A control system for oil and gas microwave electric pulse resonance shock coordinated production increase

Abstract

The present invention discloses an oil and gas microwave electric pulse resonance impact collaborative production increase control system, comprising: a vibration response acquisition module, a vibration response input module, used to input a vibration response sequence into a pre-trained resonance frequency extraction model; a main resonance frequency determination module, used to identify the main resonance frequency of the vibration response sequence by the resonance frequency extraction model, and determine the main resonance frequency of the target shale layer; an electric pulse adjustment module, used to adjust the output frequency of an electric pulse excitation device according to the determined main resonance frequency until the output frequency is consistent with the main resonance frequency; the present invention uses the main resonance frequency extracted from the frequency amplitude distribution diagram as a pseudo label, and trains with a multi-layer perceptron model, and automatically learns the main resonance frequency of the formation without manual labeling or external experience input, effectively improving the fitting ability of the nonlinear response pattern of the complex formation and enhancing the accuracy of main resonance frequency identification.

Description

Technical Field

[0001] The present invention relates to an oil and gas production increase control system, in particular to an oil and gas microwave electric pulse resonance shock coordinated production increase control system. Background Art

[0002] Physical stimulation techniques such as electric pulse excitation and microwave heating are often used to effectively develop tight oil and gas reservoirs, such as shale. Electric pulse excitation induces stress waves and electromagnetic interference in the formation by applying high-frequency, high-voltage electric pulses, enhancing the expansion of pre-existing fractures and promoting oil and gas flow. Microwave heating, on the other hand, converts electromagnetic energy into heat, increasing reservoir temperature and reducing crude oil viscosity, thereby improving permeability. Previous studies have shown that combining these two methods can significantly improve production efficiency through the synergistic effects of force and heat. However, the effectiveness of electric pulse excitation is highly dependent on the consistency of the excitation frequency with the natural resonant frequency of the formation. Only when the excitation frequency precisely matches the primary resonant frequency can maximum vibration coupling and fracture resonance propagation be induced within the shale. Patent publication number CN110485959B discloses a shale oil and gas microwave resonance shock synergistic stimulation method. This method combines the advantages of electromagnetic wave heating stimulation and spectrum resonance shock stimulation, superimposing high-energy shock waves at the natural frequency of the target reservoir on electromagnetic wave heating. It offers low pollution and low energy consumption.

[0003] However, the aforementioned patented technologies and existing technologies typically use fixed-frequency excitation or frequency sweeps based on manual experience to find the primary resonant frequency. These methods are not only inefficient and computationally intensive, but are also susceptible to interference from complex geological conditions, leading to inaccurate identification of the primary resonant frequency, which in turn affects the coupling strength and resonance efficiency of the electric pulse excitation. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an oil and gas microwave electric pulse resonance shock collaborative production increase control system, which learns the main resonance frequency through self-supervised training to solve the technical problems raised in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an oil and gas microwave electric pulse resonance shock coordinated production increase control system, comprising:

[0006] The vibration response acquisition module is used to acquire the vibration response sequence fed back by the target shale layer within a time window.

[0007] The vibration response input module is used to input the vibration response sequence into a pre-trained resonance frequency extraction model.

[0008] The main resonance frequency determination module is used to identify the main resonance frequency of the vibration response sequence using the resonance frequency extraction model to determine the main resonance frequency of the target shale layer.

[0009] The electric pulse adjustment module is used to adjust the output frequency of the electric pulse excitation device according to the determined main resonance frequency until the output frequency is consistent with the main resonance frequency.

[0010] In some embodiments, the pre-training step of the resonance frequency extraction model includes:

[0011] A1. Obtain a pseudo-label training set containing N two-tuple sample pairs; the target labels of the two-tuple sample pairs are pseudo-labels extracted from the frequency-amplitude distribution diagram.

[0012] A2. Extract the binary sample pairs of the current batch from the pseudo-label training set, input them into the multi-layer perceptron model for self-supervised training, and output the predicted frequency of the current batch.

[0013] A3. Based on the set self-supervised training loss function, calculate the error loss between the predicted frequency and pseudo label of the current batch, and perform backpropagation to update the model parameters.

