A water supply pipeline multi-leakage point positioning method based on improved EEMD and convolution ICA hybrid

Abstract

This invention relates to a method for locating multiple leaks in water supply pipelines based on a hybrid approach of improved EEEMD and convolutional ICA, belonging to the field of water supply pipeline leak detection and location technology. The method first involves placing accelerometers on both sides of the pipeline under test to collect leakage vibration signals. Then, EEEMD is used to adaptively reduce the noise of the signals, automatically determining the white noise amplitude and overall average frequency based on the high-frequency components of the signal, and using NCF to select effective intrinsic mode components to complete signal reconstruction. Subsequently, CICA based on the information maximization criterion and feedback network structure is used to blindly separate the denoised mixed signal, extracting the relative time delay from each leak source to the sensor from the peak values ​​of the orthogonal filter coefficients in the demixing filter matrix. Finally, combining the pipeline sound wave propagation velocity and the two-point linear positioning formula, the precise locations of multiple leak points are calculated. This invention has the advantages of strong adaptability, significantly improved signal-to-noise ratio, accurate time delay estimation, and high positioning accuracy.

Description

Technical Field

[0001] This invention belongs to the field of water supply pipeline leakage detection and location technology, and relates to a multi-leakage location method for water supply pipelines based on a hybrid of improved ensemble empirical mode decomposition (IEEMD) and convolutional independent component analysis (CICA), which is applicable to the location of multiple leakage points in urban water supply pipelines. Background Technology

[0002] Water supply pipeline systems are a core component of urban infrastructure, and their safe and stable operation is directly related to daily life and industrial production. However, due to factors such as pipeline aging, geological subsidence, and construction disturbance, water supply pipeline leaks occur frequently, and the simultaneous existence of multiple leak points is particularly common in old pipe networks. Therefore, achieving rapid and accurate location of multiple leak points in water supply pipelines has significant engineering practical value and social significance.

[0003] Currently, acoustic vibration method is the most mature and widely used method for leak detection in water supply pipelines. It uses sensors to pick up the vibration sound signal generated by the leak, estimates the signal propagation time delay, and combines it with a location formula to determine the location of the leak. It has advantages such as low cost, simple operation, and strong anti-interference ability. However, the research and application of existing acoustic vibration method are mainly limited to single leak point location. Commonly used time delay estimation methods such as cross-correlation function method, cross-spectral phase difference spectrum method, and energy method can only solve the signal time delay of a single leak source. They cannot separate the characteristic information of each leak source from the mixed vibration signal of multiple leak points, making it difficult to achieve multi-leak point location.

[0004] Independent component analysis (ICA) provides a new approach for separating mixed signals from multiple leaks. It can separate statistically independent source signals from mixed observation signals and obtain time delays by estimating signal propagation path information. However, in actual water supply pipeline inspection, the observation signals collected by sensors are linear convolutional mixed signals after multiple leak source signals have propagated through multiple paths and with multiple delays, requiring processing with convolutional independent component analysis (CICA). At the same time, the detection signals inevitably contain interference from environmental noise, pipeline clutter, etc., and noise will seriously reduce the time delay estimation accuracy of CICA. Therefore, effective noise reduction of the detection signals must be performed before CICA processing.

[0005] Existing signal denoising methods have many shortcomings: Fourier transform filtering requires prior knowledge of the frequency distribution of the leaked signal, which is difficult to achieve in practice; wavelet analysis denoising requires manual selection of the wavelet basis and the number of decomposition levels, and the threshold is difficult to determine accurately, resulting in poor adaptability; Empirical Mode Decomposition (EMD) denoising does not require empirical parameters, but it suffers from mode aliasing and endpoint effects, resulting in poor decomposition performance; Ensemble Empirical Mode Decomposition (EEMD) denoising alleviates mode aliasing by adding white noise, but the amplitude of the white noise and the overall average number depend on manual selection, lacking adaptive criteria, and there is no clear method for selecting effective intrinsic mode functions (IMF) components, thus limiting the denoising effect and failing to meet the signal quality requirements of CICA time delay estimation.

[0006] In addition, existing CICA methods mostly adopt feedforward separation network structures, which have problems such as time "whitening" side effects, many separation parameters, and difficulty in extracting time delays. Furthermore, they do not combine the actual propagation characteristics of water supply pipelines to build a suitable hybrid model, making it difficult to accurately estimate the relative time delays of multiple leakage sources.

