A water supply pipe network leak point positioning method based on deep learning and hydrophone array
By deploying hydrophone arrays in the water supply network and utilizing a deep learning feature extraction network to dynamically correct the sound wave propagation model, the problem of insufficient accuracy and reliability in leak location in existing technologies has been solved, achieving high-precision leak detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI SHANGYUAN WATER TECHNOLOGY GROUP CO LTD
- Filing Date
- 2026-03-17
- Publication Date
- 2026-06-09
AI Technical Summary
Existing methods for locating leaks in water supply networks are inadequate in terms of signal feature extraction, propagation model adaptability, and anti-interference capabilities, making it difficult to meet the requirements for high-precision and high-reliability location.
A method based on deep learning and hydrophone arrays is adopted. The hydrophone array is deployed to collect sound wave signals. The deep learning feature extraction network automatically learns the high-dimensional features of the water leakage sound wave. Combined with the geometric position of the hydrophone array, the first leak location parameters and the second leak location parameters are calculated. The sound wave propagation model is dynamically corrected to locate the leak.
It significantly improves the accuracy and robustness of feature extraction, enhances the precision and reliability of leak location, and can adapt to complex and ever-changing pipeline environments, providing an efficient and accurate leak detection solution.
Smart Images
Figure CN121897877B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water supply network leak detection technology, and in particular to a method for locating water supply network leaks based on deep learning and hydrophone arrays. Background Technology
[0002] Pipeline leaks are a common and serious problem in the operation of water supply networks, not only wasting water resources but also potentially causing secondary disasters such as road collapses and water pollution. Therefore, quickly and accurately locating leaks is crucial for ensuring water supply safety and reducing economic losses. Currently, acoustic leak detection technology is widely used due to its advantages such as non-contact operation and real-time performance. This technology infers the leak location by picking up the acoustic signals generated when water leaks, analyzing the signal characteristics, and combining this with the sensor position. However, the actual pipeline environment is complex. The propagation of acoustic waves in pipelines is affected by various factors such as pipeline material, burial medium, branch structure, and environmental noise, resulting in non-stationary acoustic signals with low signal-to-noise ratios, posing a significant challenge to leak location.
[0003] Traditional leak location methods mainly include sound wave propagation, correlation analysis, and time delay estimation based on wave velocity. Sound wave propagation relies on human experience, resulting in low efficiency and high subjectivity. Correlation analysis calculates time delay using the cross-correlation function of two signals, but it is susceptible to noise interference and requires high signal similarity. Time delay estimation based on wave velocity requires accurate knowledge of the sound wave propagation speed, which varies with frequency and pipe conditions in reality, making precise acquisition difficult. These methods typically utilize only a single signal characteristic (such as time delay), ignoring the sound wave spectral structure and spatial attenuation information, leading to limited location accuracy, especially in scenarios with small or multiple leaks.
[0004] In recent years, with the development of sensor array technology, researchers have begun to use arrays composed of multiple hydrophones to locate leaks using beamforming or time-of-arrival techniques. Array methods can improve location resolution by utilizing spatial information, but they still face the following problems: First, the extraction of acoustic signal features usually relies on manually designed features (such as short-time energy, zero-crossing rate, etc.), which are difficult to fully characterize the complex characteristics of leaking acoustic waves; second, existing methods often assume that the acoustic wave propagation speed is constant or use empirical attenuation models, without considering the dispersion and attenuation changes during actual propagation, leading to model mismatch; in addition, environmental noise and differences in pipe structure can seriously affect signal quality, and the performance of traditional signal processing methods deteriorates sharply under low signal-to-noise ratio conditions.
[0005] Deep learning has demonstrated powerful feature learning capabilities in signal processing, automatically extracting high-level abstract features from raw data, making it particularly suitable for processing non-stationary, low signal-to-noise ratio acoustic signals. However, directly applying deep learning to leak location presents challenges: deep learning models require a large amount of labeled training data, while obtaining samples with precise leak location is costly in practice; furthermore, the features output by deep learning models often lack physical interpretability, making them difficult to directly use for parameter correction in the localization model. Therefore, combining the features extracted by deep learning with physical propagation models to improve the accuracy and robustness of localization has become a current research hotspot.
