A monitoring method and device for water supply pipe network leakage

Abstract

The application provides a kind of monitoring method and equipment for water supply network leakage, applied to data processing technical field.The application is aimed at water supply network leakage monitoring, first collects flow, pressure sensing data and pipe network topology information, reads basic parameters to preprocess original data, constructs hydraulic model mask to screen suspected leakage area, generates monitoring data-water model pair through two-way calibration and dynamic load compensation and outputs quality control information.Based on model pair, deviation index and multi-dimensional features are extracted to construct feature set, which is input into graph neural network model to identify leakage location, rate and type.Then, through multiple rounds of reasoning calculation, point-level uncertainty is calculated, sensitive ring belt is aggregated to obtain area-level score and quality control determination is completed.Finally, multi-source data superposition visualization is realized on terminal platform, and structured report containing quality control results and inspection suggestions is output.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and equipment for monitoring leakage in water supply networks. Background Technology

[0002] As a core component of urban infrastructure, water supply networks play a crucial role in ensuring water supply for residents and industrial production. Their operational stability directly impacts the normal functioning of the city and the well-being of its people. However, over long-term use, water supply network leakage is a widespread problem due to multiple factors such as pipe aging, geological subsidence, construction disturbance, and corrosion. This not only results in significant water waste but may also lead to derivative risks such as road collapses and water pollution, increasing network maintenance costs and safety management pressures.

[0003] Traditional monitoring methods rely heavily on manual inspections and leak detection instruments, which suffer from low efficiency, high subjectivity, and insufficient accuracy in leak location, making them unsuitable for the comprehensive monitoring needs of complex pipe networks. Some monitoring technologies based on single sensor data are susceptible to interference from water load fluctuations and equipment errors because they do not fully integrate the characteristics of the pipe network topology and hydraulic conduction laws, resulting in low accuracy in leak identification. Existing hydraulic models are not well-suited to actual monitoring data and lack an effective two-way calibration mechanism, making it difficult to accurately reflect the dynamic operating status of the pipe network, thus affecting the reliability of leak feature extraction and identification. At the same time, the lack of systematic integration and intuitive display of multi-source data such as monitoring data, model results, and quality control information makes it difficult for operation and maintenance personnel to quickly obtain key decision-making information, affecting the response efficiency of leak control.

[0004] With the rapid development of data processing technology and artificial intelligence, monitoring solutions based on multi-source data fusion, model simulation, and intelligent algorithms have become an industry trend. However, existing technologies still suffer from problems in practical applications, such as non-standard data preprocessing processes, lack of targeted feature extraction, weak integration of models with business logic, and imperfect quality control systems. These issues make it difficult for the accuracy, stability, and practicality of leakage monitoring to meet actual operation and maintenance needs.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] According to one aspect of this application, a method for monitoring leakage in a water supply network is provided, comprising: collecting flow monitoring data, pressure sensing data, and network topology information of the water supply network; reading basic network parameters including monitoring point locations, data acquisition frequency, and pipe diameter / material; using the basic network data as support, first performing outlier removal and smoothing preprocessing on the raw flow / pressure data; constructing a hydraulic model mask based on the network topology; performing preliminary screening of suspected leakage areas through regional hydraulic balance analysis; then achieving bidirectional calibration of the monitoring data and hydraulic model through flow-pressure coupling characteristics; introducing time-series data trend analysis to compensate for dynamic load changes in the network; generating a monitoring data-hydraulic model pair; and simultaneously outputting calibration quality control curves and log information; based on the monitoring data-hydraulic model pair, extracting flow deviation, pressure deviation, and nodal hydraulic residual indices between real and simulated data; and mining the time-series change trend of deviations. Based on spatial distribution patterns and multi-parameter coupling correlation characteristics, a leakage identification feature set is constructed. This feature set is input into a leakage identification model based on a graph neural network to generate preliminary identification results of leakage location, leakage rate, and leakage type. Based on the confidence output of the identification model, multiple inferences are performed on the same monitoring period to obtain multiple sets of confidence data. The uncertainty at the monitoring point level is calculated using the standard deviation method or the coefficient of variation method. A sensitive ring zone of the pipeline network is constructed by combining a mask of suspected leakage areas. The uncertainty at the monitoring point level in the ring zone area is aggregated to obtain a regional score. A threshold is set based on the historical leakage verification set to complete the quality control conclusion. The pipeline network basic data, monitoring data-hydraulic model pairs, leakage identification results, monitoring point level uncertainty heatmap, and regional scores are integrated and displayed in a multi-dimensional overlay visualization on the terminal monitoring platform. At the same time, a structured report including data calibration logs, scores of each region, quality control judgment results, and priority inspection suggestions is output.

