Earthquake intelligent fusion and reservoir prediction method and system under Bayesian thickness classification, medium and equipment
By using Bayesian thickness classification and multi-model fusion, the problem of accurately capturing sand body thickness information in seismic reservoir prediction has been solved, achieving high-precision and interpretable reservoir prediction, and improving the reliability and risk control capabilities of exploration.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-03-31
AI Technical Summary
Existing seismic reservoir prediction methods struggle to accurately capture sand body thickness information under complex geological conditions, and machine learning models lack physical constraints, resulting in large prediction errors and poor interpretability.
The Bayesian thickness classification method is used to divide the seismic response characteristics into four typical intervals. By combining Bayesian posterior probability and Gaussian mixture model, multiple sub-models are constructed and fused through an adaptive learning framework to output thickness prediction results.
It improves the stability and reliability of seismic reservoir prediction, enhances geological interpretability, provides an uncertainty distribution field for thickness prediction, and supports the selection of oil and gas targets and the control of exploration risks.
Smart Images
Figure CN121763380A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of seismic reservoir prediction technology, and in particular to a seismic intelligent fusion and reservoir prediction method, system, medium and equipment based on Bayesian thickness classification. Background Technology
[0002] Seismic attribute analysis technology has been widely applied to reservoir prediction. Single seismic attributes extract and enhance information from one or more aspects of seismic data, typically improving reservoir prediction accuracy, but they suffer from significant limitations, losing a large amount of valuable information. Therefore, multi-attribute comprehensive analysis is an effective way to address these issues, with intelligent multi-attribute fusion being a current hot topic and cutting-edge area in attribute analysis.
[0003] Intelligent fusion of seismic attributes is an advanced seismic analysis technique based on machine learning. It aims to spatially predict reservoir thickness in inter-well regions by constructing a nonlinear mapping relationship between seismic attributes and well logging data. Compared with traditional linear methods such as color fusion or principal component analysis, this technique offers advantages such as less subjective intervention, higher resolution, and stronger noise resistance, demonstrating promising application potential in reservoir prediction. However, current research still faces two core challenges: First, sand body size significantly impacts seismic response characteristics, especially given the marked differences in seismic response patterns across different sand body thickness ranges. Without differentiation under complex geological conditions, prediction models often struggle to fully extract effective information from the data, leading to decreased prediction accuracy. Second, while machine learning methods possess powerful nonlinear expressive capabilities and can extract statistical patterns from multidimensional attributes, their model training process typically lacks physical constraints. The internal structure of the model is highly opaque, resulting in insufficient geological interpretability and low reliability. Especially in cases of ambiguity in seismic responses, purely data-driven learning models are prone to overfitting, failing to meet the dual requirements of accuracy and reliability for refined reservoir characterization.
[0004] Bayesian inference, as a statistical reasoning framework that integrates prior knowledge and observational data, possesses strong theoretical rigor and the ability to express uncertainty. In recent years, it has been widely applied in geological tasks such as seismic inversion, lithology prediction, and reservoir parameter prediction. In these applications, Bayesian methods can significantly improve the stability and physical consistency of prediction models by introducing geological priors to effectively constrain the training process, and can output posterior probability distributions to characterize the reliability of prediction results. However, despite its significant achievements in areas such as seismic inversion, this theory remains lacking in the application of seismic attribute fusion for sandbody prediction. Therefore, it is urgent to explore a new modeling paradigm that incorporates Bayesian inference into the seismic attribute fusion process to construct a more robust, interpretable, and geologically adaptable reservoir prediction method. Summary of the Invention
[0005] To address the aforementioned issues, the present invention aims to provide a method, system, medium, and device for intelligent seismic fusion and reservoir prediction based on Bayesian thickness classification, which improves the stability and reliability of prediction results and enhances geological interpretability.
[0006] To achieve the above objectives, in a first aspect, the technical solution adopted by this invention is as follows: a Bayesian thickness classification-based intelligent seismic fusion and reservoir prediction method, comprising: based on the seismic response law of the amplitude-thickness curve, dividing the seismic response characteristics of different sand body thicknesses into four typical intervals, namely, no response interval, increasing interval, decreasing interval, and stable interval; based on the four typical intervals, combined with the actual sand body thickness distribution characteristics and the effective frequency band range of seismic data, iteratively optimizing the thickness boundaries of different seismic response intervals through a Bayesian posterior probability classification method, so as to classify the seismic response intervals corresponding to sand body thicknesses into four categories: no response sand body, thin sand body, medium-thick sand body, and thick sand body; respectively selecting sand body thickness classification sensitive attributes and sand body thickness regression prediction sensitive attributes, and performing collinearity analysis on each attribute to reduce similar attributes; constructing posterior probabilities for different sand body thickness types. The posterior probability field supports both soft classification representation and sub-model fusion weight constraints, including: modeling the distribution of each thickness type in the multi-dimensional attribute space using a Gaussian mixture model; estimating model parameters using the expectation-maximization algorithm; combining the prior probability of the sample sand body thickness proportion after iterative convergence; calculating the posterior probability of each sample belonging to different thickness types based on Bayes' theorem; generating the planar distribution probability of thickness types; constructing multiple sub-models using an adaptive learning framework based on well logging interpretation of sand body thickness and the regression prediction of the selected sand body thickness as a sensitive attribute; training sub-models for different thickness types; assigning dynamic weights to each trained sub-model through the posterior probability field to drive multi-model fusion to achieve a unified output of thickness prediction; deploying the fused model to reservoir prediction across the entire area; and using intelligent analysis of seismic attribute data in the study area to predict reservoir spatial distribution and output prediction results.
