Method and system for automatically dividing fluvial facies reservoir small layers based on intelligent image analysis

Abstract

The invention discloses a fluvial facies reservoir small layer automatic division method and system based on image intelligent analysis, and relates to the technical field of geological exploration reservoir small layer division research, and the specific steps are as follows: obtaining historical divided small layer image data of a target fluvial facies reservoir, and performing enhancement processing to obtain a divided small layer boundary enhancement feature set; obtaining a confidence score based on a preset confidence score model, and presetting a confidence score threshold to obtain a divided small-layer boundary enhancement feature set after confidence evaluation; performing data set division on the divided small-layer boundary enhancement feature set after confidence evaluation, and training a small-layer boundary feature recognition model based on the data set; inputting to-be-divided target fluvial facies reservoir image data into the trained small-layer boundary feature recognition model to obtain a small-layer boundary feature probability graph; establishing a small-layer constraint rule base to obtain a small-layer division scheme; and obtaining a small-layer division result map, constructing a result map library, and presetting an updating period to update the result map library.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological exploration reservoir sublayer division research, in particular to a river facies reservoir sublayer automatic division method and system based on image intelligent analysis. BACKGROUND

[0002] As a typical sedimentary rock reservoir type, the sublayer division of river facies reservoirs involves complex geological features and hierarchical structures, and is greatly affected by changes in the sedimentary environment; in the process of oil and gas exploration, accurate division of river facies reservoir sublayers is of great significance for oil and gas reservoir evaluation, rock layer distribution of seepage oil and gas, and determination of injection-production well pattern deployment scheme design.

[0003] In a Chinese invention application with application publication number CN115032692A, a river facies reservoir sedimentary microfacies division method and device based on seismic waveform clustering are disclosed; including: determining the sediment source direction according to the seismic amplitude attribute of the river facies reservoir; cutting in the sediment source direction to obtain a seismic section; dividing the sedimentary microfacies type on the seismic section to determine the type of sedimentary microfacies; using a seismic waveform clustering method to classify the seismic waveforms of the river facies reservoir; combining the determined type of sedimentary microfacies and the obtained seismic waveform classification to calibrate the planar seismic waveform; and dividing the planar sedimentary microfacies of the river facies reservoir according to the seismic waveform. This initial clustering center seismic waveform selection method can ensure that the seismic waveforms corresponding to different sedimentary microfacies types have obvious differences, which is beneficial to subsequent waveform clustering.

[0004] In a Chinese invention application with application publication number CN117724151A, a method for constructing a low-frequency model of a river facies dense heterogeneous reservoir is disclosed, including: dividing the river facies dense heterogeneous reservoir into lithofacies types and counting the proportion of each lithofacies; obtaining the P-wave impedance, S-wave impedance and density parameter inversion three-dimensional data volume after the first inversion; obtaining the P-wave impedance, S-wave impedance and density parameter low-frequency model after the last iteration, and the P-wave impedance, S-wave impedance and density parameter three-dimensional data volume after the last iteration of pre-stack inversion; calculating the P-S wave velocity ratio data volume to obtain the distribution rule of the river facies dense sandstone reservoir. The method for constructing a low-frequency model of a river facies dense heterogeneous reservoir disclosed in the present application provides an accurate low-frequency model for reservoir prediction, and the pre-stack inversion result can better depict the channel distribution characteristics.

[0005] The prior art has the following problems: the traditional division of small layers of fluvial facies reservoirs relies on manual experience to identify small layer boundaries, which is inefficient and prone to human judgment errors, resulting in inaccurate small layers; at the same time, the small layer boundary recognition model is not fully trained based on image analysis technology using historical small layer image data of fluvial facies reservoirs, and the image analysis technology field is less used; the effective geological constraint rules are not established combined with the sedimentary rules of fluvial facies reservoirs, resulting in deviation of the division results from the true geological structure; resulting in low accuracy and insufficient automation in the small layer division process of fluvial facies reservoirs, which is difficult to meet the needs of complex reservoir fine geological research and development. Therefore, an automatic small layer division method and system for fluvial facies reservoirs based on image intelligent analysis are needed to solve the problems existing in the prior art. SUMMARY

