Multi-working-condition gas pipeline leakage step-by-step identification method and system based on dual-channel sound wave feature decoupling

Abstract

The invention discloses a multi-working-condition gas pipeline leakage step-by-step identification method and system based on dual-channel sound wave feature decoupling, and relates to the technical field of pipeline leakage detection.The method comprises the steps that the leakage identification process is decoupled into four stages of leakage detection, working condition identification, aperture identification and pressure identification, a YOLOv8 model is adopted to process an image classification task, and a detection result is obtained; comprising leakage identification, working condition identification and aperture identification; a CNN model is adopted to process numerical data, and effective distinguishing and recognition of pipeline pressure are achieved. Whether leakage occurs or not is judged through the CCF peak value, the leakage working condition and the leakage aperture are further determined according to the CPSD image, feature vectors are constructed based on the amplitude, variance and energy parameters of leakage signals, and leakage pressure recognition is achieved. The decoupling recognition method provided by the invention is higher in recognition accuracy, better in real-time performance, lower in computer resource requirement, better in comprehensive performance in gas pipeline leakage monitoring and leakage recognition, and capable of quickly and effectively recognizing leakage.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline leakage detection, and more particularly to a multi-condition gas transmission pipeline leakage step-by-step identification method and system based on dual-channel acoustic wave feature decoupling. Background Technique

[0002] Currently, the mainstream gas transmission pipeline detection methods include time domain reflectometry, negative pressure wave method, transient wave method, acoustic wave method, etc. Among them, the acoustic wave method uses the propagation characteristics of acoustic waves to detect leakage, and has the advantages of non-invasiveness, high sensitivity, and real-time detection, and has been widely used. The leakage acoustic wave signal contains rich information related to the leakage source. Based on the leakage acoustic wave characteristics, the pipeline operation state (leakage or non-leakage) can be judged, and further the leakage amount, the number of leakage sources, and the leakage source position can be estimated. However, due to the complexity of the actual pipeline service environment, the leakage acoustic wave characteristics are easily affected by factors such as noise interference, signal propagation medium type, and propagation path, resulting in difficult feature extraction, which in turn reduces the leakage identification accuracy and increases the risk of missed reports and false alarms. Based on this, feature selection and optimization play a key role in pipeline leakage detection and identification.

[0003] The existing pipeline leakage identification methods have achieved good identification accuracy, but there are still the following problems in actual engineering applications: 1) Unstable quantitative features: The pipeline laying method, sensor installation position, operating pressure, leakage aperture, sensor type, etc. will all affect the leakage acoustic wave characteristics. Most of the existing leakage acoustic wave signal identification methods are based on single conditions or specific operating conditions, and the quantitative feature indicators used will show uncontrollable changes under the influence of multi-factor coupling under complex operating conditions. This change not only increases the difficulty of feature analysis, but also makes it difficult to effectively decouple different influencing factors. 2) High algorithm complexity and insufficient real-time performance: Under complex operating conditions, in order to ensure the identification accuracy, a high-dimensional feature vector needs to be constructed. As the introduction of non-linear mapping increases, the neural network needs to deepen the network layer to better mine the high-order features in the leakage acoustic wave signal. However, this approach will lead to an increase in the computational complexity of the algorithm, and face problems of high hardware requirements and insufficient real-time response in actual applications. 3) Insufficient interpretability of the model: The leakage identification method based on neural network belongs to an "end-to-end" problem, and the decision-making process usually lacks interpretability, and the rationality and reliability of the output are difficult to verify. This "black box" characteristic limits the popularization and application of the method.

