Seismic lithogenous phase prediction method based on semi-supervised learning
By employing a semi-supervised learning-based seismic diagenetic facies prediction method, combined with analysis of test data and rock physics experiments, sensitive seismic elastic parameters are screened, and a semi-supervised learning neural network model is created. This solves the problem of low correlation between diagenetic facies and seismic parameters, and achieves high-precision prediction and distribution interpretation of diagenetic facies in tight sandstone reservoirs in sparsely-well areas.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-24
AI Technical Summary
Existing methods for predicting seismic diagenetic facies mainly focus on the feasibility and key steps of seismic diagenetic facies research. However, the correlation between diagenetic facies and sensitive seismic elastic parameters is low and the mapping relationship is weak, resulting in low prediction accuracy. In particular, it is difficult to effectively characterize the spatial distribution of diagenetic facies in deep tight sandstone reservoirs in sparse well areas.
A semi-supervised learning-based seismic diagenetic facies prediction method was adopted. By comprehensively analyzing test data, rock physics experimental data, and well logging data, sensitive seismic elastic parameters were screened, a semi-supervised learning neural network model based on attention mechanism was created, and a nonlinear mapping relationship between core diagenetic facies and sensitive seismic elastic parameters was established. The comprehensive index data volume of diagenetic facies was obtained, and phase and slice processing was performed to interpret the spatial distribution of seismic diagenetic facies.
It significantly improves the accuracy of seismic diagenetic facies prediction, accurately identifies the spatial distribution of diagenetic facies, provides technical support for tight sandstone reservoirs in sparse well areas, reduces exploration and development risks, and enriches the theory of seismic sedimentology.
Smart Images
Figure CN121721720A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of seismic sedimentology technology and relates to a seismic diagenetic facies prediction method based on semi-supervised learning. Background Technology
[0002] Diagenetic facies is an important research topic in deep, dense sandstone. Core observation, thin section identification, and scanning electron microscopy are effective methods for identifying various diagenetic facies (Morad et al., 2010; Li et al., 2022). In single-well drilling, diagenetic facies are correlated with conventional logging curves, and vertical prediction of diagenetic facies is performed using logging responses (Cao et al., 2015; Lai et al., 2018). However, research on the spatial variation of diagenetic facies in tight sandstone is limited. In conventional multi-well areas, analytical test data and logging data can characterize the spatial distribution of diagenetic facies (Wang et al., 2023). However, in deep marine formations, analytical test data and logging data from a limited number of wells cannot effectively characterize the spatial distribution of diagenetic facies. Correlating seismic data with core diagenesis can elucidate the planar and profile distribution of diagenetic facies (Zeng et al., 2013; Zhang et al., 2018; Zhu et al., 2020; Wang et al., 2022a & 2022b; Wang et al.). Therefore, combining analysis of test data, well logging data, and seismic data, and innovating seismic diagenetic facies prediction technology guided by seismic sedimentology theory, is an important means to reveal the spatial characteristics of deep, tight sandstone reservoirs at sea.
[0003] A prerequisite for reservoir prediction using seismic data is finding seismic parameters sensitive to the reservoir. Currently, commonly used seismic parameters are velocity (or wave impedance) (Luo Qun and Huang Handong, 2012; Zeng Hongliu et al., 2013; Zong Zhaoyun, 2013; Yin Xingyao et al., 2015). However, when seismic parameters such as velocity (or wave impedance) cannot effectively distinguish between various diagenetic facies, directly using them to predict reservoirs is difficult. Therefore, selecting seismic parameters sensitive to diagenetic facies is essential, and the selection of sensitive seismic parameters is one of the important steps in improving the accuracy of seismic diagenetic facies prediction. The transition and evolution from comprehensive seismic attribute analysis to post-stack seismic inversion and then to pre-stack seismic inversion can obtain more accurate and comprehensive information (Luo Qun and Huang Handong, 2012). Petrophysical analysis of reservoir characteristics and seismic elastic parameters of different diagenetic facies helps to determine sensitive seismic elastic parameters.
[0004] The inventors proposed an analytical method for diagenetic facies-sensitive seismic elastic parameters by utilizing rock physics experimental data and analytical test data. Through rock physics analysis, correlation analysis, and the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, the diagenetic facies-sensitive seismic elastic parameters are determined.
[0005] The inventors proposed a supervised learning-based method for predicting seismic diagenetic facies using analytical test data, rock physics experimental data, and pre-stack seismic data. This method explains the spatial distribution of seismic diagenetic facies in the tight sandstone reservoir of the fourth member of the Huagang Formation in the East China Sea Basin.
[0006] However, existing methods for predicting seismic diagenesis mainly focus on the feasibility of seismic diagenesis, the three key steps, key seismic parameters, and prediction methods. The correlation between diagenesis and sensitive seismic elastic parameters is low, and research on their mapping relationship is weak. The development of supervised learning, unsupervised learning, and semi-supervised learning techniques has provided new opportunities to improve prediction accuracy. Therefore, there is an urgent need to develop semi-supervised learning algorithms for seismic diagenesis, establish more advanced linear or nonlinear mapping relationships between core diagenesis and sensitive seismic elastic parameters, improve the prediction accuracy of seismic diagenesis, and provide spatial constraint information for diagenetic numerical models. Summary of the Invention
[0007] To address the problems of existing technologies, this invention proposes a seismic diagenetic facies prediction method based on semi-supervised learning, providing a new technological approach for studying the spatial distribution of diagenetic facies in tight sandstone reservoirs.
[0008] A semi-supervised learning-based seismic diagenetic facies prediction method comprises the following steps: 1) Comprehensively utilizing analytical test data, rock physics experimental data, well logging data, and seismic data to study the characteristics of tight sandstone reservoirs in the target stratigraphic level of the study area; analyzing the main diagenetic types of the tight sandstone in the target stratigraphic level; establishing a diagenetic evolution sequence and diagenetic facies classification scheme; determining the core diagenetic facies; 2) Screening well logging curves sensitive to diagenetic facies; evaluating the vertical distribution of diagenetic facies in a single well using well logging; 3) Determining diagenetic facies-sensitive seismic elastic parameters through rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters; 4) Creating a semi-supervised learning neural network model based on an attention mechanism; using semi-supervised learning to build a nonlinear mapping relationship between core diagenetic facies and sensitive seismic elastic parameters; obtaining a comprehensive diagenetic facies index data volume; 5) Performing phase and slice processing on the comprehensive diagenetic facies index data volume; selecting typical stratigraphic slices; and interpreting the spatial distribution of seismic diagenetic facies.
