Well-seismic combined machine learning pre-stack gather amplitude compensation system and method

Abstract

The invention discloses a well-seismic combined machine learning pre-stack gather amplitude compensation system. The system comprises a data acquisition module used for acquiring well logging stratum data, post-stack seismic data and an actual seismic gather in a work area; the target well preparation module is used for selecting a target well, completing well-seismic calibration on the target well by using the post-stack seismic data, and extracting actual seismic wavelets of the target well based on the target well; the synthetic gather forward modeling module is used for carrying out synthetic gather forward modeling according to the actual seismic wavelets to obtain a forward modeling AVO curve; the data extraction module is used for extracting amplitude value data of an actual AVO curve corresponding to the mudstone top surface as feature data and extracting amplitude value data of a forward modeling AVO curve corresponding to the mudstone top surface as label data; and the model building module trains a machine learning model according to the feature data and the label data to obtain a pre-stack gather amplitude compensation model. According to the method, the problems of over-processing, poor applicability, ignoring of important features and the like can be solved, and the accuracy of processing the pre-stack gather data is improved.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration technology, and in particular to a well-seismic combined machine learning pre-stack gather amplitude compensation system and method. Background Art

[0002] In the field of seismic exploration, prestack gather data, as a crucial form of seismic data, can deeply reveal rich geological information such as underground lithological characteristics, physical properties, and oil and gas content. The quality of prestack gather data is directly and closely related to the accuracy of subsequent seismic inversion and reservoir prediction, and has a decisive impact on the success of oil and gas exploration.

[0003] In the prior art, the processing of pre-stack gather data mainly relies on traditional seismic data processing methods. Since seismic data is easily disturbed by various noises during the acquisition process, these noises will seriously interfere with the authenticity and accuracy of the data. Traditional noise removal and amplitude compensation methods often lead to processing transitions, sacrificing the amplitude preservation of seismic data, resulting in a large deviation between the actual gather AVO and the AVO trend of the well logging forward gather, and it is impossible to obtain real and effective geological information. At the same time, due to the inhomogeneity of the underground medium, the error of the observation system or the improper operation during the data processing process, the amplitude anomaly may be caused. Although the amplitude is compensated by the difference in the number of coverages at different offsets in the prior art, this method ignores the well point AVO characteristics, resulting in the inconsistency between the compensated gather and the well point AVO trend, especially under complex geological conditions, the compensation effect is not ideal; the prior art also realizes gather compensation based on well point AVO control, but when the well point compensation coefficient is extended to the entire area, especially in the well-free area, the amplitude preservation of the compensated gather still needs to be verified. These factors work together to significantly reduce the quality of pre-stack gather data, which in turn causes great trouble and obstacles to subsequent geological interpretation and analysis.

[0004] Therefore, inventing a well-seismic combined machine learning pre-stack gather amplitude compensation system that can solve the defects of the existing technology such as over-processing, insufficient regional applicability, neglect of important features, and poor adaptability to complex lithology, thereby ensuring more efficient, accurate and reliable processing of pre-stack gather data has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a well-seismic combined machine learning pre-stack gather amplitude compensation system, which can solve the problems of over-processing, insufficient regional applicability, neglect of important features and poor adaptability to complex lithology, and improve the accuracy and reliability of processing pre-stack gather data.

[0006] To achieve this purpose, the present invention designs a well-seismic combined machine learning pre-stack gather amplitude compensation system, which includes:

[0007] The data acquisition module is used to collect logging formation data, post-stack seismic data and actual seismic gathers in the work area, and obtain the actual AVO curve according to the actual seismic gathers;

[0008] The target well preparation module is used to select the target well according to the logging formation data and the shale section that meets the lithology setting, complete the well-seismic calibration of the target well using the post-stack seismic data, and extract the actual seismic wavelet of the target well based on the well-seismic calibrated target well;

[0009] The synthetic gather forward modeling module is used to perform synthetic gather forward modeling of the target well according to the actual seismic wavelet of the target well to obtain the forward AVO curve;

[0010] The data extraction module is used to extract the amplitude value data of the actual AVO curve corresponding to the top of the shale in the target well as the characteristic data; and extract the amplitude value data of the forward AVO curve corresponding to the top of the shale in the target well as the label data;

[0011] The model establishment module trains the machine learning model according to the characteristic data and the label data to obtain the pre-stack gather amplitude compensation model.

