Multi-attribute gas saturation prediction method and system

Through multi-attribute prediction method and deep neural network algorithm, the problem of gas-containing saturation prediction in tight sandstone reservoirs of Xujiahe Formation in Bazhong area was solved, and the prediction accuracy and reliability were improved, especially in porosity and gas-containing saturation. degree prediction.

CN119024433APending Publication Date: 2024-11-26CHINA PETROLEUM & CHEMICAL CORP +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202310597258.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the dense sandstone reservoir of Xujiahe Formation in Bazhong area, the fluid responds weakly in elastic parameters, and traditional methods cannot effectively predict gas saturation, and the quality of seismic data and inversion algorithms affect the prediction accuracy, resulting in low oil and gas detection accuracy.

Method used

Using multi-attribute gas-containing saturation prediction method, pre-processing is performed by acquiring seismic data, multi-mineral rock physical modeling is carried out in stratified systems, lithophagocytics are divided, phased inversion is carried out to obtain longitudinal and transverse wave impedances and density bodies, combined with geology Statistical inversion resulted in porosity bodies, and deep neural network machine learning algorithms were used to fit gas-containing saturation data bodies.

Benefits of technology

提高了含气饱和度预测的精准度,特别是在致密砂岩储层中,孔隙度预测准确度达到82%,同时增强了流体定量预测的可靠性。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119024433A_ABST
    Figure CN119024433A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-attribute gas saturation prediction method and system. The method comprises the following steps: acquiring and preprocessing a seismic data volume; carrying out multi-mineral rock physical modeling according to lithofacies characteristics and a layered system, and further carrying out rock physical analysis to divide lithofacies; performing phase control inversion according to the divided lithofacies to obtain pre-stack inversion longitudinal and transverse wave impedance bodies, a pre-stack inversion density body and a geostatistical inversion body; obtaining a porosity inversion body according to the geostatistics inversion body; according to the preprocessed seismic data volume, the pre-stack inversion longitudinal and transverse wave impedance body, the pre-stack inversion density body, the porosity inversion body and a deep neural network machine learning algorithm based on data split-phase driving, fitting a gas saturation data volume through segmented deep learning. According to the method, on the basis of effective reservoir lithofacies and porosity prediction research, the accuracy of gas saturation prediction is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of oil and gas exploration and development, and more specifically, to a multi-attribute gas saturation prediction method and system. Background Art

[0002] There are three main methods for fluid prediction using seismic data: the first is to use pre-stack AVO (Amplitude variation with offset) attribute analysis. This method mainly uses the simplified Shuey equation of the Zoeppritz equation to extract the intercept P and gradient G from the pre-stack gathers, and intersects P and G, and uses the phenomenon of deviation from the mudstone background line caused by the gas content of the reservoir to detect the gas layer; the second is to construct the fluid factor principle based on elastic theory, mainly through pre-stack elastic parameter inversion or elastic impedance inversion method to directly or indirectly calculate the three basic elastic parameters (P-wave velocity, S-wave velocity and density), and then use other combined elastic parameters such as Lame constant and Poisson's ratio to detect the oil and gas content of the reservoir; the third is based on seismic rock physics theory, establish the relationship between reservoir physical parameters such as water saturation and porosity and seismic elastic parameters, obtain elastic parameters through pre-stack seismic inversion, and then use rock physics plates to quantitatively predict gas saturation.

[0003] The first method is mainly based on qualitative prediction, and the P and G intersection effect is greatly affected by the quality of seismic data. The second and third methods both require the inversion of accurate P- and S-wave velocities and density parameters. The parameter inversion error will directly lead to an increase in the cumulative error of the combined elastic parameters, thereby affecting the accuracy of oil and gas detection. In practical applications, due to the combined influence of seismic data quality, field acquisition of seismic data, and inversion algorithms, it is difficult to accurately calculate the S-wave velocity and density. The lithology, pore structure, fluid properties, formation temperature and pressure of deeply buried rocks have a very complex influence on the elastic properties of rocks. How to obtain accurate rock physics maps through rock physics modeling is also one of the issues that affect the accuracy of fluid quantitative prediction.

