A method and system for monitoring vibration of long oil and gas pipeline based on DAS

Abstract

The application discloses a kind of long oil and gas pipeline vibration monitoring method and system based on DAS, it is related to optical fiber vibration detection technical field;After original signal is carried out ensemble empirical mode decomposition, effective IMF component is obtained by preset screening condition screening target IMF component set, according to target IMF component set, vibration signal is reconstructed, and target original signal is removed to obtain target effective signal;All target effective signals are substituted into preset vibration identification model to obtain vibration characteristics, and vibration characteristics are substituted into preset database to compare and determine vibration type;Through ensemble empirical mode decomposition and adaptive signal reconstruction technology, the interference signal superposition and parameter distortion problems that traditional pipeline vibration monitoring method faces in complex environment are solved, and then through the rapid comparison of intelligent identification model and feature database, long oil and gas pipeline full line, high-precision, anti-interference vibration state real-time perception and type discrimination are realized, and the pipeline vibration monitoring efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of fiber optic vibration detection technology, specifically relating to a vibration monitoring method and system for long-distance oil and gas pipelines based on DAS. Background Technology

[0002] As a core infrastructure for energy transportation, long-distance oil and gas pipelines are prone to abnormal vibrations during their long-term service life due to various factors such as internal medium erosion, pressure fluctuations, external third-party construction damage, geological disasters, and equipment failures.

[0003] Traditional vibration monitoring methods suffer from limitations such as limited monitoring range, high deployment costs, susceptibility to electromagnetic interference, and significant data transmission bottlenecks. They are unable to achieve long-distance, full-range, and real-time vibration status perception, and often fail to detect potential safety hazards in a timely manner, resulting in a persistently high risk of major accidents such as leaks and explosions.

[0004] Distributed optical fiber sensing (DAS) technology, with its unique advantage of "one optical fiber equals one sensor array", can achieve continuous distributed monitoring over a range of tens to hundreds of kilometers. It also features high spatial resolution, high sensitivity, resistance to electromagnetic interference, tolerance to harsh environments, and convenient deployment, providing a brand-new technical path for vibration monitoring of long-distance oil and gas pipelines.

[0005] Publication No. CN112542046A discloses a method for early warning monitoring of heavy vehicles in long-distance pipelines based on DAS. This method involves burying optical fibers along the pipeline, with the fiber's origin connected to a vibration detection system. The vibration detection system receives the signal transmitted from the fiber and demodulates the vibration signal. A Fourier transform is performed on the vibration signal to obtain three parameters: vibration amplitude, frequency energy, and frequency domain energy distribution. The type of heavy vehicle is determined based on a scoring mechanism. After acquiring the heavy vehicle signal, the system calculates the vehicle's speed and direction within the fiber's detection range based on the trajectory peak. This method, by combining optical fibers buried along long-distance pipelines with a vibration detection system, extracts parameters such as amplitude and frequency energy of the vibration signal and identifies heavy vehicles using a scoring mechanism. Simultaneously, it calculates the vehicle's speed and direction based on the trajectory peak, achieving early warning monitoring of heavy vehicles along the pipeline. However, this scheme does not fully consider the complex environmental interference factors along the pipeline. These interference signals can superimpose and merge with the actual vibration signal to be detected, leading to parameter distortion and ultimately resulting in low efficiency in pipeline vibration monitoring. Summary of the Invention

[0006] The purpose of this invention is to solve the problem that interference signals can be superimposed and merged with the actual vibration signals to be detected, resulting in parameter distortion and ultimately low efficiency of pipeline vibration monitoring. Therefore, this invention proposes a vibration monitoring method and system for long-distance oil and gas pipelines based on DAS.

[0007] In a first aspect of this invention, a vibration monitoring method for long-distance oil and gas pipelines based on DAS is first proposed, the method comprising: The distributed acoustic sensor host collects the original vibration signals of the oil and gas pipeline at a preset period to obtain the original signal set; The target original signal is subjected to ensemble empirical mode decomposition to obtain an IMF component set; the target original signal is any one of the original signal sets; The target IMF component set is obtained by filtering effective IMF components in the IMF component set using preset filtering conditions, and the vibration signal is reconstructed based on the target IMF component set to obtain the target reconstructed signal. The original target signal is noise-removed based on the reconstructed target signal to obtain the effective target signal. All valid signals from the target are substituted into a preset vibration identification model to obtain vibration characteristics. The vibration characteristics are then substituted into a preset database for comparison to determine the vibration type. The preset database stores the vibration characteristics corresponding to different vibration types.

[0008] Optionally, the target IMF component set is obtained by filtering valid IMF components from the IMF component set using preset filtering conditions, including: The mutual information value between the target preset IMF component and the original transmitted signal is calculated to obtain a similarity measure; the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensing host. The baseline similarity is obtained by calculating the mutual information value between the target preset IMF component and the target original signal; The similarity correlation coefficient is obtained by calculating the difference between the measure of similarity and the baseline similarity; The similarity coefficient set is obtained by performing max-min normalization on the similarity correlation coefficients corresponding to all IMF components. After sorting the similarity coefficients in the set of similarity coefficients in descending order, the difference between adjacent similarity coefficients is calculated, and it is determined whether the difference is less than a preset similarity coefficient threshold. If the difference is less than the preset similarity coefficient threshold, the IMF component corresponding to the smallest similarity coefficient in the difference will be recorded as the effective IMF component. Obtain all valid IMF components to obtain the target IMF component set.

