A precise and efficient identification and classification method for shale bedding structure based on deep learning

Abstract

The present invention relates to a method for accurately and efficiently identifying and classifying shale bedding structures based on deep learning. The method comprises: calculating shale bedding structure pre-classification parameters based on fast shear wave time difference curves, slow shear wave time difference curves, and well logging curves correlated with shale bedding structures in target block logging data; obtaining segmented electrical imaging logging images and preliminary classification results based on the shale bedding structure pre-classification parameters; inputting the segmented electrical imaging logging images into different bedding structure identification network models based on the preliminary classification results to obtain shale bedding structure identification results, and analyzing the identification results using image morphology to obtain final classification results. The bedding structure identification network model is constructed based on a Swin Transformer network model and trained based on a training set. The present invention proposes a new method for fine-tuning the classification of shale bedding structure types.

Description

Technical Field

[0001] The present invention relates to the technical field of shale classification, and in particular to a method for accurately and efficiently identifying and classifying shale bedding structures based on deep learning. Background Art

[0002] In recent years, significant breakthroughs have been made in the exploration and development of continental shale oil, leading to industrial production. Continental shale reservoirs are characterized by well-developed stratification, a rich variety of types, and diverse assemblages. Research has shown that frequent lamination and thin interlayers enhance shale reservoir heterogeneity, creating favorable conditions for stable shale oil seepage and favorable pathways for fracturing. Therefore, accurate lamination identification is crucial for selecting geological sweet spots.

[0003] At present, conventional logging curve feature analysis and reconstruction methods are usually used to determine the development of bedding and interlayers in shale sedimentary structures. However, due to the low accuracy of logging curves and many influencing factors, the prediction error of shale bedding structure type is large. Microresistivity imaging logging (electrical imaging logging) has a high resolution and a relatively complete coverage range, and can intuitively and visually present geological characteristic information such as formation shape and spatial position. Therefore, electrical imaging logging can play an important role in the identification of fractures and bedding in oil and gas reservoirs. When using electrical imaging logging to identify shale bedding structure types, traditional methods mainly rely on manual evaluation methods to divide bedding and thin interlayers in the reservoir, and calibrate the category of shale bedding structure by the shape and number of bedding. Manual identification methods are inefficient and are affected by human experience and subjective factors.

[0004] And the existing technology has the following shortcomings:

[0005] 1. The changing characteristics of conventional logging curves are affected by many factors. The analysis of conventional logging curves can characterize the development characteristics of shale bedding to a certain extent, but it is impossible to accurately distinguish the types of shale bedding structures, especially for shale reservoirs where laminar, layered, and massive structures coexist. The distinction effect is very unsatisfactory. 2. At present, when processing electrical imaging logging images, the images are cut at equidistant heights, which may cut a set of bedding into two images separately, resulting in errors when using machine learning methods to determine the type of shale bedding structure. In addition, the same set of artificial intelligence network models is used when evaluating the images used, resulting in low accuracy and efficiency. 3. When using machine learning methods to identify shale bedding based on electrical imaging logging results, the more efficient and accurate Swin Transformer artificial intelligence technology has not been used. Summary of the Invention

[0006] The purpose of this invention is to provide a method for accurately and efficiently identifying and classifying shale bedding structures based on deep learning. Based on traditional well logging curves, a new method for calculating characteristic parameters that characterize shale bedding structure types is proposed. Based on these characteristic parameters, a standard for segmenting electrical imaging well logging images is constructed. For different types of shale bedding structures, a new method is proposed using different Swin Transformer artificial intelligence network models to identify and obtain shale bedding geometric parameters. The type of shale bedding structure is determined by bedding density and proportion. This method further improves the accuracy and speed of shale bedding structure classification, providing theoretical and technical support for identifying "sweet spots" in shale reservoirs.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A deep learning-based method for accurately and efficiently identifying and classifying shale bedding structures includes:

[0009] Calculate shale bedding structure pre-classification parameters based on the fast shear wave time difference curve, slow shear wave time difference curve and the logging curve related to the shale bedding structure in the target block logging data;

[0010] Obtaining a segmented electrical imaging logging map and a preliminary classification result based on the shale bedding structure pre-classification parameters;

[0011] According to the preliminary classification results, the segmented electrical imaging logging map is input into different bedding recognition network models to obtain shale bedding structure recognition results, and the recognition results are analyzed using image morphology to obtain the final classification results, wherein the bedding recognition network model is constructed based on the Swin Transformer network model and obtained through training based on a training set, and the training set includes electrical imaging logging maps with shale bedding annotated with different types of bedding structures.