[0012] Preferably, it further includes a model verification module, which is used to verify the pre-stack gather amplitude compensation model with the amplitude value data in the forward AVO curve that does not participate in machine learning, and compare whether the trend of the compensated AVO curve is consistent with the trend of the actual AVO curve;

[0013] According to the actual seismic gather after being compensated by the pre-stack gather amplitude compensation model, at a certain moment on the compensated actual seismic gather, the amplitude values at different offsets are extracted to obtain the compensated AVO curve.

[0014] Preferably, it further includes a three-dimensional gather amplitude compensation module. The three-dimensional gather amplitude compensation module is used to calculate the amplitude ratio between the compensated actual seismic gather and the actual seismic gather to obtain the amplitude compensation factors at different offsets; fit the amplitude compensation factors at different offsets of the target well with the coverage times at different offsets, so as to use the fitted relationship to convert the coverage times at different offsets of other actual seismic gathers in the work area except the target well into compensation factors, thereby obtaining the three-dimensional gather amplitude compensation factor field of the target well, and compensating the three-dimensional gather of the target well using the three-dimensional gather amplitude compensation factor field.

[0015] Preferably, the specific generation method of the actual AVO curve and the forward AVO curve is: the actual AVO curve is a certain moment on the actual seismic gather, and the offset distances are arranged in order from small to large. The actual AVO curve can be determined by displaying the amplitude value of the actual seismic gather; the forward gather is generated by forward modeling of the synthetic gather, and the forward AVO curve is a certain moment on the forward gather, and the offset distances are arranged in order from small to large. The forward AVO curve can be determined by selecting the amplitude value of the forward gather.

[0016] Preferably, the specific method of extracting characteristic data and label data is as follows: the actual seismic gather data is the amplitude value data of the actual AVO curve in the vertical direction according to the time depth sequence and in the horizontal direction according to the offset size sequence, and the amplitude value of the actual AVO curve corresponding to the mudstone top surface is selected as the characteristic data; the forward modeling gather data is the amplitude value data of the forward modeling AVO curve in the vertical direction according to the time depth sequence and in the horizontal direction according to the offset size sequence, and the amplitude value of the forward modeling AVO curve corresponding to the mudstone top surface is selected as the label data.

[0017] Preferably, the specific method for obtaining the pre-stack gather amplitude compensation model is: taking the actual seismic gather AVO curve and the forward AVO curve as input, and using a machine learning algorithm to fit the mapping relationship between the two, that is, a machine learning gather compensation model.

[0018] Beneficial effects of the present invention: The present invention proposes a well-seismic combined machine learning pre-stack gather amplitude compensation system, which combines well-seismic data and seismic wavelets to perform synthetic gather forward modeling, extract actual AVO curves as feature data, and use forward modeled AVO curves as label data. This model construction method based on high-precision data can better retain the true characteristics of seismic data and avoid information loss caused by excessive processing; by fitting the compensation factor of the target well point with the number of coverages, an amplitude compensation factor field for the entire area is generated, thereby achieving efficient compensation for pre-stack gathers in the entire area; using the powerful nonlinear fitting ability of the SVM algorithm, a complex mapping relationship between logging data, pre-stack gather data and offset coverage times is established, and the fitting method based on multiple features can better capture important features in the data and avoid ignoring key information; the high-precision pre-stack gather amplitude compensation model established by the machine learning algorithm can adapt to various complex lithology formations and capture the amplitude variation law in different lithology formations. The present invention solves the problems of over-processing, insufficient regional applicability, neglect of important features and poor adaptability to complex lithology by establishing a pre-stack gather amplitude compensation model, which helps to improve the accuracy and reliability of processing pre-stack gather data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural schematic diagram of the present invention;

[0020] Figure 2 Schematic diagram of the process of the present invention;

[0021] Figure 3 Schematic diagram of the pre-stack gather beside the well, forward synthetic gather record, logging curve of the target well and AVO curve (tag data) of the top surface of the mudstone in the example of the present invention;

[0022] Figure 4 Schematic diagram of model training of a machine learning pre-stack gather amplitude compensation system combining well and seismic of the present invention;

[0023] Figure 5 Comparison chart of other forward AVO curves and AVO trend of the pre-stack gather after optimization by the application model in the application example of the present invention. Specific implementation manners