[0006] To solve the above technical problems, one aspect of the purpose of the present application provides an automatic small layer division method for fluvial facies reservoirs based on image intelligent analysis, which comprises the following steps: Step 1: Obtain historical small layer image data of the target fluvial facies reservoir, enhance the small layer boundary of the image data, and extract the enhanced small layer boundary features to obtain an enhanced small layer boundary feature set; Step 2: Based on a pre-set confidence score model, the enhanced small layer boundary features are evaluated for confidence score, a confidence score threshold is pre-set, and the enhanced small layer boundary features less than the pre-set confidence score threshold are removed to obtain the enhanced small layer boundary feature set after confidence evaluation; Step 3: The enhanced small layer boundary feature set after confidence evaluation is divided into a data set, the data set includes: a training set, a test set and a validation set; based on the data set, a small layer boundary feature recognition model is trained to obtain a trained small layer boundary feature recognition model; Step 4: Input the image data of the target fluvial facies reservoir to be divided into the trained small layer boundary feature recognition model to obtain a small layer boundary feature probability map; Step 5: Based on the sedimentary rules of fluvial facies reservoirs, a small layer constraint rule library is established, and based on the small layer constraint rule library and the small layer boundary feature probability map, a small layer division scheme is obtained; Step 6: Input the small layer division scheme into a pre-set visualization software to obtain a small layer division result map; build a result map library, store the small layer division result map in the result map library, and pre-set an update period to update the result map library.

[0007] In a preferred embodiment, the image data specifically includes: well logging image data, core image data, thin section image data, and seismic profile image data.

[0008] In a preferred embodiment, the divided sublayer boundary enhancement features specifically include: sediment grain size features, sediment facies features, permeability features, porosity features, texture features, lithology features, fluid distribution features, boundary thickness, boundary curvature.

[0009] In a preferred embodiment, the specific process of performing confidence evaluation on the divided sublayer boundary enhancement features based on a pre-set confidence scoring model includes: Converting the divided sublayer boundary enhancement features into a mode of vector to obtain a boundary enhancement feature vector; Inputting the boundary enhancement feature vector into the pre-set confidence scoring model, and performing non-linear mapping of the boundary enhancement feature vector to a high-dimensional vector space by a multi-layer perception inside the model; Compressing the boundary enhancement feature vector in the high-dimensional vector space to the interval of 0 to 1 based on a softmax function to obtain a confidence score.

[0010] In a preferred embodiment, the specific process of obtaining the trained sublayer boundary feature recognition model includes: Dividing the set of divided sublayer boundary enhancement features after confidence evaluation into a data set, the data set including: a training set, a test set, and a validation set; Building a classifier and inputting the training set into the classifier for training to obtain a trained classifier; using cross-entropy loss function as the training target, learning rate is set to , and training period is set to T; Using the validation set to verify and evaluate the trained classifier based on evaluation indexes, the evaluation indexes including: accuracy, precision, recall, and F1 score; Using the test set to test the trained classifier, and when the test result meets the evaluation indexes, obtaining the trained sublayer boundary feature recognition model.

[0011] In a preferred embodiment, the specific process of obtaining the sublayer boundary feature probability map includes: Inputting the target fluvial facies reservoir image data to be divided into the trained sublayer boundary feature recognition model; The model recognizes the sublayer boundary features of the target fluvial facies reservoir image data, and performs non-linear transformation on the sublayer boundary features to enhance the representation of complex sublayer boundaries by the model; After the non-linear transformation of the sublayer boundary features, the model calculates and converts the possibility of the pixel points in the target fluvial facies reservoir image data belonging to the sublayer boundary into a probability value through an activation function; The model outputs the sublayer boundary feature probability map from the output layer.

[0012] In a preferred embodiment, the small-layer boundary feature probability map specifically includes: A two-dimensional or three-dimensional probability numerical matrix corresponding to the spatial dimension of the target river facies reservoir image data to be delineated; Each element in the probability plot corresponds to a spatial location in the image data of the target fluvial reservoir to be delineated; the value of each element in the probability plot represents the probability that the location is a sublayer boundary. It also includes information on local probability peaks and probability gradient changes.

[0013] In a preferred embodiment, the low-level constraint rule base specifically includes: Sedimentary rhythm constraint rules, lithological combination constraint rules, physical property variation constraint rules, thickness rationality constraint rules, boundary continuity constraint rules, and well logging response characteristic constraint rules.

[0014] In a preferred embodiment, the specific process of obtaining the small-layer partitioning scheme based on the small-layer constraint rule base and the small-layer boundary feature probability map includes: Threshold segmentation is performed on the small-layer boundary feature probability map to extract all pixels or depth points with probability values ​​greater than the preset probability threshold, thus obtaining boundary candidate points. Connectivity analysis and clustering are performed on candidate boundary points to merge spatially adjacent candidate boundary points and obtain candidate boundaries; The candidate boundaries are constrained based on the small-level constraint rule base to obtain the small-level partitioning scheme; The sublayer division scheme specifically includes: the sublayer division number, the top depth boundary of the sublayer, the bottom depth boundary of the sublayer, and the sublayer thickness.

