Fracture intelligent identification system for buried hill oil and gas reservoir based on multi-attribute fusion

Abstract

The application discloses a buried hill oil and gas reservoir fracture intelligent identification system based on multi-attribute fusion, relates to the technical field of oil and gas reservoir fracture identification, and comprises the following modules: a data acquisition module, which is used for regularly collecting and identifying multi-source original data from a buried hill oil and gas reservoir; a data preprocessing module, which is used for cleaning and standardizing the multi-source original data; a multi-attribute feature extraction module, which is used for extracting fracture-related features; an identification model module, which is used for fusing the extracted fracture-related features and completing a fracture identification operation; and a result output module, which is used for displaying and storing the fracture identification result. The application integrates core images, imaging logging, thin section observation and other multi-source data, combines random forest, K nearest neighbor and other algorithms, realizes multi-scale and multi-attribute identification and classification operations on the buried hill oil and gas reservoir fractures, breaks through the limitation of traditional single data source, realizes the capture effect of multi-dimensional information, and improves the fracture identification accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of oil and gas reservoir fracture identification technology, specifically to an intelligent fracture identification system for buried hill oil and gas reservoirs based on multi-attribute fusion. Background Technology

[0002] Buried hill reservoirs refer to traps formed by paleotopic uplifts covered by overlying impermeable layers, where oil accumulates. Buried hill reservoirs are a special form of oil and gas accumulation, characterized by their formation within paleoburied hill structures below unconformities. As an important component of unconventional oil and gas resources, buried hill reservoirs have become a key area for oil and gas exploration and development due to their large reservoir size and high resource potential. However, buried hill reservoirs have undergone multiple tectonic movements and complex diagenesis, with lithology dominated by metamorphic and volcanic rocks, exhibiting complex composition and strong heterogeneity. Fractures serve as their primary reservoir spaces and seepage channels, making accurate fracture identification crucial for reservoir evaluation.

[0003] Existing fracture identification technologies have significant shortcomings: traditional human-computer interaction identification relies on human experience, and when dealing with the highly complex imaging logging data of buried hill oil and gas reservoirs, it is prone to errors due to subjective judgment, making it difficult to distinguish between natural fractures and drilling-induced fractures; single logging data identification methods cannot fully capture the multi-attribute coupling characteristics of buried hill reservoirs, and the identification accuracy is low when faced with interference such as lithological abrupt changes and well diameter enlargement; existing machine learning models are mostly designed for single data types and do not take into account the complementarity of multi-source data such as core, thin section, and imaging logging data of buried hill oil and gas reservoirs, making it difficult to adapt to the differences in buried hill reservoirs in different blocks.

[0004] Based on this, we now offer an intelligent identification system for fractures in buried hill oil and gas reservoirs based on multi-attribute fusion, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent fracture identification system for buried hill oil and gas reservoirs based on multi-attribute fusion, so as to solve the problems of the shortcomings of existing fracture identification technologies in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A multi-attribute fusion-based intelligent fracture identification system for buried hill oil and gas reservoirs includes:

[0008] The data acquisition module is used to periodically collect and identify multi-source raw data from buried hill oil and gas reservoirs, including core image data, imaging logging data, and thin section observation data from buried hill oil and gas reservoirs.

[0009] The data preprocessing module is used to clean and standardize raw data from multiple sources.

[0010] The multi-attribute feature extraction module is used to extract fracture-related features from the preprocessed data. The fracture-related features include lithological features, resistivity features, porosity features, and fracture morphology features.

[0011] The identification model module is used to fuse the extracted crack-related features and complete the crack identification operation;

[0012] The results output module is used to display and store the crack identification results in a visual form, including crack type, orientation, and degree of development.

[0013] The data acquisition module, data preprocessing module, multi-attribute feature extraction module, recognition model module, and result output module communicate with each other via a network.

[0014] Preferably, the core image data includes information on the type, development degree, filling condition, and occurrence of core fractures; the imaging logging data includes logging response characteristics, strike, dip, dip angle, and density information of buried hill fractures; and the thin section observation data includes information on the type, width, filling degree, and intersection relationship of natural fractures at the microscale.

[0015] Preferably, the data preprocessing module includes the following processing steps: denoising, image enhancement, and size normalization of the core image data; outlier removal, depth correction, and data format unification of the imaging logging data; image segmentation of the thin section observation data to separate the fracture area from the background area; and grayscale normalization processing.

[0016] Preferably, the multi-attribute feature extraction module includes:

[0017] The lithology feature extraction unit is used to calculate the correlation between each logging curve and lithology using the random forest algorithm, select logging curves with a correlation higher than a preset threshold as lithology sensitive curves, use the lithology sensitive curves as input features, and then complete the lithology division of buried hill oil and gas reservoirs through the K-nearest neighbor algorithm to obtain lithology feature vectors.