[0014] A4. After the model parameters are updated, determine whether the current model meets the preset convergence conditions.

[0015] A5. If satisfied, terminate the self-supervised training. Otherwise, repeat A2 to A4 until the current model meets the convergence conditions.

[0016] A4. If the model convergence condition is met, stop minimizing the error loss between the prediction frequency and pseudo label of the current batch; otherwise, update the model parameters at the preset learning rate to generate updated parameters.

[0017] A5. Repeat A2 to A3 according to the updated parameters until the model converges.

[0018] In some embodiments, obtaining a pseudo-label training set containing N two-tuple sample pairs includes:

[0019] A1-1. Obtain vibration response sequences fed back by the target shale layer within N time windows.

[0020] A1-2. Based on the vibration response sequence fed back in each time window, a time-frequency diagram within the time window is constructed.

[0021] A1-3. Construct a frequency amplitude distribution diagram based on the time-frequency diagram within the time window.

[0022] A1-4. Select the main resonant frequency in the frequency amplitude distribution diagram.

[0023] A1-5. Define the vibration response sequence as an input vector, define the main resonance frequency as a pseudo label, and generate the binary sample pair.

[0024] A1-6. Repeat A1-2 to A1-5 until a pseudo-label training set containing N two-tuple sample pairs is obtained.

[0025] In some embodiments, obtaining a vibration response sequence fed back by a target shale layer within N time windows includes:

[0026] A1-1-1. Anchor the target shale layer and its N excitation frequencies.

[0027] A1-1-2. According to the excitation frequency, preset the output frequency of the electric pulse excitation device.

[0028] A1-1-3. Use an electric pulse excitation device with a preset output frequency to transmit the excitation frequency to the target shale layer.

[0029] A1-1-4. Collect several vibrations corresponding to the excitation frequency of the target shale layer and arrange them in chronological order as a vibration response sequence.

[0030] A1-1-5. Repeat A1-1-2 to A-1-4 to obtain N groups of vibration response sequences with different excitation frequencies.

[0031] In some embodiments, constructing a time-frequency diagram within each time window based on the vibration response sequence fed back within each time window includes:

[0032] A1-2-1. Divide the time window covered by each vibration response sequence into several time sub-windows.

[0033] A1-2-2. Perform short-time Fourier transform on the vibration response of each time sub-window to obtain the frequency energy distribution of the time sub-window.

[0034] The frequency energy distribution is characterized as follows: within a given time subwindow, a set of frequency components are obtained after short-time Fourier transform, and the amplitude of each frequency component within the time subwindow; the amplitude is used to represent the response intensity of the corresponding frequency component within the time subwindow.

[0035] A1-2-3. Concatenate the frequency energy distribution of each time sub-window in chronological order to construct a time-frequency diagram within the time window.

[0036] The horizontal axis of the time-frequency diagram is the time sub-window, and the vertical axis is the frequency energy distribution corresponding to the time sub-window. The frequency energy distribution is the distribution of several frequency components in the time sub-window.

[0037] In some embodiments, constructing a frequency-amplitude distribution diagram based on the time-frequency diagram within the time window includes:

[0038] A1-3-1. In the time-frequency diagram, extract the amplitude sequence of each frequency component in all time sub-windows.

[0039] A1-3-2. Perform mean calculation on the amplitude sequence of each frequency component in all time sub-windows to obtain the average amplitude of each frequency component.

[0040] A1-3-3. Obtain the average amplitude of all frequency components.

[0041] A1-3-4. With the frequency component as the horizontal axis and the average amplitude of each frequency component as the vertical axis, determine several frequency amplitude points.

[0042] A1-3-5. Connect several frequency-amplitude points to generate a frequency-amplitude curve to construct a frequency-amplitude distribution diagram.

[0043] The frequency amplitude distribution diagram is used to characterize the energy concentration of each frequency component during the vibration response process.

[0044] In some embodiments, selecting a primary resonant frequency in the frequency amplitude distribution graph includes:

[0045] A1-4-1. Mark all amplitude peaks of the frequency-amplitude curve in the frequency-amplitude distribution diagram .