[0009] The advantages of this invention are: it is feasible to operate, has good operational effects, and solves practical problems. It possesses the following innovative aspects: 1) Determining the characteristics and diagenetic types of tight sandstone reservoirs in the target stratigraphic level; 2) Establishing a diagenetic evolution sequence and diagenetic facies classification scheme; 3) Screening sensitive logging curves and interpreting the vertical distribution of diagenetic facies through logging; 4) Revealing the seismic identification principle of diagenetic facies and determining the seismic elastic parameters sensitive to diagenetic facies; 5) Creating a semi-supervised learning neural network model based on an attention mechanism to obtain a comprehensive diagenetic facies index data volume; 6) Performing phase and slice processing on the comprehensive diagenetic facies index; 7) Interpreting the spatial distribution of seismic diagenetic facies. This patented method meets the current evaluation needs for the spatial distribution of diagenetic facies in deep tight sandstone reservoirs in sparsely-well areas and provides technical support for interpreting the spatial distribution of seismic diagenetic facies in tight sandstone reservoirs in sparsely-well areas.
[0010] This invention is based on data from analysis and testing, rock physics experiments, well logging, and seismic data. It involves research on the characteristics and diagenetic types of tight sandstone reservoirs in target strata, the establishment of diagenetic evolution sequences and diagenetic facies classification schemes, the screening of diagenetic facies-sensitive well logging curves, the interpretation of diagenetic facies well logging, the screening of diagenetic facies-sensitive seismic elastic parameters, the creation of semi-supervised learning neural network models, the acquisition of diagenetic facies comprehensive index data volumes, phase and slice processing, and the interpretation of seismic diagenetic facies. This results in a technical process for a seismic diagenetic facies prediction method based on semi-supervised learning.
[0011] This invention considers the characteristics of tight sandstone reservoirs in the target stratigraphic level and the main diagenetic types, establishes a diagenetic evolution sequence and diagenetic facies classification scheme for tight sandstone, selects logging curves sensitive to diagenetic facies, and evaluates the vertical distribution of diagenetic facies in a single well through logging. Using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, it reveals the seismic identification principle of diagenetic facies, determines the seismic elastic parameters sensitive to diagenetic facies, and obtains the data volume of the determined sensitive seismic elastic parameters. A semi-supervised learning neural network model based on an attention mechanism is created, and a nonlinear mapping relationship between core diagenetic facies and sensitive seismic elastic parameters is established using semi-supervised learning to obtain a comprehensive diagenetic facies index data volume. Phase and slice processing is performed on the comprehensive diagenetic facies index data volume to select typical stratigraphic slices and interpret the spatial distribution of seismic diagenetic facies. This provides technical support for the study of the spatial distribution of seismic diagenetic facies in tight sandstone reservoirs in sparsely-welled areas and enriches the theory of seismic sedimentology.
[0012] This invention introduces an innovative semi-supervised learning algorithm for seismic diagenesis, establishing a nonlinear mapping relationship between core diagenesis and sensitive seismic elastic parameters. This significantly improves the accuracy of seismic diagenesis prediction, accurately identifying the spatial distribution of diagenesis under the coupled effects of tectonic, sedimentary, and diagenetic processes, thus making genetic sandbody prediction more accurate. This patented method can provide technical support for well location deployment in oil and gas exploration and development, helping to reduce exploration and development risks. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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. As shown in the figures:
[0014] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] Example 1: As Figure 1 As shown, a semi-supervised learning-based method for predicting seismic diagenetic facies belongs to the field of seismic sedimentology. It includes the following steps: determining the reservoir characteristics and main diagenetic types of tight sandstone, establishing diagenetic evolution sequences and diagenetic facies classification schemes; selecting well logging curves sensitive to diagenetic facies, and evaluating the vertical distribution of diagenetic facies in a single well; revealing the seismic identification principle of diagenetic facies, and determining seismic elastic parameters sensitive to diagenetic facies; characterizing the distribution of sedimentary facies under various seismic attribute constraints, and obtaining the determined data volume of sensitive seismic elastic parameters; creating a semi-supervised learning neural network model based on an attention mechanism, using semi-supervised learning to build a nonlinear mapping relationship between core diagenetic facies and sensitive seismic elastic parameters, and obtaining a comprehensive diagenetic facies index data volume; performing phase and slice processing on the comprehensive diagenetic facies index data volume, selecting typical stratigraphic slices, and interpreting the spatial distribution of seismic diagenetic facies.
[0017] A semi-supervised learning-based method for predicting seismic diagenetic facies provides information on reservoir characteristics of tight sandstone, diagenetic facies classification schemes, well logging evaluation of diagenetic facies, seismically sensitive elastic parameters of diagenetic facies, sedimentary facies studies under the sequence stratigraphic framework, a semi-supervised learning neural network model, a comprehensive diagenetic facies index data volume, and the spatial distribution of seismic diagenetic facies. The method includes the following steps:
[0018] Step 1) Collect target stratigraphic analysis test data, rock physics experimental data, well logging data and seismic data in the study area.
[0019] Step 2) Conduct research on the characteristics of tight sandstone reservoirs in the target strata of the study area, including sandstone composition, sandstone structure and pore space type.
[0020] Step 3) Determine the main diagenetic types of the tight sandstone in the target stratum and establish a diagenetic evolution sequence; then, propose a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals, which includes compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies, and unstable component dissolution facies; determine the core diagenetic facies based on cast thin sections, scanning electron microscopy, and cathodoluminescence thin sections.
