A time-frequency spectrum sedimentary facies sample establishment method based on seismic forward modeling

Abstract

This invention discloses a method for establishing time-spectral sedimentary facies samples based on seismic forward modeling. The method includes: Step S1: Constructing typical sedimentary models and various sedimentary facies property templates for the actual work area based on core sedimentary characteristics; Step S2: Obtaining sedimentary facies geological profiles based on typical sedimentary models and sedimentary facies property templates; Step S3: Extracting actual seismic forward modeling parameters from actual seismic data; Step S4: Conducting forward modeling of the sedimentary facies geological model based on the actual seismic parameters; Step S5: Converting the forward modeled seismic data into time-spectral data through continuous wavelet transform; Step S6: Extracting typical seismic traces from the forward modeled seismic profiles, using the sedimentary facies in the geological model as labels and the time-spectral data as feature values ​​to establish sedimentary facies seismic samples. A large number of high-precision samples are generated under the guidance of typical models, and the sample labels are completely accurate, eliminating errors caused by time-depth calibration, thus meeting the needs of machine learning for sedimentary facies prediction.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration technology, and in particular to a method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling. Background Technology

[0002] As exploration progresses in major oilfields, target geological bodies are becoming increasingly characterized by being "deep, fragmented, small, and thin." In particular, the understanding of sedimentary facies in deep and ultra-deep targets faces challenges such as large target layer burial depths, low quality of original seismic data, and a limited number of drilled wells. This makes it extremely difficult to understand the distribution patterns of sedimentary facies and to perform 3D modeling of sedimentary facies. Currently, establishing a comprehensive sedimentary facies seismic sample library and using machine learning methods to integrate seismic, well logging, and forward modeling data to predict sedimentary facies boundaries is an important approach to solving the sedimentary facies analysis problem in areas with few or no wells.

[0003] Chinese patent application CN202210287258.9 relates to a method for creating machine learning samples based on well data preprocessing. This method includes: firstly, filtering and smoothing the logging curve to obtain the curve's baseline; then, subtracting the baseline from the original curve to obtain the amplitude difference Δt at each point. While ensuring the quality of the original data, outliers are removed, and a threshold is set to adjust or normalize the gain of Δt, resulting in logging curves with consistent baselines, thus achieving accurate machine learning samples. The essence of logging data standardization is to utilize the similar geological and geophysical characteristics of the same layer within the same oilfield or region, defining a similar distribution pattern for logging data. Therefore, once a standard distribution model for various types of logging data is established, a comprehensive analysis of logging data from each well in the oilfield can be performed, correcting inaccuracies in the calibration and achieving standardization of logging data across the entire oilfield. Currently, this calculation method not only effectively addresses the problem of indistinct target segment features in logging curves but also serves as an effective algorithm system for creating machine learning samples.

[0004] Chinese patent application CN202010681152.8 discloses a machine learning-based well test interpretation method and system. The method includes: obtaining reservoir parameters under different influencing factors at the initial well test time based on a constructed well test physical model, forming an initial parameter set; obtaining a well test sample data set based on the execution time and the initial parameter set; inputting the well test sample data set into a corresponding numerical simulator to obtain a predicted data set for the target well test time; then, combining the obtained pressure derivative observation data and the predicted data set, automatically performing data fitting using machine learning methods, and generating parameter interpretation results using the updated predicted data set. This invention addresses the shortcomings of existing technologies in terms of accuracy and practicality when applied to parameter interpretation. It achieves parameter interpretation for multiphase flow well tests in complex oil and gas reservoirs through automatic data fitting, with high execution efficiency, providing reliable guidance for well test production prediction and specifying well test construction plans.

[0005] Chinese patent application CN202211191867.0 discloses a geostatistical pattern recognition method based on Bayesian ensemble learning machine. The method includes: establishing a numerical model of multiphase flow transport of organic pollutants in groundwater; determining the pollution source characteristics, pollutant transport parameters, and prior distribution characteristics of each variable that contribute significantly to the spatiotemporal distribution of pollutants in the model; preparing a training sample set for the numerical model's input-output; identifying and approximating the input-output nonlinear mapping relationship of the multiphase flow numerical model using different single machine learning methods; establishing a Bayesian multi-objective nonlinear programming optimization model for the key parameters and combined weights of each machine learning model; solving the optimization model using particle swarm optimization algorithm; identifying the optimal model parameters and combined weights; and constructing a Bayesian ensemble learning intelligent pattern recognition method for the numerical model. This invention significantly improves the computational efficiency of pollutant transport simulation and prediction.

[0006] In the data processing stage, to obtain more machine learning samples, the commonly used method at present is well data preprocessing-based machine learning sample creation. This type of method combines well logging and seismic data by performing time-depth matching and calibration on actual drilled wells to extract corresponding samples. Its advantage lies in its ease of implementation, but its disadvantages are also obvious. In areas with few wells, the sample extraction is insufficient, ultimately affecting the accuracy of the results. Furthermore, it cannot be applied to data from un-drilled areas, wasting a significant amount of valuable seismic information.

