Methods and Systems for Predicting the Structure of Braided Delta Sand Bodies under Sparse Well Network Conditions at Sea

Abstract

This invention provides a method and system for predicting the sand body configuration of braided river deltas under conditions of sparse well networks at sea, belonging to the field of oil and gas field development geological research technology. The method includes the following steps: establishing a prototype model of the braided river delta; establishing a database of braided river delta configuration patterns and configuration unit parameters; calculating the sand body thickness of the target layer based on well data; extracting seismic attributes of the target layer and selecting the seismic attributes that can reflect the sand body thickness; establishing a learning model between the sand body thickness interpreted from well logging and the seismic attribute data selected near the well point; performing intelligent well-seismic combined sand body prediction for various types of sand bodies with different overlay relationships; and completing the sand body configuration prediction. The system includes a sand body configuration parameter database module, a data preparation module, a seismic attribute selection module, a machine learning regression model module, a colored inversion module, and a sand body configuration prediction module. This invention improves the interpretation accuracy and reliability and overcomes the problem of multiple solutions in colored inversion.

Description

Technical Field

[0001] This invention belongs to the field of geological research technology for oil and gas field development, and in particular relates to a method and system for predicting the configuration of braided river delta sand bodies under conditions of sparse well networks at sea. Background Technology

[0002] Braided river deltaic sand bodies exhibit extremely complex spatial contact relationships, resulting in highly heterogeneous reservoirs. Sand bodies that appear adjacent and connected in plan are actually separated and not interconnected by sedimentary discontinuities. As research into sand body configurations deepens, the prediction of reservoir sand body configurations becomes increasingly refined. Improving the accuracy of sand body configuration predictions remains a hot topic in geological interpretation.

[0003] Current predictions of braided river delta sandbody configurations are primarily conducted in densely networked onshore well areas, using well logging data as the core and seismic attributes as optimization constraints to achieve reservoir configuration prediction. For high-precision seismic attribute prediction, two approaches are employed: firstly, mathematical tools are used to find new and more reasonable seismic attributes; secondly, different seismic attributes are fused, some selected based on personal experience, others using specific algorithms. The aim is to improve correlation while preserving as much effective seismic information as possible, resulting in reliable and comprehensive seismic attribute fusion for sandbody configuration characteristics.

[0004] Numerous studies have shown that the seismic response of sand body thickness in the target layer is related to seismic frequency. High-frequency information has high resolution but small tuning thickness, making it suitable for predicting thin sand bodies; low-frequency information has large tuning thickness but low resolution, making it suitable for predicting thick sand bodies. When the sand body thickness is greater than the tuning thickness, amplitude and frequency information exhibit significant ambiguity. Furthermore, when seismic data resolution is low, the distribution characteristics of seismic attributes in the target layer are also related to the seismic attributes of the surrounding rock. For example, a high amplitude attribute value with moderate amplitude in the target layer could indicate a sand body of moderate thickness in the target layer, or it could be the response of a thick sand body in an adjacent stratum. Therefore, sand body distribution prediction without considering frequency division and surrounding rock interference often leads to significant errors in the prediction results.

[0005] In seismic inversion practices in densely networked well areas, model-based inversion combined with abundant well data is typically used to achieve a detailed characterization of the target sand body distribution. However, previous methods are difficult to apply to sparsely networked offshore areas. This is partly due to the scarcity of well data in these areas, and partly because numerous studies have shown that braided deltaic facies sand bodies exhibit strong lateral heterogeneity and significant thickness variations among different single sand bodies. Consequently, model-based and frequency-division inversions have struggled to accurately characterize thin sand bodies. Colored inversion, with its minimal involvement of well logging data and faithful reproduction of seismic data, responds well to thin sand bodies and provides continuous inversion results, making it suitable for seismic inversion tasks under sparsely networked well conditions. However, colored inversion results often have multiple interpretations, necessitating guidance based on the actual sand body distribution patterns of the specific work area.