[0046] A1-4-2. Starting from the marked amplitude peak, search along the left and right sides of the frequency-amplitude curve for the left half frequency point and the right half frequency point that are first lower than half of the amplitude peak.

[0047] A1-4-3. Calculate the distance between the left half frequency point and the right half frequency point, and define the distance as the half-height width of the amplitude peak .

[0048] A1-4-4. Traverse the entire frequency-amplitude curve and obtain the half-height widths corresponding to K corresponding amplitude peaks.

[0049] A1-4-5. Calculate the energy concentration of each frequency component based on the half-widths of the K corresponding amplitude peaks to obtain K energy concentrations.

[0050] The calculation expression of the energy concentration is:

[0051] .

[0052] in, represents the energy concentration of the i-th frequency component, represents the half-maximum width of the i-th frequency component, represents the peak amplitude of the i-th frequency component, The normalized amplitude weight of the i-th peak is used to dynamically balance the influence weights of the half-maximum width and amplitude; represents the regularization constant to prevent the denominator from dividing by zero error, represents the slope smoothing term.

[0053] A1-4-6. Select the frequency component corresponding to the minimum concentration from the K energy concentrations, and define the frequency component as the main resonant frequency.

[0054] In some embodiments, the target loss of the self-supervised training loss function is mean square error loss or Huber loss.

[0055] In some embodiments, the convergence condition is: when the model satisfies any of the following conditions in several consecutive rounds of training, it is determined to be converged, including:

[0056] The error loss of the current batch is lower than the preset error threshold.

[0057] The error loss change of the model within M consecutive training cycles is less than the set change threshold ε.

[0058] The number of training rounds reaches the preset maximum number of iterations.

[0059] In some embodiments, the control system also includes: a microwave adjustment module for adjusting the output frequency of the microwave emitting device so that the adjusted output frequency is within the shale layer absorption frequency band; and a module for adjusting the microwave action duration and penetration depth of the wave emitting device.

[0060] The present invention provides an oil and gas microwave electric pulse resonance shock coordinated production increase control system, which has the following beneficial effects:

[0061] The present invention uses self-supervised training constructed through vibration response sequences and pseudo-label frequencies, utilizes the main resonance frequencies extracted from the frequency amplitude distribution diagram as pseudo-labels, and performs training with a multi-layer perceptron model. This allows the main resonance frequencies of formations to be automatically learned without manual labeling or external experience input, effectively improving the ability to fit the nonlinear response patterns of complex formations and enhancing the accuracy of main resonance frequency identification.

[0062] Furthermore, by integrating the full width at half maximum with the peak amplitude and frequency slope energy concentration, the primary resonant frequency can be accurately screened from the frequency amplitude distribution. This energy concentration quantifies the concentration of frequency components and the sharpness of the response, avoiding the misidentification problems caused by unstable response amplitude or noise interference in traditional methods. It can more effectively distinguish the primary resonant frequency from non-primary frequency responses, improving the reliability of frequency extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1This is a structural block diagram of an oil and gas microwave electric pulse resonance shock coordinated production increase control system of the present invention.

[0064] Figure 2 This is a control flow chart of an oil and gas microwave electric pulse resonance shock collaborative production increase control system of the present invention.

[0065] Figure 3 Schematic diagram of the process of defining the main resonance frequency of the present invention. DETAILED DESCRIPTION

[0066] 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.

[0067] Example 1: Please refer to Figures 1 to 2 The present invention provides an oil and gas microwave electric pulse resonance shock collaborative production increase control system, which includes: a vibration response acquisition module for acquiring the vibration response sequence fed back by the target shale layer within a time window.

[0068] Exemplarily, the vibration response sequence is collected by a three-axis vibration sensor deployed underground under tentative electric pulse excitation conditions, recording the formation response of the target shale layer in the three directions of X, Y, and Z, with each direction corresponding to an independent time series.

[0069] In one embodiment of the present invention, the vibration response sequence refers to the three-axis acceleration continuously acquired at a fixed sampling frequency within a set time window, forming a three-channel time series matrix with a structure of three rows and N columns, which is used to characterize the dynamic response behavior of the target shale layer under electric pulse excitation.