[0004] Therefore, how to propose a multi-condition gas transmission pipeline leakage step-by-step identification method and system based on dual-channel acoustic wave feature decoupling, which decouples the leakage identification process into four stages: leakage detection - condition identification - aperture identification - pressure identification, and optimizes the low identification accuracy, high computational complexity, and poor real-time performance of traditional non-step-by-step identification methods through decoupled identification is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a multi - condition gas pipeline leakage step - by - step identification method and system based on dual - channel acoustic wave feature decoupling. It uses the CCF peak value to judge whether leakage occurs, and then further determines the leakage condition and leakage aperture according to the CPSD image. Finally, a feature vector is constructed based on the effective amplitude, variance, and energy parameters of the leakage signal to achieve leakage pressure identification. Compared with traditional non - step - by - step identification methods, the identification accuracy of the proposed method is significantly improved, the computational complexity is greatly reduced, and the real - time performance is better. To achieve the above - mentioned purpose, the present invention adopts the following technical solutions:

[0006] A multi - condition gas pipeline leakage step - by - step identification method based on dual - channel acoustic wave feature decoupling, comprising:

[0007] Collect dual - channel signals of pipeline leakage vibration acoustic waves, and extract amplitude, variance, and energy parameters;

[0008] Perform cross - correlation analysis on the dual - channel signals using the cross - correlation coefficient function to judge whether leakage occurs;

[0009] After confirming that leakage occurs, use the cross - power spectral density CPSD for spectral feature analysis to construct a logarithmic amplitude CPSD image, and perform leakage condition identification based on the logarithmic amplitude CPSD image;

[0010] Convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and perform leakage aperture identification based on the linear amplitude CPSD image;

[0011] After leakage aperture identification, construct a three - dimensional feature vector according to the amplitude, variance, and energy parameters, and input it into the pressure identification model for leakage pressure identification.

[0012] Optionally, the dual - channel signals of the leakage vibration acoustic waves include:

[0013]

[0014] Wherein, S(t) is the leakage source signal, S1(t) and S2(t) are respectively the dual - channel noisy signals collected by the sensors, N1(t) and N2(t) are random noises, τ is the time delay, and θ is the attenuation factor.

[0015] Optionally, only the leakage signals are correlated and the noises are not correlated. When there is no fixed strong interference around the pipeline, only the leakage signals S(t) and θS(t - τ) are correlated in the two - channel sensor signals, while the random noises N1(t), N2(t) are not correlated with S1(t), θS(t - τ), and N1(t) and N2(t) are not correlated with each other.

[0016] Optionally, the cross - correlation analysis of the dual - channel signal using the cross - correlation coefficient function includes:

[0017] Perform signal pre - processing to make the sampling rates of S1(t) and S2(t) the same and the signal lengths the same;

[0018] Calculate the CCF values of S1(t) and S2(t) at different time delays τ;

[0019] The higher the peak value of CCF(τ), the stronger the signal correlation. When the input of the cross - correlation coefficient function CCF is the leakage signal, the correlation coefficient peak is obtained at the delay τ between the two signals; when the input signal is random noise, there is no peak in the cross - correlation coefficient function CCF.

[0020] Optionally, after confirming the occurrence of leakage, using the cross - power spectral density CPSD for spectral feature analysis to construct a logarithmic amplitude CPSD image, and the leakage condition identification based on the logarithmic amplitude CPSD image includes:

[0021] Perform fast Fourier transforms on the leakage signals S1(t) and S2(t) respectively to convert them into frequency - domain signals

[0022]

[0023] where F(*) represents the Fourier transform; f is the frequency variable;

[0024] Perform the calculation of the logarithmic amplitude CPSD image:

[0025]

[0026] where, CPSD log represents the logarithmic amplitude CPSD value, * represents the complex conjugate operation, || represents the modulus operation, and T is the total duration of the signal;

[0027] Input the logarithmic amplitude CPSD image into the leakage condition identification model based on YOLOv8 for leakage condition identification.

[0028] Optionally, the conversion of the logarithmic amplitude CPSD image to a linear amplitude CPSD image includes:

[0029] Convert the logarithmic amplitude CPSD image to a linear amplitude CPSD image through antilogarithmic operation:

[0030]

[0031] where, CPSD linear represents the linear amplitude CPSD value.

[0032] Optionally, the identification of the leakage aperture based on the linear amplitude CPSD image includes: inputting the linear amplitude CPSD image into a leakage aperture identification model based on YOLOv8 for leakage aperture identification.

