Water supply pipeline leakage detection method and system

Abstract

The invention discloses a water supply pipeline leakage detection method and system, and relates to the field of artificial intelligence, and the method comprises the steps: collecting a multi-source signal through a plurality of sensor nodes disposed on a pipeline; preprocessing and space-time alignment are carried out on the collected multi-source signal at an edge computing node to obtain a standardized signal; analyzing the standardized signal by using a preset lightweight one-dimensional convolutional neural network model at the edge computing node to obtain a preliminary leakage detection result and a corresponding confidence coefficient; executing a hierarchical response strategy according to the confidence coefficient; and in the cloud analysis module, multi-model fusion analysis is performed in combination with the feature vectors, historical time sequence data and a hydraulic model simulation result, so that leakage points are accurately positioned and verified, and a leakage result is determined. According to the invention, a real-time, accurate and reliable complete solution can be provided for pipeline leakage monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a water supply pipeline leakage detection method and system. BACKGROUND

[0002] The water supply network is one of the core infrastructures that maintain the normal operation of a city, and its operation state is directly related to the water resource utilization efficiency and the stability of public services. However, for a long time, the problem of pipeline leakage has been widespread and has not been effectively controlled, resulting in a large amount of water resources being wasted in the transportation process, which not only exacerbates the pressure of water resource shortage, but also brings continuous economic losses to water supply enterprises.

[0003] At present, in the practice of pipeline leakage detection, two types of methods, namely manual inspection and acoustic wave detection, are mainly relied on. As a traditional method, manual inspection has limited efficiency, and the pipeline mileage that can be covered by the inspection personnel in one day is short, so it is difficult to achieve comprehensive and frequent inspection of a large pipeline network system. At the same time, this method highly depends on personal experience, and is prone to omissions in complex environments or weak leakage conditions, and has the problems of high labor intensity and high labor cost, which cannot meet the requirements of modern cities for rapid response to leakage.

[0004] On the other hand, the detection method based on the principle of acoustic wave improves the technical level to a certain extent, and judges by capturing the vibration signal generated when the pipeline leaks. However, this method still has obvious limitations in practical application. For example, it has a high false positive rate, which makes the operation and maintenance personnel need to spend a lot of effort to review the results, reducing the disposal efficiency; in terms of positioning accuracy, the existing technology is difficult to achieve the ideal level, and there is often a large deviation, which brings difficulties to subsequent repair construction. In addition, acoustic wave detection is easily disturbed by environmental noise, and often needs to be operated at night and other quiet periods, which not only increases the complexity of work arrangement, but also causes a long response delay between the occurrence of leakage and the final confirmation. Therefore, although this method has made some progress compared with manual methods, it still cannot break through the technical bottleneck in real-time, accuracy and automation. SUMMARY

[0005] The embodiments of the present application provide a water supply pipeline leakage detection method and system, and aim to solve the technical problem of how to provide a pipeline leakage monitoring solution that can meet the requirements of real-time, accuracy and reliability at the same time.

[0006] In a first aspect, the embodiments of the present application provide a water supply pipeline leakage detection method, which comprises: collecting multi-source signals by a plurality of sensor nodes deployed on the pipeline, wherein the multi-source signals include pipeline vibration signals, pressure signals and temperature signals; At the edge computing node, the collected multi-source signals are preprocessed and spatio-temporally aligned to obtain standardized signals; At the edge computing node, a preset lightweight one-dimensional convolutional neural network model is used to analyze the standardized signals to obtain a preliminary leakage detection result and a corresponding confidence level; According to the confidence level, a hierarchical response strategy is executed; when the confidence level is higher than a first preset threshold, a local control action is triggered at the edge computing node; when the confidence level is between the first preset threshold and a second preset threshold, a feature vector extracted based on the standardized signals is uploaded to a cloud analysis module; At the cloud analysis module, multi-model fusion analysis is performed in combination with the feature vector, historical time series data and hydraulic model simulation results to accurately locate and verify the leakage point and determine the leakage result.

[0007] A further technical solution is that the preprocessing includes signal denoising and temperature compensation; the spatio-temporal alignment includes achieving multi-node time synchronization using a preset precision time protocol, and unifying the time axes of heterogeneous data using an interpolation algorithm.

[0008] A further technical solution is that before the edge computing node uses a preset lightweight one-dimensional convolutional neural network model to analyze the standardized signals to obtain a preliminary leakage detection result and a corresponding confidence level, the method further includes: A preset low-frequency Wigner-Ville distribution algorithm is used to perform time-frequency analysis on the pipeline vibration signals to obtain time-frequency features.

[0009] A further technical solution is that the use of a preset low-frequency Wigner-Ville distribution algorithm to perform time-frequency analysis on the pipeline vibration signals to obtain time-frequency features includes: An adaptive window function mechanism is used to dynamically select a window function type and window length according to a center frequency of a frequency band of the pipeline vibration signals; Based on Bark band reconstruction, the linear frequency spectrum of the pipeline vibration signals is converted into a plurality of critical frequency bands conforming to the human ear hearing characteristics to generate a reconstructed time-frequency distribution; According to the material of the pipeline, corresponding attenuation coefficients and frequency correction weights are obtained from a preset sound wave attenuation parameter library to perform compensation calculation on the energy of the time-frequency distribution to obtain the time-frequency features.

[0010] A further technical solution is that the hierarchical response strategy includes: If the confidence level is greater than 85%, a valve closing or pressure regulating action is directly triggered at the edge computing node; If the confidence is between 40% and 85%, the feature vector after dimensionality reduction is uploaded to the cloud analysis module; If the confidence is less than 40%, a signal is triggered to trigger a reacquisition process.