[0024] To make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. Therefore, the detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] The following further elaborates on the present invention with reference to the accompanying drawings and specific embodiments:

[0026] Embodiment 1

[0027] A machine learning pre-stack gather amplitude compensation system combining well and seismic, as Figure 1 shown, it includes:

[0028] The data acquisition module is used to acquire logging formation data, post-stack seismic data and actual seismic gathers in the work area, and obtain the actual AVO curve according to the actual seismic gathers;

[0029] The target well preparation module is used to select the target well according to the logging formation data and the mudstone section that meets the lithology setting, complete the well-seismic calibration for the target well using the post-stack seismic data, and extract the actual seismic wavelet of the target well based on the well-seismic calibrated target well;

[0030] The synthetic gather forward modeling module is used to perform forward modeling of the synthetic gathers of the target well according to the actual seismic wavelet of the target well to obtain the forward AVO curve;

[0031] The data extraction module is used to extract the amplitude value data of the actual AVO curve corresponding to the top surface of the mudstone in the target well as feature data; and extract the amplitude value data of the forward AVO curve corresponding to the top surface of the mudstone in the target well as label data;

[0032] The model building module trains the machine learning model based on the feature data and label data to obtain a pre-stack gather amplitude compensation model, as Figure 4 shown.

[0033] In the above technical solution, the data acquisition module collects logging formation data, post-stack seismic data, and actual seismic gathers in the work area, providing basic data for subsequent processing and analysis, ensuring the comprehensiveness of the data, and providing rich information for subsequent processing; generating the actual AVO curve through the actual seismic gathers as a reference for subsequent model training and compensation; the actual seismic gathers and the actual AVO curve directly reflect the true seismic response of the underground geological structure, providing reliable basic data for model training.

[0034] In the above technical solution, the target well preparation module ensures the representativeness of the data by selecting the target well in the mudstone section that meets the lithology setting, and can better reflect the geological characteristics of the target area; through well-seismic calibration, the logging data and seismic data are closely combined, improving the accuracy and reliability of the data.

[0035] In the above technical solution, the synthetic gather forward modeling module can simulate the seismic response of the target well at different offsets through the forward AVO curve, providing a reference standard for subsequent amplitude compensation; the amplitude value of the forward AVO curve is used as the label data of the machine learning model, providing a clear target for model training; by comparing the forward data with the actual data, the performance and compensation effect of the model can be better evaluated.

[0036] In the above technical solution, the data extraction module excludes the influence of formation pores and fluids on the AVO curve by extracting the amplitude value of the mudstone top surface, ensuring the consistency and geological representativeness of the data, focusing on the amplitude value of the mudstone top surface, reducing the complexity of the data, and improving the efficiency of model training.

[0037] In the above technical solution, the model building module can automatically perform amplitude compensation through the machine learning model, reducing manual intervention and human errors; the model has a strong non-linear fitting ability, can adapt to various complex geological conditions, and capture the amplitude change laws in different lithology formations; through high-precision amplitude compensation, it can more accurately identify oil and gas reservoirs and predict formation properties, thereby improving the success rate of oil and gas exploration; after verification and optimization, the model can effectively compensate the data within the entire work area, enhancing the generalization ability of the model.

[0038] In the above technical solution, it further includes a model verification module, as Figure 5 shown, which is used to verify the pre-stack gather amplitude compensation model with the amplitude value data in the forward AVO curve that does not participate in machine learning, and compare whether the trend of the compensated AVO curve is consistent with the actual AVO curve trend, so as to verify whether the model accuracy meets the requirements;

[0039] According to the actual seismic gather after being compensated by the pre-stack gather amplitude compensation model, at a certain moment on the compensated actual seismic gather, the amplitude values at different offsets are extracted to obtain the compensated AVO curve.

[0040] In the above technical solution, the forward AVO curve that does not participate in machine learning is the amplitude value of the forward AVO curve except for the amplitude value data of the forward AVO curve corresponding to the top surface of the mudstone in the target well.

[0041] In the above technical solution, the model establishment module uses the SVM algorithm to establish a model that can accurately compensate the pre-stack gather amplitude by optimizing the mapping relationship between the feature data and the label data. Through steps such as data preprocessing, selecting an appropriate kernel function, parameter optimization, model training and verification, a high-precision compensation model is finally obtained.