[0009] Optionally, removing noise from the original target signal based on the reconstructed target signal to obtain the effective target signal includes: Step 1: Substitute the target reconstructed signal into a preset model to obtain the target noise level; Step 2: Initialize the decision vector. Calculate the membership degree of the target noise level with each decision value in the decision vector set using the Gaussian membership function, and then sum and normalize all membership degree values ​​to obtain a normalized membership degree value set. Step 3: Initialize the fuzzy variance vector, and obtain the fuzzy coefficients by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set; Step 4: Substitute the ambiguity coefficients into a preset formula to denoise the original target signal to obtain an intermediate denoised signal; Step 5: Substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, update the intermediate denoised level to the target noise level and return to step 2. Step 6: After step 4, the following is also included: if the intermediate denoising level is less than or equal to the preset noise level, the intermediate denoising signal is output and recorded as the target valid signal.

[0010] Optionally, the method includes training a pre-defined model: Acquire training data and input the training data into a preset model for training to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; the preset model is a deep convolutional generative adversarial network (DCGAN) model. Update the preset model according to the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is judged to be trained successfully.

[0011] Optionally, substituting the target valid signal into a preset vibration identification model to obtain vibration characteristics includes: A spatiotemporal matrix is ​​constructed based on all valid signals from the target, and the spatiotemporal matrix is ​​then normalized by mean to obtain a standard spatiotemporal matrix. Substituting the standard spatiotemporal matrix into the group convolutional backbone network yields the first convolutional feature; Substituting the first convolutional feature into the multi-scale feature pyramid yields low-scale features, medium-scale features, and high-scale features. Perform 1×1 convolution operations on the low-scale features, the medium-scale features, and the high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features, and convolutional high-scale features; The high-scale convolutional features are upsampled and then concatenated with the mid-scale convolutional features to obtain the first concatenated feature. The convolutional mid-scale features are upsampled and then concatenated with the convolutional low-scale features to obtain the second concatenated feature. Substitute the second splicing feature into the attention mechanism to obtain the attention weight; After upsampling the first splicing feature, the first splicing feature and the second splicing feature are fused according to the attention weight to obtain the target fused feature; The vibration features are obtained by performing a 1×1 convolution on the target fusion features and then feeding them into a fully connected layer.

[0012] In a second aspect of this invention, a vibration monitoring system for long-distance oil and gas pipelines based on DAS is proposed, comprising: The raw signal set generation module is used by the distributed acoustic sensor host to collect and receive the original vibration signals of the oil and gas pipeline through a preset period to obtain the raw signal set. The IMF component set generation module is used to perform ensemble empirical mode decomposition on the target original signal to obtain the IMF component set; the target original signal is any one of the original signal sets; The target IMF component set generation module is used to filter valid IMF components in the IMF component set through preset filtering conditions to obtain a target IMF component set, and to reconstruct the vibration signal based on the target IMF component set to obtain a target reconstructed signal. The target valid signal generation module is used to remove noise from the original target signal based on the target reconstructed signal to obtain the target valid signal; The vibration feature generation module is used to input all target valid signals into a preset vibration recognition model to obtain vibration features, and to input the vibration features into a preset database for comparison to determine the vibration type; the preset database stores vibration features corresponding to different vibration types.

[0013] Optionally, the target IMF component set generation module includes: The similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the original transmitted signal to obtain the similarity measurement; the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensing host. The baseline similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the target original signal to obtain the baseline similarity. The similarity correlation coefficient calculation module is used to calculate the difference between the measure of similarity and the baseline similarity to obtain the similarity correlation coefficient; The normalization module is used to perform max-min normalization on the similarity correlation coefficients corresponding to all IMF components to obtain a set of similarity coefficients. The similarity coefficient sorting module is used to sort the similarity coefficients in the similarity coefficient set in descending order, calculate the difference between adjacent similarity coefficients, and determine whether the difference is less than a preset similarity coefficient threshold. The effective IMF component determination module is used to determine the IMF component corresponding to the smallest similarity coefficient among the differences if the difference is less than the preset similarity coefficient threshold. The target IMF component set determination module is used to obtain all valid IMF components to obtain the target IMF component set.

[0014] Optionally, the target valid signal generation module includes: The target noise level generation module is used to substitute the target reconstructed signal into a preset model to obtain the target noise level; The normalized membership value set generation module is used to initialize the decision vector, calculate the membership value of the target noise level and each decision value in the decision vector set through the Gaussian membership function, and then sum and normalize all membership values ​​to obtain the normalized membership value set. The fuzzy coefficient generation module is used to initialize the fuzzy variance vector and obtain the fuzzy coefficient by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set. An intermediate denoising signal generation module is used to substitute the ambiguity coefficients into a preset formula to denoise the original target signal and obtain an intermediate denoising signal. The target noise level update module is used to substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, the intermediate denoised level is updated to the target noise level and the module returns to the normalized membership value set generation module. The target valid signal determination module is used to further include, after the intermediate denoising signal generation module, the following: if the intermediate denoising level is less than or equal to a preset noise level, then output the intermediate denoising signal as the target valid signal.

