Water supply network leakage noise detection method and system and medium

Abstract

The invention relates to the technical field of pipe network acoustic detection, in particular to a water supply pipe network leakage noise detection method and system and a medium. The method comprises the following steps: constructing a physical model of leakage noise based on fluid mechanics and a sound propagation rule; generating simulated leakage noise data based on physical model parameterization so as to simulate leakage noise data under extreme physical conditions; a set of the actually measured leakage noise data and the simulated leakage noise data is used as leakage noise sample data, and time domain statistical characteristics and frequency domain statistical characteristics of the signals are obtained through quantification according to the leakage noise sample data; based on the statistical characteristics, dynamically adjusting window function parameters and performing short-time Fourier transform processing to obtain a resolution-optimized time-frequency image; and constructing a CNN classification model, training the model based on the resolution-optimized time-frequency image, and outputting prediction probabilities of leakage noise and environmental noise. While the water supply pipe network leakage noise detection precision is improved, hardware dependence is eliminated, and efficient pure acoustic end-to-end pipe network leakage detection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of pipe network acoustic detection, and in particular to a method, system and medium for detecting leakage noise in a water supply pipe network. Background Art

[0002] Water supply network leakage detection is a core link in ensuring the sustainable use of urban water resources and is directly related to infrastructure safety. Accurate detection of water supply network leakage helps prevent secondary disasters such as foundation collapse and pipe network bursting. At the same time, it reduces energy consumption caused by ineffective pressurization of water pumps and avoids water waste through rapid repairs. It is a key technical support for smart water services to achieve water conservation and emission reduction and enhance urban resilience.

[0003] To accurately identify urban water supply network noise collected by noise recorders, determine whether there is leakage, and improve the efficiency of technicians in interpreting big data, traditional water supply network noise leakage detection technologies typically rely on manual listening and noise recording, which have high misjudgment rates and low efficiency. Later, machine learning methods based on sound spectrum analysis gradually emerged, but they are often limited to manual feature extraction (MFCC, energy spectrum), and still have the following detailed technical issues:

[0004] 1. Data level: Measured leakage data is scarce, samples of extreme scenarios (such as micro-leaks and high-pressure pipes) are insufficient, and environmental noise interference is complex, resulting in poor model generalization and low recognition rate for small leaks and low signal-to-noise ratio samples;

[0005] 2. Preprocessing: General filtering (such as bandpass filtering) cannot remove environmental noise interference that overlaps with the leakage band, resulting in a low signal-to-noise ratio in the time-frequency image and masking of effective features.

[0006] 3. Time-frequency analysis: Existing technologies generally believe that dynamically adjusting STFT parameters increases computational complexity. However, fixed STFT parameters cannot take into account both low-frequency resonance peaks and high-frequency transient pulses, resulting in blurred time-frequency structures of weak signals and incomplete extraction of leakage features. This makes it difficult to achieve a balance between computational complexity and feature information loss.

[0007] 4. Classification model level: Existing manual features (MFCC, or Mel-Frequency Cepstral Coefficients) rely on prior knowledge and have difficulty capturing nonlinear time-frequency patterns, resulting in high false positive rates in complex scenarios and insufficient model robustness.

[0008] The above technical problems form a vicious cycle, that is, low-quality data + inefficient feature extraction leads to inevitable performance bottlenecks in the classification model.

[0009] To this end, Chinese patent CN119123338A, a large-scale model-based method for monitoring urban and rural water supply network leakage, proposed a technical route of "audio signal → spectrum heat map → CNN large model → krill swarm algorithm positioning". While some improvements have been made, the following core issues still exist:

[0010] 1. Data level: To address the scarcity of sample data in extreme scenarios, this patent relies on high-density sensor deployment. It collects water pressure data by adding multiple sub-monitoring points along the target pipeline and constructs an objective function by simulating leakage conditions. This inherently results in insufficient acoustic detection accuracy and requires additional hardware compensation, resulting in high costs and violating the original intention of acoustic leak detection technology to "reduce hardware dependence."

[0011] 2. Preprocessing: Directly converting audio signals into spectral heatmaps is susceptible to interference from external noise. Although the patent uses pre-emphasis processing, it does not specifically optimize measures for environmental noise interference. Therefore, this general preprocessing operation still cannot effectively remove environmental noise, resulting in a low signal-to-noise ratio of the spectral heatmap.

[0012] 3. Time-frequency analysis: This patent still uses fixed STFT parameters for conversion (e.g., the pre-emphasis coefficient α is (0.9, 1)). This results in the inability of time-frequency analysis to take into account both low-frequency and high-frequency characteristics, which in turn blurs weak signal features, affecting the subsequent CNN large-scale model detection accuracy and krill swarm positioning.

[0013] 4. Question on the classification model level: Although the patent uses a large CNN model for audio signal classification, model training relies on historical audio data, and extreme scene samples in historical data are still scarce, resulting in poor generalization of the trained model. In addition, the positioning part of the krill swarm algorithm also relies on the joint positioning of water pressure data and audio data. There is a cognitive bias that "leakage positioning must rely on multimodal data fusion", which undoubtedly further increases the hardware deployment requirements and detection complexity. Summary of the Invention

[0014] In response to the above technical problems, the present invention proposes a water supply network leakage noise detection method, system and medium, aiming to improve the accuracy of water supply network leakage noise detection while eliminating hardware dependence and realizing efficient pure acoustic end-to-end network leakage detection.

[0015] In a first aspect, the present application provides a method for detecting leakage noise in a water supply network, comprising the following steps:

[0016] A physical model of leakage noise is constructed based on fluid mechanics and sound propagation laws, and the physical model is used to simulate noise signals under different physical conditions;

[0017] Based on the physical model of leakage noise, parameterized simulation leakage noise data is generated to simulate leakage noise data under extreme physical conditions;

[0018] Acquire measured leakage noise data, and use the set of the measured leakage noise data and the simulated leakage noise data as leakage noise sample data;

[0019] Based on the leakage noise sample data, the time domain statistical characteristics and frequency domain statistical characteristics of the signal are quantified;

[0020] Based on the time-domain and frequency-domain statistical characteristics of the signal, the window function parameters are dynamically adjusted and short-time Fourier transform processing is performed to obtain a time-frequency image with optimized resolution;

[0021] Build a CNN classification model and train it based on the time-frequency images with optimized resolution;

[0022] The measured leakage noise data is converted into a time-frequency graph and input into the trained CNN classification model, which outputs the predicted probability of leakage noise and ambient noise.