[0012] Optional parameters for calculating shale bedding structure pre-classification include:

[0013] Calculating anisotropy coefficients at different depths of the shale reservoir based on the fast shear wave time difference curve and the slow shear wave time difference curve;

[0014] Calculating the box-counting dimension of the logging curve at different depths of the reservoir based on the logging curve that is correlated with the shale bedding structure, wherein the logging curve that is correlated with the shale bedding structure is an acoustic time difference curve, a natural gamma curve, or a neutron curve;

[0015] The shale bedding structure pre-classification parameters are obtained according to the anisotropy coefficient and the box counting dimension.

[0016] Optionally, the anisotropy coefficient of the shale reservoir at different depths is calculated as:

[0017]

[0018] The box counting dimension of the logging curve at different depths of the reservoir is calculated as:

[0019]

[0020] The shale bedding structure pre-classification parameters are obtained as follows:

[0021] P p =D f ANI

[0022] Where ANI is the anisotropy coefficient of shear wave time difference; s1 is the fast shear wave time difference; s2 is the slow shear wave time difference; D f is the box counting dimension; r is the length of the box; N(r) is the minimum number of boxes covering the well logging curve, P p Pre-classification parameters for shale bedding structure.

[0023] Optionally, obtaining the segmented electrical imaging log map and preliminary classification results includes:

[0024] Normalizing the shale bedding structure pre-classification parameters to obtain normalized pre-classification parameters;

[0025] Comparing the normalized pre-classification parameter with a first preset classification standard to obtain the preliminary classification result;

[0026] According to the shale bedding structure pre-classification parameters, the electrical imaging logging map is cut in a sliding window manner to obtain the segmented electrical imaging logging map.

[0027] Optionally, the bedding identification network model includes: a complex bedding identification network model and a simple bedding identification network model;

[0028] The complex bedding recognition network model is constructed based on the Swin Transformer network model and is used to identify laminar and layered shale bedding structures;

[0029] The simple bedding identification network model is constructed based on the Swin Transformer network model and is used to identify the bedding structure of massive shale.

[0030] Among them, the Swin Transformer network model uses Swin-L as the benchmark network. The Swin-L network model for constructing the complex bedding recognition network model selects high feature dimension and number of attention heads, and sets the size of the attention layer window according to the hybrid strategy. The Swin-L network model for constructing the simple bedding recognition network model selects low feature dimension and number of attention heads, and the window for attention layer processing is set to a medium window.

[0031] Optionally, the layer recognition network model adopts a total loss function including cross entropy loss and IOU loss;

[0032] The total loss function is:

[0033] Total Loss=α·CE+β·IoU Loss

[0034]

[0035]

[0036] Where y is the true label; is the predicted output of the model; y i is the true label of the i-th sample; is the predicted probability of the i-th sample; N is the number of samples; A is the true area; B is the predicted area; α is the weight of the cross loss function; β is the weight of the IoU loss function.

[0037] Optionally, obtaining the segmented electrical imaging log includes:

[0038] Filling the blank strips in the electrical imaging logging image, performing threshold segmentation, and outputting a binary image;

[0039] The binary image is filtered using a Sobel operator to obtain the pre-processed electrical imaging logging image.

[0040] Optionally, image morphology is used to analyze the recognition results to obtain the final classification results, including:

[0041] Image morphology is used to detect the edges of the identified beddings, calculate the thickness of each bedding and count the number;

[0042] Calculate the bedding ratio and bedding density based on the thickness and number of each bedding, wherein the bedding ratio is the ratio of the total thickness of the bedding in the lithologic body area to the thickness of the lithologic body, and the bedding density is the ratio of the number of beddings in the lithologic body area to the thickness of the lithologic body;

[0043] The bedding ratio and bedding density are compared with a second preset classification standard to obtain the final classification result.