[0033] Optionally, the construction of a three-dimensional feature vector according to the amplitude, variance, and energy parameters and inputting it into a pressure identification model for leakage pressure identification includes:

[0034] Set a fixed time window, divide the leakage signal S1(t) into M time segments, and use the data within each time window as a sample;

[0035] Calculate the effective amplitude, variance, and energy indicators for each sample:

[0036]

[0037] Where RMS i 、V i 、E i respectively represent the effective amplitude, variance, and energy of the i-th sample, 1 ≤ i ≤ M; x(n) is the signal amplitude, n is the n-th sampling point within this time window, μ is the mean value of the signal within this time window, and N is the number of data points within each time window;

[0038] Obtain an M×3-dimensional feature matrix F:

[0039]

[0040] Input F into a pressure identification model based on CNN for classification to identify the leakage conditions under different pressure states.

[0041] Optionally, a multi-condition gas transmission pipeline leakage step-by-step identification system based on dual-channel acoustic wave feature decoupling includes:

[0042] Acquisition module: used to acquire the dual-channel signals of pipeline leakage vibration acoustic waves;

[0043] Extraction module: used to extract amplitude, variance, and energy parameters;

[0044] Cross-correlation analysis module: used to perform cross-correlation analysis on the dual-channel signals using the cross-correlation coefficient function to determine whether leakage has occurred;

[0045] Leakage condition identification module: used to perform spectral feature analysis using the cross-power spectral density CPSD to construct a logarithmic amplitude CPSD image after confirming that leakage has occurred, and perform leakage condition identification based on the logarithmic amplitude CPSD image;

[0046] Leakage aperture identification module: used to convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and identify the leakage aperture based on the linear amplitude CPSD image;

[0047] Leakage pressure identification module: used to construct a three-dimensional feature vector based on the amplitude, variance, and energy parameters after the leakage aperture is identified, and input it into the pressure identification model to identify the leakage pressure.

[0048] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a multi-condition gas transmission pipeline leakage step-by-step identification method and system based on dual-channel acoustic wave feature decoupling, and has the following beneficial effects:

[0049] The present invention proposes a multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling, including: collecting dual-channel signals of pipeline leakage vibration acoustic waves, and extracting amplitude, variance, and energy parameters; using the cross-correlation coefficient function to perform cross-correlation analysis on the dual-channel signals to judge whether leakage occurs; after confirming that leakage occurs, using the cross-power spectral density CPSD to perform spectral feature analysis to construct a logarithmic amplitude CPSD image, and performing leakage condition identification based on the logarithmic amplitude CPSD image; converting the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and performing leakage aperture identification based on the linear amplitude CPSD image; after the leakage aperture is identified, constructing a three-dimensional feature vector based on the amplitude, variance, and energy parameters, and inputting it into the pressure identification model to identify the leakage pressure. The present invention decouples the leakage identification process into four stages: "leakage detection - condition identification - aperture identification - pressure identification". Two classification models, YOLOv8 and CNN, are used. Among them, the YOLOv8 model is used to process image classification tasks, including leakage identification, condition identification, and aperture identification; CNN is used to process numerical data to effectively distinguish and identify pipeline pressure. The CCF peak value is used to judge whether leakage occurs. After confirming that leakage occurs, the leakage condition and leakage aperture are further determined according to the CPSD image. Finally, a feature vector is constructed based on the amplitude, variance, and energy parameters of the leakage signal to achieve leakage pressure identification. Compared with the traditional non-step-by-step identification method for gas transmission pipeline leakage, the decoupled identification method proposed by the present invention has higher identification accuracy, better real-time performance, and lower requirements for computer resources. The decoupled identification method proposed by the present invention has better comprehensive performance in gas transmission pipeline leakage monitoring and leakage identification, and can quickly and effectively identify leakage. Description of the Drawings

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0051] Figure 1 It is a schematic flow chart of a multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling provided by the present invention.

[0052] Figure 2 It is a PSD diagram of the leakage signal received by the sensor provided by the present invention.

