Leakage sound signal denoising method based on combination of optimized VMD and improved wavelet threshold

Abstract

A leakage sound signal denoising method based on a combination of optimized VMD and an improved wavelet threshold, for use in solving the problem in existing noise processing methods of low identification accuracy in processing leakage sound signals of water supply pipe networks. The present invention comprises: acquiring leakage sound signals of a real water supply pipe network, and analyzing noise components and ranges of the leakage sound signals; on the basis of a goshawk optimization algorithm, performing parameter optimization on the number K of decomposition modes and a penalty factor α of VMD to obtain optimal parameters, and using the optimal parameters to construct a variational model; using the variational model to decompose the leakage sound signals to obtain a plurality of intrinsic mode components; using a correlation coefficient method to screen the plurality of intrinsic mode components to obtain high-frequency components and low-frequency components; performing wavelet threshold denoising processing on the high-frequency components to obtain denoised high-frequency components; and reconstructing the low-frequency components and the denoised high-frequency components to obtain denoised leakage sound signals. The beneficial effects are that the signal-to-noise ratio of denoising processing is improved, and the identification accuracy is improved.

Description

A leakage sound signal denoising method based on optimized VMD combined with improved wavelet threshold TECHNICAL FIELD

[0001] The present application relates to a water supply network leakage sound signal denoising method. BACKGROUND

[0002] The collection of leakage sound emission signals inevitably introduces some noise interference, including environmental noise, in-pipe noise, and internal circuit noise; the presence of noise reduces the clarity and accuracy of the leakage sound signal, making detection and analysis more difficult, thereby reducing detection efficiency; noise also interferes with accurate identification of the leakage sound, thereby affecting the positioning and repair of the leakage location; and these factors significantly increase the time and economic cost of pipeline repair.

[0003] Existing noise processing methods include Fourier transform, wavelet transform, adaptive filter, spectral subtraction method, and neural network, but these methods have limitations; Fourier transform cannot directly process non-stationary signals, and has limited effect on frequency segmented noise; the selection of wavelet transform base functions has a greater impact on denoising effect; the convergence speed of the adaptive filter may be slow, and its performance is unstable when processing non-stationary noise; spectral subtraction method requires accurate noise spectrum estimation, and when the noise frequency estimation is inaccurate, it may cause signal distortion; neural network requires a large amount of labeled data to train the network, and the complexity and computational load of the model are large. Therefore, the existing noise processing methods have low accuracy in identifying the leakage sound signal of the water supply network, and cannot meet the pipeline repair requirements. SUMMARY

[0004] The present application aims to solve the problem of low accuracy in identifying the leakage sound signal of the water supply network using existing noise processing methods, and proposes a leakage sound signal denoising method based on optimized VMD combined with improved wavelet threshold.

[0005] The leakage sound signal denoising method based on optimized VMD combined with improved wavelet threshold comprises:

[0006] Step 1: Obtain the leakage sound signal of the leakage point;

[0007] Step 2: Optimize the decomposition modal number K and the penalty factor a of VMD based on the hawk optimization algorithm, obtain the optimal parameters, and construct a variational model using the optimal parameters;

[0008] Step 3: Decompose the leakage sound signal obtained in step 1 using the variational model constructed in step 2 to obtain multiple intrinsic modal components;

[0009] Step four, screening the multiple intrinsic modal components obtained in step three by using the correlation coefficient method, and screening out high-frequency components and low-frequency components;

[0010] Step five, carrying out wavelet threshold denoising processing on the high-frequency components screened out in step four, and obtaining denoised high-frequency components;

[0011] Step six, reconstructing the low-frequency components screened out in step four and the denoised high-frequency components obtained in step five by using wavelet inverse transform, and obtaining a denoised leakage sound signal.

[0012] Further, the specific method for obtaining the leakage sound signal of the leakage point in step one is:

[0013] The vibration sensor is used to collect the leakage sound signal of the pipeline near the leakage point, and the collection interval is 5 minutes. Each leakage point collects 72 leakage and non-leakage data in the daytime and at night, respectively. Each leakage point collects 2x2x72=288 data in total. Seven leakage points are collected, and a total of 7x288=2016 acoustic emission data are collected. The sampling frequency of each data is 50kHz, the sampling time is 0.5 seconds, and the length of each data is 25k.