[0023] The technical concept of this application is to deeply integrate physical modeling, parameterized data generation, feature quantization, dynamic short-time Fourier transform adjustment and CNN training. First, a leakage noise physical model is constructed based on fluid mechanics and sound propagation laws, and simulated leakage noise data covering extreme scenarios is generated parameterizedly. Then, it is combined with the measured data to form high-quality samples. Then, the time domain statistical characteristics and frequency domain statistical characteristics of the signal are quantified to accurately capture the signal characteristics. Based on this, the window function parameters (such as window type, window length and overlap rate) are dynamically adjusted to match the signal dynamic characteristics and short-time Fourier transform processing is performed to optimize the time-frequency image resolution to fully extract low-frequency resonance peaks and high-frequency transient pulse features. Subsequently, the CNN classification model is trained based on the resolution-optimized time-frequency image to improve the model's ability to capture nonlinear time-frequency patterns, and finally, a purely acoustic end-to-end classification prediction of the measured leakage noise data is achieved, abandoning the inertial thinking of water pressure joint positioning and breaking the cognitive bias that "leakage positioning must rely on multimodal data fusion."

[0024] In some embodiments, a physical model of leakage noise is constructed based on fluid mechanics and sound propagation laws, including:

[0025] Obtain the pipe material parameters, leakage diameter and fluid pressure in the water supply network;

[0026] Based on the pipe material parameters, leakage aperture, and fluid pressure in the water supply network, the baseline amplitude of leakage noise is determined and the fluid turbulence simulation value is generated;

[0027] Based on the pipe material parameters of the water supply network, the impulse response of the bandpass filter is determined to simulate the frequency response characteristics of different pipe materials;

[0028] Based on the baseline amplitude of leakage noise, fluid turbulence simulation value and bandpass filter impulse response, a physical model of leakage noise is constructed.

[0029] In some embodiments, parameterized generation of simulated leakage noise data based on a physical model of leakage noise includes:

[0030] Adjust the leakage aperture and the fluid pressure parameters in the pipeline, and generate enhanced parameters under extreme physical conditions based on parameterized generation rules;

[0031] Substitute the enhanced parameters into the physical model of leakage noise to obtain simulated leakage noise data.

[0032] In some embodiments, the time-domain statistical feature includes kurtosis, and the time-domain statistical feature of the signal is quantified based on the leakage noise sample data, including: , Where Kurt represents the kurtosis of the signal, x(n) represents the signal, μ represents the mean of the signal, σ represents the standard deviation of the signal, and E represents the expectation operator;

[0033] The frequency domain statistical features include the spectrum centroid, and based on the leakage noise sample data, the frequency domain statistical features of the signal are quantified, including:

[0034] Perform frame division and windowing processing on leakage noise sample data;

[0035] Perform discrete Fourier transform calculation on each frame after windowing to obtain the amplitude of the signal at each frequency component;

[0036] The spectral centroid is calculated based on the amplitude of the signal at each frequency component.

[0037] In some embodiments, based on the time-domain statistical characteristics and frequency-domain statistical characteristics of the signal, dynamically adjusting the window function parameters and performing short-time Fourier transform processing to obtain a time-frequency image with optimized resolution includes:

[0038] The kurtosis of the signal is compared with the preset kurtosis threshold in real time. If the kurtosis of the signal is greater than the preset kurtosis threshold, the window type is adjusted to the Hamming window. If the kurtosis of the signal is less than or equal to the preset kurtosis threshold, the window type is adjusted to the Hann window.

[0039] Dividing the spectrum centroid into a plurality of first frequency intervals, and adjusting the window length according to the first frequency interval in which the spectrum centroid of the signal is located, so that the spectrum centroid value is inversely proportional to the selected window length;

[0040] Dividing the spectrum centroid into a plurality of second frequency intervals, and adjusting the overlap ratio according to the second frequency interval in which the spectrum centroid of the signal is located, so that the spectrum centroid value is proportional to the selected overlap ratio;

[0041] Based on the dynamically adjusted window type, window length and overlap ratio, the signal is processed by short-time Fourier transform to obtain a time-frequency image with optimized resolution.

[0042] In some embodiments, it further includes:

[0043] Performing a delay operation on the leakage noise sample data to obtain a delayed signal;

[0044] The delayed signal is input into the adaptive filter for filtering, and the filter weights are iteratively updated using the LMS algorithm;

[0045] The signals output by the filter are coherently accumulated to obtain the coherently accumulated output signal as the leakage noise sample data after preprocessing.

[0046] In some embodiments, coherent accumulation is performed on the signals output by the filter to obtain a coherently accumulated output signal, which is expressed as: , Where m represents the number of accumulations, y(n) represents the signal output by the adaptive filter, i represents the total number of signals selected for coherent accumulation, and Δ i Indicates the small delay increment of the i-th signal.

[0047] In some embodiments, the CNN classification model adopts a ResNet network architecture.

[0048] In a second aspect, the present application provides a water supply network leakage noise detection system, comprising:

[0049] A physical model building module is used to build a physical model of leakage noise based on fluid mechanics and sound propagation laws. The physical model is used to simulate noise signals under different physical conditions.

