Well wall crack identification method and device, electronic equipment and medium

Abstract

The invention relates to a well wall crack recognition method and device, electronic equipment and a medium, and belongs to the technical field of oil exploitation. The method comprises the steps that original signals collected by an ultrasonic sensor in the drilling process are obtained; performing continuous wavelet transform on the original signal to obtain a two-dimensional time-frequency image; performing frame segmentation processing on the original signal to obtain a one-dimensional time sequence; inputting the two-dimensional time-frequency image into a RepLKNet network to obtain a first high-dimensional feature corresponding to the underground target crack in a time-frequency domain; inputting the one-dimensional time sequence into a BiGRU network based on GlobalAttention optimization, and obtaining a second high-dimensional feature corresponding to a target crack in a time sequence domain; splicing and fusing the first high-dimensional feature and the second high-dimensional feature to obtain a fused feature; and inputting the fusion feature into the full connection layer to obtain an identification result of the target crack. According to the invention, the oil-gas exploration well wall tiny crack identification precision is improved.

Description

Technical Field

[0001] This invention relates to the field of oil extraction technology, and in particular to a method, device, electronic equipment, and medium for identifying wellbore fractures. Background Technology

[0002] As oil and gas exploration advances into deeper and more complex reservoirs (such as tight sandstone and shale gas reservoirs), wellbore fractures are a key control factor for reservoir permeability. Among them, micro-fractures (fracture width < 0.1 mm) are important channels for oil and gas seepage and may also induce well collapse and well leakage. However, due to their small size and weak signal difference, traditional logging-while-drilling technology is unable to identify them with high precision, which has become a core bottleneck restricting the development of complex reservoirs.

[0003] The current mainstream drilling wall fracture identification technologies all have significant limitations in the identification of micro fractures: (1) Sonic logging while drilling: relies on the propagation characteristics of elastic waves to invert fractures, but micro fractures are extremely weak to elastic wave disturbances and are affected by formation heterogeneity and drilling vibration, resulting in low sensitivity and easy omission or misjudgment; (2) Resistivity logging while drilling: based on the resistivity difference between fracture fluid and matrix, but the fluid volume in micro fractures is small, and the change in resistivity is lower than the instrument detection limit, and is basically ineffective due to the influence of formation fluid mineralization and wellbore enlargement; (3) Traditional Ultrasonic logging while drilling: Although theoretically it has a high resolution (millimeter level), it has core defects: ① Although low frequency (≤2MHz) has a wide coverage, the resolution is only 0.3~0.5mm, which cannot identify fractures <0.1mm. Increasing the frequency results in large signal attenuation and low signal-to-noise ratio; ② It relies on traditional algorithms such as threshold segmentation and edge detection, which can only extract strong edges and easily filter weak edge signals of small fractures or misjudge noise; ③ It lacks real-time interference compensation for downhole environment such as drilling fluid coupling and drill string vibration. Simple filtering cannot separate effective signals, and the accuracy is further reduced.

[0004] In summary, there is a lack of existing technologies for identifying wellbore fractures to improve the accuracy of identifying minute fractures in oil and gas exploration wellbores. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device, electronic equipment and medium for identifying wellbore cracks, so as to improve the accuracy of identifying micro-cracks in oil and gas exploration wellbore.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for identifying wellbore fractures, comprising: Acquire the raw signals collected by the ultrasonic sensor during the drilling process; The original signal is subjected to continuous wavelet transform to obtain a two-dimensional time-frequency image; The original signal is segmented into frames to obtain a one-dimensional time sequence. The two-dimensional time-frequency image is input into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain; By inputting a one-dimensional time series into a BiGRU network optimized based on GlobalAttention, the second high-dimensional feature corresponding to the target crack in the time domain is obtained. The first high-dimensional feature and the second high-dimensional feature are concatenated and fused to obtain the fused feature; The fused features are input into the fully connected layer to obtain the identification results of the target crack.

[0007] In one possible implementation, the RepLKNet network uses a large 31x31 convolutional kernel.

[0008] In one possible implementation, the formula for the continuous wavelet transform is:

[0009]

[0010] In the formula, Represents continuous wavelet transform. This represents the translation factor, used to control the position of the wavelet window in the time domain. This represents the scaling factor, which controls the size and position of the wavelet window in the frequency domain. Describe the wavelet basis functions. Represents the original signal function. Indicates time.