[0004] In recent years, there are some technologies that use seismic attributes to detect gas content, such as using the frequency-varying characteristics of seismic data to detect gas content. However, unconventional oil and gas reservoirs such as tight sandstone have strong heterogeneity and anisotropy, which makes the reservoir geology and logging response and seismic prediction results present a more complex nonlinear relationship. Compared with conventional oil and gas prediction, it has stronger multi-solutions, and the effect of gas content identification through a single seismic attribute is poor.

[0005] The tight sandstone reservoirs of the Xujiahe Formation in the Bazhong area of ​​the study area have diverse mineral compositions, complex pore structures, poor matrix properties, and developed but unevenly distributed fractures. Due to the low physical properties of the complex tight sandstones of the Xujiahe Formation in the Bazhong area, the fluid has a weak response in elastic parameters, and traditional methods such as co-simulation cannot effectively predict gas saturation. Therefore, it is necessary to develop a multi-attribute gas saturation prediction method and system. The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as an admission or in any form of implication that the information constitutes prior art already known to those skilled in the art. Summary of the invention

[0006] Aiming at the technical difficulties such as weak AVO response of fluids in Bazhong area and difficulty in effectively predicting low gas saturation, a multi-attribute gas saturation prediction method and system was proposed based on the research on effective reservoir lithofacies and porosity prediction to improve the accuracy of quantitative prediction of tight sandstone fluids in Xujiahe Formation in Bazhong area.

[0007] In a first aspect, an embodiment of the present disclosure provides a multi-attribute gas saturation prediction method, comprising:

[0008] Acquire seismic data volume and perform preprocessing;

[0009] According to the characteristics of lithofacies, multi-mineral rock physics modeling is carried out in the stratified system, and then rock physics analysis is carried out to divide the lithofacies;

[0010] Performing phase-controlled inversion according to the divided lithofacies to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies;

[0011] Obtaining a porosity inversion volume based on the geostatistical inversion volume;

[0012] According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

[0013] Preferably, the preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

[0014] Preferably, the rock physical analysis comprises:

[0015] Geology-logging-seismic progressive lithofacies classification;

[0016] Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0017] Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0018] Preferably, the porosity inversion volume is a geostatistical stochastic co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

[0019] Preferably, according to the preprocessed seismic data body, the prestack inversion longitudinal and shear wave impedance body, the prestack inversion density body, and the porosity inversion body, based on the data phase-driven deep neural network machine learning algorithm, the gas saturation data body is fitted by segmented deep learning, including:

[0020] Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes;

[0021] The pre-stack AVO attributes, the post-stack absorption attenuation attributes, the pre-stack inversion P- and S-wave impedance bodies, the pre-stack inversion density body and the porosity inversion body are used as initial input data, and the logging gas saturation curve is used as a constraint. A deep neural network machine learning algorithm driven by data phase separation is used to fit the gas saturation data body through segmented deep learning.

[0022] As a specific implementation method of the embodiment of the present disclosure,

[0023] In a second aspect, the embodiments of the present disclosure further provide a multi-attribute gas saturation prediction system, including:

[0024] a memory storing computer executable instructions;

[0025] A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps:

[0026] Acquire seismic data volume and perform preprocessing;

[0027] According to the characteristics of lithofacies, multi-mineral rock physics modeling is carried out in the stratified system, and then rock physics analysis is carried out to divide the lithofacies;

[0028] Performing phase-controlled inversion according to the divided lithofacies to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies;

[0029] Obtaining a porosity inversion volume based on the geostatistical inversion volume;

[0030] According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

[0031] Preferably, the preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

[0032] Preferably, the rock physical analysis comprises:

[0033] Geology-logging-seismic progressive lithofacies classification;

[0034] Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0035] Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0036] Preferably, the porosity inversion volume is a geostatistical stochastic co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