[0044] The beneficial effects of the present invention are as follows: the present invention proposes for the first time the pre-classification parameters of shale bedding structure, and uses it as the basis for segmenting electrical imaging images, which can segment electrical imaging logging images more accurately. Using the Swin Transformer deep learning network, two sets of network models for bedding identification were established for structurally complex bedding (laminated, layered) and structurally simple bedding (blocky). For shale reservoirs where laminar, layered, and blocky bedding coexist, when using a dual network model to identify bedding, the efficiency of bedding identification is further improved on the basis of ensuring the accuracy of bedding identification. On this basis, a new method using dual parameters of bedding density and proportion is proposed to identify the type of shale bedding structure, which provides an important basis for the economy, efficiency and rationality of shale oil reservoir evaluation and "sweet spot" identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] 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. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 This is a diagram showing the calculation of pre-classification parameters of a well logging curve and the classification results of shale bedding structure using electrical imaging logging according to an embodiment of the present invention;

[0047] Figure 2 The segmentation results of different types of shale bedding structures in the electrical imaging logging diagram according to an embodiment of the present invention;

[0048] Figure 3 This is the Swin Transformer result of identifying laminated shale bedding in an embodiment of the present invention;

[0049] Figure 4 This is the result of Swin Transformer identifying layered shale bedding in an embodiment of the present invention;

[0050] Figure 5 This is the result of Swin Transformer identifying massive shale bedding in an embodiment of the present invention;

[0051] Figure 6 This is a flow chart of a method for accurately and efficiently identifying and classifying shale bedding structures based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] Electrical imaging logging technology can accurately characterize the development of shale bedding, and artificial intelligence systems can accurately and efficiently identify bedding in electrical imaging logging images. Therefore, building an artificial intelligence system that can automatically and accurately identify bedding in electrical imaging logging images is of great significance. It can effectively solve the time-consuming and labor-intensive problem of human-computer interaction in bedding identification and avoid the problem of errors in shale bedding structure identification caused by subjective evaluation based on human experience.

[0055] like Figure 6 As shown, this embodiment provides a method for accurately and efficiently identifying and classifying shale bedding structures based on deep learning, including:

[0056] Calculate shale bedding structure pre-classification parameters based on the fast shear wave time difference curve, slow shear wave time difference curve and the logging curve related to the shale bedding structure in the target block logging data;

[0057] According to the pre-classification parameters of shale bedding structure, the segmented electrical imaging logging map and preliminary classification results are obtained;

[0058] Based on the preliminary classification results, the segmented electrical imaging log images were input into different bedding recognition network models to obtain shale bedding structure recognition results. These recognition results were then analyzed using image morphology to obtain the final classification results. The bedding recognition network model was constructed based on the Swin Transformer network model and trained with a training set containing electrical imaging log images labeled with different types of bedding structures. The recognition results included classifying the regional features of bedding in the segmented electrical imaging log images as white and the formation matrix as black.

[0059] Furthermore, the calculation of shale bedding structure pre-classification parameters includes:

[0060] Calculate the anisotropy coefficient of shale reservoir at different depths based on the fast shear wave time difference curve and the slow shear wave time difference curve;

[0061] Calculating the box-counting dimension of the well logging curves at different depths of the reservoir based on the well logging curves that are correlated with the shale bedding structure, wherein the well logging curves that are correlated with the shale bedding structure are acoustic transit time curves, natural gamma ray curves, or neutron curves;

[0062] The pre-classification parameters of shale bedding structure are obtained based on the anisotropy coefficient and box counting dimension.

[0063] Furthermore, the anisotropy coefficient of the shale reservoir at different depths is calculated as:

[0064]

[0065] The box counting dimension of the logging curve at different depths of the reservoir is calculated as:

[0066]

[0067] The pre-classification parameters of shale bedding structure are obtained as follows:

[0068] P p =D f ANI (3)

[0069] Where ANI is the anisotropy coefficient of shear wave time difference; s1 is the fast shear wave time difference; s2 is the slow shear wave time difference; D f is the box counting dimension; r is the length of the box; N(r) is the minimum number of boxes covering the well logging curve, P p Pre-classification parameters for shale bedding structure.