[0007] In summary, there is an urgent need for a new method for locating multiple leaks in water supply pipelines to solve the problems of low accuracy in locating multiple leaks, blind noise reduction parameters, inaccurate time delay estimation, and poor adaptability in existing technologies. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a method for locating multiple leaks in water supply pipelines based on an improved hybrid EEMD and convolutional ICA, which achieves adaptive noise reduction of the detection signal and improves the time delay estimation accuracy of convolutional independent component analysis, thereby accurately locating multiple leaks in water supply pipelines and solving the problems of low accuracy in existing multi-leak location, blind selection of noise reduction parameters, and poor effective signal extraction.

[0009] To achieve the above objectives, the present invention provides the following technical solution: Solution 1: A method for locating multiple leaks in water supply pipelines based on a hybrid of improved EEMD and convolutional ICA, specifically including the following steps: S1: Install acceleration sensors on both sides of the water supply pipeline with multiple leaks to collect the pipeline's acoustic vibration signals, which are recorded as the detection signals. and ; S2: An improved ensemble empirical mode decomposition (IEEMD) algorithm is used to analyze the detected signal. , Adaptive denoising was performed separately to obtain the reconstructed denoised signal. and ; S3: Employs a Convolutional Independent Component Analysis (CICA) algorithm based on the information maximization criterion and feedback network model to... and Blind source separation is performed, and the optimal orthogonal filter coefficients of the unmixed filter matrix are obtained by iterative updating using the stochastic gradient method. The peak positions of the optimal orthogonal filter coefficients are extracted, and the relative time delay from each leakage source to the two sensors is estimated. S4: Combining the propagation velocity v of the acoustic vibration signal in the water supply pipeline and the actual distance S between the two sensors, the distance from each leak source to the preset sensor is calculated using the two-point linear positioning method, thereby determining the specific location of multiple leak points.

[0010] Furthermore, in step S2, the adaptive noise reduction processing of the IEEMD algorithm specifically includes the following steps: S21: Calculate the standard deviation of the amplitude of the detected signal. The detection signal is then subjected to a single EMD decomposition, and the first IMF component obtained from the decomposition is taken as the high-frequency component of the signal. Calculate the standard deviation of the amplitude of this high-frequency component. ; S22: Calculate the ratio coefficient between the standard deviation of the high-frequency component amplitude and the standard deviation of the original detected signal amplitude. Substitute X into the standard deviation of white noise amplitude Calculation formula The amplitude ratio coefficient of the white noise required for EEMD is obtained. (Usually K=X / 4); S23: Decompose the relative error e( based on the desired signal) e is generally taken as 1%), combined with the formula Determine the overall mean frequency N of EEMD. If N≤20, then take N=20. S24: According to the coefficient K and the total average number N, different random Gaussian white noise is added to the detection signal multiple times. The signal after each noise addition is decomposed into EMD to obtain the corresponding IMF component. The IMF components of the same order obtained by N decompositions are averaged to obtain the final set of IMF components of the detection signal. Wherein IMF represents the intrinsic mode function. S25: Define the peak factor CF and normalized peak factor NCF. Using one detection signal as the decomposed signal and the other as the reference signal, calculate the cross-correlation function between each IMF component and the reference signal. Select the effective IMF components based on the NCF values. S26: Reconstruct the selected effective IMF components to obtain two reconstructed signals after noise reduction, denoted as... and .

[0011] Furthermore, in step S25, the formula for calculating the peak coefficient CF is:

[0012] in, The peak value of the k-th IMF component is denoted as . The peak value of the cross-correlation function between the k-th IMF component and the reference signal. Let be the i-th value of the cross-correlation function, and L be the data length of the cross-correlation function.

[0013] Furthermore, in step S3, the CICA based on the information maximization criterion and the feedback network model specifically includes the following steps: S31: Combining the signal propagation characteristics of water supply pipelines, a linear convolutional hybrid model for multiple leak points in water supply pipelines is established, and the reconstructed noise-reduced detection signal is represented as the convolution sum of the signals from each leak source and the transfer function of the propagation channel; S32: Construct a feedback-type convolutional blind separator network based on the information maximization criterion, use the stochastic gradient method as the optimization algorithm, and derive the iterative formula for the filter coefficients in the time domain; S33: Solve for the optimal orthogonal filter coefficients of the demixing filter matrix using an iterative algorithm to separate the equivalent leakage source signal; S34: Extract the optimal orthogonal filter coefficient sequence from the demixing filter matrix, determine the order corresponding to the maximum peak value of the optimal orthogonal filter coefficient sequence, and this order is the number of relative time delay points from the leakage source to the sensor. Combine this with the sampling frequency to convert it into the relative time delay between the two sensors.