[0042] In the above technical solution, if the trend of the compensated AVO curve is consistent with the actual AVO curve trend, it indicates that the pre-stack gather amplitude compensation model can accurately reflect the underground geological characteristics, and the fitting and prediction capabilities of the model are strong, which helps to more accurately identify oil and gas reservoirs and predict formation properties, improving the success rate of oil and gas exploration; if the trend of the compensated AVO curve is inconsistent with the actual AVO curve trend, it indicates that the model fails to accurately capture the characteristics of seismic data in some aspects, and there may be problems of overfitting or underfitting, which indicates that the fitting ability or generalization ability of the model is insufficient and needs further optimization.

[0043] In the above technical solution, in order to verify the performance and accuracy of the machine learning pre-stack gather amplitude compensation model, by using the amplitude value data in the forward AVO curve that does not participate in model training as verification data, the performance of the model on independent data can be tested. This not only improves the model reliability and verifies the model generalization ability, but also adjusts the model parameters or optimizes the algorithm accordingly, thereby reducing the risk caused by model errors.

[0044] In the above technical solution, it further includes a three-dimensional gather amplitude compensation module. The three-dimensional gather amplitude compensation module is used to calculate the amplitude ratio between the compensated actual seismic gather and the actual seismic gather, so as to obtain the amplitude compensation factors for different offsets; fit the amplitude compensation factors for different offsets of the target well with the coverage times for different offsets, and then use the fitted relationship to convert the coverage times for different offsets of other actual seismic gathers in the work area except the target well into compensation factors, thereby obtaining the three-dimensional gather amplitude compensation factor field of the target well, and using the three-dimensional gather amplitude compensation factor field to compensate the three-dimensional gather of the target well.

[0045] In the above technical solution, the coverage times are determined by the number of shot points and the effective number of seismic receptions at geophone points in the actual seismic gathers collected in the work area.

[0046] In the above technical solution, by extending the compensation factors of the target well to other actual seismic gathers in the entire work area and combining with the machine learning correction model, fitting the relationship between the amplitude compensation factors and the coverage times for different offsets at different well points reduces the data discontinuity caused by local compensation, enhances the adaptability of the model to different geological conditions, improves the generalization ability of the model and the overall quality of seismic data, and reduces manual intervention and human errors.

[0047] In the above technical solution, the specific method for extracting the actual seismic wavelet is as follows: calculate the reflection coefficient sequence according to the logging data, and obtain the actual seismic wavelet through the convolution model based on the reflection coefficient sequence and the seismic trace at the well point of the target well;

[0048] The specific calculation formula for calculating the reflection coefficient sequence according to the logging data is as follows:

[0049] Reflection coefficient

[0050] where IPn is the wave impedance value of the lower layer of the interface, and IPn-1 is the wave impedance value of the upper layer of the interface.

[0051] In the above technical solution, calculating the reflection coefficient sequence through logging data and extracting the actual seismic wavelet in combination with the convolution model can more accurately reflect the propagation characteristics of seismic waves in underground media; extracting the actual seismic wavelet plays a crucial role in the subsequent forward modeling of synthetic gathers. The synthetic gather is obtained by convolving the seismic wavelet with the reflection coefficient sequence, which can simulate the true response of seismic data, thereby providing a reference standard for pre-stack gather amplitude compensation. Calculating a more accurate seismic wavelet can improve the accuracy of amplitude compensation.

[0052] In the above technical solution, the specific method for carrying out the forward modeling of the synthetic gather of the target well based on the actual seismic wavelet is as follows:

[0053] Perform forward modeling of the synthetic gather for the target well by combining the actual seismic wavelet with Zoeppritz's equation. Among them, based on Zoeppritz's approximation formula, calculate the reflection coefficients corresponding to different offsets of the target well:

[0054]

[0055] Among them, θ is the seismic incident angle; R(θ) is the reflection coefficient corresponding to different incident angles; Vp is the P-wave velocity of the underground formation; Vs is the S-wave velocity of the underground formation; ρ is the density of the underground formation; S(θ) is the seismic trace obtained by convolving R(θ) with the seismic wavelet when the incident angle is θ;

[0056] The seismic record S(θ) of the target well can be expressed in the form of the convolution of the wavelet and the reflection coefficients at different offsets:

[0057] S(θ) = R(θ) * W

[0058] Based on the convolution principle, S(θ) is the amplitude value corresponding to different incident angles, R(θ) is the reflection coefficient corresponding to different incident angles, and W is the actual seismic wavelet.