[0053] Figure 3(a) is the leakage CCF result diagram provided by the present invention.

[0054] Figure 3(b) is the non-leakage CCF result diagram provided by the present invention.

[0055] Figure 4 It is the logarithmic amplitude CPSD diagram provided by the present invention.

[0056] Figure 5 It is the linear amplitude CPSD diagram provided by the present invention.

[0057] Figure 6 It is the normalized characteristic radar diagram provided by the present invention.

[0058] Figure 7 It is the variation trend diagram of the method identification performance with the input signal SNR provided by the present invention. Specific embodiments

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0060] The embodiments of the present invention disclose a multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling, including:

[0061] Collect the dual-channel signals of the pipeline leakage vibration acoustic wave, and extract the amplitude, variance, and energy parameters;

[0062] Use the cross-correlation coefficient function to perform cross-correlation analysis on the dual-channel signals to determine whether leakage has occurred;

[0063] After confirming the occurrence of leakage, use the cross-power spectral density (CPSD) to analyze the spectral characteristics and construct a logarithmic amplitude CPSD image, and identify the leakage condition based on the logarithmic amplitude CPSD image;

[0064] Convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and identify the leakage aperture based on the linear amplitude CPSD image;

[0065] After identifying the leakage aperture, construct a three-dimensional feature vector according to the amplitude, variance and energy parameters, and input it into the pressure identification model to identify the leakage pressure.

[0066] In a specific embodiment, a multi-condition leakage step-by-step identification method for gas transmission pipelines based on dual-channel acoustic wave feature decoupling is as Figure 1 shown, and the specific implementation steps include:

[0067] Step 1: Data acquisition. Place two sensors on the ground above the leakage pipeline or on the pipeline wall of the leakage pipeline to pick up the pipeline leakage vibration acoustic wave signal;

[0068] Step 2: Leakage detection. Use the cross-correlation coefficient function (CCF) to analyze the cross-correlation of the signals S1(t) and S2(t) collected by the two sensors, and judge whether leakage has occurred according to the peak value of the CCF;

[0069] Step 3: Leakage condition identification. After confirming the occurrence of leakage, use the cross-power spectral density (CPSD) to analyze the spectral characteristics of S1(t) and S2(t), and determine the leakage condition based on the logarithmic amplitude CPSD image;

[0070] Step 4: Leakage aperture identification. Convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and determine the leakage aperture according to the linear amplitude CPSD image;

[0071] Step 5: Leakage pressure identification. On the premise of knowing the leakage condition and leakage aperture, extract the effective amplitude, variance and energy parameters from the leakage signal to construct a three-dimensional feature vector. The obtained normalized feature radar chart is as Figure 6 shown; realize leakage pressure identification based on the feature vector.

[0072] Further, in the first step, the mathematical models of the signals S1(t) and S2(t) collected by the two sensors are as follows:

[0073]

[0074] Where \(S(t)\) is the leakage signal, \(S1(t)\) and \(S2(t)\) are the noisy signals received by sensors 1 and 2, \(N1(t)\) and \(N2(t)\) are random noises, \(\tau\) is the time delay, and \(\theta\) is the attenuation factor.

[0075] Furthermore, the signals \(S1(t)\) and \(S2(t)\) collected by the sensors have the following characteristics: only the leakage signal is correlated while the noise is not correlated. That is, when there is no fixed strong interference around the pipeline, only the leakage signals \(S(t)\) and \(\theta S(t - \tau)\) are correlated among the two-channel sensor signals, while the random noises \(N1(t)\) and \(N2(t)\) are not correlated with \(S1(t)\) and \(\theta S(t - \tau)\), and \(N1(t)\) and \(N2(t)\) are not correlated with each other. The power spectral density (PSD) of the leakage acoustic signals \(S1(t)\) and \(S2(t)\) collected by the two-channel sensors is as Figure 2 shown.