[0070] Specifically, the calculation of shale bedding structure pre-classification parameters includes: ① Based on the fast shear wave time difference curve and slow shear wave time difference curve in the conventional logging curve, the anisotropy at different depths of the shale reservoir is calculated according to formula (1). ② Based on the box dimension theory in fractal geometry, the box dimension of the logging curve at different depths of the reservoir is calculated according to formula (2) using the logging curve that is correlated with the shale bedding structure. The box dimension of the logging curve is calculated using a sliding window method, where the sliding window size is set to n data points, the sliding step is set to 1 logging data point, and the depth corresponding to the middle point of the window is the depth of the box dimension. ③ Anisotropy reflects the number, shape and type of bedding, and the box dimension represents the complexity of the bedding. Therefore, based on anisotropy and box dimension, a calculation method for shale bedding structure pre-classification parameters is established, as shown in formula (3).

[0071] Furthermore, the segmented electrical imaging log map and preliminary classification results are obtained, including:

[0072] Normalizing the shale bedding structure pre-classification parameters to obtain normalized pre-classification parameters;

[0073] Comparing the normalized pre-classification parameters with the first preset classification standard to obtain a preliminary classification result;

[0074] According to the pre-classification parameters of shale bedding structure, the electrical imaging logging map is cut using a sliding window method to obtain the segmented electrical imaging logging map.

[0075] Specifically, the segmentation and preliminary classification of electrical imaging logging images include: ① Calculating the pre-classification parameters (P p ) and normalize it. ② Based on the normalized pre-classification parameters, the classification standard of shale bedding structure is proposed, namely the first preset classification standard: P p <0.3 is massive shale; 0.3≤P p ≤0.6 is layered shale; 0.6 <P p ③ Establish cutting standards for electrical imaging logs. Based on pre-classification parameters calculated from conventional well logs, dynamically cut electrical imaging logs using a sliding window approach. ④ Due to the thin bedding, numerous, and complex structure of laminated and layered shales, identification is much more difficult than that of layered shales. Therefore, to improve identification accuracy, it is necessary to set an upper limit on the cutting height for electrical imaging maps of laminated and layered shales.

[0076] Specifically, the bedding identification network model includes: complex bedding identification network model and simple bedding identification network model;

[0077] The complex bedding recognition network model is built based on the Swin Transformer network model and is used to identify laminar and layered shale bedding structures.

[0078] The simple bedding identification network model is built based on the Swin Transformer network model and is used to identify the bedding structure of massive shale.

[0079] Among them, the Swin Transformer network model uses Swin-L as the benchmark network. The Swin-L network model for constructing a complex layer recognition network model uses high feature dimensions and attention heads, and sets the size of the attention layer window according to the hybrid strategy. The Swin-L network model for constructing a simple layer recognition network model uses low feature dimensions and attention heads, and the window for attention layer processing is set to a medium window.

[0080] Specifically, the network model for shale bedding recognition based on Swin Transformer deep learning includes the following: ① Based on Swin Transformer deep learning technology, two network models for shale bedding recognition are designed, one for identifying complex laminar and layered bedding structures, and the other for identifying simple blocky bedding structures. ② For laminar and layered bedding structures with complex structures and densely interlaced thin and thick layers, Swin Transformer network models with high recognition accuracy but high computational load are constructed. Using Swin-L as the baseline network, a smaller patch size is set, the attention layer window size is set according to a hybrid strategy, and high feature dimensions and number of attention heads are selected in the four stages of image processing. ③ For simple and small blocky bedding structures, a Swin Transformer network model with low computational load and high recognition efficiency is constructed while ensuring recognition accuracy. Using Swin-T as the baseline network, a larger patch size is selected, the attention layer processing window is set to a medium window, and low feature dimensions and number of attention heads are selected in the four stages of image processing. ④ Both Swin Transformer network models use the Adam algorithm structure to establish adaptive learning rates. The loss function is calculated by combining cross entropy loss (Formula (4)) and IOU loss (Formula (5)). The total loss function is shown in Formula (6). ⑤ Create n sample images by manual annotation, set the training set and test set in a ratio of 9:1, use the mean intersection over union (MIoU) indicator, recall rate (Recall) and precision rate (Precision) to evaluate the accuracy of the model for bedding classification and recognition, train the network model, and complete the construction of the network model.

[0081] Furthermore, the layer recognition network model adopts a total loss function including cross entropy loss and IOU loss;

[0082] The total loss function is:

[0083] Total Loss=α·CE+β·IoU Loss (4)

[0084]

[0085] Where y is the true label; is the predicted output of the model; y i is the true label of the i-th sample; is the predicted probability of the i-th sample; N is the number of samples; A is the true area; B is the predicted area; α is the weight of the cross loss function; β is the weight of the IoU loss function.

