An elastic parameter prestack inversion method based on frequency blending modeling

Abstract

This invention discloses a pre-stack inversion method for elastic parameters based on frequency fusion modeling. The inversion method includes: constructing an optimal initial frequency model integrating multi-scale characteristics according to the target reservoir prediction requirements; performing frequency division processing on a portion of the stacked data volume using variable time window wavelet transform to obtain frequency-divided seismic data volumes; establishing low-frequency, mid-frequency, and high-frequency models using multiple model construction methods respectively; fusing the low-frequency, mid-frequency, and high-frequency models into an optimal frequency model through a linear hybrid model according to the target reservoir prediction requirements; constructing a P-impedance inversion objective function constrained by the optimal frequency model to perform the inversion; solving for the elastic parameters using a stochastic direction search method constrained by well logging prior information; and providing guidance for reservoir prediction and comprehensive evaluation of compressed air energy storage traps. This invention improves the objectivity and accuracy of the inversion results.

Description

Technical Field

[0001] This invention relates to the fields of oil and gas geophysical exploration and underground porous media compressed air energy storage, and particularly to a pre-stack inversion method for elastic parameters based on frequency fusion modeling. Background Technology

[0002] As oil and gas exploration across the country intensifies, exploration targets become increasingly concealed, raising the exploration difficulty level. Geophysical exploration technologies such as reservoir prediction and sweet spot identification face higher requirements and greater challenges. Channel sands, as important carriers of oil and gas, have always been a key target in oil and gas exploration due to their good physical properties and high permeability. Especially in recent years, with the green, low-carbon, and new energy transformation of oilfields, shallow and medium-depth water-bearing channel sands with certain tectonic backgrounds have become the preferred targets for compressed air energy storage due to their shallow burial depth and good physical properties. However, these reservoirs often develop into thin, small reservoirs of varying thicknesses, often overlapping each other, with complex longitudinal and lateral boundaries, connectivity, and tightness, requiring high resolution for prediction results. Furthermore, the reservoirs are geographically dispersed, with blurred boundaries, and their seismic reflections suffer from poor imaging quality due to weak energy at long offsets, making range characterization extremely difficult. In recent years, thanks to the continuous efforts of many scholars, seismic post-stack inversion technology has gradually matured and is currently the most commonly used reservoir prediction method. Based on their different methodologies, inversion methods can be categorized into three main types: direct inversion, seismic attribute inversion, and model-based inversion. Direct inversion and seismic attribute inversion, being directly derived from seismic data, lack low-frequency components in their results. Their resolution, signal-to-noise ratio, and reliability all depend on the seismic data itself, making them difficult to use for identifying thin and small reservoirs. Model-based inversion methods, utilizing well logging and structural information to build an initial model, obtain inversion results through continuous forward modeling and iterative comparisons. While this avoids the limitations of direct seismic data inversion and can overcome the limitations of seismic resolution to obtain higher-resolution inversion results, the use of conventional well interpolation modeling methods to construct the initial model still results in problems such as excessive modeling and inconsistencies with actual geological structural characteristics. Therefore, more reasonable and applicable inversion models are needed to optimize the inversion effect. Meanwhile, in recent years, pre-stack inversion methods using elastic parameters have become a hot topic in oil and gas exploration research. Compared to post-stack seismic inversion, using pre-stack data offers several advantages, such as: higher fidelity; richer information content; and higher accuracy in reservoir description, making it more suitable for the current needs of exploring and developing complex lithological oil and gas reservoirs. However, commonly used angle-domain pre-stack inversion techniques require the use of velocity models to convert the initial offset gathers into angle-domain gathers. This process demands high accuracy from the overlying strata velocity information, inevitably introducing certain errors.

[0003] Chinese patent application CN201811541667.7 discloses a method for analyzing the spatial structure of sedimentary body waveforms and performing volume-controlled inversion modeling. The method comprises the following steps: S1, acquiring seismic data volumes and extracting waveform structural feature attributes from the seismic data volumes to obtain attribute data volumes characterizing the waveform structural features of the seismic data volumes; S2, performing cluster analysis on the attribute data volumes corresponding to the waveform structural features of the seismic data volumes using a pattern feature-based clustering method to form a three-dimensional seismic facies body; S3, using the three-dimensional seismic facies body to control deep learning-based sedimentary body lithofacies identification and guide facies-controlled inversion modeling under sedimentary body control. In comparison, this invention establishes an optimal frequency inversion model based on a frequency fusion modeling method, which can highlight advantageous frequency bands according to the prediction requirements such as the thickness and scale of the target reservoir, and the resolution of the inversion results is more conducive to the identification of target geological bodies.