[0011] Further technical solutions thereof are that the multi-model fusion analysis comprises: A long short-term memory network model is used to perform time series verification on the received feature vector based on historical time series data, and output a leakage loss probability; A hydraulic simulation model is run to dynamically calibrate pipe network parameters, generate a pressure anomaly gradient field, and determine a hydraulic anomaly area; Based on acoustic feature coordinates and the hydraulic anomaly area, a spatial matching optimization model is constructed, and the optimal coordinates of the leakage point are iteratively solved by the least square method, wherein the acoustic feature coordinates are determined based on the time-frequency features.

[0012] Further technical solutions thereof are that the method further comprises: When the network between the edge computing node and the cloud analysis module is interrupted, an offline analysis mode is enabled on the edge computing node, and the collected data is cached by using a local ring buffer; After the network is restored, the cached data is automatically uploaded to the cloud.

[0013] Further technical solutions thereof are that the method further comprises: Through a federated learning framework, model gradients trained by each edge computing node are aggregated, and global model updating is performed in the cloud; The updated incremental model is distributed to each edge computing node.

[0014] Further technical solutions thereof are that the method further comprises: According to the leakage result, a corresponding response action is performed, and relevant data of the leakage result is fed back to a digital twin model of the pipe network to optimize a preventive maintenance strategy of the pipe network.

[0015] In a second aspect, an embodiment of the present application proposes a water supply pipeline leakage detection system for executing the method as described in the first aspect, comprising an edge perception module, an edge computing node, and a cloud analysis module, wherein the edge perception module comprises a plurality of sensor nodes deployed on the pipeline.

[0016] The embodiment of the present application provides a water supply pipeline leakage detection method and system. Wherein, the method comprises: collecting multi-source signals through a plurality of sensor nodes arranged on the pipeline, the multi-source signals comprising pipeline vibration signals, pressure signals and temperature signals; preprocessing and space-time aligning the collected multi-source signals to obtain standardized signals at an edge computing node; analyzing the standardized signals by using a preset lightweight one-dimensional convolutional neural network model to obtain a preliminary leakage detection result and a corresponding confidence level at the edge computing node; executing a hierarchical response strategy according to the confidence level; wherein, when the confidence level is higher than a first preset threshold, triggering a local control action at the edge computing node; when the confidence level is between the first preset threshold and a second preset threshold, uploading a feature vector extracted based on the standardized signals to a cloud analysis module; combining the feature vector, historical time series data and hydraulic model simulation results to perform multi-model fusion analysis at the cloud analysis module, so as to accurately locate and verify the leakage point and determine the leakage result. The present application realizes rapid preliminary judgment and emergency control of leakage through multi-source signal collection and edge-cloud collaborative architecture, combined with lightweight neural network edge real-time analysis and confidence level hierarchical response mechanism; further through cloud multi-model fusion and space-time matching verification, the detection accuracy and positioning reliability are significantly improved. The whole method guarantees real-time performance while ensuring the accuracy of the result through multi-level verification, providing a real-time, accurate and reliable complete solution for pipeline leakage monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A flowchart of a water supply pipeline leakage detection method provided by the embodiment of the present application; Figure 2 A structural block diagram of a water supply pipeline leakage detection system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] It should be understood that the terms "comprises" and "comprising" when used in this specification and accompanying claims, signify the presence of the stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0021] It should also be understood that the terms used in the specification and the appended claims are intended to be interpreted broadly and in a manner consistent with the principles of the prior art, and that references to a particular embodiment or a particular use of the application are intended to refer only to that specific embodiment or use and not to be limiting of the application in general.

[0022] It should further be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items, and that the term "at least one of' as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0023] As used in this specification and claims, the terms "if' and "when" can each be interpreted to mean "when" or "if," depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can each be interpreted to mean "if it is determined that," or "if [the described condition or event] is detected that," or "when it is determined that" or "when [the described condition or event] is detected," depending on the context.

[0024] Referring now to Figure 2 The embodiment of the present application provides a water supply pipeline leakage detection system for executing a water supply pipeline leakage detection method provided by the following embodiments of the present application, which comprises an edge perception module 30, an edge computing node 20 and a cloud analysis module 10, wherein the edge perception module 30 comprises a plurality of sensor nodes deployed on the pipeline.

[0025] Further referring to Figure 1 The embodiment of the present application provides a water supply pipeline leakage detection method, which comprises the following steps: S1, collecting multi-source signals through a plurality of sensor nodes deployed on the pipeline, wherein the multi-source signals comprise pipeline vibration signals, pressure signals and temperature signals.

[0026] In specific implementation, multiple sensor nodes deployed on the pipeline synchronously collect multi-source signals such as vibration, pressure and temperature. This multi-dimensional sensing mechanism provides a rich data basis for subsequent analysis. The vibration signal can directly capture the acoustic characteristics generated during pipeline leakage, the pressure signal reflects the abnormal changes in the hydraulic state of the pipeline network, and the temperature signal is used to compensate for the influence of environmental factors on detection accuracy. The complementarity of the three signals significantly enhances the adaptability of the system to complex leakage scenarios.

[0027] S2, at the edge computing node, pre-processing and space-time alignment of the collected multi-source signals are performed to obtain standardized signals.

[0028] In specific implementation, at the edge computing node, the multi-source signals are pre-processed and space-time aligned based on the OpenHarmony system to fully utilize the low-latency characteristics of edge computing. The pre-processing operation and space-time alignment can effectively improve data quality. This standardized processing completed at the data source not only reduces the amount of transmitted data, but more importantly, provides high-quality input for real-time analysis.

[0029] The OpenHarmony system supports on-demand elastic deployment through component-based design and can adapt to various hardware platforms from resource-constrained embedded sensors to fully functional intelligent gateways. The system provides a unified device communication framework and security base to ensure the reliability and confidentiality of data transmission, providing an efficient and stable underlying operating environment for Internet of Things applications.

