VTI medium fluid parameter bayesian inversion method based on independent prior information

Abstract

The application discloses a VTI medium fluid parameter Bayesian inversion method based on independent prior information. The method extracts independent prior information between to-be-inverted fluid parameters (fluid factors, shear modulus, density) and anisotropy parameters ( and ), and applies the independent prior information to a Bayesian inversion technology, and finally, through inverse independent prior information transformation, the final inversion result is obtained. The technology solves the coupling problem of conventional fluid parameter inversion results, and can obtain fluid parameter inversion results closer to real values.

Description

Technical Field

[0001] This invention belongs to the field of shale reservoir exploration and development technology, and specifically relates to a Bayesian inversion method for VTI medium fluid parameters based on independent prior information. Background Technology

[0002] Fluid identification and lithological prediction are of great significance in shale reservoir exploration and development. Among them, the accuracy of fluid identification directly affects the prediction of reservoir oil and gas distribution. Shale gas reservoirs have well-developed horizontal bedding and exhibit obvious VTI medium characteristics, which significantly alters seismic reflection characteristics.

[0003] In fluid identification, based on porosity elasticity theory, Russell et al. proposed the Russell fluid factor to describe the corresponding solid and fluid components in elastic reservoirs. Since this fluid property has a clear physical meaning, it reduces the uncertainty in fluid identification. In deterministic inversion theory, the relationship between broadband impedance and frequency-varying viscoelastic fluid factor under viscoelastic medium conditions was established, enabling pre-stack seismic inversion of the frequency-varying viscoelastic fluid factor. The quality of fluid detection can be effectively improved by using a joint PP wave and PS wave inversion method. Furthermore, the square of the P-wave / S-wave velocity ratio under dry rock conditions is an important parameter in fluid factor inversion. Treating it as a dynamic variable that varies with depth can overcome the limitation in fluid factor prediction accuracy when DVRS is treated as a constant. In Bayesian theory-based inversion, prior information is represented by the prior probability distribution function of the model parameters. Among various types of distribution functions, the Gaussian distribution is the most common. Hansen et al., under the condition of model parameters satisfying the Gaussian distribution, combined Gaussian linear inversion with geostatistics, proposing a linear geostatistical inversion method. Because AVO pre-stack gathers contain more accurate fluid information, pre-stack Bayesian inversion based on AVO data is widely used for fluid factor estimation and has achieved good results in practical applications.

[0004] For VTI media, Garebner derived the accurate propagation equations for reflection and transmission of plane waves in VTI media based on the Zoeppritz equations under the isotropic assumption. Hou Dongjia et al. achieved pre-stack joint inversion of multiple waves in VTI media based on Bayesian theory. Yin Xingyao et al. derived a new elastic wave impedance formula expressed in terms of fluid factor, Lamé constant, and density based on Russell's approximation formula, and obtained parameters such as fluid factor directly through elastic impedance inversion.

[0005] Most existing inversion methods are based on deterministic inversion principles. However, for shale oil and gas exploration and development, the anisotropy in the VTI medium is non-negligible, seismic data is noisy, and the inversion process involves matrix inversion. Therefore, fluid factor predictions based on deterministic inversion are not only inaccurate but also computationally expensive. Furthermore, existing Bayesian theory-based inversion methods struggle to overcome the correlations between the parameters to be inverted, inevitably leading to prediction trend coupling in the inversion results. Additionally, current technologies lack uncertainty analysis for the anisotropic parameters of the VTI medium. Summary of the Invention

[0006] To address the problems existing in the background technology, this application provides a Bayesian inversion method for VTI medium fluid parameters based on independent prior information. This application extracts the fluid parameters to be inverted (fluid factor, shear modulus, density) and anisotropy parameters (…). , This application solves the problem of coupling in conventional fluid parameter inversion results, and can obtain fluid parameter inversion results that are closer to the true values. It utilizes independent prior information between fluid parameters and applies it to the Bayesian inversion method. Finally, through anti-independent prior information transformation, the final inversion result is obtained.

[0007] The technical solution provided by this invention is: a Bayesian inversion method for VTI medium fluid parameters based on independent prior information, comprising the following steps:

[0008] Step (1) Pre-stack seismic data forward modeling and incident angle stacking: Pre-stack seismic data is stacked according to actual needs to output stacked profile data volumes with different incident angles. The number of incident angles should be at least 3.