[0008] Another aspect of this application is a monitoring device for leakage in a water supply network, used to execute executable instructions to perform the above-described monitoring method for leakage in a water supply network.

[0009] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described method for monitoring leakage in a water supply network by executing the executable instructions.

[0010] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described method for monitoring leakage in a water supply network.

[0011] The beneficial technical effects of this application are as follows: Through a closed-loop design encompassing "data acquisition - preprocessing calibration - feature extraction and identification - quality control judgment - output," accurate leakage monitoring is achieved. First, flow and pressure sensor data and pipeline topology information are collected. After reading basic parameters, the data is preprocessed, and a hydraulic model mask is constructed to initially screen suspected areas. Through bidirectional calibration and dynamic load compensation, monitoring data-hydraulic model pairs are generated. Then, based on the model pairs, deviation indicators and multi-dimensional features are extracted, and a feature set input graph neural network model is constructed to identify the location, rate, and type of leakage. Subsequently, through multiple rounds of inference calculations at the monitoring point level, sensitive ring zones are aggregated to achieve regional-level scores, and quality control judgment is completed. Finally, multi-source data overlay visualization is achieved on the terminal platform, outputting a structured report containing quality control results and inspection suggestions, ensuring standardization and traceability throughout the entire process.

[0012] By employing bidirectional calibration of monitoring data and hydraulic models, multi-dimensional feature fusion, and graph neural network recognition, environmental interference and equipment errors are effectively reduced, significantly improving the accuracy of leak location and the identification rate of type / rate. Multi-dimensional visualization and structured reports intuitively present core information, and combined with priority inspection suggestions, leak investigation time is greatly shortened, reducing manual maintenance costs. A two-tiered uncertainty assessment mechanism at the "monitoring point level - regional level" is constructed to achieve full-process quality control of identification results, improving the reliability and traceability of monitoring data and identification results. It is compatible with different pipe network zones, pipe diameters, materials, and water load scenarios, and adapts to the dynamic operating status of complex pipe networks through dynamic load compensation and flexible calculation rules.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0014] Figure 1 A flowchart of a method for monitoring leakage in a water supply network according to an embodiment of this application is shown; Figure 2 A schematic diagram of a monitoring device for leakage in a water supply network provided in one embodiment of this application is shown. Detailed Implementation

[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0016] In one implementation, Figure 1 A schematic flowchart of a method for monitoring leakage in a water supply network according to an embodiment of this application is shown.

[0017] S101 collects flow monitoring data, pressure sensing data, and pipeline topology information of the water supply network, and reads basic pipeline parameters including monitoring point location, data acquisition frequency, and pipeline diameter / material.

[0018] In one implementation, corresponding flow sensing devices are used to collect flow data in real time or at set intervals at different monitoring points such as the main pipes, branch pipes, and key nodes of the water supply network. During the collection process, the monitoring point code and collection timestamp corresponding to the data are recorded simultaneously to ensure that the flow data is accurately correlated with spatial location and time dimension. Instantaneous flow and cumulative flow data of each monitoring point in the network are continuously collected at a preset collection frequency. The collected flow data is initially formatted to ensure the uniformity of data units and data recording format, while the original collected data is retained to avoid data loss.

[0019] At key pressure monitoring locations such as pressure monitoring points in the pipeline network, pipeline ends, and pump station outlets, pressure sensing devices collect pipeline water pressure data. During the data collection process, the timing of flow data collection must be matched to achieve time synchronization between pressure and flow data. Simultaneously, the operating status data of the pressure sensing devices is recorded to ensure the validity of the collected data. Based on the set collection frequency, static and dynamic pressure data from each pressure monitoring point are collected synchronously. The collected data is verified in real time, and invalid pulse data generated by temporary equipment failures are eliminated to ensure the continuity and accuracy of the pressure data.

[0020] Comprehensive data collection of physical and attribute topology information for the water supply network was conducted. Physical topology information included pipeline routing, node connections, and the installation locations of ancillary facilities such as pump stations, valves, and water meters. Attribute topology information included network zones, water supply ranges, and connection methods for each zone. Standardized topology descriptions were used during data collection to ensure the information could be directly used for subsequent hydraulic model construction. Based on the actual network layout, the start and end nodes of each pipeline segment and the connection logic between pipelines were collected and analyzed. The relationships between pump stations, valves, and other ancillary facilities and pipelines were recorded simultaneously, forming a structured network topology information set.