[0050] A simulation data generation module is used to parameterize and generate simulated leakage noise data based on the physical model of leakage noise to simulate leakage noise data under extreme physical conditions;

[0051] A sample data integration module is used to obtain measured leakage noise data and use the set of measured leakage noise data and simulated leakage noise data as leakage noise sample data;

[0052] The statistical feature quantization module is used to quantify the time domain statistical features and frequency domain statistical features of the signal based on the leakage noise sample data;

[0053] The time-frequency image optimization module is used to dynamically adjust the window function parameters and perform short-time Fourier transform processing based on the time-domain and frequency-domain statistical characteristics of the signal to obtain a time-frequency image with optimized resolution;

[0054] The classification model building module is used to build a CNN classification model and train the CNN classification model based on the resolution-optimized time-frequency image;

[0055] The leakage noise detection module is used to convert the measured leakage noise data into a time-frequency graph, input it into the trained CNN classification model, and output the predicted probability of leakage noise and environmental noise.

[0056] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a water supply network leakage noise detection method as described above.

[0057] The beneficial technical effects of the present invention include at least:

[0058] 1. A water supply network leakage noise detection method, system and medium are used to achieve pure acoustic end-to-end classification prediction of measured leakage noise data by deeply integrating physical modeling, parameterized data generation, feature quantization, dynamic short-time Fourier transform adjustment and CNN training. Specifically, this application ensures that the simulation data in extreme scenarios conforms to the real fluid mechanics and acoustic laws through physical modeling of leakage noise, and combines it with the measured data to form high-quality leakage noise sample data, so as to more accurately quantify the time domain statistical characteristics and frequency domain statistical characteristics of the sample data, and then drive the dynamic adjustment of the window function parameters of the short-time Fourier transform to generate higher resolution time-frequency images, providing highly robust input for the CNN classification model. Parameters in each link are automatically derived from physical properties, thus globally and collaboratively breaking the vicious cycle of "data scarcity, feature ambiguity, and poor model generalization." At the same time, it abandons the inertial thinking of water pressure joint positioning and breaks the cognitive bias that "leakage positioning requires multimodal data fusion." This effectively addresses the existing problems of insufficient extreme scene samples, high environmental noise interference, the imbalance between computational complexity and feature information loss, and the high cost caused by high dependence on additional hardware (such as water pressure sensors). It significantly improves the detection accuracy of water supply network leakage noise, eliminates hardware dependence, and achieves efficient pure acoustic end-to-end classification.

[0059] 2. This application constructs a physical model of leakage noise based on fluid mechanics and the laws of sound propagation. Based on the physical model of leakage noise, it directly generates simulated noise signal samples in extreme scenarios that conform to physical laws by adjusting relevant pipeline parameters, effectively solving the technical problems of scarcity and uneven distribution of measured data. At the same time, this application combines physical modeling and data enhancement. Instead of simply generating data using existing GANs, it simulates the acoustic characteristics of leakage under different pipe materials and pressures based on fluid mechanics and the laws of sound propagation, making the subsequent analysis steps closer to real scenarios and laying a good data foundation for eliminating the need for water pressure sensors and other requirements through pure acoustic methods.

[0060] 3. The frequency distribution of environmental noise is relatively dispersed, and the main frequency band energy distribution of some signals is curved and low-frequency; while the main frequency band energy of leakage noise is relatively continuous and concentrated, with a moderate frequency range. To this end, this application designs an adaptive filtering strategy for optimizing STFT parameters based on the time-frequency characteristics of leakage noise, targeting the characteristic difference between leakage noise and environmental noise. Specifically, according to the time domain statistical characteristics of the signal waveform, the optimal solution for spectrum leakage control and resolution enhancement is adaptively selected (window function and kurtosis linkage), and according to the distribution characteristics of signal energy in the frequency domain, the balance point of time resolution and frequency resolution is adaptively adjusted (window length and spectrum centroid linkage), and the time-frequency grid density is quadratically optimized according to the signal frequency characteristics to ensure feature continuity while improving computational efficiency (overlap rate and spectrum centroid quadratically linkage), thereby better suppressing the environmental noise in the leakage noise, but retaining the inherent characteristics of the environmental noise, and providing a good classification basis for the subsequent training of the CNN classification model. At the same time, the prior art generally believes that dynamically adjusting parameters will increase computational complexity, so fixed STFT parameters are usually used. However, the present application can achieve real-time adaptive adjustment of window function parameters (window type, window length, overlap rate) through lightweight calculation of the statistical characteristics of the signal (i.e., kurtosis and spectral centroid), thereby generating a time-frequency image with optimized resolution, realizing parameterized adaptive STFT time-frequency analysis, avoiding information loss of manual features, fully capturing leakage characteristics, and effectively improving the clarity of time-frequency features, the accuracy of high-frequency transient positioning, and the computational efficiency of the low-frequency band, so that the time-frequency representation always matches the physical characteristics of the current signal, which helps to improve the micro-leakage detection rate of the water supply network;

[0061] 4. Traditional bandpass filters cannot separate environmental noise interference that overlaps with the leakage band, and the environmental noise and leakage signal have similar statistical characteristics. Conventional filtering will cause signal distortion and have the technical problem of insufficient non-stationary noise suppression. Moreover, the signal-to-noise ratio of micro-leakage is usually less than -10dB, making it difficult for existing technical methods to effectively extract weak signals. For example, the signal-to-noise ratio of the spectrum heat map is low, which will mask effective features. To this end, this application utilizes the broadband correlation of the leakage signal (which is uncorrelated with the environmental noise) and destroys the correlation of the interference signal by delaying J. Then, the current input x(n) is used as the desired output, forcing the filter to track signal changes. Without the need for a preset reference signal, it can adapt to non-stationary environments. Then, the periodic repeatability of the leakage signal (such as pipeline resonance) is utilized to achieve coherence enhancement by phase fine-tuning Δi, separate the same-frequency noise, effectively improve the signal-to-noise ratio, provide a high-signal-to-noise ratio input signal for subsequent time-frequency analysis, reduce the spectrum centroid / kurtosis calculation error, and thus optimize the STFT parameter selection to avoid parameter misselection caused by environmental noise interference.

[0062] Other features and advantages of the present invention will be disclosed in detail in the following specific embodiments and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The present invention will be further described below with reference to the accompanying drawings:

[0064] Figure 1 This is a flow chart of a method for detecting leakage noise in a water supply network according to a first embodiment of the present invention.