[0037] Preferably, according to the preprocessed seismic data body, the prestack inversion longitudinal and shear wave impedance body, the prestack inversion density body, and the porosity inversion body, based on the data phase-driven deep neural network machine learning algorithm, the gas saturation data body is fitted by segmented deep learning, including:

[0038] Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes;

[0039] The pre-stack AVO attributes, the post-stack absorption attenuation attributes, the pre-stack inversion P- and S-wave impedance bodies, the pre-stack inversion density body and the porosity inversion body are used as initial input data, and the logging gas saturation curve is used as a constraint. A deep neural network machine learning algorithm driven by data phase separation is used to fit the gas saturation data body through segmented deep learning.

[0040] Its beneficial effects are:

[0041] In view of the technical difficulties that the tight sandstone reservoirs in the study area have diverse mineral compositions, complex pore structures, and poor matrix properties, which result in weak response of the fluid in elastic parameters and the inability of traditional co-simulation methods to effectively predict gas saturation, the present invention proposes a multi-attribute gas saturation prediction method, which can effectively improve the accuracy of gas saturation prediction.

[0042] The methods and systems of the present invention have other features and advantages that will be apparent from, or will be described in detail in, the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0044] Figure 1 A schematic diagram of prestack inversion longitudinal and shear wave impedance bodies and prestack inversion density bodies of Yuanlu 702-Yuanlu 18-Yuanlu 12 wells according to an embodiment of the present invention is shown.

[0045] Figure 2 A flowchart showing the steps of a multi-attribute gas saturation prediction method according to an embodiment of the present invention.

[0046] Figure 3 A porosity prediction profile diagram of a three-segment machine learning fusion co-simulation according to an embodiment of the present invention is shown.

[0047] Figure 4a-4d The three segments Tx3 and Tx4 are shown respectively according to an embodiment of the present invention. 1 、Tx3 2 、Tx3 3 、Tx3 4+5 Plane diagram of predicted gas saturation of effective reservoirs. DETAILED DESCRIPTION

[0048] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0049] The present invention provides a multi-attribute gas saturation prediction method, comprising:

[0050] Acquire seismic data volume and perform preprocessing;

[0051] According to the characteristics of lithofacies, the stratified system conducts multi-mineral rock physics modeling, and then conducts rock physics analysis to divide the lithofacies;

[0052] According to the divided lithofacies, phase-controlled inversion is performed to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies;

[0053] According to the geostatistical inversion volume, a porosity inversion volume is obtained;

[0054] According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

[0055] In one example, the preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

[0056] In one example, the petrophysical analysis includes:

[0057] Geology-logging-seismic progressive lithofacies classification;

[0058] Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0059] Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0060] In one example, the porosity inversion volume is a geostatistical stochastic co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

[0061] In one example, based on the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume, and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation, including:

[0062] Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes;

[0063] The pre-stack AVO attributes, post-stack absorption attenuation attributes, pre-stack inversion P-wave and S-wave impedance bodies, pre-stack inversion density bodies and porosity inversion bodies are taken as initial input data. The well logging gas saturation curve is used as a constraint. Based on the data phase-driven deep neural network machine learning algorithm, the gas saturation data body is fitted through segmented deep learning.

[0064] Specifically, the seismic data volume is obtained and the quality of the seismic data volume is optimized. The original gathers of seismic data are successively subjected to AVO law compensation, automatic flattening and denoising filter optimization processing. The optimization processing technical strategy of "original gathers-AVO law compensation-automatic flattening-denoising filter parameter combination" is adopted to effectively improve the overall quality of pre-stack gathers. Among them, the AVO law compensation further corrects the AVO law anomaly caused by the inherent defects of acquisition and processing, and uses its own relatively reliable mid-segment AVO law as a constraint to restore the amplitude characteristics of the near gather, so that the AVO law is reasonably restored, which lays a foundation for improving the accuracy of subsequent inversion. The automatic optimization stacking technology avoids the problems of reduced resolution, energy imbalance, and fuzzy breakpoints in the post-stack gathers caused by the low velocity accuracy of traditional horizontal stacking, thereby ensuring that the phase, resolution, and energy of the extracted angular gathers, especially the distant gathers, are better maintained. Compared with traditional horizontal stacking, the stacking resolution is slightly improved, the fractures are clearer, and the energy is further reset, providing a basis for later fine analysis. The results show that the plane law of the optimized processing results is consistent with the original stacking law, the longitudinal phase and characteristics are consistent, the credibility is high, and the plane trends of the inversion results using the original gather and the AVO compensated gather are consistent.