[0021] Based on the collected pipeline topology information and sensor deployment information, three types of core pipeline basic parameters were extracted and accurately read. Each parameter was correlated with the actual pipeline scenario and monitoring data, specifically as follows: From the pipeline topology information and sensor deployment files, the geographical coordinates, pipeline topology node number, and pipeline zone to which each flow and pressure monitoring point belongs were read to establish a one-to-one correspondence between the monitoring point location and the monitoring data. Based on the sensor installation and registration information, the latitude and longitude coordinates and the corresponding pipeline topology node code of each monitoring point were extracted. The location parameters were then bound to the device codes of the monitoring points to achieve spatial positioning of the monitoring data.

[0022] From the operational configuration information of the sensing devices, frequency parameters such as the acquisition interval, data upload frequency, and acquisition cycle of flow and pressure data at each monitoring point are read to clarify the temporal dimension characteristics of the data, providing a basis for subsequent time-series data processing. Preset acquisition parameters for each sensing device are extracted to determine whether the acquisition frequency at different monitoring points is on the minute, hour, or other cycle levels. A unified acquisition frequency parameter table is then compiled and matched with the raw data of the corresponding monitoring points.

[0023] From the pipeline network topology information and pipeline engineering archives, parameters such as nominal pipe diameter, pipe material, and pipe wall thickness of each pipe segment are retrieved. These parameters provide the core hydraulic calculation foundation for subsequent hydraulic model mask construction and regional hydraulic balance analysis. According to the coding of each pipe segment in the pipeline network, the corresponding pipe diameter specification and material type are retrieved one by one. These parameters are then bound to the pipe segments in the pipeline network topology to form a complete set of fundamental parameters for pipeline network hydraulic calculations.

[0024] S102, supported by pipeline network basic data, first performs outlier removal and smoothing preprocessing on the raw flow / pressure data. Based on the pipeline network topology, a hydraulic model mask is constructed. The suspected leakage areas are initially screened through regional hydraulic balance analysis. Then, the bidirectional calibration of monitoring data and hydraulic model is achieved through flow-pressure coupling characteristics. Time series data trend analysis is introduced to compensate for dynamic load changes in the pipeline network, generating monitoring data-hydraulic model pairs, and simultaneously outputting calibration quality control curves and log information.

[0025] In one implementation, based on the hydraulic analysis requirements for water supply network leakage monitoring, the preprocessing specifications for multi-source sensor data, and the spatial analysis logic of the network topology, the collected raw flow / pressure data undergoes multi-dimensional preprocessing, including outlier removal, data smoothing, and time-series completion, generating standardized flow datasets, standardized pressure datasets, and time-series correlated sensor data information that can be used for hydraulic calculations. Based on the reasonable data range for hydraulic analysis, the normal operating thresholds of the sensing equipment, and the spatial flow / pressure transmission patterns of the network topology, abnormal extreme values ​​and invalid pulse data caused by equipment failure, signal interference, or sudden operating conditions are identified and removed from the raw data, retaining valid data that conforms to the actual hydraulic characteristics of the network. Reasonable upper and lower limit thresholds for flow and pressure data are set, and the data rationality is verified through the hydraulic correlation logic of adjacent monitoring points in the network. Discrete data exceeding the thresholds and without reasonable hydraulic causes are marked and removed, and the remaining data are labeled with the reasons for removal and the basis for verification.

[0026] A smoothing method adapted to the time-series characteristics of pipeline hydraulic data is employed to smooth the flow and pressure data after outlier removal, reducing random noise in the data, restoring the true trend of pipeline hydraulic parameters, and ensuring data continuity and stability. Based on the time resolution of the pipeline hydraulic data, a corresponding data smoothing algorithm is selected to perform sliding window processing on the time-series flow and pressure data, weakening short-term, irregular data fluctuations and ensuring that the processed data curves closely match the actual flow and pressure changes in the pipeline network.

[0027] To address the issue of missing time-series data due to equipment offlineness and signal interruption during preprocessing, interpolation was employed to fill in the missing data, taking into account the spatial hydraulic correlation of the pipeline network topology and the changing trends of historical data from the same period, thus ensuring the integrity of the data in the time dimension. Based on the time span of the missing data, an appropriate interpolation method was selected, referencing the changing patterns of data from adjacent monitoring points and the historical time-series characteristics of the monitoring point, to accurately fill in the missing flow and pressure data. All filled data were labeled with the filling method and reference criteria.

