Multimodal deep learning-based coupling signal recognition method, system, device and medium

Abstract

The application discloses a kind of based on multi-modal deep learning's coupling signal identification method, system, device and medium, comprising: acquisition CCL induced electromotive force signal data;And CCL induced electromotive force signal data is preprocessed;The CCL induced electromotive force signal data after pre-processing is carried out time domain feature analysis and waveform feature analysis, obtains spatial domain feature vector;CCL induced electromotive force signal data is Fourier transformed to the preprocessed, and the frequency domain feature vector of CCL induced electromotive force data is obtained;Based on MCB algorithm, space domain feature vector and frequency domain feature vector are fused;Based on CNN network and LSTM network, the feature vector after fusion is identified, obtains coupling signal, realizes the automatic positioning of coupling signal.The application can improve the efficiency of signal identification positioning, reduce the labor cost, and can guarantee the accuracy of signal identification positioning.

Description

Technical Field

[0001] This invention belongs to the field of neural network technology and relates to a method, system, device and medium for identifying coupling signals based on multimodal deep learning. Background Technology

[0002] In oil extraction, to facilitate wellbore insertion, when a cable bridge plug is pressurized and set at a predetermined position within the casing, a magnetic locator is typically used to obtain the casing coupling signal to determine the plug's position within the well. During horizontal well pumping perforation, a magnetic locator (CCL) measures the casing coupling signal to reflect the wellbore's working condition and the operational status of the downhole tools (perforating gun). The principle behind the CCL's casing coupling signal detection lies in its internal frame with two opposing magnetic poles and a coil wound around them. As it moves within the casing, it cuts magnetic field lines, generating a changing millivolt-level induced electromotive force at the ends of the coil. By observing this changing induced electromotive force, the casing coupling position can be determined.

[0003] However, in actual oil and gas well logging processes, the casing of oil and gas wells is subject to formation forces and long-term chemical effects during long-term use, which may cause various damages and imperfections such as wear, deformation, misalignment, and corrosion in certain parts. As a result, when the magnetic positioning instrument moves and measures in the casing, it often measures various useless high and low frequency signals as well as interference signals such as noise. This causes the casing coupling curve to fluctuate and become rough, resulting in the normal coupling signal of the oil and gas well casing being buried in it. Consequently, the casing coupling position is difficult to identify, and the actual position of the coupling cannot be directly identified through the CCL signal. The casing condition detection effect is unstable, and the correct position of the cable bridge plug cannot be determined. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the prior art where magnetic positioners cannot directly identify the actual position of the coupling through CCL signals, the bushing status detection effect is unstable, and the correct position of the cable bridge plug cannot be determined. This invention provides a coupling signal identification method, system, device, and medium based on multimodal deep learning.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A method for identifying coupling signals based on multimodal deep learning includes:

[0007] Acquire CCL induced electromotive force signal data; and preprocess the CCL induced electromotive force signal data;

[0008] The preprocessed CCL induced electromotive force signal data is subjected to time-domain feature analysis and waveform feature analysis to obtain spatial domain feature vectors.

[0009] Fourier transform is performed on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data.

[0010] Based on the MCB algorithm, spatial domain feature vectors and frequency domain feature vectors are fused;

[0011] Based on CNN and LSTM networks, the fused feature vectors are identified to obtain the coupling signal, thereby achieving automatic positioning of the coupling signal.

[0012] A further improvement of the present invention is that:

[0013] Acquire CCL induced electromotive force signal data; and preprocess the CCL induced electromotive force signal data; specifically:

[0014] CCL induced electromotive force signal data includes CCL signal data of the magnetization section and CCL signal data of the coupling section;

[0015] The CCL induced electromotive force signal data is stored in an NDB file; the NDB file containing the CCL induced electromotive force signal data is processed to extract the data information, and the data information is converted into a CSV file; the dataset includes number, time, depth, voltage, velocity and tension.