[0014] Further, the specific process of parameter optimization of the decomposition modal number K and the penalty factor a of VMD based on the hawk optimization algorithm in step two includes an initialization stage, a prey identification stage, and a pursuit and escape stage.

[0015] The initialization stage refers to that before the parameter is optimized by using the iterative algorithm, a certain number of initial solutions are randomly generated, the initial solution represents the position of the hawk, and the initial solution constitutes the search space. The quality of each initial solution is evaluated by the objective function of the hawk population.

[0016] The prey identification stage refers to that the hawk population performs global search on the search space to select the optimal region of the objective function.

[0017] The pursuit and escape stage refers to that as the number of iterations increases, the radius of the optimal region gradually decreases with the increase of the number of iterations, so that the sum of the modal bandwidths is minimized, and the optimal parameters are output.

[0018] Further, the position expression of the hawk population is:

[0019] Wherein, X represents the position matrix of the hawk population, X i is the initial solution of the i th individual, x i,j is the value of the j th dimension of the i th individual, N is the number of hawk population members, and m is the dimension of the problem space.

[0020] Further, the objective function of the hawk population is expressed as a target function value vector:

[0021] Wherein, F(X) is the obtained objective function of the necklace, F i is the target function value obtained by the i-th solution.

[0022] Further, the expression of the hawk population for global search on the search space to select the optimal region of the target function is: i k , i=1, 2, …, N, k=1, 2, …, i-1, i+1, …, N r∈[0,1]

[0023] Wherein, P i is the prey position of the i-th hawk, F Pi is the target value of the i-th hawk, is the new position of the i-th solution, is the value of the j-th dimension of the new position of the i-th solution, F i (new,P1) is the target function value of the first stage of hawk algorithm optimization, I is a constant, the value of I is 1 or 2.

[0024] Further, the formula for outputting the optimal parameters is:

[0025] Wherein, R is the parameter optimization range, t represents the number of current iterations, T is the maximum number of iterations, F i new,P2 is the target function value of the second stage of hawk algorithm optimization.

[0026] Further, in step two, after final optimization, the value range of the decomposition modal number K is 11-13, and the value range of the penalty factor a is 525-527.

[0027] Further, in step five, the wavelet threshold function expression of wavelet threshold denoising is:

[0028] Wherein, W j,k is the wavelet coefficient before denoising, is the wavelet coefficient after denoising, λ is the wavelet threshold, sign() is the sign function, θ is the smoothness adjustment factor, β is the approaching waveform speed adjustment factor, θ∈[1,2]; β∈(0,1000].

[0029] Further, in step six, the wavelet inverse transform formula is: ​

[0030] wherein, wherein, WT f (j, k) is the wavelet coefficient, ψ j,k (n) is the wavelet base function, A is the scale factor.

[0031] Compared with the prior art, the denoising method has the following beneficial effects: the denoising method improves the threshold function of the wavelet threshold denoising and optimizes the parameters of the VMD by using an optimization algorithm, improves the denoising effect of the wavelet threshold and the VMD, and combines the two to denoise the leakage sound signal, improves the signal-to-noise ratio, improves the denoising effect of the leakage sound signal recognition of the water supply pipe network, and improves the recognition accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0032] Fig. 1 is a flow chart of a leakage sound signal denoising method based on optimized VMD combined with improved wavelet threshold according to the first embodiment;

[0033] Fig. 2 is a principle diagram of a leakage sound signal denoising method based on optimized VMD combined with improved wavelet threshold according to the first embodiment;

[0034] Fig. 3 is a time domain and frequency domain diagram of leakage signals at different time periods according to the second embodiment; wherein, (a) is a time domain diagram of a night leakage signal; (b) is a frequency domain diagram of a night leakage signal; (c) is a time domain diagram of a daytime leakage signal; (d) is a frequency domain diagram of a daytime leakage signal; (e) is a time domain diagram of a non-leakage signal; and (f) is a frequency domain diagram of a non-leakage signal;

[0035] Fig. 4 is a comparison diagram of improved threshold and soft and hard threshold according to the first embodiment;