[0007] Furthermore, the four typical intervals are as follows: No-response range: The sand body is extremely thin, and the amplitude is close to the background noise; Increasing range: As the thickness increases, the amplitude gradually increases, exhibiting a tuning enhancement effect; Decreasing range: As the thickness continues to increase, the interface between the two wave groups begins to separate, and the amplitude weakens instead. Stable range: The waveforms of the two sets of waves are completely separated, and the amplitude tends to stabilize.
[0008] Furthermore, posterior probability fields for different sand body thickness types are constructed, including: A Gaussian mixture model is used to model the distribution of each thickness type in the multidimensional attribute space. The expectation-maximization algorithm is used to estimate the model parameters. After iterative convergence, the prior probability of the samples is combined with the Bayesian theorem to calculate the posterior probability of each sample belonging to different thickness types, thereby generating a spatial probability volume of thickness types.
[0009] Furthermore, an adaptive learning framework is used to construct multiple sub-models, and sub-models of different thickness types are trained, including: Based on the sand body structure thickness and optimized seismic properties, an input dataset is constructed and divided into a training set and a test set. By automatically adjusting the hyperparameters of each base learner, the automatic training of each sand body thickness sub-model is achieved, thus realizing model training based on an adaptive machine learning framework.
[0010] Furthermore, after model training, model reliability verification and blind testing verification are also included; Model reliability is verified by evaluating model credibility and generalization performance through cross-validation. Blind testing was conducted using blind well data to evaluate the model's prediction accuracy and generalization ability. The model was deemed qualified when the prediction accuracy exceeded the set value.
[0011] Furthermore, in the step of dividing the input dataset into training and test sets, the sand body thickness interpreted from well logging is used as the supervision data for learning. In the training set, a binning and oversampling strategy is used to balance the sample distribution within a single interval, so as to improve the model's ability to identify extreme value intervals and enhance the learning effect of edge samples.
[0012] Furthermore, the output prediction results include: thickness prediction map uncertainty index, which can be used for subsequent reservoir interpretation, sweet spot selection and exploration risk analysis.
[0013] Secondly, the technical solution adopted by this invention is as follows: a Bayesian thickness classification-based intelligent seismic fusion and reservoir prediction system, comprising: an interval division module, which, based on the seismic response law of the amplitude-thickness curve, divides the seismic response characteristics of different sand body thicknesses into four typical intervals, namely, a no-response interval, an increasing interval, a decreasing interval, and a stable interval; a sand body thickness classification module, which, based on the four typical intervals and combined with the actual sand body thickness distribution characteristics and the effective frequency band range of seismic data, iteratively optimizes the thickness boundaries of different seismic response intervals using a Bayesian posterior probability classification method, so as to classify the seismic response intervals corresponding to sand body thicknesses into four categories: no-response sand bodies, thin-layered sand bodies, medium-thick-layered sand bodies, and thick-layered sand bodies; a seismic attribute optimization module, which respectively optimizes the sand body thickness classification sensitive attributes and the sand body thickness regression prediction sensitive attributes, and performs collinearity analysis on each attribute to reduce similar attributes; and a posterior probability field construction module, which constructs the posterior probability field for different sand body thickness types. The probability field, which supports both soft classification representation and sub-model fusion weight constraints, includes: modeling the distribution of each thickness type in the multi-dimensional attribute space using a Gaussian mixture model; estimating model parameters using the expectation-maximization algorithm; combining the prior probability of the sand body thickness proportion of the sample with the iterative convergence; calculating the posterior probability of each sample belonging to different thickness types based on Bayes' theorem; generating the planar distribution probability of thickness types; a sub-model training module, which uses an adaptive learning framework to construct multiple sub-models based on well logging interpretation of sand body thickness and the regression prediction of the selected sand body thickness as a sensitive attribute, and trains sub-models for different thickness types; a multi-model fusion module, which assigns dynamic weights to each trained sub-model through the posterior probability field, driving multi-model fusion to achieve a unified output of thickness prediction; and a prediction output module, which deploys the fused model to the entire reservoir prediction area, and achieves reservoir spatial distribution prediction by intelligently analyzing the seismic attribute data of the study area, and outputs the prediction results.
[0014] Furthermore, the four typical intervals are as follows: No response interval: the sand body is extremely thin, and the amplitude is close to the background noise; increasing interval: as the thickness increases, the amplitude gradually increases, showing a tuning enhancement effect; decreasing interval: as the thickness continues to increase, the interface between the two sets of waves begins to separate, and the amplitude weakens instead; stable interval: the waveforms of the two sets of waves are completely separated, and the amplitude tends to stabilize.