[0021] Step 4) Select logging curves that are sensitive to diagenetic facies, establish the relationship between diagenetic facies and sensitive logging curves, and conduct logging evaluation of diagenetic facies. The logging evaluation evaluates the vertical distribution of diagenetic facies in a single well.
[0022] Step 5) By integrating rock physics experimental data and analytical test data, and by using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined.
[0023] Step 6) Guided by high-resolution sequence stratigraphy theory, based on drilling, logging and seismic data, identify fourth-order sequence boundaries on the third-order sequence of the target stratum. Combining well and seismic data, identify sequence boundaries of different levels based on seismic reflection characteristics on seismic profiles, and establish a sequence stratigraphic framework.
[0024] Based on post-stack seismic data, seismic attribute analysis techniques were used to extract seismic attributes such as root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude. Sedimentary facies studies were conducted by integrating multiple seismic attributes.
[0025] Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained by pre-stack elastic parameter inversion. Furthermore, multiple seismic elastic parameters are converted into porosity data volumes using empirical formulas. Thus, a data volume of sensitive seismic elastic parameters that has been determined is obtained.
[0026] Step 7) Establish a matching relationship between the core lithogenesis and sensitive seismic elastic parameters by combining well and seismic analysis, so that both the core lithogenesis and seismic elastic parameters are at the point scale at the wellhead location.
[0027] A semi-supervised learning neural network model based on an attention mechanism was created. This model was used to establish a nonlinear mapping relationship between the core diagenetic facies and sensitive seismic elastic parameters, obtaining a comprehensive diagenetic facies index data volume. The comprehensive diagenetic facies index is a parameter reflecting the type of diagenetic facies; high and low values represent favorable diagenetic facies and compacted dense facies, respectively. Due to the limited data on calcite cemented facies, it was not considered separately but was considered in conjunction with the compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolution facies, and quartz secondary enlarged cemented facies.
[0028] Step 8) Using 90° phase transformation technology, the seismic phase axis is correlated with the geological strata to obtain isochronous stratigraphic interfaces and establish an accurate temporal stratigraphic framework model. Furthermore, the diagenetic facies composite index between isochronous stratigraphic interfaces is made consistent with the core diagenetic facies and well-logged interpreted diagenetic facies. Stratigraphic slices containing the diagenetic facies composite index between isochronous stratigraphic interfaces are produced using stratigraphic slicing technology.
[0029] Based on the core diagenetic facies and well-logged interpreted diagenetic facies, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected. Through the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well-logged interpreted diagenetic facies, the typical stratigraphic slices are truncated using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, thereby interpreting the spatial distribution of seismic diagenetic facies.
[0030] Step 9), combining steps 1), 2), 3), 4), 5), 6), 7), and 8), proposes a seismic diagenetic facies prediction method based on semi-supervised learning, forming a technical process for a seismic diagenetic facies prediction method based on semi-supervised learning, and analyzes the diagenetic facies distribution characteristics of tight sandstone in the target strata of the study area.
[0031] Step 3) involves determining the type of diagenesis, establishing the diagenetic evolution sequence, and defining the diagenetic facies. The details are as follows:
[0032] The main diagenetic types of the target stratigraphic tight sandstone in the study area were determined, and a diagenetic evolution sequence was established. Then, a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals was proposed. These diagenetic facies include compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies, and unstable component dissolution facies.
[0033] For deep tight sandstone reservoirs, sedimentary material is the material basis for reservoir quality, while diagenesis is the main controlling factor. The innovation of this claim lies in: establishing a diagenetic evolution sequence and proposing a diagenetic facies classification scheme through diagenetic research on the tight sandstone of the target stratum.
[0034] Step 4) involves the screening of sensitive logging curves and the evaluation of diagenetic facies logging, as detailed below:
[0035] Well logging curves sensitive to diagenetic facies were selected, the relationship between diagenetic facies and sensitive well logging curves was established, and well logging evaluation of diagenetic facies was carried out. The vertical distribution of diagenetic facies in a single well was evaluated through well logging.
[0036] First, by analyzing the cross-plots of core diagenetic facies and logging curves, logging curves sensitive to diagenetic facies are selected. Then, supervised learning is used to establish a nonlinear relationship between core diagenetic facies and sensitive logging curves, and a diagenetic facies logging evaluation model is established. Finally, based on the diagenetic facies logging evaluation model, diagenetic facies logging evaluation is carried out to evaluate the vertical distribution of diagenetic facies in a single well.
[0037] Step 5) reveals the seismic identification principle of diagenetic facies and determines the seismically sensitive elastic parameters of diagenetic facies, as detailed below:
[0038] By combining rock physics experimental data and analytical test data, and by using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined.
[0039] Based on rock physics experimental data and analytical test data, and by comprehensively using rock physics analysis, correlation analysis, and the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, sensitive seismic elastic parameters were determined. Shear modulus, P-wave velocity ratio, P-wave impedance, and porosity data volumes are sensitive to diagenetic facies, and the correlation among the four seismic elastic parameters is weak. The determination of sensitive seismic elastic parameters makes seismic diagenetic facies studies more targeted in distinguishing multiple diagenetic facies.
[0040] The innovation lies in the fact that, based on the supporting rock physics experimental data and analytical test data, this claim determines the seismic elastic parameters sensitive to diagenetic facies through rock physics analysis, correlation analysis, and the relationship between diagenetic characteristics, mineral composition and seismic elastic parameters, thus providing an experimental and theoretical basis for seismic diagenetic facies prediction.
[0041] Step 6) involves establishing a sequence stratigraphic framework, developing sedimentary facies based on various seismic attributes, and acquiring the determined sensitive seismic elastic parameter data volume, as detailed below:
[0042] Guided by high-resolution sequence stratigraphy theory, and based on drilling, logging and seismic data, a fourth-order sequence boundary was identified on the basis of the third-order sequence of the upper Huagang Formation. By combining well and seismic data, and based on the seismic reflection characteristics on the seismic profile, sequence boundaries of different levels were identified, and a sequence stratigraphic framework was established.