[0015] Optionally, the pre-set training models include: Acquire training data and input the training data into a preset model for training to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; the preset model is a deep convolutional generative adversarial network (DCGAN) model. Update the preset model according to the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is judged to be trained successfully.

[0016] Optionally, the vibration feature generation module includes: The standard spatiotemporal matrix generation module is used to construct a spatiotemporal matrix based on all valid target signals, and to perform mean normalization on the spatiotemporal matrix to obtain a standard spatiotemporal matrix. The first convolutional feature generation module is used to substitute the standard spatiotemporal matrix into the group convolutional backbone network to obtain the first convolutional feature. The multi-scale feature extraction module is used to substitute the first convolutional feature into the multi-scale feature pyramid to obtain low-scale features, medium-scale features and high-scale features; A multi-scale feature convolution module is used to perform 1×1 convolution operations on the low-scale features, the medium-scale features, and the high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features, and convolutional high-scale features; The first splicing feature generation module is used to upsample the high-scale convolutional features and splice them with the mid-scale convolutional features to obtain the first splicing feature; The second splicing feature generation module is used to upsample the convolutional mid-scale features and splice them with the convolutional low-scale features to obtain the second splicing feature; The attention weight generation module is used to substitute the second concatenated feature into the attention mechanism to obtain the attention weight; The target fusion feature generation module is used to upsample the first spliced ​​feature and then fuse the first spliced ​​feature and the second spliced ​​feature according to the attention weight to obtain the target fusion feature; The vibration feature determination module is used to perform a 1×1 convolution on the target fused features and then input them into a fully connected layer to obtain vibration features.

[0017] The beneficial effects of this invention are: This invention proposes a vibration monitoring method for long-distance oil and gas pipelines based on DAS. By combining empirical mode decomposition and adaptive signal reconstruction techniques, it effectively solves the problems of interference signal superposition and parameter distortion faced by traditional pipeline vibration monitoring methods in complex environments. It can accurately separate and remove environmental noise from the original vibration signal, extract the real and effective vibration features, and then achieve real-time perception and type identification of the vibration status of the entire long-distance oil and gas pipeline with high precision and anti-interference through rapid comparison between the intelligent recognition model and the feature database. This significantly improves the early identification capability of safety hazards and reduces the risk of leakage and explosion caused by missed reports or misjudgments. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 A flowchart of a vibration monitoring method for long-distance oil and gas pipelines based on DAS provided for an embodiment of the present invention; Figure 2 A flowchart of a method for screening valid IMF components in an IMF component set, provided as an embodiment of the present invention; Figure 3 A flowchart of a method for noise removal of a target original signal provided in an embodiment of the present invention; Figure 4 This is a framework diagram of a DAS-based vibration monitoring system for long-distance oil and gas pipelines, provided for an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a vibration monitoring method for long-distance oil and gas pipelines based on DAS. See also... Figure 1 , Figure 1 A flowchart illustrating a vibration monitoring method for long-distance oil and gas pipelines based on DAS (Distributed Aspect-Oriented System) is provided for embodiments of the present invention. The method includes the following steps: S101, the distributed acoustic sensor host obtains the original signal set by collecting the original vibration signals of the oil and gas pipeline through a preset period; S102, Perform ensemble empirical mode decomposition on the original target signal to obtain the IMF component set; S103, select effective IMF components in the IMF component set through preset screening conditions to obtain the target IMF component set, and reconstruct the vibration signal based on the target IMF component set to obtain the target reconstructed signal; S104, Noise removal is performed on the original target signal based on the target reconstruction signal to obtain the effective target signal; S105, Substitute all effective signals of the target into the preset vibration identification model to obtain vibration characteristics, and substitute the vibration characteristics into the preset database for comparison to determine the vibration type; The target original signal is any one of the original signals in the set; the preset database stores the vibration characteristics corresponding to different vibration types.

[0023] The vibration monitoring method for long-distance oil and gas pipelines based on DAS provided in this invention firstly performs ensemble empirical mode decomposition on the original target signal, which can accurately separate and filter out environmental noise from the original vibration signal and extract the real and effective vibration signal. Then, through rapid comparison between the intelligent recognition model and the feature database, it realizes real-time perception and type identification of the vibration status of the entire long-distance oil and gas pipeline with high precision and anti-interference, thereby improving the efficiency of pipeline vibration monitoring.

[0024] In one implementation, by combining empirical mode decomposition and preset screening conditions, the original complex signal can be adaptively decomposed into components of different scales, and the effective components can be accurately screened for reconstruction. This process can effectively remove high-frequency or low-frequency noise generated by the complex environment along the pipeline (such as wind noise, vehicle vibration, etc.), thereby obtaining a high-fidelity target effective signal, fundamentally solving the problem of parameter distortion caused by interference signal superposition.

[0025] In one implementation, the original signal is first extracted into vibration features through front-end intelligent processing. When identifying the vibration features, the system no longer processes the massive waveform data, but directly matches and compares this concise feature fingerprint with various vibration feature libraries pre-stored in the database. This enables the system to quickly provide diagnostic results and improve the efficiency of vibration monitoring for long-distance oil and gas pipelines.