[0028] The preprocessed flow and pressure data are standardized according to a unified format, unit, and timestamp to generate standardized flow datasets and standardized pressure datasets respectively. At the same time, based on the pipeline network topology information, the spatial location and time dimension correlation of flow and pressure data at each monitoring point are established to generate time-series correlated sensor data information, realizing the spatiotemporal binding of sensor data and pipeline network topology.

[0029] Based on the hydraulic modeling specifications for water supply networks, standardized flow datasets, standardized pressure datasets, and time-series correlated sensor data are structured. Spatial mapping between monitoring points and hydraulic model nodes is achieved through network topology matching, generating a network hydraulic model mask, a regional hydraulic balance calculation template, and a flow-pressure coupling analysis framework. Using the network topology information as a benchmark, the physical monitoring points corresponding to the standardized sensor data are matched one-to-one with virtual nodes and pipe segments in the hydraulic model, establishing a spatial mapping relationship between physical locations and the digital model, clarifying the corresponding location and hydraulic correlation of each monitoring point's data in the hydraulic model. Topological information such as the geographical coordinates, network zones, and pipe segments of the monitoring points is extracted and matched with the attributes of virtual nodes in the hydraulic model to generate a spatial mapping lookup table, ensuring that the flow and pressure data of each physical monitoring point can be accurately mapped to the specified virtual node in the hydraulic model.

[0030] Based on the spatial mapping results, and combined with information such as pipeline routing, node connection relationships, and water supply range of the pipeline network topology, a hydraulic model mask is constructed to delineate the effective area for hydraulic calculations and shield redundant areas in the model that are irrelevant to actual leakage monitoring, thereby improving the efficiency and accuracy of subsequent hydraulic analysis. According to the actual pipeline network range for leakage monitoring, the hydraulic model is regionalized, retaining the model areas where main pipes, branch pipes, and key monitoring nodes are located, while shielding edge areas in the model where no sensors are deployed or where there is no monitoring requirement, thus forming a hydraulic model mask that closely matches actual monitoring needs.

[0031] Assume a city's water supply network comprises two main areas: the central urban area (primary water supply network) and the surrounding suburbs (secondary water supply network). The primary network consists mainly of main trunk lines with diameters ranging from 800mm to 1200mm and made of ductile iron. The secondary network consists mainly of branch lines with diameters ranging from 200mm to 500mm and made of PE pipes. Based on the network's zoning topology information, a hydraulic balance calculation template is generated as follows: The hydraulic balance calculation for the primary water supply network focuses on the flow distribution rate at main pipeline nodes, the difference between total inflow and outflow in the network, and the compliance of pressure at key nodes. The pressure at key nodes must be stable within the range of 0.35 MPa to 0.45 MPa. The calculation adheres to the law of flow conservation, where the total inflow to each node equals the sum of the total outflow and leakage. It also follows pressure balance logic, where the sum of friction head loss and local head loss is equivalent to the pressure difference at each node. The calculation requires importing three types of data: a standardized flow dataset, a standardized pressure dataset, and basic network parameters. The standardized flow dataset covers the instantaneous and cumulative flow at each monitoring point on the main pipeline. The standardized pressure dataset includes the pump station outlet pressure and the pressure at key nodes on the main pipeline. The basic network parameters include pipe diameter, pipe length, and pipe friction coefficient. After calculation, three core results are output: a flow conservation verification value, the percentage of nodes meeting pressure standards, and a preliminary location of suspected leakage areas in the primary network. The flow conservation verification uses a difference of no more than 5% as the criterion for balance determination. The preliminary location of the suspected leakage area is defined as a 3-kilometer radius around the flow imbalance node.