[0018] The resistivity feature extraction unit is used to calculate the difference in lateral resistivity between shallow and deep sides and the rate of resistivity change based on dual lateral resistivity logging data, and to construct a resistivity feature matrix.

[0019] The porosity feature extraction unit is used to combine density logging data, neutron logging data and electrical imaging porosity spectrum data to calculate the mean and variance of porosity at different depth points and obtain porosity feature vectors.

[0020] The fracture morphology feature extraction unit is used to perform image segmentation on the preprocessed imaging logging data. It uses Hough transform to initially identify sinusoidal fractures in the image, obtains the initial dip angle and length information of the fractures, and then extracts the edge contour of the fractures through an edge detection algorithm. It calculates the curvature, perimeter and area parameters of the contour to obtain the fracture morphology feature matrix.

[0021] Preferably, the recognition model module includes:

[0022] The feature fusion unit employs an attention mechanism to assign weights to the lithological feature vector, resistivity feature matrix, porosity feature vector, and fracture morphology feature matrix, and then splices the features to obtain a fused feature matrix.

[0023] The identification unit is used to complete the identification operation of fractures in buried hill oil and gas reservoirs using a deep convolutional neural network model. The input of the deep convolutional neural network model is a fusion feature matrix, the fracture labels obtained from core observation are used as supervision signals, and the output is the fracture identification result.

[0024] An optimization training unit is used to optimize the training of the deep convolutional neural network model. The model parameters are optimized by using the ReLU activation function and mini-batch gradient descent method to improve the accuracy and generalization ability of crack identification.

[0025] Preferably, the lithological feature extraction unit specifically operates as follows:

[0026] Multiple logging curves were collected from buried hill oil and gas reservoir areas, including natural gamma logging curves, deep resistivity logging curves, density logging curves, neutron logging curves, and sonic transit time logging curves.

[0027] Using the known lithology type calibrated from the core as the target variable and the values ​​of each logging curve as the input variable, the random forest algorithm is used to calculate the feature importance value of each logging curve and the lithology type. This feature importance value is used as the quantitative index of the correlation between the logging curve and the lithology. A preset threshold for correlation is set, and logging curves with a correlation higher than the preset threshold are selected and identified as lithology-sensitive curves.

[0028] The selected lithology-sensitive curves are standardized to obtain a standardized feature matrix.

[0029] Using the standardized feature matrix as input, the K-nearest neighbor algorithm is adopted, and the lithology type identified by the core is used as the training label. By calculating the Euclidean distance between the sample to be divided and the known lithology samples in the training set, the lithology types of the K nearest samples are selected for voting to determine the lithology type of the sample to be divided. The resulting lithology type is then converted into a lithology feature vector.

[0030] Preferably, the resistivity feature extraction unit specifically operates as follows:

[0031] Extract the deep lateral resistivity value and shallow lateral resistivity value at the same depth point from the dual lateral resistivity logging data, calculate the difference between the deep and shallow lateral resistivity values, and generate a difference sequence along the depth direction based on a preset depth interval.

[0032] A sliding window process is applied to the deep lateral resistivity curve, and the resistivity change rate at the center depth point within the window is calculated using the rate of change formula to generate a rate of change sequence along the depth direction.

[0033] Align the difference sequence and the rate of change sequence along the depth dimension to construct a resistivity feature matrix. The row vectors of the resistivity feature matrix correspond to depth points, and the column vectors of the resistivity feature matrix are the difference in lateral resistivity between shallow and deep sides and the rate of change of resistivity at that depth point, respectively. Each element in the resistivity feature matrix is ​​mapped to the interval [0,1] through normalization.

[0034] Preferably, the porosity feature extraction unit specifically operates as follows:

[0035] Density porosity is calculated based on density logging data, neutron porosity is calculated based on neutron logging data, imaging porosity is extracted from electrical imaging porosity spectrum data, and consistency verification is performed on density porosity, neutron porosity and imaging porosity at the same depth point, and outliers with deviations exceeding the preset porosity threshold are removed.

[0036] For the verified data, the depth segments are divided according to the preset depth interval. The arithmetic mean of density porosity, neutron porosity and imaging porosity in each depth segment is calculated to obtain the porosity mean vector of the corresponding depth segment. The sample variance of density porosity, neutron porosity and imaging porosity in the same depth segment is calculated to obtain the porosity variance vector.

[0037] The mean porosity vector and the variance porosity vector of each depth segment are concatenated by dimension to form the porosity feature vector of that depth segment. Then, the feature vectors of all depth segments are arranged in depth order to construct the porosity feature vector along the depth direction.

[0038] Preferably, the crack morphology feature extraction unit specifically operates as follows:

[0039] Image segmentation algorithms were used to segment the preprocessed imaging logging data into potential fracture regions and non-fracture background regions.