[0030] In some preferred embodiments, the pre-processing includes signal denoising and temperature compensation; and the space-time alignment includes achieving multi-node time synchronization using a pre-set precision time protocol and unifying the time axis of heterogeneous data using an interpolation algorithm.

[0031] In specific implementation, the signal denoising in pre-processing addresses the problem that pipeline vibration signals are easily disturbed by various noises in the actual environment. A pre-set filtering algorithm is used to remove background noise, effectively improving the signal-to-noise ratio, which makes the signal features in subsequent analysis clearer and more identifiable, reducing the possibility of false positives caused by noise.

[0032] Further, temperature compensation considers the influence of environmental temperature changes on the properties of pipeline materials and the performance of sensors. By establishing a correlation model between temperature and signal characteristics, the collected vibration and pressure signals are dynamically corrected, eliminating the influence of temperature fluctuations on detection accuracy and ensuring stable detection performance under different climate conditions.

[0033] Further, the key innovation in the spatio-temporal alignment is to achieve multi-node time synchronization by using the precise time protocol. Since the sound wave signals generated by the leakage propagate at a very high speed in the pipeline, the time synchronization accuracy of different sensor nodes is directly related to the accuracy of subsequent positioning calculation. This method achieves microsecond-level time synchronization, so that the data from different location sensors have a unified time reference, thereby laying a technical foundation for accurate positioning based on signal arrival time difference.

[0034] Further, the use of interpolation algorithms to unify the time axis of heterogeneous data solves the fusion problem caused by the inconsistency of sampling frequencies of different types of sensors. By unifying the data of different sampling rates such as vibration, pressure, and temperature to the same time scale, multi-source signals can be effectively correlated and analyzed, thereby more accurately capturing the change rules of various physical parameters when leakage occurs. These preprocessing and spatio-temporal alignment measures work together to build a high-quality data foundation, enabling subsequent intelligent analysis algorithms to be performed on clean and structured data, significantly improving the detection accuracy and positioning precision of the entire system.

[0035] S3, at the edge computing node, using a preset lightweight one-dimensional convolutional neural network model to analyze the standardized signal to obtain a preliminary leakage detection result and a corresponding confidence level.

[0036] In specific implementations, the use of a lightweight one-dimensional convolutional neural network model (lightweight 1D-CNN model) for real-time analysis and output of confidence levels at the edge computing node is the core breakthrough of this method for achieving intelligence. This lightweight one-dimensional convolutional neural network model is specially optimized and adapted to edge computing resources, enabling fast extraction of leakage features in standardized signals and generation of quantitative evaluation results. In some preferred embodiments, before the above step "at the edge computing node, using a preset lightweight one-dimensional convolutional neural network model to analyze the standardized signal to obtain a preliminary leakage detection result and a corresponding confidence level", the method further includes using a preset low-frequency Wigner-Ville distribution algorithm to perform time-frequency analysis on the pipeline vibration signal to obtain time-frequency features.

[0037] In specific implementation, the step of using low-frequency Wigner-Ville distribution algorithm for time-frequency analysis provides a special signal processing scheme for the particularity of the water supply pipeline leakage sound wave signal. The traditional leakage detection method often has poor effect when processing the unique low-frequency vibration signal of the water supply pipeline, because the signal energy is weak and easy to be overwhelmed by environmental noise, while the low-frequency Wigner-Ville distribution algorithm is particularly suitable for analyzing non-stationary and low-frequency vibration signals, and can provide high-resolution signal feature representation in time domain and frequency domain at the same time. The method converts one-dimensional vibration signal into two-dimensional time-frequency distribution through time-frequency analysis, which enables the time-varying characteristics and frequency distribution characteristics of the signal to be observed at the same time, which is of great significance for identifying the time-frequency mode unique to leakage. For example, micro-leakage of the pipeline often shows sustained low-frequency vibration, while burst pipe may show wideband impact signal, and these characteristics will show different patterns in the time-frequency distribution graph, providing more rich feature input for subsequent neural network identification.

[0038] Further, the time-frequency features are input into a lightweight one-dimensional convolutional neural network model for analysis, which actually constructs a two-level feature extraction architecture. The time-frequency analysis completes the preliminary feature enhancement and conversion, and the convolutional neural network further learns the deep features and identifies the patterns on this basis. This combination fully takes advantage of the respective advantages of traditional signal processing methods and modern deep learning.

[0039] Specifically, the above step of "performing time-frequency analysis on the pipeline vibration signal using a preset low-frequency Wigner-Ville distribution algorithm to obtain time-frequency features" specifically includes the following steps: adopting an adaptive window function mechanism to dynamically select a window function type and a window length according to a center frequency of a frequency band of the pipeline vibration signal; based on Bark frequency band reconstruction, converting a linear frequency spectrum of the pipeline vibration signal into a plurality of critical frequency bands conforming to human ear hearing characteristics to generate a reconstructed time-frequency distribution; and according to a material of the pipeline, obtaining corresponding attenuation coefficients and frequency correction weights from a preset sound wave attenuation parameter library to perform compensation calculation on energy of the time-frequency distribution to obtain the time-frequency features.

[0040] In specific implementation, the adaptive window function mechanism dynamically selects a window function type and a window length according to a center frequency of a frequency band of the pipeline vibration signal, which solves the problem that a fixed window function in traditional time-frequency analysis cannot adapt to different signal characteristics. In the actual application of pipeline leakage detection, the vibration signals generated by different pipe sections and different leakage types have great differences in frequency characteristics. For example, the leakage signal of a large-diameter pipeline is mainly concentrated in the low-frequency band, while a small-diameter pipeline may contain more high-frequency components. The adaptive window function can intelligently adjust the analysis parameters according to the actual situation of the current analysis signal, ensuring that the optimal time-frequency resolution can be obtained in various situations.