[0026] In one implementation, a distributed acoustic sensor host is installed in the long-distance oil and gas pipeline to transmit and receive signals; the preset period is determined by technicians; each original vibration signal of the oil and gas pipeline carries a transmission location tag, thereby determining the vibration type and vibration location corresponding to each target valid signal.

[0027] In one implementation, the process of reconstructing the vibration signal based on the target IMF component set to obtain the target reconstructed signal involves arranging the effective IMF components in the target IMF component set according to the original order of the ensemble empirical mode decomposition, and then adding all the effective IMF components one by one according to their corresponding sampling points to obtain the target reconstructed signal.

[0028] In one embodiment, see Figure 2 , Figure 2 A flowchart is provided for a method to filter valid IMF components in an IMF component set, including... S1031, calculate the mutual information value between the target preset IMF component and the original transmitted signal to obtain a measure of similarity; S1032, calculate the mutual information value between the target preset IMF component and the target original signal to obtain the baseline similarity; S1033, calculate the similarity correlation coefficient by measuring the difference between similarity and baseline similarity; S1034, perform max-min normalization on the similarity correlation coefficients corresponding to all IMF components to obtain the similarity coefficient set; S1035, After sorting the similarity coefficients in the similarity coefficient set in descending order, calculate the difference between adjacent similarity coefficients and determine whether the difference is less than the preset similarity coefficient threshold. S1036, If the difference is less than the preset similarity coefficient threshold, the IMF component corresponding to the smallest similarity coefficient in the difference is recorded as the effective IMF component. S1037, Obtain all valid IMF components to obtain the target IMF component set; Among them, the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensor host. In one implementation, the vibration signal of the oil and gas pipeline is a non-stationary, nonlinear signal, which is easily affected by noise in traditional time-domain and frequency-domain analysis. Mutual information (MI) is an indicator that measures the degree of interdependence between two random variables and can quantify the amount of information about one variable obtained through another) can quantify the "statistical dependence" between two signals, rather than just focusing on linear correlation. It can effectively capture the potential correlation between the IMF component and the original transmitted signal and avoid feature loss caused by signal nonlinearity.

[0029] In one implementation, by calculating the mutual information value between the target preset IMF component and the target original signal, the influence of pipeline background noise and the sensor system's own noise on the screening results can be offset, ensuring that the similarity correlation coefficient only reflects the signal differences caused by vibration characteristics.

[0030] In one implementation, the amplitude and energy of different IMF components may vary greatly, and directly comparing similarity correlation coefficients may lead to screening bias. After maximum-minimum normalization, all similarity coefficients are mapped to a unified interval ([0,1]), making IMF components of different magnitudes comparable and avoiding the misscreening of effective components due to amplitude differences.

[0031] In one implementation, after sorting the normalized similarity coefficients in descending order, the change in the difference between adjacent coefficients can reflect the boundary between effective and ineffective components: the difference in similarity coefficients for the first N components is small (indicating that they all contain vibration features), and the difference changes abruptly after the (N+1)th component (indicating that noise is the main component). This boundary is automatically captured by a preset similarity coefficient threshold (determined by technicians) without manual intervention. This avoids missing key feature components (such as low-frequency IMF components corresponding to weak leakage) and also avoids mis-screening noise components (such as IMF components corresponding to high-frequency electromagnetic interference). The selected target IMF component set retains only the components that are strongly correlated with vibration features. When extracting features from this set later, there is no need to perform complex noise suppression processing, reducing feature redundancy, improving the accuracy of vibration event identification, and reducing the false alarm rate (such as misjudging background noise as a fault) and the false alarm rate (such as missing weak vibrations due to noise masking).

[0032] In one embodiment, see Figure 3 , Figure 3 A flowchart of a method for noise removal of a target original signal provided in an embodiment of the present invention includes: Step 1: Substitute the target reconstructed signal into the preset model to obtain the target noise level; Step 2: Initialize the decision vector. Calculate the membership degree of the target noise level and each decision value in the decision vector set using the Gaussian membership function, and then sum and normalize all membership degree values ​​to obtain a normalized membership degree value set. Step 3: Initialize the fuzzy variance vector, and obtain the fuzzy coefficients by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set; Step 4: Substitute the ambiguity coefficients into the preset formula to denoise the original target signal to obtain the intermediate denoised signal; Step 5: Substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, update the intermediate denoised level to the target noise level and return to step 2. Step 6: After step 4, the following is also included: If the intermediate denoising level is less than or equal to the preset noise level, the intermediate denoising signal is output and recorded as the target valid signal.

[0033] In one implementation, the noise of oil and gas pipelines is uncertain, and traditional denoising methods (such as fixed threshold denoising and linear filtering) are difficult to adapt dynamically. Gaussian membership functions quantify the degree of matching between the target noise level and the decision value, transforming the fuzzy characteristics of the noise into a calculable normalized membership value (such as the 0~1 interval), avoiding denoising deviations caused by the fuzziness of the noise type. The fuzzy coefficients are obtained by weighted summation of the fuzzy variance vectors, realizing the dynamic correlation between noise intensity and denoising strength, and avoiding over-denoising or under-denoising.