[0016] The preprocessed CCL induced electromotive force signal data is subjected to time-domain feature analysis and waveform feature analysis to obtain spatial domain feature vectors, specifically:

[0017] The CCL signal data of the coupling segment presents a "peak-valley-peak" pattern, where the valley represents the midpoint of the coupling; the CCL signal data of the coupling segment is obtained based on the feature extraction method of sliding window.

[0018] Randomly sample any segment of the CCL signal S = [S1, S2, ..., S...] n ], for any sub-segment of the acquired CCL signal [S i S i+1 ,……,S k If the following conditions are met:

[0019]

[0020] Then select [S] i S i+1 ,……,S k ], S (i+k) / 2 This is the trough of the region; if the above conditions are not met, a new window is formed to the right to continue the judgment; for all selected sub-signal segments, there is a trough in the interval, so it is the CCL signal data of the coupling segment;

[0021] Where i+1<k,i> 0, k≤n, S (i+k) / 2 The center point of the sub-signal segment;

[0022] Feature vectors are extracted from the extracted sub-signal segments to obtain spatial domain feature vectors;

[0023] Perform a Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data; specifically:

[0024] The formula for continuous Fourier transform is as follows:

[0025]

[0026] The formula for the Discrete Fourier Transform is as follows:

[0027]

[0028] Based on the MCB algorithm, spatial domain feature vectors and frequency domain feature vectors are fused; specifically:

[0029] Based on the CountSketch mapping function, obtain the spatial domain feature CountSketch mapped from the spatial domain feature vector;

[0030] Based on the CountSketch mapping function, obtain the frequency domain feature CountSketch mapped from the frequency domain feature vector;

[0031] The spatial domain feature CountSketch and the frequency domain feature CountSketch are subjected to frequency domain fast Fourier transform and inverse FFT to obtain the fused feature vector.

[0032] The Count Sketch mapping function is as follows:

[0033] E[<Ψ(x,h,s),Ψ(y,h,s)>]=<s,y> (2)

[0034]

[0035] Formulas (2) and (3) represent the outer product of two vectors as a convolution of each count map.

[0036] A coupling signal recognition system based on multimodal deep learning includes:

[0037] The preprocessing module acquires CCL induced electromotive force signal data and preprocesses the CCL induced electromotive force signal data.

[0038] The first acquisition module performs time-domain feature analysis and waveform feature analysis on the preprocessed CCL induced electromotive force signal data to obtain spatial domain feature vectors.

[0039] The second acquisition module performs a Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data.

[0040] The fusion module, based on the MCB algorithm, fuses spatial domain feature vectors and frequency domain feature vectors.

[0041] The recognition module, based on CNN and LSTM networks, identifies the fused feature vector to obtain the coupling signal and achieve automatic positioning of the coupling signal.

[0042] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0043] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This invention utilizes a multimodal feature fusion network and a deep neural network to implement a method for analyzing CCL induced electromotive force data and automatically identifying and locating coupling signals in horizontal wells. Existing CCL induced electromotive force data is used as the training dataset for the deep neural network. Features are extracted from different domains within the dataset, and after final data feature processing using a multimodal fusion algorithm, a CNN+LSTM network structure is used for coupling signal identification, achieving automatic signal localization. This invention improves the efficiency of signal identification and localization, reduces labor costs, and ensures the accuracy of signal identification and localization. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of the coupling signal recognition method based on multimodal deep learning of the present invention;

[0048] Figure 2 This is a structural diagram of the coupling signal recognition system based on multimodal deep learning according to the present invention;

[0049] Figure 3 Flowchart for automatic positioning of overall coupling signals;

[0050] Figure 4 This is a flowchart of the feature extraction process;

[0051] Figure 5 This is a waveform characteristic diagram of the coupling;

[0052] Figure 6 The flowchart of the multimodal fusion algorithm MCB;

[0053] Figure 7 This is a diagram of the CNN+LSTM network structure. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0055] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0056] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0057] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0058] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0059] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0060] The present invention will now be described in further detail with reference to the accompanying drawings:

[0061] See Figure 1 This invention discloses a method for identifying coupling signals based on multimodal deep learning, comprising:

[0062] S101, acquire CCL induced electromotive force signal data; and preprocess the CCL induced electromotive force signal data.