[0011] In one possible implementation, the BiGRU network based on GlobalAttention optimization includes: The BiGRU network extracts the temporal dependencies of a one-dimensional time series sequence through two independent forward and backward hidden layers, and outputs a concatenated bidirectional hidden state sequence. GlobalAttention performs weighted fusion of the bidirectional hidden state sequence to obtain a context vector that focuses on key information. The second high-dimensional feature is obtained based on the context vector that focuses on key information.

[0012] In one possible implementation, the step of inputting the fused features into the fully connected layer to obtain the target crack identification result includes: The fused features are input into the fully connected layer, and the fully connected layer outputs a class score. The preset loss function determines the identification result of the target crack based on the loss between the category score and the true label.

[0013] In one possible implementation, the width of the target crack is less than a first preset value.

[0014] In one possible implementation, the identification result of the target crack includes: Location, width, and direction of development of the target crack.

[0015] Secondly, the present invention also provides a wellbore crack identification device, comprising: The raw signal acquisition module is used to acquire the raw signals collected by the ultrasonic sensor during the drilling process; The two-dimensional time-frequency image acquisition module is used to perform continuous wavelet transform on the original signal to obtain a two-dimensional time-frequency image; The one-dimensional time sequence acquisition module is used to perform frame segmentation processing on the original signal to obtain a one-dimensional time sequence; The first high-dimensional feature determination module is used to input the two-dimensional time-frequency image into the RepLKNet network to obtain the first high-dimensional features corresponding to the downhole target fracture in the time-frequency domain. The second high-dimensional feature determination module is used to input a one-dimensional time series into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional features corresponding to the target crack in the time series domain. The feature fusion module is used to concatenate and fuse the first high-dimensional feature and the second high-dimensional feature to obtain the fused feature; The crack identification module is used to input the fused features into the fully connected layer to obtain the identification result of the target crack.

[0016] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the wellbore fracture identification method described in any of the above implementations.

[0017] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps of the wellbore fracture identification method described in any of the above implementations.

[0018] The beneficial effects of this invention are as follows: This invention provides a wellbore fracture identification method that acquires the original signal collected by an ultrasonic sensor during drilling. The original signal undergoes continuous wavelet transform to obtain a two-dimensional time-frequency image. The original signal is then segmented into frames to obtain a one-dimensional time-series sequence. The two-dimensional time-frequency image is input into a RepLKNet network to obtain the first high-dimensional feature corresponding to the target fracture in the time-frequency domain. The one-dimensional time-series sequence is input into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional feature corresponding to the target fracture in the time-series domain. Through the collaborative design of dual-path preprocessing and a dual-branch network, multi-dimensional capture of weak signals from small fractures is achieved. The first and second high-dimensional features are concatenated and fused to obtain a fused feature. The fused feature is input into a fully connected layer to obtain the target fracture identification result. This invention, through a multi-modal feature fusion architecture of time-frequency and time-series data, overcomes the bottleneck of existing technologies in identifying weak signals from small fractures, fills the technological gap in the field of small-sized fracture identification, and significantly improves the accuracy of the identification results through interference weight suppression and global feature capture capabilities of specific networks. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an embodiment of a wellbore crack identification method provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S107; Figure 3 This is a schematic flowchart of an embodiment of a well wall crack identification device provided by the present invention; Figure 4 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0022] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0024] This invention provides a method, device, electronic device, and medium for identifying wellbore cracks, which will be described below.

[0025] Figure 1 This is a schematic flowchart of an embodiment of the wellbore crack identification method provided by the present invention, as shown below. Figure 1 As shown, the wellbore fracture identification method includes: S101. Acquire the raw signals collected by the ultrasonic sensor during the drilling process; Directly acquiring the raw signals from the ultrasonic sensors during the drilling process provides a data foundation for subsequent processing. The raw signals contain rich time and frequency domain information, which is the basis for all subsequent analyses.

[0026] S102. Perform continuous wavelet transform on the original signal to obtain a two-dimensional time-frequency image; Converting the original signal into a two-dimensional time-frequency image can simultaneously reflect the local characteristics of the signal in both time and frequency. Wavelet transform is well-suited for non-stationary signals and can effectively capture transient features.