[0036] Fig. 5 is a decomposition diagram of VMD denoising of a standard signal according to the first embodiment;

[0037] Fig. 6 is a denoising effect diagram of VMD denoising of a standard signal according to the first embodiment; wherein, (a) is a noisy standard signal; (b) is a denoised signal with non-optimized parameters; and (c) is a denoised signal with optimized parameters;

[0038] Fig. 7 is a denoising effect diagram of improved wavelet threshold denoising of a standard signal according to the first embodiment; wherein, (a) is a noisy standard signal; and (b) is a denoised signal;

[0039] Fig. 8 is a denoising effect diagram of a standard signal of a joint denoising algorithm according to the first embodiment; wherein, (a) is a noisy standard signal; and (b) is a denoised signal after joint denoising;

[0040] FIG. 9 is a practical leak sound signal denoising effect diagram of the joint denoising algorithm in embodiment one, wherein (a) is an undenoised leak signal; (b) is a denoised leak signal; (c) is an undenoised leak signal; and (d) is a denoised leak signal. EMBODIMENT

[0041] In embodiment one, in combination with FIGS. 1-9, a leak sound signal denoising method based on optimized VMD combined with improved wavelet threshold is described. The denoising method comprises:

[0042] Step one, obtaining a leak sound signal of a leak point;

[0043] Step two, parameter optimization of the decomposition modal number K and the penalty factor a of VMD based on the gyrfalcon optimization algorithm, obtaining optimal parameters, and constructing a variational model using the optimal parameters;

[0044] Step three, decomposing the leak sound signal obtained in step one using the variational model constructed in step two to obtain a plurality of intrinsic modal components;

[0045] Step four, screening the plurality of intrinsic modal components obtained in step three using the correlation coefficient method to screen out high-frequency components and low-frequency components;

[0046] Step five, wavelet threshold denoising processing of the high-frequency components screened out in step four to obtain denoised high-frequency components;

[0047] Step six, reconstructing the low-frequency components screened out in step four and the denoised high-frequency components obtained in step five using wavelet inverse transform to obtain a denoised leak sound signal.

[0048] In this embodiment, in step one, the actual water supply network is checked at night by artificial means, the leak point is found, and a vibration sensor is used to collect the leak sound signal of the pipeline near the leak point. After the collection is completed, the leak point is repaired and the non-leak sound signal is collected. The noise components and range in the leak sound signal and the non-leak sound signal are analyzed, including environmental noise, pipe noise, and internal circuit noise.

[0049] In this embodiment, the specific method for obtaining the leak sound signal of the leak point in step one is as follows:

[0050] A vibration sensor is used to collect the leak sound signal of the pipeline near the leak point. The collection interval is 5 minutes, 72 pieces of leak and non-leak data are collected for each leak point during the day and at night, a total of 2x2x72=288 pieces of data are collected for each leak point, 7 pieces of data are collected for 7 leak points, a total of 7x288=2016 pieces of acoustic emission data are collected, the sampling frequency of each data is 50 kHz, the sampling time is 0.5 seconds, and the length of each data is 25 k.

[0051] In this embodiment, the time and frequency domains of the leaked and non-leaking acoustic emission signals collected at the same leak location and at different time periods are compared; the comparison results are shown in Figure 3. Figure 3 shows that the amplitudes of the two signals differ, with the amplitude of the leaked acoustic emission signal at night being slightly lower than that collected during the day. Comparing the spectrum diagrams of the leaked and non-leaking acoustic emission signals reveals that the useful frequency band of the leaked signal is mainly concentrated in the range of 1–2.5 kHz. To address the noise components, a wavelet threshold denoising algorithm combined with VMD is used to denoise the leaked acoustic signal, and the denoising algorithm is improved and optimized to enhance the signal-to-noise ratio of the leaked acoustic signal.

[0052] VMD decomposition transforms the problem of decomposing the original signal into a variational problem, and its key lies in the construction of the variational problem.

[0053] Constructing a variational problem:

[0054] Where K is the decomposition mode number, δ(t) is the Dirac function, and u k Let ω be a set of K modes. k This represents the center frequency of the corresponding mode.