[0065] According to the complex lithofacies characteristics of the Xujiahe Formation in Bazhong, multi-mineral rock physics modeling was carried out in a stratified system. The Xujiahe Formation in Yuanba area in northeastern Sichuan is located in a mixed source area and is controlled by multiple provenances of the Longmenshan, Micangshan and Dabashan tectonic belts. It can be divided into types such as calcareous sandstone, quartz-rich sandstone and feldspar-rich sandstone. According to the complex lithofacies characteristics of the Xujiahe Formation in Yuanba, multi-mineral rock physics modeling was carried out in a stratified system using the RPM rock physics modeling module of PowerLog.

[0066] Rock physics analysis is carried out based on the established rock physics model. The Xujiahe Formation in Yuanba area in northeastern Sichuan is located in a mixed source area and in a braided river delta-lake sedimentary background. It is a tight sandstone reservoir with diverse mineral composition, poor physical properties, complex changes in rock physics parameters, and small differences in elastic parameters. Therefore, the following targeted rock physics analysis technical countermeasures are formulated:

[0067] (1) Geological-logging-seismic progressive lithofacies classification;

[0068] (2) Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0069] (3) Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0070] Figure 1 A schematic diagram of prestack inversion longitudinal and shear wave impedance bodies and prestack inversion density bodies of Yuanlu 702-Yuanlu 18-Yuanlu 12 wells according to an embodiment of the present invention is shown.

[0071] like Figure 1 As shown in the figure, under the control of the divided lithofacies, phase-controlled inversion is performed to obtain the prestack inversion P- and S-wave impedance bodies, prestack inversion density bodies and geostatistical inversion bodies. The geostatistical inversion of the Markov chain Monte Carlo algorithm is used to complete the statistical inversion of the prestack elastic parameters of each sand layer group in the third section of Xu.

[0072] Based on the geostatistical inversion volume, the co-simulation is performed to obtain the geostatistical random co-simulation porosity inversion volume. When simulating, the well points follow the well data, and the changes between wells are controlled by the reliable results of the lateral changes of seismic acoustic impedance. Through the geostatistical inversion results, the porosity variation function is fitted to obtain the geostatistical random co-simulation porosity inversion volume.

[0073] Although the porosity prediction results of machine learning and co-simulation are highly consistent with the wells, they each have limitations in terms of fine-tuning the distribution trend and accuracy of effective reservoirs. Therefore, in the research process, the deep neural network machine learning fusion co-simulation method was used to well integrate the accuracy of machine learning with the reasonable prediction trend of geostatistical co-simulation, improve the prediction accuracy and stability, and make the porosity prediction results more reasonable and reliable. Since there are too many zero values ​​in porosity, most machine learning algorithms cannot provide high prediction accuracy. However, the Gas-Mark-BP algorithm driven by data phase separation can provide high prediction accuracy. Beta The deep neural network learning algorithm completed the high-precision porosity prediction. Based on this result, the co-simulation porosity was further integrated with machine learning information and geological statistical information, and finally a reliable porosity prediction result was obtained. The accuracy of the porosity prediction results of the entire Xu 3rd section is as high as 82%.

[0074] Seismic attributes are extracted from the seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes. The pre-stack AVO attributes and post-stack absorption attenuation attributes are taken as the initial input data together with the pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies, and porosity inversion bodies. The well logging gas saturation curve is used as a constraint, and a deep neural network machine learning algorithm driven by data phase separation is adopted to fit the gas saturation data volume through segmented deep learning.