[0006] In summary, existing methods for locating leaks in water supply networks have shortcomings in signal feature extraction, propagation model adaptability, and anti-interference capabilities, making it difficult to meet the demands for high-precision and high-reliability positioning in practical engineering. There is an urgent need for an intelligent positioning method that can fully utilize multi-dimensional information from acoustic signals, adaptively adjust propagation parameters, and resist environmental interference. Summary of the Invention
[0007] To address the technical problems of insufficient signal feature extraction, poor adaptability of propagation models, and insufficient anti-interference capability in existing technologies, this invention provides a method for locating leaks in water supply networks based on deep learning and hydrophone arrays.
[0008] The technical solution provided by this invention is as follows:
[0009] This invention provides a method for locating leaks in water supply networks based on deep learning and hydrophone arrays, comprising:
[0010] S1: Deploy an array of multiple hydrophones within the monitoring area of the water supply network to acquire the acoustic signals collected by each hydrophone.
[0011] S2: Preprocess the acquired acoustic signal to obtain preprocessed signal data;
[0012] S3: Input the preprocessed signal data into a pre-trained deep learning feature extraction network to obtain a high-dimensional feature vector that reflects the characteristics of the water leakage acoustic wave.
[0013] S4: Calculate the first leak location parameter and the second leak location parameter based on the high-dimensional feature vector; wherein, the first leak location parameter is used to characterize the spectral characteristics of the leak sound source, and the second leak location parameter is used to characterize the spatial attenuation characteristics of the sound wave propagating along the pipe.
[0014] S5: Based on the first leak location parameters and the second leak location parameters, and combined with the geometric position of the hydrophone array, the specific location of the leak is determined by the location algorithm.
[0015] The beneficial effects of the technical solution provided by this invention include at least the following:
[0016] (1) In this invention, sound wave signals are collected by deploying a hydrophone array, and high-dimensional features of the leaking sound waves are automatically learned by a deep learning feature extraction network, avoiding the limitations and subjectivity of traditional manual feature extraction. The deep learning network can extract deep features related to the location of the leak from the original signal. Even if the environmental noise is strong or the signal is not stationary, it can obtain stable feature representations, thereby significantly improving the accuracy and robustness of feature extraction and laying a solid foundation for subsequent accurate positioning.
[0017] (2) In this invention, based on the high-dimensional features extracted by deep learning, the first leak location parameter and the second leak location parameter are calculated, which respectively characterize the spectral characteristics of the leak sound source and the spatial attenuation characteristics of the sound wave propagating along the pipe. These two parameters have clear physical meanings and can dynamically correct the wave velocity and attenuation coefficient in the sound wave propagation model, making the model more consistent with the actual propagation environment. Combined with the geometric position of the hydrophone array, the location algorithm is used to solve the leak location, which effectively overcomes the location deviation caused by the fixed model and single parameters in the traditional method, and greatly improves the accuracy and reliability of leak location.
[0018] (3) In this invention, principal component analysis is used to reduce the dimensionality of the high-dimensional feature vector, and the main feature components are extracted for parameter calculation, which reduces computational redundancy and improves processing efficiency. At the same time, after determining the location of the leak, the localization is repeated and the results are fused. Weighted averaging or cluster analysis is used to further suppress random errors, thereby enhancing the stability and anti-interference ability of the method. These optimization steps enable the entire localization method to adapt to the complex and ever-changing pipeline network environment, providing an efficient and accurate solution for leak detection in water supply networks. Attached Figure Description
[0019] Figure 1 This is a flowchart illustrating a method for locating leaks in a water supply network based on deep learning and a hydrophone array, provided as an embodiment of the present invention. Detailed Implementation
[0020] The technical solution of the present invention will now be described with reference to the accompanying drawings.
[0021] Reference manual attached Figure 1 The diagram illustrates a flowchart of a water supply network leak location method based on deep learning and hydrophone array provided by an embodiment of the present invention.
[0022] This invention provides a method for locating leaks in water supply networks based on deep learning and hydrophone arrays. The processing flow may include the following steps:
[0023] S1: An array consisting of multiple hydrophones is deployed within the monitoring area of the water supply network to acquire the acoustic signals collected by each hydrophone.
[0024] An array of multiple hydrophones is deployed within the monitoring area of the water supply network to acquire the acoustic signals collected by each hydrophone. Specifically, the placement of the hydrophones is determined based on the topology of the water supply network and the monitoring range. Multiple hydrophones are installed along the pipeline route to ensure that each hydrophone can effectively pick up the acoustic signals generated by fluid leakage within the pipeline. The hydrophones can be piezoelectric or fiber optic sensors, and the acoustic signals output by each hydrophone are recorded synchronously via a data acquisition card. The spatial coordinates of each hydrophone are also recorded, forming the array's geometric layout.