[0041] In a specific implementation, the adaptive window length is based on the following formula: where L represents the calculated window function length. The window length directly affects the resolution of time-frequency analysis. A longer window has higher frequency resolution, and a shorter window has higher time resolution.

[0042] T S represents the total length of the input vibration signal (in sample points).

[0043] f c represents the center frequency of the current frequency band being analyzed.

[0044] Formula logic: when analyzing low frequency bands (center frequency below 100 Hz), use 20% of the signal length as the window length to obtain higher frequency resolution to capture low frequency leakage characteristics; when analyzing higher frequency bands, use 10% of the signal length as the window length to balance time resolution.

[0045] Further, the linear frequency spectrum is converted into critical bands that conform to the human ear's hearing characteristics based on Bark band reconstruction. This innovation draws on the intelligent processing mechanism of the human auditory system, as the human ear's perception of sound is inherently nonlinear, with higher frequency resolution in the low frequency region. Pipeline leakage sound waves have similarities in frequency domain distribution with speech signals, mainly concentrated in the low frequency region and different leakage modes have discriminative characteristics in specific frequency bands. Bark band reconstruction enhances the ability to analyze key low frequency bands, making leakage characteristics more prominent and improving the detection sensitivity of micro-leakage.

[0046] In a specific implementation, Bark band conversion is based on the following formula: where this formula converts the linear frequency scale to the Bark scale, which better conforms to the human ear's hearing characteristics.

[0047] z(f): represents the Bark value corresponding to the linear frequency f. This is a dimensionless critical band scale.

[0048] f: represents the linear frequency, with units of Hertz (Hz).

[0049] arctan: is the inverse tangent function in mathematics.

[0050] The constants (13, 0.00076, 3.5, 7500) in the formula are empirical parameters used to make the converted Bark scale well match the critical bands of human hearing.

[0051] Further, Bark band reconstruction is based on the following formula: where this equation describes how to reconstruct the time-frequency distribution of traditional WVD into the time-frequency distribution based on Bark frequency band.

[0052] LWVD(t,z) represents the energy value at time t and Bark frequency band z, i.e. the reconstructed low-frequency Wigner-Ville distribution.

[0053] t represents the time point.

[0054] z represents the Bark frequency band.

[0055] K represents the total number of linear frequency division segments.

[0056] WVD(t,fk) represents the value of traditional WVD time-frequency distribution at time t and linear frequency fk.

[0057] δ(z(fk)−z) is a discrete approximation of Dirac delta function, which acts as a selector function here.

[0058] Further, the sound wave attenuation compensation according to the pipe material considers the difference in the propagation characteristics of sound waves in different material pipes, which greatly improves the practicality and accuracy of the algorithm. In actual pipe networks, different materials such as cast iron pipes, PVC pipes, and PE pipes have completely different attenuation characteristics of sound waves. For example, sound waves travel a long distance in cast iron pipes but attenuate quickly in PE pipes. If the material difference is not considered directly, it will cause serious deviation in the positioning calculation based on signal strength. Through the pre-set sound wave attenuation parameter library, the time-frequency distribution energy is accurately compensated, eliminating the influence of material factors on the analysis results, so that the algorithm can maintain stable detection performance in pipe systems of different materials. This feature has a particularly important value for actual pipe network environments with complex and varied materials.

[0059] In specific implementation, the pipe material attenuation compensation is based on the following formula: where E corrected (t,f) represents the time-frequency energy value after pipe material attenuation compensation at time t and frequency f.

[0060] LWVD(t,f) represents the original time-frequency energy value before compensation at time t and frequency f.

[0061] α represents the attenuation coefficient of the pipe material. The value is taken from the pipe material parameter library, such as 0.85 for cast iron.

[0062] β represents the frequency correction weight. The value is also taken from the pipe material parameter library, which determines the degree of attenuation change with frequency.

[0063] Specifically, the attenuation coefficient and the frequency correction weight are valued as shown in Table 1 below: Table 1 Attenuation coefficient and frequency correction weight value table Further, in some preferred embodiments, after the pipe attenuation compensation, the output time-frequency signal is re-filtered and cross-term suppressed, wherein the formula of re-filtering is as follows: This formula is a transfer function in digital signal processing, which is used to realize pre-emphasis filtering (first-order high-pass filtering).

[0064] Wherein, H(z): represents the transfer function of the digital filter, which is used to describe the characteristics of the filter in the Z domain.

[0065] z −1 : is a unit delay operator, which represents a delay of one sampling period in digital signal processing.

[0066] Formula logic: this is a first-order high-pass filter, which is used to enhance the high-frequency components in the signal and suppress the low-frequency components. In leak detection, it is used to compensate for the attenuation of high-frequency components of the vibration signal in pipeline propagation, so that the signal spectrum is more flat, which is convenient for subsequent analysis. The coefficient 0.98 determines the cutoff frequency of the filter.

[0067] Further, cross-term suppression is based on the following formula: This formula smoothes the time-frequency distribution by two-dimensional convolution to suppress the cross-term interference inherent in the WVD algorithm.

[0068] LWVD smooth (t,f): represents the time-frequency distribution value after smoothing at time t and frequency f.

[0069] LWVD(t,f): represents the original, unsmoothed time-frequency distribution value. *: represents convolution operation.

[0070] G(t): represents the time-domain Gaussian kernel function, which is used for smoothing in the time direction. H(f): represents the frequency-domain Hanning window function, which is used for smoothing in the frequency direction.

[0071] S4, according to the confidence, a hierarchical response strategy is executed; wherein, when the confidence is higher than a first preset threshold, a local control action is triggered at the edge computing node; when the confidence is between the first preset threshold and a second preset threshold, a feature vector extracted based on the standardized signal is uploaded to a cloud analysis module.