[0065] Figure 2 Schematic diagram showing the comparison before and after preprocessing of the second embodiment of the present invention.

[0066] Figure 3 This is a structural diagram of a water supply network leakage noise detection system according to embodiment three of the present invention. DETAILED DESCRIPTION

[0067] The following is an explanation and description of the technical solutions of the embodiments of the present invention in conjunction with the drawings of the embodiments of the present invention. However, the following embodiments are only preferred embodiments of the present invention and are not exhaustive. Based on the embodiments in the implementation manner, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention.

[0068] In the following description, terms such as "inside", "outside", "up", "down", "left", "right", etc. that indicate directions or positional relationships are only used to facilitate the description of the embodiments and simplify the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0069] Example 1:

[0070] Please see the attached Figure 1 , Figure 1 A flow chart of a method for detecting leakage noise in a water supply network provided in one embodiment of this specification is shown.

[0071] like Figure 1 As shown, the water supply network leakage noise detection method may include at least the following steps:

[0072] S1, a physical model of leakage noise is constructed based on fluid mechanics and the laws of sound propagation. The physical model is used to simulate noise signals under different physical conditions.

[0073] Specifically, according to the leakage noise characteristics and prior information, in this embodiment, a physical model of leakage noise is constructed based on fluid mechanics and sound propagation laws, including:

[0074] S11, obtaining pipe material parameters, leakage aperture, and fluid pressure in the water supply network;

[0075] S12 , based on the pipe material parameters of the water supply network, the leakage aperture d, and the fluid pressure P in the pipeline, a reference amplitude A0 of the leakage noise is determined, and a fluid turbulence simulation value ΔA·ξ(n) is generated.

[0076] It is understood that the reference amplitude A0 represents the average energy of the leakage noise, and is used to establish the physical relationship between pressure, aperture, and sound energy. It can be pre-designed according to the actual leakage scenario of the water supply network, and this embodiment does not limit this. For example: , Among them, ΔA is the empirical fluctuation intensity (for example, ΔA=0.3A0), represents Gaussian white noise, and the fluid turbulence simulation value ΔA·ξ(n) is used to simulate the random pulsation characteristics of fluid turbulence.

[0077] S13, based on the pipe material parameters of the water supply network, determine the bandpass filter impulse response h(n) to simulate the frequency response characteristics of different pipe materials, such as 2kHz for steel and 800Hz for PVC.

[0078] S14, based on the baseline amplitude of the leakage noise, the fluid turbulence simulation value, and the bandpass filter impulse response, a physical model of the leakage noise is constructed, which can be expressed as: , in, Indicates the slowly varying random amplitude, A0 is the reference amplitude, ΔA is the empirical fluctuation intensity (for example, ΔA=0.3A0), represents Gaussian white noise, ΔA·ξ(n) represents the fluid turbulence simulation value, which is used to simulate the random pulsation characteristics of fluid turbulence; g(n) represents Gaussian white noise, which simulates the random emission characteristics of the sound source with a constant power spectrum density; h(n) represents the impulse response of the bandpass filter, which is used to simulate the frequency response characteristics of the pipe.

[0079] S2, based on the physical model of leakage noise, parameterizes and generates simulated leakage noise data to simulate leakage noise data under extreme physical conditions.

[0080] For example, extreme scenarios include but are not limited to tiny leaks (d < 5 mm), high-pressure conditions (P > 0.5 MPa), and other scenarios.

[0081] Specifically, in this embodiment, based on the physical model of leakage noise, parameterized generation of simulated leakage noise data includes:

[0082] S21, adjust the leakage aperture and the fluid pressure parameters in the pipeline, and generate enhanced parameters under extreme physical conditions based on parameterized generation rules.

[0083] The parameterized generation rules can be pre-designed based on rare extreme scenarios, which is not limited in this embodiment. For example, the parameterized generation rules can include enhanced parameters under high-pressure pipeline (P>0.5MPa) working conditions, expressed as: , That is, by shifting the angular frequency to a low frequency, the pipe wall resonance effect caused by high pressure is simulated.

[0084] S22, substituting the enhanced parameters into the physical model of the leakage noise to obtain simulated leakage noise data.

[0085] Furthermore, the simulated leakage noise data can be combined with background environmental noise (such as pink noise with 1 / f attenuation in urban environments, and white noise + fan harmonics in suburban environments) and interference pulses or equipment vibrations to obtain a synthetic signal as the final simulated noise data. This embodiment does not limit this.

[0086] It is understandable that traditional data enhancement (such as noise addition and time shift) cannot generate acoustic features that conform to the laws of fluid mechanics. However, this embodiment constructs a physical model of leakage noise based on fluid mechanics and the laws of sound propagation. On the basis of the physical model of leakage noise, it directly generates simulated noise signal samples in extreme scenarios that conform to the laws of physics by adjusting the relevant parameters of the pipeline, effectively solving the technical problems of scarcity and uneven distribution of measured data. At the same time, the idea of combining physical modeling and data enhancement in this embodiment is not to simply generate data using existing GANs, but to simulate the acoustic features of leakage under different pipe materials and pressures based on fluid mechanics and the laws of sound propagation, so that the subsequent analysis steps are closer to the real scene, laying a good data foundation for eliminating the need for water pressure sensors and the like through pure acoustic methods.

[0087] S3, obtaining measured leakage noise data, and taking a set of the measured leakage noise data and the simulated leakage noise data as leakage noise sample data.

[0088] S4, based on the leakage noise sample data, quantify the time domain statistical characteristics and frequency domain statistical characteristics of the signal.

[0089] Specifically, in this embodiment, the time domain statistical characteristics include kurtosis, and the time domain statistical characteristics of the signal are quantified based on the leakage noise sample data, including: , Where Kurt represents the kurtosis of the signal, x(n) represents the signal, μ represents the mean of the signal, σ represents the standard deviation of the signal, and E represents the expectation operator.

[0090] It is understandable that this embodiment preferably uses the time-domain statistical feature of kurtosis to evaluate the non-stationarity and frequency distribution of the signal, which helps to capture high-frequency transient pulses.