[0015] Another aspect of this invention discloses an automatic segmentation system for fluvial facies reservoir layers based on intelligent image analysis, the system comprising the following modules: The module includes data acquisition and processing, confidence scoring, model training, probability graph generation, partitioning scheme generation, and visualization and updating. Data acquisition and processing module: acquires historical sub-layer image data of the target fluvial reservoir, enhances the boundaries of the sub-layers in the image data, and extracts the enhanced features of the sub-layer boundaries to obtain a set of enhanced features of the sub-layer boundaries; Confidence scoring module: Based on a pre-set confidence scoring model, the confidence of the divided sub-layer boundary enhancement features is evaluated to obtain a confidence score. A pre-set confidence score threshold is set, and the divided sub-layer boundary enhancement features with a score lower than the pre-set confidence score threshold are removed to obtain the set of divided sub-layer boundary enhancement features after confidence evaluation. Model training module: The set of subdivided small-layer boundary enhancement features after confidence evaluation is divided into a dataset, which includes a training set, a test set, and a validation set; a small-layer boundary feature recognition model is trained based on the dataset to obtain the trained small-layer boundary feature recognition model; Probability map generation module: Input the target fluvial reservoir image data to be divided into the trained sublayer boundary feature recognition model to obtain the sublayer boundary feature probability map; The subdivision scheme generation module: Based on the sedimentary patterns of fluvial reservoirs, a sublayer constraint rule base is established. Based on the sublayer constraint rule base and the sublayer boundary feature probability map, a sublayer subdivision scheme is obtained. Visualization and Update Module: Input the sub-layer partitioning scheme into a preset visualization software to obtain the sub-layer partitioning result map; construct a result map library, store the sub-layer partitioning result map in the result map library, and preset the update cycle to update the result map library.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention acquires historical sub-layer image data of a target fluvial reservoir, enhances the boundaries of the sub-layers in the image data, and extracts the enhanced features of the sub-layer boundaries to obtain a set of enhanced features of the sub-layer boundaries; based on a pre-set confidence scoring model, the enhanced features of the sub-layer boundaries are evaluated to obtain a confidence score; a pre-set confidence score threshold is set, and enhanced features of the sub-layer boundaries that are less than the pre-set confidence score threshold are removed to obtain a set of enhanced features of the sub-layer boundaries after confidence evaluation; the set of enhanced features of the sub-layer boundaries after confidence evaluation is then divided into datasets, and... The dataset includes a training set, a test set, and a validation set. A sublayer boundary feature recognition model is trained based on this dataset to obtain a trained sublayer boundary feature recognition model. Image data of the target fluvial reservoir to be delineated is input into the trained sublayer boundary feature recognition model to obtain a sublayer boundary feature probability map. Based on the sedimentary patterns of fluvial reservoirs, a sublayer constraint rule base is established. Based on the sublayer constraint rule base and the sublayer boundary feature probability map, a sublayer delineation scheme is obtained. The sublayer delineation scheme is input into a preset visualization software to obtain the sublayer delineation result map. A result map library is constructed, and the sublayer delineation result maps are stored in the result map library, with a preset update cycle to update the result map library. This reduces subjective human intervention, improves the accuracy and efficiency of sublayer delineation, and provides high-precision technical support for the detailed study of fluvial reservoirs and oil and gas development. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart illustrating the steps of an automatic segmentation method for fluvial reservoir sublayers based on image intelligent analysis, according to an embodiment of this application.