[0007] Current technologies rarely involve the construction of sedimentary facies samples. In practice, the most commonly used method is based on time-depth calibration, using well logging facies interpretation from actual wells as labels and well-side seismic traces as feature values. However, this method has two main problems: 1. In areas with few wells, the number of actual wells is small, and the number of sedimentary facies samples that can be constructed is insufficient to meet the needs of machine learning; 2. Seismic data may contain problems such as "heterogeneous homogeneous facies" or "homogeneous heterogeneous facies," making it difficult to accurately reflect sedimentary facies characteristics using only the amplitude information from seismic well-side traces, and also making it difficult to meet the sample accuracy requirements of machine learning; 3. When establishing the correlation between well logging facies labels and well-side seismic feature values ​​through time-depth calibration in actual seismic data, calibration errors may occur, resulting in incorrect correspondence between labels and feature values.

[0008] Therefore, it is crucial to effectively utilize existing mature geological knowledge and seismic data within the work area, and to extract machine learning samples from areas with few or no wells. New methods for establishing machine learning sample sets are urgently needed. Summary of the Invention

[0009] In view of the above problems, the present invention is proposed to provide a method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling that overcomes or at least partially solves the above problems.

[0010] According to one aspect of the present invention, a method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling is provided, the sample establishment method comprising:

[0011] Step S1: Construct typical sedimentary models and templates for various sedimentary facies properties in the actual work area based on core sedimentary characteristics;

[0012] Step S2: Obtain sedimentary facies geological profiles based on typical sedimentary models and sedimentary facies property templates;

[0013] Step S3: Extract actual earthquake forward modeling parameters from actual earthquake data;

[0014] Step S4: Conduct forward modeling of sedimentary facies geological models based on actual seismic parameters;

[0015] Step S5: Convert the forward-modeled seismic data into time-spectrum data using continuous wavelet transform;

[0016] Step S6: Extract typical seismic traces from the forward seismic profile, and establish sedimentary facies seismic samples using sedimentary facies in the geological model as labels and time-spectrum data as feature values.

[0017] Optionally, step S1: constructing typical sedimentary models and multiple sedimentary facies property templates for the actual work area based on core sedimentary characteristics specifically includes:

[0018] The sedimentary facies type of the study area was determined based on the sedimentary characteristics of the core samples.

[0019] Based on single-well facies, multi-well facies, and planar facies, typical sedimentary facies models are established;

[0020] Templates for sedimentary facies velocity and density properties were constructed based on well logging data.

[0021] Optionally, step S2: obtaining a sedimentary facies geological profile based on a typical sedimentary model and sedimentary facies property template specifically includes:

[0022] Based on typical sedimentary models and sedimentary facies property templates, sedimentary facies geology is carried out, physical property parameter profiles are constructed, and physical property parameter model conversion is performed.

[0023] Optionally, the construction of the physical property parameter profile and the transformation of the physical property parameter model specifically include:

[0024] Based on the sedimentary model determined in step S1, a sedimentary facies geological model is drawn by changing the contact relationship between sand bodies or between sand bodies and mudstone, or by changing the thickness of sand bodies and mudstone.

[0025] The geological information contained in the profile is sedimentary facies constructed from different sand and mud combinations;

[0026] In geological profiles, based on templates of various sedimentary facies physical properties, velocity and density parameters of various sedimentary facies are filled in, transforming the sedimentary facies geological model into a sedimentary facies physical property parameter model.

[0027] Optionally, the actual seismic forward modeling parameters include the dominant frequency, bandwidth, signal-to-noise ratio, and actual wavelet.

[0028] Optionally, step S3: extracting actual seismic forward modeling parameters from actual seismic data specifically includes:

[0029] Matched filtering, wavelet transform and other techniques are used to extract actual wavelet parameters such as dominant frequency, wavelength and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling.

[0030] The signal-to-noise ratio (SNR) parameter of actual seismic data is estimated using techniques such as stacking and time-domain SVD decomposition, and used as parameters for forward modeling. The time-domain SVD decomposition is expressed as follows:

[0031] P = USV T

[0032] Where, the column vector of U is represented as XX T The column vector of V is represented by X. T X; singular values ​​are non-zero elements, mainly represented by the main diagonal of the diagonal matrix S. The seismic signal X is represented as a matrix:

[0033]

[0034] Where, x nm This is the element in the nth row and mth column, and so on.

[0035] The sedimentary facies geological model was converted from the depth domain to the time domain using sedimentary facies physical property parameter models.

[0036] Optionally, step S4: conducting forward modeling of the sedimentary facies geological model based on actual seismic parameters specifically includes:

[0037] Using the extracted actual seismic wavelet, actual seismic dominant frequency, and actual seismic signal-to-noise ratio forward modeling parameters, seismic forward modeling was carried out on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, and sedimentary facies forward modeling seismic profiles were obtained.

[0038] Optionally, step S5: converting forward-modeled seismic data into time-spectrum data through continuous wavelet transform specifically includes:

[0039] The seismic waveforms of forward-modeled seismic data are transformed into two-dimensional time-spectrum maps using continuous wavelet transform (CWT).