[0091] Specifically, in this embodiment, the frequency domain statistical features include the spectrum centroid, and based on the leakage noise sample data, the frequency domain statistical features of the signal are quantified and include:

[0092] S41, performing frame division and windowing processing on the leakage noise sample data.

[0093] Among them, framing means dividing the signal x[n] into short time frames of length M, and windowing means applying a window function to reduce spectrum leakage, that is, the sampled signal x after windowing is win [n]=x[n]·w[n].

[0094] S42, performing discrete Fourier transform (DFT) calculation on each frame after windowing to obtain the amplitude X(f) of the signal at each frequency component;

[0095] S43, the spectrum centroid is calculated based on the amplitude of the signal at each frequency component, which can be expressed as: , Among them, f k represents the physical frequency (Hz) corresponding to the kth frequency component, , fs represents the sampling rate, M represents the signal x win The length of [n] automatically weakens the influence of uniformly distributed noise through the denominator.

[0096] Furthermore, the calculated spectrum centroid may be normalized to facilitate subsequent unified threshold determination.

[0097] It is understandable that traditional leakage detection uses fixed-frequency band energy as a feature and cannot adapt to the acoustic characteristics of different pipe materials (steel / PVC). However, the actual leakage frequency will shift due to changes in pressure and pipe material, resulting in unstable characteristics. Conventional spectrum analysis tends to ignore the low-frequency sound waves generated by high-pressure pipeline leakage, and the high-frequency characteristics of micro-leaks are easily masked by noise, resulting in poor adaptability to working conditions. Therefore, this embodiment preferably uses the frequency domain statistical feature of the spectrum centroid, a physically associated pipeline vibration fundamental frequency, to dynamically reflect the main frequency position.

[0098] S5, based on the time domain statistical characteristics and frequency domain statistical characteristics of the signal, dynamically adjust the window function parameters and perform short-time Fourier transform processing to obtain a time-frequency image with optimized resolution.

[0099] Specifically, in this embodiment, based on the time domain statistical characteristics and frequency domain statistical characteristics of the signal, the window function parameters are dynamically adjusted and short-time Fourier transform processing is performed to obtain a time-frequency image with optimized resolution, including:

[0100] S51, compare the kurtosis of the signal with the preset kurtosis threshold in real time. If the kurtosis of the signal is greater than the preset kurtosis threshold, adjust the window type to a Hamming window; if the kurtosis of the signal is less than or equal to the preset kurtosis threshold, adjust the window type to a Hann window.

[0101] For example, the preset kurtosis threshold value obtained from historical noise data statistics is 5.

[0102] It is understandable that, considering that high kurtosis signals usually contain sharp burst components, this embodiment uses a Hamming window with lower side lobes to help suppress spectral leakage; low kurtosis signals are relatively stable, so this embodiment uses a Hann window to provide a slightly higher main lobe resolution.

[0103] S52, dividing the spectrum centroid into a plurality of first frequency intervals, and adjusting the window length according to the first frequency interval where the spectrum centroid of the signal is located, so that the spectrum centroid value is inversely proportional to the selected window length.

[0104] For example, the normalized spectrum centroid C∈[0,1] is discretely mapped to the preset window length L set {256,512,1024,2048}, which can be expressed as: ,

[0105] It can be understood that in this embodiment, when the signal energy is concentrated in the high frequency band (C is large), a short window (256 or 512) is used to improve the time resolution and accurately capture the rapidly changing weak transient features; when the energy is biased towards the low frequency (C is small), a long window (1024 or 2048) is used to reduce the frequency sampling interval and improve the frequency resolution.

[0106] It is understandable that the extreme scene features (such as low-frequency resonance of high-pressure pipelines) in the simulation data generated by the physical model in this embodiment can trigger the long window mode to ensure the complete representation of the time-frequency image.

[0107] S53, dividing the spectrum centroid into a plurality of second frequency intervals, and adjusting the overlap ratio according to the second frequency interval where the spectrum centroid of the signal is located, so that the spectrum centroid value is proportional to the selected overlap ratio.

[0108] For example, based on the normalized spectrum centroid C∈[0,1], the overlap rate O can be expressed as: ,

[0109] It can be understood that in this embodiment, when there are more high-frequency components (C is high), a larger overlap rate is used to ensure smooth slices between consecutive frames, that is, to ensure the time-frequency continuity of the high-frequency transient signal and avoid loss of information between frames; when the low frequency is dominant, the change is slow, and the overlap rate is reduced to reduce redundant calculations.

[0110] S54, based on the dynamically adjusted window type, window length, and overlap ratio, performs short-time Fourier transform processing on the signal to obtain a time-frequency image with optimized resolution. The spectrum value of the time-frequency image with optimized resolution at time index m and frequency f can be expressed as: , Among them, L represents the dynamically adjusted window length, O represents the dynamically adjusted overlap rate, Represents the dynamically adjusted window function, n m Represents the starting leakage noise sample of the mth frame.

[0111] It is understandable that the frequency distribution of environmental noise is relatively dispersed, and the main frequency band energy distribution of some signals is curved and low-frequency; while the main frequency band energy of leakage noise is relatively continuous and concentrated, and the frequency range is moderate. For this reason, this embodiment designs an adaptive filtering strategy for optimizing STFT parameters based on the time-frequency characteristics of leakage noise, targeting the characteristic difference between leakage noise and environmental noise. Specifically, according to the pulse characteristics of the signal waveform, the optimal solution for spectrum leakage control and resolution enhancement is adaptively selected (window function and kurtosis are linked), and according to the distribution characteristics of signal energy in the frequency domain, the balance point of time resolution and frequency resolution is adaptively adjusted (window length and spectrum centroid are linked), and the time-frequency grid density is quadratically optimized according to the signal frequency characteristics to ensure feature continuity while improving computational efficiency (overlap rate and spectrum centroid are quadratically linked), thereby better suppressing environmental noise in leakage noise, but retaining the inherent characteristics of environmental noise, and providing a good classification basis for the subsequent training of CNN classification models.