[0018] Figure 2 This is a schematic diagram showing the connection of various modules in the automatic segmentation system for fluvial reservoir sublayers based on image intelligent analysis, as described in an embodiment of this application. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0020] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] Example 1 Please see Figure 1 As shown in the figure, this application provides an automatic method for dividing fluvial facies reservoir sublayers based on intelligent image analysis. The method includes the following steps: Step 1: Obtain historical sub-layer image data of the target fluvial reservoir, enhance the sub-layer boundaries of the image data, and extract the sub-layer boundary enhancement features to obtain a set of sub-layer boundary enhancement features; Based on the above embodiments, the image data of the target fluvial reservoir history divided into small layers can be obtained by downhole instruments (such as various sensors) in oil and gas exploration. The image data specifically includes, but is not limited to: well logging image data, core image data, thin section image data, and seismic profile image data. Well logging image data specifically includes: resistivity maps, sonic maps, density maps, gamma-ray maps, etc.; resistivity maps can be used to distinguish different lithologies, especially distinguishing between highly water-saturated clay rock layers and highly permeable sandstone layers. Highly water-saturated rock layers have lower resistivity, while highly permeable sandstone layers have higher resistivity; sonic maps show a close correlation between sonic velocity (or sonic transit time) and porosity. Higher porosity is usually accompanied by lower sonic velocity, therefore, analyzing sonic maps can estimate reservoir porosity; density maps measure the density of the rock... It helps identify different lithologies (such as sandstone, shale, etc.); generally, sandstone has a lower density, while shale has a higher density. Changes in rock density can help delineate the boundaries of reservoir layers; gamma-ray images can identify clay content because clay minerals (such as high-alumina ore) usually have a high gamma-ray response. Changes in gamma rays can identify different lithologies such as sandstone, mudstone, and shale. Sharp changes in gamma rays indicate the boundaries of layers of different lithologies. Therefore, acquiring well logging image data helps in the delineation of reservoir layers. Core image data is obtained from core samples extracted from downhole. High-resolution imaging techniques (such as CT scanning or microscopic imaging) are used to record three-dimensional structural image data of the core, which can provide detailed information on lithology, porosity, fractures, mineral composition, etc., to help accurately analyze the geological characteristics of the reservoir. Thin section image data are obtained by slicing core samples extracted from the well into sections thinned to a thickness of a few centimeters or micrometers (the specific thickness can be determined by the experimenter based on the actual situation) and observing and photographing them using a microscope. Thin section images help analyze the microstructural characteristics of the reservoir and provide a basis for the microstructural characteristics of the rock for the division of sublayers. Seismic profile image data, which is a stratigraphic reflection image obtained by seismic exploration technology, can help identify the structural features of reservoirs, such as geological structures like faults and folds between layers, and plays a key role in the subdivision and analysis of layers. The specific process of enhancing the image data by dividing it into small-layer boundaries and extracting the enhanced features of the divided small-layer boundaries to obtain the enhanced feature set of the divided small-layer boundaries includes: The image data is preprocessed, specifically including: noise reduction, contrast enhancement, and normalization processing using a preset normalization threshold. The boundaries of the pre-divided small-layer regions are extracted from the pre-processed image data, and the boundaries of the pre-divided small-layer regions are enhanced based on a preset region boundary enhancement algorithm. Extract the enhanced features of the subdivided sub-layer boundaries from the enhanced sub-layer regions; Collect all the boundary enhancement features of the divided sub-layers, construct the boundary enhancement feature set, and obtain the boundary enhancement feature set of the divided sub-layers; Specifically, the enhanced features of the divided sublayer boundaries include, but are not limited to: sediment grain size characteristics, sedimentary facies characteristics, permeability characteristics, porosity characteristics, texture characteristics, lithological characteristics, fluid distribution characteristics, boundary thickness, and boundary curvature. The preset region boundary enhancement algorithm, which uses machine learning or deep learning algorithms (such as edge detection algorithms, YOLO series algorithms, etc.), can extract the boundaries of the divided sub-layer regions from the historical image data of the target fluvial reservoir, enhance the boundary features, and reduce manual intervention. Step 2: Based on the pre-set reliability scoring model, evaluate the confidence of the divided sub-layer boundary enhancement features to obtain a confidence score. Set a pre-set reliability score threshold and remove the divided sub-layer boundary enhancement features that are lower than the pre-set reliability score threshold to obtain the set of divided sub-layer boundary enhancement features after confidence evaluation. Based on the above embodiments, the specific process of evaluating the confidence of the enhanced features at the divided sub-layer boundaries based on a pre-set confidence scoring model to obtain a confidence score includes: The pattern of transforming the divided small-layer boundary enhancement features into vectors is used to obtain the boundary enhancement feature vector; The boundary enhancement feature vector is input into a pre-set reliability scoring model, and the multilayer perceptron inside the model performs a nonlinear mapping of the boundary enhancement feature vector to a high-dimensional vector space. The confidence score is obtained by compressing the boundary enhancement feature vector of the high-dimensional vector space to the interval of 0 to 1 based on the softmax function. Specifically, traditional manual segmentation of layers is often affected by subjective experience and human error, leading to inaccurate segmentation of some boundary features. Confidence scoring models can effectively eliminate boundary features with low scores (i.e., low confidence), reducing segmentation errors caused by erroneous boundary features. Eliminating unreliable boundary features improves the accuracy of subsequent layer segmentation schemes and the training of layer boundary feature recognition models, making the feature set on which the model is based more precise, thus facilitating accurate identification and segmentation of layer boundaries. Low-confidence boundary features often negatively impact the performance of trained models. By removing these low-confidence features, the interference of noisy data on model training can be reduced, thereby improving the quality of training data and enhancing the generalization ability of the small-layer boundary recognition model. Under different reservoir types and geological environments, after screening through confidence scoring, the model can more stably adapt to various situations and is less susceptible to the influence of abnormal or noisy features, thus enhancing the stability and repeatability of the model. It also reduces the reliance on manual adjustments and subsequent manual checks, making the automatic segmentation process more efficient and reliable. Based on actual needs, a confidence score threshold is set. This threshold is used to determine which boundary enhancement features are reliable and which are not. For example, 0.8 means that boundary enhancement features with a score greater than or equal to this value are considered valid. If the confidence score is less than 0.7, the boundary enhancement feature is considered unreliable and needs to be removed. The threshold can be selected by cross-validation or empirical methods to determine the most suitable threshold, or it can be dynamically adjusted to optimize according to different datasets and different types of sublayers. No specific limitations are made here. Specifically, pre-set reliability scoring models include: decision tree model, random forest model, Naive Bayes network model, etc. Step 3: Divide the set of subdivided boundary enhancement features after confidence evaluation into a dataset, which includes a training set, a test set, and a validation set; train the subdivided boundary feature recognition model based on the dataset to obtain the trained subdivided boundary feature recognition model; Based on the above embodiments, the specific process of obtaining the trained small-layer boundary feature recognition model includes: The set of enhanced features with divided small-layer boundaries after confidence evaluation is divided into a dataset, which includes a training set, a test set, and a validation set. A classifier is constructed, and the training set is input into the classifier for training to obtain a trained classifier; the cross-entropy loss function is used as the training objective, and the learning rate is set to... The training period is set to T; The trained classifier is validated and evaluated using a validation set based on evaluation metrics, including: accuracy, precision, recall, and F1 score. The trained classifier is tested using a test set. When the test results meet the evaluation criteria, the trained small-layer boundary feature recognition model is obtained. Specifically, the set of enhanced features with defined sub-layer boundaries after confidence assessment is divided into datasets. The ratio of training, test, and validation sets can be (e.g., a combination of 7:2:1 or 8:1:1). Dividing the dataset ensures that the model learns effectively on the training set, is optimized and tuned on the validation set, and its true generalization ability is evaluated on the test set. This avoids overfitting and underfitting problems, improving the model's stability and adaptability. By reasonably dividing the dataset, computational resources and time are effectively utilized, ensuring the fairness, objectivity, and final performance of the model. Specifically, the classifier includes, but is not limited to, deep learning convolutional neural networks (CNN), recurrent neural networks (RNN), etc.; the cross-entropy loss function is used as the training objective to measure the difference between the actual boundary of the small layer and the boundary predicted by the model through enhanced feature recognition; the smaller the difference, the better the model's performance; the learning rate can be set to 0.001, and the learning rate can be dynamically adjusted during training through a learning rate decay strategy (such as cosine annealing or step decay) to avoid the model getting stuck in local optima and prevent overfitting; the training cycle can be adjusted according to the complexity of the dataset and the learning capacity of the model, such as 64, 128, etc. Step 4: Input the target fluvial reservoir image data to be classified into the trained sublayer boundary feature recognition model to obtain the sublayer boundary feature probability map; Based on the above embodiments, the specific process for obtaining the small-layer boundary feature probability map includes: The target fluvial reservoir image data to be delineated is input into the trained sublayer boundary feature recognition model; The model identifies sub-layer boundary features from target fluvial reservoir image data and performs nonlinear transformations on these features to enhance the model's representation of complex sub-layer boundaries. After performing a nonlinear transformation on the sublayer boundary features, the model calculates and converts the probability of a pixel in the target fluvial reservoir image data belonging to the sublayer boundary into a probability value using an activation function. The boundary feature probability map of the sub-layer is obtained from the output of the model output layer; Based on the above embodiments, the sublayer boundary feature probability map specifically includes: a two-dimensional or three-dimensional probability value matrix corresponding to the target fluvial reservoir image data to be divided in spatial dimension; each element in the probability map corresponds to a spatial location (which may be a pixel, depth point, or spatial coordinate point) in the target fluvial reservoir image data to be divided; the value of each element in the probability map represents the probability that the location is a sublayer boundary. Specifically, the probability map of sublayer boundary features intuitively reflects the probability distribution of sublayer boundaries at different depths or locations in the image data of the target fluvial reservoir to be delineated; areas with higher probability values ​​(usually with a threshold of ≥0.7) correspond to the locations of geological phenomena such as lithological abrupt changes, property jumps, and sedimentary rhythm transitions, and these locations are potential sublayer boundaries; areas with lower probability values ​​(such as less than the threshold of 0.7) correspond to relatively homogeneous reservoir sections within the sublayer. The probability map of small-layer boundary features can also contain local probability peak information and probability gradient change information. Local probability peaks correspond to potential small-layer boundary locations, and probability gradient change information helps to determine the clarity and reliability of the boundary. These multi-dimensional probability information provide a rich data foundation for the subsequent generation of small-layer partitioning schemes based on the small-layer constraint rule base. Step 5: Based on the sedimentary patterns of fluvial reservoirs, establish a small-layer constraint rule base, and obtain a small-layer partitioning scheme based on the small-layer constraint rule base and the small-layer boundary feature probability map; Based on the above embodiments, the small-layer constraint rule library specifically includes: sedimentary rhythm constraint rules, lithological combination constraint rules, physical property change constraint rules, thickness rationality constraint rules, boundary continuity constraint rules, well logging response characteristic constraint rules, etc. Specifically, the system collects and organizes knowledge of fluvial sedimentary geology. Through literature review, expert interviews, and statistical analysis of historical data, it summarizes typical sedimentary patterns of fluvial reservoirs, including the vertical evolution and planar distribution characteristics of different sedimentary microfacies such as channel sand bodies, natural dikes, breach fans, and floodplains. It also analyzes the sedimentary rhythm characteristics of different fluvial facies types (such as meandering rivers, braided rivers, and reticulated rivers), such as the combination patterns of positive rhythm (grain size becoming finer upwards), negative rhythm (grain size becoming coarser upwards), and composite rhythm, to establish a microlayer constraint rule library. For example, according to the thickness rationality constraint rules, the thickness of a single sublayer should not be less than the longitudinal resolution of the logging instrument (usually 0.2-0.5 meters). The specific process of obtaining the small-layer partitioning scheme based on the small-layer constraint rule base and the small-layer boundary feature probability map includes: Threshold segmentation is performed on the small-layer boundary feature probability map to extract all pixels or depth points with probability values ​​greater than the preset probability threshold, thus obtaining boundary candidate points. Connectivity analysis and clustering are performed on candidate boundary points to merge spatially adjacent candidate boundary points and obtain candidate boundaries; The candidate boundaries are constrained based on the small-level constraint rule base to obtain the small-level partitioning scheme; The sublayer division scheme specifically includes: the sublayer division number, the top depth boundary of the sublayer, the bottom depth boundary of the sublayer, and the sublayer thickness; Specifically, firstly, threshold segmentation is performed on the probability map of the sublayer boundary features to extract all pixels or depth points with probability values ​​greater than a preset probability threshold. This identifies high-confidence set points that may belong to the sublayer boundary, resulting in candidate boundary points. Subsequently, connected component analysis and clustering are performed on the extracted candidate boundary points. Adjacent candidate points are grouped and merged based on spatial adjacency and probability continuity to form continuous and structurally complete candidate boundaries. Next, the candidate boundaries are constrained and screened based on the sublayer constraint rule base. Candidate boundaries that do not meet the sublayer constraint conditions in the sublayer constraint rule base are eliminated to achieve optimized identification of real geological sublayer boundaries. The final output sublayer division scheme includes key information such as the sublayer division number, the top deep boundary of the sublayer, the bottom deep boundary of the sublayer, and the sublayer thickness, thereby achieving accurate stratification and quantitative division of sublayers within the sedimentary units of the target fluvial reservoir. Step 6: Input the sub-layer partitioning scheme into the preset visualization software to obtain the sub-layer partitioning result map; construct the result map library, store the sub-layer partitioning result map in the result map library, and preset the update cycle to update the result map library; Specifically, the sublayer division scheme is input into a preset visualization software, which may include petroleum geology software such as Petrel, Techlog, and GeoFrame. After automatically reading the sublayer division scheme data, the software generates sublayer division result maps, including: single-well sublayer columnar sections, well-connected sublayer comparison profiles, and planar sublayer distribution maps. The single-well sublayer columnar sections proportionally display the vertical distribution, thickness variation, and lithological characteristics of each sublayer, using different colors to distinguish different sublayers, and also labeling the sublayer number and depth boundary information. The well-connected sublayer comparison profiles horizontally connect sublayers from multiple wells for comparison, visually demonstrating the spatial continuity and thickness variation of sublayers between wells. The planar sublayer distribution maps display the thickness and boundary information of a specific sublayer using contour lines or color blocks. Specifically, a result image library is constructed, and the result images of the sub-layer division are stored in the result image library. An update cycle is preset to update the result image library. The preset update cycle can be one day, one week, or one month, etc., and the preset update cycle can be set according to the specific actual needs. By centrally managing and dynamically maintaining the reservoir sub-layer division results map data, the reservoir sub-layer division results can be systematically, traceably, and intelligently applied. The results map library, as a unified data carrier and knowledge accumulation platform, can store sub-layer division results from different regions, well sections, and times for a long time, and provide rich historical sample support for subsequent training and comparative analysis of sub-layer boundary feature recognition models for similar fluvial reservoirs. By setting an update cycle, the results map library can be automatically iterated and updated after new data, updated models, or optimized algorithms are generated, ensuring that the sub-layer subdivision results data in the results map library always reflect the latest geological understanding and model subdivision level. At the same time, the results map library can also serve as a knowledge base to provide prior constraints and reference templates for other automatic reservoir subdivision tasks, realizing intelligent iteration of "past results feeding back to new goals", thereby improving the accuracy, stability, and intelligence level of fluvial reservoir sub-layer subdivision and promoting the transformation from experience-driven to data-driven geological interpretation mode.