[0040] Hough transform is applied to the segmented potential crack regions. Pixels that conform to a sinusoidal distribution are detected by parameter accumulation in polar coordinate space. Curves with accumulated values ​​exceeding a preset length threshold are selected as initial cracks. The dip angle and pixel length of the initial cracks are calculated based on the polar coordinate parameters.

[0041] The Canny edge detection algorithm was used to extract the closed edge contour of the initial crack region;

[0042] Calculate the curvature, perimeter, and area of ​​the closed edge contour, and arrange the tilt angle, pixel length, average curvature, actual perimeter, and actual area of ​​the same crack region in a preset order to form a crack morphology feature matrix.

[0043] Preferably, the data preprocessing module, multi-attribute feature extraction module, recognition model module, and result output module are all integrated into the back-end terminal device. The back-end terminal device has a built-in data storage repository, which includes a historical database, a basic database, and a processing database. The historical database is used to store historical monitoring data, historical fracture identification results, and historical environmental background data related to buried hill oil and gas reservoirs. The basic database is used to store and receive data information about the buried hill oil and gas reservoir area and surrounding areas from big data. The processing database is used to store multi-source data, preprocessed data, and model optimization training data transmitted by each module.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] This invention establishes an intelligent fracture identification system for buried hill oil and gas reservoirs based on multi-attribute fusion. By integrating multi-source data such as core images, imaging logging, and thin section observations, and combining algorithms such as random forest and K-nearest neighbor, it achieves multi-scale and multi-attribute identification and classification of fractures in buried hill oil and gas reservoirs. It breaks through the limitations of traditional single data sources, achieves the effect of capturing multi-dimensional information, and improves the accuracy and reliability of fracture identification. By screening lithology-sensitive curves through random forest and combining multi-attribute features such as resistivity difference, rate of change, and fracture morphology parameters, it effectively distinguishes between natural and artificial fractures, reduces the false judgment rate, can adapt to complex and variable buried hill reservoir environments, significantly reduces the cost of manual interpretation, and improves exploration and development efficiency. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the intelligent identification system for fractures in buried hill oil and gas reservoirs according to the present invention.

[0047] Figure 2 This is a schematic diagram of the multi-attribute feature extraction module of the present invention.

[0048] Figure 3 This is a schematic diagram of the recognition model module of the present invention.

[0049] Figure 4 This is a schematic diagram of the fracture in the buried hill oil and gas reservoir of the present invention.

[0050] Figure label annotations: Data acquisition module 100, data preprocessing module 200, multi-attribute feature extraction module 300, lithological feature extraction unit 310, resistivity feature extraction unit 320, porosity feature extraction unit 330, fracture morphology feature extraction unit 340, identification model module 400, feature fusion unit 410, identification unit 420, optimization training unit 430, result output module 500. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0052] Example 1

[0053] In this embodiment, as Figure 1 - Figure 4 As shown, the intelligent identification system for fractures in buried hill oil and gas reservoirs based on multi-attribute fusion includes:

[0054] The data acquisition module 100 is used to periodically collect and identify multi-source raw data from buried hill oil and gas reservoirs. The multi-source raw data includes core image data, imaging logging data, and thin section observation data from buried hill oil and gas reservoirs.

[0055] Data preprocessing module 200 is used to clean and standardize multi-source raw data;

[0056] The multi-attribute feature extraction module 300 is used to extract fracture-related features from the preprocessed data. The fracture-related features include lithological features, resistivity features, porosity features, and fracture morphology features.

[0057] The identification model module 400 is used to fuse the extracted crack-related features and complete the crack identification operation;

[0058] The results output module 500 is used to display and store the crack identification results in a visual form. The crack identification results include crack type, orientation and development degree.

[0059] The data acquisition module 100, data preprocessing module 200, multi-attribute feature extraction module 300, recognition model module 400 and result output module 500 communicate with each other via a network;

[0060] In this embodiment, the data acquisition module 100 includes instruments and equipment such as core scanners, imaging logging tools, and thin-section microscopes deployed at the drilling site and logging laboratory of buried hill oil and gas reservoirs to acquire measurement data. It also includes retrieval equipment for searching big data related to buried hill oil and gas reservoirs. The data preprocessing module 200, multi-attribute feature extraction module 300, recognition model module 400, and result output module 500 are all integrated into the back-end terminal device. The back-end terminal device has a built-in data storage repository, which supports real-time display and long-term storage. The stored content includes historical data, real-time acquired data, and data involved in model training. Stable communication is established between the modules through network protocols.

[0061] In this embodiment, considering that a single data source cannot fully characterize the multi-stage, multi-scale, and highly anisotropic characteristics of the buried hill fracture system, this application uses core image data, imaging logging data, and thin section observation data for multi-scale data fusion. Since traditional single threshold methods or manual interpretation are difficult to handle multi-source, high-dimensional features, this application uses a deep learning model for intelligent identification, which facilitates the identification of fractures and the classification of their type, occurrence, and development degree.

