Method and system for quantitatively extracting gravel parameters of conglomerate reservoir

Abstract

The invention belongs to the technical field of reservoir logging identification, and discloses a method and system for quantitatively extracting gravel parameters of a conglomerate reservoir, and the method comprises the steps: carrying out the image preprocessing, image enhancement and full-hole image processing of imaging logging information, and obtaining the full-hole logging image information; identifying full-borehole logging image information, and determining a conglomerate development depth section; and performing threshold segmentation on the conglomerate development depth section by using two-dimensional wavelet transform to obtain a binary image of conglomerate features. The method provided by the invention has the advantages that 1) the conglomerate development section is automatically determined through intelligent identification of an imaging logging image mode; 2) by adopting a two-dimensional wavelet transform method, imaging logging image segmentation thresholds in different windows can be determined in a self-adaptive manner, and automatic separation of gravels and a matrix background is realized; and 3) further processing the segmented gravel features through an image analysis method to realize effective separation of the bonded gravels.

Description

Technical Field

[0001] The invention belongs to the technical field of reservoir logging identification, and in particular relates to a method and system for quantitatively extracting gravel parameters of a conglomerate reservoir. Background Art

[0002] Various types of sandy conglomerate bodies are widely developed in the eastern and western basins of my country. In recent years, major breakthroughs have been made in the sandy conglomerate section, which is an important type of oil and gas reservoir. Sandy conglomerate reservoirs generally have the characteristics of diverse rock composition, large particle size variation, and poor sorting. The sedimentary structure and structural characteristics of the reservoir show strong heterogeneity, and its logging interpretation and evaluation has always been a world-class problem.

[0003] Exploration practice shows that parameters such as gravel content and particle size control the physical properties, brittleness and compressibility of sandstone reservoirs, and are key factors affecting the high and stable production of sandstone reservoirs. How to quantitatively calculate gravel characteristic parameters is an important part of logging evaluation of sandstone reservoirs. At present, logging evaluation of gravel characteristic parameters mainly relies on electrical imaging logging, but the use of electrical imaging is only at the qualitative stage, mainly based on the image-lithology recognition mode of electrical imaging, lacking quantitative characterization, and it is difficult to establish regional evaluation standards. There are also studies that use electrical imaging logging for image segmentation. By graying and segmenting the electrical imaging logging images, the image segmentation results are used to extract gravel information. However, due to the resolution of imaging logging images, different gravel particles on the image are often mixed together. The accuracy of directly extracting gravel through image segmentation is low, which affects the comprehensive evaluation results of the reservoir. Summary of the invention

[0004] In order to solve at least one problem in the background technology, the present invention provides a method and system for quantitatively extracting gravel parameters of a conglomerate reservoir.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for quantitatively extracting gravel parameters of a conglomerate reservoir comprises the following steps:

[0007] Perform image preprocessing, image enhancement and full-bore image processing on imaging logging data to obtain full-bore logging image information;

[0008] Identify the full borehole logging image information to determine the depth section of conglomerate development;

[0009] For the conglomerate development depth section, two-dimensional wavelet transform is used to perform threshold segmentation to obtain a binary image of conglomerate characteristics;

[0010] Processing the segmented conglomerate features, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the conglomerate features;

[0011] Based on the results of fine separation of conglomerate characteristics, quantitative statistics of gravel parameters and reservoir evaluation are carried out.

[0012] Preferably, the imaging logging data is subjected to image preprocessing, image enhancement and full borehole image processing to obtain full borehole logging image information, which mainly includes the following steps:

[0013] Perform acceleration correction and equalization processing on imaging logging data to obtain original electrical imaging logging image information;

[0014] Perform static enhancement processing on the electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section;

[0015] Dynamically enhance the electrical imaging logging image information to obtain image information reflecting the changes in rock structure;

[0016] Based on the image information of rock relative resistivity changes and rock structure changes in the entire well section, probability interpolation processing is performed on the uncovered part of the image to obtain electrical imaging logging image information covering the entire wellbore.

[0017] Preferably, identifying the full borehole logging image information and determining the conglomerate development depth section includes the following steps:

[0018] Build a deep learning model for image classification based on convolutional neural networks;

[0019] Convolution calculation is performed based on the image classification deep learning model to obtain the probability of various image types in the full-bore logging image information;

[0020] Automatically classify various types of electrical imaging image features in full-bore logging image information based on the probabilities of various image types;

[0021] Based on the classification results, the conglomerate development depth section is automatically identified and the depth interval is recorded.

[0022] Preferably, the image classification deep learning model includes an input layer, a convolution layer, an activation layer and an output layer;

[0023] The full borehole logging image information is received by the input layer and then sent to the convolutional layer;

[0024] The convolution layer is used to perform convolution calculation of the full borehole logging image information;

[0025] The activation layer uses the activation function to perform linear or nonlinear mapping of image information to enhance the network's expressive power.

[0026] The output layer connects the convolution information of each channel and finally converts the input full-bore logging image information into output data.