[0032] The hydraulic balance calculation for a secondary water supply network focuses on verifying the uniformity of flow distribution in branch pipes, the range of pressure fluctuations at nodes within a zone, and the matching degree between water supply and water consumption in each community. It uses the law of conservation of flow to determine the rationality of flow distribution at branch pipe nodes, and calculates pressure changes using formulas related to local head loss. Local head loss includes losses caused by sudden narrowing of pipes and valve throttling. It also compares the total water supply of the zone with the statistical values ​​of user water consumption. The calculation requires importing three types of data: a standardized flow dataset, a standardized pressure dataset, and basic network parameters. The standardized flow dataset includes the flow rate at monitoring points in branch pipes and the total water meter flow rate in the community. The standardized pressure dataset includes the starting pressure of branch pipes and the inlet pressure of the community. The basic network parameters include pipe diameter, pipe material, and local resistance coefficient. After calculation, four core results are output: flow distribution uniformity coefficient, pressure fluctuation amplitude, supply and water consumption matching difference, and a list of suspected leakage zones in the secondary network. A flow distribution uniformity coefficient of not less than 0.85 is considered acceptable, and a pressure fluctuation amplitude of not more than 0.05 MPa is also acceptable.

[0033] The city's water supply network exhibits well-defined hydraulic characteristics, with a negative correlation between flow rate and pressure; pressure decreases linearly as flow rate increases. Standardized sensor data possesses fixed spatiotemporal correlation characteristics, with flow and pressure data collected synchronously every five minutes at the same monitoring point. Based on this, a flow-pressure coupling analysis framework is constructed. Using flow and pressure data from the same monitoring point at the same time stamp as the core, a correlation model between flow and pressure changes is established. This model is then used to infer the network's hydraulic characteristic parameters, verifying the fit between actual data and the theoretical model. The framework employs the hydraulic calculation formula H=KQ², where H represents head loss, K represents the network resistance coefficient, and Q represents flow rate. A one-hour time-series data sliding window is used to calculate the theoretical pressure values ​​corresponding to different flow ranges, which are then compared with the actual monitored pressure values.

[0034] In the primary pipeline network, for every 100 m³ / h increase in flow rate in the main pipe, the average pressure at nodes along the pipeline decreases by 0.02 MPa. In the secondary pipeline network, for every 50 m³ / h increase in flow rate in branch pipes, the average pressure at the community inlet decreases by 0.015 MPa. Flow rates exceeding these limits are considered abnormal. If the node pressure suddenly drops by more than 0.05 MPa (without valve adjustment or other human intervention), the corresponding flow rate should increase by more than 30%. If the flow rate does not change significantly or changes in the opposite direction, it is considered a suspected leak due to pressure anomaly. Theoretical flow-pressure fit (≥90% is considered acceptable), the number of abnormal correlations (≤2 times within a single window), and the flow response compliance rate during pressure anomalies are used to perform coordinated analysis of the pipeline network's hydraulic parameters. For example, if the flow rate at a monitoring point on a main pipe increases by 80 m³ / h within one hour, the pressure should decrease by 0.016 MPa according to the rule. If the actual pressure decreases by 0.03 MPa without any other human intervention, the framework determines that there is an abnormal flow-pressure coupling in that area and marks it as a suspected leak area.

[0035] Based on the analysis logic of water supply network leakage screening, regional hydraulic calculations, preliminary screening of suspected areas, and coupling feature calibration are performed on the network hydraulic model mask, regional hydraulic balance calculation template, and flow-pressure coupling analysis framework, respectively. This completes the location of suspected leakage areas and bidirectional matching of monitoring data and hydraulic model, compensating for deviations caused by dynamic load changes in the network, and generating monitoring data-hydraulic model pairs. Using the hydraulic model mask as the calculation boundary, standardized flow and pressure datasets are input into the hydraulic model to conduct full-area hydraulic simulation calculations. The hydraulic consistency between the model simulation data and the actual monitoring data is compared to verify the accuracy and adaptability of the hydraulic model. Within the defined hydraulic model mask area, hydraulic simulation calculations are run to obtain the simulated flow and pressure data of each virtual node. These are then compared one by one with the actual monitoring data of the corresponding physical monitoring points to analyze the deviation values ​​and verify the hydraulic matching degree between the model and the actual network.

[0036] Standardized sensor data from each zone is input into the corresponding regional hydraulic balance calculation template to conduct zoned hydraulic balance verification. Based on the hydraulic laws of flow conservation and pressure balance, areas of the pipe network with imbalanced hydraulic parameters are identified, completing the initial screening and location of suspected leakage areas. Flow and pressure balance calculations are performed on each pipe network zone. If a zone shows a non-conservation of inflow and outflow or a pressure distribution that deviates significantly from the normal pattern, the zone is marked as a suspected leakage area, and the pipe network scope and core abnormal indicators of the suspected area are clearly defined.