[0015] Thirdly, the technical solution adopted by the present invention is: a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any of the methods described above.
[0016] Fourthly, the technical solution adopted by the present invention is: a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.
[0017] The present invention has the following advantages due to the adoption of the above technical solutions: 1. This invention adopts a soft classification strategy based on Bayesian posterior probability and uses a Gaussian mixture model to realize the probabilistic modeling of thickness type in the attribute space. This not only avoids the boundary misjudgment and error propagation problems caused by hard classification, but also provides dynamic weight constraints for subsequent model fusion, thereby improving the stability and reliability of the prediction results.
[0018] 2. This invention constructs an integrated prediction framework based on "thickness classification and multi-model fusion," classifying sand body thickness into multiple subtypes for targeted regression modeling and introducing posterior probabilities for dynamic weighted fusion. This improves overall prediction accuracy while effectively reducing model complexity and enhancing geological interpretability. The method balances numerical prediction of sand body thickness with the expression of seismic uncertainty, outputting thickness prediction results and uncertainty distribution fields. This provides quantitative support for oil and gas target selection, development plan formulation, and exploration risk control, demonstrating promising practical application prospects. Attached Figure Description
[0019] Figure 1 These are seismic amplitude response characteristic diagrams for different sand body thickness ranges in the seismic forward modeling embodiments of the present invention; Figure 2 This is an overall flowchart of the seismic intelligent fusion and reservoir prediction method under Bayesian thickness classification in this embodiment of the invention; Figure 3 This is a detailed flowchart of the seismic intelligent fusion and reservoir prediction method under Bayesian thickness classification in this embodiment of the invention; Figure 4 This is a flowchart of the sub-model adaptive machine learning method in an embodiment of the present invention; Figure 5 This is a diagram showing the sand body thickness distribution result of intelligent fusion of Bayesian probabilistic constrained seismic attributes in an embodiment of the present invention; Figure 6 This is a reliability distribution diagram of the Bayesian probabilistic constraint earthquake attribute intelligent fusion prediction results in an embodiment of the present invention. Detailed Implementation
[0020] In existing technologies, linear fusion of seismic attributes is a traditional reservoir prediction method. Its core idea is to enhance reservoir identification and sand body distribution characterization by linearly weighting and combining multiple seismic attributes. This method is suitable for situations where the relationships between seismic attributes are relatively simple and has advantages such as ease of implementation and high computational efficiency. A typical technical process is as follows: (1) Seismic attribute extraction: Extract various representative seismic attributes from the target layer, including amplitude, phase, frequency, energy, etc., to enhance the characterization ability of reservoir-related seismic response characteristics.
[0021] (2) Attribute dimensionality reduction and principal component analysis: For the extracted multidimensional attribute set, correlation analysis, principal component analysis (PCA) or factor analysis are used to deal with the collinearity between attributes, retain the main information variables and reduce redundancy.
[0022] (3) Multi-attribute linear fusion: Based on attribute dimensionality reduction and principal component analysis, the selected attributes are linearly superimposed to generate a comprehensive attribute body.
[0023] (4) Sand body distribution characterization: The fused attribute volume is projected onto the seismic profile or three-dimensional view, and the spatial distribution of the reservoir is qualitatively or quantitatively described by means of threshold discrimination or color layering.
[0024] This type of method has been widely used in early seismic reservoir analysis, especially suitable for areas with simple structures and clear reservoir characteristics, providing a useful tool for qualitative identification of sand bodies. However, the above methods have the following problems: (1) The linear fusion method essentially relies on the linear superposition between attributes, which makes it difficult to effectively characterize the highly nonlinear and non-stationary response relationship between seismic attributes and sand body thickness, and it is difficult to meet the prediction accuracy requirements of fine reservoir characterization. (2) The linear method is highly dependent on human experience in the attribute selection and weighting process, lacks adaptive ability, and is difficult to extend to practical application scenarios with high attribute dimensions and complex distribution patterns.
[0025] Furthermore, in existing technologies, sand body thickness prediction based on multi-attribute intelligent fusion is a reservoir characterization method integrating seismic attribute analysis and computer technology, primarily used for quantitative prediction of sand body thickness. This method is based on seismic sedimentology theory and uses machine learning to fuse multiple seismic attributes to achieve quantitative prediction of sand body thickness and fine spatial distribution characterization. Its technical process is as follows: (1) Seismic attribute extraction and standardization: various seismic attributes such as amplitude, frequency and energy are extracted along the target layer and normalized to eliminate scale differences between attributes; (2) Seismic attribute optimization: Preliminary screening is carried out based on the correlation between the attributes and the well point thickness. At the same time, redundant information is eliminated by combining the correlation coefficient matrix between attributes, and a concise and effective feature set is constructed. (3) Multi-attribute intelligent fusion modeling: using machine learning or deep learning algorithms to perform nonlinear fusion modeling of multiple attributes and establish the mapping relationship between seismic attributes and sand body thickness; (4) Fine depiction of sand body distribution: By integrating the trained sand body thickness prediction model with attribute surfaces, the plane boundary of the sand body is clearly defined and the spatial variation of thickness is quantitatively represented.