[0027] Preferably, the image classification deep learning model includes:

[0028] Matrix transformation calculation formula:

[0029]

[0030] In the formula, size is the size of the input feature matrix; k is the size of the convolution kernel; step is the convolution stride; size′ is the size of the transformed matrix; p is the number of zero-padding layers;

[0031] Convolution calculation formula:

[0032]

[0033] In the formula, is the activation function; conv2D is a two-dimensional convolution; W is a different weight matrix for each convolution layer; X is the input matrix for each convolution layer; Y is a different output matrix for each convolution layer; valid is the valid percentage of the current layer, which is used to prevent the model from overfitting; bias is the bias parameter;

[0034] The loss function E of the lth layer l :

[0035]

[0036] Where d is the expected output, y l is the output vector of the lth layer;

[0037] By taking the partial derivative of the loss function E with respect to each weight, it can be expressed as the chain rule formula:

[0038]

[0039] In the formula, E total is the total error size, indicating the sum of all errors, w i is the weight parameter of the i-th layer, and L is the total number of layers of the neural network.

[0040] Preferably, for the conglomerate development depth section, two-dimensional wavelet transform is used to perform threshold segmentation to obtain a binary image of conglomerate characteristics, including the following steps:

[0041] Obtain the dimensions of each conglomerate development depth section;

[0042] Threshold segmentation is performed based on the two-dimensional wavelet transform formula and the size of the conglomerate development depth segment;

[0043] A binary image of conglomerate features is obtained based on the threshold segmentation result.

[0044] Preferably, the two-dimensional wavelet transform formula is:

[0045]

[0046] Where W 1 f is the horizontal two-dimensional wavelet transform formula; W 2 f is the two-dimensional wavelet transform formula in the vertical direction; x and y are the horizontal and vertical coordinates of the corresponding two-dimensional data; θ s is the smooth function corresponding to ; is the gradient vector identifier; f is the mapping relationship of the function; S is the size of the conglomerate development depth section, and the formula is:

[0047] s={2 j} j∈Z

[0048] Where j is the maximum point of the modulus along the angular direction; Z is the range of the angular direction.

[0049] Preferably, the segmented conglomerate features are processed, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the conglomerate features, including the following steps:

[0050] Extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features;

[0051] Assign a unique identification value to each connected domain;

[0052] Extract the relative center points of each connected domain and set them as the local extreme points of the watershed algorithm;

[0053] Generate a watershed basin based on each local extreme point, and form a watershed at the boundary of the watershed basin;

[0054] The watershed influence area is obtained based on the threshold segmentation result, and the segmentation boundary of the watershed is obtained, so as to finely separate the cohesive gravel features.

[0055] Preferably, based on the result of fine separation of conglomerate characteristics, quantitative statistics of gravel parameters and reservoir evaluation are performed, including:

[0056] Quantitative statistics were performed on the percentage of characteristic pixels, average particle size, average roundness and average number of gravel.

[0057] The sedimentary type, morphology, lithology, physical properties and distribution scale of the gravel body are evaluated based on the proportion of gravel characteristic pixels, evaluated particle size, average roundness and average number.

[0058] Preferably, the gravel feature pixel ratio is:

[0059] Gravel_por=P i / P all ;

[0060] Where Gravel_por represents the percentage of gravel feature pixels, P i Indicates the number of gravel feature pixels per unit length, P all Indicates the total number of pixels;

[0061] The average particle size Gs is:

[0062]

[0063] In the formula, Gs i is the characteristic particle size of the i-th gravel; N is the total number of gravels.

[0064] The average roundness is:

[0065] R=C 2 / A;

[0066] Where R is the roundness of the gravel feature; C is the circumference of the gravel feature; A is the area of ​​the gravel feature;

[0067] The average number is used to describe the density of feature objects in a unit depth segment, which is the number of independent connected domains in a unit depth segment.

[0068] A conglomerate reservoir gravel parameter quantitative extraction system, comprising:

[0069] A processing unit, used for performing image preprocessing, image enhancement and full-bore image processing on the imaging logging data to obtain full-bore logging image information;

[0070] An identification unit is used to identify the full borehole logging image information and determine the conglomerate development depth section;

[0071] The segmentation unit is used to perform threshold segmentation on the conglomerate development depth section using two-dimensional wavelet transform to obtain a binary image of conglomerate characteristics;

[0072] A separation unit is used to process the segmented gravel features, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the gravel features;

[0073] The statistical unit is used to conduct quantitative statistics of gravel parameters and reservoir evaluation based on the fine separation results of gravel characteristics.

[0074] Preferably, the processing unit comprises:

[0075] The initial processing module is used to perform acceleration correction and equalization processing on the imaging logging data to obtain the original electrical imaging logging image information;

[0076] Static processing module, used to perform static enhancement processing on electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section;

[0077] Dynamic processing module, used to dynamically enhance the electrical imaging logging image information to obtain image information reflecting the changes in rock structure;

[0078] The interpolation processing module is used to perform probability interpolation processing on the uncovered part of the image based on the image information of the relative resistivity change of the rock in the whole well section and the image information of the rock structure change, so as to obtain the electrical imaging logging image information covering the whole wellbore.