[0112] It is understandable that while STFT generation of time-frequency plots is a common signal processing operation, the innovative design of this embodiment is to dynamically optimize STFT parameters based on the time-frequency characteristics of leakage noise to balance low-frequency resolution and high-frequency response. For example, weak transient signal capture (short-window STFT) and low-frequency resonance separation (long-window STFT) are customized for the acoustic characteristics of the pipe network. At the same time, the prior art generally believes that dynamically adjusting parameters increases computational complexity, and therefore typically uses fixed STFT parameters. However, this embodiment, through lightweight calculation of the statistical characteristics of the signal (i.e., kurtosis and spectral centroid), can achieve real-time adaptive adjustment of window function parameters (window type, window length, overlap ratio), thereby generating a time-frequency image with optimized resolution and implementing parameterized adaptive STFT time-frequency analysis. This avoids information loss from manual features, fully captures leakage characteristics, and effectively improves time-frequency feature clarity, high-frequency transient location accuracy, and computational efficiency in the low-frequency band. This ensures that the time-frequency representation always matches the physical characteristics of the current signal, helping to improve the micro-leak detection rate in water supply pipe networks.

[0113] S6, construct a CNN classification model and train the CNN classification model based on the resolution-optimized time-frequency image.

[0114] In this embodiment, the specific architecture of the CNN classification model is not limited, as long as it can achieve classification and discrimination of time-frequency images.

[0115] Among them, the implementation method of training the CNN classification model based on resolution-optimized time-frequency images is as follows: according to the types of leakage noise and environmental noise in the leakage noise sample data, the resolution-optimized time-frequency images are labeled to create a data set for training. The data set is then divided into a training set and a validation set, which are input into the CNN classification model for training. The trained CNN classification model is saved, and the model accuracy is evaluated based on accuracy, F1 score, etc.

[0116] It can be understood that this embodiment enables the CNN classification model to effectively learn extreme scenario patterns and improve model generalization through physical modeling data enhancement. On the basis of physical modeling enhanced data, the high-resolution time-frequency map generated by dynamically optimizing the STFT parameters according to the time-frequency characteristics of the leakage noise provides high-quality input for the CNN classification model, avoiding the information loss of manual features. The CNN deep learning model learns high-resolution time-frequency image features, provides binary classification discrimination of leakage noise and environmental noise, eliminates additional hardware dependence (no auxiliary positioning such as water pressure sensors is required), and collaboratively realizes efficient pure acoustic end-to-end pipe network leakage detection, thereby achieving a qualitative improvement in acoustic detection accuracy.

[0117] Preferably, in this embodiment, the CNN classification model adopts the ResNet network architecture.

[0118] It is understandable that in ResNet:

[0119] 1. Shallow convolution: extract edge / texture features (e.g. horizontal stripes in a time-frequency image correspond to periodic interference);

[0120] 2. Deep convolution: Capture abstract pattern structures (such as the combined features of low-frequency continuous spectrum + high-frequency sparse pulses).

[0121] It can be understood that this embodiment further uses the residual skip connection of ResNet to solve the problem of vanishing gradients in deep networks, retains the multi-scale features of the adaptive STFT output, and avoids the feature attenuation of weak signals (such as low-frequency leakage) in deep networks. That is, the adaptive STFT and ResNet work together to solve the time-frequency resolution contradiction and feature loss problems.

[0122] S7, converts the measured leakage noise data into a time-frequency graph, inputs it into the trained CNN classification model, and outputs the predicted probability of leakage noise and environmental noise as the leakage noise detection result of the water supply network.

[0123] Furthermore, the leakage result can be determined according to the output prediction probability of the leakage noise and a preset decision rule. For example, if the prediction probability of the leakage noise is greater than 0.85, leakage is determined.

[0124] In summary, the water supply network leakage noise detection method provided in this embodiment achieves pure acoustic end-to-end classification prediction of measured leakage noise data by deeply integrating physical modeling, parameterized data generation, feature quantization, dynamic short-time Fourier transform adjustment and CNN training. Specifically, this application ensures that the simulation data in extreme scenarios conforms to the real fluid mechanics and acoustic laws through physical modeling of leakage noise, and combines it with the measured data to form high-quality leakage noise sample data, so as to more accurately quantify the time domain statistical characteristics and frequency domain statistical characteristics of the sample data, and then drive the dynamic adjustment of the window function parameters of the short-time Fourier transform to generate higher resolution time-frequency images, providing highly robust input for the CNN classification model. The parameters of each link are automatically derived from physical properties, thus globally and collaboratively breaking through the vicious cycle of "data scarcity - feature fuzziness - poor model generalization". At the same time, it abandons the inertial thinking of water pressure joint positioning, breaks the cognitive bias that "leakage positioning must rely on multimodal data fusion", and effectively solves the problems of insufficient extreme scene samples, large environmental noise interference, imbalance between computational complexity and feature information loss, and high cost caused by high dependence on additional hardware (such as water pressure sensors) in existing technologies. It significantly improves the detection accuracy of leakage noise in water supply networks, eliminates and reduces hardware dependence, and realizes efficient pure acoustic end-to-end classification.

[0125] Example 2:

[0126] This embodiment only compares Figure 1 The difference parts of the corresponding embodiment are described here, and the technical concepts of the rest of the design are similar to those of the embodiment 1, which will not be repeated here in this embodiment.

[0127] To prevent the calculation of kurtosis and spectral centroid from being interfered with by environmental noise (such as vehicle vibration and water pump harmonics), and to enable them to stably guide the subsequent optimization of window function parameters, a targeted noise suppression process is designed. Specifically, in this embodiment, after S3 "using the set of measured leakage noise data and simulated leakage noise data as leakage noise sample data" and before S4, the following steps are also included:

[0128] A1, delays the leakage noise sample data x(n) to obtain a delayed signal.