[0022] Example 2 Please see Figure 2 As shown, in another embodiment of the present invention, the present invention also discloses an automatic partitioning system for fluvial reservoir sublayers based on image intelligent analysis. The system includes the following modules: data acquisition and processing module, confidence scoring module, model training module, probability map generation module, partitioning scheme generation module, and visualization and updating module. The above modules are connected via wired and / or wireless means to enable data transmission between the modules; Data acquisition and processing module: acquires historical sub-layer image data of the target fluvial reservoir, enhances the boundaries of the sub-layers in the image data, and extracts the enhanced features of the sub-layer boundaries to obtain a set of enhanced features of the sub-layer boundaries; Confidence scoring module: Based on a pre-set confidence scoring model, the confidence of the divided sub-layer boundary enhancement features is evaluated to obtain a confidence score. A pre-set confidence score threshold is set, and the divided sub-layer boundary enhancement features that are less than the pre-set confidence score threshold are removed to obtain the set of divided sub-layer boundary enhancement features after confidence evaluation. The model training module divides the pre-divided small-layer boundary enhancement feature set after confidence evaluation into a dataset, which includes a training set, a test set, and a validation set; and trains a small-layer boundary feature recognition model based on the dataset to obtain the trained small-layer boundary feature recognition model. The probability map generation module inputs the target fluvial reservoir image data to be divided into the trained sub-layer boundary feature recognition model to obtain the sub-layer boundary feature probability map. A sub-layer partitioning scheme generation module is used. Based on the sedimentary patterns of fluvial reservoirs, a sub-layer constraint rule base is established. Based on the sub-layer constraint rule base and the sub-layer boundary feature probability map, a sub-layer partitioning scheme is obtained. The visualization and updating module inputs the sub-layer partitioning scheme into a preset visualization software to obtain the sub-layer partitioning result map; it constructs a result map library, stores the sub-layer partitioning result map in the result map library, and presets an update cycle to update the result map library.