[0079] Preferably, the identification unit comprises:

[0080] Building blocks for building deep learning models for image classification based on convolutional neural networks;

[0081] The convolution module is used to perform convolution calculations based on the image classification deep learning model to obtain the probabilities of various image types in the full-bore logging image information;

[0082] A classification module for automatically classifying various types of electrical imaging image features in the full-bore logging image information based on the probabilities of various image types;

[0083] The identification module is used to automatically identify the conglomerate development depth section based on the classification results and record the depth interval.

[0084] Preferably, the segmentation unit comprises:

[0085] A first acquisition module is used to acquire a two-dimensional wavelet transform formula;

[0086] The second acquisition module is used to obtain the size of each conglomerate development depth section;

[0087] A segmentation module, used for performing threshold segmentation based on a two-dimensional wavelet transform formula and the size of the conglomerate development depth segment;

[0088] The extraction module is used to obtain a binary image of conglomerate features based on the threshold segmentation result.

[0089] Preferably, the separation module comprises:

[0090] A distinguishing module is used to extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features;

[0091] A marking module is used to assign a unique identification value to each connected domain;

[0092] A setting module is used to extract the relative center points of each connected domain and set them as local extreme points of the watershed algorithm;

[0093] A module is defined for generating a water collection basin based on each local extreme point and forming a watershed at the boundary of the water collection basin;

[0094] The separation module is used to obtain the watershed influence area based on the threshold segmentation result and obtain the segmentation boundary of the watershed, so as to achieve fine separation of the cohesive gravel features.

[0095] Beneficial effects of the present invention:

[0096] 1. The method of the present invention can: 1) automatically determine the conglomerate development segment through intelligent recognition of imaging logging image patterns; 2) use a two-dimensional wavelet transform method to adaptively determine the segmentation threshold of imaging logging images in different windows to achieve automatic separation of gravel and matrix background; 3) further process the segmented gravel features through image analysis methods to achieve effective separation of cohesive gravel; 4) establish a rounding and sorting evaluation technology for conglomerate development segments through quantitative statistics of gravel parameters, laying a foundation for effective reservoir identification of conglomerate reservoirs;

[0097] 2. The present invention can quickly realize image processing processes such as identification of gravel development sections in electrical imaging logging images, automatic threshold segmentation, and adhesion feature segmentation, and accurately calculate the quantitative parameters of gravel reservoirs, providing important technical support for oil and gas field exploration and development.

[0098] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0100] Figure 1 A flow chart showing a method for quantitatively extracting gravel parameters of a conglomerate reservoir according to the present invention is shown;

[0101] Figure 2 It shows the static, dynamic and full borehole image information diagrams obtained by processing the imaging logging data;

[0102] Figure 3 The result diagram of conglomerate intelligent identification is shown;

[0103] Figure 4 A two-dimensional wavelet transform segmentation effect diagram is shown;

[0104] Figure 5A comparison diagram of the segmentation results of the adhesion gravel features of the imaging logging image is shown;

[0105] Figure 6 A quantitative evaluation parameter diagram of gravel reservoirs extracted from imaging logging images is shown;

[0106] Figure 7 A framework diagram of a conglomerate reservoir gravel parameter quantitative extraction system is shown. DETAILED DESCRIPTION

[0107] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0108] A method for quantitatively extracting gravel parameters of conglomerate reservoirs, such as Figure 1 As shown, the following steps are included:

[0109] S1: Perform image preprocessing, image enhancement and full-bore image processing on the imaging logging data to obtain full-bore logging image information; S2: Identify the full-bore logging image information to determine the conglomerate development depth section; S3: Use two-dimensional wavelet transform to perform threshold segmentation on the conglomerate development depth section to obtain a binary image of the conglomerate characteristics; S4: Process the segmented conglomerate characteristics, including using the connected domain-watershed segmentation algorithm to finely separate the bonded gravels in the characteristics; S5: Based on the fine separation results of the conglomerate characteristics, perform quantitative statistics of gravel parameters and reservoir evaluation.

[0110] Steps S1 to S5 are described in detail below. In step S1, the following steps are included:

[0111] S101: Acceleration correction and equalization processing are performed on the imaging logging data to obtain the original electrical imaging logging image information; S102: Static enhancement processing is performed on the electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section; S103: Dynamic enhancement processing is performed on the electrical imaging logging image information to obtain image information reflecting the change of rock structure; S104: Based on the image information of the change of relative resistivity of rocks in the entire well section and the image information of the change of rock structure, probability interpolation processing is performed on the uncovered parts in the static image and the dynamic image to obtain electrical imaging logging image information covering the entire wellbore.

[0112] It should be noted that in steps S101 to S104, after preprocessing the electrical imaging image (imaging logging data), the data is scaled to the pixel value range of 0-255, and the full-borehole image repair processing is performed to obtain electrical imaging logging image information with full-borehole coverage. At present, electrical imaging logging is measured by the electrode plate sticking to the well wall, and the coverage rate is usually only 60% to 80%. The full-borehole image processing method is used to perform probability interpolation processing on the uncovered parts in the static image and the dynamic image to obtain electrical imaging logging image information with full-borehole coverage. Through full-borehole image processing, the blank parts in the original image can be effectively compensated, and the accuracy of the later gravel feature extraction can be improved.