[0014] Furthermore, in step S31, the expression for the established linear convolutional mixture model is:

[0015] in, Let be the detection signal of the i-th sensor, and n be the number of leakage sources. Let be the attenuation coefficient from the j-th leakage source to the i-th sensor. For the j-th leakage source, Let be the time delay from the j-th leakage source to the i-th sensor. This is a noise signal. This represents the convolution operation.

[0016] Furthermore, in step S4, the calculation formula for the two-point straight-line positioning method is as follows:

[0017] in, , Here, S represents the distance from the leak source to the two sensors, S is the actual distance between the two sensors, and v is the propagation speed of the acoustic vibration signal inside the pipe. The relative time delay from the leakage source to the two sensors.

[0018] Furthermore, in step S4, the propagation wave velocity v of the acoustic vibration signal inside the pipe is calculated using pipe parameters, and the calculation formula is as follows:

[0019] Where K is the elastic modulus of water, ρ is the density of water, D is the inner diameter of the pipe, e is the pipe wall thickness, E is the elastic modulus of the pipe material, and C is a dimensionless parameter of the pipe constraint effect, which is usually taken as 1.

[0020] Option 2: A multi-leakage location system for water supply pipelines based on a hybrid of improved EEMD and convolutional ICA, comprising: The signal acquisition module includes two accelerometers and a signal acquisition instrument, used to collect and transmit acoustic and vibration detection signals from both sides of the water supply pipeline. The accelerometers are IEPE piezoelectric accelerometers with an adjustable sampling frequency, the maximum sampling frequency being no less than 5000Hz. The signal acquisition instrument is connected to a computer for communication, enabling the transmission, storage, and algorithm execution of the detection signals.

[0021] The signal denoising module incorporates an improved ensemble empirical mode decomposition (IEEMD) algorithm to adaptively denoise the acquired detection signal and output the denoised reconstructed detection signal.

[0022] The delay estimation module incorporates a Convolutional Independent Component Analysis (CICA) algorithm based on the information maximization criterion and feedback network model. This algorithm is used to perform blind source separation on the denoised reconstructed detection signal and estimate the relative delay from the leakage source to the sensor.

[0023] The positioning calculation module is used to obtain the acoustic vibration wave velocity of the pipeline and the sensor spacing, and calculate the actual location of the leak source by combining the relative time delay and the two-point straight line positioning method, and output the positioning result.

[0024] The beneficial effects of this invention are as follows: (1) Achieve adaptive noise reduction of detection signal with significant noise reduction effect: This invention improves the traditional EEMD noise reduction method by extracting high-frequency information of the signal to adaptively determine the white noise amplitude ratio coefficient K and the overall average frequency N, which solves the problem that traditional EEMD parameters rely on human experience; at the same time, the NCF index is defined to achieve accurate selection of effective IMF components without prior knowledge of the frequency distribution of the leakage signal. It has strong adaptive capability, and the signal-to-noise ratio of the detection signal after noise reduction can be improved by more than 4dB, effectively suppressing the impact of noise on subsequent time delay estimation.

[0025] (2) Accurately estimate the relative time delay of multiple leakage sources and solve the problem of multiple leakage point separation: This invention combines the signal propagation characteristics of water supply pipelines and constructs a feedback convolutional blind separation network based on the information maximization criterion. It eliminates the time "whitening" side effect and parameter redundancy problem of feedforward networks. The demixing filter coefficients correspond one-to-one with the leakage source propagation channels. The optimal coefficients are solved iteratively by using the stochastic gradient ascent method. The relative time delay of multiple leakage sources is accurately extracted by the peak position of the coefficient sequence. The peak prominence is improved by more than 8 times. The reliability and accuracy of time delay estimation are greatly improved, which solves the core problem that traditional methods cannot separate mixed signals of multiple leakage points and the time delay estimation is inaccurate.

[0026] (3) High positioning accuracy and good stability: This invention combines IEEMD adaptive noise reduction with CICA time delay estimation, and finally achieves multi-leakage point positioning through the two-point straight line positioning method. After full-size experiments, compared with direct CICA positioning and traditional EEMD-CICA positioning methods, the average positioning error of this invention is reduced by more than 60%, and the average error of dual-leakage point positioning is less than 0.4m. After repeated experiments, the positioning results have good stability and high repeatability.