[0063] The CCL induced electromotive force signal data includes CCL signal data of the magnetization section and CCL signal data of the coupling section; the CCL induced electromotive force signal data is stored in an NDB file; the NDB file containing the CCL induced electromotive force signal data is processed to extract the data information, and the data information is converted into a CSV file; the dataset includes number, time, depth, voltage, velocity and tension.

[0064] S102, perform time-domain feature analysis and waveform feature analysis on the preprocessed CCL induced electromotive force signal data to obtain spatial domain feature vectors.

[0065] The CCL signal data of the coupling segment exhibits a "peak-trough-peak" pattern, where the trough represents the midpoint of the coupling; the CCL signal data of the coupling segment is obtained based on the feature extraction method of sliding window.

[0066] Randomly sample any segment of the CCL signal S = [S1, S2, ..., S...] n For any sub-segment of the acquired CCL signal [S] i S i+1 ,……,S k If the following conditions are met:

[0067]

[0068] Then select [S] i S i+1 ,……,S k ], S (i+k) / 2 This is the trough of the region; if the above conditions are not met, a new window is formed to the right to continue the judgment; for all selected sub-signal segments, there is a trough in the interval, so it is the CCL signal data of the coupling segment;

[0069] Where i+1<k,i> 0, k≤n, S (i+k) / 2 The center point of the sub-signal segment;

[0070] Feature vectors are extracted from the extracted sub-signal segments to obtain spatial domain feature vectors.

[0071] S103, perform Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data.

[0072] The formula for continuous Fourier transform is as follows:

[0073]

[0074] The formula for the Discrete Fourier Transform is as follows:

[0075]

[0076] S104, based on the MCB algorithm, fuses spatial domain feature vectors and frequency domain feature vectors.

[0077] Based on the Count Sketch mapping function, obtain the spatial domain feature CountSketch mapped from the spatial domain feature vector; based on the Count Sketch mapping function, obtain the frequency domain feature CountSketch mapped from the frequency domain feature vector; perform frequency domain fast Fourier transform and inverse FFT on the spatial domain feature CountSketch and the frequency domain feature CountSketch to obtain the fused feature vector.

[0078] The Count Sketch mapping function is as follows:

[0079] E[<Ψ(x,h,s),Ψ(y,h,s)>]=<s,y> (2)

[0080]

[0081] Formulas (2) and (3) represent the outer product of two vectors as a convolution of each count map.

[0082] S105, based on CNN and LSTM networks, identifies the fused feature vectors to obtain the coupling signal and achieve automatic positioning of the coupling signal.

[0083] See Figure 2 This invention discloses a coupling signal recognition system based on multimodal deep learning, comprising:

[0084] The preprocessing module acquires CCL induced electromotive force signal data and preprocesses the CCL induced electromotive force signal data.

[0085] The first acquisition module performs time-domain feature analysis and waveform feature analysis on the preprocessed CCL induced electromotive force signal data to obtain spatial domain feature vectors.

[0086] The second acquisition module performs a Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data.

[0087] The fusion module, based on the MCB algorithm, fuses spatial domain feature vectors and frequency domain feature vectors.

[0088] The recognition module, based on CNN and LSTM networks, identifies the fused feature vector to obtain the coupling signal and achieve automatic positioning of the coupling signal.

[0089] The present invention proposes a coupling signal recognition method based on multimodal deep learning. The method mainly includes two parts: the first is the feature extraction part of the data, and the second is the coupling signal recognition and localization part of the deep neural network.