[0072] In specific implementation, the introduction of confidence provides a scientific basis for subsequent decision-making, enabling the system to distinguish between situations of different emergency levels. The hierarchical response strategy implemented according to the confidence reflects the wisdom architecture of edge-cloud collaboration, with high-confidence emergency situations being immediately addressed at the edge computing node to ensure the timeliness of emergency response, and medium-confidence suspicious situations being uploaded to the cloud for further analysis to balance processing depth and resource consumption. This differentiated processing mechanism not only ensures the rapid control of major leaks, but also ensures the accurate judgment of complex situations.

[0073] In some preferred embodiments, the hierarchical response strategy includes: if the confidence is greater than 85%, triggering valve closing or pressure regulation actions directly at the edge computing node; if the confidence is between 40% and 85%, uploading the dimensionally reduced feature vector to the cloud analysis module; if the confidence is less than 40%, triggering a signal reacquisition process.

[0074] In specific implementation, when the confidence is greater than 85%, valve closing or pressure regulation actions are triggered directly at the edge computing node, which provides extremely fast response capability for emergency situations such as pipe bursts. In pipeline leakage accidents, response time is directly related to the degree of water resource loss and secondary disaster risk. Traditional processes that require manual confirmation or cloud decision-making often take several minutes or even longer. However, this method, by giving autonomous decision-making power at the edge computing node, can complete emergency control within seconds or even milliseconds, maximizing the control of accident impact. This local fast response under high confidence not only relies on accurate detection algorithms, but also needs to establish a reliable security mechanism to ensure that only in cases with strong evidence will critical operations be performed, thereby avoiding unnecessary water supply interruptions caused by false actions.

[0075] Further, when the confidence is between 40% and 85%, the dimensionally reduced feature vector is uploaded to the cloud analysis module, which balances the contradiction between processing accuracy and resource consumption. Medium-confidence situations usually correspond to relatively complex or weak leakage characteristics, which require more in-depth analysis for accurate judgment. Uploading feature vectors rather than raw data not only ensures the information integrity of cloud analysis, but also significantly saves communication bandwidth and storage resources. Feature dimensionality reduction further optimizes data transmission efficiency, by extracting the most discriminative feature dimensions, it minimizes data volume while preserving key information, enabling the system to achieve efficient collaboration under limited network conditions.

[0076] Further, when the confidence is lower than 40%, a signal reacquisition process is triggered, which reflects the prudent handling principle of the system for uncertain situations. Low confidence often means poor signal quality or unobvious features, and blind judgment is likely to produce false positives. By reacquiring signals, higher quality data can be obtained for further analysis, and false positives caused by accidental interference can be avoided. This hierarchical response strategy as a whole builds an intelligent decision-making system, which not only ensures rapid disposal in emergency situations, but also ensures in-depth analysis of complex situations, while taking into account resource utilization efficiency, and realizes the optimal allocation of limited computing resources and communication bandwidth.

[0077] S5, in the cloud analysis module, combining the feature vector, historical time series data and hydraulic model simulation results, multi-model fusion analysis is carried out to accurately locate and verify the leakage point and determine the leakage result.

[0078] In specific implementation, the multi-model fusion analysis carried out in the cloud analysis module fully utilizes the computing and storage advantages of the cloud platform. By combining the feature vector uploaded by the edge, long-term historical time series data and hydraulic model simulation results, a multi-dimensional verification system is constructed. The time series analysis model can identify the time law of leakage development, and the hydraulic simulation model can reproduce the pressure distribution characteristics of the pipe network. Cross-validation of the two effectively eliminates accidental interference and significantly improves the reliability of the detection conclusion.

[0079] In particular, the algorithm for spatial matching of acoustic positioning results and hydraulic abnormal areas further improves the positioning accuracy of leakage points through multi-source evidence fusion, so that the system can not only detect leaks in time, but also provide accurate location guidance for repair operations.

[0080] In some preferred embodiments, the multi-model fusion analysis includes: using a long short-term memory network model to perform time series verification on the received feature vector based on historical time series data, outputting a leakage probability; running a hydraulic simulation model, dynamically calibrating pipe network parameters, generating a pressure anomaly gradient field, and determining a hydraulic anomaly area; based on the acoustic feature coordinates and the hydraulic anomaly area, constructing a spatial matching optimization model, and solving the optimal coordinates of the leakage point by least squares method, the acoustic feature coordinates are determined based on the time-frequency features.

[0081] In a specific implementation, the received feature vector is verified based on historical time series data using a long short-term memory network model, which introduces time dimension analysis and can effectively distinguish between real leaks and transient disturbances. Pipeline leaks usually have a time persistence feature, while environmental disturbances often appear instantaneously or randomly. The long short-term memory network can identify time patterns that meet the development rules of leaks by capturing time-dependent relationships through memory units, thereby reducing false positive rates. Meanwhile, by comparing and analyzing the historical data of the pipe section, the differences between normal and leak states can be established, further enhancing the accuracy of the judgment.

[0082] Further, the present application comprehensively evaluates the leakage probability to determine an accurate probability value, specifically based on the following formula: This formula is used to comprehensively evaluate the leakage probability at coordinates (x, y) in the cloud analysis layer.

[0083] P leak (x, y): represents the comprehensive leakage probability value calculated at the plane coordinates (x, y) position. The higher the value, the greater the likelihood of a leak occurring at that location.

[0084] α: represents the weight coefficient of acoustic evidence, with a value of 0.6. This means that in the fusion strategy of the present application, acoustic evidence accounts for 60% of the weight in the final decision.

[0085] E LWVD (x, y): represents the normalized acoustic energy at coordinates (x, y) obtained by analyzing the low-frequency Wigner-Ville distribution algorithm. This value reflects the intensity of the vibration signal and is the core output of acoustic analysis.

