A general gas pipeline leakage point positioning method along the cross-sectional circumference

Abstract

The application relates to a general gas pipeline leakage point positioning method along a cross-section circumference, which comprises the following steps: dividing the pipeline cross-section into multiple quadrants, obtaining an acoustic wave signal, respectively calculating the incident angle of the signal by using beam forming and instantaneous windowing MUSIC, introducing a signal-to-noise ratio estimation value, and performing nonlinear weighting to roughly determine the quadrant where the leakage point is located; based on a K-means clustering algorithm, the quadrant corresponding to the leakage point is divided into multiple subspaces; based on a generalized quadratic cross-correlation algorithm, noise resistance and time delay estimation are performed, the subspaces are searched, and the search area is reduced; based on a band-pass filter, the acoustic wave signal in the search area is filtered and processed; and for the filtered acoustic wave signal, beam forming based on an SP-ISM algorithm is used to accurately position the leakage point. Compared with the prior art, the application has the advantages of effectively avoiding the shielding of the pipeline itself to the acoustic wave, high time delay estimation accuracy, accurate positioning and the like.

Description

Technical Field

[0001] This invention relates to the field of pipeline safety monitoring equipment technology, and in particular to a method for locating leak points along the circumferential direction of a general gas pipeline cross-section. Background Technology

[0002] As the world's deepest underground laboratory, to obtain more accurate results, gas pipelines are needed to expel radon and its decay products, among other gases. Currently, most gas pipelines use PE pipes, and pipeline leaks are one of the biggest hidden dangers affecting their use. Locating leaks in inaccessible areas, where manual location is difficult, has become a major challenge. Previous researchers have attempted to directly locate leaks using CBF algorithms, MUSIC algorithms, or improved versions of these methods under free-field conditions. Existing detection methods for locating pipeline leaks primarily involve placing a microphone array at the sampling point, then denoising and enhancing the collected acoustic signals, and using multiple microphone arrays deployed along the pipeline to collaboratively detect leaks, accurately determining the lateral and cross-sectional location of the leak. The improvements in existing, more detailed methods mainly lie in: 1. estimating the time delay of the received acoustic signals; 2. dividing the detection area into several subspaces for detailed detection, resulting in smaller errors.

[0003] However, the main shortcomings of existing methods are: 1. Previous location methods were mostly based on free fields with no obstructions, where the sound waves were direct. However, PE pipes are quite thick (up to 2.7 cm), causing sound signals to pass through them. The PE pipe significantly obstructs the signal, resulting in not only direct sound waves but also diffracted waves. The presence of diffracted waves interferes with the direction and distance of sound wave propagation, greatly affecting signal propagation and causing errors in leak location. 2. Previous research focused primarily on leak location along the pipe's direction, failing to consider how to achieve more accurate circumferential cross-sectional location when human access is limited and machine detection is necessary, thus saving significant manpower and resources. Currently, research on circumferential cross-sectional location is still in its early stages, and existing methods for circumferential location of gas pipeline cross-sections have significant errors. Summary of the Invention

[0004] The purpose of this invention is to provide a method for locating leak points along the circumferential cross-section of a general gas pipeline. First, the cross-section is divided into four quadrants to roughly locate the quadrant position of the leak point. Then, the quadrant is divided into several subspaces, the peak value is enhanced, the sound wave intensity of each space is read, and the subspace with the highest peak value is compared. The subspace is then further searched and located to accurately determine the leak point.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for locating leak points along the circumferential cross-section of a general gas pipeline includes the following steps:

[0007] Step 1) Divide the cross-section of the gas pipeline into multiple quadrants, acquire the acoustic signal, and calculate the spatial incident angle of the signal using the beamforming algorithm and the instantaneous windowing MUSIC algorithm respectively. To increase the robustness and spatial resolution of the algorithm, a signal-to-noise ratio estimate is introduced, and the output of the two algorithms is nonlinearly weighted using the sigmoid function to roughly determine the quadrant where the leak point is located.

[0008] Step 2) Divide the quadrant corresponding to the leakage point into multiple subspaces based on the K-means clustering algorithm;

[0009] Step 3) Based on the generalized quadratic cross-correlation algorithm, noise resistance and time delay estimation are performed, and the subspace is searched to reduce the search area;

[0010] Step 4) Filter the acoustic signal within the search area using a bandpass filter;

[0011] Step 5) For the filtered acoustic signal, beamforming based on the SP-ISM algorithm is used to accurately locate the leakage point.

[0012] Step 1) of the above-mentioned coarse positioning using a beamforming algorithm includes the following steps:

[0013] Step 1-1-1) Establish a two-dimensional rectangular coordinate system in the cross-section of the gas pipeline and divide the cross-section into four quadrants;

[0014] Step 1-1-2) Use beamforming algorithm to perform beamforming on the acquired sound wave signal for rough localization, estimate the spatial spectrum, locate the distribution location of the target sound source and the intensity of the sound field distribution;

[0015] Step 1-1-3) Compare the sound field intensity in the four quadrants and determine the quadrant interval where the highest peak intensity in the sound field is significantly higher than that in other sound fields, and use this quadrant as the quadrant corresponding to the leakage point.