[0040] The extracted time-spectrum diagrams of different sedimentary facies were analyzed to extract typical spectral characteristics of different sedimentary facies, and a spectral characteristic chart of different sedimentary facies in the actual study area was constructed.

[0041] Optionally, the step of converting the seismic waveform of forward-modeled seismic data into a two-dimensional time-spectrum map using continuous wavelet transform (CWT) specifically includes:

[0042] The continuous wavelet transform of a seismic signal is represented as follows:

[0043]

[0044] in, Let f(t) be the basic wavelet or mother wavelet, a be the scaling factor, b be the translation factor, and f(t) be the travel time length function.

[0045] The wavelet basis used in Continuous Wavelet Transform (CWT) is the Morlet wavelet, and the expression for the Morlet wavelet basis function is as follows:

[0046]

[0047] in, τ is the conjugate of the basic wavelet or the mother wavelet, a is the scale parameter, τ is the displacement parameter, and f(t) is the travel time length function.

[0048] Optionally, step S6: extracting typical seismic traces from the forward-modeled seismic profile, establishing sedimentary facies seismic samples using sedimentary facies in the geological model as labels and time-spectrum data as feature values ​​specifically includes:

[0049] Using the interpretation of the stratigraphic development model of the target layer in the actual work area and the existing understanding of stratigraphic contact relationships, typical seismic traces are extracted from all forward seismic profiles.

[0050] The typical seismic traces extracted are used as the basis for sample extraction. The sedimentary facies information in the forward geological model is used as the label, and the two-dimensional time-spectrum information of the seismic waves transformed by continuous wavelet transform is used as the feature value to establish a sedimentary facies seismic sample dataset.

[0051] This invention provides a method for establishing time-spectral sedimentary facies samples based on seismic forward modeling. The method includes: Step S1: Constructing typical sedimentary models and various sedimentary facies property templates for the actual work area based on core sedimentary characteristics; Step S2: Obtaining sedimentary facies geological profiles based on typical sedimentary models and sedimentary facies property templates; Step S3: Extracting actual seismic forward modeling parameters from actual seismic data; Step S4: Conducting forward modeling of the sedimentary facies geological model based on the actual seismic parameters; Step S5: Converting the forward modeled seismic data into time-spectral data through continuous wavelet transform; Step S6: Extracting typical seismic traces from the forward modeled seismic profiles, using the sedimentary facies in the geological model as labels and the time-spectral data as feature values ​​to establish sedimentary facies seismic samples. A large number of high-precision samples are generated under the guidance of typical models, and the sample labels are completely accurate, eliminating errors caused by time-depth calibration, thus meeting the needs of machine learning for sedimentary facies prediction.

[0052] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart illustrating a method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling, provided in an embodiment of the present invention;

[0055] Figure 2 This is a sedimentary geological profile in a specific embodiment of the present invention;

[0056] Figure 3 This is a seismic forward modeling profile in a specific embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of earthquake forward model sample extraction in a specific embodiment of the present invention. Detailed Implementation

[0058] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0059] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0060] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0061] Example 1

[0062] Based on core and regional sedimentary background studies, this invention constructs typical sedimentary models through single-well-to-well facies analysis and obtains a large number of sedimentary facies geological profiles based on these models. Simultaneously, it extracts actual dominant frequency, bandwidth, signal-to-noise ratio, and actual wavelet from actual seismic data, achieving forward modeling data that approximates the characteristics of actual seismic events, thereby improving the accuracy of sedimentary facies samples. Furthermore, it converts one-dimensional seismic traces into two-dimensional time-spectral data through continuous wavelet transform, reducing the ambiguity of seismic-based sedimentary facies analysis by expanding the dimensionality of sample information and further improving sample accuracy. Finally, using sedimentary facies in the geological model as labels and two-dimensional time-spectral information as feature values, a large number of high-precision sedimentary facies seismic samples are established, thus laying a solid foundation for machine learning-based sedimentary facies prediction. This method includes the following steps:

[0063] 1. Construction of typical sedimentary models and physical property templates for different sedimentary facies in actual work areas based on core sedimentary characteristics. The construction of typical sedimentary models is mainly achieved through the following steps:

[0064] (1) Determine the sedimentary facies type of the study area based on the core sedimentary characteristics;

[0065] (2) Establish typical sedimentary facies models based on single-well facies, interconnected-well facies, and planar facies;

[0066] (3) Construct templates for sedimentary facies velocity, density and other physical properties based on well logging data.

[0067] 2. Based on typical sedimentary models and physical property templates, construct numerous sedimentary facies geological and physical property parameter profiles and perform physical property parameter model conversion. The profile construction and physical property parameter model conversion can be achieved through the following steps:

[0068] (1) Based on the sedimentary model determined in step 1, a large number of sedimentary facies geological models are drawn by changing the contact relationship between sand bodies or between sand bodies and mudstone, or by changing the thickness of sand bodies and mudstone. It is worth noting that the geological information contained in this profile is sedimentary facies constructed by different sand-mud combinations.

[0069] (2) In the geological profile, based on the template of different sedimentary facies physical property parameters, the velocity and density parameters of different sedimentary facies are filled in, so as to transform the sedimentary facies geological model into a sedimentary facies physical property parameter model.

