A multi-parameter perception-based water supply network leakage intelligent diagnosis method and system

Abstract

The application discloses a kind of based on multi-parameter perception's water supply pipe network leakage intelligent diagnosis method and system, belong to wisdom water affair technical field;System includes three-in-one intelligent water meter terminal, and cloud diagnosis platform and pipe network management terminal.Through three-in-one intelligent water meter synchronous acquisition pipe network pressure, flow and vibration noise data, data cleaning is carried out in edge, feature extraction and time synchronization are uploaded to cloud after;Cloud is based on the coupling relationship of pressure fluctuation, flow continuity and acoustic leakage characteristics, combined with artificial intelligence model and water hammer model executes multi-source fusion diagnosis, identifies normal, leakage or abnormal water state, outputs risk level and alarm information and pushes to management terminal.The application realizes multi-parameter integrated synchronous perception, improves leakage identification precision and concealed leak location ability, reduces false positive rate and operation and maintenance cost, supports end cloud cooperation and model online iteration, is suitable for large-scale online monitoring and fine management of urban water supply pipe network.

Description

Technical Field

[0001] This invention relates to the field of smart water management and municipal water supply network monitoring technology, and in particular to a technical solution for network anomaly identification and leakage diagnosis based on pressure, flow and acoustic parameter sensing, which is applicable to engineering application scenarios such as online monitoring, leakage identification and operation status assessment of urban water supply networks. Background Technology

[0002] Urban water supply networks are characterized by their large scale, complex structure, and variable operating environment, making it difficult to identify leaks and abnormal water usage patterns in a timely and accurate manner. Existing leak detection devices in urban water supply networks are mostly portable devices made from manual listening rods or handheld leak detectors. During acoustic testing, the vibration sound from the pipe wall is transmitted from valves or fittings to a human earpiece for judgment. While these manual detection devices are simple in structure, they rely solely on human experience for short-term inspections and cannot meet the requirements for long-term online monitoring, identification of minute leaks, and automatic judgment in complex noise environments. Furthermore, their level of automation is low. With continuous technological advancements, electronic leak listening devices and some machine learning-based water supply network analysis methods have emerged.

[0003] Existing acoustic signal acquisition devices mostly combine piezoelectric or acoustic emission sensors with simple fixed structures to record sound waves. For example, Chinese utility model patent CN203190050U discloses an underground pipeline leakage detection system, which describes the detectors placed at both ends of the pipeline and their signal acquisition methods. It can detect leakage signals. However, such devices usually require manual adjustment of the sensor position to ensure signal quality during actual deployment. After each deployment, the sensor position needs to be manually disassembled, repositioned, and the parameters adjusted.

[0004] Existing pipe network leakage monitoring devices mostly utilize a single communication method and fixed sampling mode to achieve data transmission. For example, Chinese invention patent CN110030506A discloses an "Internet of Things-based water supply network leakage monitoring system," which uses sensor nodes and gateways to control data collection and uploads it to the cloud for analysis, possessing a certain degree of automated monitoring capability. However, such devices struggle to effectively filter interference noise in complex urban environments and are unable to achieve accurate identification of minute leaks and comprehensive leakage control.

[0005] Therefore, there is an urgent need for an intelligent diagnostic method and system for water supply network leakage that integrates multi-parameter sensing to achieve high-precision identification and location of water supply network leakage events. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a method and system for intelligent diagnosis of water supply network leakage based on multi-parameter perception, which addresses the shortcomings of the prior art. The invention innovates and improves the water supply network noise leakage identification device by introducing pressure parameters on the basis of acoustic monitoring, resulting in a three-in-one intelligent water meter based on pressure, flow rate, and acoustics to judge changes in the hydraulic state of the network. Furthermore, it combines artificial intelligence algorithms to conduct multi-source data fusion analysis, thereby achieving higher accuracy and more reliable judgment of network leakage and abnormal water use behavior.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A smart diagnostic system for water supply network leakage based on multi-parameter sensing includes a three-in-one smart water meter terminal, a cloud diagnostic platform, and a network management terminal. Among them, the three-in-one smart water meter terminal integrates a flow meter, pressure sensor, water sound monitoring sensor, edge processing module, communication module and power module for multi-parameter synchronous acquisition and edge preprocessing; The cloud-based diagnostic platform is used for multi-source fusion analysis, leakage risk assessment, abnormal water use identification, alarm management, and visualization. The pipeline management terminal is used to provide maintenance personnel with a real-time monitoring interface, alarm information, and pipeline health assessment results.

[0008] As a further preferred embodiment of the intelligent diagnostic system for water supply network leakage based on multi-parameter sensing of the present invention, the flow meter, pressure sensor and water sound monitoring sensor in the three-in-one smart water meter terminal are all built into the same water meter housing, realizing synchronous acquisition of multiple parameters at the same location and time reference; the water sound monitoring sensor is a hydrophone, the pressure sensor is a high-frequency response sensor, and the flow meter is an electromagnetic or ultrasonic flow meter.

[0009] As a further preferred embodiment of the intelligent diagnostic system for water supply network leakage based on multi-parameter perception of the present invention, the edge processing module is configured to perform at least one of the following processes: performing outlier removal, time alignment, and sliding window smoothing on flow data; performing bandpass filtering, noise reduction enhancement, short-time Fourier transform, or wavelet decomposition on acoustic signals; performing pulsation decomposition, abrupt change detection, and time domain normalization on pressure time series; and performing multi-source time synchronization and data quality marking; the communication module adopts a three-mode converged communication architecture of NB-IoT, 4G, and Bluetooth.

[0010] As a further preferred embodiment of the intelligent diagnostic system for water supply network leakage based on multi-parameter perception of the present invention, the cloud diagnostic platform is further configured to: collect historical monitoring data and actual maintenance results to construct a labeled sample library; dynamically update alarm thresholds and feature discrimination boundaries; periodically retrain the model based on new samples, and send the updated model parameters to the edge processing module.