[0034] In one implementation, the initial decision vector is defined as vector T=[0,1,2,3,4], with noise levels from high to low (0 is high noise, 4 is low noise); the fuzzy variance vector is M=[0.0001,0.0025,0.005,0.0075,0.01].

[0035] In one implementation, after each denoising step, an intermediate denoising level is calculated using a preset model and compared with a preset noise level (determined by technical personnel). If the level does not meet the standard, the target noise level is updated and iterated again. The denoising parameters (fuzzy coefficients) can be adaptively corrected until the noise residue is below the threshold, ensuring the purity of the target effective signal. This approach thoroughly removes noise while fully preserving the key features of the vibration event, providing a high-quality signal foundation for subsequent feature extraction and event recognition.

[0036] In one implementation, the preset formula is the expression for the reverse diffusion process in the diffusion model. ,in Let be the intermediate denoised signal obtained in the t-th iteration. This is the intermediate denoised signal obtained in the (t-1)th iteration (when it is the first loop, it is represented as the original target signal). The decay coefficient obtained in the (t-1)th iteration is a predefined parameter for forward diffusion (fixed before training), and its value ranges from (0,1). It is the product of all attenuation coefficients from the 1st to the (t-1th)th iteration. Let be the (t-1)th Gaussian noise, a random noise vector following a standard normal distribution, with dimension . The signals are consistent, Let be the fuzzy coefficient for the (t-1)th iteration, where t is greater than or equal to 2.

[0037] In one embodiment, the method includes training a pre-defined model: Acquire training data and input the training data into the preset model to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; the preset model is a deep convolutional generative adversarial network (DCGAN) model. Update the preset model based on the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is considered to have passed the training.

[0038] In one implementation, the training data includes pipeline vibration signals and real noise signals under different noise conditions, rather than data from a single noise scenario; it can cover all typical noises in actual pipeline operation; after model training, it can adapt to complex and changing working conditions, avoiding prediction failures caused by incomplete noise type coverage; the model's qualification is judged by the loss value between predicted noise and real noise (MAE loss), directly quantifying the noise prediction error of the model.

[0039] In one embodiment, substituting the target valid signal into a preset vibration identification model to obtain vibration characteristics includes: A spatiotemporal matrix is ​​constructed based on all valid signals from the targets, and the spatiotemporal matrix is ​​normalized by mean to obtain a standard spatiotemporal matrix. Substituting the standard spatiotemporal matrix into the group convolutional backbone network yields the first convolutional feature; Substituting the first convolutional feature into the multi-scale feature pyramid yields low-scale, medium-scale, and high-scale features; Perform 1×1 convolution operations on low-scale features, medium-scale features, and high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features, and convolutional high-scale features; The first concatenated feature is obtained by upsampling the high-scale features of the convolutional system and then concatenating them with the mid-scale features of the convolutional system. The convolutional mid-scale features are upsampled and then concatenated with the convolutional low-scale features to obtain the second concatenated feature. Substitute the second concatenation feature into the attention mechanism to obtain the attention weight; After upsampling the first spliced ​​feature, the first spliced ​​feature and the second spliced ​​feature are fused according to the attention weight to obtain the target fused feature; Vibration features are obtained by performing a 1×1 convolution on the target fusion features and then feeding them into a fully connected layer.

[0040] In one implementation, the vibration of oil and gas pipelines is a spatiotemporally coupled event. Constructing a spatiotemporal matrix can fuse the spatial location information of distributed DAS sensors with the time series information of the signals, avoiding feature loss caused by isolated analysis of signals at a single point or at a single moment. Mean normalization can offset the baseline drift caused by signal amplitude deviations at different monitoring points and changes in ambient temperature, ensuring that the spatiotemporal matrix data are on the same order of magnitude, ensuring the stability of subsequent convolution operations, and avoiding feature extraction deviations caused by uneven data distribution.

[0041] In one implementation, constructing a spatiotemporal matrix based on all target valid signals specifically involves: extracting target valid signals for a preset period (e.g., 1 second duration), sampling them at a 1 GHz rate, and discretizing them into 2000 time sampling points (corresponding to matrix rows) to reflect the dynamic changes of vibration over time; dividing the monitoring area into 1-meter intervals (e.g., 100 spatial nodes corresponding to a 100-meter pipeline), extracting vibration data at each node (corresponding to matrix columns) to reflect the spatial distribution of vibration, and constructing a spatiotemporal matrix based on the time sampling points and the vibration data at the corresponding locations.

[0042] In one implementation, traditional ordinary convolution performs global convolution on the entire spatiotemporal matrix, which is computationally intensive and prone to introducing redundant information; group convolution divides the input channel into multiple independent groups and performs convolution operations on each group; it adapts to the local propagation characteristics of pipeline vibration (the spatiotemporal characteristics of leakage vibration in the local area of ​​the pipeline are correlated, and global convolution is not required), and can accurately extract local spatiotemporal features (vibration propagation speed, local amplitude changes).

[0043] In one implementation, multi-scale features are extracted: low-scale features correspond to low-frequency vibrations (such as large-area pipeline leaks, low-frequency vibrations from third-party construction and excavation, which have long propagation distances and long durations); medium-scale features correspond to medium-frequency vibrations such as loose pipe joints; and high-scale features correspond to high-frequency vibrations such as micro-cracks in pipelines.