[0027] (4) Strong adaptability and easy to promote in engineering: The method of the present invention does not rely on prior information about pipe material and pipe diameter, and is applicable to water supply pipes of various materials such as cast iron and plastic. The detection equipment is a conventional accelerometer and signal acquisition instrument, which is low in cost and easy to operate. The whole process does not require manual intervention in parameter selection, has a high degree of automation, and can quickly realize on-site detection and positioning, which is easy to promote and apply in engineering in urban water supply network detection.

[0028] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is an overall flowchart of the method for locating multiple leakage sources in a water supply pipeline according to the present invention; Figure 2 This is a schematic diagram illustrating the leakage location of the present invention; Figure 3 EEMD decomposition flowchart; Figure 4 A schematic diagram of the feedback separation network structure; Figure 5A schematic diagram of a blind separation network structure that maximizes information. Figure 6 For detecting signals; Figure 7 The signal after EEMD noise reduction; Figure 8 These are the filter coefficients of the demixing filter matrix. Detailed Implementation

[0030] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] Please see Figures 1-5 This invention provides a method for locating multiple leaks in water supply pipelines based on a hybrid of improved EEMD and convolutional ICA, such as... Figure 1 As shown, the specific steps include: Step 1: Signal Acquisition Accelerometers were fixedly installed at predetermined locations on both sides of the water supply pipeline suspected of having multiple leak sources. Two leakage vibration detection signals (such as...) were simultaneously acquired by a signal acquisition device. Figure 2 ), denoted as and Set the sampling frequency to The sampling time is T.

[0032] Step 2, IEEMD adaptive noise reduction processing, includes the following steps: Step 2.1: Detect the signal , The improved ensemble empirical mode decomposition (IEEMD) algorithm is executed separately. By extracting high-frequency information of the signal, the white noise amplitude ratio coefficient K and the overall averaging number N are adaptively determined. The detected signal is decomposed into multiple rounds of EMD and averaged overall to obtain several intrinsic mode function (IMF) components.

[0033] The specific implementation process of the IEEMD algorithm is as follows: Step 2.1.1: Calculate the standard deviation of the amplitude of the input detection signal x(t). The formula is:

[0034] in, Let x(t) be the average value of the detection signal, and M be the data length of the detection signal. .

[0035] Step 2.1.2: Perform a single EMD decomposition on the detection signal x(t), and take the first IMF component obtained from the decomposition as the high-frequency component of the signal. Calculate the standard deviation of the amplitude of the high-frequency components. .

[0036] Step 2.1.3: Calculate the ratio coefficient X of the standard deviation of the high-frequency component amplitude to the standard deviation of the original signal amplitude. .

[0037] Step 2.1.4: Determine the standard deviation of the white noise amplitude. This leads to the white noise amplitude ratio coefficient. .

[0038] Step 2.1.5: Decompose the relative error e (take e = 1%) according to the desired signal decomposition, and combine it with the formula Determine the overall average frequency N. If the calculated N ≤ 20, then take N = 20 to avoid excessive decomposition error.

[0039] Step 2.1.6: According to the determined K and N, add different random Gaussian white noise to the detection signal x(t) multiple times, and perform EMD decomposition on the signal after each noise addition to obtain the corresponding IMF component.

[0040] Step 2.1.7: Perform an overall average of the IMF components of the same order obtained from N decompositions to obtain the final IMF component set of the detection signal x(t) (e.g., Figure 3 ).

[0041] Step 2.2: Define the peak factor CF and normalized peak factor NCF. Use one detection signal as the decomposed signal and the other as the reference signal. Calculate the cross-correlation function between each IMF component and the reference signal. Select the effective IMF component based on the NCF value.

[0042] The definitions of the peak factor CF and the normalized peak factor NCF, as well as the selection method for the effective IMF components, are as follows: Step 2.2.1: Let the IMF component of the decomposed signal be... (k=1,2,…,m, where m is the total number of IMF components), the reference signal is ,calculate and cross-correlation function .

[0043] Step 2.2.2: Define the peak coefficient CF, which characterizes the prominence of the peak of the cross-correlation function. The formula is:

[0044] in, Let L be the absolute value of the peak value of the cross-correlation function between the k-th IMF component and the reference signal, and L be the data length of the cross-correlation function.

[0045] Step 2.2.3: For all IMF components The values ​​are normalized to obtain the normalized peak coefficient. The formula is:

[0046] in, For all The maximum value among the values.

[0047] Step 2.2.4, Select The IMF components are considered as valid IMF components, and these components contain a large number of leaked valid signals.