[0009] Step (2) Initial model acquisition: Combine well logging information and stratigraphic information to output the initial model of the fluid parameters and anisotropy parameters to be inverted, which serves as a low-frequency constraint suitable for Bayesian inversion of VTI media;

[0010] Step (3) Independent Prior Information Operator Acquisition: Based on well logging information, calculate the independent prior information operator to update the seismic data forward modeling operator;

[0011] Step (4) Perform Bayesian inversion of pre-stack VTI medium fluid parameters and anisotropic parameters based on independent prior information, and output the inversion results containing independent prior information;

[0012] Step (5) Apply the anti-independent prior information operator to restore the true fluid parameters and anisotropic parameters inversion results.

[0013] In step (1) above, the specific process is as follows:

[0014] Since different fluids have different fluid factors, the fluid type in saturated rock can be predicted based on the inverted fluid factor. The expression for the fluid factor is:

[0015] (1);

[0016] in, For fluid factor terms, The density of saturated rock, and These represent the P-wave velocity and S-wave velocity under saturated rock conditions, respectively. The square of the ratio of P-wave to S-wave velocity under dry rock conditions;

[0017] An approximate formula in the form of the Aki-Richards equations is proposed using fluid factor, shear modulus, and density terms to construct the seismic reflection coefficient. This approximation is expressed as:

[0018] (2);

[0019] in, Represents the reflection coefficient of an isotropic medium. , and These represent the reflection coefficient variables corresponding to the fluid factor, shear modulus, and density terms, respectively. and Representing the shear modulus term and the density term, respectively, and Together they are called the three parameters of fluid. , and These are the mean values ​​of the fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. , and These are the differences in fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. The square of the ratio of P-wave to S-wave velocity under saturated rock conditions;

[0020] Combining formula (2) with the VTI anisotropy parameter term, we obtain the fluid parameter reflection coefficient equation applicable to VTI media, which is expressed as:

[0021] (3);

[0022] in, Represents the reflection coefficient under VTI medium conditions. Let represent the anisotropic reflection coefficient, and write it as:

[0023] (4);

[0024] in, and These are the anisotropic parameters of the media on both sides of the reflecting interface. and anisotropy parameters The difference;

[0025] Forward modeling operator for earthquake records Writing the seismic wavelet matrix Seismic reflection coefficient matrix and first-order difference matrix The folding form:

[0026] (5);

[0027] Among them, the earthquake reflection coefficient matrix The elements in the first-order difference matrix are obtained using equation (3). It can be written as:

[0028] (6);

[0029] Within the framework of convolution forward modeling, seismic records applicable to VTI media, obtained through forward modeling of fluid factor, shear modulus, and density, can be expressed as:

[0030] (7);

[0031] in, This is the earthquake record obtained through forward modeling; For earthquake forward modeling operators; This includes fluid factor, shear modulus, density, and anisotropic parameters. and anisotropy parameters A model with 5 objective parameters; Noise item;

[0032] In the simulation example, pre-stack seismic data with different incident angles are obtained based on the formula. In actual seismic data, the required pre-stack seismic data are obtained by sorting the incident angles.

[0033] In step (2) above: In the simulation data, the initial model is obtained by Gaussian smoothing of the existing model. For the actual seismic data, the unknown information between wells and between layers is interpolated and Gaussian smoothed based on the existing well logging information and layer information to establish the initial model, which is used as a low-frequency constraint for Bayesian inversion; the unknown information between wells and between layers is interpolated and Gaussian smoothed to establish the initial model.

[0034] Based on the above formula (7), the model parameters are... The higher-order covariance matrix is ​​written as:

[0035] (8);

[0036] in, Let be the number of sampling points for the seismic data. The submatrices on the main diagonal represent the higher-order variance matrices of the five parameters to be inverted, and the submatrices on the secondary diagonal represent the higher-order covariance matrices between each pair of the five parameters to be inverted. The size of each of the above submatrices is . ;

[0037] First, the independent parameter information at a certain point in the parametric model is obtained; then the model parameters are extracted. The three parameters at a certain point constitute the target parameter vector. Its corresponding fifth-order variance matrix is ​​expressed as:

[0038] (9);

[0039] The elements on the main diagonal represent the variance of each parameter with respect to itself, while the elements on the side diagonal represent the covariance between each pair of the three parameters. Through the analysis of... Principal component analysis yielded the following results:

[0040] (10);

[0041] in, For the target parameter vector Principal component analysis operator, for The principal component matrices, where T denotes the matrix transpose, are represented as follows:

[0042] (11);

[0043] (12);

[0044] In Equation (12), the five parameters of the main diagonal represent the covariance of the five parameters to be inverted after principal component analysis.