[0011] As a further preferred embodiment of the intelligent diagnostic system for water supply network leakage based on multi-parameter perception of the present invention, the cloud-based diagnostic platform also integrates geographic information system and digital twin technology to dynamically mark abnormal locations, pressure drop areas and leakage heat maps on a three-dimensional network topology map, supporting multi-view linkage display and intelligent scheduling decision-making.

[0012] A method for intelligent diagnosis of water supply network leakage based on multi-parameter sensing, using the system described in any one of claims 1 to 5, includes the following steps: Step 1: Simultaneously collect network pressure data, instantaneous flow data, and pipeline vibration and noise data using a three-in-one smart water meter installed at the nodes of the water supply network; Step 2: Perform edge preprocessing and feature construction on the collected multi-parameter data, including noise reduction filtering, outlier removal, temporal resampling, and multi-scale feature extraction; Step 3: Upload the preprocessed feature data to the cloud diagnostic platform. Based on the coupling relationship between pressure fluctuation features, flow continuity features and acoustic leakage features, establish a multi-source fusion diagnostic model to identify whether the current operating condition is normal water supply, pipeline leakage, or abnormal water use. Step 4: When the identification result meets the preset anomaly criteria, generate the leakage risk level, anomaly type label and alarm information, and push them to the pipeline management terminal; Step 5: Based on historical operational data and manual verification results, periodically update the parameters of the diagnostic model to achieve continuous learning and adaptive optimization of the model.

[0013] As a further preferred embodiment of the intelligent diagnosis method for water supply network leakage based on multi-parameter sensing of the present invention, in step 2, data preprocessing and feature construction further include: Perform outlier removal, time alignment, and sliding window smoothing on the traffic data; Perform bandpass filtering, noise reduction and enhancement, short-time Fourier transform or wavelet decomposition on acoustic signals to extract time-domain, frequency-domain and time-frequency-domain features; Empirical mode decomposition, accumulation, and abrupt change detection, along with time-domain normalization, are performed on the stress time series. Perform multi-source time synchronization and data quality marking on three types of data.

[0014] As a further preferred embodiment of the intelligent diagnostic method for water supply network leakage based on multi-parameter perception of the present invention, in step 3, the multi-source fusion diagnostic model adopts one or more artificial intelligence algorithms among XGBoost, support vector machine, convolutional neural network or long short-term memory network, takes the fusion feature matrix as input, outputs the abnormal probability or risk score, and performs double verification in combination with threshold discrimination mechanism.

[0015] As a further preferred embodiment of the intelligent diagnosis method for water supply network leakage based on multi-parameter perception of the present invention, in step 4, the leakage risk level is calculated by a multi-index weighted fusion method, wherein the multi-index includes: a macroscopic leakage index calculated based on flow balance deviation, a local leakage probability calculated based on acoustic anomaly intensity, and a hydraulic anomaly confidence level calculated based on the consistency between pressure transient response and water hammer model.

[0016] As a further preferred embodiment of the intelligent diagnosis method for water supply network leakage based on multi-parameter perception of the present invention, in step 5, the model update adopts an incremental learning algorithm, the model is periodically retrained based on the labeled samples after manual verification, and the updated model parameters are sent to the edge processing module to realize the collaborative evolution between the cloud and the edge.

[0017] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention achieves multi-parameter integrated monitoring. By integrating flow, pressure and noise sensing units into a single water meter terminal, it realizes multi-dimensional synchronous perception of the pipeline network operation status, avoiding the problems of complex installation and data asynchrony caused by the traditional decentralized deployment of multiple devices. 2. This invention can significantly improve the accuracy of leakage identification. By utilizing the collaborative discrimination mechanism of macroscopic flow screening, fine acoustic sensing and pressure-hydraulic constraints, it can effectively reduce the false alarm rate and false alarm rate of single acoustic or single flow methods and improve the ability to identify hidden leakage. 3. This invention reduces system deployment and maintenance costs. Through the integrated design of the three-in-one terminal and the edge preprocessing mechanism, it reduces the number of field devices and data transmission pressure, thereby improving the economic feasibility of large-scale system application. 4. This invention supports edge-cloud collaboration and online model iteration, making it suitable for large-scale engineering deployment and long-term online monitoring. By combining artificial intelligence models to continuously learn and analyze multi-source time-series data, it can also identify abnormal water use and potential leakage risks in advance, thereby improving the level of refined management of water supply networks. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structural composition of the three-in-one smart water meter terminal of the present invention; Figure 2 This is a flowchart of the traffic data processing of the present invention; Figure 3 This is a flowchart of the acoustic data processing of the present invention; Figure 4 This is a flowchart of the pressure data processing of the present invention; Figure 5 This is the overall flowchart of the intelligent diagnosis method for water supply network leakage of the present invention.

[0019] The specific labels in the diagram are as follows: 1—Flow meter; 2—Pressure sensor; 3—Underwater acoustic monitoring sensor; 4—Edge processing module; 5—Communication module; 6—Power supply module. Detailed Implementation

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings: The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0021] To further illustrate the technical means and effects adopted by this invention to achieve its intended purpose, the following detailed description, in conjunction with the accompanying drawings, details the specific implementation, structure, features, and effects of a multi-parameter sensing-based intelligent diagnostic method and system for water supply network leakage proposed according to this invention. The following detailed description of the invention can be better understood in conjunction with the accompanying drawings, wherein the elements and features of the invention are identified by reference numerals.

[0022] like Figure 1 As shown, this invention discloses an intelligent diagnostic method and system for water supply network leakage based on "flow-pressure-water sound" data sensing. Its core lies in the integrated design of a "flow-pressure-water sound" three-in-one smart water meter terminal. This terminal mainly consists of a flow meter 1, a pressure sensor 2, a water sound monitoring sensor 3, an edge processing module 4, a communication module 5, and a power supply module 6.