[0070] The vibration response input module is used to input the vibration response sequence into a pre-trained resonance frequency extraction model.

[0071] The main resonance frequency determination module is used to identify the main resonance frequency of the vibration response sequence using the resonance frequency extraction model to determine the main resonance frequency of the target shale layer.

[0072] The electric pulse adjustment module is used to adjust the output frequency of the electric pulse excitation device according to the determined main resonance frequency until the output frequency is consistent with the main resonance frequency.

[0073] The control system further comprises:

[0074] The microwave adjustment module is used to adjust the output frequency of the microwave transmitter to within the shale absorption band, thereby enhancing the microwave energy deposition efficiency in the formation. It is also used to adjust the microwave action duration and penetration depth of the microwave transmitter, ensuring that the microwave energy fully covers the target reservoir area in the shale layer and reduces the oil layer temperature to the viscosity reduction threshold, thereby synergistically improving oil and gas mobility and fracture expansion capacity, enhancing the resonant fracture effect of electric pulse excitation, and achieving a synergistic production increase effect of microwave heating and electric pulse shock.

[0075] In this embodiment, by inputting the triaxial vibration response sequence of the target shale layer under tentative electric pulse excitation into a pre-trained resonant frequency extraction model, the optimal primary resonant frequency can be quickly identified, precisely matching the output frequency of the electric pulse excitation device with the natural resonant frequency of the shale layer, thereby improving the effectiveness of the resonant excitation. After the matching is completed, by adjusting the output frequency of the microwave transmitter to the shale layer absorption band and controlling the microwave action duration and penetration depth, the microwave thermal effect is fully covered within the shale reservoir, driving the oil layer temperature to the critical point of viscosity drop, helping to improve crude oil fluidity and reduce reservoir toughness, thereby enhancing the fracture triggering strength of the electric pulse excitation, and ultimately achieving the goal of synergistic production control.

[0076] Example 2: The technical solution of Example 2 differs from that of Example 1 in that it discloses a pre-training step of the resonance frequency extraction model in Example 1, and the pre-training step includes:

[0077] A1. Obtain a pseudo-label training set containing N two-tuple sample pairs; the target labels of the two-tuple sample pairs are pseudo-labels extracted from the frequency-amplitude distribution diagram.

[0078] A2. Extract the binary sample pairs of the current batch from the pseudo-label training set, input them into the multi-layer perceptron model for self-supervised training, and output the predicted frequency of the current batch.

[0079] A3. Based on the set self-supervised training loss function, calculate the error loss between the predicted frequency and pseudo label of the current batch, and perform backpropagation to update the model parameters.

[0080] A4. After the model parameters are updated, determine whether the current model meets the preset convergence conditions.

[0081] A5. If satisfied, terminate the self-supervised training. Otherwise, repeat A2 to A4 until the current model meets the convergence conditions.

[0082] A4. If the model convergence condition is met, stop minimizing the error loss between the prediction frequency and pseudo label of the current batch; otherwise, update the model parameters at the preset learning rate to generate updated parameters.

[0083] A5. Repeat A2 to A3 according to the updated parameters until the model converges.

[0084] In this example, a pseudo-labeled training set consisting of two-tuples of vibration response sequences and the main resonance frequency was constructed for the main resonance frequency identification task. Self-supervised training was then performed based on a multi-layer perceptron (MLP) model. This training mechanism can efficiently fit the nonlinear mapping relationship between the shale layer response sequence and the main resonance frequency without manual labeling.

[0085] For example, in step A1, the steps of obtaining the pseudo-label training set are:

[0086] A1-1. Obtain vibration response sequences fed back by the target shale layer within N time windows.

[0087] For example, during the data construction phase of model pre-training, the system sequentially excites the target shale layer at a set frequency by controlling an electric pulse excitation device. A triaxial vibration sensor deployed downhole simultaneously collects the response signals. The sensor continuously records vibration acceleration data in the X, Y, and Z directions at a fixed sampling frequency within a set time window, generating a three-channel time series of samples. Each set of samples is individually associated with its corresponding excitation frequency.

[0088] A1-2. Based on the vibration response sequence fed back in each time window, a time-frequency diagram within the time window is constructed.