[0016] Step 1-1-2) includes the following steps:

[0017] Step 1-1-2-1) Obtain the discrete received acoustic signal matrix X(l) from the uniform linear array, and calculate its autocorrelation matrix.

[0018]

[0019] Where L represents a snapshot, X H (l) is the transpose of X(l);

[0020] Step 1-1-2-2) Construct the power spectrum based on the autocorrelation matrix:

[0021]

[0022] Among them, P CBF (y) represents the output power of y, which is determined by the input acoustic signal matrix and an N×1 column vector w(θ) that is angle-dependent, where w satisfies a norm of 1 and α k (θ) represents the steering vector form of the acoustic signal. For α k The transpose of (θ), where θ is the incident angle of the signal, i.e., the pitch angle;

[0023] The column vector w(θ) is:

[0024]

[0025] in,

[0026]

[0027] Where c is the speed of sound transmission, f k Let d be the wave number of the sound wave and d be the spacing between the elements in the microphone array.

[0028] Step 1-1-2-3) By scanning w(θ) in the range of [-90°, 90°), the angle θ corresponding to the peak value of the power spectrum is obtained, which is the pitch angle of the source, thereby obtaining the distribution location of the target sound source and the intensity of the sound field distribution.

[0029] Step 1) of the above-mentioned coarse localization using the instantaneous windowing MUSIC algorithm includes the following steps:

[0030] Step 1-2-1) Construct the covariance matrix: Combine the acoustic signals into a data matrix, apply the Hanning window function to the data matrix to obtain the windowed data matrix, and calculate the covariance matrix between the windowed data matrix and its transpose.

[0031] Step 1-2-2) Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors;

[0032] Steps 1-2-3) Use eigenvalues ​​and eigenvectors to construct a spatial spectrum estimation function to represent the energy distribution of the signal source in different directions;

[0033] Steps 1-2-4) Analyze and search for the peak value of the spatial spectrum estimation function to determine the quadrant corresponding to the leakage point.

[0034] Steps 1-2-3) include the following steps:

[0035] Step 1-2-3-1) Select the first N eigenvalues ​​and their corresponding eigenvectors, where N is the number of signal sources;

[0036] Step 1-2-3-2) Normalize the selected feature vectors;

[0037] Step 1-2-3-3) Constructing the spatial spectrum estimation function: For a given direction angle θ, the beamforming coefficient θ is obtained by taking the inner product of the unit vector of that direction with the normalized eigenvector. T *v, where θ T It is the transpose of θ, and v is the normalized eigenvector; the square of the beamforming coefficient is regarded as the energy distribution of the signal in that direction, and is used as the spatial spectrum estimation function value f = (θ) in that direction. T *v) 2 ;

[0038] Steps 1-2-3-4) For multiple directional angles, calculate the spatial spectrum estimation function value in each direction to obtain the energy distribution of the signal source in different directions.

[0039] The method for determining the window length of the instantaneous windowing MUSIC algorithm is as follows:

[0040] Let r be the vertical distance between the sound source and the acoustic center of the microphone array, and t be the time point at which the sound source emits the signal. s The signal arrives at the array at time t. r The incident angle of the signal is θ(t) r The straight-line distance from the sound source to the center of the microphone array is denoted as t. r function R c (t r ),but:

[0041] t r =t s +R c (t r ) / c

[0042] R c (t r )=r / cosθ(t r )

[0043] Where d is the spacing between the elements in the microphone array, and c is the speed of sound transmission;

[0044] The derivation yields:

[0045] [1+vsinθ(t r ) / c]dt r =dt s

[0046] Where v is the vertical distance from the sound source to the microphone array divided by the speed of sound, c, i.e., v = r / c, and t is the time required for the sound wave to travel from the sound source to the microphone array according to the basic principle of sound wave propagation, and the time difference tt. r sinθ(t) is the time difference between the sound wave arriving at the microphone array from the sound source. r )≈θ(t r ), representing the angle at which the sound wave reaches the microphone array from the sound source;

[0047] According to the definition of instantaneous frequency, the above equation is equivalent to:

[0048]

[0049] Among them, f tr f represents the instantaneous frequency of the signal received by the microphone array. ts Indicates the instantaneous frequency of the signal itself;

[0050] For both sides of t r Taking the derivative, we get:

[0051]

[0052] Since v / c << 1, the above equation simplifies to:

[0053]

[0054] Assuming the window length T is very small, then approximately:

[0055]

[0056] Among them, f s The sampling frequency;

[0057] To ensure that the signal within each instantaneous window meets the narrowband signal condition, we get:

[0058]

[0059] Among them, W B It is the ratio of bandwidth to the signal frequency range, and f0 is the center frequency of the signal;

[0060] After simplification, the window length T is obtained as follows:

[0061]

[0062] The specific steps of using the sigmoid function to perform non-linear weighting of the outputs of the two algorithms are as follows:

[0063] The weights output by the beamforming algorithm and the instantaneous windowed MUSIC algorithm are solved using the sigmoid function, and then nonlinear weighting is performed based on the output values ​​and corresponding weights. The formula for calculating the output weights is as follows:

[0064]

[0065] w2 = 1 - w1

[0066] Where w1 is the weight output by the beamforming algorithm, w2 is the weight output by the instantaneous windowing MUSIC algorithm; SNR represents the introduced signal-to-noise ratio estimate; ε is a threshold used to control SNR; s represents a scaling factor used to control how w1 and w2 change with SNR, s = SNR / ε.