[0070] 3. Collect forward modeling parameters such as dominant frequency, signal-to-noise ratio, and actual wavelet from actual seismic data. The extraction of forward modeling parameters is mainly achieved through the following steps:

[0071] (1) Use matched filtering, wavelet transform and other techniques to extract actual wavelet parameters such as main frequency, wavelength and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling.

[0072] (2) The signal-to-noise ratio parameters of actual seismic data are estimated using techniques such as the stacking method and time-domain SVD decomposition, and used as parameters for forward modeling. The time-domain SVD decomposition can be expressed as:

[0073] P = USV T

[0074] Here, the column vector of U can be expressed as XX T The column vector of V can be expressed as X T X; singular values ​​are non-zero elements, mainly represented by the main diagonal of the diagonal matrix S. The seismic signal X can be expressed as a matrix:

[0075]

[0076] Where, x nm This is the element in the nth row and mth column, and so on.

[0077] (3) Use sedimentary facies physical property parameter model to convert the drawn sedimentary facies geological model from the depth domain to the time domain.

[0078] 4. Conduct forward modeling of sedimentary facies geological models based on actual seismic parameters. Using the forward modeling parameters extracted in step 3, such as actual seismic wavelet, actual seismic dominant frequency, and actual seismic signal-to-noise ratio, conduct seismic forward modeling on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, and obtain a large number of sedimentary facies forward modeling seismic profiles.

[0079] 5. Convert forward-modeled seismic data into time-spectrum data using continuous wavelet transform to suppress waveform ambiguity. The conversion of forward-modeled seismic traces into time-spectrum data is mainly achieved through the following steps:

[0080] (1) In order to suppress the ambiguity that often exists when using information such as seismic waveforms and amplitudes to distinguish different sedimentary facies and different lithologies, such as heteromorphism, this patent transforms the seismic waveforms of forward modeling seismic data into two-dimensional time spectrum diagrams by using continuous wavelet transform (CWT), thereby effectively reducing the ambiguity of seismic waveforms by expanding the information dimension.

[0081] The continuous wavelet transform of a seismic signal is represented as follows:

[0082]

[0083] in, Let f(t) be the basic wavelet or mother wavelet, a be the scaling factor, b be the translation factor, and f(t) be the travel time length function.

[0084] (2) The wavelet basis used in the Continuous Wavelet Transform (CWT) in this patent is the Morlet wavelet. The following is the expression of the Morlet wavelet basis function used in this application:

[0085]

[0086] in, τ is the conjugate of the basic wavelet or the mother wavelet, a is the scale parameter, τ is the displacement parameter, and f(t) is the travel time length function.

[0087] (3) Analyze the extracted time-spectrum diagrams of different sedimentary facies, extract the typical spectral characteristics of different sedimentary facies, and construct a spectral characteristic chart of different sedimentary facies in the actual study area.

[0088] 6. Extract typical seismic traces from the forward-modeled seismic profiles, and establish sedimentary facies seismic samples using sedimentary facies information from the geological model as labels and time-spectrum information obtained from forward-modeled seismic models as feature values. The establishment of sedimentary facies seismic samples is specifically achieved through the following steps:

[0089] (1) Using the interpretation of the stratigraphic development model of the target layer in the actual work area and the existing understanding of stratigraphic contact relationships, typical seismic traces are extracted from all forward seismic profiles.

[0090] (2) The extracted typical seismic traces are used as the basis for sample extraction. The sedimentary facies information in the forward geological model is used as the label, and the two-dimensional time-frequency information of the seismic waves transformed by continuous wavelet transform is used as the feature value to establish a sedimentary facies seismic sample dataset.

[0091] By employing rigorous quality control methods, such as guidance from typical sedimentary models and control of actual seismic parameters, a large amount of forward modeling data that closely approximates actual seismic characteristics has been acquired, effectively solving the problem of the number of sedimentary facies samples.

[0092] By using continuous wavelet transform, the information dimension of one-dimensional seismic data is expanded to two-dimensional time-spectral data. The spectral characteristics of different sedimentary facies are significantly different, which effectively suppresses the ambiguity of sedimentary facies analysis based on seismic waveforms and solves the problem of sedimentary facies sample accuracy.

[0093] This invention provides a method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling continuous wavelet transform. It is mainly used for constructing sedimentary facies seismic samples in oil exploration and lays a large-scale, high-precision sample foundation for three-dimensional sedimentary facies prediction based on machine learning.

[0094] Example 2

[0095] In a specific embodiment 2 of the present invention, such as Figure 1 As shown, Figure 1 The flowchart below shows a method for establishing a machine learning sample set for seismic forward modeling according to the present invention. In this embodiment, it includes the following six steps and related parameter settings:

[0096] The first step involved constructing typical sedimentary models and physical property templates for different sedimentary facies in the actual work area based on the sedimentary characteristics of the core samples. This case study uses 3D seismic data from the central Junggar Basin in western China as an example. Through analysis of typical lithofacies combinations from five drilled wells in the work area, analysis of their logging curves, and observation of core samples, the single-well and interconnected well sedimentary facies of the study area were clarified. The typical sedimentary model of the study area was also identified: the first section of the Karamay Formation is a shallow-water braided river delta, while the second and third sections are predominantly littoral-latitude facies. The 3D work area covers 413 km². 2 With only 5 wells drilled and a small number of actual wells in the study area, it is necessary to use the machine learning sample set establishment method of seismic forward modeling to provide a large sample base for machine learning.