[0062] Among them, such as Figure 4 As shown, the core image data includes information on the type, development degree, filling condition, and occurrence of core fractures; the imaging logging data includes logging response characteristics, strike, dip, dip angle, and density information of buried hill fractures; and the thin section observation data includes information on the type, width, filling degree, and intersection relationship of natural fractures at the microscale. The data acquisition module 100 scans the core of the core section of the buried hill oil and gas reservoir using a core scanner, and simultaneously records raw data such as core depth and core well number. The fracture type, development degree, filling condition, and occurrence are manually marked using an image annotation tool to form a core fracture label library. In practical applications, the logging data, core, and thin section data of the drilling exploration wells encountered in the target identification area can be obtained through the data storage library. The logging data is relevant information recorded during the drilling process, the core data can be obtained through drilling and core sampling, and the thin section data can be obtained by sampling during core observation and then obtaining relevant data through grinding and analysis.

[0063] Among them, such as Figure 1As shown, the processing flow of the data preprocessing module 200 includes: denoising, image enhancement, and size normalization of the core image data. Gaussian filtering can be used to remove random noise in the core image. Histogram equalization is used to enhance the gray-level contrast between the fracture and matrix regions, making the fracture region and matrix region significantly different. Outlier removal, depth correction, and data format unification are performed on the imaging logging data. For example, the Raida criterion can be used to remove outliers. Data offset is corrected based on the well depth correction algorithm. Image segmentation is performed on the thin section observation data to separate the fracture region from the background region. Then, gray-level normalization is performed. For example, the Otsu method can be used to perform threshold segmentation on the thin section image to separate the fracture region from the background region. The data preprocessing module 200 outputs images and curve data with a unified style.

[0064] Among them, such as Figure 1 and Figure 2 As shown, the multi-attribute feature extraction module 300 includes:

[0065] The lithological feature extraction unit 310 is used to calculate the correlation between each logging curve and lithology using the random forest algorithm, select logging curves with a correlation higher than a preset threshold as lithological sensitive curves, use the lithological sensitive curves as input features, and then complete the lithological division of buried hill oil and gas reservoirs through the K-nearest neighbor algorithm to obtain lithological feature vectors.

[0066] The resistivity feature extraction unit 320 is used to calculate the difference in deep and shallow lateral resistivity and the rate of resistivity change based on dual lateral resistivity logging data, and to construct a resistivity feature matrix.

[0067] Porosity feature extraction unit 330 is used to combine density logging data, neutron logging data and electrical imaging porosity spectrum data to calculate the mean and variance of porosity at different depth points and obtain porosity feature vector. This operation can reflect the reservoir space characteristics of the reservoir from both macroscopic and microscopic scales and help distinguish between fracture porosity and matrix porosity.

[0068] The fracture morphology feature extraction unit 340 is used to perform image segmentation on the preprocessed imaging logging data. It uses Hough transform to initially identify sinusoidal fractures in the image, obtains the initial dip angle and length information of the fracture, and then extracts the edge contour of the fracture through the edge detection algorithm. It calculates the curvature, perimeter and area parameters of the contour to obtain the fracture morphology feature matrix. This operation can directly quantify the geometric shape of the fracture from the image, providing a basis for identifying the fracture type.

[0069] In this embodiment, the buried hill reservoir matrix is ​​dense, and fractures are the main storage space and seepage channels. The lithology of buried hill oil and gas reservoirs is complex and varied, and lithology directly controls the brittleness of rocks, thereby controlling the degree of fracture development. Different lithologies will produce completely different fracture systems under the same tectonic stress. The bedrock of buried hills is usually a high resistivity background. Resistivity characteristics include abnormally low resistivity caused by fracture capture and changes in resistivity differences between shallow and deep layers. The porosity of buried hill reservoirs is mainly provided by fractures and dissolution cavities. Porosity characteristics include abnormal rock physical parameters caused by fracture development, such as reduced density and abnormal sonic time difference. Fracture morphology characteristics can be directly obtained from imaging logging data and core image data to distinguish high-angle fractures, low-angle fractures, and network fractures.

[0070] Among them, such as Figure 1 and Figure 3 As shown, the recognition model module 400 includes:

[0071] The feature fusion unit 410 employs an attention mechanism to assign weights to the lithological feature vector, resistivity feature matrix, porosity feature vector, and fracture morphology feature matrix, and then concatenates the features to obtain a fused feature matrix. For the four feature branches, including the lithological feature vector and resistivity feature matrix, a fully connected layer is used to generate their respective attention weight scalars. These weight scalars represent the importance of the attribute for fracture identification. Subsequently, the features of each branch are multiplied by their corresponding weights and then concatenated to form the fused feature matrix. Before concatenation, an alignment operation is performed, mapping all features to a unified feature dimension through a fully connected layer or convolutional layer. For example, global average pooling can be introduced for matrix features to compress them into vectors of fixed length. Then, the feature vectors with unified dimensions are concatenated. This operation overcomes the drawbacks of traditional methods that simply concatenate features, and can dynamically focus on the information most relevant to the fracture, thereby improving the discriminative ability of feature representation.