[0090] The method includes the following steps:

[0091] Step 1: The experimental data for this invention comes from real-time cluster perforation pumping operation data collected in the field from an oil and gas well in Southwest China, in NDB file format. This type of data file mainly consists of two parts: one part is the CCL signal data of the magnetization section, and the other part is the CCL signal data of the coupling section. To standardize the format and facilitate subsequent operations, the exported NDB file is processed using software to extract the valid data information and convert the data information into a CSV file. The original dataset of real-time cluster perforation pumping operation data includes six attributes: number, time, depth, voltage, velocity, and tension.

[0092] Step 2: Extract the time-domain and waveform features from the processed clustered perforation pumping real-time operation data. The main feature of the CCL signal data of the coupling section is a "peak-trough-peak" pattern, where the trough represents the midpoint of the coupling. To accurately describe the signal characteristics, a feature extraction method based on a sliding window is adopted, which not only solves the problem of signal drift from the sensor, but also accurately obtains the signal trough features, facilitating coupling identification and positioning.

[0093] First, set the initial sliding window width to L. This will allow you to obtain all discrete waveform points within a depth of L within the sliding window. To more accurately distinguish between the magnetization segment and the coupling segment, the window value L should be slightly larger than the length of the characteristic signal segment. Based on well data, it was statistically found that over 75% of the characteristic signal segments are less than 0.8m in length. Therefore, the initial sliding window width is set to 0.8m; that is, L = 0.8m. Furthermore, it is ensured that the minimum voltage range is at the midpoint of the signal segment. The specific calculation process is as follows:

[0094] Randomly sample any segment of the CCL signal S = [S1, S2, ..., S...] n ], for any sub-segment of the acquired CCL signal [S i S i+1 ,……,S k If the following conditions are met:

[0095]

[0096] Then select [S] i S i+1 ,……,S k ], S (i+k) / 2 This is the trough of the region; if the above conditions are not met, a new window is formed to the right to continue the judgment; for all selected sub-signal segments, there is an interval trough, so they are considered possible coupling segments. Where i+1<k,i> 0, k≤n, S (i+k) / 2 The center point of the sub-signal segment;

[0097] Feature vectors are extracted from the extracted sub-signal segments to obtain spatial domain feature vectors; temporal features and shape features are selected for classification feature extraction.

[0098] Temporal feature extraction involves extracting characteristic parameters from the signal's time series to describe its characteristics. The mean and root mean square (RMS) are chosen for feature description. The mean is the average value of the signal and can be used to describe its amplitude and intensity. The RMS, also known as the effective value, reflects the strength of signal fluctuations.

[0099] Waveform feature extraction involves extracting various characteristic information from signal waveforms, including skewness, kurtosis, waveform factor, spurious-to-peak (SOP) factor, impulse factor, and margin factor. Skewness and kurtosis both represent the distribution pattern of a signal. Skewness describes the asymmetry of the signal distribution. Kurtosis indicates the smoothness of the signal waveform and reflects the amplitude distribution of signal segments. The waveform factor, the ratio of the root mean square (RMS) value to the rectified average value, describes the fullness of the signal waveform. The SOP factor, the ratio of the signal peak value to the RMS value, represents the extreme degree of the peak value in the waveform and can be used to describe the amplitude variation of the signal. The impulse factor, the ratio of the signal peak value to the rectified average value (average of absolute values), and the margin factor, the ratio of the signal peak value to the RMS amplitude, both reflect the characteristics of signal abrupt changes.

[0100] After the above data feature extraction, the clustered perforation pumping real-time operation dataset yielded a spatial domain feature vector, which includes 10 attributes: minimum voltage of the signal segment, average signal value, root mean square signal value, signal skewness, signal kurtosis, signal waveform factor, signal peak factor, signal impulse factor, and signal margin factor. These attributes can reflect the characteristics of the signal segment.

[0101] Step 3: Obtain the frequency domain feature vector of the CCL induced electromotive force data from the real-time operation dataset of clustered perforation pumps using Fourier transform.