[0075] The present invention also provides a multi-attribute gas saturation prediction system, comprising:

[0076] a memory storing computer executable instructions;

[0077] A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps:

[0078] Acquire seismic data volume and perform preprocessing;

[0079] According to the characteristics of lithofacies, multi-mineral rock physics modeling is carried out in the stratified system, and then rock physics analysis is carried out to divide the lithofacies;

[0080] According to the divided lithofacies, phase-controlled inversion is performed to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies;

[0081] According to the geostatistical inversion volume, a porosity inversion volume is obtained;

[0082] According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

[0083] In one example, the preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

[0084] In one example, the petrophysical analysis includes:

[0085] Geology-logging-seismic progressive lithofacies classification;

[0086] Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0087] Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0088] In one example, the porosity inversion volume is a geostatistical stochastic co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

[0089] In one example, based on the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume, and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation, including:

[0090] Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes;

[0091] The pre-stack AVO attributes, post-stack absorption attenuation attributes, pre-stack inversion P-wave and S-wave impedance bodies, pre-stack inversion density bodies and porosity inversion bodies are taken as initial input data. The well logging gas saturation curve is used as a constraint. Based on the data phase-driven deep neural network machine learning algorithm, the gas saturation data body is fitted through segmented deep learning.

[0092] Specifically, the seismic data volume is obtained and the quality of the seismic data volume is optimized. The original gathers of seismic data are successively subjected to AVO law compensation, automatic flattening and denoising filter optimization processing. The optimization processing technical strategy of "original gathers-AVO law compensation-automatic flattening-denoising filter parameter combination" is adopted to effectively improve the overall quality of pre-stack gathers. Among them, the AVO law compensation further corrects the AVO law anomaly caused by the inherent defects of acquisition and processing, and uses its own relatively reliable mid-segment AVO law as a constraint to restore the amplitude characteristics of the near gather, so that the AVO law is reasonably restored, which lays a foundation for improving the accuracy of subsequent inversion. The automatic optimization stacking technology avoids the problems of reduced resolution, energy imbalance, and fuzzy breakpoints in the post-stack gathers caused by the low velocity accuracy of traditional horizontal stacking, thereby ensuring that the phase, resolution, and energy of the extracted angular gathers, especially the distant gathers, are better maintained. Compared with traditional horizontal stacking, the stacking resolution is slightly improved, the fractures are clearer, and the energy is further reset, providing a basis for later fine analysis. The results show that the plane law of the optimized processing results is consistent with the original stacking law, the longitudinal phase and characteristics are consistent, the credibility is high, and the plane trends of the inversion results using the original gather and the AVO compensated gather are consistent.

[0093] According to the complex lithofacies characteristics of the Xujiahe Formation in Bazhong, multi-mineral rock physics modeling was carried out in a stratified system. The Xujiahe Formation in Yuanba area in northeastern Sichuan is located in a mixed source area and is controlled by multiple provenances of the Longmenshan, Micangshan and Dabashan tectonic belts. It can be divided into types such as calcareous sandstone, quartz-rich sandstone and feldspar-rich sandstone. According to the complex lithofacies characteristics of the Xujiahe Formation in Yuanba, multi-mineral rock physics modeling was carried out in a stratified system using the RPM rock physics modeling module of PowerLog.

[0094] Rock physics analysis is carried out based on the established rock physics model. The Xujiahe Formation in Yuanba area in northeastern Sichuan is located in a mixed source area and in a braided river delta-lake sedimentary background. It is a tight sandstone reservoir with diverse mineral composition, poor physical properties, complex changes in rock physics parameters, and small differences in elastic parameters. Therefore, the following targeted rock physics analysis technical countermeasures are formulated:

[0095] (1) Geological-logging-seismic progressive lithofacies classification;

[0096] (2) Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0097] (3) Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0098] like Figure 1As shown in the figure, under the control of the divided lithofacies, phase-controlled inversion is performed to obtain the prestack inversion P- and S-wave impedance bodies, prestack inversion density bodies and geostatistical inversion bodies. The geostatistical inversion of the Markov chain Monte Carlo algorithm is used to complete the statistical inversion of the prestack elastic parameters of each sand layer group in the third section of Xu.

