A set smoothing time-lapse seismic difference inversion method based on probability pattern conversion

Abstract

The application discloses a set smoothing time-lapse seismic difference inversion method based on probability mode conversion. The method comprises the following steps: acquiring time-lapse seismic data and logging data; constructing a linear forward model between difference seismic data and relative wave impedance difference parameters based on a convolution model; extracting a logging label reflecting real underground wave impedance difference, calculating a cumulative probability distribution and a functional inverse function thereof based on a probability integral transformation, and constructing a probability mode conversion operator; in a set smoothing framework, updating a model set of a multivariate Gaussian distribution by using a Kalman gain, and mapping the updated model to a model set conforming to a logging real distribution through the conversion operator; and iteratively performing the above process to obtain a posterior inversion result of the relative wave impedance difference. The application can directly invert a relative difference parameter, and effectively improves the inversion precision of a time-lapse seismic difference parameter with a non-Gaussian distribution characteristic and the reliability of reservoir dynamic characterization.

Description

Technical Field

[0001] This invention belongs to the field of exploration geophysics and oil and gas field dynamic monitoring technology, and particularly relates to a statistical inversion method for time-shifted seismic difference parameters based on probability mode conversion. Background Technology

[0002] Time-shifted seismic inversion is an important geophysical monitoring method that characterizes the evolution of subsurface media properties and fluid states over time by comparing seismic responses at different periods. It provides crucial evidence for identifying reservoir dynamics and quantitatively representing remaining hydrocarbons. In the field of time-shifted seismic inversion, existing methods are mostly based on a Bayesian framework, estimating differences in subsurface elastic or physical parameters by constructing an objective function. Among these, ensemble-based stochastic inversion methods, while maintaining high computational efficiency, can update the model ensemble by assimilating observational data, thereby obtaining the posterior probability distribution of parameters. Currently, ensemble methods, such as ensemble smoothing, are widely used in elastic parameter inversion, rock physics inversion, and full-waveform inversion.

[0003] However, existing ensemble Kalman inversion methods have significant limitations in practical applications. The core issue lies in the fact that these methods rely on statistical quantities such as parameter covariance when updating the model, thus typically depending strictly on multivariate Gaussian prior assumptions. For time-lapse seismic differential inversion, parameter changes caused by subsurface reservoir dynamics often exhibit a sparse distribution across the entire region, while showing significant clustering characteristics in local areas, with highly complex correlation structures between different differential parameters. These parameter changes caused by reservoir dynamics generally do not conform to Gaussian distributions, often exhibiting non-Gaussian skewed distributions. Using a single multivariate Gaussian model to construct the prior distribution fails to accurately reflect the spatial statistical characteristics of the differential parameters, making it susceptible to influence and bias when updating model parameters using an ensemble framework. This leads to artifacts in non-reservoir areas, severely limiting the inversion accuracy of time-lapse seismic differential parameters.

[0004] Furthermore, traditional time-lapse seismic inversion requires prior knowledge of the baseline impedance values ​​from the first period when calculating the absolute impedance difference. This means that a conventional independent inversion of the baseline seismic data is usually necessary to construct an initial model before conducting time-lapse difference inversion. However, for areas with few wells and complex geological conditions, establishing an accurate initial model for inversion is extremely difficult. Under such conditions, the inversion accuracy of single-period baseline seismic data is inherently low, and the errors in this baseline model are directly propagated and amplified, severely impacting the accuracy and reliability of subsequent time-lapse seismic difference inversion. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention proposes an ensemble smoothing time-shifted seismic differential inversion method based on probability mode transformation. This method enables direct inversion of elastic parameters that conform to the actual non-Gaussian distribution characteristics, improving the accuracy and geological rationality of time-shifted seismic differential inversion, and avoiding excessive reliance on the basic absolute wave impedance model.