[0037] For the flow-pressure coupling characteristics of suspected leakage areas and the entire pipe network, a flow-pressure coupling analysis framework was used to calibrate the coupling characteristics. The core parameters of the hydraulic model were adjusted to ensure a high degree of match between the simulated flow-pressure coupling characteristics and actual monitoring data, achieving bidirectional calibration between the monitoring data and the hydraulic model. Based on the flow-pressure coupling variation characteristics of the actual monitoring data, core parameters in the hydraulic model, such as the pipe friction coefficient and local resistance coefficient, were adjusted to ensure that the simulated flow-pressure correlation is consistent with reality, completing the bidirectional adaptation between the model and the monitoring data.

[0038] By introducing time-series data trend analysis methods, the water load variation patterns of the pipeline network at different times are analyzed, and dynamic load compensation is performed on the simulation results of the hydraulic model to eliminate hydraulic deviations caused by dynamic load changes such as peak and off-peak water use. Finally, the actual monitoring data after two-way calibration and dynamic load compensation are bound to the hydraulic simulation model to generate a precisely correlated monitoring data-hydraulic model pair.

[0039] The calibration accuracy and data consistency of the monitoring data-hydraulic model pair are verified. Combined with the quality control evaluation standards for pipeline network monitoring, calibration quality control curves and data calibration log information are generated, achieving full-process recording and quality control evaluation of the monitoring data and hydraulic model calibration process. Accuracy evaluation indicators and thresholds for the calibration of the water supply network hydraulic model are set. The deviation values ​​between the actual monitoring data and the model simulation data in the monitoring data-hydraulic model pair are quantitatively analyzed to verify whether the matching accuracy meets the hydraulic analysis requirements for leakage monitoring. Calibration accuracy evaluation indicators such as flow deviation rate and pressure deviation value are set. The indicator values ​​of the actual data and simulation data at each monitoring point are calculated. If the indicator values ​​are within the preset threshold range, the calibration accuracy is determined to meet the requirements; otherwise, the model calibration needs to be repeated.

[0040] Data consistency verification of the monitoring data-hydraulic model pair is performed from both temporal and spatial dimensions. This verifies whether the temporal variation trends and spatial distribution patterns of the actual monitoring data are consistent with the model simulation data, eliminating data misalignment and inconsistencies caused by calibration operations. The verification also checks whether the temporal variation trends of the actual data and simulation data at different time periods at the same monitoring point are consistent, and whether the spatial distribution patterns of the actual data and simulation data at various monitoring points within the same pipeline area match, ensuring data consistency across both temporal and spatial dimensions.

[0041] Based on the results of calibration accuracy verification and data consistency verification, quantitative data of core evaluation indicators are extracted to generate calibration quality control curves, which intuitively display the variation law of calibration deviation and matching accuracy level between monitoring data and hydraulic model. A calibration accuracy quality control curve is generated with monitoring point number as the horizontal axis and flow deviation rate and pressure deviation value as the vertical axis; a time-series consistency quality control curve is generated with time as the horizontal axis and data consistency matching degree as the vertical axis, clearly presenting the calibration effect.

[0042] The entire process of data preprocessing, hydraulic modeling, model calibration, and quality control verification was meticulously recorded, generating a data calibration log. The log includes core information such as the operation time, operation content, methods / parameters used, processing results, abnormal situations, and handling measures for each step. Simultaneously, the calibration quality control curve was integrated with the log information to achieve full-process traceability and quality control evaluation of the monitoring data and hydraulic model calibration process. All quality control information was retained according to a unified standard, providing a basis for tracing and verifying each stage of subsequent leakage monitoring.

[0043] S103, based on the monitoring data-hydraulic model pair, extracts the flow deviation, pressure deviation, and nodal hydraulic residual indexes between real data and simulation data, mines the temporal variation trend, spatial distribution law and multi-parameter coupling correlation characteristics of the deviation, constructs a leakage identification feature set, inputs the feature set into the leakage identification model based on graph neural network, and generates preliminary identification results of leakage location, leakage rate and leakage type.