[0023] The system comprises the following components: Flowmeter 1 acquires instantaneous and cumulative flow data, providing foundational data for subsequent water conservation analysis; Pressure Sensor 2 collects pipeline pressure time series data to construct transient hydraulic response characteristics; and Water Acoustic Monitoring Sensor 3 collects leak-related acoustic signals propagating within the pipe, capturing weak pressure fluctuations and transient pressure wave disturbances caused by minute leaks, particularly the water hammer effect and specific frequency acoustic disturbances caused by leakage. Edge processing module 4 employs an embedded microcontroller running a real-time operating system, providing data cleaning, feature extraction, and preliminary anomaly identification capabilities to reduce cloud computing pressure and communication load. Communication module 5 utilizes a NB-IoT, 4G, and Bluetooth tri-mode converged communication architecture for remote data transmission, serving as a bridge for interaction between the terminal and external cloud platforms and mobile terminals. Power module 6 provides a stable power supply to all components.

[0024] The aforementioned flow meter, pressure sensor, and underwater acoustic monitoring sensor are all built into the inside of the water meter housing. This integrated design overcomes the data alignment issues caused by the dispersed deployment of multiple devices in existing technologies. Through this high level of physical integration, this invention fundamentally achieves synchronous acquisition of flow, pressure, and acoustic parameters at the same location and time reference, based on the hardware.

[0025] Figure 5 This is the overall flowchart of the intelligent diagnosis method for water supply network leakage of the present invention. Figure 5 As shown, this method first collects flow, pressure, and acoustic data synchronously through three-in-one smart water meters deployed at pipeline nodes; then, it preprocesses and extracts features from the three types of data at the edge; after completing multi-source time synchronization, the feature data is uploaded to the cloud diagnostic platform; multi-source fusion analysis is performed in the cloud, and leakage diagnosis and location are performed by combining water hammer models and artificial intelligence algorithms; finally, alarm information is generated and pushed to the pipeline management terminal, while the model is continuously learned and updated using feedback data.

[0026] The following is a combination of the sub-flowcharts ( Figure 2 , Figure 3 , Figure 4 Each step is explained in detail: like Figure 2 As shown, the traffic data processing flow includes the following steps: Step S201: Obtain the flow information to be diagnosed from the flow sensor built into the housing of the three-in-one smart water meter. The flow sensor is preferably an electromagnetic flow meter, ultrasonic flow meter, or other flow measurement device suitable for pipeline metering. It continuously measures the instantaneous and cumulative flow through the water meter at a preset sampling frequency, and outputs a standard signal (digital or analog signal) to the edge processing module for subsequent water volume analysis and conservation calculations. The flow sensor is installed in a distributed manner at the user-side nodes along with the water meter, forming a metering sensing network for each water-using node at the system level, thereby obtaining the basic water input and output data for each section of the pipeline network.

[0027] Step S202: Perform data cleaning on the flow information to be diagnosed collected by the flow sensor built into the housing of the three-in-one smart water meter, and remove outliers.

[0028] Step S203: Perform time synchronization processing under a unified time reference. Since the responses of flow, pressure, and underwater acoustic signals to the same hydraulic disturbance have different time scales and propagation delays, in order to ensure the physical consistency and comparability of multi-source characteristics (i.e., the above-mentioned flow, pressure, and underwater acoustic signals), this step uses a high-precision time synchronization module to time-align the three types of data, and uses a sliding window weighted moving average for smoothing to ensure that the time deviation is controlled within an acceptable range.

[0029] Step S204: Based on the flow information to be diagnosed, perform signal analysis to determine the flow change information within the collection range of the first flow monitoring device, including extracting features such as the trend and abrupt change points of the flow curve.

[0030] Step S205: Perform the water balance analysis based on the flow change information collected within the range of the first flow monitoring device. The edge processing module or cloud diagnostic platform performs water balance calculations on the flow data of multiple water meter nodes according to the preset pipeline topology. By constructing node flow balance equations and comparing the difference between upstream water supply and downstream total water consumption, the suspected leakage or abnormal water consumption of the corresponding pipe section is obtained, achieving a preliminary judgment of the macroscopic leakage risk of the pipeline network.

[0031] Step S206: Determine whether the leakage water volume exceeds a preset threshold. If it is less than the threshold (at which point the water meter determines that no leakage has occurred), return to step S201 to continue monitoring; if it exceeds the threshold, proceed to step S207.

[0032] Step S207: Determine the initial leakage range based on flow rate change information. When the leakage volume exceeds the threshold in multiple consecutive monitoring cycles, it is determined that there is a macroscopic leakage risk in the area, thus completing the screening of large-scale risk sections.

[0033] Step S208 prepares for subsequent refined diagnosis. Flow balance analysis, as the first-layer macroscopic screening method, provides regional guidance for subsequent underwater acoustic feature analysis and pressure response analysis. Flow analysis based on water balance is suitable for large-scale judgments and can effectively identify abnormal water loss, but its spatial resolution capability for hidden leak locations is limited. Therefore, this invention uses flow balance analysis as the first-layer macroscopic screening method, in conjunction with subsequent underwater acoustic feature analysis and pressure response analysis, to form a layered diagnostic mechanism.

[0034] like Figure 3 As shown, the acoustic data processing flow includes the following steps: Step S301: Obtain the acoustic information to be diagnosed from the underwater acoustic monitoring equipment in the water supply network to be diagnosed, i.e., obtain the acoustic information to be diagnosed from the underwater acoustic monitoring sensor built into the housing of the three-in-one smart water meter. The underwater acoustic monitoring sensor is preferably a hydrophone, whose frequency response range covers the characteristic frequency band of the leakage signal, and whose sensitivity is sufficient to capture acoustic disturbances caused by minute leaks. The sensor is in direct contact with the water in the pipe through a coupling medium, receives the acoustic signal propagating in the water and converts it into an electrical signal. The sampling frequency is set to a high-frequency value sufficient to preserve the signal characteristics according to the Nyquist theorem, ensuring that the signal is not distorted.