[0067] Step 2) includes the following steps:

[0068] Step 2-1) Feature selection: Select features from the spatial spectrum estimated in Step 1) that reflect the quadrant where the leak point is located;

[0069] Step 2-2) Feature Standardization: Standardize the selected features;

[0070] Steps 2-3) Determine the value of K: Determine the number of clusters K, that is, determine the number of subspaces to be divided;

[0071] Steps 2-4) Initialize cluster centers: Based on the selected features, initialize K cluster centers using the K-means++ algorithm;

[0072] Steps 2-5) Iterative optimization: Use the K-means algorithm for iterative optimization until the cluster centers no longer change significantly or the predetermined number of iterations is reached. The iterative process includes the following two steps:

[0073] a) Sample allocation: Assign each data point to the nearest cluster center to form K clusters;

[0074] b) Update cluster centers: Recalculate the cluster centers for each cluster based on the assigned samples;

[0075] Steps 2-6) Result Interpretation: For each cluster, interpret and analyze the data points within it, observe the characteristics and distribution of each subspace, and determine its quadrant.

[0076] Steps 2-7) Result Validation: Validate the results using independent datasets related to pipeline leaks or expert systems in the field.

[0077] Step 3) specifically refers to:

[0078] Assuming the sound source location is estimated by weighted fusion of the beamforming algorithm and the instantaneous windowed MUSIC algorithm, then... Azimuth If the maximum absolute error is ε1 and the maximum absolute error of the pitch angle θ is ε2, then the reduced search range S determined based on the generalized quadratic cross-correlation algorithm is:

[0079]

[0080] At this point, the number of search regions for the algorithm is:

[0081]

[0082] Where a and b are the step sizes corresponding to the azimuth and elevation angles during the search.

[0083] Step 5) includes the following steps:

[0084] Step 5-1) Divide the received filtered broadband signal into sub-band signals and use the narrowband DOA estimation algorithm to obtain the unnormalized azimuth spectrum of each sub-band signal.

[0085] Step 5-2) Retain the spectral values ​​at the azimuth of the local peaks in each sub-band signal azimuth spectrum, while setting the spectral values ​​at other azimuths to a pre-configured minimum value, to obtain the sub-band signal azimuth spectrum after local peak extraction processing;

[0086] Step 5-3) The processed sub-band signal azimuth spectra are superimposed and summed to obtain the final broadband signal azimuth spectrum, which is then normalized and displayed to accurately locate the leak point.

[0087] Compared with the prior art, the present invention has the following beneficial effects:

[0088] (1) By roughly locating the quadrant, this invention effectively avoids the influence of sound waves from other directions on the sound waves near the leak point, and also avoids the pipe itself blocking the sound waves, so that the subsequent experimental results are more accurate.

[0089] (2) The subspace of this invention is smaller, the noise resistance is better, the time delay estimation accuracy is higher, and the error can be greatly reduced.

[0090] (3) The present invention performs effective filtering and enhancement of signals, resulting in better leakage detection and more accurate positioning;

[0091] (4) The present invention has conducted detailed research and design on the location of leakage points in the cross-section circumferential direction, which reduces the error in the overall positioning. Attached Figure Description

[0092] Figure 1 This is a flowchart of the method of the present invention;

[0093] Figure 2 This is a diagram showing the input-output relationship of the beamforming algorithm.

[0094] Figure 3 This is an experimental result of detecting the position angle of a sound source using the CBF algorithm, as simulated in one embodiment.

[0095] Figure 4 This is a schematic diagram of the quadrant corresponding to the leak point determined in one embodiment;

[0096] Figure 5 This is the result of instantaneously windowing a segment of sound wave signal in one embodiment;

[0097] Figure 6 This is a schematic diagram illustrating the use of a generalized quadratic cross-correlation algorithm to narrow down the search region in one embodiment.

[0098] Figure 7 This is the result of a bandpass filter processing a signal within a threshold in one embodiment;

[0099] Figure 8 This is a flowchart of the SP-ISM algorithm. Detailed Implementation

[0100] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0101] This embodiment provides a method for locating leak points along the circumferential cross-section of a general gas pipeline, such as... Figure 1 As shown, it includes the following steps:

[0102] Step 1) The pipe cross-section is divided into multiple quadrants, and relevant acoustic signals are acquired. Then, beamforming and instantaneous windowing MUSIC algorithms are used to calculate the incident angle of the signals, respectively. To increase the robustness and accuracy of the algorithms, a signal-to-noise ratio estimate is introduced, and the outputs of the two algorithms are nonlinearly weighted using the sigmoid function to roughly determine the quadrant where the leak point is located.