[0072] The identification unit 420 is used to complete the identification operation of fractures in buried hill oil and gas reservoirs using a deep convolutional neural network model. The input of the deep convolutional neural network model is the fused feature matrix, and the fracture labels obtained from core observation are used as supervision signals. The output is the fracture identification result. The deep convolutional neural network has powerful local feature extraction and nonlinear fitting capabilities, which can further refine the fracture boundary, correct misjudgments, and improve the calculation accuracy. The model abstracts features layer by layer through multi-layer convolution and pooling operations, and finally outputs the classification or segmentation results corresponding to the fracture type, occurrence and development degree through fully connected layers or 1x1 convolutional layers.

[0073] The optimized training unit 430 is used to optimize the training of the deep convolutional neural network model. The model parameters are optimized by using the ReLU activation function and mini-batch gradient descent. During model training, label data is retrieved from the historical database. The ReLU activation function is used to introduce nonlinearity during training, and the mini-batch gradient descent method is used to optimize the model parameters to minimize the loss function between the prediction result and the true label, thereby improving the accuracy and generalization ability of crack identification.

[0074] Among them, such as Figure 1 and Figure 2 As shown, the specific operations of the lithological feature extraction unit 310 include:

[0075] Multiple logging curves were collected from buried hill oil and gas reservoirs, including natural gamma logging curves, deep resistivity logging curves, density logging curves, neutron logging curves, and sonic transit time logging curves.

[0076] Using the known lithology type calibrated from the core as the target variable and the values ​​of each logging curve as the input variable, the random forest algorithm is used to calculate the feature importance value of each logging curve and the lithology type. This feature importance value is used as the quantitative index of the correlation between the logging curve and the lithology. A preset threshold for correlation is set, and logging curves with a correlation higher than the preset threshold are selected and identified as lithology-sensitive curves.

[0077] The selected lithology-sensitive curves are standardized to obtain a standardized feature matrix.

[0078] Using the standardized feature matrix as input, the K-nearest neighbor algorithm is adopted, and the lithology type of the core is used as the training label. By calculating the Euclidean distance between the sample to be divided and the known lithology samples in the training set, the lithology types of the K nearest samples are selected for voting to determine the lithology type of the sample to be divided. The lithology types obtained by the division are then converted into lithology feature vectors.

[0079] In this embodiment, considering that a high dimensionality in the original data would increase the complexity of data analysis, and that there is a certain correlation between the information in the data, this application uses correlation analysis to remove redundant and repetitive variables and variables with low correlation, making the new data uncorrelated. These uncorrelated data retain as much information as possible from the original data. The random forest algorithm, a classifier composed of a set of decision trees, is a type of ensemble learning that can effectively analyze data and assess the importance of each logging curve to lithology. Based on conventional logging curves, the most sensitive curve is selected. Conventional logging curves include, but are not limited to, natural gamma ray logging (GR), deep resistivity logging (RLLD), density logging (DEN), neutron logging (CNL), and sonic transit time logging (AC). The preset threshold is formulated based on existing historical data and expert experience, and can be adjusted according to actual needs. In this embodiment, the preset threshold is set to 0.2. Then, the curves are selected based on the calculation results with a correlation ≥ 0.2. Assuming the calculation results are: Natural Gamma Logging (GR) curve has a correlation of 0.35, Deep Resistivity Logging (RLLD) curve has a correlation of 0.28, Density Logging (DEN) curve has a correlation of 0.22, Neutron Logging (CNL) curve has a correlation of 0.18, Acoustic Transit Logging (AC) curve has a correlation of 0.15, Spontaneous Potential Logging (SP) curve has a correlation of 0.15, etc., then the selected curves are Natural Gamma Logging (GR), Deep Resistivity Logging (RLLD), and Density Logging (DEN). The K-nearest neighbor algorithm can avoid overfitting.

[0080] Among them, such as Figure 1 and Figure 2 As shown, the specific operations of the resistivity feature extraction unit 320 include:

[0081] Extract the deep lateral resistivity value and shallow lateral resistivity value at the same depth point from the dual lateral resistivity logging data, calculate the difference between the deep and shallow lateral resistivity values, and generate a difference sequence along the depth direction based on a preset depth interval.

[0082] A sliding window process is applied to the deep lateral resistivity curve, and the resistivity change rate at the center depth point within the window is calculated using the rate of change formula to generate a rate of change sequence along the depth direction.