[0043] Based on post-stack seismic data, seismic attribute analysis techniques were used to extract seismic attributes such as root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude. Sedimentary facies studies were conducted by integrating multiple seismic attributes.
[0044] Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained by pre-stack elastic parameter inversion. Furthermore, multiple seismic elastic parameters are converted into porosity data volumes using multiple linear regression equations. Thus, a data volume of sensitive seismic elastic parameters that has been determined is obtained.
[0045] Based on rock physics experimental data and analytical test data, an empirical formula was established using a multiple linear regression equation to relate porosity to seismic elastic parameters (such as P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio). Then, this empirical formula was used to convert the seismic elastic parameters into a porosity data volume. The multiple linear regression equation for calculating the porosity data volume is as follows:
[0046] (1)
[0047] In the formula, φ is porosity, Zp is longitudinal wave impedance, Zs is transverse wave impedance, Vp / Vs is the ratio of longitudinal to transverse wave velocities, and a, b, c, and d are constant terms.
[0048] The innovations are: 1) conducting sedimentary facies analysis based on multiple seismic attributes to obtain the planar distribution of tight sandstone sedimentary facies in the target layer; 2) using pre-stack elastic parameter inversion to obtain various seismic elastic parameters, and using a multiple linear regression method to establish a calculation formula for the porosity data volume to obtain the porosity data volume.
[0049] Step 7) involves creating a semi-supervised learning neural network model and obtaining the comprehensive diagenetic index data volume, as detailed below:
[0050] By combining well and seismic analysis, a matching relationship is established between the core lithogenic facies and sensitive seismic elastic parameters, so that both the core lithogenic facies and seismic elastic parameters are at the point scale at the wellhead location.
[0051] A semi-supervised learning neural network model with an attention mechanism was created. This model was used to establish a nonlinear mapping relationship between the core diagenetic facies and sensitive seismic elastic parameters, obtaining a comprehensive diagenetic facies index data volume. The comprehensive diagenetic facies index is a parameter reflecting the type of diagenetic facies; high and low values represent favorable diagenetic facies and compacted dense facies, respectively. Due to the limited data on calcite cemented facies, it was not considered separately but was considered in conjunction with the compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolution facies, and quartz secondary enlarged cemented facies.
[0052] A semi-supervised learning neural network model with an attention mechanism is created to capture the nonlinear mapping relationship between four diagenetic facies and sensitive seismic elastic parameters, thereby improving prediction accuracy.
[0053] Based on the TensorFlow deep learning framework, a U-shaped convolutional neural network model with a convolutional encoder and decoder based on an attention mechanism is defined for seismic lithogenesis prediction. The U-net network structure based on the attention mechanism consists of an encoder, a decoder, an attention mechanism module, and connecting bridges.
[0054] The supervised learning part uses seismically sensitive elastic parameters as input data and four diagenetic facies as labels. The unsupervised learning part only requires the seismically sensitive elastic parameters. For the labeled training set, the relationship between the four diagenetic facies estimated by the deep neural network and the seismically sensitive elastic parameters is utilized... L 1 The norm is used as the objective optimization function; for unlabeled training sets, a custom objective function is defined.
[0055] That is, the supervised learning objective function is expressed as
[0056] (2)
[0057] The unsupervised objective function is expressed as:
[0058] (3)
[0059] The semi-supervised objective function is
[0060] (4)
[0061] in, Represents the cost function; This indicates the network parameters that need to be estimated. b is the bias, and w is the weight vector. The input values for network activation are the four diagenetic facies estimated by the neural network; These are the four predicted diagenetic facies; P 0 This represents the four actual diagenetic facies; Indicates the weighting coefficient; N This indicates the amount of labeled data. M This represents the amount of unlabeled data. In the optimization problem described above, the following approach is adopted: L 1 The norm estimates the inconsistency between the label data and the predicted data, and a custom objective function is used to quantify the sparsity of the four diagenetic facies estimated by the network.
[0062] By using a semi-supervised learning neural network structure with an innovative attention mechanism, sensitive seismic elastic parameters are input into the trained neural network structure to conduct seismic lithogenesis studies.
[0063] The innovation lies in: innovatively establishing a semi-supervised learning neural network structure, building a nonlinear mapping relationship between core lithogenesis facies and sensitive seismic elastic parameters, establishing a seismic lithogenesis prediction model based on semi-supervised learning, and improving the prediction accuracy of seismic lithogenesis facies.
[0064] Step 8) involves obtaining typical stratigraphic slices and interpreting the spatial distribution of seismic diagenetic facies, as detailed below:
[0065] The 90° phase transformation technique was used to align seismic phase axes with geological strata, obtaining isochronous stratigraphic interfaces and establishing an accurate temporal stratigraphic framework model. Furthermore, the comprehensive diagenetic facies index between isochronous stratigraphic interfaces was made consistent with the core diagenetic facies and well-logged interpreted diagenetic facies. Stratigraphic slices containing the comprehensive diagenetic facies index between isochronous stratigraphic interfaces were created using stratigraphic slicing technology.
[0066] Based on the diagenetic facies of the core and the diagenetic facies interpreted from well logging, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected.
[0067] By interpreting the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well logging, and using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, typical stratigraphic slices are truncated to interpret the spatial distribution of seismic diagenetic facies.
[0068] The innovation lies in the phase and slice processing of the comprehensive diagenetic facies index data volume, which enables the comprehensive diagenetic facies index between isochronous stratigraphic interfaces to be consistent with the core diagenetic facies and well-logged interpreted diagenetic facies, thereby selecting typical stratigraphic slices and interpreting the spatial distribution of seismic diagenetic facies in the target strata.
[0069] By combining and innovating steps 1), 2), 3), 4), 5), 6), 7), 8), and 9), an evaluation process for a seismic lithogenesis prediction method based on semi-supervised learning is formed.
[0070] Example 2: As Figure 1 As shown, a seismic lithogenesis prediction method based on semi-supervised learning is proposed, which aims to overcome the problems of low correlation between core lithogenesis and sensitive seismic elastic parameters and weak research on nonlinear mapping relationships.