[0129] Specifically, the delayed signal vector X(nJ)=[x(nJ),x(nJ-1),…,x(nJ-N+1)] T , where N represents the adaptive filter order and J represents the delay length, which can be determined based on the signal sampling rate fs and the noise correlation time, such as J=fs / 2f min , f min is the lowest frequency of the leakage signal.

[0130] A2: Input the delayed signal into the adaptive filter for filtering, and use the LMS algorithm to iteratively update the filter weights.

[0131] Specifically, it can be expressed as: , , , Among them, y(n) represents the output signal of the adaptive filter, W(n) represents the weight vector of the adaptive filter, e(n) represents the error, and the current signal x(n) is used as the expected output. The filter weights are adjusted through adaptive error feedback to overcome the technical problem of non-stationary noise interference. μ represents the update step factor of the LMS algorithm, and W(n+1) represents the update of the weights along the negative gradient direction to achieve LMS lightweight iteration.

[0132] A3, performing coherent accumulation on the signals output by the filter to obtain the coherently accumulated output signal as the leakage noise sample data after preprocessing.

[0133] Specifically, it can be expressed as: , Where m represents the number of coherent accumulations, y(n) represents the signal output by the adaptive filter, i represents the total number of signals selected for coherent accumulation, and Δ i Indicates the small delay increment of the i-th signal.

[0134] It can be understood that in this embodiment, coherent accumulation is performed on the signals output by the filter, that is, phase alignment and superposition of multiple signals are performed to enhance the coherent components of the leakage signal, thereby solving the technical problem of low energy of weak signals.

[0135] Please see the attached Figure 2 , Figure 2 FIG. 1 shows a schematic diagram of a comparison before and after preprocessing provided by an embodiment of this specification. Figure 2 As shown, Figure 2 2(a) is the time-frequency image before filtering, Figure 2 2 (b) is the time-frequency image after filtering using the preprocessing solution provided in this embodiment. It can be seen that Figure 2 2 (a) has strong environmental noise interference, and the frequency band of the leakage noise cannot be clearly identified; Figure 2 2(b) clearly shows that the leakage noise is mainly concentrated in the frequency band near the center frequency of 900 Hz. Therefore, it shows that after the adaptive filtering of the preprocessing scheme provided in this embodiment, the environmental noise interference can be effectively suppressed and the leakage noise can be accurately extracted, thereby providing a good signal data foundation for subsequent time-frequency analysis and deep learning model training, which helps to improve detection accuracy.

[0136] It is understandable that traditional bandpass filters cannot separate environmental noise interference that overlaps with the leakage band, and the environmental noise has similar statistical characteristics to the leakage signal. Conventional filtering will cause signal distortion and there is a technical problem of insufficient non-stationary noise suppression. Moreover, the micro-leakage signal-to-noise ratio is usually lower than -10dB. Existing technical methods are difficult to effectively extract weak signals. For example, the spectrum heat map signal-to-noise ratio is low, which will mask effective features. To this end, this embodiment utilizes the broadband correlation of the leakage signal (the environmental noise is irrelevant), destroys the correlation of the interference signal by delaying J, and then uses the current input x(n) as the expected output to force the filter to track signal changes. It can adapt to non-stationary environments without presetting a reference signal, and then utilizes the periodic repeatability of the leakage signal (such as pipeline resonance) to fine-tune the phase Δ i It achieves coherence enhancement, separates co-frequency noise, effectively improves the signal-to-noise ratio, provides a high signal-to-noise ratio input signal for subsequent time-frequency analysis, reduces the spectrum centroid / kurtosis calculation error, and thus optimizes the STFT parameter selection to avoid parameter misselection caused by environmental noise interference.

[0137] Example 3:

[0138] Please see the attached Figure 3 , Figure 3 This is a schematic diagram of the structure of a water supply network leakage noise detection system provided in one embodiment of this specification.

[0139] like Figure 3 As shown, the water supply network leakage noise detection system may at least include:

[0140] Physical model construction module 1 is used to construct a physical model of leakage noise based on fluid mechanics and sound propagation laws. The physical model is used to simulate noise signals under different physical conditions.

[0141] A simulation data generation module 2 is used to parameterize and generate simulation leakage noise data based on a physical model of leakage noise to simulate leakage noise data under extreme physical conditions;

[0142] The sample data integration module 3 is used to obtain the measured leakage noise data and use the set of the measured leakage noise data and the simulated leakage noise data as leakage noise sample data;

[0143] The statistical feature quantization module 4 is used to quantify the time domain statistical features and frequency domain statistical features of the signal based on the leakage noise sample data;

[0144] The time-frequency image optimization module 5 is used to dynamically adjust the window function parameters and perform short-time Fourier transform processing based on the time-domain statistical characteristics and frequency-domain statistical characteristics of the signal to obtain a time-frequency image with optimized resolution;

[0145] A classification model building module 6 is used to build a CNN classification model and train the CNN classification model based on the resolution-optimized time-frequency image;

[0146] The leakage noise detection module 7 is used to convert the measured leakage noise data into a time-frequency diagram, input the trained CNN classification model, and output the predicted probability of leakage noise and environmental noise.

[0147] It can be understood that the technical concept of the water supply network leakage noise detection system provided in this embodiment is similar to the technical concept of the aforementioned water supply network leakage noise detection method, and this embodiment will not be repeated here.

[0148] Example 4:

[0149] Another embodiment of the present disclosure provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform one or more steps of the aforementioned embodiments. If the components of the aforementioned electronic device are implemented as software functional units and used as independent downstream task predictions or tasks, they can be stored in the computer-readable storage medium.

[0150] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. A computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of this specification are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. Available media may be magnetic media (eg, floppy disks, hard disks, tapes), optical media (eg, digital versatile discs (DVDs)), or semiconductor media (eg, solid state drives (SSDs)).