[0004] Chinese patent application CN201910467123.9 discloses a pre-stack seismic inversion method and apparatus. The method includes: acquiring pre-stack seismic data of the target reservoir; modeling to determine a prior model and an initial inversion model as the current inversion iteration model; determining the energy error based on the seismic angle gathers of the target reservoir and the current inversion iteration model; if the energy error is not less than a preset energy error, determining an updated inversion iteration model based on the prior model of the pre-stack seismic data and the current inversion iteration model; using the updated inversion iteration model as the current inversion iteration model; iteratively determining the energy error; if the energy error is less than a preset energy error, using the current inversion iteration model with the energy error less than the preset energy error as the inversion result of the pre-stack seismic data. Compared to this, the present invention performs inversion based on wide-azimuth five-dimensional seismic data, which can fully exploit multi-dimensional seismic response information and improve the prediction accuracy of complex reservoirs. Furthermore, addressing the reliance on accurate overlying strata velocity information in existing angle-domain pre-stack inversion methods, the wide-azimuth seismic data is converted into P-value domain ray gathers for inversion, avoiding velocity model errors.

[0005] Chinese patent application CN202210625625.1 discloses a method and apparatus for pre-stack facies inversion of reservoirs. The method includes: acquiring basic data; establishing a stratigraphic morphology model based on the basic data; acquiring sedimentary plane facies characteristics; establishing the relationship between the sedimentary plane facies characteristics and each elastic parameter; establishing a three-dimensional spatial model based on the relationship between the sedimentary plane facies characteristics and each elastic parameter, and the stratigraphic morphology model; the basic data includes pre-stack seismic data, which is divided according to the incident angle to obtain pre-stack seismic data at different angles; wavelets of different angles are obtained based on the pre-stack seismic data at different angles; the relationship between the pre-stack seismic data at different angles, the wavelets of different angles, the sedimentary plane facies characteristics, and each elastic parameter is used as input to the stratigraphic morphology model, and an elastic parameter data volume is obtained through inversion; the elastic parameter data volume is intersected, and the reservoir is depicted by combining the results of single-well reservoir analysis. Compared to this method, the present invention differs significantly in its disclosed technical features, implementation method, and application field. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a pre-stack inversion method for elastic parameters based on frequency fusion modeling to overcome or at least partially solve the above problems.

[0007] According to one aspect of the present invention, a pre-stack inversion method for elastic parameters based on frequency fusion modeling is provided, the inversion method comprising:

[0008] Construct an optimal frequency initial model that integrates multi-scale characteristics based on the target reservoir prediction requirements;

[0009] Frequency division processing was performed on a portion of the superimposed data volume using variable time-window wavelet transform to obtain frequency-divided seismic data volume. Low-frequency, medium-frequency, and high-frequency models were established using various model building methods.

[0010] Based on the target reservoir prediction requirements, low-frequency, medium-frequency, and high-frequency models are fused into an optimal frequency model through a linear hybrid model.

[0011] An optimal frequency model-constrained P-impedance inversion objective function is constructed for inversion. The elastic parameters are solved using a stochastic direction search method constrained by well logging prior information, and this solution guides reservoir prediction.

[0012] Optionally, the step of constructing an optimal frequency initial model that integrates multi-scale features based on the target reservoir prediction requirements specifically includes:

[0013] A model building method based on frequency fusion modeling is adopted to construct an optimal initial frequency model that comprehensively considers multi-scale characteristics according to the target reservoir prediction requirements.

[0014] Optionally, the multi-scale features specifically include: geology, well logging, and seismic data.

[0015] Optionally, the step of using variable time-window wavelet transform to perform frequency division processing on the partially stacked data volume to obtain frequency-divided seismic data volume, and establishing low-frequency, mid-frequency, and high-frequency models using multiple model building methods, specifically includes:

[0016] Optimize and process wide-azimuth seismic data;

[0017] Based on the characteristics of actual seismic data, the optimized wide-azimuth seismic data is subjected to P-value transformation to obtain P-value domain ray gathers;

[0018] Based on the P-value range of the actual seismic data, multiple partial data volumes with different P-value ranges are stacked as needed to form seismic data with a central P-value. The stacking formula is as follows:

[0019]

[0020] Where S represents a single-channel seismic record, P i Let P be the P-value parameter for the i-th channel. m P n The cutoff range for the superposition;

[0021] Spectral analysis was performed on the P-value stacked seismic data to determine the seismic bandwidth. Based on the thickness and scale characteristics of the reservoir in the target stratigraphic section of the work area, the optimal frequency was determined, and the frequency ranges were divided into a mid-frequency band (f1–f2), a low-frequency band (0–f1), and a high-frequency band (f2–f2). h ;

[0022] Based on the geological target, a wavelet function is selected or constructed, and the corresponding frequency-division seismic data volume is obtained by using the adaptive variable time window wavelet transform method.