[0045] If If we consider it as a new variance matrix, then we denote its corresponding new parameter model as... At this point, the variance matrix of the new parameter model only contains the variance of the parameters themselves, and the covariance between the parameters becomes zero, which means that the new parameters are independent of each other.

[0046] Next, the third-order principal component analysis operator is extended to satisfy the number of sampling points. Format:

[0047] (13);

[0048] in, Each submatrix has the same form, with For example, it can be written as:

[0049] (14);

[0050] Then matrix To be applicable to a number of sampling points parametric model Independent prior information operator.

[0051] In step (4) above: based on the linear Bayesian inversion result as the maximum probability distribution characteristic of the parameter to be inverted, the uncertainty analysis of the inversion result is performed using the posterior mean, so that the target parameter and the noise term both follow a multivariate Gaussian distribution:

[0052] (15);

[0053] (16);

[0054] Where: N represents a Gaussian distribution; and Representing the model respectively The mean and covariance; Noise item The mean and covariance;

[0055] The new independent model parameters after principal component analysis are defined as follows:

[0056] (17);

[0057] in, This represents the parametric model after principal component analysis;

[0058] The parameters of the independent model still satisfy a Gaussian distribution:

[0059] (18);

[0060] in, and Representing the model respectively The mean and covariance;

[0061] Furthermore, its corresponding forward operator Updated to:

[0062] (19);

[0063] in, Representation Model The corresponding forward operator;

[0064] Therefore, the relationship between the seismic data and the updated parametric model can be written as follows:

[0065] (20);

[0066] According to the linear Gaussian principle, the distribution function of seismic data can be expressed as:

[0067] (twenty one);

[0068] in, and Let represent the mean and covariance of the earthquake record, respectively.

[0069] Since it is assumed that both the target parameter and the noise term are Gaussian distributed, according to statistical properties, the posterior distribution of the target parameter is also of multivariate Gaussian type. Therefore, the posterior Gaussian distribution of the target parameter under seismic information constraints is obtained, and its posterior mean is... and posterior covariance It can be represented as:

[0070] (twenty two).

[0071] In step (5) above: the initial model parameters are restored using the independent information operator:

[0072] (twenty three).

[0073] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0074] 1. This method is a Bayesian inversion technique for VTI medium fluid parameters based on independent prior information. Independent prior information technology is being applied to the geophysical inversion of VTI medium fluid parameters for the first time, and has achieved good results in simulation data testing and actual data application.

[0075] 2. This method combines the independent prior information operator extracted from well logging information with Bayesian theory to update the linear Bayesian inversion algorithm, avoiding the trend coupling phenomenon of inversion results that is difficult to solve by conventional inversion techniques, and solving the current situation of difficulty in inverting fluid terms and anisotropic parameter terms.

[0076] 3. This method has good algorithm integration and can achieve modular processing, making it easy for seismic data interpreters to use and giving it high promotional value. Attached Figure Description

[0077] Figure 1 This is a flowchart of the Bayesian inversion method for VTI medium fluid parameters based on independent prior information;

[0078] Figure 2 This is a diagram of the SEG Overthrust model; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters ;

[0079] Figure 3 These are pre-stack seismic data maps for different incident angles in Example 1; where: (a) incident angle is 8°; (b) incident angle is 13°; (c) incident angle is 18°; (d) incident angle is 23°;

[0080] Figure 4 This is the initial model diagram of Example 1; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters ;

[0081] Figure 5 It is the independent prior information operator calculated based on the Overthrust model in Example 1;

[0082] Figure 6 This is a two-dimensional inversion result diagram from Example 1; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters .

[0083] Figure 7 This is a one-dimensional inversion result diagram of Example 1; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters ;

[0084] Figure 8 This is a map of actual pre-stack seismic data from Example 2; where: (a) incident angle is 5°; (b) incident angle is 15°; (c) incident angle is 25°;

[0085] Figure 9 This is the initial model diagram of Example 2; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters ;

[0086] Figure 10 Example 2 is based on Figure 8 A graph showing the results of operators with independent prior information obtained from data computation;

[0087] Figure 11This is a one-dimensional inversion result diagram from Example 2; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters ;

[0088] Figure 12 This is a two-dimensional inversion result diagram from Example 2; where: (a) fluid factor; (b) shear modulus; (c) density; (d) anisotropy parameter. (e) Anisotropy parameters . Detailed Implementation

[0089] To clarify the technical advantages of the present invention, the design scheme of the present invention will be described in further detail and clearly below with reference to the accompanying drawings and embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0090] Example 1: A flowchart of a Bayesian linear inversion method for fluid parameters based on independent prior information is shown below. Figure 1 As shown, this application mainly includes five steps:

[0091] The SEG Overthrust model was used for testing. This model consists of five parts: fluid factor, shear modulus, density, anisotropy parameter, and so on. and anisotropy parameters The models, and the results of each model are as follows: Figure 2 As shown in (a)-(e). Each model has 800 channels, and each channel contains 181 sampling points.

[0092] Step 1: Pre-stack seismic data forward modeling and incident angle stacking: Four incident angles are used, namely 8°, 13°, 18° and 23°. These are used to output stacked profile data volumes with different incident angles.

[0093] Different fluids have different fluid factors, therefore the fluid type in saturated rocks can be predicted based on the inverted fluid factors. The expression for the fluid factor is:

[0094] (1);

[0095] in, For fluid factor terms, The density of saturated rock, and These represent the P-wave velocity and S-wave velocity under saturated rock conditions, respectively. This represents the square of the P-wave to S-wave velocity ratio under dry rock conditions. Extensive experimental studies have revealed that in sandstone reservoirs... A value of 2.333 is usually more reasonable.

[0096] An approximate formula in the form of the Aki-Richards equations is proposed using fluid factor, shear modulus, and density terms to construct the seismic reflection coefficient. This approximation can be expressed as:

[0097] (2);

[0098] in, Represents the reflection coefficient of an isotropic medium. , and These represent the reflection coefficient variables corresponding to the fluid factor, shear modulus, and density terms, respectively. and Representing the shear modulus term and the density term, respectively, and Together they are called the three parameters of fluid. , and These are the mean values ​​of the fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. , and These are the differences in fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. It is the square of the ratio of P-wave to S-wave velocity under saturated rock conditions, and is usually taken as 4; The incident angle of the seismic wave is denoted as . This expression for the reflection coefficient clearly describes the relationship between the seismic reflection coefficient and three parameters: fluid factor, shear modulus, and density.

[0099] Combining formula (2) with the VTI anisotropy parameter term, we can obtain the fluid parameter reflection coefficient equation applicable to VTI media, which can be expressed as:

[0100] (3);

[0101] in, Represents the reflection coefficient under VTI medium conditions. Let represent the anisotropic reflection coefficient, and it can be written as:

[0102] (4);

[0103] in, and These are the anisotropic parameters of the media on both sides of the reflecting interface. and anisotropy parameters The difference.

[0104] Forward modeling operator for earthquake records It can be written as a seismic wavelet matrix. Seismic reflection coefficient matrix and first-order difference matrix The convolutional form (Buland and More, 2003):

[0105] (5);

[0106] Among them, the earthquake reflection coefficient matrix The elements in the first-order difference matrix are obtained using equation (3). It can be written as:

[0107] (6);

[0108] Within the framework of convolution forward modeling, seismic records applicable to VTI media, obtained through forward modeling of fluid factor, shear modulus, and density, can be expressed as:

[0109] (7);

[0110] in, This is the earthquake record obtained through forward modeling; For earthquake forward modeling operators; This includes fluid factor, shear modulus, density, and anisotropic parameters. and anisotropy parameters A model with 5 objective parameters; This is the noise item.

[0111] In the simulation example, pre-stack seismic data with different incident angles can be obtained based on the formula. In actual seismic data, the required pre-stack seismic data can be obtained by sorting the incident angles. The results are as follows: Figure 3 As shown in (a)-(d).

[0112] Step 2: Initial Model Acquisition

[0113] In the simulation data, the initial model is obtained by Gaussian smoothing the existing model. For the actual seismic data, the unknown information between wells and between layers is interpolated and Gaussian smoothed based on the existing well logging information and layer information to establish the initial model, which is used as a low-frequency constraint for Bayesian inversion. The results are shown in Figure 4(a)-(e).