[0113] Furthermore, in step S2, the following steps are included:

[0114] S201: Construct an image classification deep learning model based on a convolutional neural network; S202: Perform convolution calculation based on the image classification deep learning model to obtain the probabilities of various image types in the full-bore logging image information; S203: Automatically classify various types of electrical imaging image features in the full-bore logging image information based on the probabilities of various image types; S204: Automatically identify the conglomerate development depth section based on the classification results, and record the depth interval.

[0115] It should be noted that in steps S201 to S204, a deep learning method is used to perform imaging logging image pattern recognition. The image classification deep learning model is mainly constructed through a convolutional neural network. Then, the imaging logging image is convolved using convolution kernels of different sizes to extract feature values ​​between different receptive domains. Finally, the probabilities of various image types are calculated through an activation function, and various types of electrical imaging image features are automatically classified. The conglomerate development depth section is automatically identified based on the classification results, and the depth interval is recorded. No calculation is performed in the non-conglomerate characteristic layer section, which can improve the processing efficiency to a certain extent.

[0116] In addition, in steps S201 to S203, the image classification deep learning model includes an input layer, a convolution layer, an activation layer and an output layer; wherein the full-bore logging image information is received by the input layer and then transmitted to the convolution layer, through which the convolution calculation of the full-bore logging image information is performed. The convolution calculation results in a matrix size as shown in formula (1). The activation layer uses an activation function to perform linear or nonlinear mapping of the image information to enhance the expression ability of the network; the output layer connects the convolution information of each channel and finally converts the input full-bore logging image information into output data.

[0117]

[0118] In the formula, size is the size of the input feature matrix, k is the size of the convolution kernel, step is the convolution stride, and size′ is the size of the transformed matrix. In addition, since the image size is not fixed, different convolution operations will cause the convolution operation to fail. Therefore, the number of zero-padding layers p is set so that the convolution kernel moves just to the edge of the feature matrix, which is the size of the output feature matrix.

[0119] Secondly, for each convolution layer, each convolution layer has a different weight matrix W, input matrix X and output matrix Y. Using conv2D to replace the two-dimensional convolution operation process, the convolution calculation formula can be expressed as formula (2), which mainly includes functions such as sigmoid, tanh, relu, leakyrelu, etc.

[0120] Convolution calculation formula:

[0121]

[0122] In the formula, is the activation function; conv2D is a two-dimensional convolution; W is a different weight matrix for each convolution layer; X is the input matrix for each convolution layer; Y is a different output matrix for each convolution layer; valid is the effective percentage of the current layer, which is used to prevent the model from overfitting; bias is the bias parameter, which is equivalent to the linear additional parameter of each layer, which can better handle unbalanced data sets;

[0123] The loss function E of the lth layer l :

[0124]

[0125] Where d is the expected output, y l is the output vector of the lth layer;

[0126] Finally, like the traditional BP neural network, the convolutional neural network is also trained by gradient descent and back propagation. As the name implies, back propagation is from the output layer to the input layer. By comparing the error between the result propagated to the input layer and the original input layer, for each weight parameter, the partial derivative is used to measure the impact on the whole. The chain rule formula used is shown below.

[0127] Chain rule formula:

[0128]

[0129] In the formula, E total is the total error size, indicating the sum of all errors, w i is the weight parameter of the i-th layer, and L is the total number of layers of the neural network.

[0130] It should be noted that formulas (1) to (4) are important components of the image classification deep learning model.

[0131] Furthermore, in step S3, the following steps are included:

[0132] S301: obtaining a two-dimensional wavelet transform formula; S302: obtaining the size of each conglomerate development depth segment; S303: performing threshold segmentation based on the two-dimensional wavelet transform formula and the size of the conglomerate development depth segment; S304: obtaining a binary image of conglomerate characteristics based on the threshold segmentation result.

[0133] It should be noted that in step S3, since the depth segment length of each conglomerate development is different, different thresholds need to be used for threshold segmentation to obtain a fine binary image. Using two-dimensional wavelet transform for threshold segmentation, there is no need to manually adjust the threshold size, and the automatic segmentation of gravel features can be achieved to obtain the corresponding binary image.

[0134] The extreme value points of the gradient change in a certain direction are found by using three-point peak search on the second-order wavelet domain transform image. The maximum value points of these gradient changes in the image are the edge points between the gravel target and the background rock in the imaging data. The imaging data is preliminarily segmented according to the average value of the imaging data pixels corresponding to the edge point coordinates and related parameters as the segmentation threshold.

[0135] In step S3, the derivation process of the two-dimensional wavelet transform formula is as follows:

[0136] Let θ(ξ) be a smooth function and define two sub-waves,

[0137]

[0138] make Corresponding to the smoothing functions in the horizontal and vertical directions.