[0048] Step 2.2.5: Exchange the decomposed signal and the reference signal, and repeat steps 2.2.1 to 2.2.4 to complete the selection of the effective IMF components of the two detection signals.

[0049] Step 2.3: Reconstruct the selected effective IMF components to obtain two reconstructed signals after noise reduction, denoted as... and .

[0050] Step 3: CICA relative time delay estimation based on the information maximization criterion and feedback network model, specifically including the following steps: Step 3.1: Based on the signal propagation characteristics of water supply pipelines, establish a linear convolutional hybrid model for multiple leakage sources, and use the denoised reconstructed signal as the model input.

[0051] Step 3.2: Construct a feedback convolutional blind separation network structure based on the information maximization (informax) criterion, using the reconstructed signal as input, design the demixing filter matrix and determine the ideal separation conditions.

[0052] Step 3.3: Derive the information maximization objective function and use the stochastic gradient method to iteratively update the orthogonal filter coefficients of the unmixed filter matrix until the objective function converges, thus obtaining the optimal orthogonal filter coefficients.

[0053] Step 3.4: Extract the peak positions of the optimal orthogonal filter coefficient sequence and convert the peak indices into the actual relative time delays from each leakage source to the two sensors. (j=1,2,…,n, where n is the number of leakage sources).

[0054] Step 4, Multi-Leakage Point Location Calculation, specifically includes the following steps: Step 4.1: Calculate the propagation velocity v of the leakage sound wave in the pipe based on parameters such as the material, inner diameter, and wall thickness of the water supply pipe.

[0055] The propagation velocity of sound waves is calculated based on the physical parameters of the pipe and water. The formula is:

[0056] Where K is the elastic modulus of water, ρ is the density of water, D is the inner diameter of the pipe, e is the pipe wall thickness, E is the elastic modulus of the pipe material, and C is the dimensionless parameter of the pipe constraint effect, which is taken as C=1.

[0057] Step 4.2: Measure the actual straight-line distance S between the two sensors, and combine it with the relative time delay obtained in Step 3. The two-point straight-line positioning method is used to calculate the actual distance from each leakage source to the preset sensor, and to determine the specific location of multiple leakage points.

[0058] Two-point linear positioning method: Calculate the distances from a leak source to sensor 1 and sensor 2 as follows: , The distance between the two sensors is S, and the relative time delay is... , United The distance from the leakage source to sensor 1 can be obtained by solving:

[0059] Based on the calculation The system determines the specific location of the leak source between the two sensors, and calculates the location of all leak sources in sequence to achieve multi-leak location.

[0060] Example 1: This embodiment provides a CICA relative delay estimation method based on information maximization and feedback network model (corresponding to step 3 of the method of this invention). Taking a dual-leakage source as an example (which can be directly extended to multiple leakage sources), the specific process is as follows: Step 3.1: Establish a linear convolutional mixture model Ignoring the dispersion effect of sound wave propagation in the pipeline (or classifying it as noise), a noisy linear convolutional hybrid model with two leakage sources and two sensors is established:

[0061] in, Let be the vibration signal of the j-th leakage source, which satisfies a statistically independent and non-Gaussian distribution; Let be the propagation path transfer function from the j-th leakage source to the i-th sensor, and be a causal finite impulse response filter; This is a temporal convolution operation; To detect noise.

[0062] The propagation channel transfer function is simplified to a time-delay attenuation filter, and the model is further simplified to:

[0063] in, This is the amplitude attenuation coefficient. Let be the absolute propagation delay from the j-th leakage source to the i-th sensor.

[0064] Applying a z-transform to the simplified model transforms the convolution operation into a multiplication operation. The z-transform domain model is as follows:

[0065] in, To transform the column vector of the detection signal z, Let A(z) be the z-transform column vector of the leakage source signal, and let A(z) be the z-transform domain mixing matrix containing attenuation and time delay information.

[0066] Step 3.2: Construct a feedback-based convolutional blind separator network structure To overcome the shortcomings of feedforward networks, a feedback-type convolutional blind separator network (such as...) is constructed. Figure 4 ), to reconstruct the signal after noise reduction The input is the signal source, and the output is the estimated value of the leakage source signal. The input-output relationship of the network in the z-transform domain is as follows:

[0067] Summarized as follows:

[0068] Where I is a 2-order identity matrix; This is the demixing filter matrix, with zeros on the main diagonal, containing only orthogonal filter terms. , All of them are finite-order FIR filters, and the peak positions of their coefficients correspond to the relative time delay of the leakage source.