[0044] In one implementation, the monitoring data-hydraulic model pair, the node spatial data of the pipeline network topology, and the time-series data acquired by sensor monitoring are classified, extracted, and aligned in dimension to generate a real monitoring dataset, a simulation calculation dataset, and pipeline network node association matching parameters. Based on data attributes and sources, the three types of core data are specifically classified and extracted. From the monitoring data-hydraulic model pair, the actual sensor-acquired real monitoring data and the simulation data from the hydraulic model simulation calculation are separated. From the pipeline network topology information, node spatial data such as geographic coordinates, zone affiliation, and pipeline connection relationships are extracted. From the raw sensor monitoring data, time-series data of flow and pressure with timestamps are extracted. All types of data are initially standardized according to a unified format, retaining core and valid information. The monitoring data-hydraulic model pair is decoupled, and the actual flow / pressure values ​​of each monitoring point and time period are extracted to form the real monitoring dataset. The simulated flow / pressure values ​​of the corresponding nodes and time periods in the model are extracted to form the simulation calculation dataset. Both datasets retain core fields such as monitoring point codes, timestamps, and parameter values.

[0045] Based on the spatiotemporal characteristics of pipeline network monitoring, bidirectional alignment of various data types after classification and extraction is performed in both time and spatial dimensions. This ensures a one-to-one match between the real monitoring dataset, simulation dataset, and pipeline node spatial data and time-series acquired data in dimensions such as monitoring point codes, timestamps, and pipeline node numbers, eliminating analytical biases caused by spatiotemporal misalignment. Using the binding relationship between pipeline node numbers and monitoring point codes as the spatial benchmark and a unified timestamp as the temporal benchmark, row-dimensional matching is performed between the real monitoring dataset and the simulation dataset. This ensures accurate correspondence between real data and simulation data for the same node and time period. Simultaneously, the temporal resolution of the time-series acquired data is unified to the temporal resolution of the hydraulic model simulation, achieving complete alignment in the temporal dimension.

[0046] The real monitoring data and simulation calculation data that have been classified, extracted and aligned in dimensions are packaged into standardized real monitoring datasets and simulation calculation datasets, respectively. The datasets contain information such as data identifiers, spatiotemporal parameters, and core hydraulic parameters. At the same time, information such as pipeline node numbers, monitoring point codes, and spatiotemporal matching rules in the data dimension alignment process are extracted to generate pipeline node association matching parameters, which are used for subsequent association analysis and feature fusion of various types of data.

[0047] Based on the goal of accurate leakage identification, the deviation calculation rules of the real monitoring dataset and the simulation calculation dataset are correlated. Combined with the hydraulic calculation logic of the pipeline network, the numerical information and spatiotemporal distribution information of flow deviation, pressure deviation, and node hydraulic residuals are integrated to generate the basis for extracting core leakage features. Taking node A (code J001) of the main pipe of a primary water supply network and node B (code J128) of a secondary water supply network in a certain city as monitoring objects, the preset pipeline hydraulic deviation calculation rules conform to the law of flow conservation and the formula for head loss along the pipeline. The specific calculations are as follows. The flow deviation rate is calculated using the relative deviation formula (flow deviation rate = |actual flow - simulated flow| / actual flow × 100%). For node A, the actual flow rate during a certain period is 800 m³ / h, and the simulated flow rate is 760 m³ / h, resulting in a flow deviation rate of 5%. For node B, the actual flow rate during the same period is 120 m³ / h, and the simulated flow rate is 108 m³ / h, resulting in a flow deviation rate of 10%. Both types of data are bound to node codes and timestamps (e.g., "2024-05-20 10:00").

[0048] For pressure deviation rate calculation, using the same relative deviation formula, the actual pressure at node A is 0.4 MPa, the simulated pressure is 0.38 MPa, and the pressure deviation rate is 5%; the actual pressure at node B is 0.32 MPa, the simulated pressure is 0.28 MPa, and the pressure deviation rate is 12.5%, with spatiotemporal identifiers linked synchronously. For node hydraulic residual calculation, based on the hydraulic model residual logic (residual value = actual node head - simulated node head - friction loss - local head loss), the actual head at node A is 0.45 m, the simulated head is 0.42 m, and the total friction and local head loss is 0.02 m, resulting in a residual value of 0.01 m; the actual head at node B is 0.35 m, the simulated head is 0.3 m, the total loss is 0.03 m, and the residual value is 0.02 m, ultimately forming a set of deviation index values ​​with spatiotemporal identifiers.

[0049] To supplement the above deviation index numerical set with spatiotemporal information, the spatial information is supplemented as follows: Node A is supplemented with latitude and longitude (30.67°N, 114.32°E) and its corresponding central urban area zone of the primary water supply network; Node B is supplemented with latitude and longitude (30.72°N, 114.38°E) and its corresponding suburban eastern district zone of the secondary water supply network, thus binding the deviation index with spatial location and network zone. Analysis of the 24-hour flow deviation rate changes at Node A reveals that the deviation rate is stable at 3%-5% from 0-6 (off-peak water use), and rises to 5%-7% from 18-22 (peak water use); Node B's deviation rate reaches 12%-15% from 12-14 (peak water use in the community), forming a temporal trend. This information is ultimately integrated into a comprehensive deviation index information set containing numerical, spatial, and temporal characteristics.