[0023] Optionally, the optimization processing of wide-azimuth seismic data specifically includes: predicting and suppressing random noise through seismic data denoising processing; and processing the actual seismic data using an inverse Q-filtering method to compensate for amplitude attenuation and frequency loss.

[0024] Optionally, the wavelet function specifically includes: Morlet, Marr, and Gabor wavelets.

[0025] Optionally, the step of establishing low-frequency, mid-frequency, and high-frequency models using multiple model building methods specifically includes:

[0026] By comprehensively utilizing well logging and frequency-division seismic data, three frequency models of low, medium, and high frequencies were constructed respectively.

[0027] Based on the linear hybrid model fusion method, the optimal frequency inversion model is obtained by assigning weight constraints to the advantageous frequency bands of the target reservoir type and fusing them together.

[0028] Model fusion is performed in the frequency domain, based on the frequency fusion modeling formula of the improved linear mixing model:

[0029]

[0030] Where, α i A is an adjustable weighting parameter. m M is obtained by quantifying the characteristics of the target body, such as geological body type and thickness. f For the optimal frequency model, M i Partially dominant frequency model, where ε represents noise.

[0031] Optionally, the inversion of the P-impedance inversion objective function constrained by the optimal frequency model specifically includes:

[0032] Construct an objective function for elastic impedance inversion with P-value constraints under the optimization model, and use an iterative algorithm to obtain elastic impedance inversion volumes with different P-values.

[0033] Formula for elastic impedance of P value:

[0034]

[0035] Where: v p v s Here, ρ represents the longitudinal wave velocity, p represents the transverse wave velocity, ρ represents the density, and p represents the ray parameter.

[0036] REI(p) is the extension of the wave impedance in the P-domain. When the ray parameter is zero, REI is consistent with AI.

[0037] REI(0)=ρv p =AI

[0038] Referring to the method for determining the reflection coefficient based on wave impedance, the expression for the P-value reflection coefficient is:

[0039]

[0040] The convolution model is expressed as: S(p)=R(p)*W(p) where S(p) is the P-value seismic trace, R(p) is the P-value reflection coefficient, and W(p) is the P-value wavelet;

[0041] By establishing a correlation between the P-value impedance and seismic data, an objective function for the P-value impedance constrained by the optimization model is constructed, in the following form:

[0042] G i (n)=||S0-S i || P +α||REIM i -REIM0|| P

[0043] Where S0 represents the actual earthquake record; S i REIM is a synthetic seismic record calculated using a perturbed model. i REIM0 is the P-value impedance model after perturbation; REIM0 is the optimal initial P-value impedance model; α is the weighting coefficient of the optimal initial model constraint, which is selected according to actual needs.

[0044] The objective function includes the error between the synthesized seismic record and the actual seismic record in each iteration, and the inversion process is constrained by the optimal frequency initial model;

[0045] By continuously perturbing the model to minimize the objective function, the inversion results obtained under the constraint of the optimal frequency model have vertical and horizontal resolution characteristics guided by the identification of target geological bodies.

[0046] Optionally, the stochastic direction search method constrained by well logging prior information, which solves for elastic parameters and guides reservoir prediction, specifically includes:

[0047] The inverted elastic impedance inversion bodies of different ranges of P-value are used to solve for reservoir-related elastic parameters based on the P-value elastic impedance formula and a random direction search algorithm constrained by well logging information.

[0048] By using well logging statistics to constrain the initial value selection and search space of the random direction search method, the elastic parameters of the reservoir are solved.

[0049] Optionally, the random direction search method iterates and gradually optimizes the solution until a least-squares solution with relatively small error is found, thus finding the optimal solution and enabling the elastic parameters to obtain inversion results while ensuring their physical meaning.

[0050] The objective function for solving the elastic parameters is:

[0051]

[0052] In the formula: ρ is density; v p V represents the longitudinal wave velocity; s The transverse wave velocity is represented by REI, and the elastic impedance is represented by the P-value.