[0035] Step S302: Analyze the acoustic information to be diagnosed, make a preliminary judgment on whether a pipe burst event exists, and record the installation location of the water meter where the event was detected. Through the large-scale deployment of water meters in the pipeline network, a wide-area sensing coverage of the water supply network's operating status can be formed, achieving blind-spot-free monitoring.

[0036] Step S303: Perform time-frequency analysis. When a water pipe leaks, the water flow is obstructed by the leak, generating a specific frequency sound wave signal and pipe wall vibration. The system first applies a bandpass filter to remove low-frequency water flow friction noise, and then uses short-time Fourier transform (STFT, with a Hamming window size of 1024 points) or multi-scale wavelet transform (such as using Daubechies wavelet for 5-level decomposition) to convert the non-stationary acoustic signal from the time domain to the time-frequency domain, effectively removing noise and background environmental interference, and accurately extracting the hidden high-frequency leakage signal components.

[0037] Step S304: Extract the corresponding feature vectors. Statistical features are extracted through time-domain analysis, spectral features are extracted through frequency-domain analysis, and wavelet energy, wavelet entropy, and other features are extracted through time-frequency analysis, forming feature vectors for anomaly detection.

[0038] The aforementioned eigenvectors specifically include: time domain: peak value, valley value, root mean square, etc.; frequency domain: power spectral density, dominant frequency, etc.; time-frequency domain: wavelet energy, wavelet entropy. When a pipe leaks or has a minor seepage, the water flow is obstructed by the leak, and the flow velocity and pressure change, which in turn generates a sound wave signal of a specific frequency and vibrates the pipe wall, producing the corresponding eigenvectors mentioned above.

[0039] Step S305: Determine whether the feature value exceeds the adaptive threshold. The threshold is set using an adaptive method: statistically analyze the feature value distribution of the same time period (e.g., 2-4 AM) over the past 7 days, and use 3 times the mean as the upper threshold and 0.3 times the mean as the lower threshold. If the threshold is not exceeded, return to step S301; if it is exceeded, proceed to step S306.

[0040] Step S306: Mark the presence of acoustic anomalies in the water supply network.

[0041] Step S307: Obtain the prediction information generated by the water hammer model. The water hammer model is a transient hydraulic analysis model built based on the pipeline hydraulic model, used to characterize and simulate the transient process of water flow in the water supply network to be diagnosed. Through the water hammer model, the pressure changes, pressure wave propagation and attenuation characteristics of the fluid in the pipeline under different operating conditions can be calculated and analyzed.

[0042] Water hammer refers to the process by which pressure fluctuations and oscillations propagate within a pipeline due to transient disturbances such as sudden changes in flow velocity, rapid valve opening and closing, pipe rupture, or sudden leakage. These transient pressure waves propagate within the pipeline, forming a series of fluctuations. Strong pressure waves propagate along the pipeline and reflect and superimpose at nodes. Their waveform characteristics are correlated with the type and location of the abnormal event. The pressure wave velocity depends on the elastic modulus and density of water, the elastic modulus of the pipe material, the pipe diameter, wall thickness, and pipeline constraints, and can be calculated using classical formulas.

[0043] The water hammer model is constructed based on the structural parameters and topological relationships of the components such as pipes, valves, and pumps in the water supply network. The model comprehensively considers key parameters such as the inner diameter, wall thickness, pipe material type, length, elastic modulus, and mass, inertial, and damping characteristics related to the transient process for each pipe segment. The method of characteristics (MOC) is used to numerically solve the transient hydraulic response of the network. The MOC transforms the partial differential equations of water hammer into a system of ordinary differential equations along characteristic lines. By integrating along the characteristic lines, the pressure and flow rates at each node at each moment are obtained.

[0044] Step S308: Based on the flow screening results, acoustic anomaly location, and water hammer model prediction information, the location of the pipe burst is preliminarily determined.

[0045] like Figure 4As shown, the pressure data processing flow includes the following steps: Step S401: Obtain the pressure information to be diagnosed collected by the pressure sensing unit built into the housing of the three-in-one smart water meter. The pressure sensor is a high-frequency response sensor, whose response speed and accuracy meet the requirements for capturing transient pressure waves. The sampling frequency is set according to the characteristics of the pressure wave signal to a frequency sufficient to reconstruct the waveform, and the resolution is sufficient to identify minute pressure changes (such as <0.1mmH2O).

[0046] Step S402, Pulsation Decomposition and Empirical Mode Decomposition (EMD). Due to the significant hydraulic noise generated by water usage fluctuations during actual pipeline operation, directly using the raw pressure data can easily lead to misjudgments. The system adaptively decomposes the non-stationary raw pressure signal into a finite number of intrinsic mode functions (IMFs) and a residual sequence using the EMD algorithm. During the iteration process, a stopping criterion is set at a standard deviation less than 0.01, and the first three high-frequency and mid-frequency IMF components are extracted. This effectively separates the transient low-frequency pressure drop caused by pipe bursts, eliminating background hydraulic noise.

[0047] Step S403: Perform the CUSUM (Cumulative Summation) algorithm to identify abrupt changes. The system applies the CUSUM algorithm to the pressure time series reconstructed by EMD. By continuously accumulating small deviations of the pressure signal from the mean, it amplifies the abrupt changes in the arrival of transient pressure waves at the monitoring point, accurately locking the initial arrival time (timestamp) of the negative pressure wave (pressure wave caused by pipe rupture) at the first pressure monitoring device, providing high-precision time input for subsequent Time Difference of Arrival (TDOA) positioning.

[0048] Step S404, Time Domain Normalization. Time domain normalization is performed. To ensure the physical consistency and comparability of multi-source features, the pressure data is time-synchronized and normalized. A high-precision time synchronization module is used to ensure that the time deviation with the flow and acoustic data is controlled within an acceptable range.