[0006] Previous studies have largely relied on well logging data under dense well networks to predict reservoir sandbody configurations, rarely considering configuration guidance from sedimentary models. This makes accurate sandbody configuration prediction difficult under sparse well networks in offshore environments with limited logging data, resulting in significant ambiguity and an inability to accurately predict reservoir sandbody distribution. Therefore, in addition to comprehensively utilizing various seismic attributes and well logging data, machine learning methods can be employed to consider seismic data from different frequency bands, as well as the seismic attributes of the surrounding rock and the target layer. Furthermore, a sandbody configuration prototype parameter library should be established to filter the sandbody prediction results. This approach can effectively reduce ambiguity and improve accuracy. Summary of the Invention

[0007] The present invention aims to propose a method and system for predicting the configuration of braided river delta sand bodies under conditions of sparse well networks at sea, which can solve the above-mentioned problems.

[0008] To achieve the above objectives, the technical solution of this invention is as follows: a method for predicting the configuration of braided river delta sand bodies under conditions of sparse well networks at sea, comprising the following steps:

[0009] S1: Identify sand bodies and mudstone interlayers in the prototype model area, conduct a comprehensive dissection of the sand body configuration inside the reservoir and analyze the internal characteristics of the configuration units, and then measure the geometric parameter characteristics of the configuration units to establish a prototype model of the braided river delta.

[0010] S2: Using the prototype model of braided river deltas as a guide, and combining the well network analysis data of typical blocks in the study area, establish the configuration model of braided river deltas and the parameter library of braided river delta configuration units in the study area.

[0011] S3: Establish a high-precision stratigraphic-structural framework for the study area based on core, well logging, and seismic data;

[0012] S4: Calculate the thickness of the target layer sand body based on core and logging data;

[0013] S5: Extract the seismic attributes of the target layer in the study area, and select the seismic attributes that can reflect the sand body thickness by analyzing the correlation between the seismic attributes and the sand body thickness interpreted by well logging.

[0014] S6: Using the sand body thickness interpreted by well logging in the target layer as the supervised dataset, a supervised learning method is adopted to conduct machine learning between the sand body thickness interpreted by well logging and the seismic attribute data selected near the well point, thereby establishing a learning model between the two.

[0015] S7: By utilizing the trained learning model, the selected seismic attributes are fused. The fusion of the selected frequency-division seismic attributes is the frequency-division multi-attribute intelligent fusion method, and the result is the frequency-division intelligent fused attribute. The fusion of the seismic attributes of the target layer and the surrounding rock layer is the attribute intelligent fusion method for reducing surrounding rock interference, and the result is the intelligent fused attribute for reducing surrounding rock interference.

[0016] S8: Intelligent well-seismic combined sand body prediction based on frequency division multi-attribute intelligent fusion method, attribute intelligent fusion method to reduce surrounding rock interference and colored inversion method;

[0017] S9: Based on the intelligent well-seismic combination of sand body prediction, the prediction results are constrained according to the established braided river delta configuration unit parameter library, and the most reasonable prediction results are selected to complete the sand body configuration prediction.

[0018] Furthermore, step S1 includes:

[0019] S11: To obtain high-precision morphological characteristics, lithofacies assemblages, and quantitative parameters of braided river delta morphological units through instruments and field investigations;

[0020] S12: Summarize the information in S11 and extract the braided river delta prototype model according to type.

[0021] Furthermore, the braided river delta configuration unit parameter library includes a lithofacies library, a morphological structure library, and a scale library.

[0022] Furthermore, the braided river delta configuration model and braided river delta configuration unit parameter library include lithofacies assemblage, morphological structure, and empirical formulas.

[0023] Furthermore, step S3 includes:

[0024] S31: Time-depth calibration is performed using synthetic seismic records;

[0025] S32: Based on the stratigraphic interpretation data in the seismic data and the stratigraphic data provided by the well logging data, establish the isochronous stratigraphic framework of the study area.