[0053] Optionally, the random search algorithm can select the direction that causes the objective function to decrease the fastest from several random directions as a feasible search direction. The random search formula for well logging information constraints is:

[0054]

[0055] Where, ρ j VP j vs jLet e ​​be a random value within the well logging statistics range, and select the feasible direction that converges fastest among the k search directions as the forward direction. j min The process generates input, iterates repeatedly until L approaches its minimum value, and then outputs the optimal solution.

[0056] This invention provides a pre-stack inversion method for elastic parameters based on frequency fusion modeling. The inversion method includes: constructing an optimal initial frequency model integrating multi-scale characteristics according to the target reservoir prediction requirements; performing frequency division processing on a portion of the stacked data volume using variable time-window wavelet transform to obtain frequency-divided seismic data volumes; establishing low-frequency, mid-frequency, and high-frequency models using multiple model construction methods; fusing the low-frequency, mid-frequency, and high-frequency models into an optimal frequency model through a linear hybrid model according to the target reservoir prediction requirements; constructing a P-impedance inversion objective function constrained by the optimal frequency model to perform the inversion; solving for elastic parameters using a stochastic direction search method constrained by well logging prior information; and guiding reservoir prediction. The optimal initial frequency model not only has high resolution in both vertical and horizontal directions but also effectively characterizes the sand body distribution morphology, thereby improving the objectivity and accuracy of the inversion results. This invention is used for the inversion of elastic parameters related to lithology, connectivity, and hydrocarbon content of channel sand bodies in the comprehensive evaluation of oil exploration and compressed air storage traps, laying the foundation for predicting the vertical and horizontal boundaries, thickness, and spatial volume of sand bodies.

[0057] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart illustrating a pre-stack inversion method for elastic parameters based on frequency fusion modeling, provided for an embodiment of the present invention;

[0060] Figure 2 This is a specific embodiment of the invention where the P-value portion is superimposed on a gather;

[0061] Figure 3 This is a specific embodiment of the present invention involving seismic data frequency band analysis;

[0062] Figure 4This is an initial model for dominant frequency fusion in a specific embodiment of the present invention;

[0063] Figure 5 This is the impedance inversion result of the P-value in a specific embodiment of the present invention;

[0064] Figure 6 This is the result of elastic parameter inversion in a specific embodiment of the present invention. Detailed Implementation

[0065] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0066] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0067] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0068] Example 1:

[0069] like Figure 1 As shown, this invention provides a pre-stack inversion method for elastic parameters based on frequency fusion modeling. The method specifically includes the following steps:

[0070] Step 1: Optimize the wide-azimuth seismic data to improve the signal-to-noise ratio and resolution;

[0071] In this embodiment, the existing seismic data has a low dominant frequency and narrow bandwidth. Before inversion, it needs to be optimized to improve the resolution of the target layer, providing a foundation for subsequent prediction of thin sandstone reservoirs. Frequency extension processing based on compressed sensing is performed on the target data. Before frequency extension, the dominant frequency of the seismic data was 32Hz, and the bandwidth was 5–58Hz. After frequency extension, the dominant frequency was increased to 38Hz, and the bandwidth was 8–78Hz. Furthermore, the reflected signal of the seismic data was enhanced, effectively improving the resolution of the seismic data.

[0072] Step 2: Based on the characteristics of the seismic data in the embodiment, the optimized wide-azimuth seismic data is first converted to obtain P-value gathers. In this embodiment, the distribution range of P-values ​​is 0-0.17. These are then divided into three intervals as needed: 0.02-0.07, 0.07-0.12, and 0.12-0.17. These intervals are then overlaid to obtain three overlaid data volumes. Figure 2This is a three-part superimposed seismic data volume in a specific embodiment 1 of the present invention.

[0073] Step 3: Conduct spectral analysis on the partially stacked seismic data volumes, and determine the optimal frequency for reservoir identification by combining the characteristics of the target layer reservoir in the work area of ​​Example 1, and further divide the frequency band range of low, medium and high frequency models. Figure 3 This is the frequency band analysis result of seismic data for the target layer in a specific embodiment of the present invention. It can be seen that after optimization, the dominant frequency of the target layer in this embodiment is 38Hz, and the bandwidth is 8–78Hz, which is significantly improved compared to the original data. Combined with the predicted sandstone reservoir characteristics, the optimal frequency range for reservoir identification is clearly high frequency. The frequency division scheme in this embodiment is determined as follows: low frequency range 0–10Hz, mid frequency range 10–78Hz, and high frequency range 78–350Hz. Then, the corresponding frequency-divided seismic data volume is obtained using the adaptive variable-time-window wavelet transform method.