[0139] The corresponding wavelet transforms in the two directions are:

[0140]

[0141]

[0142] So we get the following two-dimensional wavelet transform formula:

[0143]

[0144] Where W 1 f is the horizontal two-dimensional wavelet transform formula; W 2 f is the two-dimensional wavelet transform formula in the vertical direction; x and y are the horizontal and vertical coordinates of the corresponding two-dimensional data; θ s is the corresponding smoothing function; is the gradient vector identifier; f is the mapping relationship of the function; S is the size of the conglomerate development depth section, and the formula is:

[0145] s={2j}j∈Z

[0146] Where j is the maximum point of the modulus along the angular direction; Z is the range of the angular direction.

[0147] In addition, the modulus of S is the square root of the sum of the two-dimensional wavelet transform in the vertical direction and the two-dimensional wavelet transform in the horizontal direction:

[0148]

[0149] The argument of S is the angle corresponding to the ratio of the two-dimensional wavelet transform in the vertical direction to the two-dimensional wavelet transform in the horizontal direction:

[0150]

[0151] For a given j value, the modulus maximum along the amplitude direction can be calculated. The modulus maximum point is the spatial point with the steepest change. These modulus maximum points are the dividing points between the target and the background rock.

[0152] Furthermore, in step S4, the following steps are included:

[0153] S401: extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features; S402: assign a unique identification value to each connected domain; S403: extract the relative center points of each connected domain and set them as local extreme points of the watershed algorithm; S404: generate a watershed based on each local extreme point and form a watershed at the boundary of the watershed; S405: obtain the watershed influence area based on the threshold segmentation result and obtain the segmentation boundary of the watershed, so as to achieve fine separation of the cohesive gravel features.

[0154] It should be noted that the connected domain labeling algorithm is a type of algorithm that identifies each unconnected independent area in an image, and generally selects a 4-domain or 8-domain identification type; the watershed segmentation algorithm regards the image as a geodesic topological landform. The grayscale value of each pixel in the image represents the altitude of the point. Each local extreme value and its affected area are called a watershed, and the boundary of the watershed forms a watershed.

[0155] It should be further explained that the connected domain-watershed segmentation algorithm proposed in the present invention combines the advantages of the connected domain identification method and the watershed segmentation algorithm, and has an excellent effect on segmenting the features of adhered gravel. First, the distance formula is used to transform the image distance as shown in formula (11). Distance is a metric distance formula. Common metric distances include Euclidean distance as shown in formula (12), city block distance as shown in formula (13), chessboard distance as shown in formula (14), Q i is the starting point of the pixel in the image, Q j is a relative point, and the connected domain identification method is used to limit the range of distance transformation to the same connected domain C. The so-called connected domain is a pixel set composed of adjacent pixels with the same pixel value.

[0156]

[0157]

[0158] distance city =|x1-x2|+|y1-y2|; (13)

[0159] distance chess =max(|x1-x2|+|y1-y2|); (14)

[0160] In the formula, F i is the shortest distance currently stored; Q i is a fixed point; Q j is the j-th point with a pixel value of 0; distance is the distance calculation formula; x and y are the horizontal and vertical coordinates of point Q.

[0161] Furthermore, in step S5, the quantitative statistics mainly include the percentage of gravel characteristic pixels, average particle size, average roundness and average number. The reservoir space of the sandy conglomerate reservoir generally includes dissolved pores, primary pores and fractures, and mold pores and matrix micropores are seen locally. Based on parameters such as the percentage of gravel characteristic pixels, average particle size, average roundness and average number, the sedimentary type, morphology, lithology, physical properties and distribution scale of the sandy conglomerate body can be evaluated.

[0162] The present invention counts the number of pixels in each connected domain based on the fine segmentation result, and uses the number to express the color of the display.

[0163] Among them, the gravel characteristic pixel ratio describes the overall development of gravel in each depth segment, and the single pixel ratio is the ratio of the number of gravel characteristic pixels to the total number of pixels per unit length, as shown in formula (15):

[0164] Gravel_por=P i / P all ; (15)

[0165] Where Gravel_por represents the percentage of gravel feature pixels, P i Indicates the number of gravel feature pixels per unit length, P all Indicates the total number of pixels;

[0166] The particle size Gs describes the characteristic average size and is described by Feret diameter. The single characteristic particle size Gsi is obtained by measuring the diameter of the characteristic object in multiple directions and taking the average value; the characteristic average particle size within a unit depth range is as shown in formula (16):

[0167]

[0168] In the formula, Gs i is the characteristic particle size of the i-th gravel; N is the total number of gravels.

[0169] The average circularity describes the circularity of the feature object as follows:

[0170] R=C 2 / A; (17)

[0171] Where R is the roundness of the gravel feature; C is the circumference of the gravel feature; A is the area of ​​the gravel feature;

[0172] The average number is used to describe the density of feature objects in a unit depth segment, which is the number of independent connected domains in a unit depth segment.

[0173] Taking a certain oil field as an example, steps S1 to S5 are implemented as follows:

[0174] Among them, the S1 stage mainly carries out electrical imaging logging data preprocessing, image enhancement and full borehole image processing. The processed static and dynamic images are as follows Figure 2 The first and Figure 2 As shown in the second track, the blank strips in the image account for about 20%; the dynamic image uses the data within the window length as a comparison for image enhancement, and the local features are more obvious; the static image uses the entire image as a comparison for image enhancement, and the overall contrast features are retained. Figure 2 As shown in the third track, the original blank area is almost indistinguishable to the naked eye. Obviously, through the full borehole image processing, the blank part of the original image is completed, the image features are more complete and intuitive, and lays the foundation for the subsequent automatic recognition of image features and parameter extraction.