[0151] The above description is merely a description of the preferred embodiments disclosed in this application and the technical principles employed. Those skilled in the art should understand that the scope of protection provided by this disclosure is not limited to technical solutions formed by a specific combination of the aforementioned technical features, but also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents without departing from the scope of the disclosure. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0152] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

Claims

1. A method for detecting leakage noise in a water supply network, characterized in that: The following steps are involved: A physical model of leakage noise is constructed based on fluid mechanics and sound propagation laws, and the physical model is used to simulate noise signals under different physical conditions; Based on the physical model of leakage noise, parameterized simulation leakage noise data is generated to simulate leakage noise data under extreme physical conditions; Acquire measured leakage noise data, and use the set of the measured leakage noise data and the simulated leakage noise data as leakage noise sample data; Based on the leakage noise sample data, the time domain statistical characteristics and frequency domain statistical characteristics of the signal are quantified; Based on the time-domain and frequency-domain statistical characteristics of the signal, the window function parameters are dynamically adjusted and short-time Fourier transform processing is performed to obtain a time-frequency image with optimized resolution; Build a CNN classification model and train it based on the time-frequency images with optimized resolution; The measured leakage noise data is converted into a time-frequency graph and input into the trained CNN classification model, which outputs the predicted probability of leakage noise and ambient noise.

2. A water supply network leakage noise detection method according to claim 1, characterized in that: A physical model of leakage noise is constructed based on fluid mechanics and sound propagation laws, including: Obtain the pipe material parameters, leakage diameter and fluid pressure in the water supply network; Based on the pipe material parameters, leakage aperture, and fluid pressure in the water supply network, the baseline amplitude of leakage noise is determined and the fluid turbulence simulation value is generated; Based on the pipe material parameters of the water supply network, the impulse response of the bandpass filter is determined to simulate the frequency response characteristics of different pipe materials; Based on the baseline amplitude of leakage noise, fluid turbulence simulation value and bandpass filter impulse response, a physical model of leakage noise is constructed.

3. A water supply network leakage noise detection method according to claim 2, characterized in that: Based on the physical model of leakage noise, parameterized simulation leakage noise data is generated, including: Adjust the leakage aperture and the fluid pressure parameters in the pipeline, and generate enhanced parameters under extreme physical conditions based on parameterized generation rules; Substitute the enhanced parameters into the physical model of leakage noise to obtain simulated leakage noise data.

4. A method for detecting leakage noise in a water supply network according to claim 1, characterized in that: The time domain statistical characteristics include kurtosis, and the time domain statistical characteristics of the signal are quantified based on the leakage noise sample data, including: ， Where Kurt represents the kurtosis of the signal, x(n) represents the signal, μ represents the mean of the signal, σ represents the standard deviation of the signal, and E represents the expectation operator; The frequency domain statistical features include the spectrum centroid, and based on the leakage noise sample data, the frequency domain statistical features of the signal are quantified, including: Perform frame division and windowing processing on leakage noise sample data; Perform discrete Fourier transform calculation on each frame after windowing to obtain the amplitude of the signal at each frequency component; The spectral centroid is calculated based on the amplitude of the signal at each frequency component.

5. A water supply network leakage noise detection method according to claim 4, characterized in that: Based on the time-domain and frequency-domain statistical characteristics of the signal, the window function parameters are dynamically adjusted and short-time Fourier transform processing is performed to obtain a time-frequency image with optimized resolution, including: The kurtosis of the signal is compared with the preset kurtosis threshold in real time. If the kurtosis of the signal is greater than the preset kurtosis threshold, the window type is adjusted to the Hamming window. If the kurtosis of the signal is less than or equal to the preset kurtosis threshold, the window type is adjusted to the Hann window. Dividing the spectrum centroid into a plurality of first frequency intervals, and adjusting the window length according to the first frequency interval in which the spectrum centroid of the signal is located, so that the spectrum centroid value is inversely proportional to the selected window length; Dividing the spectrum centroid into a plurality of second frequency intervals, and adjusting the overlap ratio according to the second frequency interval in which the spectrum centroid of the signal is located, so that the spectrum centroid value is proportional to the selected overlap ratio; Based on the dynamically adjusted window type, window length and overlap ratio, the signal is processed by short-time Fourier transform to obtain a time-frequency image with optimized resolution.

6. A method for detecting leakage noise in a water supply network according to claim 1, characterized in that: Also includes: Performing a delay operation on the leakage noise sample data to obtain a delayed signal; The delayed signal is input into the adaptive filter for filtering, and the filter weights are iteratively updated using the LMS algorithm; The signals output by the filter are coherently accumulated to obtain the coherently accumulated output signal as the leakage noise sample data after preprocessing.

7. A method for detecting leakage noise in a water supply network according to claim 6, characterized in that: The signals output by the filter are coherently accumulated to obtain the output signal after coherent accumulation, which is expressed as: ， Where m represents the number of accumulations, y(n) represents the signal output by the adaptive filter, i represents the total number of signals selected for coherent accumulation, and Δ i Indicates the small delay increment of the i-th signal.

8. A method for detecting leakage noise in a water supply network according to claim 1, characterized in that: The CNN classification model adopts the ResNet network architecture.

9. A water supply network leakage noise detection system, characterized in that: include: A physical model building module is used to build a physical model of leakage noise based on fluid mechanics and sound propagation laws. The physical model is used to simulate noise signals under different physical conditions. A simulation data generation module is used to parameterize and generate simulated leakage noise data based on the physical model of leakage noise to simulate leakage noise data under extreme physical conditions; A sample data integration module is used to obtain measured leakage noise data and use the set of measured leakage noise data and simulated leakage noise data as leakage noise sample data; The statistical feature quantization module is used to quantify the time domain statistical features and frequency domain statistical features of the signal based on the leakage noise sample data; The time-frequency image optimization module is used to dynamically adjust the window function parameters and perform short-time Fourier transform processing based on the time-domain and frequency-domain statistical characteristics of the signal to obtain a time-frequency image with optimized resolution; The classification model building module is used to build a CNN classification model and train the CNN classification model based on the resolution-optimized time-frequency image; The leakage noise detection module is used to convert the measured leakage noise data into a time-frequency graph, input it into the trained CNN classification model, and output the predicted probability of leakage noise and environmental noise.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting leakage noise in a water supply network according to any one of claims 1 to 8 is implemented.

Images