[0026] This type of method improves the resolution and automation of thickness prediction by fusing information from a high-dimensional attribute space, and has broad application prospects in reservoir characterization. However, the above methods have the following problems: (1) Lack of clear physical constraints and prior guidance. Existing methods are data-driven modeling, and the model parameters are difficult to trace back to specific geological features, resulting in poor interpretability of the prediction results and difficulty in providing reliable geological basis for reservoir evaluation or development decisions. (2) Failure to consider the thickness sensitivity mechanism of seismic response. Existing methods mostly use a uniform model to process all samples, ignoring the piecewise characteristics of nonlinear response caused by tuning effects, which easily causes the model prediction results to deviate from the actual sedimentary laws, especially in thin reservoirs or complex geological conditions where the error is more prominent.
[0027] In order to solve the following problems in the existing technology: (1) The seismic response mechanism is different in different sand body thickness ranges. If a unified model is used, it will be difficult for the model to accurately capture key geological information, and the fusion effect and prediction accuracy will be significantly limited. At the same time, the model lacks a physical constraint mechanism, the structure is seriously "black box", and it is difficult to output interpretable prediction basis and credibility evaluation, which restricts its application value in actual exploration deployment. (2) Although Bayesian methods have been widely applied in areas such as seismic inversion, proving their effectiveness in integrating prior knowledge and outputting uncertain information, the following problems remain unresolved in the seismic attribute fusion sand body thickness prediction method: ① Bayesian classification is highly sensitive to thickness type division. If there is a lack of a division basis based on physical response mechanisms, it is easy to lead to category confusion and accumulation of prediction errors; ② Reservoir thickness prediction usually involves two tasks at the same time: classification and regression. How to construct a unified and collaborative fusion framework to balance type discrimination ability and continuous prediction accuracy is an important challenge currently facing Bayesian modeling in this field; ③ Constructing the likelihood function of each thickness type in the high-dimensional attribute space depends on the accurate modeling of the joint distribution of attributes. However, the current lack of effective feature selection and modeling strategies restricts the accuracy and reliability of probability inference.
[0028] This invention proposes a method, system, medium, and device for intelligent seismic fusion and reservoir prediction based on Bayesian thickness classification. It aims to construct a prediction framework that combines physical response constraints, model robustness, and uncertainty expression capabilities, thereby solving the problems of insufficient accuracy, strong ambiguity, and poor geological interpretation in existing seismic reservoir prediction methods.
[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.
[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0031] The technical terms and abbreviations used in this invention are defined as follows: Reservoir architecture: The morphology, scale, orientation, and stacking relationships of reservoir units of different levels, also known as sedimentary architecture. This concept reflects the differences in the spatial configuration and distribution of reservoir units and seepage barriers of different levels.
[0032] Earthquake attributes: Geometric, kinematic, dynamic, or statistical characteristics of seismic waves derived from seismic data through mathematical transformations, used to describe information such as underground geological structure, lithology, and fluid properties.
[0033] Seismic attribute optimization: For a specific problem to be solved in a specific region, the combination of attributes that is most sensitive to and effective for specific geological targets is selected from a large number of seismic attributes to improve interpretation efficiency and accuracy.
[0034] Seismic attribute fusion: This involves organically combining multiple seismic attributes through mathematical, statistical, or geological methods to form a new information body that comprehensively represents underground geological features, thereby improving the reliability and resolution of seismic interpretation.
[0035] Bayesian posterior probability: Bayesian posterior probability is an estimate of the probability that an event or category is true given observed data. This probability value is calculated based on Bayes' theorem, taking into account both prior probability (i.e., subjective or statistical experience before observing data) and likelihood function (i.e., the probability of observed data occurring under a certain assumption).
[0036] Gaussian Mixture Model: A Gaussian mixture model is a clustering and modeling method based on probability density estimation, used to express the statistical characteristics of data composed of a mixture of multiple Gaussian distributed components. Each Gaussian component represents a subgroup, and its weight represents the proportion of that component in the whole.
[0037] Soft classification is a classification method that outputs probability values or confidence scores, as opposed to hard classification, which directly assigns discrete category labels. Its result is the probability distribution of a sample belonging to each category.
[0038] Adaptive machine learning refers to machine learning methods that automatically adjust model parameters, structure, or combinations of input features to enable the model to adapt to different data characteristics, task requirements, or sample distribution conditions. This method typically combines automatic algorithm selection, parameter optimization, and model ensemble strategies to reduce human intervention, improve modeling efficiency and predictive generalization ability, and is suitable for highly robust modeling tasks in complex or changing environments.
[0039] In one embodiment of the present invention, a method for intelligent seismic fusion and reservoir prediction based on Bayesian thickness classification is provided. In this embodiment, as... Figure 2 , Figure 3 As shown, the method includes the following steps: a) Seismic response thickness type interval classification. Based on the seismic response characteristics of the amplitude-vertus-thickness (AVT) curve, the seismic response characteristics of different sand body thicknesses are divided into four typical intervals: no response interval, increasing interval, decreasing interval, and stable interval. Figure 1 As shown.