[0103] A two-dimensional Cartesian coordinate system is established within the pipe's cross-section, dividing the section into four intervals. Beamforming and instantaneous windowing MUSIC algorithms are used to locate the distribution of the target sound source and the intensity of the sound field. By comparing the sound field intensities in the four quadrants, the quadrant interval with the highest peak intensity significantly higher than the others is identified. Subsequently, only the parameter values ​​of the acoustic sensors located in this quadrant are read, and only the microphone array near the leak point is read. This pre-coarse localization approach effectively avoids the pipe's obstruction of the sound signal from the other three quadrants, ensuring that the angle between the acoustic sensors and the leak point does not exceed 90 degrees, thus preventing sound signal obstruction.

[0104] In one embodiment, a beamforming algorithm (CBF algorithm) is used for coarse localization, including the following steps:

[0105] Step 1-1-1) Establish a two-dimensional rectangular coordinate system in the cross-section of the gas pipeline and divide the cross-section into four quadrants.

[0106] Step 1-1-2) Use beamforming algorithm to perform beamforming on the acquired sound wave signal for rough localization, estimate the spatial spectrum, locate the distribution location of the target sound source and the intensity of the sound field distribution.

[0107] Step 1-1-2-1) Obtain the discrete received acoustic signal matrix X(l) from the uniform linear array, and calculate its autocorrelation matrix.

[0108]

[0109] Where L represents a snapshot, X H (l) is the transpose of X(l).

[0110] Step 1-1-2-2) Construct the power spectrum based on the autocorrelation matrix:

[0111]

[0112] Among them, P CBF (y) represents the output power of y, which is determined by the input acoustic signal matrix and an N×1 column vector w(θ) that is angle-dependent, where w satisfies a norm of 1 and α k (θ) represents the steering vector form of the acoustic signal. For α k The transpose of (θ), where θ is the incident angle of the signal, i.e., the pitch angle.

[0113] First, construct an N×1 column vector w(θ) that is related to the angle:

[0114]

[0115] The conjugate of each element of w(θ) is used as a weighting coefficient and sequentially compared with the input acoustic signal x1(t), x2(t), ..., x N Multiply the products (t) and sum them to get the new output y(t):

[0116]

[0117] Among them, w k * Indicates w k The conjugate of the above equation can be written in matrix form as follows:

[0118] y(t)=w(θ) H X(t)

[0119] Where X(t) represents the input acoustic wave matrix.

[0120] like Figure 2 As shown, the output power P of y(t) is... CBF (y) is:

[0121] P CBF (y)=E(yy H )=E(w(θ) H X(t)X H (t)w(θ))

[0122] P CBF (y)=w(θ) H R xx w(θ)

[0123] Among them, R xx Here is the autocorrelation matrix of X(t):

[0124] R xx =E(X(t)X H (t))

[0125] Find P under these conditions. CBF The maximum value of (y):

[0126]

[0127] When the angle θ of w(θ) is related to the pitch angle β of a certain information source in space... k When the signals coincide, the output power reaches a peak value, from which the one-dimensional parameter β of a certain signal source can be calculated. k Therefore, w(θ) can be simply designed as:

[0128]

[0129] Where, α k (θ) represents the steering vector form of the signal.

[0130]

[0131] Where c is the speed of sound transmission, f k Let d be the wave number of the sound wave, and d be the spacing between the elements in the microphone array.

[0132] Step 1-1-2-3) By performing an angle scan of w(θ) within the range of [-90°, 90°), the angle θ corresponding to the peak value of the power spectrum is obtained, which is the elevation angle β of the signal source. k This allows us to obtain the distribution location of the target sound source and the intensity of the sound field distribution.

[0133] Figure 3 The results are from an experiment simulating the detection of the sound source position angle using the CBF algorithm.

[0134] Step 1-1-3) Compare the sound field intensity in the four quadrants and determine the quadrant interval where the highest peak intensity in the sound field is significantly higher than that in other sound fields, and use this quadrant as the quadrant corresponding to the leakage point.

[0135] In one embodiment, according to as follows Figure 3 The experimental results determine the quadrant corresponding to the leak point, such as Figure 4 As shown.

[0136] In another embodiment, a coarse localization is performed using an instantaneous windowing MUSIC algorithm, including the following steps:

[0137] Step 1-2-1) Constructing the covariance matrix: Combine the acoustic signals into a data matrix, where each column represents a signal received by a sensor. To process instantaneous signals in the time domain, the data matrix is ​​windowed using the Hanning window function to obtain the windowed data matrix X, and the covariance matrix between the windowed data matrix X and its transpose is calculated:

[0138]

[0139] Where C is the covariance matrix, N is the number of data points, and X is the windowed data matrix. H This represents the conjugate transpose of X.

[0140] Step 1-2-2) Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors.

[0141] Step 1-2-2-1) Singular Value Decomposition: Use singular value decomposition to decompose the covariance matrix into U*∑*V H The form is where U is the left singular vector matrix, ∑ is a diagonal matrix with singular values ​​of the covariance matrix C on the diagonal, and V H It is the conjugate transpose of the right singular vector matrix.