[0069] The ideal separation condition is that the output estimated signal and the actual leakage source signal only differ in amplitude scaling, i.e. , The matrix is ​​a non-zero diagonal scaling matrix. In this case, the demixing filter matrix and the mixing matrix satisfy... This achieves a one-to-one correspondence between the demixing filter and the propagation channel.

[0070] Step 3.3: Derive the information maximization objective function Selecting a sigmoid monotonic nonlinear activation function Network output After activation function

[0071] Based on the information maximization principle, maximize The joint entropy is equivalent to minimizing the mutual information of the output signals (e.g., Figure 5 This enables independent separation of signal statistics; By combining variable substitution and the Jacobian determinant, the joint entropy is transformed into a form with the filter coefficients w in the unmixed filter matrix as variables, ultimately yielding the practical objective function:

[0072] Among them, E[ [This represents the mathematical expectation.] It is the first derivative of the activation function.

[0073] Step 3.4: Iteratively solve for the coefficients of the optimal unmixed filter using the stochastic gradient method. The stochastic gradient method is used to iteratively update the filter coefficients of the unmixed filter matrix, approximating the mathematical expectation with instantaneous values ​​to improve iteration efficiency; the core formula for coefficient iteration update is:

[0074] Where k is the number of iterations, and η is the learning rate (taken as 0.001). 0.01), Let the gradient of the objective function at step k be denoted; derive the orthogonal filtering terms respectively. , The iterative formula for the FIR filter coefficients is used to update the coefficients successively and determine convergence. (ε=10) 6 If the convergence threshold is reached or the maximum number of iterations is reached, the iteration stops, and the optimal orthogonal filter coefficient sequence is obtained. , .

[0075] Step 3.5: Extract the relative time delay, which specifically includes the following steps: Step 3.5.1: Extract the optimal orthogonal filter coefficient sequence , L is the filter order; Step 3.5.2: Find the maximum absolute value of the two coefficient sequences respectively, and the corresponding subscripts are the relative time delay points n1 and n2; Step 3.5.3, combined with sampling frequency The formula for converting latency points to actual relative latency is as follows:

[0076] in, , The relative propagation delay from the two leakage sources to the two sensors; Step 3.5.4: Introduce the peak factor CF to verify the validity of the peak value. If CF > 5, it is a valid peak value, ensuring accurate time delay extraction.

[0077] Example 2: This embodiment uses the dual-leakage point location of a cast iron water supply pipeline as an example to verify the effectiveness of the method of the present invention. The water supply pipeline has an inner diameter of 200mm, a wall thickness of 10mm, and a burial depth of 1.5m. The specific implementation steps are as follows: Step 1: Signal Acquisition Two IEPE piezoelectric accelerometers (sensor 1 and sensor 2) were installed at 38m intervals on both sides of the cast iron water supply pipeline suspected of having two leaks. The sensors were fixed to the outer wall of the pipeline inside the inspection well. The sampling frequency of the signal acquisition instrument was set. =5000Hz, sampling time T=2s, synchronously acquire two leakage vibration detection signals. and Data length M=10000, the collected inspection signals are as follows Figure 6 As shown.

[0078] Step 2: IEEMD Adaptive Noise Reduction Processing 2.1 Detection signal Perform IEEMD decomposition: 2.1.1 Calculation yields average Amplitude standard deviation ; 2.1.2 After a single EMD decomposition, the first IMF component is taken as the high-frequency component, and the following is calculated: Ratio coefficient ; 2.1.3 Determine the white noise amplitude ratio coefficient Combining the expected decomposition error e=1%, the overall average frequency is calculated. ; 2.1.4, according to , right Multiple rounds of EMD decomposition and overall averaging were performed to obtain 10 IMF components; Similarly, for Performing IEEMD decomposition yields... , This yields 10 IMF components.

[0079] 2.2 Selecting effective IMF components: 2.2.1, with To decompose the signal, Using the reference signal as a reference, calculate the cross-correlation function between each IMF component and the reference signal to obtain the peak coefficient of each component. ; 2.2.2. After normalization, we get Select IMF2, IMF3, and IMF4 are valid IMF components; 2.2.3. Exchange the decomposed signal and the reference signal. Complete the selection of effective IMF components, and the results are consistent with... Consistent.

[0080] 2.3 Signal Reconstruction: The effective IMF components of the two detection signals are reconstructed separately to obtain the noise-reduced reconstructed signal. and ,like Figure 7 As shown.