[0049] Step S405: Based on pressure change information, a method of characteristics (MOC) analysis is performed using a water hammer model. The system constructs transient hydraulic response characteristics of the pipeline network from the collected pressure time series and uses a water hammer propagation model to identify pressure anomalies. By simulating and analyzing the propagation, reflection, and superposition processes of pressure waves in the pipeline network, the pressure change characteristics of each pipe section under typical disturbance conditions can be obtained. Based on this, potential high-risk sections in the water supply network to be diagnosed can be identified and predicted, thereby determining multiple burst-prone locations and providing a model basis for subsequent pipe burst location and risk assessment.

[0050] The water hammer model is constructed based on parameters such as the pipe network topology, pipe diameter, wall thickness, and pipe material elastic modulus. The partial differential equations describing the transient flow in the pipe are transformed into a system of ordinary differential equations propagating along characteristic lines using the method of characteristics. These equations are then solved numerically on spatial and temporal grids. The system sets multiple assumed initial leakage points in the water hammer model and generates theoretical simulated pressure waveforms and theoretical pressure changes for each assumed leakage point through forward MOC simulation.

[0051] Step S406, wave velocity calculation. The propagation speed of transient pressure waves in the pipeline is the core parameter for positioning. The system first calculates the theoretical wave velocity based on classical theoretical formulas, and then introduces an adaptive wave velocity correction mechanism to address the complex characteristics of old pipelines such as aging pipe materials and internal scaling. The theoretical wave velocity is reversed by using the actual pressure wave propagation time between adjacent known sensor nodes to obtain a high-precision actual wave velocity.

[0052] Step S407: Determine the initial leak location based on pressure change information. A dual positioning verification mechanism is adopted: First, the arrival time difference of adjacent sensors obtained by the CUSUM algorithm and the corrected actual wave velocity are used to construct a set of positioning equations to calculate the physical coordinates based on the time difference; Second, the measured pressure change information (pressure drop, waveform slope) is matched with the theoretical pressure change information generated by MOC in step S405 using a similarity method (least squares method). When the difference between the measured pressure change and the simulated pressure information of a certain assumed leak point is the smallest, the point is weighted and fused with the coordinates calculated by TDOA to determine the accurate initial leak location.

[0053] Step S408 proceeds to the next diagnostic step. At this point, edge preprocessing and preliminary analysis of the pressure data are complete. Combined with the processing results of the aforementioned flow and acoustic data, the edge preprocessing stages for all three types of data are now ready, laying the foundation for subsequent multi-source fusion diagnostics.

[0054] After completing the edge preprocessing of the above three types of data, the cloud-based fusion diagnostic stage begins. The complete process of this stage is as follows: Figure 5 As shown.

[0055] Step S501, Data Upload. After performing multi-source time synchronization verification, the traffic, pressure, and acoustic feature data extracted from the edge are uploaded to the cloud diagnostic platform via the communication module. Since the edge has already completed dimensionality reduction and feature extraction of a large amount of high-frequency data, this greatly reduces communication bandwidth usage and improves the system's real-time response capability.

[0056] The aforementioned preprocessing and feature extraction processes are all completed by the edge processing module built into the three-in-one smart water meter. The purpose is to reduce communication bandwidth usage, improve the system's real-time response capability, and ensure the quality and effectiveness of the data uploaded to the cloud diagnostic platform.

[0057] After receiving edge feature data from multiple pipeline nodes, the aforementioned cloud-based diagnostic platform constructs a multi-source fusion analysis model oriented towards the pipeline network's operational status to quantitatively assess the leakage risk of the target pipeline segment. The data, after edge preprocessing and feature compression, is uploaded to the cloud-based diagnostic platform via a communication module, significantly reducing the upload pressure and system power consumption of the original high-frequency data while ensuring diagnostic effectiveness.

[0058] The aforementioned edge processing module employs an embedded microcontroller, runs a real-time operating system, and possesses a certain level of computing power and storage space. Specifically, the edge processing module is configured to execute one or more of the following processing procedures: Perform outlier removal, time alignment, and sliding window smoothing on the traffic data.

[0059] Perform bandpass filtering, noise reduction and enhancement, short-time Fourier transform (STFT), or wavelet decomposition on acoustic signals.

[0060] Perform pulsation decomposition, abrupt change detection, and time-domain normalization on the pressure time series. Perform multi-source time synchronization and data quality marking.

[0061] The preprocessing of the aforementioned acoustic signals includes at least one or more of the following: noise filtering, denoising enhancement, bandpass filtering, baseline correction, and signal normalization. Noise filtering is used to suppress environmental interference, and signal enhancement is used to highlight suspected leakage frequency bands. After preprocessing, the system extracts features from the acoustic information: statistical features (such as peak value, valley value, root mean square value, kurtosis factor, etc.) are extracted through time-domain analysis; spectral features (such as power spectral density, dominant frequency position, frequency band energy distribution, etc.) are extracted through frequency-domain analysis; and wavelet energy, wavelet entropy, and other features are extracted through time-frequency analysis, forming a feature vector for subsequent discrimination.

[0062] In one basic discrimination method, when the peak value exceeds a first preset threshold or the trough value is lower than a second preset threshold, it is determined that there is an abnormal acoustic event in the acoustic information to be diagnosed, thereby identifying a suspected pipe burst or leakage risk in the water supply network to be diagnosed. In other words, when the key amplitude characteristics exceed the normal operating statistical range, an acoustic anomaly marker is triggered. The threshold setting adopts an adaptive method: statistically analyzing the feature value distribution of the same time period (e.g., 2-4 AM) over a historical 7-day period, taking 3 times the mean as the upper threshold and 0.3 times the mean as the lower threshold.