[0076] Furthermore, the specific process of performing cross-correlation analysis on the signals \(S1(t)\) and \(S2(t)\) collected by the two-channel sensors using the cross-correlation coefficient function CCF in step 2 is as follows:

[0077] 1) Signal preprocessing: Ensure that the sampling rates of \(S1(t)\) and \(S2(t)\) are the same and the signal lengths are the same;

[0078] 2) Calculate the CCF values of \(S1(t)\) and \(S2(t)\) at different time delays \(\tau\). The CCF calculation formula is as follows:

[0079]

[0080] 3) Signal correlation analysis: The higher the peak value of CCF(\(\tau\)), the stronger the signal correlation. When the input of CCF is the leakage signal, the peak value of the correlation coefficient can be obtained at the delay \(\tau\) between the two signals; when the input signal is random noise, there is no obvious peak in CCF. The obtained CCF images are shown in Figures 3(a) and 3(b);

[0081] 4) Through the above process, identify whether the input signal is a leakage signal, so as to achieve the purpose of leakage detection.

[0082] Furthermore, in step 3, the cross-power spectral density CPSD quantifies the linear correlation between two signals in the frequency domain, provides the energy information of the common frequency components of the two signals, can enhance the signal PSD and weaken the noise PSD at the same time, and realizes spectrum feature enhancement. Under the same working conditions, the changes in the leakage aperture and pipeline pressure will affect the spectrum details of the acoustic signal (such as the amplitude sizes of the high-frequency part and the low-frequency part, and the high-amplitude frequency response positions), but cannot fundamentally change the generation mechanism and propagation characteristics of the leakage acoustic wave. Therefore, the cross-power spectral density CPSD under the same working conditions remains consistent in the overall distribution.

[0083] Further, the specific steps for obtaining the logarithmic amplitude CPSD image in step 3 are as follows:

[0084] 1) Perform the Fast Fourier Transform (FFT) on the leakage signals S1(t) and S2(t) respectively to convert them into frequency-domain signals The FFT calculation formula is as follows:

[0085]

[0086] In the formula, F() represents the Fourier transform; f is the frequency variable;

[0087] 2) Calculate the logarithmic amplitude CPSD image of the signals to obtain the correlation between the two signals in the frequency domain. The calculation formula for the logarithmic amplitude CPSD image is as follows:

[0088]

[0089] In the formula, CPSD log represents the logarithmic amplitude CPSD value, with the unit of decibel (dB); "*" represents the complex conjugate operation; "||" represents the modulus operation; T is the total duration of the signal, with the unit of second (s);

[0090] 3) Generate the logarithmic amplitude CPSD image. Based on the differences in the logarithmic amplitude CPSD images of the signals under different working conditions, identify the leakage conditions. The obtained logarithmic amplitude CPSD image is as Figure 4 shown.

[0091] Further, the logarithmic amplitude CPSD image reduces the influence of the high-amplitude frequency response on the signal spectrum by amplifying the low-amplitude frequency response and reducing the high-amplitude frequency response. Even if the frequency and magnitude of the high-amplitude frequency response change, it is difficult to be reflected in the logarithmic amplitude CPSD image.

[0092] Further, in step 4, an antilogarithm operation is required to convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image. The linear amplitude CPSD image is as Figure 5 shown, and the specific calculation formula is as follows:

[0093]

[0094] In the formula, CPSD linear represents the linear amplitude CPSD value, with the unit of V 2 / Hz or the corresponding unit.

[0095] Furthermore, the linear amplitude CPSD image can magnify the detailed changes in the high-amplitude frequency response, more sensitively capture the frequency response differences caused by aperture and pressure changes, especially the changes in the high-amplitude frequency response part.

[0096] Furthermore, in step 5, the constructing a three-dimensional feature vector according to the amplitude, variance, and energy parameters and inputting it into the pressure recognition model for leak pressure recognition includes:

[0097] Set a fixed time window, divide the leakage signal S1(t) into M time segments, and use the data within each time window as a sample;

[0098] Calculate the effective amplitude, variance, and energy indicators for each sample:

[0099]

[0100] In the formula, RMS i , V i , E i respectively represent the effective amplitude, variance, and energy of the i-th sample, 1 ≤ i ≤ M; x(n) is the signal amplitude, n is the n-th sampling point within this time window, μ is the mean value of the signal within this time window, and N is the number of data points within each time window;

[0101] Obtain the M×3-dimensional feature matrix F:

[0102]

[0103] Input F into the pressure recognition model based on CNN for classification to identify the leakage conditions under different pressure states.