[0044] In one implementation, the first and second concatenated features are fused based on attention weights, which can enhance the expression of core features, solve the problem of excessive redundant information after multi-scale feature fusion leading to recognition confusion, and improve the accuracy of vibration event classification. Finally, after further simplifying the fused features through 1×1 convolution, the high-dimensional features are substituted into a fully connected layer to map the high-dimensional features into a low-dimensional structured vector to determine the vibration type.

[0045] Based on the same inventive concept, this invention also provides a DAS-based vibration monitoring system for long-distance oil and gas pipelines. See also Figure 4 , Figure 4 A framework diagram of a DAS-based vibration monitoring system for long-distance oil and gas pipelines, provided for embodiments of the present invention, includes: The raw signal set generation module is used by the distributed acoustic sensor host to collect and receive the original vibration signals of the oil and gas pipeline through a preset period to obtain the raw signal set. The IMF component set generation module is used to perform ensemble empirical mode decomposition on the target original signal to obtain the IMF component set; the target original signal is any one of the original signal sets; The target IMF component set generation module is used to filter valid IMF components in the IMF component set through preset filtering conditions to obtain the target IMF component set, and to reconstruct the vibration signal based on the target IMF component set to obtain the target reconstructed signal. The target effective signal generation module is used to remove noise from the original target signal based on the target reconstruction signal to obtain the target effective signal. The vibration feature generation module is used to input all valid signals of the target into a preset vibration recognition model to obtain vibration features, and to input the vibration features into a preset database for comparison to determine the vibration type; the preset database stores the vibration features corresponding to different vibration types.

[0046] The vibration monitoring system for long-distance oil and gas pipelines based on DAS provided in this invention first performs ensemble empirical mode decomposition on the original target signal, which can accurately separate and filter out environmental noise from the original vibration signal and extract the real and effective vibration signal. Then, through rapid comparison between the intelligent recognition model and the feature database, it realizes real-time perception and type identification of the vibration status of the entire long-distance oil and gas pipeline with high precision and anti-interference, thereby improving the efficiency of pipeline vibration monitoring.

[0047] In one embodiment, the target IMF component set generation module includes: The similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the original transmitted signal to obtain the similarity measurement; the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensor host; The baseline similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the target original signal to obtain the baseline similarity. The similarity correlation coefficient calculation module is used to calculate the similarity correlation coefficient by measuring the difference between the similarity and the baseline similarity. The normalization module is used to perform max-min normalization on the similarity correlation coefficients corresponding to all IMF components to obtain a set of similarity coefficients. The similarity coefficient sorting module is used to sort the similarity coefficients in the similarity coefficient set in descending order, calculate the difference between adjacent similarity coefficients, and determine whether the difference is less than a preset similarity coefficient threshold. The effective IMF component determination module is used to determine the IMF component corresponding to the smallest similarity coefficient among the differences if the difference is less than the preset similarity coefficient threshold. The target IMF component set determination module is used to obtain all valid IMF components to obtain the target IMF component set.

[0048] In one embodiment, the target valid signal generation module includes: The target noise level generation module is used to input the target reconstructed signal into a preset model to obtain the target noise level; The normalized membership value set generation module is used to initialize the decision vector. After calculating the membership value of the target noise level and each decision value in the decision vector set through the Gaussian membership function, the module sums and normalizes all membership values ​​to obtain the normalized membership value set. The fuzzy coefficient generation module is used to initialize the fuzzy variance vector and obtain the fuzzy coefficient by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set. The intermediate denoising signal generation module is used to denoise the target original signal by substituting the ambiguity coefficients into a preset formula to obtain an intermediate denoising signal; The target noise level update module is used to substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, the intermediate denoised level is updated to the target noise level and the module returns to the normalized membership value set generation module. The target valid signal determination module, which is used after the intermediate denoising signal generation module, further includes: if the intermediate denoising level is less than or equal to the preset noise level, then the intermediate denoising signal is output as the target valid signal.

[0049] In one embodiment, training a pre-defined model includes: Acquire training data and input the training data into the preset model to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; the preset model is a deep convolutional generative adversarial network (DCGAN) model. Update the preset model based on the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is considered to have passed the training.

[0050] In one embodiment, the vibration feature generation module includes: The standard spatiotemporal matrix generation module is used to construct a spatiotemporal matrix based on all valid target signals, and to perform mean normalization on the spatiotemporal matrix to obtain the standard spatiotemporal matrix. The first convolutional feature generation module is used to substitute the standard spatiotemporal matrix into the group convolutional backbone network to obtain the first convolutional feature. The multi-scale feature extraction module is used to substitute the first convolutional features into the multi-scale feature pyramid to obtain low-scale features, medium-scale features, and high-scale features. The multi-scale feature convolution module is used to perform 1×1 convolution operations on low-scale features, medium-scale features and high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features and convolutional high-scale features; The first concatenation feature generation module is used to upsample the high-scale convolutional features and concatenate them with the mid-scale convolutional features to obtain the first concatenation feature. The second concatenation feature generation module is used to upsample the convolutional mid-scale features and concatenate them with the convolutional low-scale features to obtain the second concatenation feature. The attention weight generation module is used to input the second concatenation feature into the attention mechanism to obtain the attention weight; The target fusion feature generation module is used to upsample the first concatenated feature and then fuse the first concatenated feature and the second concatenated feature according to the attention weight to obtain the target fusion feature; The vibration feature determination module is used to perform a 1×1 convolution on the target fusion features and then input them into a fully connected layer to obtain vibration features.