[0063] Through the collaborative processing of signal preprocessing, feature extraction, and intelligent classification and recognition, this embodiment can effectively suppress interference caused by environmental noise, hydraulic fluctuations, and user water usage behavior, and significantly improve the accuracy, stability, and engineering practicality of identifying pipe bursts and hidden leaks.

[0064] Step S502, Multi-source data fusion: Based on the multi-source feature data that has undergone edge preprocessing and time synchronization, multi-source data fusion is performed. The feature vectors of the three types of data—flow, pressure, and acoustic data—are spatiotemporally aligned and jointly represented to form a unified fusion feature matrix, providing input for the subsequent diagnostic model. During the fusion process, the system integrates the response characteristics of the three types of data to the same hydraulic event, and enhances the signal strength of abnormal events through feature-level fusion or decision-level fusion strategies, effectively suppressing false alarms that may occur from a single sensor.

[0065] Step S503: Input the fused data into the leak diagnosis model (XGBoost, neural network). Through artificial intelligence model operation (CNN extracts spatial features and LSTM captures temporal dependencies), especially by using the time difference of arrival (TDOA) positioning model and combining it with an adaptive wave velocity correction algorithm to overcome reflection and attenuation interference in multi-branch pipe networks, high-precision positioning of the leak point is achieved (error can be controlled within 3-5 meters).

[0066] In step S504, the system then performs a risk score calculation.

[0067] The aforementioned risk assessment employs a multi-indicator weighted fusion method, integrating three indicators: macroscopic leakage index, local leakage probability, and hydraulic anomaly confidence level. The macroscopic leakage index is calculated based on flow balance deviation, reflecting the degree of leakage in different zones; the local leakage probability is calculated based on acoustic anomaly intensity, reflecting the multiple of acoustic anomaly frequency and amplitude relative to normal fluctuations; and the hydraulic anomaly confidence level is calculated based on the consistency between pressure transient response and water hammer model, obtained through the similarity between measured pressure waveforms and model simulation waveforms. The weights of each indicator can be dynamically adjusted according to the characteristics of the pipeline network to adapt to different application scenarios.

[0068] In step S505, the cloud-based diagnostic platform performs a multi-source fusion leakage risk assessment and outputs warnings. Based on the risk score from step S504 above, and combined with preset alarm thresholds, the cloud-based diagnostic platform performs the final multi-source fusion leakage risk assessment.

[0069] Specifically, the location and risk assessment process integrates diagnostic results from three levels: flow, acoustics, and pressure. First, based on the flow monitoring data of each node, a zoned flow balance analysis is performed according to the pipeline topology. For each DMA zone, a water balance equation is established, the zone's leakage is calculated, and the nighttime minimum flow method is used to select data from the nighttime low-water-use period for analysis. When the leakage is detected to exceed the preset threshold for several consecutive days, it is determined that there is a suspected leakage or pipe burst risk in the area, thus completing the large-scale risk section screening.

[0070] Further, the acoustic wave information to be diagnosed was retrieved from the built-in underwater acoustic monitoring sensors of each three-in-one smart water meter within the risk section. After preprocessing the acoustic signals, time-frequency analysis was performed using Short-Time Fourier Transform (STFT) or Wavelet Transform to extract the characteristic frequency band of the leakage signal. By calculating the energy integral of the signal at each monitoring point within the leakage characteristic frequency band, the node with the strongest energy was identified as the most likely location of the leakage-sensitive node. Simultaneously, cross-correlation analysis was used to calculate the time delay between signals from different monitoring points to preliminarily estimate the direction of the leakage source.

[0071] Simultaneously, a water hammer model based on the hydraulic transient mechanism was introduced to simulate and analyze the pressure propagation behavior of the pipeline network under disturbance conditions, identifying multiple potentially explosive locations in the water supply network to be diagnosed. Combined with transient pressure change data collected by the pressure sensing unit built into the water meter, the Time Difference of Arrival (TDOA) method was used for precise location. For a set of monitoring points, a system of equations was established with the leak point coordinates as unknowns. The time difference was obtained through accurate acquisition of the pressure wave's initial arrival time, and the wave velocity was calculated based on pipe material parameters and corrected through on-site verification. Numerical solutions were used to obtain estimated leak point coordinates. The location error depends on the accuracy of time measurement and wave velocity. By optimizing the monitoring point layout and signal processing algorithm, the error can be controlled within an acceptable engineering range.

[0072] By weighted and fused diagnostic results from flow screening, acoustic localization, and pressure verification, the burst pipe space is converged under multiple constraints to ultimately determine the burst location in the water supply network to be diagnosed. The burst intensity can be estimated based on pressure drop, and the severity of the burst can be judged based on flow rate changes. This embodiment integrates three types of sensors—flow rate, noise, and pressure—and combines transient hydraulic analysis using a water hammer model to achieve a layered diagnostic mechanism from large-scale anomaly screening to small-scale precise localization, effectively reducing the risk of misjudgment and improving the accuracy and reliability of burst pipe location and early warning. Furthermore, the diagnostic process in this embodiment can be executed cyclically according to a set time window to continuously track the anomaly evolution process, determine the occurrence time, duration, and development trend of the burst pipe event, and provide quantitative basis for operation and maintenance decisions and emergency repair scheduling. This method has good adaptability and engineering feasibility for water supply networks with different pipe diameters, materials, and topological complexities.

[0073] Step S506, generate an alarm message. When the above risk score exceeds the threshold, the system generates a detailed leakage warning message. The alarm message includes basic event alarms and integrates the in-depth diagnosis results output by the AI model, including but not limited to: coordinates of suspected pipe bursts / water leakage points accurate to longitude and latitude, estimated leakage water volume calculated based on pressure drop and flow rate changes (accurate to L / s), leakage risk grading evaluated from multiple dimensions, and an automated control plan automatically generated by the intelligent decision support system (by remotely controlling the opening of the upstream electric valve or reducing the power of the pump station frequency converter through the linkage of the remote control module to curb the expansion of leakage before maintenance).