[0055] Solving variational problems:

[0056] VMD decomposition is essentially finding the optimal solution to a variational problem. By transforming a constrained variational problem into an unconstrained one, and leveraging the advantages of quadratic penalty terms and the Lagrange multiplier method, an augmented Lagrange function is introduced, transforming the problem into the following function:

[0057] Among them, u k λ(t) is the k-th modal component, α is the penalty factor, λ(t) is the Lagrange multiplier, and f(t) is the signal to be decomposed. The function is solved by fixing the values ​​of the other two variables and updating the value of the third variable. First, for all ω≥0, iterative updates are performed. The value, specifically the mathematical expression, is:

[0058] Then iteratively update ω k The value of is given by the specific iterative expression:

[0059] Then, for all cases where ω≥0, double promotion is performed, and its expression is:

[0060] Finally, repeat the above steps until the constraints are met:

[0061] For each modal component after decomposition, in order to screen out useful signals and remove noise contained in the original signal, Pearson correlation coefficients (PCCs) of each mode are calculated, a threshold is set to screen the modal components, and the screened modal components are reconstructed to obtain the denoised signal.

[0062] When the decomposition mode number K is too large, false IMF components may appear after VMD decomposition;

[0063] When the value is too small, the VMD decomposition IMF component may not be sufficient, and the characteristics of the original signal cannot be fully extracted.

[0064] The size of the penalty factor affects the bandwidth of the IMF component. When a is too large, the bandwidth of the IMF component is too small, which may cause the characteristics of the original signal to be lost, resulting in the IMF component being unable to describe the characteristics of the original signal;

[0065] When a is too small, the bandwidth of the IMF component is too large, which may cause the IMF component to have a modal aliasing phenomenon.

[0066] In order to reduce the overlarge or too small setting of K and a parameters, an optimization algorithm is used to optimize the parameters of the variational mode decomposition, K and a are optimized by setting a certain threshold range, and the optimal parameters are found, thereby improving the decomposition effect of the variational mode decomposition.

[0067] In the embodiment, the parameter optimization of the decomposition mode number K and the penalty factor a of VMD based on the hawk optimization algorithm in step two includes the following specific processes: an initialization stage, a prey identification stage, and a pursuit and escape stage.

[0068] The initialization stage refers to that before the parameter is optimized by using an iterative algorithm, a certain number of initial solutions are randomly generated, the initial solutions represent the positions of hawks, and the initial solutions constitute a search space, and the quality of each initial solution is evaluated by the objective function of the hawk population.

[0069] The prey identification stage refers to that the hawk population performs global search on the search space to select the optimal region of the objective function.

[0070] The pursuit and escape stage refers to that as the number of iterations increases, the radius of the optimal region gradually decreases with the increase of the number of iterations, so that the sum of the bandwidths of each mode is minimized, and the optimal parameters are output.

[0071] In the present embodiment, the hawk optimization (NGO) algorithm is a method of gradient descent with adaptive learning rate, by simulating the behavior of a hawk in the process of hunting, specifically, simulating the hawk's behavior of identifying, attacking, chasing and escaping, etc. In the hawk algorithm, each solution is regarded as a "hawk", and is evaluated and compared according to its fitness value, and the solution is constantly updated by iteration algorithm to find the optimal solution.

[0072] In the present embodiment, the position expression of the hawk population is:

[0073] wherein X represents the position matrix of the hawk population, X i is the initial solution of the i-th individual, x i,j is the value of the i-th individual in the j-th dimension, N is the number of members of the hawk population, and m is the dimension of the problem space.

[0074] In the present embodiment, the objective function of the hawk population is expressed as a vector of objective function values:

[0075] wherein F(X) is the item chain of the objective function obtained, F i is the objective function value obtained by the i-th solution.

[0076] In the present embodiment, F(X) is the item chain of the objective function obtained, F i is the objective function value obtained by the i-th solution, and the objective function value F i is a key indicator for evaluating the quality of each solution. This objective function value is used to guide the optimization process to find the optimal parameter combination. F i here is determined by the sum of the signal reconstruction error and the modal bandwidth, the signal reconstruction error measures the gap between the decomposed modal and the original signal, and in the ideal case, the modal components and the residual of VMD should reconstruct the original signal as accurately as possible; the sum of the modal bandwidth measures the total bandwidth of all modes in the frequency domain. The bandwidth of each mode reflects the width of its frequency spectrum, and the target of the bandwidth sum is to make it as small as possible to ensure the effectiveness of the modal.