[0006] To achieve the above objectives, this invention provides an ensemble smoothing time-shifted seismic difference inversion method based on probabilistic mode transformation, comprising the following steps:

[0007] (1) Acquire time-lapse seismic data and well logging data of the underground area;

[0008] (2) Based on the time-shifted seismic difference forward model, establish a linear forward modeling relationship between the observed seismic data with difference and the parameters of the relative wave impedance difference model;

[0009] (3) Extract wave impedance difference label data from the well logging data and construct a probabilistic mode conversion operator; use stochastic simulation method to generate an initial model parameter set that follows a multivariate Gaussian distribution, and generate a corresponding set of seismic records based on the linear forward modeling relationship;

[0010] (4) Subtract the time-shifted seismic data to obtain the actual observation difference seismic data, and combine the actual observation difference seismic data, the seismic record set, and the probability mode conversion operator to form a joint iterative update framework;

[0011] (5) Based on the set smoothing theory, the Kalman gain is used to iteratively update the model parameter set, and the probability mode conversion operator is applied to perform distribution mapping in each update to obtain the inversion result of the relative difference of the posterior wave impedance.

[0012] In step (2) above, based on the time-shifted seismic difference forward model, a linear forward modeling relationship is established between the observed differential seismic data and the parameters of the relative acoustic impedance difference model, including the following steps:

[0013] (1) Represent a single seismic record as the convolution of a seismic wavelet and a reflection coefficient, and use Taylor expansion to approximate the reflection coefficient as a linear relationship with the logarithm of wave impedance;

[0014] (2) Establish forward modeling equations for basic and monitoring seismic data respectively, and calculate the difference between the two periods of seismic data to eliminate the static consistency background;

[0015] (3) Define new model parameters that include the characteristics of relative wave impedance differences, and construct a direct linear forward modeling equation between the observed differential seismic data and the model parameters; after solving the model parameters by inversion, use the exponential relationship to directly calculate the relative difference value of wave impedance, without the need for the absolute value of the basic wave impedance to participate in this process.

[0016] Step (3) above involves extracting wave impedance difference label data from the well logging data and constructing a probabilistic mode conversion operator, including the following steps:

[0017] (1) Extract data representing the actual changes in underground wave impedance from well logging dynamic curves or known reservoir dynamic information as a tag data set;

[0018] (2) Perform statistical analysis on the tag data set to obtain its true cumulative probability distribution functional and its inverse function;

[0019] (3) Perform statistics on the parameter set of the conventional model that follows a Gaussian distribution to obtain its Gaussian cumulative probability distribution functional;

[0020] (4) Using the quantiles of the probability distribution as a connecting bridge, while keeping the quantiles consistent, a mapping function is defined to transform the Gaussian distribution into the real distribution of the well, forming a probability mode conversion operator.

[0021] Step (4) above uses a stochastic simulation method to generate an initial set of model parameters that follows a multivariate Gaussian distribution, and generates a corresponding set of seismic records based on the linear forward modeling relationship, including the following steps:

[0022] (1) Generate an initial set of model parameters that follows a multivariate Gaussian distribution based on the Gaussian simulation method of geostatistical sequences;

[0023] (2) Using the linear forward modeling relation constructed by the seismic wavelet matrix, the initial model parameter set is synthesized into the corresponding seismic record set.

[0024] In step (5) above, based on set smoothing theory, the Kalman gain is used to iteratively update the model parameter set, including the following steps:

[0025] (1) Calculate the cross-correlation matrix and autocorrelation matrix using the model parameter set that follows a Gaussian distribution and its corresponding synthetic seismic record set;

[0026] (2) Calculate the Kalman gain of the current iteration step by combining the error correlation matrix and the inflation factor;

[0027] Calculate the residuals between observed seismic data and the synthetic seismic record set;

[0028] (3) Using the Kalman gain and the residual, the current Gaussian distribution model parameter set is updated and fitted.

[0029] The above-mentioned application of the probability mode transformation operator to perform distribution mapping in each update to obtain the posterior impedance relative difference inversion result includes the following steps:

[0030] (1) The set of model parameters after Kalman gain adjustment is directly input into the probability mode conversion operator for forced distribution mapping;

[0031] (2) Output a set of model parameters that strictly follow the characteristics of the actual probability distribution in the well, and use them to synthesize the set of real seismic records required for the next iteration;

[0032] (3) The model parameter set with real distribution characteristics is transformed into a Gaussian distribution state again to participate in the next Kalman gain calculation. After repeated iterations and convergence, the average value of the parameter set is calculated to obtain the final wave impedance relative difference inversion result.