[0086] β: represents the weight coefficient of hydraulic evidence, with a value of 0.4. Complementary to the acoustic weight, it represents a weight of 40% for hydraulic evidence.

[0087] P real (x): represents the actual pressure value measured at position x by the pressure sensor.

[0088] P model (x): represents the theoretical pressure value calculated at position x by the hydraulic model simulation.

[0089] Furthermore, by running a hydraulic simulation model to dynamically calibrate pipeline parameters and generate anomaly gradient fields, the physical simulation model is introduced into the detection system. By establishing a digital twin of the pipeline network's hydraulics, the pressure distribution patterns under different operating conditions can be simulated. When a significant deviation occurs between the actual monitored pressure and the simulation results, potential leak areas can be identified. This verification method based on physical principles and the detection method based on signal analysis complement each other effectively. The hydraulic model is particularly suitable for detecting slowly developing leaks, which may not produce obvious vibration signals but will cause gradual changes in the pipeline network's pressure characteristics. Therefore, the combination of the two methods expands the range of leak types that the system can detect.

[0090] Furthermore, a spatial matching optimization model was constructed based on acoustic feature coordinates and hydraulic anomaly zones, and the optimal coordinates of the leakage point were iteratively solved using the least squares method. This technique achieves the organic integration of acoustic localization and hydraulic pressure localization. Acoustic localization is based on the propagation time and intensity attenuation of vibration signals, featuring rapid response and direct location, but it may introduce errors in complex pipe networks due to sound wave reflection and refraction. Hydraulic pressure localization is based on abnormal pressure distribution in the pipe network, reflecting the overall impact area of ​​the leak, but the boundaries are relatively vague. Integrating these two types of evidence through a spatial matching model and using the least squares method to find the optimal solution is essentially using mathematical optimization methods to find a location point that simultaneously satisfies both acoustic and hydraulic characteristics. This cross-validation mechanism significantly improves the accuracy and reliability of localization, especially in areas with complex pipe network structures, effectively overcoming the limitations of single methods.

[0091] Specifically, the process of solving for the optimal coordinates of the leakage point is as follows: Step 1: Multi-source data spatial mapping First, acoustic feature coordinate extraction is performed: energy peak points are extracted from the low-frequency Wigner-Ville distribution time-frequency matrix to form an acoustic feature coordinate set P. a = {(x i , y i , t i E i )}, where (x i , y i ) represents the geographic coordinates of the i-th vibration sensor, t i E represents the timestamp of the energy peak. i This represents the corresponding normalized energy value.

[0092] Calculate the sound wave propagation distance based on the sound wave propagation speed model: d i =v w ·(t i -t0), where d i v represents the distance sound waves travel. wt0 represents the speed at which sound waves propagate in a specific pipe material, and t0 represents the initial time when the leak occurs.

[0093] Secondly, perform hydraulic anomaly zone generation: run the EPANET hydraulic model to obtain the pressure anomaly network G. p = {(x j ,y j , Δp j )}, where (x j , y j ) represents the coordinates of the pressure monitoring point, Δp j This indicates the abnormal pressure value at that point.

[0094] Construct the pressure gradient field: ∇p j = √[(∂p / ∂x) 2 + (∂p / ∂y) 2 ], where ∇p j Indicates the coordinate (x) j ,y j The pressure gradient value at ().

[0095] Step 2: Construction of Spatial Matching Optimization Model Establish the objective function: min Σ k=1 n w k || L est - L k || 2 , where L est = (x est , y est ) represents the coordinates of the leak point to be determined, L k w represents the projected coordinates of the k-th evidence point. k Indicates the weight of evidence.

[0096] Weight of evidence calculation: w k = αE k + β|∇p k | where α is the acoustic evidence weighting coefficient, β is the hydraulic evidence weighting coefficient, and E k For the acoustic energy normalization value, |∇p k | represents the absolute value of the pressure gradient.

[0097] Step 3: Optimize the solution using the least squares method Constructing an overdetermined system of equations: The x-coordinate of the point to be identified (e.g., longitude or X-coordinate in a certain coordinate system).

[0098] The ordinate (e.g. latitude or Y coordinate in a certain coordinate system) representing the leakage point to be solved.

[0099] Perform an iterative solution process: First, use the weighted centroid method to calculate the initial position: x0 = (Σ k=1 n w k x k ) / (Σ k=1 n w k )y0 = (Σ k=1 n w k y k ) / (Σ k=1 n w k ) Then perform the Gauss-Newton iteration: Set the maximum number of iterations, for each iteration: Calculate the residual vector: r i = w i ‖ L est - L i ‖; Construct the Jacobian matrix: J ij = ∂r i / ∂x j ; Solve the increment: Δx = - (J T J) -1 J T r ; Update the estimate: L est ← L est + μΔx, where μ is the step size; Terminate the iteration when the increment norm ‖Δx‖ is less than the preset threshold, and obtain the result.

[0100] In some preferred embodiments, the method further comprises the following steps: when the network between the edge computing node and the cloud analysis module is interrupted, enable the offline analysis mode at the edge computing node, and cache the collected data using the local ring buffer; after the network is restored, automatically resume the cached data to the cloud.

[0101] In specific implementation, the water supply network often spans a wide geographical area, and some pipe sections may be located in remote areas with poor network signal coverage, or communication may be interrupted due to natural disasters or human damage. In such cases, the traditional system that relies entirely on cloud processing may be paralyzed and unable to perform monitoring and protection. The method enables offline analysis mode on the edge computing node, so that even if the connection with the cloud is lost, the local node can still maintain basic detection and analysis capabilities and continue to monitor the pipeline status. This distributed intelligent architecture ensures that the core functions are not lost due to network problems.