[0071] A semi-supervised learning-based method for predicting seismic diagenetic facies analyzes the characteristics of tight sandstone reservoirs in target strata, determines the type of diagenesis and diagenetic evolution sequence, establishes a diagenetic facies classification scheme, evaluates the vertical distribution of diagenetic facies through well logging, determines seismically sensitive elastic parameters of diagenetic facies, produces sedimentary facies studies with multiple seismic attributes, creates a semi-supervised learning neural network model, obtains comprehensive diagenetic facies index data, and interprets the spatial distribution of seismic diagenetic facies. The method includes the following steps:
[0072] Step 1) Collect target stratigraphic analysis test data, rock physics experimental data, well logging data and seismic data in the study area.
[0073] Step 2) Conduct research on the characteristics of tight sandstone reservoirs in the target strata of the study area, including sandstone composition, sandstone structure and pore space type.
[0074] Step 3) Determine the main diagenetic types of the tight sandstone in the target stratum and establish a diagenetic evolution sequence; then, propose a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals. These diagenetic facies include compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies, and unstable component dissolution facies; determine the core diagenetic facies based on cast thin sections, scanning electron microscopy, and cathodoluminescence thin sections.
[0075] Step 4) Select logging curves that are sensitive to diagenetic facies, establish the relationship between diagenetic facies and sensitive logging curves, and conduct logging evaluation of diagenetic facies. The logging evaluation evaluates the vertical distribution of diagenetic facies in a single well.
[0076] Step 5) By integrating rock physics experimental data and analytical test data, and by using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined.
[0077] Step 6) Guided by high-resolution sequence stratigraphy theory, based on drilling, logging and seismic data, identify fourth-order sequence boundaries on the third-order sequence of the target stratum. Combining well and seismic data, identify sequence boundaries of different levels based on seismic reflection characteristics on seismic profiles, and establish a sequence stratigraphic framework.
[0078] Based on post-stack seismic data, seismic attribute analysis techniques were used to extract seismic attributes such as root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude. Sedimentary facies studies were conducted by integrating multiple seismic attributes.
[0079] Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained by pre-stack elastic parameter inversion. Furthermore, multiple seismic elastic parameters are converted into porosity data volumes using empirical formulas. Thus, a data volume of sensitive seismic elastic parameters that has been determined is obtained.
[0080] Step 7) Establish a matching relationship between the core lithogenesis and sensitive seismic elastic parameters by combining well and seismic analysis, so that both the core lithogenesis and seismic elastic parameters are at the point scale at the wellhead location.
[0081] A semi-supervised learning neural network model based on an attention mechanism was created. This model was used to establish a nonlinear mapping relationship between the core diagenetic facies and sensitive seismic elastic parameters, obtaining a comprehensive diagenetic facies index data volume. The comprehensive diagenetic facies index is a parameter reflecting the type of diagenetic facies; high and low values represent favorable diagenetic facies and compacted dense facies, respectively. Due to the limited data on calcite cemented facies, it was not considered separately but was considered in conjunction with the compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolution facies, and quartz secondary enlarged cemented facies.
[0082] Step 8) Using 90° phase transformation technology, the seismic phase axis is correlated with the geological strata to obtain isochronous stratigraphic interfaces and establish an accurate temporal stratigraphic framework model. Furthermore, the diagenetic facies composite index between isochronous stratigraphic interfaces is made consistent with the core diagenetic facies and well-logged interpreted diagenetic facies. Stratigraphic slices containing the diagenetic facies composite index between isochronous stratigraphic interfaces are produced using stratigraphic slicing technology.
[0083] Based on the diagenetic facies of the core and the diagenetic facies interpreted from well logging, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected.
[0084] By interpreting the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well logging, and using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, typical stratigraphic slices are truncated to interpret the spatial distribution of seismic diagenetic facies.
[0085] Step 9), combining steps 1), 2), 3), 4), 5), 6), 7), and 8), proposes a seismic diagenetic facies prediction method based on semi-supervised learning, forming a technical process for a seismic diagenetic facies prediction method based on semi-supervised learning, and analyzes the diagenetic facies distribution characteristics of tight sandstone in the target strata of the study area.
[0086] A semi-supervised learning-based method for predicting seismic diagenetic facies involves: studying tight sandstone reservoir characteristics; analyzing diagenetic types; establishing diagenetic evolution sequences and diagenetic facies classification schemes; analyzing diagenetic facies-sensitive logging curves and evaluating diagenetic facies logging; determining diagenetic facies-sensitive seismic elastic parameters; studying sedimentary facies with multiple seismic attributes; acquiring sensitive seismic elastic parameters; creating a semi-supervised learning neural network model; acquiring comprehensive diagenetic facies index data; phase and slice processing; screening typical stratigraphic slices; and interpreting seismic diagenetic facies. This method ultimately interprets the spatial distribution of seismic diagenetic facies, providing technical support for the study of the spatial distribution of diagenetic facies in tight sandstone reservoirs in sparsely-welled areas.
[0087] This invention achieves innovations in establishing diagenetic facies classification schemes for tight sandstone reservoirs, interpreting diagenetic facies logging, determining seismically sensitive elastic parameters of diagenetic facies, creating semi-supervised learning neural network models, acquiring comprehensive diagenetic facies index data, and phase and slice processing techniques. Furthermore, by comprehensively considering and analyzing test data, rock physics experimental data, logging data, and seismic data, a semi-supervised learning-based seismic diagenetic facies prediction method has been developed. An attention-based semi-supervised learning neural network model has been created to interpret the spatial distribution of seismic diagenetic facies, enriching seismic sedimentology theory and related technical methods. This invention fills a gap in the application of semi-supervised learning neural network structures in the field of seismic diagenetic facies prediction methods for tight sandstone reservoirs in sparsely-well areas. It is the first innovative method to create an attention-based semi-supervised learning neural network model, acquire comprehensive diagenetic facies indices, and perform phase and slice processing of the comprehensive diagenetic facies index data. This invention presents a semi-supervised learning-based seismic diagenetic facies prediction method and establishes a technical process for the spatial distribution of seismic diagenetic facies based on semi-supervised learning, which can provide good technical support for the spatial distribution of seismic diagenetic facies in tight sandstone reservoirs in sparsely-well areas.