[0104] In a specific embodiment, a multi-condition gas transmission pipeline leakage step-by-step recognition system based on dual-channel acoustic wave feature decoupling includes:

[0105] Acquisition module: used to acquire the dual-channel signals of pipeline leakage vibration acoustic waves;

[0106] Extraction module: used to extract amplitude, variance, and energy parameters;

[0107] Cross-correlation analysis module: used to perform cross-correlation analysis on the dual-channel signals using the cross-correlation coefficient function to determine whether leakage has occurred;

[0108] Leakage condition recognition module: used to perform spectrum feature analysis using the cross-power spectral density CPSD to construct a logarithmic amplitude CPSD image after confirming that leakage has occurred, and perform leakage condition recognition based on the logarithmic amplitude CPSD image;

[0109] Leakage aperture recognition module: used to convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and perform leakage aperture recognition based on the linear amplitude CPSD image;

[0110] Leakage pressure recognition module: after the leakage aperture is recognized, it is used to construct a three-dimensional feature vector based on the effective amplitude, variance, and energy parameters, and input it into the pressure recognition model to perform leakage pressure recognition.

[0111] Furthermore, the leakage recognition process is decoupled into four stages: "leakage detection - operating condition recognition - aperture recognition - pressure recognition". Two classification models, YOLOv8 and CNN, are used. Among them, the YOLOv8 model is used to process image classification tasks, including leakage recognition, operating condition recognition, and aperture recognition. The YOLOv8 can be YOLOv8n; CNN is used to process numerical data, identify the constructed three-dimensional feature vector for pipeline pressure recognition, and effectively distinguish and identify pipeline pressure. The recognition results of each classification task are shown in Table 1.

[0112] Table 1 Recognition results of each classification task

[0113]

[0114] In the specific implementation, the performance of the proposed denoising method is evaluated under different signal-to-noise ratios (SNR) of the input noisy signal. The SNR change range is set to -4dB to 4dB, and the change step is 2dB. The change trend of the performance of the proposed decoupled recognition method with the SNR of the input signal is as Figure 7 shown, where Figure 7 (a) in is the change trend of the leakage detection performance with the SNR of the input signal, Figure 7 (b) in is the change trend of the leakage operating condition recognition performance with the SNR of the input signal, Figure 7 (c) in is the change trend of the leakage aperture recognition performance with the SNR of the input signal, Figure 7 (d) in is the change trend of the leakage pressure recognition performance with the SNR of the input signal.

[0115] To prove the advantages of the decoupled recognition method proposed in the present invention, it is compared with the currently common non-stepwise recognition methods for gas pipeline leakage. It should be noted that the method proposed in this embodiment decouples the leakage recognition task into 4 subtasks: leakage detection - leakage operating condition recognition - leakage aperture recognition - pipeline pressure recognition. The stepwise recognition method proposed in the present invention performs pipeline pressure recognition after completing the first 3 subtasks, while the one-step recognition method directly performs pipeline pressure recognition. Therefore, the pressure recognition results of different methods are directly compared here.