[0114] Step 3: Independent Prior Information Operator Acquisition: Based on well logging information, calculate the independent prior information operator to update the seismic data forward modeling operator;

[0115] Based on formula (7) in step 1, the model parameters The higher-order covariance matrix can be written as:

[0116] (8);

[0117] in, This represents the number of sampling points for the seismic data. The submatrices on the main diagonal represent the higher-order variance matrices of the five parameters to be inverted, and the submatrices on the secondary diagonal represent the higher-order covariance matrices between each pair of the five parameters to be inverted. The size of each of these submatrices is [missing value]. Covariance matrix It not only represents the correlation between parameters of different types of target models, but also reflects the spatial correlation between different sampling points. This spatial correlation has a significant impact on the accuracy of the inversion.

[0118] First, the independent parameter information at a certain point in the parametric model is obtained. Then, the model parameters are extracted. The three parameters at a certain point constitute the target parameter vector. Its corresponding fifth-order variance matrix can be expressed as:

[0119] (9);

[0120] The elements on the main diagonal represent the variance of each parameter with respect to itself, while the elements on the side diagonal represent the covariance between each pair of the three parameters. Through analysis of... Principal component analysis yields the following results:

[0121] (10);

[0122] in, For the target parameter vector Principal component analysis operator, for The principal component matrix, T, denotes the matrix transpose, and can be represented as follows:

[0123] (11);

[0124] (12);

[0125] In Equation (12), the five parameters of the main diagonal represent the covariance of the five parameters to be inverted after principal component analysis.

[0126] If Viewed as a new variance matrix, its corresponding new parameter model can be denoted as... At this point, the variance matrix of the new parameter model only contains the variance of the parameters themselves, and the covariance between the parameters becomes zero, which means that the new parameters are independent of each other.

[0127] Next, the third-order principal component analysis operator is extended to satisfy the number of sampling points. Format:

[0128] (13);

[0129] in, Each submatrix has the same form, with For example, it can be written as:

[0130] (14);

[0131] Then matrix To be applicable to a number of sampling points parametric model The independent prior information operator. The result is as follows: Figure 5 As shown.

[0132] Step 4: Perform pre-stack VTI medium fluid parameters Bayesian inversion based on independent prior information. The input to this module is the output of the first three modules, and the output is the inversion result containing independent prior information.

[0133] The results of linear Bayesian inversion represent the maximum probability distribution characteristics of the parameters to be inverted. The posterior mean can be used to analyze the uncertainty of the inversion results. Let the target parameters and noise term both follow a multivariate Gaussian distribution:

[0134] (15);

[0135] (16);

[0136] The new independent model parameters after principal component analysis are defined as follows:

[0137] (17);

[0138] The parameters of the independent model still satisfy a Gaussian distribution:

[0139] (18);

[0140] Furthermore, its corresponding forward operator It can be updated to:

[0141] (19);

[0142] Therefore, the relationship between the seismic data and the updated parametric model can be written as:

[0143] (20);

[0144] According to the linear Gaussian principle, the distribution function of seismic data can be expressed as:

[0145] (twenty one);

[0146] in, and These represent the mean and covariance of the earthquake record, respectively.

[0147] Since it is assumed that both the target parameter and the noise term are Gaussian distributed, according to statistical properties, the posterior distribution of the target parameter is also multivariate Gaussian. Ultimately, the posterior Gaussian distribution of the target parameter under seismic information constraints can be obtained, and its posterior mean is... and posterior covariance It can be represented as:

[0148] (twenty two);

[0149] Step 5: Calculation of the anti-prior information operator: This module is used to eliminate the influence of independent prior information and restore the true fluid parameters and anisotropic parameters inversion results.

[0150] Finally, the independent information operators are used to restore the initial model parameters from their inverse independent information state:

[0151] (twenty three);

[0152] The final Bayesian linear inversion result based on independent prior information is as follows: Figure 6 As shown. It is worth noting that the output of step 5 is the final inversion result, which is based on the result of step 4. It can be calculated and restored in one step. Moreover, the result of step 4 does not have a clear physical meaning, so the result of step 4 does not need to be output separately.