[0088] Example 3: As Figure 1 As shown, a semi-supervised learning-based seismic diagenetic facies prediction method is proposed. Based on analytical testing, rock physics experiments, well logging, and seismic data, it mainly includes the study of tight sandstone reservoir characteristics, determination of major diagenetic types, establishment of diagenetic evolution sequences and diagenetic facies classification schemes, screening of diagenetic facies-sensitive well logging curves, interpretation of diagenetic facies well logging, revelation of the seismic identification principle of diagenetic facies, determination and acquisition of diagenetic facies-sensitive seismic elastic parameters, creation of a semi-supervised learning neural network model, acquisition of diagenetic facies comprehensive index data volume, phase and slice processing, and interpretation of seismic diagenetic facies.
[0089] A semi-supervised learning-based method for predicting seismic lithogenesis facies, such as Figure 1 The technical flowchart shows the following steps:
[0090] (1) Collect analytical test data, rock physics experimental data, well logging data and seismic data of tight sandstone in the fourth member of Huagang Formation (H4 gas field) in XX Depression of XX Basin; analytical test data include cast thin section, scanning electron microscopy, cathodoluminescence thin section, whole rock X-ray diffraction analysis data and clay X-ray diffraction analysis data; well logging curves include natural gamma, sonic transit time, density, neutron, deep lateral resistivity, shallow lateral resistivity and other curves; seismic data and pre-stack seismic data and post-stack seismic data.
[0091] (2) Conduct research on the characteristics of tight sandstone reservoirs in the target strata of the study area, including sandstone composition, sandstone structure and pore space type.
[0092] (3) Determine the main diagenetic types of the target stratigraphic tight sandstone and establish a diagenetic evolution sequence; then, propose a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals, which includes compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies and unstable component dissolution facies; determine the core diagenetic facies based on cast thin sections, scanning electron microscopy and cathodoluminescence thin sections.
[0093] (4) Screen out logging curves that are sensitive to diagenetic facies, establish the relationship between diagenetic facies and sensitive logging curves, conduct logging evaluation of diagenetic facies, and evaluate the vertical distribution of diagenetic facies in a single well.
[0094] (5) By combining rock physics experimental data and analytical test data, and by using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined.
[0095] (6) Under the guidance of high-resolution sequence stratigraphy theory, based on drilling, logging and seismic data, fourth-level sequence interfaces are identified on the basis of the third-level sequence of the target stratum. By combining well and seismic data, different levels of sequence interfaces are identified based on the seismic reflection characteristics on the seismic profile, and a sequence stratigraphic framework is established.
[0096] Based on post-stack seismic data, seismic attribute analysis techniques were used to extract seismic attributes such as root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude. Sedimentary facies studies were conducted by integrating multiple seismic attributes.
[0097] Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained by pre-stack elastic parameter inversion. Furthermore, multiple seismic elastic parameters are converted into porosity data volumes using empirical formulas. Thus, a data volume of sensitive seismic elastic parameters that has been determined is obtained.
[0098] Obtain the identified sensitive seismic elastic parameter data volume.
[0099] Based on rock physics experimental data and analytical test data, an empirical formula was established using a multiple linear regression equation to relate porosity to seismic elastic parameters (such as P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio). Then, this empirical formula was used to convert the seismic elastic parameters into a porosity data volume. The multiple linear regression equation for calculating the porosity data volume is as follows:
[0100] , (1)
[0101] In the formula, φ is porosity, Zp is longitudinal wave impedance, Zs is transverse wave impedance, and Vp / Vs is the ratio of longitudinal wave velocity to transverse wave velocity.
[0102] (7) By combining well and seismic analysis, a matching relationship is established between the core lithogenic facies and the sensitive seismic elastic parameters, so that both the core lithogenic facies and the seismic elastic parameters are at the point scale at the wellhead location.
[0103] A semi-supervised learning neural network model based on an attention mechanism was created. This model was used to establish a nonlinear mapping relationship between the core diagenetic facies and sensitive seismic elastic parameters, obtaining a comprehensive diagenetic facies index data volume. The comprehensive diagenetic facies index is a parameter reflecting the type of diagenetic facies; high and low values represent favorable diagenetic facies and compacted dense facies, respectively. Due to the limited data on calcite cemented facies, it was not considered separately but was considered in conjunction with the compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolution facies, and quartz secondary enlarged cemented facies.
[0104] (8) Using 90° phase transformation technology, the seismic phase axis is correlated with the geological strata to obtain isochronous stratigraphic interfaces and establish an accurate temporal stratigraphic framework model; and, to achieve consistency between the comprehensive diagenetic facies index between isochronous stratigraphic interfaces and the core diagenetic facies and well logging interpretation diagenetic facies. Stratigraphic slices with the comprehensive diagenetic facies index between isochronous stratigraphic interfaces are produced using stratigraphic slicing technology.
[0105] Based on the diagenetic facies of the core and the diagenetic facies interpreted from well logging, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected.
[0106] By interpreting the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well logging, and using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, typical stratigraphic slices are truncated to interpret the spatial distribution of seismic diagenetic facies.
[0107] A seismic diagenetic facies prediction method based on semi-supervised learning is proposed, and a set of technical procedures for seismic diagenetic facies prediction based on semi-supervised learning is formed. The diagenetic facies distribution characteristics of tight sandstone in the target stratigraphic position in the study area are analyzed.