[0074] Step 4: Using the aforementioned frequency-divided seismic data, perform well interpolation with low-frequency seismic constraints, attribute inversion of mid-frequency seismic data, and phased-array waveform similarity interpolation of high-frequency seismic data to obtain low-, mid-, and high-frequency models. By assigning greater weight to the high-frequency model and highlighting its high-frequency characteristics, frequency fusion is performed to obtain the optimal initial frequency model. Figure 4 This is an optimal frequency initial inversion model in a specific embodiment of the present invention. In the low-frequency model, the vertical resolution of sand bodies is low, but the lateral variation is natural; in the high-frequency model, the vertical resolution of sand bodies is high, but some chaotic outliers exist in the lateral direction, causing interference. The optimal frequency model obtained by frequency fusion modeling has full-band information characteristics and better characterizes the morphological distribution of geological bodies. Because the high-frequency components are highlighted through weight adjustment during the fusion process, the model exhibits high vertical resolution, which is beneficial for the accurate identification of thin and small reservoirs. Compared with conventional models, the optimal frequency model shows more balanced seismic energy at different depths, better alignment of stratigraphic morphology and seismic phase axis, and better characterization of sand body boundary features in some detailed locations.

[0075] Step 5: Using the optimal frequency model-constrained P-value elastic impedance inversion method, obtain the three superimposed P-value impedance inversion bodies respectively; Figure 5 The figures show the impedance inversion results for P-values ​​of 0.045, 0.095, and 0.145 in a specific embodiment of the present invention. The three inversion results are constrained by the optimal frequency model. It can be seen that the basic shape of the impedances with different P-values ​​tends to be consistent, but at the target layer location, there are subtle differences in amplitude intensity, lateral distribution, etc., on the impedance profiles of different P-value ranges. This contains unique velocity, density, and other characteristic information of the superimposed data from different P-value ranges. Deeply mining these differences provides a foundation for solving reservoir parameters.

[0076] Step 6: Using the inverted impedance inversion bodies with different P-value ranges obtained from the inversion, and based on the P-value elastic impedance formula, solve for the elastic parameters using a random direction search algorithm constrained by well logging information. Figure 6 This is the result of the vp / vs elastic parameter inversion in a specific embodiment of the present invention. Under the constraint of well logging statistics, the value range of the inversion result is closer to the actual well data, and has higher reliability and accuracy. The process ends with using the vp / vs elastic parameter inversion results to guide the prediction of channel sandstone reservoirs.

[0077] Example 2

[0078] In response to the current need for lithological exploration and high-precision reservoir prediction, this invention aims to optimize the inversion effect by establishing a more reasonable and applicable inversion model based on the frequency fusion modeling method and carrying out pre-stack inversion in the P-value domain, thereby improving the vertical and horizontal resolution and accuracy of the inversion results.

[0079] This invention utilizes wide-azimuth five-dimensional seismic data for inversion, fully extracting multi-dimensional seismic response information and improving the prediction accuracy of complex reservoirs. Furthermore, addressing the limitation of existing angle-domain pre-stack inversion methods that rely on overlying strata velocity information, this invention converts wide-azimuth seismic data into P-domain ray gathers for inversion, thus avoiding velocity model errors.

[0080] To address the shortcomings of existing modeling methods, a frequency fusion modeling approach was adopted. Based on the target reservoir prediction requirements, an optimal initial frequency model was constructed, comprehensively considering multi-scale characteristics from geology, well logging, and seismic data. Wavelet function applicability optimization was performed, and frequency division was applied to a portion of the stacked data volume using variable-time-window wavelet transform to obtain frequency-divided seismic data volumes. Low-frequency, mid-frequency, and high-frequency models were established using different model construction methods. Based on the prediction requirements for the target reservoir's thickness and size, the dominant frequency band was identified. The low, mid, and high-frequency models were then fused into an optimal frequency model using an improved linear hybrid model. The optimal frequency fusion model results better match the sand body's occurrence and structural characteristics, maintain lateral variation consistent with seismic waveforms, and offer higher resolution for target geological body identification. Then, an inversion objective function constrained by the optimal frequency model was constructed for inversion. Finally, a stochastic direction search method constrained by well logging prior information was used to solve for lithological and hydrocarbon-bearing elastic parameters, guiding reservoir prediction.