[0081] Step 3: CICA Relative Delay Estimation Based on Information Maximization and Feedback Network 3.1 Establish a linear convolutional hybrid model with dual leakage sources, simplify the propagation channel to a time-delayed attenuation filter, and ignore dispersion effects and residual noise; 3.2 Construct a feedback-type convolutional blind separator network, setting the demixing filter order L=1000, learning rate η=0.005, and maximum iteration steps to 5000. Reconstruct the signal... and Input network; 3.3. Deriving the information maximization objective function and iteratively updating the orthogonal filter terms using the stochastic gradient ascent method. , The coefficients of the objective function converge at step 3200, with a convergence threshold ε=10. 6. Obtain the optimal orthogonal filter coefficient sequence. , ; 3.4 Extracting the peak positions of the coefficient sequence, such as... Figure 8 As shown, The peak index n1=108, The peak index n2=70, and the peak coefficients CF are 16.21 and 14.02 respectively, both greater than 5, indicating that they are effective peak values; 3.5. Convert to actual relative time delay: , .

[0082] Step 4: Calculation of multiple leak points 4.1 Calculation of sound wave propagation velocity in the pipe: Elastic modulus of water K = 2.1 × 10⁻⁶ 9 Pa, the density of water ρ = 1000 kg / m³, and the elastic modulus of cast iron E = 1.1 × 10⁻⁶. 11 Given Pa, pipe inner diameter D = 0.2 m, pipe wall thickness e = 0.01 m, constraint parameter C = 1, substituting into the formula yields:

[0083] 4.2 Two-point linear positioning calculation: The distance between the two sensors is S=38m. Substituting the relative time delay and wave velocity into the positioning formula, we obtain: the distance from leakage source 1 to sensor 1: Distance from leakage source 2 to sensor 1: The results are highly consistent with the actual locations of the leak source (31.5m and 10.8m) as preset in the experiment, with positioning errors of 0.08m and 0.02m, respectively.

[0084] Example 3: The method of this invention is not only applicable to dual-leakage location, but can also be directly extended to the location of water supply pipelines with three or more leaks. It only requires expanding the dimensions of the convolutional hybrid model and the feedback convolutional blind separation network according to the number of leak sources, and iteratively solving the demixing filter coefficients of the corresponding dimensions to extract the relative time delay of each leak source and complete the location. Its core principle and steps are consistent with the dual-leakage embodiment.

[0085] Example 4: Furthermore, the method of the present invention is applicable to water supply pipes made of other materials such as plastic and steel. It only requires replacing the elastic modulus E of the corresponding pipe material when calculating the sound wave propagation velocity. No other adjustments are needed to the method itself, making it highly adaptable.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for locating multiple leaks in water supply pipelines based on a hybrid of improved EEMD and convolutional ICA, characterized in that, Includes the following steps: S1: Install acceleration sensors on both sides of the water supply pipeline with multiple leaks to collect the pipeline's acoustic vibration signals, which are recorded as the detection signals. and ; S2: The IEEMD algorithm is used to detect the signal. , Adaptive denoising was performed separately to obtain the reconstructed denoised signal. and Where IEEMD represents the improved set empirical mode decomposition; S3: The CICA algorithm based on the information maximization criterion and feedback network model is used to... and Blind source separation is performed, and the optimal orthogonal filter coefficients of the unmixed filter matrix are obtained by iterative updating using the stochastic gradient method. The peak positions of the optimal orthogonal filter coefficients are extracted, and the relative time delay from each leakage source to the two sensors is estimated; where CICA represents convolutional independent component analysis. S4: Combining the propagation speed of acoustic vibration signals within the water supply pipeline v and the actual distance between the two sensors S The distance from each leak source to the preset sensor is calculated using the two-point straight-line positioning method, thereby determining the specific location of multiple leak points.

2. The method for locating multiple leaks in a water supply pipeline according to claim 1, characterized in that, In step S2, the adaptive noise reduction processing of the IEEMD algorithm specifically includes the following steps: S21: Calculate the standard deviation of the amplitude of the detected signal. The detection signal is then subjected to a single EMD decomposition, and the first IMF component obtained from the decomposition is taken as the high-frequency component of the signal. Calculate the standard deviation of the amplitude of this high-frequency component. ; S22: Calculate the ratio coefficient between the standard deviation of the high-frequency component amplitude and the standard deviation of the original detected signal amplitude. ,Will X Substitute the standard deviation of the white noise amplitude Calculation formula The amplitude ratio coefficient of white noise is obtained. ; S23: Decompose the relative error according to the desired signal e Combined with formula Determine the overall average frequency N ; S24: According to the coefficient K and the overall average frequency N, different random Gaussian white noise is added to the detection signal multiple times. EMD decomposition is performed on the signal after each noise addition to obtain the corresponding IMF component; N The IMF components of the same order obtained from the decomposition are averaged to obtain the final set of IMF components of the detected signal; where IMF represents the intrinsic mode function. S25: Define the peak factor CF and normalized peak factor NCF. Using one detection signal as the decomposed signal and the other as the reference signal, calculate the cross-correlation function between each IMF component and the reference signal. Select the effective IMF components based on the NCF values. S26: Reconstruct the selected effective IMF components to obtain two reconstructed signals after noise reduction.