[0108] 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, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A seismic lithogenesis prediction method based on semi-supervised learning, characterized in that, By comprehensively utilizing analytical test data, rock physics experimental data, well logging data, and seismic data, this study investigates the characteristics of tight sandstone reservoirs in the target strata of the study area, analyzes the main diagenetic types of tight sandstone in the target strata, establishes diagenetic evolution sequences and diagenetic facies classification schemes, and determines the core diagenetic facies. Well logging curves sensitive to diagenetic facies are selected, and well logging evaluation is performed on the vertical distribution of diagenetic facies in a single well. By utilizing rock physics analysis, correlation analysis, and the relationship analysis between diagenetic characteristics, mineral composition, and seismic elastic parameters, sensitive seismic elastic parameters of diagenetic facies are determined. A semi-supervised learning neural network model based on an attention mechanism is created, and a nonlinear mapping relationship between core diagenetic facies and sensitive seismic elastic parameters is established using semi-supervised learning to obtain a comprehensive diagenetic facies index data volume. The comprehensive diagenetic facies index data volume is processed by phase and slice analysis to select typical stratigraphic slices and interpret the spatial distribution of seismic diagenetic facies.
2. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Includes the following steps: Step 1) Collect target stratigraphic analysis test data, rock physics experimental data, well logging data, and seismic data for the study area; Step 2) Conduct research on the characteristics of tight sandstone reservoirs in the target stratigraphic level in the study area, including sandstone composition, sandstone structure and pore space type; Step 3) Determine the main diagenetic types of the tight sandstone in the target stratum and establish a diagenetic evolution sequence; then, propose a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals. These diagenetic facies include compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies, and unstable component dissolution facies; determine the core diagenetic facies based on cast thin sections, scanning electron microscopy, and cathodoluminescence thin sections. Step 4) Select logging curves that are sensitive to diagenetic facies, establish the relationship between diagenetic facies and sensitive logging curves, conduct logging evaluation of diagenetic facies, and evaluate the vertical distribution of diagenetic facies in a single well. Step 5) By integrating rock physics experimental data and analytical test data, and by using rock physics analysis, correlation analysis, and analysis of the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined. Step 6) Guided by high-resolution sequence stratigraphy theory, based on drilling, logging and seismic data, identify fourth-order sequence boundaries on the third-order sequence of the target stratum. Combine well and seismic data, identify sequence boundaries of different levels based on seismic reflection characteristics on seismic profiles, and establish a sequence stratigraphic framework. Based on post-stack seismic data, seismic attribute analysis techniques were used to extract root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude seismic attributes, and sedimentary facies studies were conducted using multiple seismic attributes. Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained by pre-stack elastic parameter inversion. Furthermore, multiple seismic elastic parameters are converted into porosity data volumes using empirical formulas. Thus, a data volume of sensitive seismic elastic parameters that has been determined is obtained. Step 7) Establish a matching relationship between core diagenetic facies and sensitive seismic elastic parameters by combining well and seismic analysis, so that both core diagenetic facies and seismic elastic parameters are at the point scale at the wellhead location; A semi-supervised learning neural network model based on an attention mechanism was created. The nonlinear mapping relationship between the core diagenetic facies and sensitive seismic elastic parameters was established using semi-supervised learning to obtain the comprehensive diagenetic facies index data. The comprehensive diagenetic facies index is a parameter reflecting the type of diagenetic facies. High and low values represent favorable diagenetic facies and compacted dense facies, respectively. Since there is little data on calcite cemented facies, it is not considered separately, but is considered together with compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolution facies, and quartz secondary enlarged cemented facies; Step 8) Use 90° phase transformation technology to align seismic phase axes with geological strata, obtain isochronous stratigraphic interfaces, and establish an accurate temporal stratigraphic framework model; furthermore, ensure that the comprehensive diagenetic facies index between isochronous stratigraphic interfaces is consistent with the core diagenetic facies and well logging interpretation diagenetic facies. Stratigraphic slices with comprehensive diagenetic facies indices between isochronous stratigraphic interfaces are produced using stratigraphic slicing technology. Based on the core diagenetic facies and well-logged interpreted diagenetic facies, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected. Through the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well-logged interpreted diagenetic facies, the typical stratigraphic slices are truncated using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, thereby interpreting the spatial distribution of seismic diagenetic facies and obtaining the typical stratigraphic slices and the spatial distribution of the interpreted seismic diagenetic facies.
3. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 3) involves determining the type of diagenesis, establishing the diagenetic evolution sequence, and defining the diagenetic facies. The details are as follows: The main diagenetic types of the target stratigraphic tight sandstone in the study area were determined, and a diagenetic evolution sequence was established. Then, a diagenetic facies classification scheme based on diagenetic processes and diagenetic minerals was proposed. These diagenetic facies include compacted tight facies, chlorite film cemented facies, calcite cemented facies, quartz secondary enlarged cemented facies, and unstable component dissolution facies. For deep tight sandstone reservoirs, sedimentary material is the material basis for the quality of tight sandstone reservoirs, while diagenesis is the main controlling factor for the quality of deep tight sandstone reservoirs. By studying the diagenesis of tight sandstone in the target strata, a diagenetic evolution sequence is established and a diagenetic facies classification scheme is proposed.
4. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 4) involves the screening of sensitive logging curves and the evaluation of diagenetic facies logging, as detailed below: Well logging curves sensitive to diagenetic facies were selected, the relationship between diagenetic facies and sensitive well logging curves was established, and well logging evaluation of diagenetic facies was carried out, evaluating the vertical distribution of diagenetic facies in a single well; First, by analyzing the cross-plots of core diagenetic facies and logging curves, logging curves sensitive to diagenetic facies are selected. Then, supervised learning is used to establish a nonlinear relationship between core diagenetic facies and sensitive logging curves, and a diagenetic facies logging evaluation model is established. Finally, based on the diagenetic facies logging evaluation model, diagenetic facies logging evaluation is carried out to evaluate the vertical distribution of diagenetic facies in a single well.
5. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 5) reveals the seismic identification principle of diagenetic facies and determines the seismically sensitive elastic parameters of diagenetic facies, as detailed below: By combining rock physics experimental data and analytical test data, and by utilizing rock physics analysis, correlation analysis, and the relationship analysis between diagenetic characteristics, mineral composition, and seismic elastic parameters, the seismic identification principle of diagenetic facies is revealed, and the seismic elastic parameters sensitive to diagenetic facies are determined. Based on rock physics experimental data and analytical test data, and by comprehensively using rock physics analysis, correlation analysis, and the relationship between diagenetic characteristics, mineral composition, and seismic elastic parameters, sensitive seismic elastic parameters were determined. Shear modulus, P-wave velocity ratio, P-wave impedance, and porosity data volumes are sensitive to diagenetic facies, and the correlation among the four seismic elastic parameters is weak. The determination of sensitive seismic elastic parameters makes seismic diagenetic facies studies more targeted in distinguishing multiple diagenetic facies.
6. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 6) involves establishing a sequence stratigraphic framework, developing sedimentary facies based on various seismic attributes, and acquiring the determined sensitive seismic elastic parameter data volume, as detailed below: Guided by high-resolution sequence stratigraphy theory and based on drilling, logging, and seismic data, fourth-order sequence boundaries were identified on the basis of the third-order sequence stratigraphy of the upper Huagang Formation. By combining well and seismic data and identifying sequence boundaries of different levels based on seismic reflection characteristics on seismic profiles, a sequence stratigraphic framework was established. Based on post-stack seismic data, seismic attribute analysis techniques were used to extract seismic attributes such as root mean square amplitude, mean absolute amplitude, and maximum absolute amplitude. Sedimentary facies studies were then conducted by integrating multiple seismic attributes. Various seismic elastic parameters, including P-wave modulus, shear modulus, P-wave impedance, S-wave impedance, and P-wave / S-wave velocity ratio, are obtained through pre-stack elastic parameter inversion. Furthermore, multiple linear regression equations are used to convert these seismic elastic parameters into porosity data volumes, thus acquiring data volumes of the identified sensitive seismic elastic parameters. Based on rock physics experimental data and analytical test data, an empirical formula was established using a multiple linear regression equation to relate porosity to the seismic elastic parameters P-wave impedance, S-wave impedance, and the P-wave / S-wave velocity ratio. Then, the empirical formula was used to convert the seismic elastic parameters into a porosity data volume. The multiple linear regression equation for calculating the porosity data volume is as follows: （1） In the formula, φ is porosity, Zp is longitudinal wave impedance, Zs is transverse wave impedance, Vp / Vs is the ratio of longitudinal to transverse wave velocities, and a, b, c, and d are constant terms.
7. The seismic diagenetic facies prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 7) involves creating a semi-supervised learning neural network model and obtaining the comprehensive diagenetic index data volume, as detailed below: By combining well-seismic analysis, a matching relationship is established between the core diagenetic facies and sensitive seismic elastic parameters, ensuring that both the core diagenetic facies and seismic elastic parameters are at point scale at the wellhead location. A semi-supervised learning neural network model with an attention mechanism was created. The nonlinear mapping relationship between the core diagenetic facies and the elastic parameters of sensitive earthquakes was established using semi-supervised learning to obtain the comprehensive diagenetic facies index data. The comprehensive diagenetic facies index is a parameter that reflects the type of diagenetic facies. High and low values represent favorable diagenetic facies and compacted dense facies, respectively. Since there is little data on calcite cemented facies, it is not considered separately, but is considered together with compacted dense facies. The comprehensive diagenetic facies index can distinguish between compacted dense facies, chlorite film cemented facies, unstable component dissolved facies, and quartz secondary enlarged cemented facies; A semi-supervised learning neural network model with an attention mechanism is created to capture the nonlinear mapping relationship between four diagenetic facies and sensitive seismic elastic parameters, thereby improving prediction accuracy; Based on the TensorFlow deep learning framework, a U-shaped convolutional neural network model with a convolutional encoder and a convolutional decoder based on an attention mechanism is defined for seismic lithogenesis prediction. The U-net network structure based on the attention mechanism consists of an encoder, a decoder, an attention mechanism module, and connection bridges. The supervised learning component uses sensitive seismic elastic parameters as input data and four diagenetic facies as labels. The unsupervised learning part only requires the sensitive seismic elastic parameters; for the labeled training set, the relationship between the four diagenetic facies estimated by the deep neural network and the sensitive seismic elastic parameters is utilized. L 1 Norm as the objective optimization function; for unlabeled training sets, a custom objective function is used. That is, the supervised learning objective function is expressed as （2） The unsupervised objective function is expressed as: （3） The semi-supervised objective function is （4） in, Represents the cost function; This indicates the network parameters that need to be estimated. ; b For bias, w This is the weight vector; The input values for network activation are the four diagenetic facies estimated by the neural network; P represents the four predicted diagenetic facies; P0 represents the four actual diagenetic facies. Indicates the weighting coefficient; N This indicates the amount of labeled data. M This represents the amount of unlabeled data; in the optimization problem described above, the following approach is adopted: L 1 Norm estimates the inconsistency between label and predicted data, and a custom objective function is used to quantify the sparsity of the four diagenetic facies estimated by the network; By using a semi-supervised learning neural network structure with an innovative attention mechanism, sensitive seismic elastic parameters are input into the trained neural network structure to conduct seismic lithogenesis studies.
8. The seismic lithogenesis prediction method based on semi-supervised learning according to claim 1, characterized in that, Step 8) involves obtaining typical stratigraphic slices and interpreting the spatial distribution of seismic diagenetic facies, as detailed below: By employing 90° phase transformation technology, seismic phase axes are correlated with geological strata, isochronous stratigraphic interfaces are obtained, and an accurate temporal stratigraphic framework model is established; the comprehensive diagenetic facies index between isochronous stratigraphic interfaces is consistent with the diagenetic facies of core samples and well logging interpretations. Stratigraphic slices with comprehensive diagenetic facies indices between isochronous stratigraphic interfaces were produced using stratigraphic slicing techniques. Based on the diagenetic facies of core samples and well logging interpretations, typical stratigraphic slices that best reflect the diagenetic characteristics of the target strata are selected. By interpreting the synergistic constraints of sedimentary facies planar distribution, core diagenetic facies, and well logging, and using the specific distribution range of the comprehensive diagenetic facies index for each diagenetic facies, typical stratigraphic slices are truncated to interpret the spatial distribution of seismic diagenetic facies.