[0142] Step 1-2-2-2) Eigenvalue and eigenvector extraction: Extract eigenvalues ​​λ and eigenvectors μ from the eigenvalue decomposition results. The eigenvalue λ represents the energy of the covariance matrix C in the direction of the eigenvector, while the eigenvector μ represents the direction corresponding to the eigenvalue.

[0143] Step 1-2-2-3) Eigenvalue sorting: Sort the eigenvalues ​​λ in descending order, λ1>λ2>...>λ M >0. The larger the eigenvalue, the stronger the energy in the direction represented by the corresponding eigenvector, where M represents the number of eigenvalues ​​λ.

[0144] Step 1-2-2-4) Correspondence between eigenvalues ​​and eigenvectors: There is a one-to-one correspondence between eigenvectors μ and eigenvalues. Each eigenvector μ represents a spatial direction, describing the beamforming of the signal source in that direction.

[0145] Steps 1-2-3) Use eigenvalues ​​and eigenvectors to construct a spatial spectrum estimation function to represent the energy distribution of the signal source in different directions.

[0146] Step 1-2-3-1) Select the first N eigenvalues ​​and their corresponding eigenvectors μ, where N is the number of signal sources.

[0147] Step 1-2-3-2) Normalize the selected eigenvectors to make their magnitude 1, ensuring that the energy range of different eigenvectors is consistent, which facilitates comparison and analysis.

[0148] Step 1-2-3-3) Constructing the spatial spectrum estimation function: For a given direction angle θ (representing a direction in space), the beamforming coefficient θ is obtained by taking the inner product of the unit vector of that direction with the normalized eigenvector. T *v, where θ T It is the transpose of θ, and v is the normalized eigenvector; the square of the beamforming coefficient is regarded as the energy distribution of the signal in that direction, and is used as the spatial spectrum estimation function value f = (θ) in that direction. T *v) 2 .

[0149] Steps 1-2-3-4) For multiple directional angles, calculate the spatial spectrum estimation function value in each direction to obtain the energy distribution of the signal source in different directions.

[0150] Steps 1-2-4) Analyze and search for the peak value of the spatial spectrum estimation function to determine the quadrant corresponding to the leakage point.

[0151] Step 1-2-4-1) Analyze the peak position: Observe the peak position of the spatial spectrum estimation function. The peak value indicates the direction of stronger signal energy. By finding the peak value in the spatial spectrum estimation function, the direction of the leakage point can be preliminarily determined.

[0152] Step 1-2-4-2) Peak Detection Threshold: According to the application requirements of the present invention, a peak detection threshold is set using the ratio of peak value to background noise. When the peak value exceeds this threshold, it is considered that there is a high probability of a leakage point in that direction.

[0153] Step 1-2-4-3) Determine the quadrant where the leak point is located: Based on the position of the peak value, determine the quadrant where the leak point is located.

[0154] a) Set the origin of the coordinate system as the center or reference point of the array, and divide it into four quadrants.

[0155] b) If the peak value is located in the first or fourth quadrant, the leak point is considered to be located in the quadrant directly above or to the right.

[0156] c) If the peak value is located in the second or third quadrant, the leak point is considered to be located in the quadrant directly above or to the left.

[0157] In this embodiment, the method for determining the window length of the instantaneous windowing MUSIC algorithm is as follows:

[0158] Let r be the vertical distance between the sound source and the acoustic center of the microphone array, and t be the time point at which the sound source emits the signal. s The signal arrives at the array at time t. r The incident angle of the signal is θ(t) r The straight-line distance from the sound source to the center of the microphone array is denoted as t. r function R c (t r ),but:

[0159] t r =t s +R c (t r ) / c

[0160] R c (t r )=r / cosθ(t r )

[0161] Where d is the spacing between the elements in the microphone array, and c is the speed of sound transmission;

[0162] The derivation yields:

[0163] [1+vsinθ(t r ) / c]dt r =dt s

[0164] Where v is the vertical distance from the sound source to the microphone array divided by the speed of sound, c, i.e., v = r / c, and t is the time required for the sound wave to travel from the sound source to the microphone array according to the basic principle of sound wave propagation, and the time difference tt. r sinθ(t) is the time difference between the sound wave arriving at the microphone array from the sound source. r )≈θ(t r ), representing the angle at which the sound wave reaches the microphone array from the sound source;

[0165] According to the definition of instantaneous frequency, the above equation is equivalent to:

[0166]

[0167] Among them, f tr f represents the instantaneous frequency of the signal received by the microphone array. ts Indicates the instantaneous frequency of the signal itself;

[0168] For both sides of t r Taking the derivative, we get:

[0169]

[0170] Since v / c << 1, the above equation simplifies to:

[0171]

[0172] Assuming the window length T is very small, then approximately:

[0173]

[0174] Among them, f s The sampling frequency;

[0175] To ensure that the signal within each instantaneous window meets the narrowband signal condition, we get:

[0176]

[0177] Among them, W B It is the ratio of bandwidth to the signal frequency range, and f0 is the center frequency of the signal;

[0178] After simplification, the window length T is obtained as follows:

[0179]

[0180] Figure 5 This is the result of instantaneous windowing processing of a segment of sound wave signal in one embodiment.