[0089] A1-3. Construct a frequency amplitude distribution diagram based on the time-frequency diagram within the time window.

[0090] A1-4. Select the main resonant frequency in the frequency amplitude distribution diagram.

[0091] A1-5. Define the vibration response sequence as an input vector, define the main resonance frequency as a pseudo label, and generate the binary sample pair.

[0092] A1-6. Repeat A1-2 to A1-5 until a pseudo-label training set containing N two-tuple sample pairs is obtained.

[0093] In this example, by systematically collecting triaxial vibration response sequences of the target shale formation at different excitation frequencies, constructing a time-frequency plot using a short-time Fourier transform, and then extracting the primary resonant frequencies as pseudo-labels based on the frequency-amplitude distribution, high-quality training data can be generated without manual annotation. This construction process, based on the actual shale formation response, ensures that the pseudo-label frequencies are sensitive to the formation structure and enables the automatic generation of frequency pseudo-labels with high confidence under unsupervised conditions.

[0094] Furthermore, the A1-1 step specifically includes:

[0095] A1-1-1. Anchor the target shale layer and its N excitation frequencies.

[0096] A1-1-2. According to the excitation frequency, preset the output frequency of the electric pulse excitation device.

[0097] A1-1-3. Use an electric pulse excitation device with a preset output frequency to transmit the excitation frequency to the target shale layer.

[0098] A1-1-4. Collect several vibrations corresponding to the excitation frequency of the target shale layer and arrange them in chronological order as a vibration response sequence.

[0099] A1-1-5. Repeat A1-1-2 to A-1-4 to obtain N groups of vibration response sequences with different excitation frequencies.

[0100] In this example, by anchoring the target shale layer and sequentially applying N set excitation frequencies, combined with the frequency presetting of the electric pulse excitation device and the dynamic acquisition of the formation response, a systematic construction of multiple sets of vibration response sequences corresponding to different excitation conditions was achieved. This acquisition process ensured the consistency of sample sources and comprehensive frequency coverage.

[0101] Furthermore, the step A1-2 specifically includes:

[0102] A1-2-1. Divide the time window covered by each vibration response sequence into several time sub-windows.

[0103] A1-2-2. Perform short-time Fourier transform on the vibration response of each time sub-window to obtain the frequency energy distribution of the time sub-window.

[0104] The frequency energy distribution is characterized as follows: within a given time subwindow, a set of frequency components are obtained after short-time Fourier transform, and the amplitude of each frequency component within the time subwindow; the amplitude is used to represent the response intensity of the corresponding frequency component within the time subwindow.

[0105] A1-2-3. Concatenate the frequency energy distribution of each time sub-window in chronological order to construct a time-frequency diagram within the time window.

[0106] The horizontal axis of the time-frequency diagram is the time sub-window, and the vertical axis is the frequency energy distribution corresponding to the time sub-window. The frequency energy distribution is the distribution of several frequency components in the time sub-window.

[0107] In an embodiment of the present invention, the system sets a sliding analysis window for the complete vibration response sequence and performs a short-time Fourier transform in each window to obtain the frequency energy distribution of the time segment; the spectrum results of all windows are integrated into a complete time-frequency diagram according to the time series, in which the horizontal axis represents time and the vertical axis represents frequency components. The amplitude of each point in the diagram represents the energy intensity of a certain frequency component in a specific time window, which is used to characterize the multi-frequency response characteristics of the formation.

[0108] Furthermore, the steps A1-3 specifically include:

[0109] A1-3-1. In the time-frequency diagram, extract the amplitude sequence of each frequency component in all time sub-windows.

[0110] A1-3-2. Perform mean calculation on the amplitude sequence of each frequency component in all time sub-windows to obtain the average amplitude of each frequency component.

[0111] A1-3-3. Obtain the average amplitude of all frequency components.

[0112] A1-3-4. With the frequency component as the horizontal axis and the average amplitude of each frequency component as the vertical axis, determine several frequency amplitude points.

[0113] A1-3-5. Connect several frequency-amplitude points to generate a frequency-amplitude curve to construct a frequency-amplitude distribution diagram.