[0102] The formula for continuous Fourier transform is as follows:

[0103]

[0104] The formula for the Discrete Fourier Transform is as follows:

[0105]

[0106] Step four: Having obtained the spatial domain feature vector through step two, including feature factors such as skewness, kurtosis, waveform factor, peak factor, impulse factor, and margin factor, and the frequency domain feature vector obtained in step three, we need to fuse the feature vectors of the two different modes. Feature fusion involves taking the feature vectors of the two modes as input and outputting the fused vector. The most commonly used methods are concatenation, element-wise product, and element-wise sum. This invention uses the Multimodal Compact Bilinear (MCB) method. MCB maps the result of the outer product to a low-dimensional space and does not require explicit calculation of the outer product, cleverly avoiding the problem of high computational complexity of the outer product.

[0107] Specific details of the MCB algorithm implementation:

[0108] Bilinear pooling is the calculation of the outer product of vectors. It pools the features obtained after bilinear fusion. First, bilinear pooling calculates the outer product of the feature vectors at each location in the feature map obtained from convolution. Then, it performs sum pooling on the results of all outer products to obtain the feature vector x. x is then processed through signed square root and L2 normalization to obtain the final feature vector.

[0109] The operation of bilinear pooling can be expressed by the following formula:

[0110]

[0111] In the case of a linear kernel, because:

[0112]

[0113] Only then:

[0114]

[0115] The mapping function Φ, which approximates the polynomial kernel in a low dimension, can be used to compress bilinear pooling.

[0116] MCB is suitable for fusing features from different modalities. First, the feature vectors of the two modalities are mapped using the CountSketch function to obtain the feature count sketch. Then, the fused features are obtained through frequency domain Fast Fourier Transform and Inverse FFT.

[0117] Step 5: A complex model based on Convolutional Neural Networks (CNN) and Long Short-Term Memory Neural Networks (LSTM) is used. This model adds consideration of temporal features to the CNN network to improve recognition performance. The network consists of three parts: a local spatiotemporal network, a gating mechanism, and an attention transfer mechanism. The local spatiotemporal network uses CNN to handle local spatial dependencies in the data, the original LSTM to handle temporal dependencies, the gating mechanism to capture dynamic dependencies in the hierarchical structure, and the attention mechanism to improve network accuracy.

[0118] Step 6: Obtain the automatic identification and positioning results of the coupling signal based on CCL induced electromotive force data analysis.

[0119] See Figures 3-7 The present invention provides a method for identifying coupling signals based on multimodal deep learning, comprising the following steps:

[0120] Figure 3This is a flowchart illustrating the overall automatic positioning of coupling signals according to the present invention. The flowchart provides an overview of the research process. First, the software converts real-time operational data from clustered perforation pumping into a processable data file. Next, the data undergoes preprocessing, including data cleaning and feature engineering. Then, spatial and frequency domain feature vectors are extracted, and the MCB multimodal method is used to fuse the data features. Finally, the obtained feature parameter vectors are input into a deep neural network for training, ultimately yielding a coupling signal identification and positioning scheme.

[0121] Figure 4 This is a flowchart of the feature extraction process for this invention. Processing the raw data reveals that the original dataset includes six attributes: number, time, depth, voltage, velocity, and tension. Feature engineering transforms the raw data into features, and feature selection allows features to better describe these data and improve model performance. Magnetic locators can only acquire real-time information such as induced electromotive force, velocity, and depth within the well. The dataset is characterized by a large volume of data but very few features. Therefore, feature selection and construction are considered for the raw data. Thus, multimodal processing is used to transform the data into spatial and frequency domain feature vectors through corresponding features, and a multimodal fusion algorithm is used to fuse these features for subsequent input into the neural network. The overall operation steps are as follows:

[0122] Step 1: Process the raw cluster perforation pumping real-time operation data, that is, preprocess the data, including data cleaning, feature engineering, etc., to remove complicated and useless data.