[0074] Step S507, multi-channel push. To ensure the real-time nature of event response and data security, the system uses 4G / 5G or NB-IoT communication networks and combines encryption transmission technology to push structured alarm instructions and control plans to terminals at all levels in milliseconds.

[0075] Step S508, display the above alarm message on the visual dashboard. The system deeply integrates the Geographic Information System (GIS) and dynamic digital twin technology. The visual dashboard not only provides traditional data reports but also accurately marks abnormal shock sources and pressure drop areas in the form of heat map rendering, dynamic red dot flashing, and ripple animation on the three-dimensional pipe network topology map. At the same time, it supports multi-perspective linkage display of multi-source data, such as real-time reconstruction of the holographic state of the pipe network, such as transient pressure waveform playback and abnormal flow trend comparison, providing intuitive and accurate decision-making support for water supply management personnel.

[0076] Step S509, work order dispatch and intelligent scheduling decision. When the cloud diagnosis platform generates an alarm message, the system automatically enters the work order management process. Operation and maintenance management personnel can dispatch work orders at the intelligent scheduling center based on the decision-making support provided by the visual dashboard, or the system can automatically trigger the creation of work orders according to preset rules. The core of this step is to convert leakage events into executable task instructions, laying a foundation for subsequent resource scheduling and on-site handling.

[0077] Step S510, optimized resource allocation and work order push. The system has a built-in resource allocation optimization algorithm. According to the work order requirements generated in Step S511, combined with multi-dimensional information such as the severity of pipe bursts, the real-time geographical location of maintenance personnel, current workload, and skill matching degree, it realizes the optimal automated distribution of work orders. The optimized work orders are pushed to the mobile terminals of corresponding maintenance personnel in milliseconds through 4G / 5G or NB-IoT networks. The information received by the terminals includes coordinates of suspected pipe burst points accurate to longitude and latitude, the pipe network topology map, and relevant diagnostic data, providing navigation guidance for precise maintenance.

[0078] Step S511: On-site handling and feedback submission. Maintenance personnel arrive at the site based on the precise coordinates and pipeline information received via their mobile devices. Leveraging the high-precision positioning achieved through prior algorithms, the investigation time for traditional manual leak listening and large-scale excavation is significantly reduced. After completing on-site handling, maintenance personnel upload the survey data via their mobile devices, including verification tags such as the actual leak location, photos of pipe damage, and the actual leakage amount, and submit the handling results. This closed-loop feedback provides accurate labeled data for subsequent model optimization.

[0079] Step S512: Perform continuous learning and adaptive model updates. After the closed loop of handling real business scenarios is completed, the system enters the closed-loop data feedback stage. Based on the accurate labeled data from the aforementioned manual verification and on-site feedback, it will be written into the cloud historical database in real time as the highest quality supervision signal, driving subsequent incremental training of AI models, adaptive evolution of parameters, and dynamic calibration of edge computing thresholds.

[0080] Step S513: Historical Data Import. All verified field feedback data and corresponding raw monitoring data are imported into the cloud-based historical database in a unified format. The imported data includes timestamps, event tags, sensor waveforms, positioning results, maintenance records, etc., forming a structured sample library for subsequent analysis and training.

[0081] Step S514: Sample Labeling and Manual Verification. The system periodically extracts typical samples from the historical database and performs sample labeling and verification through a combination of semi-automatic and manual intervention. Labeling content includes event type (pipe burst, leak, normal water use, etc.), severity level, and actual leak location coordinates, ensuring sample diversity and representativeness. The manual verification process further eliminates mislabeled samples, improving the quality of the training data.

[0082] Step S515: Incremental Model Training. Based on the high-quality labeled samples generated in step S516, the system performs incremental model training. Specifically, algorithms such as XGBoost incremental learning and neural network fine-tuning can be used to absorb new sample features while retaining the original model knowledge, optimizing classification boundaries and regression accuracy. The incremental training process fully utilizes new data, avoids catastrophic forgetting, and allows the model to continuously adapt to trends such as network aging and environmental changes.

[0083] Step S516: Model parameter update and distribution. After incremental training is completed, the updated model parameters (such as CNN convolutional kernel weights, LSTM biases, decision tree structure, etc.) are deployed to the cloud diagnostic platform and can be distributed to the edge processing module of the three-in-one smart water meter according to the strategy. After obtaining the latest model, the edge device can achieve more accurate anomaly identification locally, forming an intelligent closed loop of collaborative evolution between the cloud and the edge. At this point, a complete leakage diagnosis, handling, and learning cycle ends, and the system enters the next monitoring cycle.

[0084] The cloud-based diagnostic platform is configured to collect historical monitoring data and actual maintenance results to form a labeled sample library. Typical samples are periodically extracted from the database and labeled in conjunction with the results confirmed by manual inspections to ensure sample diversity and representativeness. Alarm thresholds and feature discrimination boundaries are dynamically updated. A sliding window method or adaptive algorithm is used to adjust threshold parameters based on the statistical distribution of recent data, recalculate the statistical distribution (mean, standard deviation, quantiles) of each feature value within the window, and update threshold parameters to adapt to the aging trend of the pipeline network. The model is periodically retrained based on new samples, and an incremental learning algorithm is used to fine-tune the model while retaining historical knowledge.

[0085] The analysis of the diagnostic information is implemented using artificial intelligence algorithms. Specifically, it can adopt high-dimensional classification models (such as gradient boosting trees, support vector machines, neural networks, etc.) based on supervised learning, semi-supervised learning multimodal learning, or self-supervised learning (for automatic extraction of emergent features from unlabeled data). The deep feature vector extracted from the edge is used as input, and the abnormal probability or risk score is output. At the same time, a threshold discrimination mechanism is used for double verification to reduce the false alarm rate.