[0023] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0024] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An automatic segmentation method for fluvial facies reservoir sublayers based on intelligent image analysis, characterized in that, The method includes the following steps: Step 1: Obtain historical sub-layer image data of the target fluvial reservoir, enhance the sub-layer boundaries of the image data, and extract the sub-layer boundary enhancement features to obtain a set of sub-layer boundary enhancement features; Step 2: Based on the pre-set reliability scoring model, evaluate the confidence of the divided sub-layer boundary enhancement features to obtain a confidence score. Set a pre-set reliability score threshold and remove the divided sub-layer boundary enhancement features that are lower than the pre-set reliability score threshold to obtain the set of divided sub-layer boundary enhancement features after confidence evaluation. Step 3: Divide the set of subdivided boundary enhancement features after confidence evaluation into a dataset, which includes a training set, a test set, and a validation set; train the subdivided boundary feature recognition model based on the dataset to obtain the trained subdivided boundary feature recognition model; Step 4: Input the target fluvial reservoir image data to be classified into the trained sublayer boundary feature recognition model to obtain the sublayer boundary feature probability map; Step 5: Based on the sedimentary patterns of fluvial reservoirs, establish a small-layer constraint rule base, and obtain a small-layer partitioning scheme based on the small-layer constraint rule base and the small-layer boundary feature probability map; Step 6: Input the sub-layer partitioning scheme into the preset visualization software to obtain the sub-layer partitioning result map; construct the result map library, store the sub-layer partitioning result map in the result map library, and preset the update cycle to update the result map library.