[0077] The expression of the hawk population for global search of the search space to select the optimal region of the objective function is: i = X k , i = 1, 2, …, N, k = 1, 2, …, i-1, i+1, …, N r ∈ [0, 1]

[0078] wherein P i is the prey position of the i-th hawk, F Pi is the target value of the i-th hawk, is the new position of the i-th solution, is the value of the j-th dimension of the new position of the i-th solution, F i (new,P1) is the objective function value of the first phase of the hawk optimization algorithm, I is a constant, and the value of I is 1 or 2.

[0079] In this embodiment, in the pursuit and escape stage, when the hawk attacks the prey, the prey will evade the pursuit of the hawk within a certain area, but eventually the hawk will complete the hunting of the prey after a period of pursuit; through the simulation of such pursuit and hunting behavior, the search ability of the optimization algorithm in the local space is improved, and the formula of the optimal parameter is output as:

[0080] wherein R is the parameter optimization range, t represents the number of current iterations, T is the maximum number of iterations, F i new,P2 is the objective function value of the second phase of the hawk optimization algorithm.

[0081] In this embodiment, the Doppler noise signal is used to verify the denoising effect of the optimized variational mode decomposition algorithm, and the initial variational mode decomposition parameters K and a are set to 10 and 5000.

[0082] The parameters K and a are optimized by the NGO algorithm, and the optimization ranges of K and a are [4, 15] and [100, 8000] respectively.

[0083] In this embodiment, the value range of the decomposition mode number K is 11-13, and the value range of the penalty factor a is 525-527. And the effect is best when the decomposition mode number K is 12 and the penalty factor a is 526.

[0084] In this embodiment, the wavelet threshold function expression of wavelet threshold denoising in step five is:

[0085] wherein W j,k is the wavelet coefficient before denoising, is the wavelet coefficient after denoising, λ is the wavelet threshold, sign() is the sign function, θ is the smoothness adjustment factor, β is the approaching waveform speed adjustment factor, θ ∈ [1, 2]; β ∈ (0, 1000].

[0086] In the embodiment, the decomposed modal components are denoised by using the correlation coefficient method, and a noise-added standard Doppler signal is tested by using a verification wavelet threshold denoising algorithm. The signal-to-noise ratio (SNR) and the root mean square error (RSME) after noise addition are 20.275 and 0.0982, respectively. The SNR and the RSME before parameter optimization are 24.653 and 0.0593, respectively. The SNR and the RSME after parameter optimization are 27.351 and 0.0435, respectively.

[0087] The denoising effect of the wavelet threshold denoising algorithm is greatly affected by the wavelet basis function and the wavelet threshold function. The characteristics of different wavelet basis functions and the denoising effects of different wavelet threshold functions are compared and analyzed.

[0088] The selection of the wavelet basis function usually needs to consider the following properties:

[0089] Orthogonality: Whether the wavelet basis function has orthogonality determines whether the signal can be decomposed into mutually orthogonal subspaces;

[0090] Compact support: The compactly supported wavelet has the localization property and can accurately locate the characteristics in the signal. The compactly supported wavelet can usually better localize the signal in time and frequency;

[0091] Symmetry: The symmetric wavelet basis function can avoid phase distortion of the signal when performing signal decomposition and transformation, which helps to maintain the phase information of the signal;

[0092] Regularity: Regularity is a measure of the reconstruction effect of the wavelet basis function. The wavelet basis function with regularity can effectively reduce errors and distortions in the signal reconstruction process;

[0093] Vanishing moment order: The vanishing moment order reflects the adaptability of the wavelet basis function to the smoothness of the signal. The larger the vanishing moment order, the better the wavelet basis function can compress the signal and suppress noise when the signal changes;

[0094] Support length: The support length is generally proportional to the support length. The longer the support length, the longer the calculation time of the wavelet transform, and the boundary problem will occur. The shorter the support length, the more difficult it is to concentrate the signal energy. Generally, a wavelet with a support length of 5-9 is selected;

[0095] The orthogonality, compact support, support length, symmetry, regularity, and vanishing moment order of several common wavelet basis functions are compared and analyzed.