[0175] In the S2 stage, the imaging logging image pattern is intelligently recognized to automatically determine the conglomerate development segments. The convolutional neural network model established in this example has a total of 12 layers, including an input layer, an output layer, 3 convolutional pooling layers, and 7 convolutional layers. The specific structure is shown in Table 1 below:

[0176] Table 1 Convolutional neural network model structure

[0177]

[0178] Figure 3 It is the lithology identification result of the corresponding layer segment. Tracks 2-3 in the figure are static and dynamic images of imaging logging, and Track 4 is the identification result of the conglomerate segment. It can be seen from the figure that based on the image features of electrical imaging, different layers are divided into two categories (conglomerate segment and non-conglomerate segment) in the vertical direction. The automatic classification result is basically the same as the naked eye identification. This method can be used to identify and distinguish the conglomerate development segments well, laying the foundation for subsequent quantitative processing.

[0179] Among them, in the S3 stage, the automatic threshold segmentation processing uses the two-dimensional wavelet transform method to calculate the segmentation threshold in different gravel depth segments, and uses the segmented threshold to perform threshold segmentation on the data in different window lengths. The segmentation effect is as follows: Figure 4 As shown in the figure, the second track is the full-bore dynamic image of imaging logging, and the third track is the result of two-dimensional wavelet transform automatic threshold segmentation processing. It can be seen from the figure that the gravel features can be automatically segmented into the binary image regardless of whether the overall feature is bright or dark, and the segmented gravel part has a good consistency with the image.

[0180] Among them, in the S4 stage, the adhesion of gravel features is processed. The adhesion of gravel is usually weak, and the contact area between spheres is small. Therefore, the connected domain-watershed algorithm is used, and the maximum value of the distance transformation result is used as the minimum value of the watershed algorithm depression. The connected domain is used to divide different areas. The advantages of the connected domain identification algorithm and the watershed segmentation algorithm are combined. Based on the dynamic image of the electrical imaging logging of the whole wellbore, the segmented independent gravel features are extracted. For larger multi-elliptical image features, the separated gravel features are separated in different connected domains with a circular watershed boundary as the boundary on the image; for slender small adhesion gravel features, the slender image features will be divided into several small circular gravels according to the different segmentation ratio settings. The segmentation effect is as follows: Figure 5 As shown in the figure, the dark color of the segmentation result indicates the size of the gravel, where the features of the adhered gravel are effectively separated, and different colors indicate the size after segmentation. Overall, the sum of the areas occupied by each separated region is approximately equal to the area of ​​the original image.

[0181] Among them, in S5 stage 5, the quantitative statistics of gravel parameters are carried out, mainly based on the segmentation results of the adhesion gravel characteristics, and the quantitative evaluation parameters of the gravel development section are calculated. Figure 6 The calculated particle size distribution parameters, roundness distribution parameters and gravel parameters are displayed. The degree of gravel development can be analyzed through the size of the flat particle size and the particle size distribution within the window length. The trend of the calculated gravel parameters is basically consistent with the image characteristics. The degree of gravel roundness can be analyzed through the gravel roundness distribution and the maximum and minimum values ​​of the roundness, and then the pore type can be determined and the effective reservoir can be identified.

[0182] A system for quantitatively extracting gravel parameters from conglomerate reservoirs, such as Figure 7 As shown, it includes a processing unit, an identification unit, a segmentation unit, a separation unit and a statistical unit. The processing unit is used to perform image preprocessing, image enhancement and full-bore image processing on the imaging logging data to obtain full-bore logging image information; the identification unit is used to identify the full-bore logging image information and determine the conglomerate development depth section; the segmentation unit is used to perform threshold segmentation on the conglomerate development depth section using two-dimensional wavelet transform to obtain a binary image of the conglomerate characteristics; the separation unit is used to process the segmented conglomerate characteristics, including using the connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the conglomerate characteristics; the statistical unit is used to perform quantitative statistics of gravel parameters and reservoir evaluation based on the fine separation results of the conglomerate characteristics.

[0183] Furthermore, the processing unit includes:

[0184] The initial processing module is used to perform acceleration correction and equalization processing on the imaging logging data to obtain the original electrical imaging logging image information;

[0185] Static processing module, used to perform static enhancement processing on electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section;

[0186] Dynamic processing module, used to dynamically enhance the electrical imaging logging image information to obtain image information reflecting the changes in rock structure;

[0187] The interpolation processing module is used to perform probability interpolation processing on the uncovered part of the image based on the image information of the relative resistivity change of the rock in the whole well section and the image information of the rock structure change, so as to obtain the electrical imaging logging image information covering the whole wellbore.