2. The automatic segmentation method for fluvial facies reservoir sublayers based on image intelligent analysis according to claim 1, characterized in that, The image data specifically includes: well logging image data, core image data, thin section image data, and seismic profile image data.

3. The automatic segmentation method for fluvial facies reservoir sublayers based on image intelligent analysis according to claim 1, characterized in that, The enhanced features of the boundary of the subdivided layers specifically include: sediment grain size characteristics, sedimentary facies characteristics, permeability characteristics, porosity characteristics, texture characteristics, lithological characteristics, fluid distribution characteristics, boundary thickness, and boundary curvature.

4. The automatic segmentation method for fluvial facies reservoir sublayers based on intelligent image analysis according to claim 1, characterized in that, The specific process of evaluating the confidence of the enhanced features at the defined sub-layer boundaries based on a pre-set confidence scoring model to obtain a confidence score includes: The pattern of transforming the divided small-layer boundary enhancement features into vectors is used to obtain the boundary enhancement feature vector; The boundary enhancement feature vector is input into a pre-set reliability scoring model, and the multilayer perceptron inside the model performs a nonlinear mapping of the boundary enhancement feature vector to a high-dimensional vector space. The confidence score is obtained by compressing the boundary enhancement feature vector of the high-dimensional vector space to the interval of 0 to 1 using the softmax function.