[0086] Furthermore, obtaining the segmented electrical imaging logging image includes:

[0087] Fill in the blank strips in the electrical imaging logging image, perform threshold segmentation, and output a binary image;

[0088] The binary image is filtered using the Sobel operator to obtain the preprocessed electrical imaging logging image.

[0089] Furthermore, the identified shale bedding is analyzed using image morphology to obtain the final classification results including:

[0090] Image morphology is used to detect the edges of the identified beddings, calculate the thickness of each bedding and count the number;

[0091] Based on the thickness and number of each bedding, the bedding ratio and bedding density are calculated. The bedding ratio is the ratio of the total thickness of the bedding within the lithologic body to the thickness of the lithologic body, and the bedding density is the ratio of the number of bedding within the lithologic body to the thickness of the lithologic body.

[0092] The bedding ratio and bedding density are compared with the second preset classification standard to obtain a final classification result.

[0093] Specifically, the image preprocessing and analysis include: ① first fill in the blank strips in the electrical imaging logging image. ② threshold segmentation of the image is performed to output a binary image. ③ Sobel operator is used to filter the binary image to enhance the extraction of horizontal features in the image. ④ Based on the pre-classification parameters of the shale bedding structure, the electrical imaging logging image is segmented. ⑤ According to the results of the segmentation of the electrical imaging logging image based on the pre-classification parameters, the laminar and layered bedding images are imported into a complex Swin Transformer network model for analysis, and the blocky bedding images are imported into a simple Swin Transformer network model for analysis to identify the shale bedding structure. ⑥ Image morphology is used to detect the edges of the identified beddings, calculate the thickness of each bedding and count the number. ⑦ Bedding density is defined as the ratio of the number of beddings in the lithologic body area to the thickness of the lithologic body, and the bedding ratio is the ratio of the total thickness of the bedding in the lithologic body area to the thickness of the lithologic body, as shown in formulas (6) and (7). Based on the density and ratio of bedding, the classification standards for laminar, layered and blocky shale bedding structures are proposed. When the bedding ratio is <40% and the bedding density is >0.4, it is classified as laminated shale; when the bedding ratio is <40% and the bedding density is 0.2 < ≤ 0.4, it is classified as layered shale; and when the bedding ratio is >40%, it is classified as massive shale. ⑧ When the simple network model determines that the shale bedding is laminated or layered, it needs to be imported into the complex bedding recognition network model for reclassification.

[0094]

[0095] Where H is the height of a single image, cm; is the total number of cracks in the figure; hi is the height of the ith bedding, cm, F d is the bedding density, F r is the bedding ratio.

[0096] The present invention will be further described below with reference to the accompanying drawings:

[0097] A deep learning-based method for accurately and efficiently identifying and classifying shale bedding structures includes:

[0098] (1) Calculation of shale bedding structure pre-classification parameters:

[0099] ① Collect and organize the logging data of the target block, obtain the fast shear wave time difference curve and slow shear wave time difference curve in the logging data, and calculate the heterogeneity of the target layer (3410m~3460m) according to formula (1). ② Use the acoustic wave time difference curve of the target block to calculate the box dimension. The calculation of the box dimension is implemented in Matlab, and the analysis window length is set to 11 logging data points. The TXT data of the acoustic wave time difference curve is read through Matlab, and the box dimension is calculated. The calculation results are stored in the TXT text and then exported. ③ Based on the heterogeneity and box dimension, the pre-classification parameters of the shale bedding structure at different depths of the reservoir are calculated according to formula (3). The calculation results are as follows Figure 1 ④ Coarsening the pre-classification parameters: reduce the sampling rate of the pre-classification parameters from 12.5cm to 25cm according to the mean calculation method.