[0114] The frequency amplitude distribution diagram is used to characterize the energy concentration of each frequency component during the vibration response process.

[0115] In this embodiment, by performing statistical mean processing on the amplitudes of each frequency component within a time subwindow in the time-frequency diagram, a frequency-amplitude distribution diagram is constructed. This effectively removes temporal perturbations from the formation response and extracts steady-state characteristics reflecting the strength of the formation frequency response. This frequency-amplitude distribution diagram, with frequency as the horizontal axis and average amplitude as the vertical axis, intuitively displays the energy contribution of each frequency component throughout the entire response process.

[0116] Further, see Figure 3 , the steps A1-4 specifically further include:

[0117] A1-4-1. Mark all amplitude peaks of the frequency-amplitude curve in the frequency-amplitude distribution diagram .

[0118] For example, a local maximum detection algorithm may be used to determine, in the frequency-amplitude curve, a turning point where adjacent frequency points rise on the left and fall on the right as an amplitude peak.

[0119] A1-4-2. Starting from the marked amplitude peak, search along the left and right sides of the frequency-amplitude curve for the left half frequency point and the right half frequency point that are first lower than half of the amplitude peak.

[0120] A1-4-3. Calculate the distance between the left half frequency point and the right half frequency point, and define the distance as the half-height width of the amplitude peak .

[0121] A1-4-4. Traverse the entire frequency-amplitude curve and obtain the half-height widths corresponding to K corresponding amplitude peaks.

[0122] A1-4-5. Calculate the energy concentration of each frequency component based on the half-widths of the K corresponding amplitude peaks to obtain K energy concentrations.

[0123] The calculation expression of the energy concentration is:

[0124] .

[0125] in, Indicates the energy concentration of the i-th frequency component, which is used to compare multiple frequency peaks. The smaller the energy concentration, the more likely the frequency component is the main resonant frequency. represents the half-height width of the i-th frequency component, which indicates the width of the energy distribution of this frequency (the smaller the more concentrated), Indicates the peak amplitude of the i-th frequency component, indicating the response intensity of the frequency (the larger the more obvious), The normalized amplitude weight of the i-th peak is used to dynamically balance the influence weights of the half-width and amplitude. In this embodiment, the normalized amplitude weight can adopt a minimum-maximum scaling method; represents the regularization constant to prevent the denominator from dividing by zero error, Represents the slope smoothing term, which is used to prevent the amplitude / width ratio from being 0 and maintain the continuity of concentration.

[0126] Specifically, the calculation expression of energy concentration consists of two parts. The first part is , represents power adaptive weighting. Large means strong peaks, emphasizing the importance of amplitude → the denominator becomes larger and the score becomes smaller (tending to retain strong peaks); Small means weak peaks, emphasizing half-width penalty → the numerator becomes larger, the score becomes larger (tends to exclude wide and weak peaks); through power adaptive weighting, dynamically balance "amplitude vs. concentration". Part 2 represents the slope penalty factor, where It is the unit width slope of the frequency peak (i.e., "sharpness"), the sharper (larger slope) → large denominator → small score (better), the flatter (smaller slope) → small denominator → large score (penalty).

[0127] A1-4-6. Select the frequency component corresponding to the minimum concentration from the K energy concentrations, and define the frequency component as the main resonant frequency.

[0128] In other words, the energy concentration comprehensively considers the frequency amplitude, half-width and sharpness, adaptively adjusts the influence of each parameter through the normalized amplitude weight, and combines the peak slope index to finally select the frequency corresponding to the lowest concentration as the main frequency.

[0129] In this example, by extracting amplitude peaks from the frequency-amplitude distribution diagram and calculating their half-width, combined with energy concentration, a comprehensive quantitative assessment of multiple frequency components across three dimensions: energy concentration, amplitude intensity, and response sharpness is achieved. Compared to traditional dominant frequency extraction methods that select only based on peak amplitude, this method balances the influence between width and height through normalized weights, while introducing frequency slope as an auxiliary indicator of distribution sharpness. This method effectively suppresses interference from wide and weak perturbation peaks in the formation response, accurately identifies the frequency components with the most concentrated performance in the resonance response, and provides highly reliable pseudo-labels for self-supervised training.