[0033] Technical Effects of this Invention: This invention discloses an ensemble-smoothed time-shifted seismic differential inversion method based on probabilistic mode transformation. By deeply integrating a probabilistic distribution mode transformation mechanism into an ensemble-like Kalman inversion framework, it effectively improves the inversion accuracy and geological fidelity of time-shifted differential parameters with non-Gaussian distribution characteristics. The transformation operator constructed by introducing probabilistic integral transformation can act as a distribution filter, ensuring that each model parameter update process is strictly constrained by the real non-Gaussian physical laws of the reservoir. This significantly suppresses the inconsistency errors and false geological anomalies generated in non-reservoir areas by forced smoothing in traditional methods, overcoming the theoretical bottleneck of inversion distortion in complex dynamic monitoring areas by traditional ensemble-smoothed differential inversion methods. Based on the forward modeling strategy of directly reconstructed differential relative parameters, it realizes the relative difference calculation without the participation of a basic high-precision absolute acoustic impedance model, completely avoiding the risk of amplification of differential prediction due to the limited accuracy of the single-period basic model caused by few wells or complex geology. This method not only cleverly maintains the core advantages of high computational efficiency and reduced quantification uncertainty in processing high-dimensional nonlinear problems through multi-data assimilation, but also significantly suppresses and filters noise-induced artifacts in measured data with high background noise through a forced morphological transformation mechanism, demonstrating strong noise robustness and computational stability. In the dynamic monitoring of water injection development in complex, loose sandstone oilfields, this invention can accurately reconstruct the physical property changes of the water injection scour zone continuously distributed along the main channel sand body. The inversion results highly match the long-term dynamic production history of more than ten years, demonstrating significant practical value and promising prospects in engineering fields such as quantitative prediction of remaining oil and gas and monitoring of carbon dioxide geological sequestration. Attached Figure Description

[0034] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0035] Figure 1The diagram shows a real wave impedance model in an embodiment of the present invention, where (a) is the basic wave impedance model; (b) is the monitoring wave impedance model; and (c) is the differential wave impedance model.

[0036] Figure 2 This is a schematic diagram of seismic data synthesized according to an embodiment of the present invention, wherein (a) is basic seismic data; (b) is monitoring seismic data; and (c) is differential seismic data.

[0037] Figure 3 The following is a comparison of the relative wave impedance inversion results of the embodiments of the present invention, wherein (a) is the relative wave impedance inversion result of conventional ensemble smoothing differential inversion; and (b) is the relative wave impedance inversion result of ensemble smoothing differential inversion based on probability mode conversion.

[0038] Figure 4 This is a comparison chart of the single-channel inversion results of the relative wave impedance difference under noise-free conditions in an embodiment of the present invention. The black dashed line represents the actual relative wave impedance difference model, the red solid line represents the inversion result of the conventional ensemble smoothing difference inversion method, and the blue solid line represents the inversion result of the ensemble smoothing difference inversion method based on probability mode conversion of the present invention.

[0039] Figure 5 The following is a comparison of the synthesized noisy differential seismic records and inversion results according to an embodiment of the present invention, wherein (a) is the synthesized noisy differential seismic record; (b) is the inversion result of the conventional ensemble smoothing differential inversion method; and (c) is the inversion result of the ensemble smoothing differential inversion method based on probability mode conversion.

[0040] Figure 6 The image shows a comparison of the single-channel inversion results of the relative wave impedance difference when the signal-to-noise ratio is 7 in an embodiment of the present invention. The black dashed line represents the actual relative wave impedance difference model, the red solid line represents the inversion result of the conventional ensemble smoothing difference inversion method, and the blue solid line represents the inversion result of the ensemble smoothing difference inversion method based on probability mode conversion of the present invention.

[0041] Figure 7 The following is a time-lapse seismic analysis diagram of six wells in an embodiment of the present invention, wherein (a) is a comparison of time-lapse seismic waveforms; (b) is the time-lapse seismic difference value; and (c) is the corresponding relative difference value of wave impedance.