[0025] S2: Preprocess the acquired acoustic signal to obtain preprocessed signal data.
[0026] The acquired acoustic signals are preprocessed to obtain preprocessed signal data. Preprocessing includes basic operations such as denoising, filtering, and normalization of the original acoustic signals to eliminate environmental interference and sensor noise, enhance the effective components of the signal, and ensure the signal meets the input requirements for subsequent deep learning feature extraction. The preprocessed signal data maintains its time-series structure and has a unified sampling rate and amplitude range.
[0027] S3: Input the preprocessed signal data into a pre-trained deep learning feature extraction network to obtain a high-dimensional feature vector that reflects the characteristics of the water leakage acoustic wave.
[0028] The preprocessed signal data is input into a pre-trained deep learning feature extraction network to obtain high-dimensional feature vectors reflecting the characteristics of the leaking sound waves. This deep learning feature extraction network adopts a convolutional neural network structure and is trained on a large number of historical leaking sound wave samples. It can automatically learn and extract deep features related to the location of the leak from the preprocessed signal and output a set of high-dimensional feature vectors. These feature vectors condense the frequency domain, time domain, and spatial domain information of the leaking sound waves.
[0029] S4: Calculate the first leak location parameter and the second leak location parameter based on the high-dimensional feature vector. The first leak location parameter characterizes the spectral characteristics of the leak sound source, and the second leak location parameter characterizes the spatial attenuation characteristics of the sound wave propagating along the pipe.
[0030] Based on the high-dimensional feature vector, a first leak location parameter and a second leak location parameter are calculated. The first leak location parameter characterizes the spectral characteristics of the leaking sound source; by performing frequency domain analysis on the high-dimensional feature vector, values related to the sound source's frequency distribution are extracted. The second leak location parameter characterizes the spatial attenuation characteristics of sound waves propagating along the pipe; by performing spatial domain analysis on the high-dimensional feature vector, the degree of sound wave energy attenuation with propagation distance is calculated. These two parameters together describe the propagation law of leaking sound waves in the pipe.
[0031] S5: Based on the first leak location parameters and the second leak location parameters, and combined with the geometric position of the hydrophone array, the specific location of the leak is determined by the location algorithm.
[0032] Based on the first and second leak location parameters, and combined with the geometric position of the hydrophone array, a location algorithm is used to determine the specific location of the leak. Specifically, the first leak location parameter is used to correct the propagation speed of sound waves in the pipe, and the second leak location parameter is used to correct the energy attenuation model of the sound waves. Then, combined with the spatial coordinates of each hydrophone, location methods such as time difference of arrival or beamforming are used to calculate the precise position of the leak relative to the array, thereby achieving automatic location of leaks in the water supply network.
[0033] In one possible implementation, calculating the first leak location parameter in step S4 specifically includes the following steps:
[0034] S401: Perform wavelet packet decomposition on the preprocessed signal of each hydrophone to obtain energy sequences of several sub-bands;
[0035] S402: Calculate the weighted average of the energy sequences of each sub-band to obtain the spectral energy distribution vector;
[0036] S403: Based on the spectral energy distribution vector, the first leak location parameter α is calculated using the following formula:
[0037] ;
[0038] Where N represents the number of hydrophones and M represents the number of sub-bands. Let i be the energy of the i-th hydrophone in the j-th sub-band. Let j be the center frequency of the j-th sub-band. The preset reference frequency, Let be the straight-line distance from the i-th hydrophone to the preset reference point. σ represents the average distance from all hydrophones to the reference point, and σ is a preset distance distribution scale parameter.
[0039] Steps S401 to S403 further define the calculation of the first leak location parameter in step S4 of weight 1. Specifically, the pre-processed signal from each hydrophone is first decomposed into wavelet packets, dividing the signal into multiple sub-bands and obtaining the energy sequence of each sub-band. Then, the weighted average of the energy sequences of each sub-band is calculated to obtain the spectral energy distribution vector. Finally, the first leak location parameter α is calculated based on the spectral energy distribution vector using a given formula, where N is the number of hydrophones and M is the number of sub-bands. Let i be the energy of the i-th hydrophone in the j-th sub-band. Let j be the center frequency of the j-th sub-band. The preset reference frequency, Let be the straight-line distance from the i-th hydrophone to the preset reference point. σ represents the average distance from all hydrophones to the reference point, and σ is a preset distance distribution scale parameter.