[0040] Specifically, this invention proposes for the first time a sand body thickness classification strategy based on the differences in the characteristics of the earthquake amplitude-thickness relationship (AVT curve). From the perspective of physical response mechanisms, it divides continuous thickness variables into several physically meaningful type intervals, significantly simplifying the complexity of the mapping relationship between earthquake attributes and thickness (e.g., Figure 1 As shown in the figure, this effectively improves the interpretability of the model.
[0041] b) Determination of thickness type interval boundaries. Based on four typical intervals, and combined with the actual sand body thickness distribution characteristics and the effective frequency band range of seismic data, the thickness boundaries of different seismic response intervals are iteratively optimized using the Bayesian posterior probability classification method to classify the seismic response intervals corresponding to sand body thickness into four categories: non-responsive sand bodies, thin-layered sand bodies, medium-thick-layered sand bodies, and thick-layered sand bodies.
[0042] Specifically, this invention addresses the characteristics of seismic response where thickness type boundaries are not fixed and drift due to frequency components. It employs a soft classification strategy based on Bayesian posterior probability, introducing a Gaussian mixture model for probabilistic modeling of the attribute space and using iterative optimization to determine interval boundaries. This strategy effectively avoids the boundary uncertainty and error amplification problems of traditional hard classification, thereby enhancing the physical consistency and predictive robustness of the model classification.
[0043] c) Seismic attribute optimization. Sensitive attributes for sand body thickness classification and sensitive attributes for sand body thickness regression prediction are optimized separately, and collinearity analysis is performed on each attribute to reduce similar attributes.
[0044] Specifically, attribute optimization is performed for different task objectives to fully leverage the discriminative and interpretative advantages of each attribute in classification and regression. For seismic attribute optimization in classification problems, the classification effectiveness of attributes needs to be quantitatively evaluated. For attribute optimization in regression problems, correlation analysis with various sand body thickness types is conducted to select the seismic attribute corresponding to the sand body thickness type with the highest correlation. Subsequently, collinearity analysis is performed on each attribute to reduce similar attributes.
[0045] Specifically, under the conditions of numerous earthquake attributes and complex responses, this invention proposes a differentiated attribute optimization strategy to meet the different needs of classification and regression tasks. For classification problems, an attribute "classification efficiency factor" is constructed, and for regression problems, a correlation analysis method is applied to eliminate redundant information.
[0046] d) Construction of posterior probability fields for different sand body thickness types. Posterior probability fields are constructed for different sand body thickness types, supporting both soft classification representation and sub-model fusion weight constraints.
[0047] e) Sub-model adaptive machine learning. Based on the sensitivity attributes of well logging interpretation of sand body thickness and preferred sand body thickness regression prediction, an adaptive learning framework is used to construct multiple sub-models, and sub-models of different thickness types are trained.
[0048] f) Probabilistically Constrained Multi-Model Fusion. Dynamic weights are assigned to each trained sub-model through the posterior probability field, driving multi-model fusion to achieve a unified output for thickness prediction, thereby improving the continuity of overall spatial representation while preserving thickness fractal features.
[0049] g) Seismic reservoir prediction. The fused model is deployed to reservoir prediction across the entire region, especially between wells. By intelligently analyzing the seismic attribute data of the study area, the spatial distribution of reservoirs is predicted, and the prediction results are output.
[0050] In step a) above, the four typical intervals are as follows: No-response range: The sand body is extremely thin, and the amplitude is close to the background noise; Increasing range: As the thickness increases, the amplitude gradually increases, exhibiting a tuning enhancement effect; Decreasing range: As the thickness continues to increase, the interface between the two wave groups begins to separate, and the amplitude weakens instead. Stable range: The waveforms of the two sets of waves are completely separated, and the amplitude tends to stabilize.
[0051] In step d) above, the posterior probability field for different sand body thickness types is constructed. Specifically, a Gaussian mixture model is used to model the distribution of each thickness type in the multidimensional attribute space, the expectation-maximization (EM) algorithm is used to estimate the model parameters, and after iterative convergence, the prior probability of the sand body thickness proportion of the sample is combined with the Bayes theorem to calculate the posterior probability of each sample belonging to different thickness types, thereby generating the planar distribution probability of the thickness type.
[0052] In this embodiment, for the attribute optimization of the classification task, a classification performance factor F is introduced to quantitatively evaluate the ability of different attributes to distinguish between thickness types. This classification performance factor F measures the inter-class differences and intra-class stability of attributes across different thickness types, and attributes with high F values or a set value are selected as classification modeling features.
[0053] in:
[0054] In the formula, Indicates the first Classification efficiency factors for each attribute, For the number of types, For the first Number of samples per class For the first Class in The mean of each attribute, For the full sample at the 1st The mean of each attribute For the first The first in the class The value of each sample on this attribute. The larger the factor value, the more effectively the attribute can distinguish different thickness types.