[0042] Figure 8 This is a profile of actual seismic data in an embodiment of the present invention, wherein (a) is basic seismic data; (b) is monitoring seismic data; and (c) is differential seismic data.

[0043] Figure 9The following is a comparison chart of the relative wave impedance difference inversion of actual data in the embodiments of the present invention, wherein (a) is the relative wave impedance difference of the conventional ensemble smoothing difference inversion method; and (b) is the relative wave impedance difference of the ensemble smoothing difference inversion method based on probability mode conversion.

[0044] Figure 10 This is a one-dimensional difference inversion comparison diagram at channel 350 in an embodiment of the present invention;

[0045] Figure 11 This is a comparison diagram of the inversion results of the embodiment of the present invention and the distribution of differential wave impedance labels on the well. Detailed Implementation

[0046] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0048] Example 1

[0049] A method for ensemble smoothing time-shifted seismic difference inversion based on probabilistic mode transformation is provided, comprising the following steps:

[0050] (1) Acquire time-lapse seismic data and well logging data of the underground area;

[0051] (2) Based on the time-shifted seismic difference forward model, establish a linear forward modeling relationship between the observed differential seismic data and the parameters of the relative acoustic impedance difference model; including the following steps:

[0052] a represents a single-channel seismic record as the convolution of a seismic wavelet and a reflection coefficient, and uses Taylor expansion to approximate the reflection coefficient as a linear relationship with the logarithm of wave impedance;

[0053] b. Establish forward modeling equations for basic and monitoring seismic data respectively, and calculate the difference between the two seismic data to eliminate the static consistency background;

[0054] c defines new model parameters that include the characteristics of relative wave impedance differences, and constructs a direct linear forward modeling equation between the observed differential seismic data and these model parameters.

[0055] Specifically, the implementation process of this embodiment includes:

[0056] When a seismic wave is incident at a perpendicular angle, the relationship between the reflection coefficient and the seismic wave impedance can be expressed as:

[0057] (1);

[0058] in, When indicating a round trip, Indicates underground wave impedance , and Let represent density and P-wave velocity, respectively. Based on Taylor expansion, the reflection coefficient equation can be approximated as a linear relationship:

[0059] (2);

[0060] For single-channel seismic records, the above formula can be transformed into matrix form:

[0061] (3);

[0062] in This is a first-order difference matrix. Considering that seismic records can be represented as the result of the convolution of the seismic wavelet and the reflection coefficient, combining formulas (2) and (3), we can obtain the linear relationship between wave impedance and seismic data in the case of a single channel:

[0063] (4);

[0064] in, It is a banded matrix constructed from seismic wavelets. It is the reflection coefficient vector. It is an orthogonal matrix. This represents the noise term. Considering the two periods of seismic data, they are defined as basic data respectively. With monitoring data The linear forward relationships of the two are as follows:

[0065] (5);

[0066] (6);

[0067] The relationship between the differences between two earthquake data sets and their corresponding model parameters can be expressed as follows:

[0068] (7);

[0069] in, This represents the combined noise and observation error term after subtracting and merging two sets of seismic data. New model parameters are defined. The above formula is transformed into a direct differential forward modeling equation:

[0070] (8);

[0071] Among them, model parameters The expression is:

[0072] (9);

[0073] By solving the parameters The relative wave impedance difference can be directly calculated, and its expression is:

[0074] (10);

[0075] in, This represents the absolute change in the difference in ground wave impedance between the monitoring point and the foundation. This represents the parameters of the composite model obtained from the inversion solution. The inverse exponential transformation operation is performed using the natural constant. Using the base as the equation, this process transforms the logarithmic domain parameters back to the linear physical domain. This step, which calculates the relative wave impedance difference, does not require the participation of the absolute value of the fundamental wave impedance. Therefore, it eliminates the need for independent single-period wave impedance inversion before the difference inversion, thus avoiding error propagation caused by inaccuracies in the fundamental model.