[0051] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A vibration monitoring method for long-distance oil and gas pipelines based on DAS, characterized in that, The method includes: The distributed acoustic sensor host collects the original vibration signals of the oil and gas pipeline at a preset period to obtain the original signal set; The target original signal is subjected to ensemble empirical mode decomposition to obtain an IMF component set; the target original signal is any one of the original signal sets; The target IMF component set is obtained by filtering effective IMF components in the IMF component set using preset filtering conditions, and the vibration signal is reconstructed based on the target IMF component set to obtain the target reconstructed signal. The original target signal is noise-removed based on the reconstructed target signal to obtain the effective target signal. All valid signals from the target are substituted into a preset vibration identification model to obtain vibration characteristics. The vibration characteristics are then substituted into a preset database for comparison to determine the vibration type. The preset database stores the vibration characteristics corresponding to different vibration types.

2. The vibration monitoring method for long-distance oil and gas pipelines based on DAS according to claim 1, characterized in that, The target IMF component set is obtained by filtering valid IMF components from the IMF component set using preset filtering conditions, including: The mutual information value between the target preset IMF component and the original transmitted signal is calculated to obtain a similarity measure; the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensing host. The baseline similarity is obtained by calculating the mutual information value between the target preset IMF component and the target original signal; The similarity correlation coefficient is obtained by calculating the difference between the measure of similarity and the baseline similarity; The similarity coefficient set is obtained by performing max-min normalization on the similarity correlation coefficients corresponding to all IMF components. After sorting the similarity coefficients in the set of similarity coefficients in descending order, the difference between adjacent similarity coefficients is calculated, and it is determined whether the difference is less than a preset similarity coefficient threshold. If the difference is less than the preset similarity coefficient threshold, the IMF component corresponding to the smallest similarity coefficient in the difference will be recorded as the effective IMF component. Obtain all valid IMF components to obtain the target IMF component set.

3. The vibration monitoring method for long-distance oil and gas pipelines based on DAS according to claim 1, characterized in that, The process of removing noise from the original target signal based on the reconstructed target signal to obtain the effective target signal includes: Step 1: Substitute the target reconstructed signal into a preset model to obtain the target noise level; Step 2: Initialize the decision vector. Calculate the membership degree of the target noise level with each decision value in the decision vector set using the Gaussian membership function, and then sum and normalize all membership degree values ​​to obtain a normalized membership degree value set. Step 3: Initialize the fuzzy variance vector, and obtain the fuzzy coefficients by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set; Step 4: Substitute the ambiguity coefficients into a preset formula to denoise the original target signal to obtain an intermediate denoised signal; Step 5: Substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, update the intermediate denoised level to the target noise level and return to step 2. Step 6: After step 4, the following is also included: if the intermediate denoising level is less than or equal to the preset noise level, the intermediate denoising signal is output and recorded as the target valid signal.

4. The vibration monitoring method for long-distance oil and gas pipelines based on DAS according to claim 3, characterized in that, The method includes training a pre-defined model: Acquire training data and input the training data into a preset model for training to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; The preset model is the Deep Convolutional Generative Adversarial Network (DCGAN) model. Update the preset model according to the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is judged to be trained successfully.

5. The vibration monitoring method for long-distance oil and gas pipelines based on DAS according to claim 1, characterized in that, Substituting the target valid signal into a preset vibration identification model yields vibration characteristics including: A spatiotemporal matrix is ​​constructed based on all valid signals from the target, and the spatiotemporal matrix is ​​then normalized by mean to obtain a standard spatiotemporal matrix. Substituting the standard spatiotemporal matrix into the group convolutional backbone network yields the first convolutional feature; Substituting the first convolutional feature into the multi-scale feature pyramid yields low-scale features, medium-scale features, and high-scale features. Perform 1×1 convolution operations on the low-scale features, the medium-scale features, and the high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features, and convolutional high-scale features; The high-scale convolutional features are upsampled and then concatenated with the mid-scale convolutional features to obtain the first concatenated feature. The convolutional mid-scale features are upsampled and then concatenated with the convolutional low-scale features to obtain the second concatenated feature. Substitute the second splicing feature into the attention mechanism to obtain the attention weight; After upsampling the first splicing feature, the first splicing feature and the second splicing feature are fused according to the attention weight to obtain the target fused feature; The vibration features are obtained by performing a 1×1 convolution on the target fusion features and then feeding them into a fully connected layer.

