, back-propagated, and the parameters are iteratively updated until the training is completed.

[0146] Preferably, the method further includes step S4: using the P-wave and S-wave velocity prediction model established in S3, and performing comparative analysis with the measured P-wave and S-wave velocities to verify the accuracy of the model. The specific accuracy comparative analysis is as follows:

[0147]

[0148] Where y is the measured value of P-wave and S-wave velocity, and f is the predicted value output by the model. is the average of the measured values.

[0149] Usually, R 2 The closer it is to 1, the better the explanation of the corresponding variable of the independent variable in the regression analysis, and the higher the model accuracy. Otherwise, repeat the sensitive parameter optimization in step S2 and adjust the cluster centroid in step S3, and optimize the prediction model based on the deep learning method of the attention mechanism and the bidirectional long short-term memory network until the model prediction accuracy meets the actual needs.

[0150] The method of the present invention first analyzes the influence of logging rock physical parameters and mineral components on P-wave and S-wave velocities; then, an equivalent medium model is established using information such as natural gamma, resistivity, clay, quartz, calcite and carbonate to construct a constrained optimization equation, in which natural gamma and resistivity are constrained, and clay, quartz, calcite and carbonate information are optimized variables to find the optimal solution; finally, the method is verified using actual drilling data from a well in the southern Sichuan Basin, and the calculated P-wave and S-wave velocities are highly consistent with the actual data.

[0151] Example

[0152] See also Figures 1 to 15 In this embodiment, a method for predicting P- and S-waves based on well logging rock physics and mineral composition data is provided, comprising the following steps:

[0153] S1. Standardization of logging data based on regional geological and oil and gas testing data: Based on geological and oil and gas testing data, determine that there is no collapse and no expansion of mudstone marker layers, and use Well A as the standard well to standardize logging data. The statistics of standardized parameters are shown in Table 1.

[0154] Table 1. Standardized information of logging parameters

[0155]

[0156] S2. Carry out correlation analysis between logging parameters, mineral components and P- and S-wave velocities, and select sensitive parameters: Through statistical analysis, it is found that natural gamma (GR), quartz content (Quartz) and clay content (Clay) in mineral components are negatively correlated with P- and S-wave velocities; deep lateral resistivity (Rt), calcite content (Clacite) and carbonate content (Carbonate) in mineral components are positively correlated with P- and S-wave velocities. Among them, Well B was not involved in the correlation analysis as a verification well.

[0157] Table 2 Analysis of the sensitivity of logging parameters, mineral composition and P-wave velocity

[0158]

[0159] Table 3. Sensitivity analysis of logging parameters, mineral composition and shear wave velocity

[0160]

[0161] S3, taking the sensitivity parameters selected in S2 as independent variables and the longitudinal and transverse wave velocities as dependent variables, K-Means cluster analysis was performed, and the deep learning method based on the attention mechanism and bidirectional long short-term memory network was combined to finally determine the hyperparameters and network structure as follows: single-layer attention layer, double-layer BLSTM layer, 15 neurons each, 10 time steps, 0.001 learning rate, 45 iterations, and Sigmoid activation function. This example uses this structure to construct an AT-BLSTM network, thereby establishing a longitudinal and transverse wave velocity prediction model.

[0162] S4. Using the P-wave and S-wave velocity prediction calculation model established in S3, P-wave and S-wave prediction calculations were performed on actual well data. The calculation results were compared with the measured P-wave and S-wave data. The P-wave velocity prediction agreement rate of the main target layer was 93.52%, and the S-wave velocity prediction agreement rate was 95.17%. The model has high accuracy.

[0163] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for predicting longitudinal and transverse waves based on well logging rock physics and mineral composition data, characterized in that: The following steps are involved: S1. Standardize the logging data by combining regional geological and oil and gas testing data; S2, combining the standardized processing results of step S1 to carry out correlation analysis between logging parameters, mineral components and P- and S-wave velocities, and optimize sensitive parameters; S3. Using the K-Means multi-parameter clustering analysis method, a deep learning method based on the attention mechanism and the bidirectional long short-term memory network is used to establish a P-wave and S-wave velocity prediction model in combination with the sensitive parameters obtained in step S2.

2. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 1, characterized in that: In step S1, the standardization of logging data is carried out in combination with regional geological and oil and gas testing data, which specifically includes the following steps: S101. Conduct borehole environment correction based on regional geological and lithological logging data; S102. Based on the oil and gas testing data, determine the standard wells and standard layers in the study area and carry out standardized processing of well logging data.

3. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 1, characterized in that: In step S2, the Pearson correlation analysis method is used to carry out correlation analysis on the P- and S-wave velocities of the main target layers with the logging parameters and mineral components to obtain the correlation coefficient R between the logging parameters, mineral components and the P- and S-wave velocities, and the parameters with the correlation coefficient R close to 1 or -1 are selected as sensitivity parameters.

4. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 3, characterized in that: The Pearson correlation analysis method is used to conduct correlation analysis on the P- and S-wave velocities of the main target layers with logging parameters and mineral components as follows: Assuming the independent variable X and the dependent variable Y, the steps to calculate the correlation between the two are as follows: 1) Calculate the covariance between variables X and Y: Cov(X,Y)=E[(XE(X))(YE(Y))] (1) Where X is the logging parameter and mineral component parameter, Y is the measured P-wave velocity Vp or the measured S-wave velocity Vs, Cov(X,Y) is the covariance of X and Y, E(X) and E(Y) are the expected values ​​of variables X and Y respectively; The expected values ​​E(X) and E(Y) of variables X and Y are calculated as follows: The distribution law of discrete random variable X is: P{x i }=p i i=1,2,3…n (2) If the following series converges absolutely: Then the expected value E(X) of variable X is: Where i is the sampling point of well logging parameters and mineral composition parameters; E(Y) can be calculated in the same way as above; 2) Calculate the standard deviation of variables X and Y: Calculate the variance Var(X), Var(Y) of variables X and Y: Var(X)=E[(XE(X)) 2 ],Var(Y)=E[(YE(Y)) 2 ] (5) The standard deviations of variables X and Y are: 3) Calculate Pearson's coefficient R: The Pearson coefficient R is used to judge the sensitivity of the above X variable to the Y variable (P- and S-wave velocities). When R>0, it means that the variables X and Y are positively correlated, and when R<0, it means that the variables X and Y are negatively correlated. The closer |R| is to 1, the better the correlation between variables X and Y, and the more sensitive it is. The closer |R| is to 0, the worse the correlation between variables X and Y.

5. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 1, characterized in that: In step S3, the preferred sensitivity parameters are used as independent variables and the longitudinal and transverse wave velocities are used as dependent variables, and a K-Means multi-parameter clustering analysis is performed. In combination with a deep learning method based on an attention mechanism and a bidirectional long short-term memory network, a longitudinal and transverse wave velocity prediction model is established.

6. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 5, characterized in that: The K-Means multi-parameter cluster analysis method includes the following steps: 1) Initialization: That is, from the above sensitive parameter data set D = {x1, x2, ..., x m } randomly select K={u1,u2,…,u k } cluster points as the initial centroid; 2) Assign clusters: Calculate the distance from each sensitive parameter data point to all centroids (usually using Euclidean distance); for data point x i and the center of mass u j , the Euclidean distance calculation formula is: Assign each data point to the cluster corresponding to the nearest centroid, that is: In the formula, x i is the above sensitive parameter data point, u j The clustering point in the sensitive parameter set is the centroid; 3) Update the centroid: For each sensitive parameter cluster (C i ), calculate the average value of all data points in the cluster and use the average value as the new centroid; the new centroid calculation formula is: 4) Iteration: Repeat the steps of assigning sensitive parameter clusters and updating the centroid until the centroid position no longer changes or the maximum number of iterations is reached; 5) Output cluster partition C = {C1, C2, ..., C k }; The optimization goal of this clustering algorithm is to minimize the sum of the squares of the distances from all sensitive parameter data points to the centroid of the cluster to which they belong. The optimization function is: This function is also called the total square error within the sensitive parameter set, and the smaller the error, the higher the accuracy.

7. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 5, characterized in that: The deep learning method based on attention mechanism and bidirectional long short-term memory network includes the following steps: 1) Input feature matrix x t Normalization, x t is a two-dimensional matrix, k is the number of columns, representing the number of input curves: 2) Obtain the attention distribution of each element through the fully connected layer, and the output dimension is consistent with the number of features: z=w·x+b (13) In the formula, x represents the input feature matrix, that is, the sensitive parameter cluster mentioned above, z represents the fully connected layer neurons, that is, the corresponding longitudinal and transverse wave data clusters, the number of which is consistent with the number of columns in x, w is the weight matrix, and b is the bias term of each neuron; 3) Normalize z through the Softmax function to obtain the weights a of different sensitive parameters: 4) Multiply the weight matrix by the corresponding element of the input feature matrix to get l, and get the output value of the attention layer, which is the input value of the BLSTM layer: 5) BLSTM layer: This layer uses the output value l of the attention layer as the input value and obtains the BLSTM by connecting two LSTM hidden layers with opposite time series; Where T is the sequence length, the output value of the BLSTM neural network is: 6) Output the predicted shear wave velocity value of the BLSTM layer.

8. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 7, characterized in that: The method also includes step S4: using the P-wave and S-wave velocity prediction model established in S3 to perform P-wave and S-wave prediction calculations on actual well data, comparing the calculation results with the measured P-wave and S-wave data, and performing accuracy analysis on the calculation results.

9. The method for predicting P- and S-waves based on well logging rock physics and mineral composition data according to claim 8, characterized in that: In step S4, the P-wave and S-wave velocity prediction model established in S3 is used and compared with the measured P-wave and S-wave velocities to verify the accuracy of the model. The specific accuracy comparison analysis is as follows: Where y is the measured value of P-wave and S-wave velocities, and f is the predicted value output by the model. is the average of the measured values; When training a neural network, the output result h of the output layer is t After denormalization, the values ​​are compared with the measured P- and S-wave velocities to obtain the error, perform back propagation, and iteratively update the parameters until the training is completed.

Images