[0076] The results are as follows Figure 1 As shown, using Figure 1 The two-dimensional test model shown was used for verification. This benchmark model contains 168 sampling points in the vertical direction and 361 channels in the horizontal direction, with a channel spacing of 10 meters. Figure 1 The diagram shows a real wave impedance model, where (a) is the basic wave impedance model; (b) is the monitoring wave impedance model; and (c) is the differential wave impedance model. Figure 1 In (a), the black arrows indicate the locations of the main reservoirs for water injection oil recovery. (Comparison) Figure 1 (b) It can be seen that the wave impedance at this point is significantly reduced. Figure 1 (c) The actual difference shown can be up to 20%, while the actual difference in non-reservoir areas is strictly zero.

[0077] Obtained using convolution forward modeling theory Figure 2 The synthetic seismic data shown are as follows: (a) is the basic seismic data; (b) is the monitoring seismic data; and (c) is the noise-free differential seismic data obtained by subtracting the two periods.

[0078] (3) Extract wave impedance difference label data from the well logging data and construct a probabilistic mode conversion operator; generate an initial model parameter set that follows a multivariate Gaussian distribution using a stochastic simulation method, and generate a corresponding seismic record set based on the linear forward modeling relationship; including the following steps:

[0079] a. Extract the real acoustic impedance difference from well logging dynamic curves or known reservoir dynamic information as a tag data set;

[0080] b performs statistical analysis on the tag data set to obtain its cumulative probability distribution functional and its inverse function;

[0081] c performs statistics on the parameter set of a conventional model that follows a Gaussian distribution to obtain its Gaussian cumulative probability distribution functional and its inverse function;

[0082] Using the quantiles of the probability distribution as a bridge, and while maintaining consistency of the quantiles, d defines a function to transform the Gaussian distribution into the true distribution at the bottom of the well.

[0083] Specifically, the implementation process of this embodiment includes:

[0084] Assuming the wellbore in the study area is constructed with respect to the difference in wave impedance... A non-Gaussian distributed label is defined as:

[0085] (11);

[0086] in, This represents the set of non-Gaussian impedance difference data actually measured from wells in the study area, which can represent the true dynamic evolution characteristics of subsurface impedance. Statistical analysis was performed on the well data to obtain its cumulative probability distribution. The reverse mapping is:

[0087] (12);

[0088] in, This represents a functional for finding the cumulative probability distribution. This represents the inverse function of the real cumulative probability distribution functional. This represents the quantile of the probability distribution of well logging data, with a value space of [0,1]. The same transformation also applies to the set of model parameters that follow a Gaussian distribution generated by conventional stochastic simulations. : ,in, This represents the set of parameters for a conventional model that is initially generated using stochastic simulation methods such as geostatistics and strictly follows a multivariate Gaussian distribution. This represents the mathematical functional for obtaining the Gaussian cumulative probability distribution of the parameter set of the Gaussian model; This represents the equal probability quantile of the Gaussian cumulative probability distribution corresponding to the Gaussian distribution model set. Its inverse mapping is:

[0089] (13);

[0090] After transformation using the cumulative distribution function, both the well logging data distribution and the Gaussian model ensemble distribution are mapped to the interval [0,1]. This is achieved when the following conditions are met... Under these conditions, quantiles serve as a bridge connecting two distributions, defining a probability pattern transition operator. :

[0091] (14);

[0092] This transformation function can be used to transform any generated set of Gaussian distributed random models. The mapping is transformed into a set of models that conform to the true distribution law of the well. ,Right now

[0093] (15);

[0094] in, Indicates the process after probabilistic mode conversion operator After mapping and reshaping, the output model parameter set strictly follows the actual physical distribution law of the target work area above the well.

[0095] (4) Subtract the time-shifted seismic data to obtain the actual observation difference seismic data, and combine the actual observation difference seismic data, the seismic record set and the probability mode conversion operator to form a joint iterative update framework.

[0096] (5) Based on the ensemble smoothing theory, the model parameter set is iteratively updated using Kalman gain, and the probability mode transformation operator is applied to perform distribution mapping in each update to obtain the posterior impedance relative difference inversion result. This includes the following steps:

[0097] Generate a set of model parameters that follow a Gaussian distribution and the corresponding set of synthetic seismic records;

[0098] b. Calculate the Kalman gain using the cross-correlation matrix, autocorrelation matrix, noise correlation matrix, and expansion factor;

[0099] c. Use Kalman gain and residuals to update the state of the model;

[0100] d inputs the updated model into the probability pattern transformation operator, forcibly reshaping it into a model with true distribution characteristics, and uses it for the next iteration.