[0059] In the above technical solution, Zoeppritz's approximation equation takes into account the influence of various geological parameters on the reflection coefficient, can more accurately simulate the seismic response at different offsets, make the comparison between the synthetic gather and the actual seismic gather more reliable, and thus improve the accuracy of amplitude compensation; under complex geological conditions, the propagation characteristics of seismic waves may be affected by various factors, and Zoeppritz's approximate equation can effectively capture these changes. Especially in complex lithology formations such as gas-bearing sandstone and carbonate rock, it can better reflect the amplitude change law of seismic waves.

[0060] In the above technical solution, the specific method for performing forward modeling of the synthetic gather for the target well according to the actual seismic wavelet is:

[0061] Perform forward modeling of the synthetic gather for the target well by combining the actual seismic wavelet with the approximate equation. Among them, based on the approximate equation, calculate the reflection coefficients corresponding to different offsets of the target well: For example: offset 100m, at a depth of 2000m, for an incident angle of 2.87°:

[0062]

[0063] Offset 200m, at a depth of 2000m, for an incident angle of 5.71°:

[0064]

[0065] ……

[0066] Offset n for an incident angle of θ:

[0067]

[0068] Among them, R(θ) is the reflection coefficient corresponding to different incident angles; Vp is the P-wave velocity of the underground formation; Vs is the S-wave velocity of the underground formation; ρ is the density of the underground formation; θ is the seismic incident angle, and the specific calculation formula for the incident angle is:

[0069] θ = arctan(offset / depth)

[0070] Among them, offset is the offset (half of the distance from the shot point to the geophone position), and depth is the depth of the underground reflection interface;

[0071] Then calculate the amplitude values at different offsets:

[0072] S(2.87) = R(2.87) * W

[0073] S(5.71) = R(5.71) * W

[0074] ……

[0075] S(θ) = R(θ) * W

[0076] Based on the convolution principle, S(θ) is the amplitude value corresponding to different incident angles, R(θ) is the reflection coefficient corresponding to different incident angles, and W is the actual seismic wavelet. Finally, arrange S(2.87), S(5.71)……S(θ) in ascending order of offset to obtain the synthetic gather.

[0077] In the above technical solution, as Figure 3 shown, the specific generation methods of the actual AVO curve and the forward AVO curve are:

[0078] The actual AVO curve is at a certain moment on the actual seismic gather. Arrange the offsets in ascending order, and the amplitude values of the actual seismic gather can be displayed to determine the actual AVO curve; generate the forward gather through forward modeling of the synthetic gather. The forward AVO curve is at a certain moment on the forward gather. Arrange the offsets in ascending order, and select the amplitude values of the forward gather to determine the forward AVO curve.

[0079] In the above technical solution, the actual AVO curve and the forward AVO curve are generated through a standardized generation method. This standardized generation method ensures the consistency and comparability of the data, provides a unified benchmark for subsequent amplitude compensation and model training. High-quality input data can improve the training effect and prediction accuracy of the model, and improve the efficiency and reliability of data processing.

[0080] In the above technical solution, the specific method for extracting the characteristic data and label data is as follows: The actual seismic trace gather data is the amplitude value data of the actual AVO curve in the vertical direction following the time-depth sequence and in the horizontal direction following the offset size sequence. The amplitude value of the actual AVO curve corresponding to the top of the shale is selected as the characteristic data from it; The forward modeling trace gather data is the amplitude value data of the forward modeling AVO curve in the vertical direction following the time-depth sequence and in the horizontal direction following the offset size sequence. The amplitude value of the forward modeling AVO curve corresponding to the top of the shale is selected as the label data from it.

[0081] In the above technical solution, the amplitude value of the AVO curve corresponding to the top of the shale is selected to exclude the influence of pores and fluids in the formation on the AVO curve, which can ensure that the extracted data has good geological representativeness; This data extraction method can better reflect the seismic response of the underground geological structure, thus providing more accurate information for subsequent geological interpretation.