[0027] S103. Perform frame segmentation on the original signal to obtain a one-dimensional time sequence; The original signal is segmented into multiple one-dimensional time series sequences, which facilitates subsequent time series modeling. This segmentation process reduces the data dimensionality while preserving local features, making it suitable for processing long sequence data.

[0028] S104. Input the two-dimensional time-frequency image into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain; RepLKNet extracts high-dimensional features from time-frequency images, effectively capturing the distribution patterns of cracks in the frequency domain. This lightweight network structure is suitable for embedded or edge computing scenarios.

[0029] S105. Input the one-dimensional time sequence into the BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional feature corresponding to the target crack in the time domain. BiGRU (Bidirectional Gated Recurrent Unit) extracts high-dimensional features from one-dimensional time series, enabling it to capture the temporal dependencies of signals. The GlobalAttention mechanism can dynamically adjust the weights at different time steps to highlight key features.

[0030] S106. The first high-dimensional feature and the second high-dimensional feature are concatenated and fused to obtain the fused feature; High-dimensional features from the time-frequency domain and the time-series domain are concatenated to form a fused feature. This multimodal fusion combines information from different perspectives, improving the robustness and accuracy of crack identification.

[0031] S107. Input the fused features into the fully connected layer to obtain the identification results of the target crack.

[0032] The fully connected layer classifies or regresses the fused features, outputting the final crack identification result. This layer maps high-dimensional features to the target space through nonlinear transformation, achieving end-to-end identification.

[0033] Compared with existing technologies, this embodiment provides a wellbore fracture identification method. It acquires the raw signal collected by an ultrasonic sensor during drilling, performs continuous wavelet transform on the raw signal to obtain a two-dimensional time-frequency image, and then performs frame segmentation to obtain a one-dimensional time-series sequence. The two-dimensional time-frequency image is input into a RepLKNet network to obtain the first high-dimensional feature corresponding to the target fracture in the time-frequency domain. The one-dimensional time-series sequence is input into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional feature corresponding to the target fracture in the time-series domain. Through the collaborative design of dual-path preprocessing and dual-branch networks, multi-dimensional capture of weak signals from small fractures is achieved. The first and second high-dimensional features are concatenated and fused to obtain a fused feature. The fused feature is input into a fully connected layer to obtain the target fracture identification result. This invention, through a multi-modal feature fusion architecture of time-frequency and time-series data, overcomes the bottleneck of existing technologies in identifying weak signals from small fractures, fills the technological gap in the field of small-sized fracture identification, and significantly improves the accuracy of the identification results through interference weight suppression and global feature capture capabilities of specific networks.

[0034] It should be noted that the application scenarios of this invention are not only applicable to the identification of cracks in complex downhole environments, but also applicable to scenarios with good lighting, such as land.

[0035] In specific embodiments of this invention, in the specific environment of downhole fracture identification, signals may be interfered with by various noise sources, especially for micro-fractures, where the interference is particularly significant. Considering the computational power and real-time requirements of the downhole environment, this paper proposes a multimodal feature fusion model combining time-frequency images and one-dimensional time-series signals. This model utilizes RepLKNet and BiGRU-GlobalAttention for classification, aiming to achieve fast and accurate identification of micro-fractures and improve the robustness of the algorithm to enhance its resistance to environmental interference, thereby improving the model's generalization performance. The overall framework of the model consists of four core components: input data preprocessing, dual-branch feature extraction, feature fusion, and classification output.

[0036] The first step is the input data preprocessing stage. For the original ultrasonic echo signal detected on the well wall, the model adopts a dual-path processing strategy to take into account both time-frequency and time-series information: on the one hand, the one-dimensional original signal is converted into a 2D time-frequency image through continuous wavelet transform (CWT). With the help of the time-frequency joint representation capability of CWT, the frequency variation law of the fracture signal at different times is preserved; on the other hand, the long time-series original signal is divided into equal-length 1D time-series subframes through framing operation, which not only reduces the computational complexity of long time-series data, but also provides local sequence units for subsequent time-series feature extraction.