[0116] The recognition results of different methods are shown in Table 2. Among them, the convolutional neural network recognition method based on the fusion of time-domain features (TF) and frequency-domain features (FF), namely the TF+FF-CNN method, combines time-domain features and frequency-domain features to construct feature vectors (time-domain features include peak value, effective amplitude, variance, peak factor, kurtosis, gap factor, shape factor, and frequency-domain features include peak frequency, center frequency, and mean square frequency). The convolutional neural network recognition method based on multi-modal entropy features (MEF), namely the MEF-CNN method, extracts 7 kinds of entropy from the pipeline leakage signal to reflect the signal characteristics, including permutation entropy, envelope entropy, approximate entropy, fuzzy entropy, energy entropy, sample entropy, and dispersion entropy. The convolutional neural network recognition method based on multi-feature fusion, namely the TF+FF+EF-CNN method, in addition to the time-domain and frequency-domain features used in the TF+FF-CNN method, also selects two entropy indexes of permutation entropy and sample entropy to describe the irregularity and complexity of the pipeline signal. The above features are used to construct feature vectors and input into CNN for classification. The convolutional neural network recognition method based on variational mode decomposition (VMD) and multi-feature fusion, namely the VMD-TF+FF+EF-CNN method, uses VMD to decompose the leakage signal, uses the correlation coefficient between the original signal and the decomposed IMF components to screen effective modes, and the feature selection is the same as that of the TF+FF+EF-CNN method. To ensure fairness, in the comparison method, the network structure of the CNN model is the same as that of the method proposed in the present invention, and the model hyperparameters are obtained through multiple experiments.

[0117] Table 2 Pipeline pressure recognition results obtained by different methods

[0118] Method mAP (%) T (s) <![CDATA[GFLOPs(×10 -6 )]]> TF+FF-CNN method 99.98 10.32 30.08 MEF-CNN method 96.83 8.36 17.02 TF+FF+EF-CNN method 94.81 8.16 41.3 VMD-TF+FF+EF-CNN method 96.47 14.25 41.34 The method proposed in the present invention 96.40 6.07 7.68

[0119] Experiments show that the step-by-step gas pipeline leakage recognition method proposed in the present invention has achieved good results. In the comparative experiment, the present invention is superior to the existing common recognition algorithms in terms of recognition accuracy, real-time performance, etc.

[0120] The present invention discloses a multi - condition gas transmission pipeline leakage step - by - step identification method based on dual - channel acoustic wave feature decoupling. Aiming at the analysis results of the acoustic wave characteristics of pipeline leakage: the cross - correlation coefficient function (CCF) between leakage signals has obvious peaks, which can effectively distinguish leakage signals from background noise. Different factors have different influence mechanisms on the cross - power spectral density (CPSD) characteristics of leakage signals. The logarithmic amplitude CPSD characteristics are determined by the leakage condition, while the linear amplitude CPSD characteristics are mainly affected by the leakage aperture. Based on this, first, the CCF peak is used to judge whether leakage occurs. After confirming the occurrence of leakage, the leakage condition and leakage aperture are further determined according to the CPSD image. Finally, a feature vector is constructed based on the effective amplitude, variance and energy parameters of the leakage signal to realize the identification of leakage pressure. Compared with the traditional non - step - by - step identification method, the identification accuracy of the proposed method is significantly improved, the computational complexity is greatly reduced, and the real - time performance is better.

[0121] In the present specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0122] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling, characterized in that Including: Collect the dual-channel signals of leakage vibration sound waves of the pipeline, and extract the amplitude, variance, and energy parameters; Perform cross-correlation analysis on the dual-channel signals using the cross-correlation coefficient function to determine whether leakage occurs; After confirming the occurrence of leakage, use the cross-power spectral density CPSD for spectral feature analysis to construct a logarithmic amplitude CPSD image, and identify the leakage condition based on the logarithmic amplitude CPSD image; Convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image, and identify the leakage aperture based on the linear amplitude CPSD image; After identifying the leakage aperture, construct a three-dimensional feature vector based on the amplitude, variance, and energy parameters, and input it into the pressure identification model for leakage pressure identification.

2. The multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 1, characterized in that The dual-channel signals of leakage vibration sound waves include: Where S(t) is the leakage signal, S1(t) and S2(t) are the dual-channel noisy signals collected by the sensors respectively, N1(t) and N2(t) are random noises, τ is the time delay, and θ is the attenuation factor.

3. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 2, characterized in that, Only the leakage signals are correlated while the noises are not. When there is no fixed strong interference around the pipeline, only the leakage signals S(t) and θS(t - τ) are correlated in the two-channel sensor signals, while the random noises N1(t) and N2(t) are not correlated with S1(t) and θS(t - τ), and N1(t) and N2(t) are not correlated with each other.

4. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 1, characterized in that The performing cross-correlation analysis on the dual-channel signals using the cross-correlation coefficient function includes: Perform signal preprocessing to make the sampling rates of S1(t) and S2(t) the same and the signal lengths the same; Calculate the CCF values of S1(t) and S2(t) at different time delays τ; The higher the peak value of CCF(τ), the stronger the signal correlation. When the input of the cross-correlation coefficient function CCF is the leakage signal, a correlation coefficient peak is obtained at the delay τ between the two signals; when the input signal is random noise, there is no peak in the cross-correlation coefficient function CCF.

5. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 1, characterized in that After confirming the occurrence of leakage, using the cross-power spectral density CPSD for spectral feature analysis to construct a logarithmic amplitude CPSD image, and identifying the leakage condition based on the logarithmic amplitude CPSD image includes: Perform fast Fourier transforms on the leakage signals S1(t) and S2(t) respectively to convert them into frequency-domain signals Where F(*) represents the Fourier transform; f is the frequency variable; Perform the calculation of the logarithmic amplitude CPSD image: Among them, CPSD log represents the logarithmic amplitude CPSD value, * represents the complex conjugate operation, || represents the modulus operation, and T is the total duration of the signal; Input the logarithmic amplitude CPSD image into the leakage condition identification model based on YOLOv8 for leakage condition identification.

6. The multi - condition gas transmission pipeline leakage step - by - step identification method based on dual - channel acoustic wave feature decoupling according to claim 1, characterized in that, The converting the logarithmic amplitude CPSD image into a linear amplitude CPSD image includes: Convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image through antilogarithmic operation: Among them, CPSD linear represents the linear amplitude CPSD value.

7. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 1, characterized in that The identifying the leakage aperture based on the linear amplitude CPSD image includes: Input the linear amplitude CPSD image into the leakage aperture identification model based on YOLOv8 for leakage aperture identification.

8. A multi-condition gas transmission pipeline leakage step-by-step identification method based on dual-channel acoustic wave feature decoupling according to claim 1, characterized in that The constructing a three-dimensional feature vector based on the amplitude, variance, and energy parameters, and inputting it into the pressure identification model for leakage pressure identification includes: Set a fixed time window, divide the leakage signal S1(t) into M time segments, and the data within each time window is used as a sample; Calculate the effective amplitude, variance, and energy indexes of each sample: where RMS i , V i , E i represent the effective amplitude, variance, and energy of the i-th sample, respectively, where 1 ≤ i ≤ M; x(n) is the signal amplitude, n is the n-th sampling point within the time window, μ is the mean of the signal within the time window, and N is the number of data points within each time window; Obtain the M×3-dimensional feature matrix F: Input F into the CNN-based pressure recognition model for classification to identify leakage conditions under different pressure states.

9. A multi-condition gas transmission pipeline leakage step-by-step identification system based on dual-channel acoustic wave feature decoupling, characterized in that It includes: Acquisition module: used to acquire the dual-channel signals of pipeline leakage vibration sound waves; Extraction module: used to extract amplitude, variance, and energy parameters; Cross-correlation analysis module: used to perform cross-correlation analysis on the dual-channel signals using the cross-correlation coefficient function to determine whether leakage has occurred; Leakage condition recognition module: used to perform spectral feature analysis using the cross-power spectral density CPSD to construct a logarithmic amplitude CPSD image after confirming leakage, and perform leakage condition recognition based on the logarithmic amplitude CPSD image; Leakage aperture recognition module: used to convert the logarithmic amplitude CPSD image into a linear amplitude CPSD image and perform leakage aperture recognition based on the linear amplitude CPSD image; Leakage pressure recognition module: used to construct a three-dimensional feature vector based on the amplitude, variance, and energy parameters after leakage aperture recognition, and input it into the pressure recognition model for leakage pressure recognition.

Images