[0181] Compared to traditional beamforming algorithms, which do not optimize for the sidelobes of the power spectrum, resulting in large sidelobes and low resolution, the MUSIC algorithm offers high angle estimation accuracy and resolution, making it more suitable for subspace partitioning in such scenarios.

[0182] Subsequently, the outputs of the two algorithms are non-linearly weighted using the sigmoid function:

[0183] If SNR represents the introduced signal-to-noise ratio estimate, and the weights output by the beamforming algorithm are w1 and the weights output by the instantaneous windowing MUSIC algorithm are w2, then the essence of using the sigmoid function to achieve nonlinear weighting is to use the sigmoid function to solve for w1 or w2, where w1 and w2 satisfy the condition w1 + w2 = 1. The formula for determining the weights in this embodiment is as follows:

[0184]

[0185] w2 = 1 - w1

[0186] In the formula for w1, the SNR value depends on the actual operating conditions and is calculated from real-time measurements. ε is a threshold used to control SNR, selected by experts in the field of pipeline leakage based on actual operating conditions. s represents a scaling factor, controlling how w1 and w2 change with SNR. s = SNR / ε, and its magnitude reflects the degree of deviation of SNR from ε. When SNR ≥ ε, w1 is close to 1, meaning that the coarse determination of the quadrant of the leak point tends to rely on the MUSIC algorithm because the MUSIC algorithm has higher spatial resolution in high signal-to-noise ratio environments; conversely, the coarse determination of the quadrant of the leak point tends to rely on the beamforming algorithm because the beamforming algorithm is more robust. Therefore, the strategy of using the sigmoid function to nonlinearly weight the outputs of the beamforming algorithm and the MUSIC algorithm can improve the accuracy and robustness of the coarse determination of the quadrant of the leak point.

[0187] Step 2) Divide the quadrant corresponding to the leakage point into multiple subspaces based on the K-means clustering algorithm.

[0188] Step 2-1) Feature selection: Select features from the spatial spectrum estimated in Step 1) that reflect the quadrant where the leak point is located.

[0189] Step 2-2) Feature Standardization: Standardize the selected features to ensure they have similar scales. This can prevent certain features from having too much influence on the results during the clustering process.

[0190] Steps 2-3) Determine the K value: Determine the number of clusters K, that is, the number of subspaces to be divided; the K value can be selected using rules of thumb, elbow rule or other suitable methods.

[0191] Steps 2-4) Initialize cluster centers: Based on the selected features, initialize K cluster centers using the K-means++ algorithm; the K-means++ algorithm can better select initial cluster centers and improve the algorithm's performance.

[0192] Steps 2-5) Iterative optimization: Use the K-means algorithm for iterative optimization until the cluster centers no longer change significantly or the predetermined number of iterations is reached. The iterative process includes the following two steps:

[0193] a) Sample allocation: Assign each data point to the nearest cluster center to form K clusters;

[0194] b) Update cluster centers: Recalculate the cluster centers for each cluster based on the assigned samples.

[0195] Steps 2-6) Result Interpretation: For each cluster, interpret and analyze the data points within it, observe the characteristics and distribution of each subspace, and determine its quadrant.

[0196] Steps 2-7) Result Validation: Validate the results using an independent dataset related to pipeline leakage or an expert system in the field to ensure that the partitioned subspaces match the actual situation.

[0197] Step 3) Perform noise immunity and time delay estimation based on the improved generalized quadratic cross-correlation algorithm of the phat function, and search the subspace to reduce the search area.

[0198] Assuming the sound source location is estimated by weighted fusion of the beamforming algorithm and the instantaneous windowed MUSIC algorithm, then... Azimuth If the maximum absolute error is ε1 and the maximum absolute error of the pitch angle θ is ε2, then the reduced search range S determined based on the generalized quadratic cross-correlation algorithm is:

[0199]

[0200] At this point, the number of search regions for the algorithm is:

[0201]

[0202] Where a and b are the step sizes corresponding to the azimuth and elevation angles during the search.

[0203] like Figure 6As shown, the improved search region is reduced to one-quarter of the original. The Phat function-weighted generalized quadratic cross-correlation algorithm has better stability and can further reduce computation time.

[0204] Step 4) Filter the acoustic signal in the search area based on the bandpass filter.

[0205] The threshold for leakage under normal conditions, ranging from 20Hz to 20kHz, was determined through multiple acoustic signal tests at different locations along the pipeline. A bandpass filter was then used to extract and process signals within this threshold, while signals outside the threshold were eliminated without further processing or analysis. Figure 7 The results of processing signals within the threshold using a bandpass filter are shown.

[0206] Step 5) For the filtered acoustic signal, beamforming based on the SP-ISM algorithm is used to accurately locate the leakage point.