[0026] Furthermore, step S4 includes:

[0027] S41: The method for calculating sand body thickness utilizes the characteristics of spontaneous potential logging curves and gamma logging curves, combined with the lithofacies characteristics provided by core data, to interpret sand bodies in a single well and calculate the sand body thickness of the target layer.

[0028] Furthermore, step S5 includes: performing a correlation analysis between the sand body thickness interpreted from well logging and the seismic attribute data, and selecting seismic attributes with a correlation coefficient higher than 0.5.

[0029] Furthermore, if the selected seismic attribute is a frequency-division seismic attribute, then a frequency-division multi-attribute intelligent fusion method is adopted to fuse multiple frequency-division attributes in the trained supervised learning model; if the selected seismic attribute includes the seismic attribute of the surrounding rock layer, then an attribute intelligent fusion method to reduce surrounding rock interference is adopted to fuse the seismic attributes of the target layer and the surrounding rock layer in the trained supervised learning model.

[0030] Furthermore, the colored inversion method involves performing spectral analysis on the single-well impedance and seismic impedance, fitting the corresponding energy spectrum curves, setting a matching operator in the frequency domain to match the single-well impedance spectrum curve with the seismic impedance spectrum curve, and then returning to the time domain to apply the matching operator to the seismic data for inversion.

[0031] A system for predicting the sand body configuration of braided river deltas under marine sparse well network conditions includes a sand body configuration parameter library module, a data preparation module, a seismic attribute selection module, a machine learning regression model module, a colored inversion module, and a sand body configuration prediction module.

[0032] The data preparation module is used to extract the sand body thickness and seismic attribute data of the target layer from the well logging interpretation. The seismic attribute selection module is used to perform correlation analysis between the sand body thickness and seismic attribute data of the target layer extracted by the data preparation module from the well logging interpretation.

[0033] The machine learning regression model module is used to obtain the trained learning model. The sand body configuration prediction module uses the learning model obtained by the machine learning regression model module as a basis, integrates the earthquake attributes selected by the earthquake attribute selection module, and then combines the colored inversion results of the colored inversion module to preliminarily predict the sand body distribution. Then, according to the prototype parameter library of braided river delta sand body configuration units established by the sand body configuration parameter library module, the prediction results of the sand body configuration prediction module are constrained and the most reasonable prediction results are selected.

[0034] Compared with existing technologies, the method and system for predicting the configuration of braided river delta sand bodies under marine sparse well network conditions described in this invention have the following advantages:

[0035] (1) This invention establishes a parameter library of braided river delta sand body configuration units based on the actual and typical characteristics of various braided river delta facies, which has guiding significance for actual configuration analysis;

[0036] (2) Machine learning algorithms were introduced to intelligently fuse earthquake attributes, and a nonlinear learning model was established, which can more accurately represent the complex relationship between earthquake attributes and geological parameters, and is more objective than human operation.

[0037] (3) There are corresponding prediction methods for different target layers. The interference of the surrounding rock on the seismic properties of the target layer is comprehensively considered, which improves the interpretation accuracy and reliability and has universality in actual production.

[0038] (4) Considering the scarcity of well logging data under sparse well network conditions, a color inversion method with strong objectivity is used to invert seismic data. In order to further reduce the ambiguity of seismic interpretation, this invention also constrains the prediction results by establishing a braided river delta configuration unit parameter library, thus overcoming the problem of ambiguity in color inversion.