[0102] Further, the design of the local ring buffer takes into account the extreme case of long-term network disconnection. Through the cyclic storage mechanism, it can save the key data of the recent period in the limited local storage space, avoiding data loss due to storage overflow, and ensuring that the complete state trajectory during network disconnection can be reconstructed after network recovery. The cached data is automatically transferred to the cloud after network recovery, which ensures data integrity and analysis continuity.

[0103] Further, in traditional systems, data during network interruption is often permanently lost, resulting in gaps in monitoring records and making it impossible to conduct post-analysis and trace responsibility. The automatic transfer function enables the cloud to obtain complete time series data, providing complete information for analyzing leakage development patterns and optimizing detection models. At the same time, the transferred data can be correlated with other regional data received by the cloud during this period to reveal the operation status of the pipeline network in a larger range. This network interruption emergency handling capability is particularly suitable for monitoring and protection of critical infrastructure, as pipeline leakage accidents may be accompanied by communication interruption (such as optical cable damage caused by pipe explosion impact). The system can maintain basic functions after the accident, which is of great significance for disaster control and emergency command. In addition, this function reduces the dependence of the system on network quality, making it possible to deploy in rural areas or developing countries with poor network conditions, expanding the application scope of the technology.

[0104] In some preferred embodiments, the method further comprises the steps of: aggregating the model gradients trained by each edge computing node through a federated learning framework, and performing global model updating in the cloud; and distributing the updated incremental model to each edge computing node.

[0105] In specific implementation, traditional centralized machine learning requires aggregation of raw data from each node to the cloud, which poses two major challenges in practical application: first, large data transmission consumes a large amount of bandwidth resources, especially in pipeline monitoring scenarios with a large number of sensor nodes; second, pipeline operation data may contain sensitive geographic information and water operation information, and direct transmission poses a risk of privacy leakage.

[0106] The application fundamentally avoids the large-scale flow of private data by using a federal learning framework, in which each edge computing node locally trains a model using collected data and only uploads the model gradient obtained by training to the cloud for aggregation. The original data is always kept locally, which meets the increasingly stringent data security regulations. A special aggregation algorithm is used for global model updating in the cloud, which can extract common knowledge from the local training results of each node, eliminate model bias caused by local data bias, and form a global model with stronger generalization ability. The updated incremental model is distributed to each edge computing node, so that the entire system can continuously learn and improve. With the accumulation of running time, the detection algorithm continuously adapts to the characteristics of the specific pipe network environment, gradually improving the detection performance under different working conditions, different pipe sections, and different seasons. This gradual optimization process is particularly suitable for complex tasks such as pipeline leakage detection, because the environmental noise characteristics and leakage forms of actual pipe networks have great regional differences. Personalized adaptation achieved through federal learning can significantly improve practical effectiveness.

[0107] In addition, the federal learning framework also has good scalability. When new sensor nodes or new pipe network areas are added, they can quickly integrate into the existing learning system, share existing knowledge accumulation, and accelerate the formation of new node capabilities. This continuous evolution mechanism enables the system to adapt to the effects of long-term factors such as pipe network aging and water usage pattern changes, maintaining the stability of detection performance and providing long-term reliable technical support for water supply pipe networks.

[0108] In some preferred embodiments, the method further comprises the steps of: performing corresponding response actions according to the leakage results, and feeding back the related data of the leakage results to the digital twin model of the pipe network to optimize the preventive maintenance strategy of the pipe network.

[0109] In specific implementation, performing corresponding response actions according to the leakage results makes the detection system no longer just provide alarm information, but directly drive the processing flow, forming a complete monitoring-decision-execution closed loop. For example, for confirmed major leaks, automatically generate maintenance work orders and assign them to the nearest maintenance team, for suspected leaks that need further investigation, arrange inspection verification, and for frequently abnormal key pipe sections, increase monitoring frequency. This intelligent operation and maintenance scheduling based on detection results greatly improves work efficiency and reduces the delay and uncertainty of human intervention.

[0110] Further, the method is characterized in that the relevant data of the leakage result is fed back to the digital twin model of the pipe network. The digital twin model is an accurate mapping of the physical pipe network in the virtual space. By continuously integrating actual operation data and fault records, the digital twin model can more and more accurately reflect the real-time state and evolution trend of the pipe network. Each confirmed leakage event contains rich information such as location, size, time, and repair condition. After the information is reversely labeled to the digital twin model, it can be correlated and analyzed with static parameters such as pipe material, service life, soil environment, and load pressure, to mine the rules and causes of leakage. Based on these accumulated knowledge, the digital twin model can scientifically evaluate and predict the health of the pipe network, identify high-risk pipe sections and potential weak links, and provide data-driven decision support for preventive maintenance.

[0111] For example, the model may find that pipes of a certain material have a significantly increased leakage probability after a certain service life, or pipes in a certain area have a shortened service life in an environment with strong soil corrosion. These insights can guide targeted inspections and old pipe network renovation plans, reducing leakage from the source. This preventive maintenance strategy based on the digital twin changes the pipe network management from passive emergency repair to active risk prevention and control, not only reducing water resource loss and maintenance cost, but more importantly, improving the reliability and public service quality of the water supply system, and optimizing asset management.