[0099] Based on the geostatistical inversion volume, the co-simulation is performed to obtain the geostatistical random co-simulation porosity inversion volume. When simulating, the well points follow the well data, and the changes between wells are controlled by the reliable results of the lateral changes of seismic acoustic impedance. Through the geostatistical inversion results, the porosity variation function is fitted to obtain the geostatistical random co-simulation porosity inversion volume.

[0100] Although the porosity prediction results of machine learning and co-simulation are highly consistent with the wells, they each have limitations in terms of fine-tuning the distribution trend and accuracy of effective reservoirs. Therefore, in the research process, the deep neural network machine learning fusion co-simulation method was used to well integrate the accuracy of machine learning with the reasonable prediction trend of geostatistical co-simulation, improve the prediction accuracy and stability, and make the porosity prediction results more reasonable and reliable. Since there are too many zero values ​​in porosity, most machine learning algorithms cannot provide high prediction accuracy. However, the Gas-Mark-BP algorithm driven by data phase separation can provide high prediction accuracy. Beta The deep neural network learning algorithm completed the high-precision porosity prediction. Based on this result, the co-simulation porosity was further integrated with machine learning information and geological statistical information, and finally a reliable porosity prediction result was obtained. The accuracy of the porosity prediction results of the entire Xu 3rd section is as high as 82%.

[0101] Seismic attributes are extracted from the seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes. The pre-stack AVO attributes and post-stack absorption attenuation attributes are taken as the initial input data together with the pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies, and porosity inversion bodies. The well logging gas saturation curve is used as a constraint, and a deep neural network machine learning algorithm driven by data phase separation is adopted to fit the gas saturation data volume through segmented deep learning.

[0102] To facilitate understanding of the solutions and effects of the embodiments of the present invention, two specific application examples are given below. Those skilled in the art should understand that the examples are only for facilitating understanding of the present invention, and any specific details thereof are not intended to limit the present invention in any way.

[0103] Example 1

[0104] Figure 2 A flowchart showing the steps of a multi-attribute gas saturation prediction method according to an embodiment of the present invention.

[0105] like Figure 2 As shown, the multi-attribute gas saturation prediction method includes: step 101, obtaining a seismic data body and performing preprocessing; step 102, conducting multi-mineral rock physical modeling in layers according to lithofacies characteristics, and then performing rock physical analysis to divide lithofacies; step 103, performing phase-controlled inversion according to the divided lithofacies to obtain a prestack inversion longitudinal and transverse wave impedance body, a prestack inversion density body and a geostatistical inversion body; step 104, obtaining a porosity inversion body according to the geostatistical inversion body; step 105, fitting the gas saturation data body through segmented deep learning based on the preprocessed seismic data body, the prestack inversion longitudinal and transverse wave impedance body, the prestack inversion density body and the porosity inversion body, and a deep neural network machine learning algorithm driven by data phase separation.

[0106] Figure 3 A porosity prediction profile diagram of a three-segment machine learning fusion co-simulation according to an embodiment of the present invention is shown.

[0107] In order to verify the stability of machine learning-assisted co-simulation porosity, three wells, Yuanlu 12, Yuanlu 15 and Yuanba 12 in three different regions, were selected as blind wells for the first round of machine learning verification. The verification results showed that the porosity prediction coincidence rates of the three wells were 81%, 80% and 81% respectively, with an average of 81%, proving a high degree of reliability. Based on this, the machine learning-constrained porosity co-simulation with the participation of all wells was completed. The porosity prediction profile of the three-stage machine learning fusion co-simulation is shown as follows: Figure 2 shown.