[0039] (5) The configuration plan view and cross-sectional view obtained by the present invention are more intuitive and can quantitatively characterize the sand body configuration. Attached Figure Description

[0040] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0041] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0042] Figure 2 This is a schematic diagram of the parameter library for braided river deltaic facies sand body configuration units according to an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the frequency division multi-attribute intelligent fusion method according to an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of the intelligent attribute fusion method for reducing surrounding rock interference as described in an embodiment of the present invention. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0046] Due to the complexity of the deltaic facies deposition process, the contact relationships between sand body configuration units are also extremely complex. Therefore, this invention establishes a parameter library of braided deltaic sand body configuration units using laser scanners, ground-penetrating radar, modern sedimentary satellite imagery, and dense well network analysis data from typical blocks for model guidance. Combined with core, well logging, and seismic data, precise stratigraphic-structural framework delineation is completed under sparse well network conditions. Based on this, machine learning methods are used to establish a nonlinear mapping relationship between sand body thickness and the target layer and surrounding seismic attribute data. The invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0047] This invention provides a method and system for predicting the configuration of braided river delta sand bodies under conditions of sparse well networks at sea. For example... Figure 1 As shown, this study integrates ground-penetrating radar data, laser scanning data, and modern sedimentary satellite imagery, combined with dense well networks and dissection data, to conduct a prototype configuration model study of braided river deltas and establish a parameter library for braided river delta configuration units. This library includes lithofacies, morphological structure, and scale databases. Guided by the establishment of corresponding models based on the braided river delta sandbody configuration parameter library, and using well logging interpretation of sandbody thickness in the target layer as a basis, intelligent attribute fusion and colored inversion are employed to obtain seismic response characteristic maps of different types of sandbody stacking relationships. Under the guidance of the model, predictions of sandbody distribution and sandbody stacking relationships are completed. This embodiment includes the following steps:

[0048] S1: Identify sand bodies and mudstone interlayers in the prototype model area, conduct a comprehensive dissection of the sand body configuration within the reservoir, analyze the internal characteristics of the configuration units, and then measure the geometric parameters of the configuration units to establish a prototype model of the braided river delta. Specifically, high-precision information on the morphological characteristics, lithofacies assemblage, and quantitative scale of various configuration units of the braided river delta is obtained through various instruments and field investigations. Lithofacies refers to the color, composition, structure, and original occurrence of paleontology and sediments contained in the rocks. Due to limitations in core data, lithofacies in this invention mainly refers to the color, composition, and structure of the rocks. Quantitative scale refers to the length, width, and height of the configuration units, as well as the correlation between the above parameters. All the above information is summarized and extracted according to type to form the prototype model of the braided river delta. Figure 2As shown, by acquiring, processing, interpreting, and analyzing the properties of ground-penetrating radar (GPR) data, sand bodies and mudstone interlayers in the prototype model area are identified. The response markers of the GPR profile at the overall structural boundary are also analyzed. A laser scanner can be used to perform a comprehensive dissection of the sand body configuration within the reservoir. Satellite maps can be used to intuitively measure the geometric parameters of the configuration units, thereby establishing a three-dimensional configuration model of the modern braided river delta. Three-dimensional laser data of the braided river delta outcrop profiles are acquired in the field. Indoors, the laser scanning data is preprocessed by stitching, denoising, and data reduction. After encapsulation, repair, and mapping, a digital model of the field outcrop is established. Google Earth software is used to measure the basic data of modern sediments in the braided river delta to conduct a study on the configuration of the prototype model of the braided river delta.

[0049] S2: Guided by a braided river delta prototype model and combined with well network analysis data from typical blocks in the study area (target area), a braided river delta configuration model and a braided river delta configuration unit parameter library were established for the study area. The braided river delta configuration unit parameter library includes a lithofacies library, a morphological structure library, and a scale library. The lithofacies library includes lithofacies types and lithofacies assemblages; the morphological structure library includes individual morphologies and assemblage relationships; and the scale library includes absolute and relative scales. The braided river delta configuration model and the braided river delta configuration unit parameter library contain lithofacies assemblages, morphological structures, and empirical formulas.