[0081] The method of the present invention includes the following steps:

[0082] 1. Optimize the processing of wide-azimuth seismic data to improve the signal-to-noise ratio and resolution; for example, improve the signal-to-noise ratio by predicting and suppressing random noise through seismic data denoising; use inverse Q filtering to process actual seismic data to compensate for amplitude attenuation and frequency loss, etc.

[0083] 2. Based on the characteristics of actual seismic data, the optimized wide-azimuth seismic data undergoes P-value transformation to obtain P-value domain ray gathers. According to the P-value range of the actual seismic data, multiple partial stacked data volumes with different P-value ranges are formed as the seismic data with the central P-value. The stacking formula is as follows:

[0084] Where S represents a single-channel seismic record, P i Let P be the P-value parameter for the i-th channel. m P n The cutoff range for the superposition;

[0085] 3. Conduct spectral analysis on the P-value superimposed seismic data volume to determine the seismic bandwidth. Based on the characteristics of the target reservoir layer in the work area, such as thickness and scale, determine the optimal frequency and further divide it into mid-frequency range (f1~f2), low-frequency range (0~f1), and high-frequency range (f2~ ... h Based on different geological targets, appropriate wavelet functions are selected or constructed, such as Morlet, Marr, and Gabor wavelets, and the corresponding frequency-division seismic data volume is obtained using the adaptive variable time window wavelet transform method.

[0086] 4. A three-part frequency model (low, medium, and high) was constructed by comprehensively utilizing well logging and frequency-division seismic data. Based on a linear hybrid model fusion method, the advantageous frequency bands of the model were assigned weights according to the target reservoir type, and the optimal frequency inversion model was obtained through fusion. This model not only has high resolution in both vertical and horizontal directions but also effectively depicts the sand body distribution morphology, ensuring the objectivity and accuracy of the inversion results.

[0087] Model fusion is performed in the frequency domain, based on the frequency fusion modeling formula of the improved linear mixing model:

[0088]

[0089] Where α i A is an adjustable weighting parameter. m M is obtained by quantifying the characteristics of the target body, such as geological body type and thickness. f For the optimal frequency model, M i Partially dominant frequency model, where ε represents noise.

[0090] The optimal frequency model utilizes low-frequency information from wells for interpolation in its low-frequency components to compensate for the low-frequency information in the fused model, accurately reflecting the absolute value of impedance. The mid-frequency components are obtained by bandpass filtering of mid-frequency seismic attribute inversion results. These mid-frequency components enable the inversion model to characterize the morphological features of geological sedimentary bodies, improving the consistency between the lateral distribution of the inversion results and the seismic record. The high-frequency components are obtained by extrapolating high-frequency seismic constrained well data driven by the sensitive waveform characteristics of well logging in the well-side traces, offering high-resolution advantages. Waveform matching is performed, considering both waveform similarity and distance from known wells, with the specific weighting: w = βw. c +(1-β)w d , where: w c For waveform similarity, w d It is a function of distance, and β is the waveform influence coefficient.

[0091] 5. Construct the objective function for elastic impedance inversion with P-value constraints of the optimization model, and use an iterative algorithm to obtain the elastic impedance inversion volume for different P-values; the formula for elastic impedance with P-value is as follows:

[0092]

[0093] Where: v p v s Let ρ be the longitudinal wave velocity, p be the transverse wave velocity, ρ be the density, and p be the ray parameter.

[0094] REI(p) is the extension of the wave impedance in the P-domain. When the ray parameter is zero, REI is consistent with AI, that is...

[0095] REI(0)=ρv p =AI

[0096] Referring to the method for determining the reflection coefficient based on wave impedance, the expression for the P-value reflection coefficient is:

[0097]

[0098] The convolution model can be expressed as: S(p)=R(p)*W(p) where S(p) is the P-value seismic trace, R(p) is the P-value reflection coefficient, and W(p) is the P-value wavelet.

[0099] By establishing a relationship between the P-value impedance and seismic data, and then constructing the objective function of the P-value impedance constrained by the optimization model, the specific form is as follows:

[0100] G i (n)=||S0-S i || P +α||REIM i -REIM0|| P

[0101] Where S0 represents the actual earthquake record; S i REIM is a synthetic seismic record calculated using a perturbed model. i REIM0 represents the P-value impedance model after perturbation; REIM0 represents the optimal initial P-value impedance model. α is the weighting coefficient of the optimal initial model constraint, and its value is selected according to actual needs.

[0102] The objective function includes both the error between the synthesized seismic record and the actual seismic record in each iteration, and the inversion process is constrained by the optimal frequency initial model. By continuously perturbing the model to minimize the objective function, the inversion results under the constraint of the optimal frequency model have vertical and horizontal resolution characteristics guided by the identification of target geological bodies.