[0097] The second step involved constructing numerous sedimentary facies geological and physical property profiles and converting physical property models based on typical sedimentary models and property templates. In this study area, sand body contact relationships were primarily surface and conformable, with fewer point contacts. The sandstone thicknesses for different sedimentary facies ranged from 2-6 meters for littoral-shallow lacustrine facies and 12-25 meters for shallow-water braided river deltas. Through parameter transformation, a total of 15 sedimentary facies geological profiles and 15 physical property profiles were constructed, each 3000 meters long.

[0098] The third step is to collect forward modeling parameters such as dominant frequency, signal-to-noise ratio, and actual wavelet from actual seismic data.

[0099] Matched filtering and wavelet transform techniques are used to extract actual wavelet parameters such as dominant frequency, wavelength, and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling. The signal-to-noise ratio (SNR) parameters of the actual seismic data are estimated using techniques such as superposition and time-domain SVD, which are then used as parameters for forward modeling. In this case, the dominant frequency of the seismic data is 20Hz, the actual wavelet is an approximate Ricker wavelet that is not completely zero-phase, the sampling rate is 2ms, and the actual seismic SNR is 6. Based on this, the sedimentary facies physical property parameter model obtained in step 1 is used to transform the drawn sedimentary facies geological model from the depth domain to the time domain, laying the foundation for parameter settings in subsequent forward modeling.

[0100] The fourth step involves using the forward modeling parameters extracted in step 3, such as the actual seismic wavelet, the actual seismic dominant frequency, and the actual seismic signal-to-noise ratio, to perform forward modeling on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, thereby obtaining a large number of sedimentary facies forward modeled seismic profiles. In this case, a total of 15 seismic profiles were obtained through forward modeling, with a trace spacing of 5 meters, and each profile contains 600 seismic traces.

[0101] The fifth step is to convert the forward modeling seismic data into time-spectrum data through continuous wavelet transform to suppress waveform ambiguity.

[0102] To mitigate the ambiguity often present when using seismic waveforms and amplitudes to distinguish different sedimentary facies and lithologies, such as isomorphism, this patent utilizes Continuous Wavelet Transform (CWT) to convert the seismic waveforms of forward-modeled seismic data into two-dimensional time-frequency spectra. By expanding the information dimension, the ambiguity of seismic waveforms is effectively reduced. The wavelet basis used in this CWT is the Morlet wavelet. In this case, by analyzing the extracted time-frequency spectra of different sedimentary facies, it is found that the larger the grain size and thickness of the sedimentary facies, the higher the mid-to-high frequency energy, while the finer the grain size and the smaller the thickness, the weaker the mid-to-high frequency energy and the stronger the low-frequency energy. At the same time, the characteristics of samples of the same lithology collected from different locations show good consistency in the spectra.

[0103] The sixth step is to extract typical seismic traces from the forward seismic profiles, and establish sedimentary facies seismic samples using sedimentary facies information from the geological model as labels and time-spectrum information obtained from forward seismic modeling as feature values.

[0104] Utilizing the interpreted stratigraphic development patterns of the target strata in the actual work area and existing understanding of stratigraphic contact relationships, typical seismic traces were extracted from all forward-modeled seismic profiles. In this case, three strata—T3b, T2k, and T1b—were used, all exhibiting conformable contact relationships. Based on this understanding, 100 typical seismic traces were extracted from each profile. Furthermore, using sedimentary facies information from the forward-modeled geological model as labels and the two-dimensional time-spectral information of seismic waves transformed by continuous wavelet transform as feature values, a sedimentary facies seismic sample dataset was established. In this case, a total of 2341 sedimentary facies seismic samples were constructed. This method provides a large sample dataset for sedimentary facies machine learning in this case, greatly improving the accuracy of the method in predicting sedimentary facies boundaries, and has practical significance for widespread application.

[0105] Example 3

[0106] The first step involved constructing typical sedimentary models and physical property templates for different sedimentary facies in the actual work area based on the sedimentary characteristics of the core samples. This case study uses a 3D seismic data set from the central Junggar Basin in western China as an example. Through analysis of typical lithofacies assemblages from 18 drilled wells in the work area, analysis of their logging curves, and observation of core samples, the single-well and interconnected well sedimentary facies of the study area were clarified. The typical sedimentary model of the study area was also identified: the first member of the Sangonghe Formation mainly developed shallow-water delta front sedimentary facies, while the second and third members of the Sangonghe Formation were dominated by littoral-shallow lacustrine facies. The 3D seismic work area covers 356 km². 2 The number of wells is only 18, but their distribution in the plane is uneven. Therefore, it is necessary to use the machine learning sample set establishment method of seismic forward modeling to provide a large sample basis for machine learning.