[0037] After preprocessing, the model performs feature extraction through two parallel branches: the first is the RepLKNet branch for time-frequency domain feature extraction. This branch targets 2D time-frequency images, first performing initial convolution and downsampling with Stem layers, and then further processing through four cascaded feature extraction stages. RepLKBlock uses large-size convolutional kernels (31×31) to capture cross-dimensional global correlation features of the time-frequency image, ConvFN refines local details with multiple layers of small convolutional kernels, and the Transition layer achieves downsampling and feature abstraction through stride convolution, ultimately outputting time-frequency domain global abstract features. The second branch is the BiGRU+Global Attention branch for time-domain feature extraction. This branch targets one-dimensional time-series subframes, first capturing the forward and backward temporal dependencies of the sequence simultaneously with a bidirectional gated recurrent unit (BiGRU), outputting latent state features containing bidirectional temporal information, and then using Global Attention. The correlation between the features at each time step and the global features is calculated, and weights are assigned to enhance the key temporal features corresponding to the cracks and suppress redundant noise features, ultimately obtaining the weighted key temporal features.

[0038] Finally, through the feature fusion stage, the time-frequency domain global features of the RepLKNet branch and the time-domain key features of the BiGRU+Attention branch are dimensionally concatenated to achieve multimodal feature complementarity of "time-frequency global information - time-domain key information", which greatly improves the comprehensiveness of feature representation. The fused high-dimensional features are input into the fully connected layer, transformed into the classification space through linear mapping, and finally output as the identification result of well wall cracks.

[0039] In a specific embodiment of the present invention, in steps S102 and S103, the raw input data of the drilling ultrasonic test is preprocessed in parallel with dual channels to selectively preserve the weak signal characteristics of micro-cracks: Branch 1 (Time-Frequency Feature Path): The one-dimensional original signal is converted into a two-dimensional time-frequency image using continuous wavelet transform (CWT) to capture the high-frequency and low-frequency variation trends in the signal and fully preserve the weak frequency domain anomaly features corresponding to the tiny cracks. Continuous wavelet transform is a time-frequency analysis method for time-varying, non-stationary signals. This method can convert a one-dimensional original signal into a two-dimensional image representation that simultaneously contains time and frequency domain information. This representation is more suitable for feature extraction and pattern recognition in neural networks. Compared to the short-time Fourier transform with a fixed window function, the continuous wavelet transform's window function can be adaptively adjusted according to the frequency, achieving a good balance between time and frequency resolution when processing non-stationary signals. In some embodiments of this invention, for any signal x(t), the formula for the continuous wavelet transform is:

[0040]

[0041] In the formula, Represents continuous wavelet transform. This represents the translation factor, used to control the position of the wavelet window in the time domain. This represents the scaling factor, which controls the size and position of the wavelet window in the frequency domain. Describe the wavelet basis functions. Represents the original signal function. Indicates time.

[0042] In the analysis of ultrasonic echo signals from wellbore fractures, the selection of wavelet basis functions is a crucial step in continuous wavelet transform, and their waveforms should match the characteristics of the signal being analyzed. Considering the common transient impacts and damped oscillation modes in fracture signals, the Morlet wavelet (cmor100-1) is selected as the basis function for continuous wavelet transform in the model. Its frequency is 100Hz, its bandwidth parameter is 1, and it exhibits waveforms highly similar to the signal characteristics as well as good time-frequency localization properties.

[0043] In the actual algorithm implementation, the key parameters are set as follows: the sampling frequency of the input signal is 2MHz, corresponding to a sampling period of 0.5 microseconds; the scale length totalscal is set to 128, and its scale sequence is obtained through the following formula:

[0044] in Here, c represents the wavelet center frequency, and c is a constant. The coefficient matrix obtained after transformation is processed using absolute value, and then color mapping is used to visualize the amplitude information. The generated time-frequency images are uniformly standardized to 224×224 pixel RGB format, and the coordinate axes and margins are removed to meet the input requirements of subsequent convolutional neural networks. This preprocessing pipeline converts each ultrasonic echo signal into a standardized two-dimensional time-frequency image, providing a standardized input to the deep learning model that simultaneously contains time-frequency features.

[0045] Branch 2 (Temporal Feature Path): The original signal is divided into a one-dimensional time sequence through frame segmentation, which preserves the time dependence characteristics of the signal and reflects the dynamic evolution of micro-fractures during the drilling process.

[0046] In some embodiments of the present invention, the RepLKNet network uses a large 31x31 convolutional kernel.