[0050] S3: Establish a high-precision stratigraphic-structural framework for the study area based on core, well logging, and seismic data. Specifically, firstly, use synthetic seismic records (or inter-well VSP velocity data) for time-depth calibration, that is, establish a spatial correspondence between seismic data in the time domain and well data in the depth domain. Secondly, based on the stratigraphic interpretation data in the seismic data and the stratification data provided by the well logging data, establish an isochronous stratigraphic framework (i.e., stratified data) for the study area.

[0051] S4: Calculate the sand body thickness of the target layer based on core and logging data; interpret the sand body of a single well by utilizing the characteristics of spontaneous potential logging curves and gamma logging curves, combined with the lithofacies characteristics provided by core data, and calculate the sand body thickness of the target layer.

[0052] S5: Extract the seismic attributes of the target layer in the study area. By analyzing the correlation between the seismic attributes and the sand body thickness interpreted by well logging, select one or more seismic attributes that can reflect the sand body thickness (Pearson correlation coefficient greater than 0.5). That is, perform correlation analysis between the sand body thickness interpreted by well logging and the seismic attribute data, and select the seismic attributes with a correlation coefficient higher than 0.5.

[0053] S6: Using the sand body thickness interpreted from well logging in the target layer as the supervised dataset, a supervised learning method (such as support vector machine, neural network, etc.) is employed to perform machine learning between the sand body thickness data interpreted from well logging and the seismic attribute data selected near the well point, thereby establishing a learning model (non-linear mapping relationship) between the two. The "near the well point" refers to a circular area centered on the well point location. The radius of this circular area can be set according to actual conditions and adjusted appropriately based on the seismic lateral sampling interval, generally taking a diameter of 25m, approximately equal to the lateral sampling interval. Then, the average value of the seismic attributes is calculated. Using the sand body thickness interpreted from well data as supervised data, a supervised learning method is employed to perform supervised learning between the sand body thickness data interpreted from well logging and the selected seismic attributes, establishing a supervised learning model.

[0054] S7: By utilizing the trained learning model, the selected seismic attributes are fused. The fusion of selected frequency-division seismic attributes is the frequency-division multi-attribute intelligent fusion method, resulting in frequency-division intelligent fused attributes. The fusion of seismic attributes of the target layer and the surrounding rock layer is the attribute intelligent fusion method for reducing surrounding rock interference, resulting in intelligent fused attributes for reducing surrounding rock interference. Specifically, if the selected seismic attributes are frequency-division seismic attributes, the frequency-division multi-attribute intelligent fusion method is used to fuse multiple frequency-division attributes in the trained supervised learning model. If the selected seismic attributes include the seismic attributes of the surrounding rock layer, the attribute intelligent fusion method for reducing surrounding rock interference is used to fuse the seismic attributes of the target layer and the surrounding rock layer in the trained supervised learning model.

[0055] S8: Intelligent well-seismic combined sand body prediction is performed on various types of sand bodies with different superposition relationships based on frequency-division multi-attribute intelligent fusion method, attribute intelligent fusion method to reduce surrounding rock interference, and colored inversion method. The intelligent fusion method includes frequency-division multi-attribute intelligent fusion method and attribute intelligent fusion method to reduce surrounding rock interference, and the appropriate intelligent fusion method is selected according to the differences in development patterns of different stratigraphic positions.

[0056] The frequency-division multi-attribute intelligent fusion method selects seismic data volumes of different frequencies for sand bodies of different thicknesses, optimizes amplitude and frequency-based seismic attributes, and uses machine learning algorithms for fusion. It is suitable for target layers with large differences in sand body development scale.

[0057] The aforementioned intelligent attribute fusion method for reducing surrounding rock interference extracts amplitude and frequency seismic attributes of the target layer and its adjacent layers above and below, and fuses them using machine learning algorithms to reduce surrounding rock interference and decrease the uncertainty in seismic attribute interpretation. It is applicable to target layers where adjacent layers have sand bodies but lack stable mudstone interlayers.