[0101] Specifically, the implementation process of this embodiment includes:

[0102] In the set-based Kalman framework for smoothing multiple data assimilation, consider A set of models that follow a Gaussian distribution and its corresponding synthetic seismic record set In the i-th iteration, the Kalman gain is... Represented as:

[0103] (16);

[0104] in, This is the cross-correlation matrix between the model set and the seismic data set. The autocorrelation matrix of the earthquake data, The noise correlation matrix is... Represents the expansion factor.

[0105] The constructed probability pattern transformation operator The update process is modified in the implanted iterative formula as follows:

[0106] (17);

[0107] After this update, a set of parameters that strictly conforms to the true distribution characteristics of the wellbore is obtained. When synthesizing the seismic record set required for the next iteration, a true morphological model with probabilistic mapping is used uniformly. To ensure the authenticity of the forward response, it is transformed back into a Gaussian distribution for the next Kalman gain calculation. After repeated iterations and convergence, the average value of the output parameter set is calculated to obtain the final high-precision relative impedance difference inversion result. A comparative inversion is performed using conventional ensemble smoothing difference inversion and the ensemble smoothing difference inversion based on probability mode conversion proposed in this embodiment. The results are as follows: Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown.

[0108] like Figure 3 The comparison chart of relative wave impedance inversion results is shown below. Figure 3 (a) Conventional ensemble smoothing differential inversion is severely limited by the Gaussian smoothing assumption, failing to characterize parameter sparsity and resulting in a large number of strong false anomalies remaining in non-major reservoir areas. In contrast, probabilistic mode conversion-based... Figure 3 (b) The inversion results of ensemble smoothing difference inversion and Figure 1 (c) is a perfect match for the true difference model and successfully suppresses background noise.

[0109] Figure 4 A comparison chart of single-channel inversion results for relative impedance differences under noise-free conditions is presented. By extracting typical single-channel curves, it can be intuitively seen that the ensemble smoothing difference inversion based on probability mode transformation, due to the forced constraint matching the actual well distribution pattern in each update, reduces the inversion uncertainty in non-reservoir backgrounds to almost zero, and extremely accurately characterizes the local abrupt changes and sparsity features of time-shifted difference signals.

[0110] In addition, in the noise resistance assessment, adding random Gaussian white noise to the differential seismic records significantly reduced its signal-to-noise ratio to 7. Figure 5The images show a comparison of the synthesized noisy differential seismic records and the inversion results, where (a) is the noisy differential seismic record under strong background interference; (b) is the inversion result of conventional ensemble smoothing differential inversion; and (c) is the inversion result of ensemble smoothing differential inversion based on probability mode transformation.

[0111] Figure 6 The comparison chart of single-channel inversion results for the relative difference in wave impedance when the signal-to-noise ratio is 7 is further shown. Figure 5 and Figure 6 The results fully demonstrate that the distribution reshaping mechanism based on probability mode transformation and ensemble smoothing difference inversion can serve as an extremely effective mathematical filter, significantly suppressing and filtering out false difference artifacts caused by complex background noise, and maintaining extremely high interpolation stability and noise robustness even under severe interference from strong background noise.

[0112] Example 2

[0113] Actual data was tested using the method described in Example 1: The actual test data came from two periods of marine time-shifted seismic data with a 15-year interval from a certain oilfield. This oilfield is a typical strongly heterogeneous loose sandstone reservoir with a relatively loose rock skeleton. Long-term large-volume water injection development has led to a rock skeleton erosion effect on the main sandstone bodies of the reservoir, resulting in a significant reduction in wave impedance and forming a typical negative differential skewed distribution. Authentic non-Gaussian differential logging labels were extracted from six drilled wells in the study area with complete long-term injection-production dynamic monitoring data.