[0153] Furthermore, comparing the one-dimensional inversion results of our method with those of conventional inversion techniques, such as... Figure 7 As shown, where Figure 7 (a) The results are obtained by applying Bayesian inversion techniques for VTI medium fluid parameters based on independent prior information. Figure 7 (b) The figures show the inversion results obtained using conventional inversion techniques. The dashed lines represent the true values, and the solid lines represent the inversion results. It can be seen that existing inversion techniques applicable to VTI media are affected by the correlation between parameters. In the inversion results, especially the density term and the two anisotropic parameters, the deviations from the true values ​​are significant, and the trends of the true model cannot be reproduced. However, the Bayesian inversion technique for VTI media fluid parameters based on independent prior information can accurately reproduce the true model.

[0154] Example 2: After verifying the feasibility of this method in Example 1, the method was used to estimate the fluid parameters and anisotropy parameters of a well in a certain actual work area. Actual seismic data from that work area was used for testing. This data contains 180 seismic traces, with a sampling time of 2 seconds and a sampling rate of 2 milliseconds. The steps are as follows:

[0155] Step 1: Using seismic data processing software, stack the actual seismic data by incident angle. The result is as follows: Figure 8 As shown, the incident angles are 5°, 15° and 25° respectively.

[0156] Step 2: Establish an initial model for the work area using actual well logging curves and stratigraphic information, such as... Figure 9 As shown.

[0157] Step 3: Calculate the independent prior information operator based on this data, such as... Figure 10 As shown.

[0158] Steps 4 and 5: The final Bayesian linear inversion results based on independent prior information are as follows Figure 11 and Figure 12 As shown in the figure. First, a one-dimensional inversion test was performed. The one-dimensional inversion results and the actual logging curves were compared at the selected well location. The results are as follows. Figure 11 As shown in the figure, the solid lines represent well logging curves, and the dashed lines represent inversion results, from left to right: three fluid parameters and two anisotropic parameters. It can be seen that the inversion results match the actual well logging curves well, demonstrating the effectiveness and accuracy of the proposed method. Finally, two-dimensional data application is performed, and the two-dimensional inversion results for the five parameters are shown below. Figure 12 As shown in (a)-(e), it can be seen that the two-dimensional inversion results have high resolution and lateral continuity.

Claims

1. A Bayesian inversion method for VTI medium fluid parameters based on independent prior information, comprising the following steps: Step (1) Pre-stack seismic data forward modeling and incident angle stacking: Pre-stack seismic data is stacked according to actual needs to output stacked profile data volumes with different incident angles. The number of incident angles should be at least 3. Step (2) Initial model acquisition: Combine well logging information and stratigraphic information to output the initial model of the fluid parameters and anisotropy parameters to be inverted, which serves as a low-frequency constraint suitable for Bayesian inversion of VTI media; Step (3) Independent Prior Information Operator Acquisition: Based on well logging information, calculate the independent prior information operator to update the seismic data forward modeling operator; Step (4) Perform Bayesian inversion of pre-stack VTI medium fluid parameters and anisotropic parameters based on independent prior information, and output the inversion results containing independent prior information; Step (5) Apply the anti-independent prior information operator to restore the true fluid parameters and anisotropic parameters inversion results; In step (4): based on the linear Bayesian inversion result as the maximum probability distribution feature of the parameters to be inverted, the uncertainty analysis of the inversion result is performed using the posterior mean, so that the target parameter and the noise term both follow a multivariate Gaussian distribution: (15); (16); Where: N represents a Gaussian distribution; and Representing the model respectively The mean and covariance; Noise item The mean and covariance; The new independent model parameters after principal component analysis are defined as follows: (17); in, This represents the parametric model after principal component analysis; The parameters of the independent model still satisfy a Gaussian distribution: (18); in, and Representing the model respectively The mean and covariance; Furthermore, its corresponding forward operator Updated to: (19); in, Representation Model The corresponding forward operator; Therefore, the relationship between the seismic data and the updated parametric model can be written as follows: (20); According to the linear Gaussian principle, the distribution function of seismic data can be expressed as: (21); in, and Let represent the mean and covariance of the earthquake record, respectively. Since it is assumed that both the target parameter and the noise term are Gaussian distributed, according to statistical properties, the posterior distribution of the target parameter is also of multivariate Gaussian type. Therefore, the posterior Gaussian distribution of the target parameter under seismic information constraints is obtained, and its posterior mean is... and posterior covariance It can be represented as: (22)。 2. The method according to claim 1, characterized in that, In step (1), the specific process is as follows: Since different fluids have different fluid factors, the fluid type in saturated rock can be predicted based on the inverted fluid factor. The expression for the fluid factor is: (1)； in, For fluid factor terms, The density of saturated rock, and These represent the P-wave velocity and S-wave velocity under saturated rock conditions, respectively. The square of the ratio of P-wave to S-wave velocity under dry rock conditions; An approximate formula in the form of the Aki-Richards equations is proposed using fluid factor, shear modulus, and density terms to construct the seismic reflection coefficient. The approximate formula is expressed as: (2)； in, Represents the reflection coefficient of an isotropic medium. , and These represent the reflection coefficient variables corresponding to the fluid factor, shear modulus, and density terms, respectively. and Representing the shear modulus term and the density term, respectively, and Together they are called the three parameters of fluid. , and These are the mean values ​​of the fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. , and These are the differences in fluid factor, shear modulus, and density of the media on both sides of the reflective interface, respectively. The square of the ratio of P-wave to S-wave velocity under saturated rock conditions; Combining formula (2) with the VTI anisotropy parameter term, we obtain the fluid parameter reflection coefficient equation applicable to VTI media, which is expressed as: (3)； in, Represents the reflection coefficient under VTI medium conditions. Let represent the anisotropic reflection coefficient, and write it as: (4); in, and These are the anisotropic parameters of the media on both sides of the reflecting interface. and anisotropy parameters The difference; Forward modeling operator for earthquake records Writing the seismic wavelet matrix Seismic reflection coefficient matrix and first-order difference matrix The folding form: (5)； Among them, the earthquake reflection coefficient matrix The elements in the first-order difference matrix are obtained using equation (3). It can be written as: (6)； Within the framework of convolution forward modeling, seismic records applicable to VTI media, obtained through forward modeling of fluid factor, shear modulus, and density, can be expressed as: (7)； in, This is the earthquake record obtained through forward modeling; For earthquake forward modeling operators; This includes fluid factor, shear modulus, density, and anisotropic parameters. and anisotropy parameters A model with 5 objective parameters; Noise item; In the simulation example, pre-stack seismic data with different incident angles are obtained based on the formula. In actual seismic data, the required pre-stack seismic data are obtained by sorting the incident angles.