[0086] Furthermore, it supports a distributed edge-cloud collaborative architecture with the edge system. (For example, CNN convolutional kernel weights, LSTM biases, etc.) The updated model parameters of the global large model can be distributed to the edge processing module of the three-in-one smart water meter according to the strategy, enabling the edge to have the latest anomaly recognition capabilities from the cloud. This architecture mechanism, similar to federated learning, facilitates lightweight model collaborative evolution, protects the privacy of the underlying high-frequency raw data, and adapts to self-learning. This continuously improves the intelligence level of the entire pipeline monitoring system.

[0087] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0088] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention. All technical features in this embodiment can be freely combined according to actual needs.

[0089] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart diagnostic system for water supply network leakage based on multi-parameter sensing, characterized in that: It includes a three-in-one smart water meter terminal, a cloud-based diagnostic platform, and a pipeline management terminal; Among them, the three-in-one smart water meter terminal integrates a flow meter, pressure sensor, water sound monitoring sensor, edge processing module, communication module and power module for multi-parameter synchronous acquisition and edge preprocessing; The cloud-based diagnostic platform is used for multi-source fusion analysis, leakage risk assessment, abnormal water use identification, alarm management, and visualization. The pipeline management terminal is used to provide maintenance personnel with a real-time monitoring interface, alarm information, and pipeline health assessment results.

2. The intelligent diagnostic system for water supply network leakage based on multi-parameter sensing according to claim 1, characterized in that: The flow meter, pressure sensor, and water acoustic monitoring sensor in the three-in-one smart water meter terminal are all built into the same water meter housing, enabling synchronous acquisition of multiple parameters at the same location and time reference. The water acoustic monitoring sensor is a hydrophone, the pressure sensor is a high-frequency response sensor, and the flow meter is an electromagnetic or ultrasonic flow meter.

3. The intelligent diagnostic system for water supply network leakage based on multi-parameter sensing according to claim 1, characterized in that: The edge processing module is configured to perform at least one of the following processes: outlier removal, time alignment, and sliding window smoothing on traffic data; bandpass filtering, noise reduction and enhancement, short-time Fourier transform, or wavelet decomposition on acoustic signals; pulsation decomposition, abrupt change detection, and time-domain normalization on pressure time series; and multi-source time synchronization and data quality marking. The communication module adopts a tri-mode converged communication architecture of NB-IoT, 4G, and Bluetooth.

4. The intelligent diagnostic system for water supply network leakage based on multi-parameter sensing according to claim 1, characterized in that, The cloud-based diagnostic platform is also configured to: collect historical monitoring data and actual maintenance results to construct a labeled sample library; The alarm threshold and feature discrimination boundary are dynamically updated; the model is periodically retrained based on new samples, and the updated model parameters are sent to the edge processing module.

5. The intelligent diagnostic system for water supply network leakage based on multi-parameter sensing according to claim 1, characterized in that, The cloud-based diagnostic platform also integrates geographic information systems and digital twin technology to dynamically mark abnormal locations, areas of sudden pressure drops, and leakage heat maps on a three-dimensional pipeline network topology map, supporting multi-view linkage display and intelligent scheduling decisions.

6. A method for intelligent diagnosis of leakage in water supply networks based on multi-parameter sensing, characterized in that, The system according to any one of claims 1 to 5 comprises the following steps: Step 1: Simultaneously collect network pressure data, instantaneous flow data, and pipeline vibration and noise data using a three-in-one smart water meter installed at the nodes of the water supply network; Step 2: Perform edge preprocessing and feature construction on the collected multi-parameter data, including noise reduction filtering, outlier removal, temporal resampling, and multi-scale feature extraction; Step 3: Upload the preprocessed feature data to the cloud diagnostic platform. Based on the coupling relationship between pressure fluctuation features, flow continuity features and acoustic leakage features, establish a multi-source fusion diagnostic model to identify whether the current operating condition is normal water supply, pipeline leakage, or abnormal water use. Step 4: When the identification result meets the preset anomaly criteria, generate the leakage risk level, anomaly type label and alarm information, and push them to the pipeline management terminal; Step 5: Based on historical operational data and manual verification results, periodically update the parameters of the diagnostic model to achieve continuous learning and adaptive optimization of the model.

7. The intelligent diagnostic method for water supply network leakage based on multi-parameter sensing according to claim 6, characterized in that, In step 2, data preprocessing and feature construction further include: Perform outlier removal, time alignment, and sliding window smoothing on the traffic data; Perform bandpass filtering, noise reduction and enhancement, short-time Fourier transform or wavelet decomposition on acoustic signals to extract time-domain, frequency-domain and time-frequency-domain features; Empirical mode decomposition, accumulation, and abrupt change detection, along with time-domain normalization, are performed on the stress time series. Perform multi-source time synchronization and data quality marking on three types of data.

8. The intelligent diagnostic method for water supply network leakage based on multi-parameter sensing according to claim 6, characterized in that: In step 3, the multi-source fusion diagnostic model uses one or more artificial intelligence algorithms, such as XGBoost, support vector machine, convolutional neural network or long short-term memory network, as input to the fusion feature matrix, outputs the abnormal probability or risk score, and performs double verification by combining a threshold discrimination mechanism.

9. The intelligent diagnostic method for water supply network leakage based on multi-parameter sensing according to claim 6, characterized in that: In step 4, the leakage risk level is calculated using a multi-indicator weighted fusion method. The multi-indicators include: a macroscopic leakage index calculated based on flow balance deviation, a local leakage probability calculated based on acoustic anomaly intensity, and a hydraulic anomaly confidence level calculated based on the consistency between pressure transient response and water hammer model.

10. The intelligent diagnostic method for water supply network leakage based on multi-parameter sensing according to claim 6, characterized in that: In step 5, the model update adopts an incremental learning algorithm, which periodically retrains the model based on the labeled samples after manual verification, and sends the updated model parameters to the edge processing module to achieve collaborative evolution between the cloud and the edge.

Images