[0082] In the above technical solution, the specific method for obtaining the pre-stack trace gather amplitude compensation model is as follows: Using the actual seismic trace gather AVO curve and the forward modeling AVO curve as inputs, and using machine learning algorithms to fit the mapping relationship between the two, that is, the machine learning trace gather compensation model.

[0083] In the above technical solution, using machine learning algorithms to fit the mapping relationship between the actual seismic trace gather and the forward modeling AVO curve can effectively improve the accuracy, adaptability and efficiency of pre-stack trace gather amplitude compensation, thereby improving the quality of seismic data and the success rate of oil and gas exploration.

[0084] Embodiment 2

[0085] A machine learning pre-stack trace gather amplitude compensation method combining well and seismic data, as Figure 2 shown, collect the required data in the work area, and select a well with well logging data of longitudinal wave velocity (Vp), shear wave velocity (Vs), and underground formation density (Den) and a shale section (low porosity section) with stable lithology development as the target well. Then, complete the well-seismic calibration for the target well and extract the actual seismic wavelet of the target well; Apply Zoeppritz equation (or approximate equation) to carry out forward modeling of the synthetic trace gather of the target well; Extract the amplitude values of the actual AVO curves corresponding to the top of 4 sets of shales in the target well as characteristic data, and extract the amplitude values of the forward modeling AVO curves corresponding to the top of 4 sets of shales in the target well as label data; Input the characteristic data and label data, use machine learning algorithms to establish a high-precision pre-stack trace gather optimization model, and conduct comparison and verification on the model.

[0086] The pre-stack trace gather amplitude compensation method includes the following steps:

[0087] Collect well logging formation data, post-stack seismic data and actual seismic trace gather in the work area, and obtain the actual AVO curve according to the actual seismic trace gather;

[0088] Select a target well according to the logging formation data and the shale section set according to the lithology. Use the post-stack seismic data to complete the well-seismic calibration of the target well. Extract the actual seismic wavelet of the target well based on the well-seismic calibrated target well.

[0089] Carry out forward modeling of the synthetic gather of the target well according to the actual seismic wavelet of the target well to obtain the forward AVO curve.

[0090] Extract the amplitude value data of the actual AVO curve corresponding to the top surface of the shale in the target well as the characteristic data; and extract the amplitude value data of the forward AVO curve corresponding to the top surface of the shale in the target well as the label data.

[0091] The model establishment module trains the machine learning model according to the characteristic data and the label data to obtain a pre-stack gather amplitude compensation model.

[0092] Embodiment 3

[0093] A computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps of the method described in Embodiment 2.

[0094] The content not detailedly described in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. A well-seismic combined machine learning pre-stack gather amplitude compensation system, characterized in that: It includes: The data acquisition module is used to collect well logging formation data, post-stack seismic data and actual seismic gathers in the work area, and obtain the actual AVO curve based on the actual seismic gathers; The target well preparation module is used to select the target well according to the well logging formation data and the mudstone section that meets the lithology setting, complete the well-seismic calibration of the target well using the post-stack seismic data, and extract the actual seismic wavelet of the target well based on the well-seismic calibration; The synthetic gather forward modeling module is used to carry out synthetic gather forward modeling of the target well according to the actual seismic wavelet of the target well to obtain the forward modeling AVO curve; The data extraction module is used to extract the amplitude value data of the actual AVO curve corresponding to the top surface of the mudstone in the target well as feature data; and to extract the amplitude value data of the forward AVO curve corresponding to the top surface of the mudstone in the target well as label data; The model building module trains the machine learning model according to the feature data and label data to obtain the pre-stack gather amplitude compensation model.

2. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: It also includes a model verification module, which is used to verify the amplitude compensation model of the pre-stack gathers using the amplitude value data in the forward AVO curve that does not participate in machine learning, and to compare whether the trend of the compensated AVO curve is consistent with the trend of the actual AVO curve; According to the actual seismic gathers compensated by the pre-stack gather amplitude compensation model, the amplitude values ​​at different offsets are extracted at a certain moment on the compensated actual seismic gathers to obtain the compensated AVO curve.

3. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 2, characterized in that: It also includes a three-dimensional gather amplitude compensation module, which is used to calculate the amplitude ratio of the actual seismic gather after compensation to the actual seismic gather, and obtain the amplitude compensation factor of different offset distances; fit the amplitude compensation factor of different offset distances of the target well with the number of coverages of different offset distances, and then use the fitted relationship to convert the number of coverages of different offset distances of other actual seismic gathers within the work area except the target well into the compensation factor, so as to obtain the three-dimensional gather amplitude compensation factor field of the target well, and use the three-dimensional gather amplitude compensation factor field to compensate the three-dimensional gather of the target well.

4. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: The specific method of extracting the actual seismic wavelet is: calculating the reflection coefficient sequence according to the well logging data, and obtaining the actual seismic wavelet through the convolution model according to the reflection coefficient sequence and the seismic trace of the target well point position; The specific calculation formula for calculating the reflection coefficient sequence based on the logging data is as follows: Reflection coefficient Among them, IPn is the wave impedance value of the lower layer of the interface, and IPn-1 is the wave impedance value of the upper layer of the interface.

5. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: The specific method of conducting synthetic gather forward modeling of the target well based on the actual seismic wavelet is: The actual seismic wavelet is combined with the Zobnitz equation to carry out the synthetic gather forward modeling of the target well. Based on the Zobnitz approximation formula, the reflection coefficient of the target well at different offsets is calculated: Wherein, θ is the seismic incident angle; R(θ) is the reflection coefficient corresponding to different incident angles; Vp is the P-wave velocity of the underground stratum; Vs is the S-wave velocity of the underground stratum; ρ is the density of the underground stratum; S(θ) is the seismic trace obtained by convolution of R(θ) and the seismic wavelet when the incident angle is θ; The target well seismic record S(θ) can be expressed as the convolution of the wavelet and the reflection coefficients at different offsets: S(θ)=R(θ)*W Based on the convolution principle, S(θ) is the amplitude value corresponding to different incident angles, R(θ) is the reflection coefficient corresponding to different incident angles, and W is the actual seismic wavelet.

6. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: The specific generation method of the actual AVO curve and the forward AVO curve is: The actual AVO curve is a certain moment on the actual seismic gather, and the offset distances are arranged in ascending order. The actual AVO curve can be determined by displaying the amplitude value of the actual seismic gather. The forward gather is generated by forward modeling of the synthetic gather. The forward AVO curve is a certain moment on the forward gather, and the offset distances are arranged in ascending order. The forward AVO curve can be determined by selecting the amplitude value of the forward gather.

7. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: The specific method of extracting feature data and label data is as follows: The actual seismic gather data are the amplitude value data of the actual AVO curve in the vertical direction with the time depth sequence and in the horizontal direction with the offset size sequence, from which the amplitude value of the actual AVO curve corresponding to the top surface of the mudstone is selected as the characteristic data; the forward modeling gather data are the amplitude value data of the forward modeling AVO curve in the vertical direction with the time depth sequence and in the horizontal direction with the offset size sequence, from which the amplitude value of the forward modeling AVO curve corresponding to the top surface of the mudstone is selected as the label data.

8. The well-seismic combined machine learning pre-stack gather amplitude compensation system according to claim 1, characterized in that: The specific method for obtaining the pre-stack gather amplitude compensation model is: taking the actual seismic gather AVO curve and the forward modeling AVO curve as input, and using the machine learning algorithm to fit the mapping relationship between the two, that is, the machine learning gather compensation model.

9. A method for amplitude compensation of pre-stack gathers by machine learning combined with well-seismic learning, characterized in that: It includes the following steps: Collect well logging formation data, post-stack seismic data and actual seismic gathers in the work area, and obtain actual AVO curves based on the actual seismic gathers; Select the target well according to the well logging formation data and the mudstone section that meets the lithology setting, use the post-stack seismic data to complete the well-seismic calibration of the target well, and extract the actual seismic wavelet of the target well based on the well-seismic calibration; According to the actual seismic wavelet of the target well, the synthetic gather of the target well is forward modeled to obtain the forward modeled AVO curve; The amplitude value data of the actual AVO curve corresponding to the top surface of the mudstone in the target well is extracted as characteristic data; and the amplitude value data of the forward AVO curve corresponding to the top surface of the mudstone in the target well is extracted as label data; The model building module trains the machine learning model according to the feature data and label data to obtain the pre-stack gather amplitude compensation model.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method described in claim 9 are implemented.

Images