[0108] The qualitative to quantitative gas content detection under the constraints of high-quality reservoirs mainly starts from the following two aspects:

[0109] Qualitative: accurate prediction by means of pre-stack AVO attribute analysis and post-stack absorption attenuation attribute analysis;

[0110] Quantitative: Based on the control of porosity inversion results, the multi-attribute participation constraint and data phase-driven deep neural network machine learning method are adopted to improve the rationality and accuracy of gas saturation prediction. The specific implementation steps are to use pre-stack AVO attributes, pre-stack inversion longitudinal and shear wave impedance bodies, pre-stack inversion density, post-stack absorption attenuation attributes, porosity prediction and other data bodies as initial input data, and use the well logging gas saturation curve as a constraint, and use the Gas-Mark-BP based on data phase-driven. Beta The deep neural network machine learning algorithm fits the current gas state through segmented deep learning, which can not only ensure the resolution of gas saturation prediction, but also solve the problem of unreliable gas saturation inversion in co-simulation.

[0111] Since there are too many 1 values ​​in the gas saturation interpreted by logging, most machine learning algorithms cannot achieve high prediction accuracy. After multiple rounds of attempts, we finally chose Gas-Mark-BP, the same BP deep neural network technology as porosity. Beta The gas saturation is learned, and then the porosity body learned by deep neural network machine learning is used as control, and the AVO attributes and post-stack absorption attenuation attributes are combined to predict the gas saturation, ultimately achieving a high-precision prediction of gas saturation.

[0112] In order to verify the stability of machine learning gas saturation, three wells, Yuanlu 12, Yuanlu 15 and Yuanba 12 in three different areas, were selected as blind wells to carry out the first round of machine learning verification. The verification results showed that the gas saturation prediction coincidence rates of the three wells were 79%, 81% and 80% respectively, with an average of 80%, proving a high degree of reliability. Based on this, the machine learning gas saturation prediction with the participation of all wells was completed. The machine learning gas saturation was compared with the porosity results, and the two had a strong correlation and consistency, indicating that after being constrained by porosity, the prediction results of machine learning gas saturation were more consistent with the logical relationship between sedimentation, lithofacies and storage parameters.

[0113] Figure 4a-4d The three segments Tx3 and Tx4 are shown respectively according to an embodiment of the present invention. 1 、Tx3 2 、Tx3 3 、Tx3 4+5 The predicted plan view of gas saturation of effective reservoirs reveals the distribution characteristics of effective gas-bearing reservoirs, which are controlled by high-quality lithofacies development zones and physical property conditions. Among them, the main river channel zone has the largest gas content and is characterized by local sweet spots.

[0114] By comparing and verifying with the logging gas saturation parameters, the verification results show that the logging gas saturation parameters of the verification wells in multiple regions have a high consistency rate, which reflects that the method has a high reliability and high gas saturation prediction accuracy.

[0115] Example 2

[0116] The multi-attribute gas saturation prediction system includes:

[0117] a memory storing computer executable instructions;

[0118] A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps:

[0119] Acquire seismic data volume and perform preprocessing;

[0120] According to the characteristics of lithofacies, the stratified system conducts multi-mineral rock physics modeling, and then conducts rock physics analysis to divide the lithofacies;

[0121] According to the divided lithofacies, phase-controlled inversion is performed to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies;

[0122] According to the geostatistical inversion volume, a porosity inversion volume is obtained;

[0123] According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

[0124] In one example, the preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

[0125] In one example, the petrophysical analysis includes:

[0126] Geology-logging-seismic progressive lithofacies classification;

[0127] Reconstruction of sensitivity curves of elastic parameters through machine learning;

[0128] Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

[0129] In one example, a co-simulation is performed based on a geostatistical inversion volume to obtain a porosity inversion volume of geostatistical stochastic co-simulation.

[0130] In one example, based on the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume, and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation, including:

[0131] Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes;

[0132] The pre-stack AVO attributes, post-stack absorption attenuation attributes, pre-stack inversion P-wave and S-wave impedance bodies, pre-stack inversion density bodies and porosity inversion bodies are taken as initial input data. The well logging gas saturation curve is used as a constraint. Based on the data phase-driven deep neural network machine learning algorithm, the gas saturation data body is fitted through segmented deep learning.