[0047] In some embodiments of the present invention, the BiGRU network optimized based on GlobalAttention includes: The BiGRU network extracts the temporal dependencies of a one-dimensional time series sequence through two independent forward and backward hidden layers, and outputs a concatenated bidirectional hidden state sequence. GlobalAttention performs weighted fusion of the bidirectional hidden state sequence to obtain a context vector that focuses on key information. The second high-dimensional feature is obtained based on the context vector that focuses on key information.

[0048] In a specific embodiment of the present invention, in step S104, the two-dimensional time-frequency image is input into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain, specifically including: The time-frequency image feature extraction branch (based on the RepLKNet network) takes the time-frequency image as input and extracts weak frequency domain features through a large convolutional kernel network. It sequentially passes through the Stem module (for initial dimensionality reduction and feature encoding of the time-frequency image) and four cascaded Stage modules (Stage 1~Stage 4) for deep feature extraction, ultimately outputting high-dimensional features corresponding to the microcracks in the time-frequency domain. Each Stage module contains three types of components: RepLKBlock: uses large convolutional kernels to capture global and local correlation features of the time-frequency image, adapting to the dispersed distribution of weak signals from microcracks; ConvFN module: enhances the non-linear expressive power of features through a convolutional feedforward network, amplifying the difference between microcracks and background signals; Transition module: combines 1×1 convolution and depthwise separable convolution (DWConv) to achieve feature dimension transformation and channel attention enhancement; finally, it outputs high-dimensional features corresponding to the microcracks in the time-frequency domain.

[0049] In a specific embodiment of the present invention, in step S105, a one-dimensional time-series sequence is input into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional feature corresponding to the target crack in the time-series domain, specifically including: One-dimensional temporal feature extraction branch (based on BiGRU+Global Attention network): Taking a one-dimensional temporal sequence as input, dynamic features are extracted through temporal encoding and attention enhancement. First, a Bidirectional Gated Recurrent Unit (BiGRU) is used to simultaneously extract the forward and backward propagation features of the ultrasonic echo time-domain signal. In the BiGRU, a backward propagation is added to the single-layer GRU network to fully utilize the input information, extracting past and future information using two hidden layers. These two hidden layers are connected to the same output layer. The first-layer GRU calculates the forward propagation information of the data in chronological order, while the second-layer GRU reads the input signal in reverse order and calculates the backward propagation information in reverse chronological order. These two GRU layers do not interfere with each other when calculating their own hidden states. The final output information of the BiGRU network is jointly determined by both layers, and its output formula is:

[0050] in, This is the output of the positive GRU. This is the output of the reverse GRU. It passes through the forward pass hidden layer. and backward transfer hidden layer The hidden state obtained at time step t .

[0051] The Global Attention layer is connected: global dynamic weights are assigned to the hidden state output by BiGRU, which specifically enhances the weights of key temporal features corresponding to micro-fractures and suppresses interference noise such as drilling vibration and mud coupling; finally, the high-dimensional features corresponding to micro-fractures in the temporal domain are output.

[0052] In some embodiments of the present invention, in step S107, such as Figure 2 As shown, the step of inputting the fused features into the fully connected layer to obtain the target crack identification result includes: S201. Input the fused features into the fully connected layer, and the fully connected layer outputs the category score; S202. The preset loss function determines the identification result of the target crack based on the loss between the category score and the true label.

[0053] In some embodiments of the present invention, the width of the target crack is less than a first preset value.

[0054] In some embodiments of the present invention, the identification result of the target crack includes: Location, width, and direction of development of the target crack.

[0055] In a specific embodiment of the present invention, in steps S106 and S107, the feature fusion and recognition output module first performs feature stitching by stitching together the high-dimensional features output from the time-frequency image branch and the one-dimensional time-series branch, fusing complementary information of "frequency domain anomaly + time-series dynamics" to fully cover the multi-dimensional features of the microcracks; finally, the recognition output inputs the stitched fused features into the fully connected layer (FC), and outputs the final results (Outputs) through classification and regression operations, including core parameters such as the location, width, and development direction of the microcracks.

[0056] In summary, this invention, through a multimodal feature fusion architecture combining time-frequency and time-series data, overcomes the bottleneck of existing technologies in identifying weak signals from micro-cracks, filling a technological gap in the field of micro-crack identification. Addressing the interference issues in complex downhole environments, it significantly improves the stability of identification results by leveraging the interference weight suppression and global feature capture capabilities of a specific network, overcoming the shortcomings of existing technologies with high rates of missed and false positives. Furthermore, the invention employs a lightweight network structure, resolving the contradiction between real-time performance and identification accuracy in existing technologies, achieving both real-time processing synchronized with drilling operations and ensuring accurate identification. The dual-branch, multi-dimensional feature extraction method also avoids the information limitations of single-modal features in existing technologies, fully covering the characteristics of micro-cracks and further enhancing the reliability of identification.