6. A vibration monitoring system for long-distance oil and gas pipelines based on DAS, characterized in that, The system includes: The raw signal set generation module is used by the distributed acoustic sensor host to collect and receive the original vibration signals of the oil and gas pipeline through a preset period to obtain the raw signal set. The IMF component set generation module is used to perform ensemble empirical mode decomposition on the target original signal to obtain the IMF component set; the target original signal is any one of the original signal sets; The target IMF component set generation module is used to filter valid IMF components in the IMF component set through preset filtering conditions to obtain a target IMF component set, and to reconstruct the vibration signal based on the target IMF component set to obtain a target reconstructed signal. The target valid signal generation module is used to remove noise from the original target signal based on the target reconstructed signal to obtain the target valid signal; The vibration feature generation module is used to input all target valid signals into a preset vibration recognition model to obtain vibration features, and to input the vibration features into a preset database for comparison to determine the vibration type; the preset database stores vibration features corresponding to different vibration types.

7. The vibration monitoring system for long-distance oil and gas pipelines based on DAS according to claim 6, characterized in that, The target IMF component set generation module includes: The similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the original transmitted signal to obtain the similarity measurement; the target preset IMF component is any one of the IMF component sets; the original transmitted signal is the signal transmitted by the distributed acoustic sensing host. The baseline similarity calculation module is used to calculate the mutual information value between the target preset IMF component and the target original signal to obtain the baseline similarity. The similarity correlation coefficient calculation module is used to calculate the difference between the measure of similarity and the baseline similarity to obtain the similarity correlation coefficient; The normalization module is used to perform max-min normalization on the similarity correlation coefficients corresponding to all IMF components to obtain a set of similarity coefficients. The similarity coefficient sorting module is used to sort the similarity coefficients in the similarity coefficient set in descending order, calculate the difference between adjacent similarity coefficients, and determine whether the difference is less than a preset similarity coefficient threshold. The effective IMF component determination module is used to determine the IMF component corresponding to the smallest similarity coefficient among the differences if the difference is less than the preset similarity coefficient threshold. The target IMF component set determination module is used to obtain all valid IMF components to obtain the target IMF component set.

8. A vibration monitoring system for long-distance oil and gas pipelines based on DAS according to claim 6, characterized in that, The target valid signal generation module includes: The target noise level generation module is used to substitute the target reconstructed signal into a preset model to obtain the target noise level; The normalized membership value set generation module is used to initialize the decision vector, calculate the membership value of the target noise level and each decision value in the decision vector set through the Gaussian membership function, and then sum and normalize all membership values ​​to obtain the normalized membership value set. The fuzzy coefficient generation module is used to initialize the fuzzy variance vector and obtain the fuzzy coefficient by weighted summation of the values ​​in the fuzzy variance vector according to the normalized membership value set. An intermediate denoising signal generation module is used to substitute the ambiguity coefficients into a preset formula to denoise the original target signal and obtain an intermediate denoising signal. The target noise level update module is used to substitute the intermediate denoised signal into the preset model to obtain the intermediate denoised level. If the intermediate denoised level is greater than the preset noise level, the intermediate denoised level is updated to the target noise level and the module returns to the normalized membership value set generation module. The target valid signal determination module is used to further include, after the intermediate denoising signal generation module, the following: if the intermediate denoising level is less than or equal to a preset noise level, then output the intermediate denoising signal as the target valid signal.

9. A vibration monitoring system for long-distance oil and gas pipelines based on DAS according to claim 8, characterized in that, The pre-set training model includes: Acquire training data and input the training data into a preset model for training to obtain model update parameters; the training data includes vibration signals of oil and gas pipelines collected under different noise conditions and real noise signals; the preset model is a deep convolutional generative adversarial network (DCGAN) model. Update the preset model according to the model update parameters; Vibration signals of oil and gas pipelines collected under different noise conditions are substituted into the updated preset model to obtain the predicted noise. The loss value is calculated based on the predicted noise and the actual noise. If the loss value is less than the preset value, the model is judged to be trained successfully.

10. A vibration monitoring system for long-distance oil and gas pipelines based on DAS according to claim 6, characterized in that, The vibration feature generation module includes: The standard spatiotemporal matrix generation module is used to construct a spatiotemporal matrix based on all valid target signals, and to perform mean normalization on the spatiotemporal matrix to obtain a standard spatiotemporal matrix. The first convolutional feature generation module is used to substitute the standard spatiotemporal matrix into the group convolutional backbone network to obtain the first convolutional feature. The multi-scale feature extraction module is used to substitute the first convolutional feature into the multi-scale feature pyramid to obtain low-scale features, medium-scale features and high-scale features; A multi-scale feature convolution module is used to perform 1×1 convolution operations on the low-scale features, the medium-scale features, and the high-scale features respectively to obtain convolutional low-scale features, convolutional medium-scale features, and convolutional high-scale features; The first splicing feature generation module is used to upsample the high-scale convolutional features and splice them with the mid-scale convolutional features to obtain the first splicing feature; The second splicing feature generation module is used to upsample the convolutional mid-scale features and splice them with the convolutional low-scale features to obtain the second splicing feature; The attention weight generation module is used to substitute the second concatenated feature into the attention mechanism to obtain the attention weight; The target fusion feature generation module is used to upsample the first spliced ​​feature and then fuse the first spliced ​​feature and the second spliced ​​feature according to the attention weight to obtain the target fusion feature; The vibration feature determination module is used to perform a 1×1 convolution on the target fused features and then input them into a fully connected layer to obtain vibration features.

Images