[0123] Step 2: Using the sliding window feature extraction method, and based on time-domain feature analysis and waveform feature analysis, extract the spatial domain feature vector from the CCL signal data.

[0124] Step 3: Use Fourier transform to extract the frequency domain feature vector from the CCL signal data.

[0125] Step 4: Since the above two steps yield two types of feature vectors with different modalities, based on the multimodal concept, the feature vectors of the above different types are fused. This invention uses the MCB algorithm to complete the feature vector fusion.

[0126] Step 5: Complete the extraction of data feature parameter vectors.

[0127] Figure 5The waveform characteristics of the couplings are shown. Each coupling generates an induced electromotive force change in a shape resembling "M". The waveforms show that the amplitude and fluctuation of the signal in the coupling segment are greater than those in the non-coupling segment. The waveforms of the noise segment and the coupling segment are similar, with more spikes in the noise segment, but they are not easily distinguishable visually, making it difficult to judge using only voltage as a characteristic parameter. To address the difficulty of characterizing a signal using a single characteristic parameter, this study proposes to construct characteristic parameters for the selected signal segment from both time-domain and waveform perspectives using signal time-domain analysis methods.

[0128] Figure 6 This is a flowchart of the Multimodal Feature Fusion (MCB) algorithm. Current multimodal tasks require fusing features from two modalities. Feature fusion involves taking the feature vectors of two modalities as input and outputting a fused vector. This invention uses the MCB method for multimodal feature fusion. The process generally involves the following steps:

[0129] Step 1: Obtain the feature count map (CountSketch) from the feature vectors of the spatial domain using the CountSketch mapping function.

[0130] The formula for the Count Sketch mapping function is as follows:

[0131] E[<Ψ(x,h,s),Ψ(y,h,s)>]=<s,y>

[0132]

[0133] The above formula shows that the count map of the outer product of two vectors can be represented as the convolution of each count map.

[0134] Step 2: Similarly, use the Count Sketch mapping function to obtain the feature count map of the frequency domain feature vector.

[0135] Step 3: Both are subjected to frequency domain fast Fourier transform (FFT) and inverse FFT to obtain the final fused features.

[0136] Figure 7 The diagram shows the CNN+LSTM network structure. Module a is an attention transfer mechanism that captures long-term dependencies and temporal shifts in depth. For different well depths, this invention also uses LSTM to capture sequence information. Module b uses an LSTM to capture short-term dependencies. Module c's flow gating mechanism tracks dynamic spatial similarity representations by controlling the propagation of spatial information, where FC represents a fully connected layer and Conv represents multiple convolutional layers. Module d is a local spatiotemporal network using CNN to perform the final convolution operation on the overall hoop signal. The process roughly involves the following steps:

[0137] Step 1: First, the feature data output by the multimodal fusion algorithm is input into module C. In this flow gating mechanism module, the spatial dependencies in the hierarchical structure of the data are captured through local CNN.

[0138] Step 2: In module a, the LSTM network is used to process the dependence of the CCL value changing with depth in the real-time operation data of clustered perforation pumping. Then, the attention mechanism is used to capture the relevant dependencies and signal features to obtain a weighted representation of the previous depth.

[0139] Step 3: By capturing the time series dependencies in the real-time operation data of clustered perforation pumping using Long Short-Term Memory (LSTM), the burst and vanishing gradient problems existing in traditional Recurrent Neural Networks (RNNs) are solved.

[0140] Step 4: Input the previously obtained dependency relationships into the CNN network structure to obtain the final CCL coupling signal recognition and localization results.