[0112] The embodiment of the present application provides a water supply pipeline leakage detection method, comprising: collecting multi-source signals through a plurality of sensor nodes arranged on the pipeline, wherein the multi-source signals include pipeline vibration signals, pressure signals and temperature signals; pre-processing and spatio-temporal alignment are performed on the collected multi-source signals at an edge computing node to obtain standardized signals; a pre-set lightweight one-dimensional convolutional neural network model is used to analyze the standardized signals at the edge computing node to obtain a preliminary leakage detection result and a corresponding confidence level; a hierarchical response strategy is executed according to the confidence level; wherein when the confidence level is higher than a first pre-set threshold, a local control action is triggered at the edge computing node; when the confidence level is between the first pre-set threshold and a second pre-set threshold, a feature vector extracted based on the standardized signals is uploaded to a cloud analysis module; the cloud analysis module combines the feature vector, historical time series data and hydraulic model simulation results to perform multi-model fusion analysis, so as to accurately locate and verify a leakage point and determine a leakage result. The present application realizes rapid preliminary judgment and emergency control of leakage through multi-source signal collection and edge-cloud collaborative architecture, combined with lightweight neural network edge real-time analysis and confidence level hierarchical response mechanism. Further, through cloud multi-model fusion and spatio-temporal matching verification, the detection accuracy and positioning reliability are significantly improved. The whole method ensures real-time performance while ensuring the accuracy of the result through multi-level verification, thereby providing a real-time, accurate and reliable complete solution for pipeline leakage monitoring.

[0113] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0114] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0115] The steps in the method embodiments of the present application can be adjusted, combined and reduced in sequence according to actual needs. The units in the device embodiments of the present application can be combined, divided and reduced according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0116] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application.

[0117] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0118] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, these modifications and variations of the present application are intended to be included within the scope of the claims of the present application and their equivalent technologies. Therefore, the present application is also intended to include these modifications and variations.

[0119] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for detecting leakage in water supply pipelines, characterized in that, include: Multiple sensor nodes deployed on the pipeline collect multi-source signals, including pipeline vibration signals, pressure signals, and temperature signals. At the edge computing node, the acquired multi-source signals are preprocessed and spatiotemporally aligned to obtain standardized signals; At the edge computing node, the standardized signal is analyzed using a preset lightweight one-dimensional convolutional neural network model to obtain preliminary leakage detection results and corresponding confidence levels. A tiered response strategy is executed based on the confidence level; wherein, when the confidence level is higher than a first preset threshold, a local control action is triggered at the edge computing node. When the confidence level is between the first preset threshold and the second preset threshold, the feature vector extracted based on the standardized signal will be uploaded to the cloud analysis module. The cloud-based analysis module combines the feature vectors, historical time-series data, and hydraulic model simulation results to perform multi-model fusion analysis, thereby accurately locating and verifying leakage points and determining the leakage results.

2. The water supply pipeline leakage detection method according to claim 1, characterized in that, The preprocessing includes signal noise reduction and temperature compensation; the spatiotemporal alignment includes using a preset precision time protocol to achieve time synchronization of multiple nodes and using an interpolation algorithm to unify the time axis of heterogeneous data.

3. The water supply pipeline leakage detection method according to claim 1, characterized in that, Before the standardized signal is analyzed using a preset lightweight one-dimensional convolutional neural network model at the edge computing node to obtain preliminary leakage detection results and corresponding confidence levels, the method further includes: The time-frequency characteristics of the pipeline vibration signal are obtained by performing time-frequency analysis using a preset low-frequency Wigner-Ville distribution algorithm.

4. The water supply pipeline leakage detection method according to claim 3, characterized in that, The step of performing time-frequency analysis on the pipeline vibration signal using a preset low-frequency Wigner-Ville distribution algorithm to obtain time-frequency characteristics includes: An adaptive window function mechanism is adopted to dynamically select the window function type and window length based on the center frequency of the frequency band of the pipeline vibration signal; Based on Bark band reconstruction, the linear frequency spectrum of the pipeline vibration signal is converted into multiple critical frequency bands that conform to the characteristics of human hearing, generating a reconstructed time-frequency distribution. Based on the material of the pipe, the corresponding attenuation coefficient and frequency correction weight are obtained from the preset acoustic attenuation parameter library, and the energy of the time-frequency distribution is compensated and calculated to obtain the time-frequency characteristics.

5. The method for detecting leakage in water supply pipelines according to claim 1, characterized in that, The hierarchical response strategy includes: If the confidence level is greater than 85%, the valve closing or pressure regulation action will be directly triggered at the edge computing node. If the confidence level is between 40% and 85%, the dimensionality-reduced feature vector will be uploaded to the cloud analysis module. If the confidence level is below 40%, the signal reacquisition process is triggered.

6. The water supply pipeline leakage detection method according to claim 1, characterized in that, The multi-model fusion analysis includes: The received feature vectors are verified using a long short-term memory network model based on historical time-series data, and the probability of missing data is output. Run the hydraulic simulation model, dynamically calibrate the pipeline parameters, generate a pressure anomaly gradient field, and determine the hydraulic anomaly zone; Based on acoustic feature coordinates and hydraulic anomaly zones, a spatial matching optimization model is constructed, and the optimal coordinates of leakage points are solved iteratively using the least squares method. The acoustic feature coordinates are determined based on the time-frequency features.

7. The method for detecting leakage in water supply pipelines according to claim 1, characterized in that, The method further includes: When the network between the edge computing node and the cloud analytics module is interrupted, the offline analytics mode is enabled on the edge computing node, and the collected data is cached using a local circular buffer. Once the network is restored, the cached data will be automatically uploaded to the cloud.

8. The method for detecting leakage in water supply pipelines according to claim 1, characterized in that, The method further includes: By using a federated learning framework, the model gradients trained on each edge computing node are aggregated, and the global model is updated in the cloud. The updated incremental model is then distributed to each edge computing node.

9. The method for detecting leakage in water supply pipelines according to claim 1, characterized in that, The method further includes: Based on the leakage results, corresponding response actions are executed, and the relevant data of the leakage results are fed back to the digital twin model of the pipeline network to optimize the preventive maintenance strategy of the pipeline network.

10. A water supply pipeline leakage detection system, characterized in that, The method for performing any one of claims 1-9 includes an edge sensing module, an edge computing node, and a cloud analytics module, wherein the edge sensing module includes multiple sensor nodes deployed on the pipeline.

Images