[0058] The colored inversion method mainly involves performing spectral analysis on the impedance of single well waves and seismic waves, fitting the corresponding energy spectrum curves, setting a matching operator in the frequency domain to match the impedance spectrum curve of single well waves with the impedance spectrum curve of seismic waves, and then returning to the time domain to apply the matching operator to the seismic data to complete the colored inversion.

[0059] For target stratigraphic horizons with significant differences in sand body development scale, a frequency-division multi-attribute intelligent fusion method is employed to fuse the optimized frequency-division seismic attributes, such as... Figure 3 As shown, firstly, wavelet frequency division technology is used to decompose the original data volume into frequency-divided data volumes with different center frequency bands. Based on the sand body thickness distribution range of the target layer, low-frequency, medium-frequency, and high-frequency data volumes are selected, with high frequencies suitable for thin layers and low frequencies suitable for thick layers. Various seismic attributes, such as amplitude, frequency, and phase, commonly found in the target layer are extracted from the original seismic data volume. Correlation analysis is performed between these attributes and sand body thickness, selecting one or more attributes with a high correlation to sand body thickness (correlation coefficient greater than 0.5). The average value of each attribute around each well is set as the training dataset, and the sand body thickness interpreted from well logging in the target layer is set as the supervision dataset. The support vector machine (SVM) algorithm is selected to perform machine learning between the training and supervision datasets, establishing a nonlinear regression model between well logging interpreted sand thickness and fused attributes. Finally, the trained regression model is applied to obtain a fused attribute map that quantitatively represents the planar distribution of sand body thickness in the target layer.

[0060] For target strata where adjacent layers contain sand bodies but lack stable mudstone interlayers, an intelligent attribute fusion method to reduce surrounding rock interference is employed to fuse the seismic attributes of the target strata with those of the surrounding rock layers. Figure 4 As shown, firstly, various seismic attributes, such as amplitude, frequency, and phase, commonly found in the target layer, are extracted from the original seismic data volume. Correlation analysis is performed on these attributes with sand body thickness, and one or more attributes with a high correlation to sand body thickness (correlation coefficient greater than 0.5) are selected. Using a 1 / 4 wavelength time window, 2-3 preferred seismic attributes of the surrounding rocks and the target layer are extracted in the form of stratigraphic slices along the top and bottom interfaces of the target layer, respectively. These are set as the training dataset, and the sand body thickness interpreted from well logging in the target layer is set as the supervision dataset. The support vector machine (SVM) algorithm is selected to perform machine learning between the training and supervision datasets to establish a nonlinear regression model between well logging interpreted sand thickness and fused attributes. Finally, the trained regression model is applied to fuse the seismic attributes of the surrounding rocks and the target layer into a new attribute that can quantitatively reflect the planar distribution of sand body thickness in the target layer.

[0061] In this example, the machine learning algorithm used is Support Vector Machine (SVM). The kernel function uses radial basis function. The input training data can be represented as (xi, yi), where i = 1, 2, 3...n, n is the number of training samples, xi belongs to the input data (seismic attributes), denoted as xi∈R*; yi is the target data (sand body thickness), denoted as yi∈R.

[0062] The expression for the support vector machine regression model is as follows:

[0063] f(x) =<w,x> +b

[0064] Where w, x∈R*, b∈R, f(x) is the output after running the program, and w is the weight value that changes as the gradient decreases.<w,x> Let w represent the dot product of w and x, and b be a constant adjustment factor.

[0065] In addition, genetic neural networks and deep learning algorithms can also be used as machine learning algorithms.

[0066] The reliability of the obtained regression model is evaluated as follows: if the accuracy of the test set is greater than or equal to 80%, the model is accepted; otherwise, the above steps are repeated to adjust the initial parameters until the accuracy of the test set is greater than or equal to 80%, and then the model is output.

[0067] Because there is limited well data under sparse well network conditions at sea, model-based inversion and frequency division inversion are not applicable. This example uses a colored inversion method, which can overcome the shortcomings of sparse well networks and more realistically reflect the distribution characteristics of sand bodies in the profile.