[0114] Figure 7 The diagram shows the time-lapse seismic analysis near six wells, where (a) is a comparison of time-lapse seismic waveforms; (b) is the time-lapse seismic difference value; and (c) is the corresponding relative difference value of wave impedance obtained from well logging mapping. This ensemble smoothing difference inversion method based on probabilistic mode transformation is applied to the actual difference data volume of the entire work area.

[0115] Figure 8 The images show actual seismic data profiles, including (a) basic seismic data, (b) monitored seismic data, and (c) differential seismic data. It can be seen that after more than ten years of long-term water injection, the sand bodies in the main river channels of this area exhibit significant reflection anomalies in the monitoring data.

[0116] Figure 9 This is a comparison chart showing the inversion of relative wave impedance differences from actual data. Figure 9(a) indicates that the traditional ensemble smoothing difference inversion method suffers from strong and dispersed non-uniform background noise interference throughout the entire profile space, and the physical property differences in the main meandering channel appear discontinuous and unclear due to the smoothing effect. In contrast, the ensemble smoothing difference inversion method based on probability mode transformation used in this embodiment significantly filters out and removes background spurious differences in non-reservoir areas. The strong negative wave impedance difference anomaly obtained by the inversion is highly continuously distributed along the microfacies distribution direction of the sand body in the main meandering channel, which is completely consistent with the spatial distribution of the historical dynamic record of the water injection sweep range of the core water injection wells in the oilfield for more than ten years.

[0117] To further verify the fidelity of the inversion of deep, minute anomalies, the 350th trace, which is far from the edge and has weak reflection, was selected for one-dimensional curve detail comparison, such as... Figure 10 As shown in the one-dimensional difference inversion comparison diagram at channel 350, the inversion results of the set smoothing difference inversion based on probability mode transformation provide a more incisive and accurate characterization of the boundary of the main layer.

[0118] at last, Figure 11 A comparison chart of the inversion results and the distribution of differential impedance labels on the wellbore is presented. The global probability density curve shows that the distribution of the 3D inversion results based on ensemble smoothing differential inversion using probability mode transformation perfectly matches the highly skewed non-Gaussian distribution of the actual single-well labels. Both qualitative and quantitative verifications fully demonstrate that ensemble smoothing differential inversion based on probability mode transformation can accurately remove inconsistent acquisition noise in practical and complex engineering applications, and faithfully reproduce the actual reservoir dynamics of the water injection scour zone, demonstrating significant industrial application value.

[0119] This invention discloses an ensemble-smoothed time-shifted seismic differential inversion method based on probabilistic mode transformation, achieving direct inversion of elastic parameters that conform to the actual non-Gaussian distribution characteristics, thus improving the accuracy and geological rationality of time-shifted seismic differential inversion. This method cleverly maintains the advantages of ensemble multi-data assimilation methods in terms of high computational efficiency and the uncertainty of quantifying nonlinear inverse problems. By introducing a probabilistic mode transformation operator based on cumulative probability integral transformation, each model update is forced to conform to the non-Gaussian skewed or sparse distribution characteristics of real reservoir dynamics, suppressing the inconsistency artifacts caused by conventional Gaussian assumptions from a fundamental mathematical mechanism. Simultaneously, by innovatively reconstructing the relative acoustic impedance difference forward model, it avoids excessive reliance on high-precision absolute fundamental models, greatly reducing the risk of inversion error propagation under complex geological conditions. Under experimental conditions containing strong random disturbances, this method exhibits excellent interpolation stability and strong noise robustness, demonstrating high engineering practical value in fields such as fine prediction of remaining oil in complex old oilfields, fluid dynamic monitoring, and dynamic safety assessment of carbon capture and storage.