3. The method for locating multiple leaks in a water supply pipeline according to claim 2, characterized in that, In step S25, the formula for calculating the peak coefficient CF is: in, For the first k Peak coefficients of each IMF component, For the first k The peak value of the cross-correlation function between each IMF component and the reference signal. The first cross-correlation function i Each possible value L The data length of the cross-correlation function.

4. The method for locating multiple leaks in a water supply pipeline according to claim 1, characterized in that, In step S3, the CICA based on the information maximization criterion and the feedback network model specifically includes the following steps: S31: Combining the signal propagation characteristics of water supply pipelines, a linear convolutional hybrid model for multiple leak points in water supply pipelines is established, and the reconstructed noise-reduced detection signal is represented as the convolution sum of the signals from each leak source and the transfer function of the propagation channel; S32: Construct a feedback-type convolutional blind separator network based on the information maximization criterion, use the stochastic gradient method as the optimization algorithm, and derive the iterative formula for the filter coefficients in the time domain; S33: Solve for the optimal orthogonal filter coefficients of the demixing filter matrix using an iterative algorithm to separate the equivalent leakage source signal; S34: Extract the optimal orthogonal filter coefficient sequence from the demixing filter matrix, determine the order corresponding to the maximum peak value of the optimal orthogonal filter coefficient sequence, and this order is the number of relative time delay points from the leakage source to the sensor. Combine this with the sampling frequency to convert it into the relative time delay between the two sensors.

5. The method for locating multiple leaks in a water supply pipeline according to claim 4, characterized in that, In step S31, the expression for the established linear convolutional mixture model is: in, For the first i The detection signals from each sensor n Number of leak sources For the first j The leak source to the first i The attenuation coefficient of each sensor, For the first j Acoustic and vibration signals from a leakage source For the first j The leak source to the first i The delay of each sensor, This is a noise signal. This represents the convolution operation.

6. The method for locating multiple leaks in a water supply pipeline according to claim 4, characterized in that, In step S4, the calculation formula for the two-point straight-line positioning method is as follows: in, , These represent the distances from the leak source to the two sensors, respectively. S This represents the actual distance between the two sensors. v The propagation wave speed of acoustic vibration signals inside the pipe. The relative time delay from the leakage source to the two sensors.

7. The method for locating multiple leaks in a water supply pipeline according to claim 6, characterized in that, In step S4, the propagation wave speed of the acoustic vibration signal inside the pipe v The calculation is obtained through pipeline parameters, and the calculation formula is as follows: in, K The elastic modulus of water, ρ The density of water, D The inner diameter of the pipe. e For pipe wall thickness, E The elastic modulus of the pipe material. C This is a dimensionless parameter representing the constraint effect of the pipeline.

8. A system for implementing the method for locating multiple leaks in a water supply pipeline according to any one of claims 1 to 7, characterized in that, The system includes: The signal acquisition module, including two accelerometers and a signal acquisition instrument, is used to acquire and transmit acoustic and vibration detection signals from both sides of the water supply pipeline. The signal noise reduction module has a built-in IEEMD algorithm, which is used to perform adaptive noise reduction processing on the acquired detection signal and output the reconstructed detection signal after noise reduction. The delay estimation module incorporates the CICA algorithm based on the information maximization criterion and feedback network model, which is used to perform blind source separation on the denoised reconstructed detection signal and estimate the relative delay from the leakage source to the sensor. The positioning calculation module is used to obtain the acoustic vibration wave velocity of the pipeline and the sensor spacing, and calculate the actual location of the leak source by combining the relative time delay and the two-point straight line positioning method, and output the positioning result.

9. The system according to claim 8, characterized in that, The accelerometer is an IEPE piezoelectric accelerometer with an adjustable sampling frequency, the maximum sampling frequency being no less than 5000Hz; the signal acquisition instrument is connected to a computer for communication, enabling the transmission, storage, and algorithm execution of the detection signal.

Images