[0100] (2) Segmentation of electrical imaging logging images:

[0101] ① According to the depth of the electrical imaging logging map corresponding to the pre-classification parameters of the shale bedding structure, the electrical imaging logging map of 3410m~3460m is segmented. According to the calculation results of the pre-classification parameters, there are 20 intervals of laminar shale, 25 intervals of layered shale, and 7 intervals of massive shale in the target block. ② In this embodiment, the minimum unit of the pre-classification window is 25cm, and the upper limit of the segmentation height of the electrical imaging map of the laminar and layered shale area is set to 50cm, and the segmentation height of the massive shale area is not adjusted. According to the segmentation standard of the height of the electrical imaging map of the laminar and layered shale, a total of 33 laminar shale pictures, 32 layered shale pictures, and 7 massive shale pictures are finally obtained, such as Figure 2 shown.

[0102] (3) A shale bedding structure recognition network model is constructed based on Swin Transformer deep learning, where the shale bedding structure recognition network model is used to recognize the input segmented image:

[0103] First, a network model for complex bedding recognition was constructed based on Swin Transformer deep learning and the Swin-L network model. This model is primarily used to identify laminar and layered shale bedding structures. The model's image feature extraction process consists of four stages: image segmentation and preliminary feature extraction in the first stage, further feature extraction in the second stage, high-level feature extraction in the third stage, and refinement and output in the fourth stage. These four stages include image segmentation, linear embedding, the Transformer encoder, and shifting window attention. The model uses a 3×3 patch size and a hybrid strategy for window size: 3×3 for the first two image processing stages and 7×7 for the last two stages. The number of Transformer blocks in each of the four image processing stages is set to 2, 2, 18, and 2, respectively. The feature dimensions are set to 192, 384, 768, and 1536, respectively. The number of attention heads is set to 6, 12, 24, and 48, respectively. ② Based on Swin Transformer deep learning, a network model for simple bedding recognition was constructed with the Swin-T network model as the benchmark. It is mainly used for the recognition of massive shale bedding structures. The model sets the image feature extraction process to four stages, which is the same as the complex bedding recognition network model. The patch size in the network model is set to 4×4; the window size of the four stages of image processing is set to 7×7; the number of Transformer blocks in the four stages of image processing is set to 2, 2, 6, and 2 respectively, the feature dimensions are set to 96, 192, 384, and 768 respectively, and the number of heads is set to 4, 8, 16, and 32 respectively. ④ When training the two Swin Transformer network models, the Adam algorithm is used to construct an adaptive learning rate. The initial learning rate is set to 0.001, and the upper and lower limits of the learning rate adjustment are set to 0.1 to 0.0001. The loss function is calculated by combining cross entropy loss and IOU loss. The total loss function is shown in formula (6). ⑤ 1,000 sample images were created by manually annotating different types of bedding structures in electrical imaging logs. The model was trained using a training set and a test set with a ratio of 9:1. The model's accuracy in bedding classification and recognition was evaluated using the Mean Intersection over Union (MIoU) metric, recall, and precision. The results from both models demonstrated high bedding recognition accuracy, meeting application requirements. The network model's recognition accuracy evaluation parameters are shown in Table 1.

[0104] Table 1

[0105]

[0106] (4) Identification and classification of shale bedding structure:

[0107] ① Use the Filtersim algorithm to complete blank bands in the electrical imaging log image. ② Perform threshold segmentation on the image and output a binary image. Use the Sobel operator to filter the binary image and enhance the horizontal representation of the image features. ③ Segment the electrical imaging log image based on the pre-classification parameters of the shale bedding structure. Based on the classification of the cut image, different deep learning network models are used to identify the bedding and stratigraphic matrix in the electrical imaging log image, classifying the bedding-related regional features in the image as white and the stratigraphic matrix as black. ④ Use Python-based edge detection to identify the number of beddings in the image. Import the OpenCV and Numpy libraries into Python and use the cv2.GaussianBlur function to smooth the image with a 5x5 convolution kernel to reduce noise. Furthermore, use the cv2.Canny edge detection function to enhance the edges of each bedding in the image, setting the upper and lower thresholds of the cv2.Canny function to 10 and 220, respectively. Finally, the Rtre_External retrieval mode in the cv2.findContours function is used to extract the external contour of the bedding, and the information and number of the external contours of the bedding in the image are obtained. ⑤ The thickness of each bedding is calculated by image profile analysis based on Python. The thickness of the bedding used is reduced to 1 pixel using the Zhang-Suen algorithm to obtain the center line and centroid of each bedding. Taking the centroid as the benchmark, 5 measurement points are selected on the center line of the bedding with equal interval thickness, and the top and bottom white pixels of the 5 measurement points are selected by index. The distance between them is calculated to obtain the thickness of the bedding at that position. The average value of the thickness of the 5 measurement points of the bedding is calculated and set as the thickness of the bedding. ⑥ The density of the bedding is calculated according to formula (6), and the proportion of the bedding is calculated according to formula (7). The bedding ratio and bedding density are compared with the second preset classification standard to obtain the final classification results: when the bedding ratio is <40% and the bedding density is >0.4, it is judged as laminated shale; when the bedding ratio is <40% and 0.2≤bedding density≤0.4, it is judged as layered shale; when the bedding ratio is >40%, it is judged as massive shale. The electrical imaging images are calculated and analyzed based on the Swin Transformer deep learning model. The final calculation and judgment results of three types of shale bedding structures are selected, such as Figure 3 、 Figure 4 、 Figure 5 As shown. Figure 3 The results of Swin Transformer identification of laminar shale bedding geometry data are shown in Table 2. There are 40 beddings in the 50cm thick rock body, with a maximum bedding thickness of 0.47cm, a minimum thickness of 0.23cm, and an average thickness of 0.37cm; the density of bedding is 0.80, and the proportion of bedding is 30.13%; therefore, it is determined to be laminar shale. Figure 4 The results of Swin Transformer identification of layered shale bedding geometry data are shown in Table 3. There are 17 beddings in the 50cm thick rock body, with a maximum bedding thickness of 1.52cm, a minimum thickness of 0.67cm, and an average thickness of 0.92cm. The density of the bedding is 0.34, and the proportion of the bedding is 31.16%. Therefore, it is determined to be layered shale. Figure 5 As shown in Table 4, the results of Swin Transformer identification of massive shale bedding geometry data show that there are 7 beddings in the 50cm thick lithology body, with a maximum bedding thickness of 16.36cm, a minimum thickness of 0.76cm, and an average thickness of 4.32cm; the density of the bedding is 0.14, and the proportion of bedding is 60.45%; therefore, it is determined to be massive shale. ⑦ When using a simple bedding identification model to determine that the electrical imaging logging image is laminated or layered shale, the image needs to be re-imported into the complex bedding identification network model to re-refine the bedding identification and the judgment of shale bedding structure. ⑧ As Figure 1 As shown in the figure, the electrical imaging logging map was analyzed and calculated using the Swin Transformer deep learning network model to classify the types of shale bedding structures at different depths. Among them, 12 areas are laminated shale, 9 areas are layered shale, and 3 areas are massive shale.

[0108] Table 2

[0109]

[0110] Table 3

[0111]

[0112]

[0113] Table 4

[0114]

[0115] The present invention first calculates the reservoir anisotropy through conventional logging curves, analyzes the fractal characteristics of the curves, and constructs the classification characteristic parameters of shale bedding structure through these two parameters, and preliminarily evaluates and divides the types of shale bedding structures at different depths. On this basis, according to the preliminary classification characteristic parameters, the electrical imaging logging map is cut into pictures of different sizes according to the type of shale bedding structure in the form of a sliding window. Based on the Swin Transformer artificial intelligence technology, different bedding network recognition models are designed for complex shale bedding structures (laminated, layered) and simple massive shale bedding structures, so as to achieve efficient and accurate identification of the shape, number and thickness of bedding in shale. Finally, based on the discrimination method of the dual parameters of bedding density and proportion, the accurate division of shale bedding types is achieved.

[0116] Based on traditional well logging curves, a new method for calculating characteristic parameters that characterize shale bedding structure types was proposed. Based on these characteristic parameters, a standard for segmenting electrical imaging logs was established. For different types of shale bedding structures, a new method for identifying and obtaining shale bedding geometric parameters using different Swin Transformer artificial intelligence network models was proposed. The type of shale bedding structure was then determined using bedding density and proportion. This method further improved the accuracy and speed of shale bedding structure classification, providing theoretical and technical support for identifying "sweet spots" in shale reservoirs.