3. The method according to claim 2, characterized in that, In step (2) of the above: In the simulation data, the initial model is obtained by Gaussian smoothing the existing model. For the actual seismic data, the unknown information between wells and between layers is interpolated and Gaussian smoothed based on the existing well logging information and layer information to establish the initial model, and the fluid factor, shear modulus, density, and anisotropy parameters to be inverted are output. and anisotropy parameters The initial model is used for Bayesian inversion as a low-frequency constraint.

4. The method according to claim 2, characterized in that, In step (3) above: based on formula (7) above, the model parameters... The higher-order covariance matrix is ​​written as: (8)； in, Let be the number of sampling points for the seismic data. The submatrices on the main diagonal represent the higher-order variance matrices of the five parameters to be inverted, and the submatrices on the secondary diagonal represent the higher-order covariance matrices between each pair of the five parameters to be inverted. The size of each of the above submatrices is . ; First, the independent parameter information at a certain point in the parametric model is obtained; then the model parameters are extracted. The three parameters at a certain point constitute the target parameter vector. Its corresponding fifth-order variance matrix is ​​expressed as: (9)； The elements on the main diagonal represent the variance of each parameter with respect to itself, while the elements on the side diagonal represent the covariance between each pair of the three parameters. Through the analysis of... Principal component analysis yielded the following results: (10)； in, For the target parameter vector Principal component analysis operator, for The principal component matrices, where T denotes the matrix transpose, are represented as follows: (11)； (12)； In Equation (12), the five parameters of the main diagonal represent the covariance of the five parameters to be inverted after principal component analysis. If If we consider it as a new variance matrix, then we denote its corresponding new parameter model as... At this point, the variance matrix of the new parameter model only contains the variance of the parameters themselves, and the covariance between the parameters becomes zero, which means that the new parameters are independent of each other. Next, the third-order principal component analysis operator is extended to satisfy the number of sampling points. Format: (13)； in, Each submatrix has the same form, with For example, it can be written as: (14)； Then matrix To be applicable to a number of sampling points parametric model Independent prior information operator.

5. The method according to claim 4, characterized in that, In step (5): the initial model parameters are restored using the independent information operator: (23)。

Images