[0083] Align the difference sequence and the rate of change sequence along the depth dimension to construct a resistivity feature matrix. The row vectors of the resistivity feature matrix correspond to depth points, and the column vectors of the resistivity feature matrix are the difference between shallow and deep lateral resistivity and the rate of change of resistivity at that depth point, respectively. Each element in the resistivity feature matrix is ​​mapped to the [0,1] interval through normalization.

[0084] In this embodiment, resistivity characteristics are sensitive to high-angle, open-cut cracks and are a key indicator for identifying effective cracks. The deep lateral resistivity value is... The shallow lateral resistivity value is The difference in lateral resistivity between the two is... The preset depth interval is [value], and the difference sequence is [value]. A 3-point sliding window is used to avoid the window being too large, causing a smooth transition of the rate of change, or the window being too small, introducing noise interference. The formula for calculating the rate of change of resistivity is: ,in, Let be the rate of change of resistivity at a certain depth. The deep lateral resistivity value is the value of the point adjacent to the center depth point i within the sliding window. Let be the deep lateral resistivity value of the adjacent depth point below the center depth point i within the sliding window, h be the depth interval between adjacent depth points, matching the well logging data sampling rate, k be the half-length of the window, the rate of change sequence be , and the resistivity characteristic matrix be represented as . R is the resistivity characteristic matrix;

[0085] Among them, such as Figure 1 and Figure 2 As shown, the specific operations of the porosity feature extraction unit 330 include:

[0086] Density porosity is calculated based on density logging data, neutron porosity is calculated based on neutron logging data, imaging porosity is extracted from electrical imaging porosity spectrum data, and consistency verification is performed on density porosity, neutron porosity and imaging porosity at the same depth point, and outliers with deviations exceeding the preset porosity threshold are removed.

[0087] For the verified data, the depth segments are divided according to the preset depth interval. The arithmetic mean of density porosity, neutron porosity and imaging porosity in each depth segment is calculated to obtain the porosity mean vector of the corresponding depth segment. The sample variance of density porosity, neutron porosity and imaging porosity in the same depth segment is calculated to obtain the porosity variance vector.

[0088] The mean porosity vector and the variance porosity vector of each depth segment are concatenated by dimension to form the porosity feature vector of that depth segment. Then, the feature vectors of all depth segments are arranged in depth order to construct the porosity feature vector along the depth direction.

[0089] In this embodiment, the preset porosity threshold is formulated based on expert experience and historical data, and can be adjusted according to actual needs. The density porosity calculation formula is as follows: ,in, ρ represents density and porosity, and m represents the cementation index. For example, the value ranges from 2.0 to 2.2 for buried hill metamorphic rocks and from 1.8 to 2.0 for volcanic rocks. These values ​​are determined based on historical data and expert experience. For formation water resistivity, The resistivity of the formation when it contains water is given. Neutron porosity is measured directly using neutron logging equipment. Imaging porosity is the value corresponding to the peak porosity in the electrical imaging porosity spectrum data at that depth point. The mean and standard deviation of the three porosities at the same depth point are calculated. The sample variance is calculated using the formula... The calculation shows that, where n is the number of valid data points within the depth segment, This represents the average porosity at this depth.

[0090] Among them, such as Figure 1 and Figure 2 As shown, the specific operations of the crack morphology feature extraction unit 340 include:

[0091] Image segmentation algorithm is used to segment the preprocessed imaging logging data into potential fracture areas and non-fracture background areas. The image segmentation algorithm can smooth the imaging logging image through Gaussian filtering, and then use an adaptive threshold method to determine the initial segmentation mark. The criteria for determining potential fracture areas are set, such as areas with gray values ​​more than 30% lower than the surrounding matrix and a continuous pixel length of more than 5.

[0092] Hough transform is applied to the segmented potential crack regions. Pixels conforming to a sinusoidal distribution are detected by accumulating parameters in polar coordinate space. Curves with accumulated values ​​exceeding a preset length threshold are selected as initial cracks. The dip angle and pixel length of the initial cracks are calculated based on the polar coordinate parameters. The preset length threshold for accumulated values ​​is determined according to the image resolution. When the accumulated value corresponding to a certain set of parameters in polar coordinate space exceeds the preset length threshold, it is determined to be an initial crack. The dip angle is calculated through polar coordinate angle conversion, and the pixel length is obtained by multiplying the number of pixels corresponding to the accumulated value by the actual pixel size.

[0093] The Canny edge detection algorithm is used to extract the closed edge contour of the initial crack region. The Canny edge detection algorithm has high and low thresholds, which can suppress and eliminate false edges.

[0094] The curvature, perimeter, and area of ​​the closed edge contour are calculated. The tilt angle, pixel length, average curvature, actual perimeter, and actual area of ​​the same crack region are arranged in a preset order and combined to form a crack morphology feature matrix. The curvature of the closed edge contour is calculated by fitting the contour curve using the least squares method. Several sampling points are evenly distributed on the contour. The curvature of each point is calculated and the average value is taken as the average curvature. The actual perimeter is obtained by accumulating the Euclidean distances between the contour pixels. The actual area is obtained by counting the number of pixels in the contour using the scan line method and then multiplying it by the actual area of ​​the pixels.