[0055] In step e) above, such as Figure 4 As shown, an adaptive learning framework is used to construct multiple sub-models, and sub-models of different thickness types are trained, including the following steps: (1) Construction of input data set and training dataset: Based on the sand body thickness interpreted by well logging and the preferred sand body thickness regression prediction sensitive attribute, the input data set (i.e., sand body thickness interpreted by well logging, sand body thickness classification label, and preferred seismic attribute) is constructed, and the input data set is divided into training set and test set.
[0056] In this embodiment, the available samples are randomly divided into a training set and a test set in a 7:3 ratio. In the step of dividing the input dataset into training and test sets, the sand body thickness interpreted from well logging is used as supervised learning data. Considering the uneven distribution of thickness samples, a binning and oversampling strategy is employed in the training set to balance the sample distribution within a single interval. This improves the model's ability to identify extreme value intervals, enhances the learning effect of edge samples, and reduces discontinuous jumps caused by class mutations.
[0057] (2) Model training based on adaptive machine learning framework: By automatically adjusting the hyperparameters of each base learner, the automatic training of each sand body thickness sub-model is achieved, thus realizing model training based on adaptive machine learning framework. This avoids the tedious process of manually adjusting parameters and reduces the impact of human factors on model performance.
[0058] After model training, the process includes model reliability verification and blind testing: Model reliability is verified by evaluating the model's credibility and generalization performance through cross-validation; blind testing uses blind well data to test the model's prediction accuracy and generalization ability, and the model is deemed qualified when the prediction accuracy is greater than a set value. For example, the model is deemed qualified when the prediction accuracy is ≥90%; otherwise, it is re-optimized.
[0059] In step g) above, the output prediction results include: an uncertainty index for the thickness prediction map, which is used for subsequent reservoir interpretation, sweet spot selection, and exploration risk analysis, such as... Figure 5 , Figure 6 As shown.
[0060] In summary, this invention constructs a multi-attribute fusion framework that integrates thickness type classification with a Bayesian dynamic weighting mechanism. Each thickness type trains an independent regression model, and during the prediction phase, the outputs of each sub-model are dynamically weighted and fused using posterior probabilities as weights, thereby achieving a unified expression and spatially continuous prediction of sand body thickness. This method simultaneously considers the physical constraints of seismic-geological mapping, the ability to quantify the uncertainty of prediction results, and the interpretability of the model structure.
[0061] In one embodiment of the present invention, a seismic intelligent fusion and reservoir prediction system based on Bayesian thickness classification is provided, comprising: The interval division module, based on the seismic response law of the amplitude-thickness curve, divides the seismic response characteristics of different sand body thicknesses into four typical intervals: no response interval, increasing interval, decreasing interval, and stable interval. The sand body thickness classification module, based on four typical intervals, combines the actual sand body thickness distribution characteristics with the effective frequency band range of seismic data. It uses the Bayesian posterior probability classification method to iteratively optimize the thickness boundaries of different seismic response intervals, so as to classify the seismic response intervals corresponding to sand body thickness into four categories: non-responding sand bodies, thin sand bodies, medium-thick sand bodies, and thick sand bodies. The earthquake attribute optimization module optimizes the sand body thickness classification sensitive attributes and the sand body thickness regression prediction sensitive attributes respectively, and performs collinearity analysis on each attribute to reduce similar attributes; The posterior probability field construction module constructs posterior probability fields for different sand body thickness types. This posterior probability field simultaneously supports soft classification expression and sub-model fusion weight constraints. The sub-model training module, based on the sensitive attributes of well logging interpretation of sand body thickness and preferred sand body thickness regression prediction, uses an adaptive learning framework to construct multiple sub-models and trains sub-models for different thickness types. The multi-model fusion module assigns dynamic weights to each trained sub-model through the posterior probability field, driving multi-model fusion to achieve a unified output for thickness prediction. The prediction output module deploys the fused model to the entire reservoir prediction area. By intelligently analyzing the seismic attribute data of the study area, it realizes the prediction of reservoir spatial distribution and outputs the prediction results.
[0062] In the above embodiments, the four typical intervals are as follows: No-response range: The sand body is extremely thin, and the amplitude is close to the background noise; Increasing range: As the thickness increases, the amplitude gradually increases, exhibiting a tuning enhancement effect; Decreasing range: As the thickness continues to increase, the interface between the two wave groups begins to separate, and the amplitude weakens instead. Stable range: The waveforms of the two sets of waves are completely separated, and the amplitude tends to stabilize.
[0063] In the above embodiments, the posterior probability fields for different sand body thickness types are constructed, including: A Gaussian mixture model is used to model the distribution of each thickness type in the multidimensional attribute space. The expectation-maximization algorithm is used to estimate the model parameters. After iterative convergence, the prior probability of the sand body thickness ratio of the sample is combined with the Bayesian theorem to calculate the posterior probability of each sample belonging to different thickness types, and the planar distribution probability of the thickness type is generated.