5. The automatic segmentation method for fluvial facies reservoir sublayers based on intelligent image analysis according to claim 1, characterized in that, The specific process of obtaining the trained small-layer boundary feature recognition model includes: The set of enhanced features with divided small-layer boundaries after confidence evaluation is divided into a dataset, which includes a training set, a test set, and a validation set. A classifier is constructed, and the training set is input into the classifier for training to obtain a trained classifier; the cross-entropy loss function is used as the training objective, and the learning rate is set to... The training period is set to T; The trained classifier is validated and evaluated using a validation set based on evaluation metrics, including: accuracy, precision, recall, and F1 score. The trained classifier is tested using a test set. When the test results meet the evaluation criteria, the trained small-layer boundary feature recognition model is obtained.

6. The automatic segmentation method for fluvial facies reservoir sublayers based on intelligent image analysis according to claim 1, characterized in that, The specific process of obtaining the probability map of small-layer boundary features includes: The target fluvial reservoir image data to be delineated is input into the trained sublayer boundary feature recognition model; The model identifies sub-layer boundary features from target fluvial reservoir image data and performs nonlinear transformations on these features to enhance the model's representation of complex sub-layer boundaries. After performing a nonlinear transformation on the sublayer boundary features, the model calculates and converts the probability of a pixel in the target fluvial reservoir image data belonging to the sublayer boundary into a probability value using an activation function. The boundary feature probability map of the sub-layer is obtained from the output of the model output layer.

7. The automatic segmentation method for fluvial facies reservoir sublayers based on image intelligent analysis according to claim 1, characterized in that, The small-layer boundary feature probability map specifically includes: A two-dimensional or three-dimensional probability numerical matrix corresponding to the spatial dimension of the target river facies reservoir image data to be delineated; Each element in the probability plot corresponds to a spatial location in the image data of the target fluvial reservoir to be delineated; the value of each element in the probability plot represents the probability that the location is a sublayer boundary. It also includes information on local probability peaks and probability gradient changes.

8. The automatic segmentation method for fluvial facies reservoir sublayers based on image intelligent analysis according to claim 1, characterized in that, The lower-level constraint rule base specifically includes: Sedimentary rhythm constraint rules, lithological combination constraint rules, physical property variation constraint rules, thickness rationality constraint rules, boundary continuity constraint rules, and well logging response characteristic constraint rules.

9. The automatic segmentation method for fluvial facies reservoir sublayers based on image intelligent analysis according to claim 1, characterized in that, The specific process of obtaining the small-layer partitioning scheme based on the small-layer constraint rule base and the small-layer boundary feature probability map includes: Threshold segmentation is performed on the small-layer boundary feature probability map to extract all pixels or depth points with probability values ​​greater than the preset probability threshold, thus obtaining boundary candidate points. Connectivity analysis and clustering are performed on candidate boundary points to merge spatially adjacent candidate boundary points and obtain candidate boundaries; The candidate boundaries are constrained based on the small-level constraint rule base to obtain the small-level partitioning scheme; The sublayer division scheme specifically includes: the sublayer division number, the top depth boundary of the sublayer, the bottom depth boundary of the sublayer, and the sublayer thickness.

10. An automatic segmentation system for fluvial facies reservoir layers based on intelligent image analysis, employing the automatic segmentation method for fluvial facies reservoir layers based on intelligent image analysis as described in any one of claims 1-9, characterized in that, The system includes the following modules: data acquisition and processing module, confidence scoring module, model training module, probability graph generation module, partitioning scheme generation module, and visualization and update module; Data acquisition and processing module: acquires historical sub-layer image data of the target fluvial reservoir, enhances the boundaries of the sub-layers in the image data, and extracts the enhanced features of the sub-layer boundaries to obtain a set of enhanced features of the sub-layer boundaries; Confidence scoring module: Based on a pre-set confidence scoring model, the confidence of the divided sub-layer boundary enhancement features is evaluated to obtain a confidence score. A pre-set confidence score threshold is set, and the divided sub-layer boundary enhancement features with a score lower than the pre-set confidence score threshold are removed to obtain the set of divided sub-layer boundary enhancement features after confidence evaluation. Model training module: The set of subdivided small-layer boundary enhancement features after confidence evaluation is divided into a dataset, which includes a training set, a test set, and a validation set; a small-layer boundary feature recognition model is trained based on the dataset to obtain the trained small-layer boundary feature recognition model; Probability map generation module: Input the target fluvial reservoir image data to be divided into the trained sublayer boundary feature recognition model to obtain the sublayer boundary feature probability map; The subdivision scheme generation module: Based on the sedimentary patterns of fluvial reservoirs, a sublayer constraint rule base is established. Based on the sublayer constraint rule base and the sublayer boundary feature probability map, a sublayer subdivision scheme is obtained. Visualization and Update Module: Input the sub-layer partitioning scheme into a preset visualization software to obtain the sub-layer partitioning result map; construct a result map library, store the sub-layer partitioning result map in the result map library, and preset the update cycle to update the result map library.

Images