[0068] S9: Based on intelligent well-seismic combined sand body prediction, the prediction results are constrained according to the established braided river delta facies sand body configuration parameter library, and the most reasonable prediction results are selected to complete the sand body configuration prediction. The constraint of the braided river delta facies sand body configuration parameter library is to select prediction results that meet both the parameter library requirements and the actual situation by screening the intelligent attribute fusion results and the colored inversion results through the established configuration prototype parameter library.

[0069] A system for predicting the sand body configuration of braided river deltas under marine sparse well network conditions includes a sand body configuration parameter library module, a data preparation module, a seismic attribute selection module, a machine learning regression model module, a colored inversion module, and a sand body configuration prediction module.

[0070] The sand body configuration parameter library module is a knowledge base of quantitative parameters of braided river delta facies characteristics built in various ways, used for model guidance;

[0071] The data preparation module is used to extract the sand body thickness and seismic attribute data of the target layer from the well logging interpretation. The seismic attribute selection module is used to perform correlation analysis between the sand body thickness and seismic attribute data of the target layer extracted by the data preparation module and select seismic attributes with a correlation coefficient higher than 0.5.

[0072] The machine learning regression model module uses the sand body thickness as the target dataset, the seismic attribute values ​​of the target layer and the adjacent upper and lower layers as the training sample dataset, and then uses both the target data and the training data as input data. A nonlinear mapping is established through machine learning algorithms to obtain the trained learning model.

[0073] The inversion module performs spectral analysis on the single-well impedance and seismic impedance, fits the corresponding energy spectrum curves, sets a matching operator in the frequency domain to match the single-well impedance spectrum curve with the seismic impedance spectrum curve, and then returns to the time domain to apply the matching operator to the seismic data to complete the colored inversion.

[0074] The sand body configuration prediction module uses the learning model obtained by the machine learning regression model module as a basis, integrates the seismic attributes selected by the seismic attribute selection module, and then combines the colored inversion results of the colored inversion module to preliminarily predict the sand body distribution. Then, based on the prototype parameter library of braided river delta sand body configuration units established by the sand body configuration parameter library module, the prediction results of the sand body configuration prediction module are constrained, the distribution characteristics and superposition relationships of the target layer sand body profile are analyzed, and the most reasonable prediction results that best represent the actual sand body distribution characteristics are selected, thereby completing the sand body configuration prediction.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the configuration of braided river delta sand bodies under conditions of sparse well networks at sea, characterized in that, Includes the following steps: S1: Identify sand bodies and mudstone interlayers in the prototype model area, conduct a comprehensive dissection of the sand body configuration inside the reservoir and analyze the internal characteristics of the configuration units, and then measure the geometric parameter characteristics of the configuration units to establish a prototype model of the braided river delta. S2: Using the prototype model of braided river deltas as a guide, and combining the well network analysis data of typical blocks in the study area, establish the configuration model of braided river deltas and the parameter library of braided river delta configuration units in the study area. S3: Establish a high-precision stratigraphic-structural framework for the study area based on core, well logging, and seismic data; S4: Calculate the thickness of the target layer sand body based on core and logging data; S5: Extract the seismic attributes of the target layer in the study area, and select the seismic attributes that can reflect the thickness of the sand body by analyzing the correlation between the seismic attributes and the sand body thickness interpreted by well logging. S6: Using the sand body thickness interpreted by well logging in the target layer as the supervised dataset, a supervised learning method is adopted to conduct machine learning between the sand body thickness interpreted by well logging and the seismic attribute data selected near the well point, thereby establishing a learning model between the two. S7: By utilizing the trained learning model, the selected seismic attributes are fused. The fusion of selected frequency-division seismic attributes is the frequency-division multi-attribute intelligent fusion method, and the result is the frequency-division intelligent fused attribute. The fusion of the seismic attributes of the target layer and the surrounding rock layer is the attribute intelligent fusion method for reducing surrounding rock interference, and the result is the intelligent fused attribute for reducing surrounding rock interference. S8: Intelligent well-seismic combined sand body prediction based on frequency division multi-attribute intelligent fusion method, attribute intelligent fusion method to reduce surrounding rock interference and colored inversion method; S9: Based on the intelligent well-seismic combination of sand body prediction, the prediction results are constrained according to the established braided river delta configuration unit parameter library, and the most reasonable prediction results are selected to complete the sand body configuration prediction.

2. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that... Step S1 includes: S11: To obtain high-precision morphological characteristics, lithofacies assemblages, and quantitative parameters of braided river delta morphological units through instruments and field investigations; S12: Summarize the information in S11 and extract the braided river delta prototype model according to type.

3. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that: The braided river delta configuration unit parameter library includes a lithofacies library, a morphological structure library, and a scale library.

4. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that: The braided river delta configuration model and braided river delta configuration unit parameter library include lithofacies assemblage, morphological structure and empirical formulas.

5. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that... Step S3 includes: S31: Time-depth calibration is performed using synthetic seismic records; S32: Based on the stratigraphic interpretation data in the seismic data and the stratigraphic data provided by the well logging data, establish the isochronous stratigraphic framework of the study area.

6. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that... Step S4 includes: S41: The method for calculating sand body thickness utilizes the characteristics of spontaneous potential logging curves and gamma logging curves, combined with the lithofacies characteristics provided by core data, to interpret sand bodies in a single well and calculate the sand body thickness of the target layer.

7. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that... Step S5 includes: performing a correlation analysis between the sand body thickness interpreted from well logging and the seismic attribute data, and selecting seismic attributes with a correlation coefficient higher than 0.

5.

8. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that: If the selected earthquake attribute is a frequency-division earthquake attribute, then the frequency-division multi-attribute intelligent fusion method is adopted to fuse multiple frequency-division attributes in the trained supervised learning model; If the selected seismic attributes include those of the surrounding rock layer, an intelligent attribute fusion method that reduces interference from the surrounding rock layer is adopted to fuse the seismic attributes of the target layer and the surrounding rock layer in the trained supervised learning model.

9. The method for predicting the sand body configuration of braided river deltas under sparse well network conditions at sea, as described in claim 1, is characterized in that: The described inversion method involves performing spectral analysis on the impedance of single-well waves and seismic waves, fitting the corresponding energy spectrum curves, setting a matching operator in the frequency domain to match the impedance spectrum curves of single-well waves and seismic waves, and then returning to the time domain to apply the matching operator to the seismic data for inversion.

10. A system for predicting the configuration of braided delta sand bodies under marine sparse well network conditions using the method described in any one of claims 1-9, characterized in that: It includes a sand body configuration parameter library module, a data preparation module, a seismic attribute selection module, a machine learning regression model module, a colored inversion module, and a sand body configuration prediction module; The sand body configuration parameter library module is a knowledge base of quantitative parameters of braided river delta facies built in various ways, used to establish model guidance; the data preparation module is used to extract the sand body thickness and seismic attribute data of the target layer from the well logging interpretation; the seismic attribute selection module is used to perform correlation analysis between the sand body thickness and seismic attribute data of the target layer extracted by the data preparation module from the well logging interpretation. The machine learning regression model module is used to obtain the trained learning model. The sand body configuration prediction module uses the learning model obtained by the machine learning regression model module as a basis, integrates the seismic attributes selected by the seismic attribute selection module, and then combines the colored inversion results of the colored inversion module to preliminarily predict the sand body distribution. Then, according to the prototype parameter library of braided river delta sand body configuration units established by the sand body configuration parameter library module, the prediction results of the sand body configuration prediction module are constrained and the most reasonable prediction results are selected.

Images