[0057] To verify the reliability of this invention in fracture identification, the data measured in the experimental well was processed using the processing flow of the method in this embodiment. It should be noted that this is merely an application example of this invention, and this embodiment is only used to illustrate the invention and not to limit the scope of the invention. Equivalent substitutions and modifications made by those skilled in the art are considered to fall within the scope of protection of the claims. Table 1 shows the efficiency difference between graph neural networks and one-dimensional neural networks in wellbore fracture identification, comparing them with traditional one-dimensional convolutional neural networks (1D-CNN), LSTM (Long Short-Term Memory network), and one-dimensional bidirectional gated recurrent units (BiGRU): 1D-CNN consists of two convolutional layers, two pooling layers, and a flattening layer. Both convolutional layers use 1×3 kernels with 32 and 64 kernels respectively. Each convolutional layer is followed by a 1×2 max-pooling layer. The ReLU activation function is used.

[0058] LSTM: It consists of three LSTM layers and dropout layers, with the number of hidden neurons set to 128, 64 and 32 respectively.

[0059] GRU: This network has a similar overall structure to the BiGRU used in this paper, consisting of an input layer, three bidirectional GRU layers, a batch normalization layer, a Dropout layer, and a fully connected output layer. The bidirectional GRU layers have 128, 64, and 32 hidden units, respectively. The dropout rate of the Dropout layer is 0.3 to prevent model overfitting. This example further illustrates that the method of this invention can be used for fracture identification, providing technical support for effective early warning during logging-while-drilling.

[0060] Table 1: Comparison of Accuracy on Test Sets for Different Algorithms

[0061] To better implement the wellbore fracture identification method in this embodiment of the invention, based on a wellbore fracture identification method, correspondingly, as follows: Figure 3 As shown, this embodiment of the invention also provides a wellbore crack identification device. A wellbore crack identification device 300 includes: The raw signal acquisition module 301 is used to acquire the raw signals collected by the ultrasonic sensor during the drilling process; The two-dimensional time-frequency image acquisition module 302 is used to perform continuous wavelet transform on the original signal to obtain a two-dimensional time-frequency image. The one-dimensional time sequence acquisition module 303 is used to perform frame segmentation processing on the original signal to obtain a one-dimensional time sequence; The first high-dimensional feature determination module 304 is used to input the two-dimensional time-frequency image into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain. The second high-dimensional feature determination module 305 is used to input a one-dimensional time series sequence into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional feature corresponding to the target crack in the time series domain. The feature fusion module 306 is used to concatenate and fuse the first high-dimensional feature and the second high-dimensional feature to obtain the fused feature; The crack recognition module 307 is used to input the fused features into the fully connected layer to obtain the recognition result of the target crack.

[0062] The well wall crack identification device 300 provided in the above embodiments can realize the technical solution described in the above embodiment of the well wall crack identification method. The specific implementation principle of each module or unit can be found in the corresponding content in the above embodiment of the well wall crack identification method, which will not be repeated here.

[0063] like Figure 4 As shown, the present invention also provides an electronic device 400. The electronic device 400 includes a processor 401, a memory 402, and a display 403. Figure 4 Only some components of the electronic device 400 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0064] In some embodiments, processor 401 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 402 or process data, such as a well wall fracture identification method in this invention.

[0065] In some embodiments, processor 401 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 401 may be local or remote. In some embodiments, processor 401 may be implemented on a cloud platform. In some embodiments, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, or any combination thereof.

[0066] In some embodiments, memory 402 may be an internal storage unit of electronic device 400, such as a hard disk or memory of electronic device 400. In other embodiments, memory 402 may also be an external storage device of electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 400.

[0067] Furthermore, the memory 402 may include both internal storage units of the electronic device 400 and external storage devices. The memory 402 is used to store application software and various types of data installed on the electronic device 400.

[0068] In some embodiments, display 403 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 403 is used to display information from electronic device 400 and to display a visual user interface. Components 401-403 of electronic device 400 communicate with each other via a system bus.