[0117] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for accurate and efficient identification and classification of shale bedding structure based on deep learning, characterized by: include: Calculate shale bedding structure pre-classification parameters based on the fast shear wave time difference curve, slow shear wave time difference curve and the logging curve related to the shale bedding structure in the target block logging data; The anisotropy coefficient of shale reservoir at different depths is calculated as: The box counting dimension of the logging curve at different depths of the reservoir is calculated as: The shale bedding structure pre-classification parameters are obtained as follows: Where ANI is the anisotropy coefficient of shear wave time difference; s1 is the fast shear wave time difference; s2 is the slow shear wave time difference; D f is the box counting dimension; r is the length of the box; N(r) is the minimum number of boxes covering the well logging curve, Pre-classification parameters for shale bedding structure; Obtaining a segmented electrical imaging logging map and a preliminary classification result based on the shale bedding structure pre-classification parameters; According to the preliminary classification results, the segmented electrical imaging logging map is input into different bedding recognition network models to obtain shale bedding structure recognition results, and the recognition results are analyzed using image morphology to obtain the final classification results, wherein the bedding recognition network model is constructed based on the Swin Transformer network model and obtained through training based on a training set, and the training set includes electrical imaging logging maps annotated with different types of bedding structures.

2. The method for accurately and efficiently identifying and classifying shale bedding structure based on deep learning according to claim 1 is characterized in that: Calculation of shale bedding structure pre-classification parameters include: Calculating anisotropy coefficients at different depths of the shale reservoir based on the fast shear wave time difference curve and the slow shear wave time difference curve; According to the logging curve correlated with the shale bedding structure, the box counting dimension of the logging curve at different depths of the reservoir is calculated, wherein the logging curve correlated with the shale bedding structure is an acoustic time difference curve, a natural gamma curve or a neutron curve.

3. The method for accurate and efficient identification and classification of shale bedding structure based on deep learning according to claim 1 is characterized in that: Obtaining segmented electrical imaging logs and preliminary classification results includes: Normalizing the shale bedding structure pre-classification parameters to obtain normalized pre-classification parameters; Comparing the normalized pre-classification parameter with a first preset classification standard to obtain the preliminary classification result; According to the shale bedding structure pre-classification parameters, the electrical imaging logging map is cut in a sliding window manner to obtain the segmented electrical imaging logging map.

4. The method for accurately and efficiently identifying and classifying shale bedding structure based on deep learning according to claim 1 is characterized in that: The bedding identification network model includes: a complex bedding identification network model and a simple bedding identification network model; The complex bedding recognition network model is constructed based on the Swin Transformer network model and is used to identify laminar and layered shale bedding structures; The simple bedding identification network model is constructed based on the Swin Transformer network model and is used to identify the bedding structure of massive shale. Among them, the Swin Transformer network model uses Swin-L as the benchmark network. The Swin-L network model for constructing the complex bedding recognition network model selects high feature dimension and number of attention heads, and sets the size of the attention layer window according to the hybrid strategy. The Swin-L network model for constructing the simple bedding recognition network model selects low feature dimension and number of attention heads, and the window for attention layer processing is set to a medium window.

5. The method for accurately and efficiently identifying and classifying shale bedding structure based on deep learning according to claim 1 is characterized in that: The layer recognition network model adopts a total loss function including cross entropy loss and IOU loss; The total loss function is: Where y is the true label; is the predicted output of the model; is the true label of the i-th sample; is the predicted probability of the i-th sample; N is the number of samples; A is the true area; B is the predicted area; α is the weight of the cross loss function; β is the weight of the IoU loss function.

6. The method for accurate and efficient identification and classification of shale bedding structure based on deep learning according to claim 1 is characterized in that: Before obtaining the segmented electrical imaging log map, the following steps are involved: Filling the blank strips in the electrical imaging logging image, performing threshold segmentation, and outputting a binary image; The binary image is filtered using a Sobel operator to obtain the pre-processed electrical imaging logging image.

7. The method for accurate and efficient identification and classification of shale bedding structure based on deep learning according to claim 1 is characterized in that: Image morphology is used to analyze the recognition results and obtain the final classification results, including: Image morphology is used to detect the edges of the identified beddings, calculate the thickness of each bedding and count the number; Calculate the bedding ratio and bedding density based on the thickness and number of each bedding, wherein the bedding ratio is the ratio of the total thickness of the bedding in the lithologic body area to the thickness of the lithologic body, and the bedding density is the ratio of the number of beddings in the lithologic body area to the thickness of the lithologic body; The bedding ratio and bedding density are compared with a second preset classification standard to obtain the final classification result.

Images