[0133] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any given examples.

[0134] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A multi-attribute gas saturation prediction method, characterized in that: include: Acquire seismic data volume and perform preprocessing; According to the characteristics of lithofacies, the stratified system conducts multi-mineral rock physics modeling, and then conducts rock physics analysis to divide the lithofacies; Performing phase-controlled inversion according to the divided lithofacies to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies; Obtaining a porosity inversion volume based on the geostatistical inversion volume; According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

2. The multi-attribute gas saturation prediction method according to claim 1, wherein: The preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

3. The multi-attribute gas saturation prediction method according to claim 1, wherein: The rock physical analysis includes: Geology-logging-seismic progressive lithofacies classification; Reconstruction of sensitivity curves of elastic parameters through machine learning; Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

4. The multi-attribute gas saturation prediction method according to claim 1, wherein: The porosity inversion volume is a geostatistical random co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

5. The multi-attribute gas saturation prediction method according to claim 1, wherein: According to the pre-processed seismic data volume, pre-stack inversion P-wave and S-wave impedance volume, pre-stack inversion density volume, and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation, including: Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes; The pre-stack AVO attributes, the post-stack absorption attenuation attributes, the pre-stack inversion P- and S-wave impedance bodies, the pre-stack inversion density body and the porosity inversion body are used as initial input data, and the logging gas saturation curve is used as a constraint. A deep neural network machine learning algorithm driven by data phase separation is used to fit the gas saturation data body through segmented deep learning.

6. A multi-attribute gas saturation prediction system, characterized in that: The system includes: a memory storing computer executable instructions; A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps: Acquire seismic data volume and perform preprocessing; According to the characteristics of lithofacies, the stratified system conducts multi-mineral rock physics modeling, and then conducts rock physics analysis to divide the lithofacies; Performing phase-controlled inversion according to the divided lithofacies to obtain pre-stack inversion P- and S-wave impedance bodies, pre-stack inversion density bodies and geostatistical inversion bodies; Obtaining a porosity inversion volume based on the geostatistical inversion volume; According to the preprocessed seismic data volume, prestack inversion longitudinal and shear wave impedance volume, prestack inversion density volume and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation.

7. The multi-attribute gas saturation prediction system according to claim 6, wherein: The preprocessing includes AVO law compensation, automatic leveling and denoising filtering.

8. The multi-attribute gas saturation prediction system according to claim 6, wherein: The rock physical analysis includes: Geology-logging-seismic progressive lithofacies classification; Reconstruction of sensitivity curves of elastic parameters through machine learning; Automatic clustering of multi-well and multi-attribute PDF probabilistic lithofacies.

9. The multi-attribute gas saturation prediction system according to claim 6, wherein: The porosity inversion volume is a geostatistical random co-simulation porosity inversion volume obtained by co-simulating the geostatistical inversion volume.

10. The multi-attribute gas saturation prediction system according to claim 6, wherein: According to the pre-processed seismic data volume, pre-stack inversion P-wave and S-wave impedance volume, pre-stack inversion density volume, and porosity inversion volume, the gas saturation data volume is fitted through segmented deep learning based on the deep neural network machine learning algorithm driven by data phase separation, including: Extract seismic attributes from the pre-processed seismic data volume to obtain pre-stack AVO attributes and post-stack absorption attenuation attributes; The pre-stack AVO attributes, the post-stack absorption attenuation attributes, the pre-stack inversion P- and S-wave impedance bodies, the pre-stack inversion density body and the porosity inversion body are used as initial input data, and the logging gas saturation curve is used as a constraint. A deep neural network machine learning algorithm driven by data phase separation is used to fit the gas saturation data body through segmented deep learning.

Citation Information

Cited By

  • Deep stratum rock physical modeling method under complex lithologic background

    CN120762138A