[0095] Example 2

[0096] Unlike Example 1, the data preprocessing module 200, multi-attribute feature extraction module 300, identification model module 400, and result output module 500 are all integrated into the back-end terminal device. The back-end terminal device has a built-in data storage repository, which includes a historical database, a basic database, and a processing database. The historical database is used to store historical monitoring data, historical fracture identification results, and historical environmental background data related to buried hill oil and gas reservoirs. The data stored in the historical database includes logging data, core analysis reports, and thin section identification results of exploration wells in buried hill oil and gas reservoirs in recent years. It supports retrieval by key fields such as well number, depth, and lithology. The basic database is used to store and receive data information about the buried hill oil and gas reservoir area and surrounding areas from big data. The basic database connects to the industry big data platform through an API interface to obtain geological structure maps, seismic interpretation results, and production dynamic data of adjacent wells in the buried hill area. The processing database is used to store multi-source data, preprocessed data, and model optimization training data transmitted by various modules. The processing database adopts a distributed storage architecture and is divided into a real-time processing area and a historical processing area. The real-time processing area stores data generated by each module in the past 24 hours, and the historical processing area stores processed data for more than 24 hours.

[0097] In summary, this invention effectively solves the problems of traditional methods relying on human experience and having limited identification capabilities based on a single data source. It has good engineering application value and promotion prospects. The model parameters can be adjusted according to the geological characteristics of different blocks, making it suitable for fracture identification tasks under various complex reservoir conditions.

[0098] 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 technical scope 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.

Claims

1. A multi-attribute fusion-based intelligent fracture identification system for buried hill oil and gas reservoirs, characterized in that, include: The data acquisition module (100) is used to periodically collect and identify multi-source raw data from buried hill oil and gas reservoirs, including core image data, imaging logging data, and thin section observation data from buried hill oil and gas reservoirs. The data preprocessing module (200) is used to clean and standardize multi-source raw data; A multi-attribute feature extraction module (300) is used to extract fracture-related features from preprocessed data, including lithological features, resistivity features, porosity features and fracture morphology features; The identification model module (400) is used to fuse the extracted crack-related features and complete the crack identification operation; The result output module (500) is used to display and store the crack identification results in a visual form, including crack type, orientation and development degree. The data acquisition module (100), data preprocessing module (200), multi-attribute feature extraction module (300), recognition model module (400), and result output module (500) communicate with each other via a network; The multi-attribute feature extraction module (300) include: The lithological feature extraction unit (310) is used to calculate the correlation between each logging curve and lithology using the random forest algorithm, select logging curves with a correlation higher than a preset threshold as lithological sensitive curves, use the lithological sensitive curves as input features, and then complete the lithological division of buried hill oil and gas reservoirs through the K-nearest neighbor algorithm to obtain lithological feature vectors. The resistivity feature extraction unit (320) is used to calculate the difference in deep and shallow lateral resistivity and the rate of resistivity change based on the dual lateral resistivity logging data, and to construct a resistivity feature matrix. Porosity feature extraction unit (330) is used to combine density logging data, neutron logging data and electrical imaging porosity spectrum data to calculate the mean and variance of porosity at different depth points and obtain porosity feature vectors. The fracture morphology feature extraction unit (340) is used to perform image segmentation on the preprocessed imaging logging data, use Hough transform to initially identify sinusoidal fractures in the image, obtain the initial dip angle and length information of the fracture, and then extract the closed edge contour of the fracture through the edge detection algorithm, calculate the curvature, perimeter and area parameters of the contour, and obtain the fracture morphology feature matrix. The specific operations of the lithological feature extraction unit (310) include: Multiple logging curves were collected from buried hill oil and gas reservoir areas, including natural gamma logging curves, deep resistivity logging curves, density logging curves, neutron logging curves, and sonic transit time logging curves. Using the known lithology type calibrated from the core as the target variable and the values ​​of each logging curve as the input variable, the random forest algorithm is used to calculate the feature importance value of each logging curve and the lithology type. This feature importance value is used as the quantitative index of the correlation between the logging curve and the lithology. A preset threshold for correlation is set, and logging curves with a correlation higher than the preset threshold are selected and identified as lithology-sensitive curves. The selected lithology-sensitive curves are standardized to obtain a standardized feature matrix. Using the standardized feature matrix as input, the K-nearest neighbor algorithm is adopted, and the lithology type of the core is used as the training label. By calculating the Euclidean distance between the sample to be divided and the known lithology samples in the training set, the lithology types of the K nearest samples are selected for voting to determine the lithology type of the sample to be divided. The lithology types obtained by the division are then converted into lithology feature vectors. The specific operations of the resistivity feature extraction unit (320) include: Extract the deep lateral resistivity value and shallow lateral resistivity value at the same depth point from the dual lateral resistivity logging data, calculate the difference between the deep and shallow lateral resistivity values, and generate a difference sequence along the depth direction based on a preset depth interval. A sliding window process is applied to the deep lateral resistivity curve, and the resistivity change rate at the center depth point within the window is calculated using the rate of change formula to generate a rate of change sequence along the depth direction. Align the difference sequence and the rate of change sequence along the depth dimension to construct a resistivity feature matrix. The row vectors of the resistivity feature matrix correspond to depth points, and the column vectors of the resistivity feature matrix are the difference in lateral resistivity between shallow and deep sides and the rate of change of resistivity at that depth point, respectively. Each element in the resistivity feature matrix is ​​mapped to the [0,1] interval through normalization. The specific operations of the porosity feature extraction unit (330) include: Density porosity is calculated based on density logging data, neutron porosity is calculated based on neutron logging data, imaging porosity is extracted from electrical imaging porosity spectrum data, and consistency verification is performed on density porosity, neutron porosity and imaging porosity at the same depth point, and outliers with deviations exceeding the preset porosity threshold are removed. For the verified data, the depth segments are divided according to the preset depth interval. The arithmetic mean of density porosity, neutron porosity and imaging porosity in each depth segment is calculated to obtain the porosity mean vector of the corresponding depth segment. The sample variance of density porosity, neutron porosity and imaging porosity in the same depth segment is calculated to obtain the porosity variance vector. The mean porosity vector and the variance porosity vector of each depth segment are concatenated by dimension to form the porosity feature vector of that depth segment. Then, the feature vectors of all depth segments are arranged in depth order to construct the porosity feature vector along the depth direction. The specific operations of the crack morphology feature extraction unit (340) include: Image segmentation algorithms were used to segment the preprocessed imaging logging data into potential fracture regions and non-fracture background regions. Hough transform is applied to the segmented potential crack regions. Pixels that conform to a sinusoidal distribution are detected by parameter accumulation in polar coordinate space. Curves with accumulated values ​​exceeding a preset length threshold are selected as initial cracks. The dip angle and pixel length of the initial cracks are calculated based on the polar coordinate parameters. The Canny edge detection algorithm was used to extract the closed edge contour of the initial crack region; Calculate the curvature, perimeter, and area of ​​the closed edge contour, and arrange the tilt angle, pixel length, average curvature, actual perimeter, and actual area of ​​the same crack region in a preset order to form a crack morphology feature matrix.