[0107] The second step involved constructing a large number of sedimentary facies geological and physical property profiles and converting physical property models based on typical sedimentary models and property templates. The sand body contact relationships in this study area were primarily conformable. The sandstone thicknesses of different sedimentary facies ranged from 2-5 meters for lacustrine facies and 8-16 meters for shallow deltaic facies. Through parameter transformation, a total of 12 sedimentary facies geological profiles and 12 physical property profiles were drawn, each 2000 meters long.

[0108] The third step is to collect forward modeling parameters such as dominant frequency, signal-to-noise ratio, and actual wavelet from actual seismic data.

[0109] Matched filtering and wavelet transform techniques are used to extract actual wavelet parameters such as dominant frequency, wavelength, and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling. The signal-to-noise ratio (SNR) parameters of the actual seismic data are estimated using techniques such as superposition and time-domain SVD, which are then used as parameters for forward modeling. In this case, the dominant frequency of the seismic data is 26Hz, the actual wavelet is an approximate Ricker wavelet with zero phase, the sampling rate is 2ms, and the actual seismic SNR is 8. Based on this, the sedimentary facies physical property parameter model obtained in step 1 is used to transform the drawn sedimentary facies geological model from the depth domain to the time domain, laying the foundation for parameter settings in subsequent forward modeling.

[0110] The fourth step involves using the forward modeling parameters extracted in step 3, such as the actual seismic wavelet, the actual seismic dominant frequency, and the actual seismic signal-to-noise ratio, to perform forward modeling on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, thereby obtaining a large number of sedimentary facies forward modeled seismic profiles. In this case, a total of 12 seismic profiles were obtained through forward modeling, with a trace spacing of 5 meters, and each profile contains 400 seismic traces.

[0111] The fifth step is to convert the forward modeling seismic data into time-spectrum data through continuous wavelet transform to suppress waveform ambiguity.

[0112] To mitigate the ambiguity often present when using seismic waveforms and amplitudes to distinguish different sedimentary facies and lithologies, such as isomorphism, this patent utilizes Continuous Wavelet Transform (CWT) to convert the seismic waveforms of forward-modeled seismic data into two-dimensional time-frequency spectra. By expanding the information dimension, the ambiguity of seismic waveforms is effectively reduced. The wavelet basis used in this CWT is the Morlet wavelet. In this case, analysis of the extracted time-frequency spectra of different sedimentary facies shows that the spectral energy is not significantly affected by the sedimentary facies grain size, but the greater the facies thickness, the higher the low, mid, and high frequency energies. Conversely, the smaller the facies thickness, the weaker the mid and high frequency energies and the stronger the low frequency energies. Furthermore, the characteristics of samples collected from different locations of the same lithofacies show good consistency in the spectral spectra.

[0113] The sixth step is to extract typical seismic traces from the forward seismic profiles, and establish sedimentary facies seismic samples using sedimentary facies information from the geological model as labels and time-spectrum information obtained from forward seismic modeling as feature values.

[0114] Based on the interpreted stratigraphic development patterns of the target layer in the actual work area and existing understanding of stratigraphic contact relationships, typical seismic traces were extracted from all forward-modeled seismic profiles. In this case, three stratigraphic levels, J1s3, J1s2, and J1s1, were used. J1s3 and J1s2 represent conformable contacts, while J1s2 and J1s1 represent erosive contacts. Based on this understanding, 80 typical seismic traces were extracted from each profile. Furthermore, using sedimentary facies information from the forward-modeled geological model as labels and the two-dimensional time-spectral information of the seismic waves transformed by continuous wavelet transform as feature values, a sedimentary facies seismic sample dataset was established. In this case, a total of 1560 sedimentary facies seismic samples were constructed. This method provides a large sample dataset for sedimentary facies machine learning in this case, greatly improving the accuracy of the method in predicting sedimentary facies boundaries, and has practical significance for widespread application.

[0115] Example 4

[0116] Similar to Example 2, it also includes the following six implementation steps, but uses different parameter settings:

[0117] The first step involved constructing typical sedimentary models and physical property templates for different sedimentary facies in the actual work area based on the sedimentary characteristics of the core samples. This case study uses 3D seismic data from the western margin of the Junggar Basin in western China as an example. Through analysis of typical lithofacies assemblages from 12 drilled wells in the work area, analysis of their logging curves, and observation of core samples, the single-well and interconnected-well sedimentary facies of the study area were clarified. The typical sedimentary models of the study area were also identified: the Hutubi Formation mainly exhibits fan deltaic sedimentary facies, while the Lianmuqin Formation is primarily characterized by littoral-shallow lacustrine facies. The 3D work area covers 420 km². 2 The number of wells is only 12, but their distribution in the plane is uneven. Therefore, it is necessary to use the machine learning sample set establishment method of seismic forward modeling to provide a large sample basis for machine learning.

[0118] The second step involved constructing a large number of sedimentary facies geological and physical property profiles and converting physical property models based on typical sedimentary models and property templates. The sand body contact relationships in this study area were primarily conformable. The sandstone thicknesses of different sedimentary facies ranged from 3-7 meters for lacustrine facies and 8-13 meters for shallow deltaic facies. Through parameter transformation, a total of 13 sedimentary facies geological profiles and 13 physical property profiles were drawn, each 2000 meters long.