[0069] In one embodiment, when processor 401 executes a wellbore fracture identification program in memory 402, the following steps can be performed: Acquire the raw signals collected by the ultrasonic sensor during the drilling process; The original signal is subjected to continuous wavelet transform to obtain a two-dimensional time-frequency image; The original signal is segmented into frames to obtain a one-dimensional time sequence. The two-dimensional time-frequency image is input into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain; By inputting a one-dimensional time series into a BiGRU network optimized based on GlobalAttention, the second high-dimensional feature corresponding to the target crack in the time domain is obtained. The first high-dimensional feature and the second high-dimensional feature are concatenated and fused to obtain the fused feature; The fused features are input into the fully connected layer to obtain the identification results of the target crack.

[0070] It should be understood that when the processor 401 executes a wellbore fracture identification program in the memory 402, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0071] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 400 mentioned. Electronic device 400 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 400 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0072] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0073] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A borehole wall fracture identification method characterized by, include: Acquire the raw signals collected by the ultrasonic sensor during the drilling process; The original signal is subjected to continuous wavelet transform to obtain a two-dimensional time-frequency image; The original signal is segmented into frames to obtain a one-dimensional time sequence. The two-dimensional time-frequency image is input into the RepLKNet network to obtain the first high-dimensional feature corresponding to the downhole target fracture in the time-frequency domain; By inputting a one-dimensional time series into a BiGRU network optimized based on GlobalAttention, the second high-dimensional feature corresponding to the target crack in the time domain is obtained. The first high-dimensional feature and the second high-dimensional feature are concatenated and fused to obtain the fused feature; The fused features are input into the fully connected layer to obtain the identification results of the target crack.

2. The wellbore crack identification method according to claim 1, characterized in that, The RepLKNet network uses a large 31x31 convolutional kernel.

3. The wellbore crack identification method according to claim 1, characterized in that, The formula for the continuous wavelet transform is: In the formula, Represents continuous wavelet transform. This represents the translation factor, used to control the position of the wavelet window in the time domain. This represents the scaling factor, which controls the size and position of the wavelet window in the frequency domain. Describe the wavelet basis functions. Represents the original signal function. Indicates time.

4. The wellbore crack identification method according to claim 1, characterized in that, The BiGRU network optimized based on GlobalAttention includes: The BiGRU network extracts the temporal dependencies of a one-dimensional time series sequence through two independent forward and backward hidden layers, and outputs a concatenated bidirectional hidden state sequence. GlobalAttention performs weighted fusion of the bidirectional hidden state sequence to obtain a context vector that focuses on key information. The second high-dimensional feature is obtained based on the context vector that focuses on key information.

5. The wellbore crack identification method according to claim 1, characterized in that, The step of inputting the fused features into the fully connected layer to obtain the target crack identification result includes: The fused features are input into the fully connected layer, and the fully connected layer outputs a class score. The preset loss function determines the identification result of the target crack based on the loss between the category score and the true label.

6. The wellbore crack identification method according to claim 1, characterized in that, The width of the target crack is less than a first preset value.

7. The wellbore crack identification method according to claim 1, characterized in that, The identification results of the target crack include: Location, width, and direction of development of the target crack.

8. A well wall crack identification device, characterized in that, include: The raw signal acquisition module is used to acquire the raw signals collected by the ultrasonic sensor during the drilling process; The two-dimensional time-frequency image acquisition module is used to perform continuous wavelet transform on the original signal to obtain a two-dimensional time-frequency image; The one-dimensional time sequence acquisition module is used to perform frame segmentation processing on the original signal to obtain a one-dimensional time sequence; The first high-dimensional feature determination module is used to input the two-dimensional time-frequency image into the RepLKNet network to obtain the first high-dimensional features corresponding to the downhole target fracture in the time-frequency domain. The second high-dimensional feature determination module is used to input a one-dimensional time series into a BiGRU network optimized based on GlobalAttention to obtain the second high-dimensional features corresponding to the target crack in the time series domain. The feature fusion module is used to concatenate and fuse the first high-dimensional feature and the second high-dimensional feature to obtain the fused feature; The crack identification module is used to input the fused features into the fully connected layer to obtain the identification result of the target crack.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the wellbore fracture identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the wellbore fracture identification method according to any one of claims 1 to 7.

Images