[0207] The SP-ISM algorithm's core idea and steps include the following: When synthesizing the broadband azimuth spectrum from the sub-band signal azimuth spectra, only the spectral values ​​of the local peak azimuth of each sub-band signal are retained, while the spectral values ​​of other azimuths are set to minimum values. The spectral values ​​of other azimuths are not set to zero here to facilitate logarithmic calculation when calculating the azimuth spectrum. Based on the "spatial consistency" of the azimuth spectra of each sub-band signal, this synthesis of the broadband signal spectrum can enhance the signal at the target azimuth, thereby significantly suppressing noise interference in other azimuths and improving the resolution of the broadband spectrum. Figure 8 As shown, the specific steps include:

[0208] Step 5-1) Divide the received filtered broadband signal into sub-band signals and use the narrowband DOA estimation algorithm to obtain the unnormalized azimuth spectrum of each sub-band signal.

[0209] Step 5-2) Retain the spectral values ​​at the local peak azimuth of each sub-band signal azimuth spectrum, while setting the spectral values ​​at other azimuths to a pre-configured minimum value (e.g., 10). -4 This yields the sub-band signal azimuth spectrum after local peak extraction processing;

[0210] Step 5-3) The processed sub-band signal azimuth spectra are superimposed and summed to obtain the final broadband signal azimuth spectrum, which is then normalized and displayed to accurately locate the leak point.

[0211] The CBF method based on the SP-ISM algorithm is used to compare the remaining acoustic signal parameters and further filter the signal to find the acoustic wave with the highest peak and a duration of more than 3 seconds. The target location angle is calculated from the source of the acoustic wave. The advantages of this method include higher target orientation resolution, less computation, and more robust performance.

[0212] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for locating leak points along the circumferential cross-section of a general gas pipeline, characterized in that, Includes the following steps: Step 1) Divide the cross-section of the gas pipeline into multiple quadrants, acquire the acoustic signal, calculate the spatial incident angle of the signal using the beamforming algorithm and the instantaneous windowing MUSIC algorithm respectively, introduce the signal-to-noise ratio estimate, and use the sigmoid function to nonlinearly weight the outputs of the two algorithms to roughly determine the quadrant where the leak point is located. Step 2) Divide the quadrant corresponding to the leakage point into multiple subspaces based on the K-means clustering algorithm; Step 3) Based on the generalized quadratic cross-correlation algorithm, noise resistance and time delay estimation are performed, and the subspace is searched to reduce the search area; Step 4) Filter the acoustic signal within the search area using a bandpass filter; Step 5) For the filtered acoustic signal, beamforming based on the SP-ISM algorithm is used to accurately locate the leakage point.

2. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, Step 1) of the above-mentioned coarse positioning using a beamforming algorithm includes the following steps: Step 1-1-1) Establish a two-dimensional rectangular coordinate system in the cross-section of the gas pipeline and divide the cross-section into four quadrants; Step 1-1-2) Use beamforming algorithm to perform beamforming on the acquired sound wave signal for rough localization, estimate the spatial spectrum, locate the distribution location of the target sound source and the intensity of the sound field distribution; Step 1-1-3) Compare the sound field intensity in the four quadrants and determine the quadrant interval where the highest peak intensity in the sound field is significantly higher than that in other sound fields, and use this quadrant as the quadrant corresponding to the leakage point.

3. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 2, characterized in that, Step 1-1-2) includes the following steps: Step 1-1-2-1) Obtain the discrete received acoustic signal matrix X(l) from the uniform linear array, and calculate its autocorrelation matrix. Where L represents a snapshot, X H (l) is the transpose of X(l); Step 1-1-2-2) Construct the power spectrum based on the autocorrelation matrix: Among them, P CBF (y) represents the output power of y, which is determined by the input acoustic signal matrix and an N×1 column vector w(θ) that is angle-dependent, where w satisfies a norm of 1 and α k (θ) represents the steering vector form of the acoustic signal. For α k The transpose of (θ), where θ is the incident angle of the signal, i.e., the pitch angle; The column vector w(θ) is: in, Where c is the speed of sound transmission, f k Let d be the wave number of the sound wave and d be the spacing between the elements in the microphone array. Step 1-1-2-3) By scanning w(θ) in the range of [-90°, 90°), the angle θ corresponding to the peak value of the power spectrum is obtained, which is the pitch angle of the source, thereby obtaining the distribution location of the target sound source and the intensity of the sound field distribution.

4. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, Step 1) of the above-mentioned coarse localization using the instantaneous windowing MUSIC algorithm includes the following steps: Step 1-2-1) Construct the covariance matrix: Combine the acoustic signals into a data matrix, apply the Hanning window function to the data matrix to obtain the windowed data matrix, and calculate the covariance matrix between the windowed data matrix and its transpose. Step 1-2-2) Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; Steps 1-2-3) Use eigenvalues ​​and eigenvectors to construct a spatial spectrum estimation function to represent the energy distribution of the signal source in different directions; Steps 1-2-4) Analyze and search for the peak value of the spatial spectrum estimation function to determine the quadrant corresponding to the leakage point.

5. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 4, characterized in that, Steps 1-2-3) include the following steps: Step 1-2-3-1) Select the first N eigenvalues ​​and their corresponding eigenvectors, where N is the number of signal sources; Step 1-2-3-2) Normalize the selected feature vectors; Step 1-2-3-3) Constructing the spatial spectrum estimation function: For a given direction angle θ, the beamforming coefficient θ is obtained by taking the inner product of the unit vector of that direction with the normalized eigenvector. T *v, where θ T It is the transpose of θ, and v is the normalized eigenvector; the square of the beamforming coefficient is regarded as the energy distribution of the signal in that direction, and is used as the spatial spectrum estimation function value f = (θ) in that direction. T *v) 2 ; Steps 1-2-3-4) For multiple directional angles, calculate the spatial spectrum estimation function value in each direction to obtain the energy distribution of the signal source in different directions.

6. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 4, characterized in that, The method for determining the window length of the instantaneous windowing MUSIC algorithm is as follows: Let r be the vertical distance between the sound source and the acoustic center of the microphone array, and t be the time point at which the sound source emits the signal. s The signal arrives at the array at time t. r The incident angle of the signal is θ(t) r The straight-line distance from the sound source to the center of the microphone array is denoted as t. r function R c (t r ),but: t r =t s +R c (t r ) / c R c (t r )=r / cosθ(t r ) Where d is the spacing between the elements in the microphone array, and c is the speed of sound transmission; The derivation yields: [1+vsinθ(t r ) / c]dt r =dt s Where v is the vertical distance from the sound source to the microphone array divided by the speed of sound, c, i.e., v = r / c, and t is the time required for the sound wave to travel from the sound source to the microphone array according to the basic principle of sound wave propagation, and the time difference tt. r sinθ(t) is the time difference between the sound wave arriving at the microphone array from the sound source. r )≈θ(t r ), representing the angle at which the sound wave reaches the microphone array from the sound source; According to the definition of instantaneous frequency, the above equation is equivalent to: Among them, f tr f represents the instantaneous frequency of the signal received by the microphone array. ts Indicates the instantaneous frequency of the signal itself; For both sides of t r Taking the derivative, we get: Since v / c << 1, the above equation simplifies to: Assuming the window length T is very small, then approximately: Among them, f s The sampling frequency; To ensure that the signal within each instantaneous window meets the narrowband signal condition, we get: Among them, W B It is the ratio of bandwidth to the signal frequency range, and f0 is the center frequency of the signal; After simplification, the window length T is obtained as follows:

7. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, The specific steps of using the sigmoid function to non-linearly weight the outputs of the two algorithms are as follows: The weights output by the beamforming algorithm and the instantaneous windowed MUSIC algorithm are solved using the sigmoid function, and then nonlinear weighting is performed based on the output values ​​and corresponding weights. The formula for calculating the output weights is as follows: w2 = 1 - w1 Where w1 is the weight output by the beamforming algorithm, w2 is the weight output by the instantaneous windowing MUSIC algorithm; SNR represents the introduced signal-to-noise ratio estimate; ε is a threshold used to control SNR; s represents a scaling factor used to control how w1 and w2 change with SNR, s = SNR / ε.

8. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, Step 2) includes the following steps: Step 2-1) Feature selection: Select features from the spatial spectrum estimated in Step 1) that reflect the quadrant where the leak point is located; Step 2-2) Feature Standardization: Standardize the selected features; Steps 2-3) Determine the value of K: Determine the number of clusters K, that is, determine the number of subspaces to be divided; Steps 2-4) Initialize cluster centers: Based on the selected features, initialize K cluster centers using the K-means++ algorithm; Steps 2-5) Iterative optimization: Use the K-means algorithm for iterative optimization until the cluster centers no longer change significantly or the predetermined number of iterations is reached. The iterative process includes the following two steps: a) Sample allocation: Assign each data point to the nearest cluster center to form K clusters; b) Update cluster centers: Recalculate the cluster centers for each cluster based on the assigned samples; Steps 2-6) Result Interpretation: For each cluster, interpret and analyze the data points within it, observe the characteristics and distribution of each subspace, and determine its quadrant. Steps 2-7) Result Validation: Validate the results using independent datasets related to pipeline leaks or expert systems in the field.

9. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, Step 3) specifically refers to: Assuming the sound source location is estimated by weighted fusion of the beamforming algorithm and the instantaneous windowed MUSIC algorithm, then... Azimuth If the maximum absolute error is ε1 and the maximum absolute error of the pitch angle θ is ε2, then the reduced search range S determined based on the generalized quadratic cross-correlation algorithm is: At this point, the number of search regions for the algorithm is: Where a and b are the step sizes corresponding to the azimuth and elevation angles during the search.

10. The method for locating leak points along the circumferential cross-section of a general gas pipeline according to claim 1, characterized in that, Step 5) includes the following steps: Step 5-1) Divide the received filtered broadband signal into sub-band signals and use the narrowband DOA estimation algorithm to obtain the unnormalized azimuth spectrum of each sub-band signal. Step 5-2) Retain the spectral values ​​at the azimuth of the local peaks in each sub-band signal azimuth spectrum, while setting the spectral values ​​at other azimuths to a pre-configured minimum value, to obtain the sub-band signal azimuth spectrum after local peak extraction processing; Step 5-3) The processed sub-band signal azimuth spectra are superimposed and summed to obtain the final broadband signal azimuth spectrum, which is then normalized and displayed to accurately locate the leak point.

Images