[0141] An embodiment of the present invention provides a terminal device. This terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0142] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0143] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0144] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0145] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0146] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0147] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying coupling signals based on multimodal deep learning, characterized in that, include: Acquire CCL induced electromotive force signal data; The CCL induced electromotive force signal data is preprocessed. The preprocessed CCL induced electromotive force signal data is subjected to time-domain feature analysis and waveform feature analysis to obtain spatial domain feature vectors. Fourier transform is performed on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data. Based on the MCB algorithm, spatial domain feature vectors and frequency domain feature vectors are fused; specifically: Based on the Count Sketch mapping function, obtain the spatial domain feature Count Sketch mapped from the spatial domain feature vector; Based on the Count Sketch mapping function, obtain the frequency domain feature Count Sketch mapped from the frequency domain feature vector; Perform frequency domain fast Fourier transform and inverse FFT on the spatial domain feature count sketch and the frequency domain feature count sketch to obtain the fused feature vector; The Count Sketch mapping function is specifically as follows: Formulas (2) and (3) represent the outer product of two vectors as a convolution of each count map; Based on CNN and LSTM networks, the fused feature vectors are identified to obtain the coupling signal, thereby achieving automatic positioning of the coupling signal.

2. The coupling signal recognition method based on multimodal deep learning according to claim 1, characterized in that, The acquisition of CCL induced electromotive force signal data and the preprocessing of the CCL induced electromotive force signal data are specifically as follows: The CCL induced electromotive force signal data includes magnetized section CCL signal data and coupling section CCL signal data; The CCL induced electromotive force signal data is stored in an NDB file; the NDB file containing the CCL induced electromotive force signal data is processed to extract the data information, and the data information is converted into a CSV file; the dataset includes number, time, depth, voltage, velocity and tension.

3. The coupling signal recognition method based on multimodal deep learning according to claim 2, characterized in that, The step of performing time-domain feature analysis and waveform feature analysis on the preprocessed CCL induced electromotive force signal data to obtain spatial domain feature vectors is as follows: The CCL signal data of the coupling segment presents a "peak-valley-peak" pattern, where the valley represents the midpoint of the coupling; the CCL signal data of the coupling segment is obtained based on the feature extraction method of sliding window. Randomly sample any segment of the CCL signal. For any sub-segment of the acquired CCL signal If the following conditions are met: Then choose , This is the trough of the region; if the above conditions are not met, a new window is formed to the right to continue the judgment; for all selected sub-signal segments, there is a trough in the interval, so it is the CCL signal data of the coupling segment; in, , , , The center point of the sub-signal segment; Feature vectors are extracted from the extracted sub-signal segments to obtain spatial domain feature vectors.

4. The coupling signal recognition method based on multimodal deep learning according to claim 1, characterized in that, The process involves performing a Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data; specifically: The formula for continuous Fourier transform is as follows: The formula for the Discrete Fourier Transform is as follows: (k=0,1,2,…,N-1)。 5. A coupling signal recognition system based on multimodal deep learning, characterized in that, include: The preprocessing module acquires CCL induced electromotive force signal data; The CCL induced electromotive force signal data is preprocessed. The first acquisition module performs time-domain feature analysis and waveform feature analysis on the preprocessed CCL induced electromotive force signal data to obtain spatial domain feature vectors. The second acquisition module performs a Fourier transform on the preprocessed CCL induced electromotive force signal data to obtain the frequency domain feature vector of the CCL induced electromotive force data. The fusion module, based on the MCB algorithm, fuses spatial domain feature vectors and frequency domain feature vectors. Based on the Count Sketch mapping function, obtain the spatial domain feature Count Sketch mapped from the spatial domain feature vector; Based on the Count Sketch mapping function, obtain the frequency domain feature Count Sketch mapped from the frequency domain feature vector; Perform frequency domain fast Fourier transform and inverse FFT on the spatial domain feature count sketch and the frequency domain feature count sketch to obtain the fused feature vector; The Count Sketch mapping function is specifically as follows: Formulas (2) and (3) represent the outer product of two vectors as a convolution of each count map; The recognition module, based on CNN and LSTM networks, identifies the fused feature vector to obtain the coupling signal and achieve automatic positioning of the coupling signal.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-4.

Images