[0040] In one possible implementation, calculating the second leak location parameter in step S4 specifically includes the following steps:
[0041] S411: Perform a short-time Fourier transform on the preprocessed signal of each hydrophone to extract the main frequency amplitude sequence at each moment;
[0042] S412: Calculate the logarithmic attenuation rate of the main frequency amplitude between adjacent hydrophones;
[0043] S413: Based on the logarithmic attenuation rate of all adjacent hydrophone pairs, calculate the second leak location parameter β using the following formula:
[0044] ;
[0045] Where L is the number of adjacent hydrophone pairs. Let k be the straight-line distance between the kth pair of adjacent hydrophones. and are the main frequency amplitude values of the signals received by the k-th and (k+1)-th hydrophones, respectively, and γ is a preset adjustment coefficient. Let be the short-time Fourier transform spectrum amplitude of the k-th hydrophone signal at time t, where t is the time index.
[0046] Steps S411 to S413 further refine the calculation of the second leak location parameter in step S4 of weight 1. Specifically, firstly, a short-time Fourier transform is performed on the preprocessed signal of each hydrophone to extract the dominant frequency amplitude sequence at each moment. Then, the logarithmic attenuation rate of the dominant frequency amplitude between adjacent hydrophones is calculated. Finally, the second leak location parameter β is calculated using a given formula based on the logarithmic attenuation rates of all adjacent hydrophone pairs, where L in the formula represents the number of adjacent hydrophone pairs. Let k be the straight-line distance between the kth pair of adjacent hydrophones. and are the main frequency amplitude values of the signals received by the k-th and (k+1)-th hydrophones, respectively, and γ is a preset adjustment coefficient. Let be the short-time Fourier transform spectrum amplitude of the k-th hydrophone signal at time t, where t is the time index.
[0047] In one possible implementation, the deployment of the hydrophone array in S1 specifically includes: installing hydrophones at equal or non-equal intervals along the direction of the water supply pipe, and recording the spatial coordinates of each hydrophone to form an array geometric layout.
[0048] The specific method for deploying the hydrophone array involves installing hydrophones at equal or non-equal intervals along the water supply pipeline and recording the spatial coordinates of each hydrophone to form the array's geometric layout. This layout data will be used for geometric calculations in subsequent positioning algorithms.
[0049] In one possible implementation, the preprocessing in S2 includes the following steps: bandpass filtering the acquired acoustic signal to remove environmental noise, and then performing amplitude normalization to make the signal amplitude fall within a uniform range.
[0050] The specific steps of preprocessing are as follows: bandpass filtering is performed on the acquired acoustic signal to remove environmental noise, followed by amplitude normalization to ensure that the signal amplitude falls within a uniform range. The preprocessed signal has a consistent amplitude scale, which facilitates subsequent feature extraction.
[0051] In one possible implementation, the deep learning feature extraction network in S3 is a convolutional neural network, and its training process includes: using historical leak sound wave samples and their corresponding leak point location labels to train the network in a supervised learning manner, so that the high-dimensional feature vector output by the network can effectively distinguish the sound wave characteristics of leak points at different locations.
[0052] The deep learning feature extraction network is a convolutional neural network. Its training process involves using historical leak sound wave samples and their corresponding leak location labels to train the network in a supervised learning manner, enabling the high-dimensional feature vector output by the network to effectively distinguish the sound wave characteristics of leaks at different locations. The trained network can then be directly used to extract features from the input signal.
[0053] In one possible implementation, the positioning algorithm in S5 specifically includes: calculating the possible area of the leak point based on the first leak point positioning parameters and the second leak point positioning parameters, combined with the geometric position of the hydrophone array, using a triangulation positioning method based on the time difference of arrival, and obtaining the final position through iterative optimization.
[0054] The specific implementation of the localization algorithm involves calculating the possible areas of the leak based on the first and second leak location parameters, combined with the geometric position of the hydrophone array, using a triangulation method based on time difference of arrival. The final location is then obtained through iterative optimization. This algorithm utilizes two parameters to correct the propagation model, thereby improving localization accuracy.
[0055] In one possible implementation, S5, which combines the geometric position of the hydrophone array, includes: establishing a sound wave propagation path model based on the pipe topology and hydrophone coordinates, and correcting the propagation speed and attenuation coefficient in the model using the first and second leak point location parameters.