[0103] 6. The inverted elastic impedance inversion bodies of different ranges of P-value are used to solve for reservoir-related elastic parameters based on the P-value elastic impedance formula and a random direction search algorithm constrained by well logging information.

[0104] By using well logging statistics to constrain the initial value selection and search space of the stochastic direction search method, the elastic parameters of the reservoir can be solved. Well logging statistics can greatly reduce the search space and improve the solution efficiency.

[0105] The method of this invention iterates and optimizes the solution step by step until a least-squares solution with relatively small error is obtained, thus finding the optimal solution. This allows the elastic parameters to easily obtain stable and reliable solutions while ensuring their physical meaning, thereby ensuring the authenticity and accuracy of the inversion results.

[0106] The objective function for solving the elastic parameters is:

[0107]

[0108] In the formula: ρ is density; v p V represents the longitudinal wave velocity; s is the transverse wave velocity; REI is the P-value elastic impedance.

[0109] The random search algorithm can select the direction that causes the objective function to decrease the fastest from several random directions as a feasible search direction. The random search formula constrained by well logging information is as follows:

[0110]

[0111] Where, ρ j VP j vs j Let e ​​be a random value within the well logging statistics range, and select the feasible direction that converges fastest among the k search directions as the forward direction. j minThe process generates input, iterates repeatedly until L approaches its minimum value, and then outputs the optimal solution.

[0112] Beneficial Effects: The P-value gathers of this invention can be directly calculated from seismic data without relying on overlying velocity information, thus improving inversion accuracy. The optimal frequency model obtained based on the frequency fusion modeling method can comprehensively consider multi-scale characteristics such as geology, well logging, and seismic data. Based on the thickness and scale of the target reservoir, the optimal model frequency is determined, improving the accuracy of thin sand body identification. Inversion is carried out using the P-impedance inversion objective function constrained by the dominant frequency model. A stochastic direction search method constrained by well logging prior information is used to solve for lithological and hydrocarbon-bearing elastic parameters, which is faster than ordinary methods and reduces the possibility of local optima. The inversion results obtained using this technical solution have relatively higher vertical and horizontal resolution and better objectivity, effectively solving the prediction challenges of reservoirs with thin vertical thickness, rapid horizontal changes, and strong heterogeneity, achieving accurate prediction of thin channel sand body reservoirs.

[0113] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A pre-stack inversion method for elastic parameters based on frequency fusion modeling, characterized in that, The inversion method includes: Construct an optimal frequency initial model that integrates multi-scale characteristics based on the target reservoir prediction requirements; Frequency division processing was performed on a portion of the superimposed data volume using variable time-window wavelet transform to obtain frequency-divided seismic data volume. Low-frequency, medium-frequency, and high-frequency models were established using various model building methods. Based on the target reservoir prediction requirements, low-frequency, medium-frequency, and high-frequency models are fused into an optimal frequency model through a linear hybrid model. An optimal frequency model-constrained P-impedance inversion objective function is constructed for inversion. The elastic parameters are solved using a stochastic direction search method constrained by well logging prior information, and this solution guides reservoir prediction.

2. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 1, characterized in that, The construction of the optimal frequency initial model integrating multi-scale features based on the target reservoir prediction requirements specifically includes: A model building method based on frequency fusion modeling is adopted to construct an optimal initial frequency model that comprehensively considers multi-scale characteristics according to the target reservoir prediction requirements.

3. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 2, characterized in that, The multi-scale features specifically include: geology, well logging, and seismic data.

4. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 1, characterized in that, The step of using variable-time-window wavelet transform to perform frequency division processing on partially stacked data volumes to obtain frequency-divided seismic data volumes, and establishing low-frequency, mid-frequency, and high-frequency models using multiple model building methods, specifically includes: Optimize and process wide-azimuth seismic data; Based on the characteristics of actual seismic data, the optimized wide-azimuth seismic data is subjected to P-value transformation to obtain P-value domain ray gathers; Based on the P-value range of the actual seismic data, multiple partial data volumes with different P-value ranges are stacked as needed to form seismic data with a central P-value. The stacking formula is as follows: Where S represents a single-channel seismic record, P i Let P be the P-value parameter for the i-th channel. m P n The cutoff range for the superposition; Spectral analysis was performed on the P-value stacked seismic data to determine the seismic bandwidth. Based on the thickness and scale characteristics of the reservoir in the target stratigraphic section of the work area, the optimal frequency was determined, and the frequency ranges were divided into a mid-frequency band (f1–f2), a low-frequency band (0–f1), and a high-frequency band (f2–f2). h ; Based on the geological target, a wavelet function is selected or constructed, and the corresponding frequency-division seismic data volume is obtained by using the adaptive variable time window wavelet transform method.

5. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 4, characterized in that, The optimization processing of wide-azimuth seismic data specifically includes: predicting and suppressing random noise through seismic data denoising; and using the inverse Q-filter method to process the actual seismic data to compensate for amplitude attenuation and frequency loss.

6. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 4, characterized in that, The wavelet functions specifically include: Morlet, Marr, and Gabor wavelets.

7. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 1, characterized in that, The specific methods used to establish low-frequency, mid-frequency, and high-frequency models include: By comprehensively utilizing well logging and frequency-division seismic data, three frequency models of low, medium, and high frequencies were constructed respectively. Based on the linear hybrid model fusion method, the optimal frequency inversion model is obtained by assigning weight constraints to the advantageous frequency bands of the target reservoir type and fusing them together. Model fusion is performed in the frequency domain, based on the frequency fusion modeling formula of the improved linear mixing model: Where, α i A is an adjustable weighting parameter. m M is obtained by quantifying the characteristics of the target body, such as geological body type and thickness. f For the optimal frequency model, M i Partial dominant frequency model, where ε is noise.

8. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 1, characterized in that, The inversion of the P-impedance inversion objective function constrained by the optimal frequency model specifically includes: Construct an objective function for elastic impedance inversion with P-value constraints under the optimization model, and use an iterative algorithm to obtain elastic impedance inversion volumes with different P-values. Formula for elastic impedance of P value: Where: v p v s Here, ρ represents the longitudinal wave velocity, p represents the transverse wave velocity, ρ represents the density, and p represents the ray parameter. REI(p) is the extension of the wave impedance in the P-domain. When the ray parameter is zero, REI is consistent with AI. REI(0)=ρv p =AI Referring to the method for determining the reflection coefficient based on wave impedance, the expression for the P-value reflection coefficient is: The convolution model is expressed as: S(p)=R(p)*W(p) where S(p) is the P-value seismic trace, R(p) is the P-value reflection coefficient, and W(p) is the P-value wavelet; By establishing a correlation between the P-value impedance and seismic data, an objective function for the P-value impedance constrained by the optimization model is constructed, in the following form: G i (n)=||S0-S i || P +αREIM i -REIM0|| P Where S0 represents the actual earthquake record; S i REIM is a synthetic seismic record calculated using a perturbed model. i REIM0 is the P-value impedance model after perturbation; REIM0 is the optimal initial P-value impedance model; α is the weighting coefficient of the optimal initial model constraint, which is selected according to actual needs. The objective function includes the error between the synthesized seismic record and the actual seismic record in each iteration, and the inversion process is constrained by the optimal frequency initial model; By continuously perturbing the model to minimize the objective function, the inversion results obtained under the constraint of the optimal frequency model have vertical and horizontal resolution characteristics guided by the identification of target geological bodies.

9. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 1, characterized in that, The stochastic direction search method constrained by well logging prior information, which solves for elastic parameters and guides reservoir prediction, specifically includes: The inverted elastic impedance inversion bodies of different ranges of P-value are used to solve for reservoir-related elastic parameters based on the P-value elastic impedance formula and a random direction search algorithm constrained by well logging information. By using well logging statistics to constrain the initial value selection and search space of the random direction search method, the elastic parameters of the reservoir are solved.

10. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 9, characterized in that, The random direction search method iterates continuously to gradually optimize the solution until a least-squares solution with relatively small error is found, thus finding the optimal solution. This allows the elastic parameters to obtain inversion results while ensuring their physical meaning. The objective function for solving the elastic parameters is: In the formula: ρ is density; v p V represents the longitudinal wave velocity; s is the transverse wave velocity; REI is the P-value elastic impedance.

11. The pre-stack inversion method for elastic parameters based on frequency fusion modeling according to claim 9, characterized in that, The random search algorithm can select the direction that causes the objective function to decrease the fastest from several random directions as a feasible search direction. The random search formula constrained by well logging information is as follows: Where, ρ j VP j vs j Let each be a random value within the well logging statistics range, and select the feasible direction that converges fastest among the k search directions as the forward direction e. j min The process generates input, iterates repeatedly until L approaches its minimum value, and then outputs the optimal solution.

Images