[0064] In the above embodiments, an adaptive learning framework is used to construct multiple sub-models, and sub-models of different thickness types are trained, including: Based on the sensitivity attributes of well logging interpretation of sand body thickness and preferred sand body thickness regression prediction, an input dataset is constructed and divided into a training set and a test set. By automatically adjusting the hyperparameters of each base learner, the automatic training of each sand body thickness sub-model is achieved, thus realizing model training based on an adaptive machine learning framework.
[0065] In the above embodiments, after model training, model reliability verification and blind testing verification are also included; Model reliability is verified by evaluating model credibility and generalization performance through cross-validation. Blind testing was conducted using blind well data to evaluate the model's prediction accuracy and generalization ability. The model was deemed qualified when the prediction accuracy exceeded the set value.
[0066] In the above embodiments, in the step of dividing the input data set into a training set and a test set, the sand body thickness interpreted by well logging is used as the supervision data for learning. In the training set, a binning and oversampling strategy is used to balance the sample distribution within a single interval, so as to improve the model's ability to identify extreme value intervals and enhance the learning effect of edge samples.
[0067] In the above embodiments, the output prediction results include: thickness prediction map uncertainty index, which is used for subsequent reservoir interpretation, sweet spot selection and exploration risk analysis.
[0068] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.
[0069] The beneficial effects achieved by the present invention through the above technical solution are as follows: The sand body thickness classification method based on the AVT response curve utilizes the amplitude variation trend under seismic tuning effect to divide the thickness response characteristics into multiple intervals with similar physical mechanisms. By combining well logging interpretation data and seismic dominant frequency characteristics of the study area, the thickness type classification boundary is iteratively optimized within the response interval, forming a geologically and physically consistent thickness type classification standard, providing a foundation for subsequent sand body thickness classification fusion.
[0070] Furthermore, this invention proposes a posterior probability calculation method combining Gaussian mixture models and Bayes' theorem. This method fits the probability density distribution of each thickness type in a multidimensional seismic attribute space, calculates the posterior probability of each sample belonging to its respective thickness type, and achieves a soft classification representation of thickness types. Compared to traditional hard classification methods, this strategy significantly reduces the error propagation problem caused by boundary ambiguity. Simultaneously, a seismic information confidence distribution map is constructed, which can be used to quantify the prediction confidence level of the model in various regions, thereby providing support for reservoir interpretation and development deployment risk prediction.
[0071] Furthermore, based on the thickness classification results, this invention constructs different sub-models to fit the attribute response features of each thickness range, significantly reducing the complexity of a single model and improving the geological interpretability of the results. Finally, by introducing a posterior probability field as weights, the prediction results of each sub-model are dynamically fused and output, forming a unified and continuous thickness distribution prediction map, balancing prediction accuracy and consistency with geological interpretation.
[0072] In one embodiment of the present invention, a computing device is provided. This computing device can be a terminal and may include a processor, a communication interface, memory, a display screen, and an input device. The processor, communication interface, and memory communicate with each other via a communication bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. When the computer programs are executed by the processor, they implement the methods described in the above embodiments. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, a management network, NFC (Near Field Communication), or other technologies. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad mounted on the casing of the computing device, or an external keyboard, touchpad, or mouse. The processor can call logical instructions stored in the memory.
[0073] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0074] In one embodiment of the present invention, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the methods provided in the above-described method embodiments.
[0075] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, which stores server instructions that cause a computer to perform the methods provided in the above embodiments.
[0076] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.
[0077] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent fusion of seismic and reservoir prediction under Bayesian thickness classification, characterized in that, The method comprises the following steps: Based on the amplitude-thickness curve, the seismic response characteristics of different sand body thicknesses are divided into four typical intervals, namely, no response interval, increasing interval, decreasing interval and stable interval; Based on the four typical intervals, combined with the actual sand body thickness distribution characteristics and the effective frequency band range of the seismic data, the thickness boundaries of different seismic response intervals are iteratively optimized by the Bayesian posterior probability classification method, so as to divide the sand body thickness corresponding to the seismic response interval into four types, namely, no response sand body, thin layer sand body, medium-thick layer sand body and thick layer sand body; Sensitive attributes for sand body thickness classification and sensitive attributes for sand body thickness regression prediction are respectively optimized, and collinearity analysis is performed on each attribute to reduce similar attributes; A posterior probability field of different sand body thickness types is constructed, which supports soft classification expression and sub-model fusion weight constraint at the same time, including: a Gaussian mixture model is used to model the distribution of each thickness type in the multi-dimensional attribute space, the expectation maximization algorithm is used to estimate the model parameters, and after iterative convergence, combined with the prior probability of the sample sand body thickness proportion, the posterior probability of each sample belonging to different thickness types is calculated according to the Bayes theorem to generate the planar distribution probability of the thickness type; Based on the well-logging interpreted sand body thickness and the optimized sand body thickness regression prediction sensitive attributes, a plurality of sub-models are constructed by using an adaptive learning framework, and different thickness type sub-models are trained; Each trained sub-model is given a dynamic weight by the posterior probability field to drive multi-model fusion, so as to realize unified output of thickness prediction; The fused model is deployed to the whole area reservoir prediction, the seismic attribute data of the research area is intelligently analyzed, the reservoir spatial distribution prediction is realized, and the prediction result is output.