[0056] By combining the geometric position of the hydrophone array with the pipe topology and hydrophone coordinates, a sound wave propagation path model is established, and the propagation velocity and attenuation coefficient in the model are corrected using the first and second leak point location parameters. The corrected model can more accurately reflect the actual sound wave propagation characteristics.
[0057] In one possible implementation, before calculating the first leak location parameter and the second leak location parameter in S4, a step of performing principal component analysis to reduce the dimensionality of the high-dimensional feature vector is included to extract the main feature components for parameter calculation.
[0058] Before calculating the first and second leak location parameters, a principal component analysis (PCA) step is performed on the high-dimensional feature vector to reduce its dimensionality, extracting the main feature components for parameter calculation. The dimensionality-reduced feature vector retains key information and reduces computational complexity.
[0059] In one possible implementation, after determining the specific location of the leak in S5, the process further includes repeating S1 to S5 to obtain multiple positioning results, and performing weighted averaging or cluster analysis on these results to improve positioning accuracy and robustness.
[0060] After determining the specific location of the leak, steps S1 through S5 are repeated to obtain multiple positioning results. These results are then weighted and averaged or clustered to improve positioning accuracy and robustness. Multiple measurements and fusion processing can effectively reduce the error of a single positioning.
[0061] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:
[0062] (1) In this invention, sound wave signals are collected by deploying a hydrophone array, and high-dimensional features of the leaking sound waves are automatically learned by a deep learning feature extraction network, avoiding the limitations and subjectivity of traditional manual feature extraction. The deep learning network can extract deep features related to the location of the leak from the original signal. Even if the environmental noise is strong or the signal is not stationary, it can obtain stable feature representations, thereby significantly improving the accuracy and robustness of feature extraction and laying a solid foundation for subsequent accurate positioning.
[0063] (2) In this invention, based on the high-dimensional features extracted by deep learning, the first leak location parameter and the second leak location parameter are calculated, which respectively characterize the spectral characteristics of the leak sound source and the spatial attenuation characteristics of the sound wave propagating along the pipe. These two parameters have clear physical meanings and can dynamically correct the wave velocity and attenuation coefficient in the sound wave propagation model, making the model more consistent with the actual propagation environment. Combined with the geometric position of the hydrophone array, the location algorithm is used to solve the leak location, which effectively overcomes the location deviation caused by the fixed model and single parameters in the traditional method, and greatly improves the accuracy and reliability of leak location.
[0064] (3) In this invention, principal component analysis is used to reduce the dimensionality of the high-dimensional feature vector, and the main feature components are extracted for parameter calculation, which reduces computational redundancy and improves processing efficiency. At the same time, after determining the location of the leak, the localization is repeated and the results are fused. Weighted averaging or cluster analysis is used to further suppress random errors, thereby enhancing the stability and anti-interference ability of the method. These optimization steps enable the entire localization method to adapt to the complex and ever-changing pipeline network environment, providing an efficient and accurate solution for leak detection in water supply networks.
[0065] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for locating leaks in a water supply network based on deep learning and a hydrophone array, characterized in that, include: S1: Deploy an array of multiple hydrophones within the monitoring area of the water supply network to acquire the acoustic signals collected by each hydrophone. S2: Preprocess the acquired acoustic signal to obtain preprocessed signal data; S3: Input the preprocessed signal data into a pre-trained deep learning feature extraction network to obtain a high-dimensional feature vector that reflects the characteristics of the water leakage acoustic wave. S4: Based on the high-dimensional feature vector, calculate the first leak location parameter and the second leak location parameter; wherein, the first leak location parameter is used to characterize the spectral characteristics of the water leakage sound source. The calculation method includes performing wavelet packet decomposition on the preprocessed signal of each hydrophone to obtain several sub-band energy sequences and calculating their weighted average value to form a spectral energy distribution vector. Then, combined with the hydrophone spatial position weight and the logarithmic correction of the sub-band center frequency, the first leak location parameter α characterizing the spectral characteristics of the water leakage sound source is calculated using the following formula: ； Where N represents the number of hydrophones and M represents the number of sub-bands. Let i be the energy of the i-th hydrophone in the j-th sub-band. Let j be the center frequency of the j-th sub-band. The preset reference frequency, Let be the straight-line distance from the i-th hydrophone to the preset reference point. The average distance from all hydrophones to the reference point is σ, which is a preset distance distribution scale parameter. The second leak location parameter is used to characterize the spatial attenuation characteristics of sound waves propagating along the pipe. The calculation method includes performing a short-time Fourier transform on the preprocessed signals of each hydrophone, extracting the dominant frequency amplitude sequence at each moment, and calculating the logarithmic attenuation rate of the dominant frequency amplitude between adjacent hydrophones. Then, combined with the normalization of the distance between adjacent hydrophones and the dynamic correction of the difference in the time-domain spectrum amplitude of the signal, the second leak location parameter β, which characterizes the spatial attenuation characteristics of sound waves propagating along the pipe, is calculated using the following formula: ； Where L is the number of adjacent hydrophone pairs. Let k be the straight-line distance between the kth pair of adjacent hydrophones. and are the main frequency amplitude values of the signals received by the k-th and (k+1)-th hydrophones, respectively, and γ is a preset adjustment coefficient. Let be the short-time Fourier transform spectrum amplitude of the k-th hydrophone signal at time t, where t is the time index; S5: Based on the first leak location parameters and the second leak location parameters, and combined with the geometric position of the hydrophone array, the specific location of the leak is determined by the location algorithm.