[0130] In the above embodiment 2, the target loss of the self-supervised training loss function is mean square error loss or Huber loss.

[0131] Specifically, the mean square error loss can square the frequency prediction error, thereby producing a stronger penalty effect on prediction results with large deviations during training; while the Huber loss (smoothed L1 loss) manifests itself as a mean square error when the error is small, and degenerates into a linear penalty when the error is large, thereby maintaining error sensitivity while having stronger resistance to outliers; compared with the two, the Huber loss is more robust in actual collected data with geological anomalies or sensor interference, while the mean square error can achieve faster convergence and higher accuracy under high-quality pseudo-label training samples, and can be selected according to the quality of training data and model tuning strategy.

[0132] The model convergence condition in Example 2 is: when the model meets any of the following conditions in several consecutive rounds of training, it is determined to be converged, including:

[0133] The error loss of the current batch is lower than the preset error threshold.

[0134] The error loss change of the model in several consecutive training cycles is less than the set change threshold ε.

[0135] The number of training rounds reaches the preset maximum number of iterations.

[0136] The model convergence conditions in this embodiment are based on three criteria: an error threshold, a rate of change limit, and a maximum number of iterations. These conditions are used to ensure the stability and controllability of the self-supervised training process. These convergence conditions provide flexible termination criteria at different training stages to avoid overfitting or invalid iterations.

[0137] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired method (e.g., infrared, wireless, microwave, etc.).

[0138] The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that contains one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division of a waterway underwater terrain change analysis system and method. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0140] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An oil and gas microwave electric pulse resonance shock coordinated production increase control system, characterized in that: include: A vibration response acquisition module is used to acquire the vibration response sequence fed back by the target shale layer within a time window; a vibration response input module, configured to input the vibration response sequence into a pre-trained resonance frequency extraction model; a main resonance frequency determination module, configured to identify the main resonance frequency of the vibration response sequence using the resonance frequency extraction model, and determine the main resonance frequency of the target shale layer; an electric pulse adjustment module, configured to adjust the output frequency of the electric pulse excitation device according to the determined main resonant frequency until the output frequency is consistent with the main resonant frequency; The pre-training step of the resonance frequency extraction model includes: A1. Obtain a pseudo-label training set containing N two-tuple sample pairs; the target labels of the two-tuple sample pairs are pseudo-labels extracted from the frequency-amplitude distribution graph; A2. Extract the current batch of binary sample pairs from the pseudo-label training set, input them into the multi-layer perceptron model for self-supervised training, and output the predicted frequency of the current batch; A3. Based on the set self-supervised training loss function, calculate the error loss between the predicted frequency and pseudo label of the current batch, and perform backpropagation to update the model parameters; A4. After the model parameters are updated, determine whether the current model meets the preset convergence conditions; A5. If satisfied, terminate the self-supervised training. Otherwise, repeat A2 to A4 until the current model meets the convergence conditions. A4. If the model convergence condition is met, stop minimizing the error loss between the predicted frequency and pseudo-label of the current batch; otherwise, update the model parameters at the preset learning rate to generate the updated parameters; A5. Repeat A2 to A3 according to the updated parameters until the model converges. Get a pseudo-label training set containing N two-tuple sample pairs, including: A1-1. Obtain vibration response sequences fed back from the target shale layer within N time windows; A1-2. Construct a time-frequency diagram within each time window based on the vibration response sequence fed back within each time window; A1-3. Constructing a frequency amplitude distribution diagram based on the time-frequency diagram within the time window; A1-4, selecting a main resonant frequency in the frequency amplitude distribution diagram; A1-5. Define the vibration response sequence as an input vector, define the main resonance frequency as a pseudo label, and generate the binary sample pair; A1-6. Repeat A1-2 to A1-5 until a pseudo-label training set containing N two-tuple sample pairs is obtained; Selecting a main resonant frequency in the frequency amplitude distribution diagram includes: A1-4-1. Mark all amplitude peaks of the frequency-amplitude curve in the frequency-amplitude distribution diagram ; A1-4-2. Starting from the marked amplitude peak, search along the left and right sides of the frequency-amplitude curve for the left and right half-frequency points that are first lower than half of the amplitude peak; A1-4-3. Calculate the distance between the left half frequency point and the right half frequency point, and define the distance as the half-height width of the amplitude peak ; A1-4-4. Traverse the entire frequency-amplitude curve and obtain the half-height widths of the corresponding K amplitude peaks; A1-4-5. Calculate the energy concentration of each frequency component based on the half-widths of the K corresponding amplitude peaks to obtain K energy concentrations; The calculation expression of the energy concentration is: ； in, represents the energy concentration of the i-th frequency component, represents the half-maximum width of the i-th frequency component, represents the peak amplitude of the i-th frequency component, The normalized amplitude weight of the i-th peak is used to dynamically balance the influence weights of the half-maximum width and amplitude; represents the regularization constant to prevent the denominator from dividing by zero error, represents the slope smoothing term; A1-4-6. Select the frequency component corresponding to the minimum concentration from the K energy concentrations, and define the frequency component as the main resonant frequency.

2. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: Obtain the vibration response sequence of the target shale layer feedback within N time windows, including: A1-1-1, anchoring the target shale layer and its N excitation frequencies; A1-1-2. Preset the output frequency of the electric pulse excitation device according to the excitation frequency; A1-1-3. Using an electric pulse excitation device with a preset output frequency, transmit the excitation frequency to the target shale layer; A1-1-4. Collect several vibrations corresponding to the excitation frequency of the target shale layer and arrange them in chronological order as a vibration response sequence; A1-1-5. Repeat A1-1-2 to A-1-4 to obtain N groups of vibration response sequences with different excitation frequencies.

3. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: Based on the vibration response sequence fed back in each time window, a time-frequency diagram within the time window is constructed, including: A1-2-1. Divide the time window covered by each vibration response sequence into several time sub-windows; A1-2-2. Perform a short-time Fourier transform on the vibration response of each time sub-window to obtain the frequency energy distribution of the time sub-window; The frequency energy distribution is characterized by: within a given time sub-window, a set of frequency components is obtained after short-time Fourier transform, and the amplitude of each frequency component within the time sub-window; the amplitude is used to represent the response intensity of the corresponding frequency component within the time sub-window; A1-2-3. Concatenate the frequency energy distribution of each time sub-window in chronological order to construct a time-frequency diagram within the time window; The horizontal axis of the time-frequency diagram is the time sub-window, and the vertical axis is the frequency energy distribution corresponding to the time sub-window. The frequency energy distribution is the distribution of several frequency components in the time sub-window.

4. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: Constructing a frequency amplitude distribution diagram according to the time-frequency diagram within the time window, including: A1-3-1. In the time-frequency diagram, extract the amplitude sequence of each frequency component in all time sub-windows; A1-3-2. Perform mean calculation on the amplitude sequence of each frequency component in all time sub-windows to obtain the average amplitude of each frequency component; A1-3-3. Obtain the average amplitude of all frequency components; A1-3-4. Using the frequency component as the horizontal axis and the average amplitude of each frequency component as the vertical axis, determine several frequency amplitude points. A1-3-5. Connect several frequency-amplitude points to generate a frequency-amplitude curve to construct a frequency-amplitude distribution diagram; The frequency amplitude distribution diagram is used to characterize the energy concentration of each frequency component during the vibration response process.

5. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: The target loss of the self-supervised training loss function is mean square error loss or Huber loss.

6. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: The convergence condition is: The model is considered converged when it meets any of the following conditions in several consecutive rounds of training: The error loss of the current batch is lower than the preset error threshold; The error loss change of the model in M ​​consecutive training cycles is less than the set change threshold ε; The number of training rounds reaches the preset maximum number of iterations.

7. The oil and gas microwave electric pulse resonance shock coordinated production increase control system according to claim 1 is characterized in that: The control system further comprises: The microwave adjustment module is used to adjust the output frequency of the microwave emitting device so that the adjusted output frequency is within the shale layer absorption frequency band; and is used to adjust the microwave action time and penetration depth of the wave emitting device.

Images