[0120] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for ensemble smoothing time-shifted seismic difference inversion based on probabilistic mode transformation, characterized in that, Includes the following steps: (1) Acquire time-lapse seismic data and well logging data of the underground area; (2) Based on the time-shifted seismic difference forward model, establish a linear forward modeling relationship between the observed seismic data with difference and the parameters of the relative wave impedance difference model; (3) Extract wave impedance difference label data from the well logging data and construct a probabilistic mode conversion operator; use stochastic simulation method to generate an initial model parameter set that follows a multivariate Gaussian distribution, and generate a corresponding set of seismic records based on the linear forward modeling relationship; (4) Subtract the time-shifted seismic data to obtain the actual observation difference seismic data, and combine the actual observation difference seismic data, the seismic record set, and the probability mode conversion operator to form a joint iterative update framework; (5) Based on the set smoothing theory, the Kalman gain is used to iteratively update the model parameter set, and the probability mode conversion operator is applied to perform distribution mapping in each update to obtain the posterior impedance relative difference inversion result; In step (3) above, the wave impedance difference label data is extracted from the logging data, and a probabilistic mode conversion operator is constructed, including the following steps: (3.1) Extract data representing the actual changes in underground wave impedance from well logging dynamic curves or known reservoir dynamic information as a tag data set; (3.2) Perform statistical analysis on the tag data set to obtain its true cumulative probability distribution functional and its inverse function; (3.3) Perform statistics on the set of parameters of a conventional model that follows a Gaussian distribution to obtain its Gaussian cumulative probability distribution functional; (3.4) Using the quantiles of the probability distribution as a connecting bridge, while keeping the quantiles consistent, a mapping function is defined to transform the Gaussian distribution into the Inoue true distribution, forming a probability mode conversion operator; In step (5) above, the probability mode transformation operator is applied to perform distribution mapping in each update to obtain the posterior impedance relative difference inversion result, including the following steps: (5.1) The set of model parameters after Kalman gain adjustment is directly input into the probability mode conversion operator for forced distribution mapping; (5.2) Output a set of model parameters that strictly follow the characteristics of the actual probability distribution in the well, and use them to synthesize the set of real seismic records required for the next iteration; (5.3) The model parameter set with real distribution characteristics is transformed into a Gaussian distribution state again to participate in the next Kalman gain calculation. After repeated iterations and convergence, the average value of the parameter set is calculated to obtain the final wave impedance relative difference inversion result.

2. The ensemble smoothing time-shifted seismic difference inversion method based on probabilistic mode transformation as described in claim 1, characterized in that, Step (2) establishes a linear forward modeling relationship between observed differential seismic data and relative impedance difference model parameters based on the time-shifted seismic difference model, including the following steps: (2.1) The single-channel seismic record is represented as the convolution of the seismic wavelet and the reflection coefficient, and the reflection coefficient is approximated as a linear relationship of the logarithm of the wave impedance using Taylor expansion; (2.2) Establish forward modeling equations for basic and monitoring seismic data respectively, and calculate the difference between the two seismic data to eliminate the static consistency background; (2.3) Define new model parameters that include the characteristics of relative wave impedance differences, and construct a direct linear forward modeling equation between the observed differential seismic data and the model parameters. After solving the model parameters by inversion, the relative difference value of wave impedance is directly calculated using the exponential relationship. In this process, the absolute value of the basic wave impedance does not need to be involved.

3. The ensemble smoothing time-shifted seismic difference inversion method based on probabilistic mode transformation as described in claim 1, characterized in that, Step (4) uses a stochastic simulation method to generate an initial set of model parameters that follows a multivariate Gaussian distribution, and generates a corresponding set of seismic records based on the linear forward modeling relationship, including the following steps: (4.1) Generate an initial set of model parameters that follow a multivariate Gaussian distribution based on the Gaussian simulation method of geostatistical sequences; (4.2) Using the linear forward modeling relation constructed by the seismic wavelet matrix, the initial model parameter set is synthesized into the corresponding seismic record set.

4. The ensemble smoothing time-shifted seismic difference inversion method based on probabilistic mode transformation as described in claim 1, characterized in that, Step (5) uses Kalman gain to iteratively update the model parameter set based on set smoothing theory, including the following steps: (5.4) Calculate the cross-correlation matrix and autocorrelation matrix using the model parameter set that follows a Gaussian distribution and its corresponding synthetic seismic record set; (5.5) Calculate the Kalman gain of the current iteration step by combining the error correlation matrix and the inflation factor; Calculate the residuals between observed seismic data and the synthetic seismic record set; (5.6) Using the Kalman gain and the residual, the current Gaussian distribution model parameter set is updated and fitted.

Images