2. The intelligent identification system for fractures in buried hill oil and gas reservoirs based on multi-attribute fusion according to claim 1, characterized in that, The core image data includes information on the type, development degree, filling condition, and occurrence of core fractures; the imaging logging data includes logging response characteristics, strike, dip, dip angle, and density information of buried hill fractures; and the thin section observation data includes information on the type, width, filling degree, and intersection relationship of natural fractures at the microscale.

3. The intelligent fracture identification system for buried hill oil and gas reservoirs based on multi-attribute fusion according to claim 1, characterized in that, The processing flow of the data preprocessing module (200) includes: denoising, image enhancement and size normalization of the core image data; outlier removal, depth correction and data format unification of the imaging logging data; image segmentation of the thin section observation data to separate the fracture area and the background area; and grayscale normalization.

4. The intelligent fracture identification system for buried hill oil and gas reservoirs based on multi-attribute fusion according to claim 1, characterized in that, The recognition model module (400) includes: The feature fusion unit (410) uses an attention mechanism to assign weights to the lithological feature vector, resistivity feature matrix, porosity feature vector and fracture morphology feature matrix, and then splices the features to obtain the fused feature matrix. The identification unit (420) is used to complete the identification operation of fractures in buried hill oil and gas reservoirs using a deep convolutional neural network model. The input of the deep convolutional neural network model is a fusion feature matrix, the fracture labels obtained from core observation are used as supervision signals, and the fracture identification results are output. The optimization training unit (430) is used to optimize the training of the deep convolutional neural network model. The model parameters are optimized by using the ReLU activation function and mini-batch gradient descent method to improve the accuracy and generalization ability of crack recognition.

5. The intelligent identification system for fractures in buried hill oil and gas reservoirs based on multi-attribute fusion according to claim 1, characterized in that, The data preprocessing module (200), multi-attribute feature extraction module (300), recognition model module (400), and result output module (500) are all integrated in the back-end terminal device. The back-end terminal device has a built-in data storage repository, which includes a historical database, a basic database, and a processing database. The historical database is used to store historical monitoring data, historical fracture identification results, and historical environmental background data related to buried hill oil and gas reservoirs. The basic database is used to store and receive data information about buried hill oil and gas reservoir areas and surrounding areas from big data. The processing database is used to store multi-source data, preprocessed data, and model optimization training data transmitted by each module.

Images