[0096] The mortlet wavelet, the mexh wavelet and the Meyer wavelet do not have tight support, and the time-frequency characteristics of the signal cannot be effectively extracted when the signal is transformed by using the three wavelets, and the symN wavelet has tight support and orthogonality, and has better symmetry than the dbN wavelet and has better vanishing moment order and support length than the Haar wavelet.

[0097] In addition, considering the influence of the support length of the wavelet basis function on the calculation amount and the data compression and denoising effect of the vanishing moment order, the sym4 wavelet basis function is selected to decompose the leakage acoustic emission signal.

[0098] In addition to the selection of the wavelet basis function affecting the effect of the wavelet threshold denoising, the selection of the threshold function is also crucial, and the selection of the appropriate threshold function can ensure that the signal has good continuity in the reconstruction process and does not appear distorted; the hard threshold and the soft threshold will have some signal accuracy loss and distortion when processing the wavelet coefficients, in order to improve this phenomenon, the wavelet threshold function of the wavelet threshold denoising in step five is proposed; the improved wavelet threshold function can change the smoothness of the curve by adjusting θ, and can change the speed of the curve approaching the hard threshold waveform by adjusting β. The improved wavelet threshold function retains the advantages of the soft and hard threshold functions, avoids the phenomenon that the hard threshold is not continuous in the threshold interval, resulting in signal oscillation, and also improves the phenomenon that the soft threshold has a constant deviation in signal reconstruction.

[0099] In the embodiment, the wavelet inverse transform formula in step six is:

[0100] wherein, wherein, WT f (j, k) is a wavelet coefficient, ψ j,k (n) is a wavelet basis function, and A is a scale factor.

[0101] In order to verify the denoising effect of the improved wavelet threshold function, the denoising effect of different threshold functions is verified by using the Doppler standard signal with added Gaussian white noise.

[0102] The signal-to-noise ratio (SNR) and the root mean square error (RSME) of the noisy standard signal are 20.275 and 0.0982, respectively.

[0103] The signal-to-noise ratio (SNR) and the root mean square error (RSME) of the noisy standard signal are 20.275 and 0.0982, respectively.

[0104] In order to preserve the time and frequency domain characteristics of the leakage acoustic emission signal as much as possible and improve the recognition accuracy of the leakage signal.

[0105] The collected noise signal is processed by using the wavelet threshold denoising algorithm combined with the variational mode decomposition algorithm to remove the noise component. The wavelet threshold denoising algorithm can effectively filter out the high-frequency noise in the leakage acoustic emission signal, the VMD decomposition denoising algorithm has good complex data decomposition accuracy and good anti-noise interference, and the combined method can perform wavelet threshold denoising on different modal components after VMD decomposition, so as to achieve better denoising effect.

[0106] In order to ensure that the signal noise is effectively filtered out, the improved and optimized algorithm is used to perform denoising experiment on the leakage acoustic emission signal, the denoising effect is verified by using the standard signal, the signal-to-noise ratio and the root mean square error after denoising are calculated to evaluate the denoising effect, the joint denoising method is used to verify the denoising effect of the standard signal, and the SNR and RSME after denoising are 30.580 and 0.0299 respectively, it can be seen that the joint denoising effect is better than the effect of the single denoising algorithm.

[0107] After verifying that the joint denoising algorithm has better denoising effect, the real leakage acoustic signal collected is denoised;

[0108] The VMD parameters are optimized by using the optimization algorithm, and the optimal K and alpha are 12 and 526 respectively;

[0109] The parameters are used to improve the wavelet threshold to denoise the leakage acoustic emission signal, and finally realize the denoising of the leakage acoustic signal, so as to improve the signal-to-noise ratio of the leakage acoustic signal.