[0188] Furthermore, the identification unit comprises:

[0189] Building blocks for building deep learning models for image classification based on convolutional neural networks;

[0190] The convolution module is used to perform convolution calculations based on the image classification deep learning model to obtain the probabilities of various image types in the full-bore logging image information;

[0191] A classification module for automatically classifying various types of electrical imaging image features in the full-bore logging image information based on the probabilities of various image types;

[0192] The identification module is used to automatically identify the conglomerate development depth section based on the classification results and record the depth interval.

[0193] Furthermore, the segmentation unit comprises:

[0194] A first acquisition module is used to acquire a two-dimensional wavelet transform formula;

[0195] The second acquisition module is used to obtain the size of each conglomerate development depth section;

[0196] A segmentation module, used for performing threshold segmentation based on a two-dimensional wavelet transform formula and the size of the conglomerate development depth segment;

[0197] The extraction module is used to obtain a binary image of conglomerate features based on the threshold segmentation result.

[0198] Furthermore, the separation module comprises:

[0199] A distinguishing module is used to extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features;

[0200] A marking module is used to assign a unique identification value to each connected domain;

[0201] A setting module is used to extract the relative center points of each connected domain and set them as local extreme points of the watershed algorithm;

[0202] A module is defined for generating a water collection basin based on each local extreme point and forming a watershed at the boundary of the water collection basin;

[0203] The separation module is used to obtain the watershed influence area based on the threshold segmentation result and obtain the segmentation boundary of the watershed, so as to achieve fine separation of the cohesive gravel features.

[0204] It should be noted that, for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The various units and modules of the system for quantitative extraction of gravel parameters in conglomerate reservoirs are only divided according to functional logic, but are not limited to the above divisions, as long as the corresponding functions can be achieved; in addition, the specific names of the units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0205] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for quantitatively extracting gravel parameters from a conglomerate reservoir, characterized in that: The following steps are involved: Perform image preprocessing, image enhancement and full-bore image processing on imaging logging data to obtain full-bore logging image information; Identify the full borehole logging image information to determine the depth section of conglomerate development; For the conglomerate development depth section, two-dimensional wavelet transform is used to perform threshold segmentation to obtain a binary image of conglomerate characteristics; Processing the segmented conglomerate features, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the conglomerate features; Based on the results of fine separation of conglomerate characteristics, quantitative statistics of gravel parameters and reservoir evaluation are carried out.

2. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 1, characterized in that: The imaging logging data is subjected to image preprocessing, image enhancement and full borehole image processing to obtain full borehole logging image information, which mainly includes the following steps: Perform acceleration correction and equalization processing on imaging logging data to obtain original electrical imaging logging image information; Perform static enhancement processing on the electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section; Dynamically enhance the electrical imaging logging image information to obtain image information reflecting the changes in rock structure; Based on the image information of rock relative resistivity changes and rock structure changes in the entire well section, probability interpolation processing is performed on the uncovered part of the image to obtain electrical imaging logging image information covering the entire wellbore.

3. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 1, characterized in that: The identification of full borehole logging image information and determination of the conglomerate development depth section include the following steps: Build a deep learning model for image classification based on convolutional neural networks; Convolution calculation is performed based on the image classification deep learning model to obtain the probability of various image types in the full-bore logging image information; Automatically classify various types of electrical imaging image features in full-bore logging image information based on the probabilities of various image types; Based on the classification results, the conglomerate development depth section is automatically identified and the depth interval is recorded.

4. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 3, characterized in that: The image classification deep learning model includes an input layer, a convolution layer, an activation layer and an output layer; The full borehole logging image information is received by the input layer and then sent to the convolutional layer; The convolution layer is used to perform convolution calculation of the full borehole logging image information; The activation layer uses the activation function to perform linear or nonlinear mapping of image information to enhance the network's expressive power. The output layer connects the convolution information of each channel and finally converts the input full-bore logging image information into output data.

5. A method for quantitatively extracting gravel parameters from a conglomerate reservoir according to claim 3 or 4, characterized in that: The image classification deep learning model includes: Matrix transformation calculation formula: In the formula, size is the size of the input feature matrix; k is the size of the convolution kernel; step is the convolution stride; size′ is the size of the transformed matrix; p is the number of zero-padding layers; Convolution calculation formula: In the formula, is the activation function; conv2D is a two-dimensional convolution; W is a different weight matrix for each convolution layer; X is the input matrix for each convolution layer; Y is a different output matrix for each convolution layer; valid is the valid percentage of the current layer, which is used to prevent the model from overfitting; bias is the bias parameter; The loss function E of the lth layer l : Where d is the expected output, y l is the output vector of the lth layer; By taking the partial derivative of the loss function E with respect to each weight, it can be expressed as the chain rule formula: In the formula, E total is the total error size, indicating the sum of all errors, w i is the weight parameter of the i-th layer, and L is the total number of layers of the neural network.

6. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 1, characterized in that: For the conglomerate development depth section, two-dimensional wavelet transform is used to perform threshold segmentation to obtain a binary image of conglomerate characteristics, including the following steps: Obtain the dimensions of each conglomerate development depth section; Threshold segmentation is performed based on the two-dimensional wavelet transform formula and the size of the conglomerate development depth segment; A binary image of conglomerate features is obtained based on the threshold segmentation result.

7. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 6, characterized in that: The two-dimensional wavelet transform formula is: Where W 1 f(s, x, y) is the two-dimensional wavelet transform formula in the horizontal direction; W 2 f(s, x, y) is the two-dimensional wavelet transform formula in the vertical direction; x and y are the horizontal and vertical coordinates of the corresponding two-dimensional data; θ s is the corresponding smooth function; is the gradient vector identifier; f is the mapping relationship of the function; S is the size of the conglomerate development depth section, and the formula is: s={2j}j∈Z Where j is the maximum point of the modulus along the angular direction; Z is the range of the angular direction.

8. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 1, characterized in that: The segmented conglomerate features are processed, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the conglomerate features, including the following steps: Extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features; Assign a unique identification value to each connected domain; Extract the relative center points of each connected domain and set them as the local extreme points of the watershed algorithm; Generate a watershed basin based on each local extreme point, and form a watershed at the boundary of the watershed basin; The watershed influence area is obtained based on the threshold segmentation result, and the segmentation boundary of the watershed is obtained, so as to finely separate the cohesive gravel features.

9. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 1, characterized in that: Based on the fine separation results of conglomerate characteristics, quantitative statistics of gravel parameters and reservoir evaluation are carried out, including: Quantitative statistics of gravel characteristic pixel ratio, average particle size, average roundness and average number; The sedimentary type, morphology, lithology, physical properties and distribution scale of the gravel body are evaluated based on the proportion of gravel characteristic pixels, evaluated particle size, average roundness and average number.

10. The method for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 9, characterized in that: The gravel feature pixel ratio is: Gravel_by=P i / P all ; Where Gravel_por represents the percentage of gravel feature pixels, P i Indicates the number of gravel feature pixels per unit length, P all Indicates the total number of pixels; The average particle size Gs is: In the formula, Gs i is the characteristic particle size of the i-th gravel; N is the total number of gravels; The average roundness is: R=C 2 / A; Where R is the roundness of the gravel feature; C is the circumference of the gravel feature; A is the area of ​​the gravel feature; The average number is used to describe the density of feature objects in a unit depth segment, which is the number of independent connected domains in a unit depth segment.

11. A conglomerate reservoir gravel parameter quantitative extraction system, characterized in that: include: A processing unit, used for performing image preprocessing, image enhancement and full-bore image processing on the imaging logging data to obtain full-bore logging image information; An identification unit is used to identify the full borehole logging image information and determine the conglomerate development depth section; The segmentation unit is used to perform threshold segmentation on the conglomerate development depth section using two-dimensional wavelet transform to obtain a binary image of conglomerate characteristics; A separation unit is used to process the segmented gravel features, including using a connected domain-watershed segmentation algorithm to finely separate the cohesive gravels in the gravel features; The statistical unit is used to conduct quantitative statistics of gravel parameters and reservoir evaluation based on the fine separation results of gravel characteristics.

12. A conglomerate reservoir gravel parameter quantitative extraction system according to claim 11, characterized in that: The processing unit comprises: The initial processing module is used to perform acceleration correction and equalization processing on the imaging logging data to obtain the original electrical imaging logging image information; Static processing module, used to perform static enhancement processing on electrical imaging logging image information to obtain image information reflecting the change of relative resistivity of rocks in the entire well section; Dynamic processing module, used to dynamically enhance the electrical imaging logging image information to obtain image information reflecting the changes in rock structure; The interpolation processing module is used to perform probability interpolation processing on the uncovered part of the image based on the image information of the relative resistivity change of the rock in the whole well section and the image information of the rock structure change, so as to obtain the electrical imaging logging image information covering the whole wellbore.

13. The system for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 11, characterized in that: The identification unit comprises: Building blocks for building deep learning models for image classification based on convolutional neural networks; The convolution module is used to perform convolution calculations based on the image classification deep learning model to obtain the probabilities of various image types in the full-bore logging image information; A classification module for automatically classifying various types of electrical imaging image features in the full-bore logging image information based on the probabilities of various image types; The identification module is used to automatically identify the conglomerate development depth section based on the classification results and record the depth interval.

14. The system for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 11, characterized in that: The segmentation unit comprises: A first acquisition module is used to acquire a two-dimensional wavelet transform formula; The second acquisition module is used to obtain the size of each conglomerate development depth section; A segmentation module, used for performing threshold segmentation based on a two-dimensional wavelet transform formula and the size of the conglomerate development depth segment; The extraction module is used to obtain a binary image of conglomerate features based on the threshold segmentation result.

15. The system for quantitatively extracting gravel parameters of a conglomerate reservoir according to claim 11, characterized in that: The separation module comprises: A distinguishing module is used to extract all connected domains with pixel values ​​of 0 in the binary image of conglomerate features; A marking module is used to assign a unique identification value to each connected domain; A setting module is used to extract the relative center points of each connected domain and set them as local extreme points of the watershed algorithm; A module is defined for generating a water collection basin based on each local extreme point and forming a watershed at the boundary of the water collection basin; The separation module is used to obtain the watershed influence area based on the threshold segmentation result and obtain the segmentation boundary of the watershed, so as to achieve fine separation of the cohesive gravel features.

Images