[0119] The third step is to collect forward modeling parameters such as dominant frequency, signal-to-noise ratio, and actual wavelet from actual seismic data.

[0120] Matched filtering and wavelet transform techniques are used to extract actual wavelet parameters such as dominant frequency, wavelength, and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling. The signal-to-noise ratio (SNR) of the actual seismic data is estimated using techniques such as superposition and time-domain SVD, and used as parameters for forward modeling. In this case, the dominant frequency of the seismic data is 35Hz, the actual wavelet is an approximate Ricker wavelet with zero phase, the sampling rate is 2ms, and the actual seismic SNR is 7. Based on this, the sedimentary facies physical property parameter model obtained in step 1 is used to transform the drawn sedimentary facies geological model from the depth domain to the time domain, laying the foundation for parameter settings in subsequent forward modeling.

[0121] The fourth step involves using the forward modeling parameters extracted in step 3, such as the actual seismic wavelet, the actual seismic dominant frequency, and the actual seismic signal-to-noise ratio, to perform forward modeling on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, thereby obtaining a large number of sedimentary facies forward modeled seismic profiles. In this case, a total of 13 seismic profiles were obtained through forward modeling, with a trace spacing of 5 meters, and each profile contains 400 seismic traces.

[0122] The fifth step is to convert the forward modeling seismic data into time-spectrum data through continuous wavelet transform to suppress waveform ambiguity.

[0123] To mitigate the ambiguity often present when using seismic waveforms and amplitudes to distinguish different sedimentary facies and lithologies, such as isomorphism, this patent utilizes Continuous Wavelet Transform (CWT) to convert the seismic waveforms of forward-modeled seismic data into two-dimensional time-frequency spectra. By expanding the information dimension, the ambiguity of seismic waveforms is effectively reduced. The wavelet basis used in this CWT is the Morlet wavelet. In this case, analysis of the extracted time-frequency spectra of different sedimentary facies reveals that the spectral energy is significantly affected by the grain size and thickness of the sedimentary facies. Generally, the coarser the grain size, the greater the thickness of the facies band, and the higher the mid-to-high frequency energy. Conversely, the smaller the grain size, the smaller the thickness of the facies band, the weaker the mid-to-high frequency energy, and the moderate low-frequency energy. Furthermore, samples of the same lithofacies collected from different locations exhibit good consistency in their spectral characteristics.

[0124] The sixth step is to extract typical seismic traces from the forward seismic profiles, and establish sedimentary facies seismic samples using sedimentary facies information from the geological model as labels and time-spectrum information obtained from forward seismic modeling as feature values.

[0125] Based on the interpreted stratigraphic development patterns of the target layer in the actual work area and existing understanding of stratigraphic contact relationships, typical seismic traces were extracted from all forward-modeled seismic profiles. In this case, five stratigraphic levels—K1h1, K1h4, K1s, K1l, and N1s—were used. K1h1 and K1h4 represent conformable contacts, K1h4 and K1s represent conformable contacts, K1s and K1l represent erosive contacts, and K1l and N1s represent erosive contacts. Based on this understanding, 80 typical seismic traces were extracted from each profile. Furthermore, using sedimentary facies information from the forward-modeled geological model as labels and the two-dimensional time-spectral information of the seismic waves transformed by continuous wavelet transform as feature values, a sedimentary facies seismic sample dataset was established. In this case, a total of 1790 sedimentary facies seismic samples were constructed. This method provides a large sample dataset for sedimentary facies machine learning in this case, greatly improving the accuracy of the method in predicting sedimentary facies boundaries, and has practical significance for widespread application.

[0126] Beneficial Effects: Based on core and regional sedimentary background studies, this application constructs typical sedimentary models through single-well-to-well facies analysis. A large number of sedimentary facies geological profiles are obtained based on these models. Simultaneously, actual dominant frequency, bandwidth, signal-to-noise ratio, and actual wavelet are extracted from actual seismic data to achieve forward modeling data that approximates the characteristics of actual seismic events, thereby improving the accuracy of sedimentary facies samples. Furthermore, continuous wavelet transform converts one-dimensional seismic traces into two-dimensional time-spectrum data, reducing the ambiguity of seismic-based sedimentary facies analysis by expanding the sample information dimension and further improving sample accuracy. Finally, using sedimentary facies in the geological model as labels and two-dimensional time-spectrum information as feature values, a large number of high-precision sedimentary facies seismic samples are established, laying a solid foundation for machine learning-based sedimentary facies prediction. Compared with manually picked sedimentary facies seismic samples from actual seismic data, this method can generate a large number of high-precision samples under the guidance of typical models, while the sample labels (sedimentary facies) are completely accurate, without errors caused by time-depth calibration, thus better meeting the needs of machine learning for sedimentary facies prediction.