[0110] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for denoising a leakage acoustic signal based on an optimized VMD combined with improved wavelet thresholding, characterized in that, The denoising method comprises: Step one, obtaining a leakage sound signal of a leakage point; Step two, performing parameter optimization on a decomposition modal number K and a penalty factor a of VMD based on a hawk optimization algorithm, obtaining optimal parameters, and constructing a variational model using the optimal parameters; Step three, decomposing the leakage sound signal obtained in step one using the variational model constructed in step two to obtain a plurality of intrinsic modal components; Step four, screening the plurality of intrinsic modal components obtained in step three using a correlation coefficient method to screen out high-frequency components and low-frequency components; Step five, performing wavelet threshold denoising processing on the high-frequency components screened out in step four to obtain denoised high-frequency components; Step six, reconstructing the low-frequency components screened out in step four and the denoised high-frequency components obtained in step five using wavelet inverse transform to obtain a denoised leakage sound signal.

2. The method according to claim 1, wherein, The specific method for obtaining the leakage sound signal of the leakage point in step one is as follows: A vibration sensor is used to collect the leakage sound signal of the pipeline near the leakage point, the collection interval is 5 minutes, 72 pieces of leakage and non-leakage data are collected for each leakage point during the day and at night respectively, a total of 2x2x72=288 pieces of data are collected for each leakage point, 7 pieces of data of leakage points are collected, a total of 7x288=2016 pieces of sound emission data are collected, the sampling frequency of each piece of data is 50kHz, the sampling time is 0.5 seconds, and the length of each piece of data is 25k.

3. The method according to claim 1, wherein, The specific process of obtaining the optimal parameters by performing parameter optimization on the decomposition modal number K and the penalty factor a of VMD based on the hawk optimization algorithm in step two comprises an initialization stage, a prey identification stage, and a pursuit and escape stage; The initialization stage refers to that before the parameters are optimized by using an iterative algorithm, a certain number of initial solutions are randomly generated, the initial solutions represent the positions of hawks, and the initial solutions constitute a search space, and the quality of each initial solution is evaluated by a target function of a hawk population; The prey identification stage refers to that the hawk population performs global search on the search space to select an optimal region of the target function; The pursuit and escape stage refers to that as the number of iterations increases, the radius of the optimal region gradually decreases with the increase of the number of iterations, so that the sum of the modal bandwidths is minimized, and the optimal parameters are output.

4. The method according to claim 3, wherein, The position expression of the goshawk population is: where X represents the position matrix of the Goshawk population, X i is the initial solution of the ith individual, x i,j is the value of the jth dimension of the ith individual, N is the number of Goshawk population members, and m is the dimension of the problem space.

5. The method according to claim 4, wherein, The objective function of the goshawk population is represented as an objective function value vector: where F(X) is the term chain of the objective function obtained, F i is the objective function value obtained by the i-th solution.

6. The method according to claim 5, wherein, The expression of the population of the goshawk globally searching the searching space to select the optimal area where the target function is located is: i = X k i = 1, 2, …, N, k = 1, 2, …, i-1, i+1, …, N r ∈ [0, 1] where P i is the position of the prey of the ith hawk, F Pi is the target value of the ith hawk, is the new position of the i-th solution, is the value of the new position of the i-th solution in the j-th dimension, F i (new,P1) is the objective function value of the first phase of the Goshawk Optimization algorithm, I is a constant, the value of I is 1 or 2.

7. The method according to claim 6, wherein, The formula for outputting the optimal parameters is: Where, R is the parameter optimization range, t represents the number of current iterations, T is the maximum number of iterations, F i new,P2 is the objective function value of the second phase of the hawk algorithm optimization.

8. The method according to claim 7, wherein, The final optimization in step two obtains that the value range of the decomposition modal number K is 11-13, and the value range of the penalty factor a is 525-527.

9. The method according to claim 1, wherein, The wavelet threshold function expression of wavelet threshold denoising in step five is: wherein W j,k is the wavelet coefficient before denoising, The denoised wavelet coefficient is denoted as, the wavelet threshold is denoted as, the sign function is denoted as, the smoothness adjustment factor is denoted as, the approaching waveform speed adjustment factor is denoted as, and. β∈(0,1000]。 10. The method according to claim 1, wherein, The inverse wavelet transform formula in step six is: where WT f (j, k) are the wavelet coefficients, ψ j,k (n) is the wavelet basis function, A is the scale factor.

Images