2. The method of claim 1, wherein the Bayesian thickness classification under seismic intelligent fusion and reservoir prediction is characterized by, The four typical intervals are as follows: No response interval: the sand body is extremely thin, and the amplitude is approximately background noise; Increasing interval: with the increase of thickness, the amplitude gradually increases, showing a tuning enhancement effect; Decreasing interval: with the continuous increase of thickness, the interface of the two groups of waves begins to separate, and the amplitude decreases instead; Stable interval: the waveforms of the two groups of waves are completely separated, and the amplitude tends to be stable.
3. The method of claim 1, wherein the Bayesian thickness classification under seismic intelligent fusion and reservoir prediction is characterized by, A plurality of sub-models are constructed by using an adaptive learning framework, and different thickness type sub-models are trained, including: Based on the well-logging interpreted sand body thickness and the optimized sand body thickness regression prediction sensitive attributes, an input data set is constructed, and the input data set is divided into a training set and a test set; Through automatic adjustment of the hyperparameters of each base learner, the automatic training of each sand body thickness sub-model is realized, and the model training based on the adaptive machine learning framework is realized.
4. The method of claim 3, wherein the Bayesian thickness classification is used for intelligent fusion of seismic and reservoir prediction. In the step of dividing the input data set into a training set and a test set, the well-logging interpreted sand body thickness is used as the supervised data for learning. Considering the problem of uneven distribution of thickness samples, the sample distribution in a single interval is balanced by using a binning combined oversampling strategy in the training set, so as to improve the identification ability of the model to the extreme interval and enhance the learning effect of the edge samples.
5. The method of claim 3, wherein the Bayesian thickness classification is used for intelligent fusion of seismic and reservoir prediction. After the model training, model reliability verification and blind test verification are also included; The model reliability verification is performed by cross-validation to evaluate the credibility and generalization performance of the model; The blind test verification tests the prediction accuracy and generalization ability of the model by using blind well data, and the model is determined to be qualified when the prediction accuracy is greater than a set value.
6. The method of claim 1, wherein the Bayesian thickness classification under seismic intelligent fusion and reservoir prediction is characterized by, The output prediction result comprises a thickness prediction map uncertainty index, which is used for subsequent reservoir interpretation, sweet spot optimization and exploration risk analysis.
7. A system for intelligent fusion of seismic and reservoir prediction under Bayesian thickness classification, characterized in that, The method comprises the following steps: The interval division module divides the seismic response characteristics of different sand body thicknesses into four typical intervals based on the seismic response law of the amplitude-thickness curve, and the four typical intervals are a non-response interval, an increasing interval, a decreasing interval and a stable interval; The sand body thickness classification module, based on the four typical intervals, combines the actual sand body thickness distribution characteristics and the effective frequency band range of the seismic data, and iteratively optimizes the thickness boundaries of different seismic response intervals by using a Bayesian posterior probability classification method, so as to divide the sand body thickness corresponding to the seismic response interval into four types, namely, a non-response sand body, a thin layer sand body, a medium-thick layer sand body and a thick layer sand body; The seismic attribute optimization module respectively optimizes the sand body thickness classification sensitive attribute and the sand body thickness regression prediction sensitive attribute, and performs collinearity analysis on each attribute to reduce similar attributes; The posterior probability field construction module constructs a posterior probability field of different sand body thickness types, which simultaneously supports soft classification expression and sub-model fusion weight constraint, including: using a Gaussian mixture model to model the distribution of each thickness type in a multi-dimensional attribute space, using an expectation maximization algorithm to estimate the model parameters, and after iterative convergence, combining the sample sand body thickness proportion prior probability to calculate the posterior probability of each sample belonging to different thickness types according to the Bayes theorem, and generating a planar distribution probability of the thickness type; The sub-model training module, based on the well-logging interpreted sand body thickness and the optimized sand body thickness regression prediction sensitive attribute, uses an adaptive learning framework to construct multiple sub-models, and trains sub-models of different thickness types; The multi-model fusion module gives dynamic weights to each trained sub-model through the posterior probability field, drives multi-model fusion, and realizes unified output of thickness prediction; The prediction output module deploys the fused model to the whole area reservoir prediction, intelligently analyzes the seismic attribute data of the research area, realizes reservoir spatial distribution prediction, and outputs the prediction result.
8. The system for intelligent fusion of seismic and reservoir prediction under Bayesian thickness classification as claimed in claim 7, wherein, The four typical intervals are as follows: The non-response interval: the sand body is extremely thin, and the amplitude is approximately background noise; The increasing interval: as the thickness increases, the amplitude gradually increases, showing a tuning enhancement effect; The decreasing interval: as the thickness continues to increase, the interface of the two groups of waves begins to separate, and the amplitude decreases; The stable interval: the waveforms of the two groups of waves are completely separated, and the amplitude tends to be stable.
9. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method of any of claims 1-8. The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods of claims 1-6.
10. A computing device, comprising: The method comprises the following steps: One or more processors, memories and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing any of the methods of claims 1-6.