[0127] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling, characterized in that, The sample establishment method includes: Step S1: Construct typical sedimentary models and templates for various sedimentary facies properties in the actual work area based on core sedimentary characteristics; Step S2: Obtain sedimentary facies geological profiles based on typical sedimentary models and sedimentary facies property templates; Step S3: Extract actual earthquake forward modeling parameters from actual earthquake data; Step S4: Conduct forward modeling of sedimentary facies geological models based on actual seismic parameters; Step S5: Convert the forward-modeled seismic data into time-spectrum data using continuous wavelet transform; Step S6: Extract typical seismic traces from the forward seismic profile, and establish sedimentary facies seismic samples using sedimentary facies in the geological model as labels and time-spectrum data as feature values.

2. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, Step S1: Constructing typical sedimentary models and multiple sedimentary facies property templates for the actual work area based on core sedimentary characteristics specifically includes: The sedimentary facies type of the study area was determined based on the sedimentary characteristics of the core samples. Based on single-well facies, multi-well facies, and planar facies, typical sedimentary facies models are established; Templates for sedimentary facies velocity and density properties were constructed based on well logging data.

3. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, Step S2: Obtaining a sedimentary facies geological profile based on typical sedimentary models and sedimentary facies property templates specifically includes: Based on typical sedimentary models and sedimentary facies property templates, sedimentary facies geology is carried out, physical property parameter profiles are constructed, and physical property parameter model conversion is performed.

4. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 3, characterized in that, The construction of the physical property parameter profile and the transformation of the physical property parameter model specifically include: Based on the sedimentary model determined in step S1, a sedimentary facies geological model is drawn by changing the contact relationship between sand bodies or between sand bodies and mudstone, or by changing the thickness of sand bodies and mudstone. The geological information contained in the profile is sedimentary facies constructed from different sand and mud combinations; In geological profiles, based on templates of various sedimentary facies physical properties, velocity and density parameters of various sedimentary facies are filled in, transforming the sedimentary facies geological model into a sedimentary facies physical property parameter model.

5. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, The actual earthquake forward modeling parameters include the dominant frequency, bandwidth, signal-to-noise ratio, and actual wavelet.

6. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, Step S3, extracting actual earthquake forward modeling parameters from actual earthquake data, specifically includes: Matched filtering, wavelet transform and other techniques are used to extract actual wavelet parameters such as dominant frequency, wavelength and sampling rate from actual seismic data to improve the accuracy of subsequent forward modeling. The signal-to-noise ratio (SNR) parameter of actual seismic data is estimated using techniques such as stacking and time-domain SVD decomposition, and used as parameters for forward modeling. The time-domain SVD decomposition is expressed as follows: P=USV T Where, the column vector of U is represented as XX T The column vector of V is represented by X. T X; singular values ​​are non-zero elements, mainly represented by the main diagonal of the diagonal matrix S. The seismic signal X is represented as a matrix: Where, x nm This refers to the element in the nth row and mth column, and so on. The sedimentary facies geological model was converted from the depth domain to the time domain using sedimentary facies physical property parameter models.

7. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 5, characterized in that, Step S4: Forward modeling of sedimentary facies geological models based on actual seismic parameters specifically includes: Using the extracted actual seismic wavelet, actual seismic dominant frequency, and actual seismic signal-to-noise ratio forward modeling parameters, seismic forward modeling was carried out on a large number of sedimentary facies physical property parameter profiles drawn based on typical sedimentary models and physical property templates, and sedimentary facies forward modeling seismic profiles were obtained.

8. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, Step S5: Converting forward-modeled seismic data into time-spectrum data through continuous wavelet transform specifically includes: The seismic waveforms of forward-modeled seismic data are transformed into two-dimensional time-spectrum maps using continuous wavelet transform (CWT). The extracted time-spectrum diagrams of different sedimentary facies were analyzed to extract typical spectral characteristics of different sedimentary facies, and a spectral characteristic chart of different sedimentary facies in the actual study area was constructed.

9. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 8, characterized in that, The process of converting the seismic waveform of forward-modeled seismic data into a two-dimensional time-spectrum image using continuous wavelet transform (CWT) specifically includes: The continuous wavelet transform of a seismic signal is represented as follows: in, Let f(t) be the basic wavelet or mother wavelet, a be the scaling factor, b be the translation factor, and f(t) be the travel time length function. The wavelet basis used in Continuous Wavelet Transform (CWT) is the Morlet wavelet, and the expression for the Morlet wavelet basis function is as follows: in, τ is the conjugate of the basic wavelet or the mother wavelet, a is the scale parameter, τ is the displacement parameter, and f(t) is the travel time length function.

10. The method for establishing time-spectrum sedimentary facies samples based on seismic forward modeling according to claim 1, characterized in that, Step S6: Extracting typical seismic traces from the forward-modeled seismic profile, and establishing sedimentary facies seismic samples using sedimentary facies in the geological model as labels and time-spectrum data as feature values, specifically includes: Using the interpretation of the stratigraphic development model of the target layer in the actual work area and the existing understanding of stratigraphic contact relationships, typical seismic traces are extracted from all forward seismic profiles. The typical seismic traces extracted are used as the basis for sample extraction. The sedimentary facies information in the forward geological model is used as the label, and the two-dimensional time-spectrum information of the seismic waves transformed by continuous wavelet transform is used as the feature value to establish a sedimentary facies seismic sample dataset.

Images