2. The method for locating leaks in a water supply network based on deep learning and a hydrophone array as described in claim 1, characterized in that, The calculation of the first leak location parameters in S4 specifically includes the following steps: S401: Perform wavelet packet decomposition on the preprocessed signal of each hydrophone to obtain energy sequences of several sub-bands; S402: Calculate the weighted average of the energy sequences of each sub-band to obtain the spectral energy distribution vector; S403: Calculate the first leak location parameter α using the formula based on the spectral energy distribution vector.
3. The method for locating leaks in a water supply network based on deep learning and a hydrophone array as described in claim 1, characterized in that, The calculation of the second leak point location parameters in S4 specifically includes the following steps: S411: Perform a short-time Fourier transform on the preprocessed signal of each hydrophone to extract the main frequency amplitude sequence at each moment; S412: Calculate the logarithmic attenuation rate of the main frequency amplitude between adjacent hydrophones; S413: Calculate the second leak location parameter β using the formula based on the logarithmic attenuation rate of all adjacent hydrophone pairs.
4. The method for locating leaks in a water supply network based on deep learning and a hydrophone array as described in claim 1, characterized in that, The deployment of the hydrophone array in S1 specifically includes: installing hydrophones at equal or non-equal intervals along the direction of the water supply pipeline, and recording the spatial coordinates of each hydrophone to form an array geometric layout.
5. The method for locating leaks in a water supply network based on deep learning and a hydrophone array according to claim 1, characterized in that, The preprocessing in S2 includes the following steps: bandpass filtering is performed on the acquired acoustic signal to remove environmental noise, and then amplitude normalization is performed to make the signal amplitude fall within a uniform range.
6. The method for locating leaks in a water supply network based on deep learning and a hydrophone array as described in claim 1, characterized in that, The deep learning feature extraction network in S3 is a convolutional neural network. Its training process includes: using historical leak sound wave samples and their corresponding leak point location labels to train the network in a supervised learning manner, so that the high-dimensional feature vector output by the network can effectively distinguish the sound wave characteristics of leak points at different locations.
7. The method for locating leaks in a water supply network based on deep learning and a hydrophone array according to claim 1, characterized in that, The positioning algorithm in S5 specifically includes: based on the first leak location parameters and the second leak location parameters, combined with the geometric position of the hydrophone array, a triangulation positioning method based on the time difference of arrival is used to calculate the possible area of the leak, and the final position is obtained through iterative optimization.
8. The method for locating leaks in a water supply network based on deep learning and a hydrophone array according to claim 1, characterized in that, The S5 step of combining the geometric position of the hydrophone array includes: establishing a sound wave propagation path model based on the pipe topology and hydrophone coordinates, and using the first and second leak point location parameters to correct the propagation speed and attenuation coefficient in the model.
9. The method for locating leaks in a water supply network based on deep learning and a hydrophone array according to claim 1, characterized in that, Before calculating the first and second leak location parameters in step S4, a step of principal component analysis is also included to reduce the dimensionality of the high-dimensional feature vector in order to extract the main feature components for parameter calculation.
10. A method for locating leaks in a water supply network based on deep learning and a hydrophone array, as described in claim 1, characterized in that... After determining the specific location of the leak in S5, the process further includes repeating S1 to S5 to obtain multiple positioning results, and performing weighted averaging or cluster analysis on these results to improve positioning accuracy and robustness.