[0050] Based on the comprehensive information set analyzed by the hydraulic calculation logic of the pipeline network, the dimensions of leakage-related deviation characteristics are clarified. For deviation value thresholds: flow deviation rate > 6%, pressure deviation rate > 5%, hydraulic residual > 0.015m in primary pipelines; flow deviation rate > 10%, pressure deviation rate > 10%, hydraulic residual > 0.02m in secondary pipelines, these are identified as suspected leakage-related deviations. If three or more adjacent nodes simultaneously exhibit excessive deviations and show a clustered distribution (e.g., the deviation rates of nodes J129 and J130 around node B both exceed 10% during the same period), it is defined as a spatially clustered leakage characteristic. A continuously increasing deviation rate during non-peak water usage periods (e.g., the deviation rate of node B rises from 8% to 12% between 2:00 AM and 6:00 AM), or a deviation rate during peak periods far exceeds the historical average for the same period (exceeding 30%), is defined as a temporally abnormal leakage characteristic. The final basis for extracting core features of leakage is generated, and the core indicators are defined as flow / pressure deviation rate and hydraulic residual. The analysis dimensions cover numerical threshold, spatial clustering and time series trend. The selection criterion is that if any leakage-related feature dimension is met, it will be included in the subsequent feature set.

[0051] Based on the construction goals of the leakage identification feature set, the core leakage feature data, the mining rules for the temporal changes of deviation, and the analysis strategy for multi-parameter coupling and correlation are respectively mapped to the feature quantification model and the feature set construction standard, generating core parameters for leakage identification feature calculation and initial information for feature screening. A linear quantification model is selected as the carrier for leakage feature quantification. The spatiotemporal characteristics of flow deviation (deviation rate increases from 8% to 12% between 2-6 AM, showing a continuous upward trend) and the spatial characteristics of pressure deviation (pressure deviation rates of the three surrounding nodes all exceed 10%, showing a clustered distribution) of node B (coded J128) in the secondary pipeline network are input into the model. The model transforms the qualitative trend of "continuous increase" into a quantitative trend slope (0.002% / min) and the "clustered distribution" into a spatial aggregation quantification value (0.85), achieving a quantitative expression of leakage features.

[0052] The rules for mining time-series deviation changes (an absolute value of the trend slope ≥ 0.001% / min is judged as a time-series abrupt change) and the multi-parameter coupling correlation analysis strategy (a flow deviation rate exceeding 10% and a pressure deviation rate exceeding 8% are judged as a strong coupling anomaly) are incorporated into the feature set construction criteria. For example, node B has a flow deviation time-series slope of 0.002% / min, and both flow and pressure deviations exceed the standard, satisfying the feature mining and correlation logic in the criteria, and is marked as a core candidate feature.

[0053] Based on the hydraulic characteristics of the pipeline network and historical leakage data, the core parameters for feature calculation are set as follows: For the primary pipeline network, the spatial dimension weight is set to 0.4, the time dimension to 0.3, and the numerical dimension to 0.3; for the secondary pipeline network, since leakage is mostly related to local spatial distribution, the spatial dimension weight is increased to 0.5, the time dimension to 0.25, and the numerical dimension to 0.25. The secondary pipeline network to which node B belongs is calculated according to this weight. The quantification coefficient for temporal abrupt change features is set to 1.2 (higher than the 1.0 of ordinary temporal features). The quantification value of the temporal slope of node B (0.002% / min) multiplied by 1.2 highlights its temporal anomaly characteristics. The quantification coefficient for spatial clustering is set to 1.1. The spatial clustering degree of node B (0.85) multiplied by 1.1 strengthens the spatial correlation characteristics. The proportions of flow deviation features are 0.4, pressure deviation features are 0.3, and hydraulic residual features are 0.3. The quantification values ​​of flow deviation, pressure deviation, and hydraulic residual of node B are 0.12, 0.125, and 0.02, respectively. After proportional fusion, the comprehensive feature value is 0.101.

[0054] The feature correlation threshold was set to 0.6 (below this value is considered weakly correlated), and the feature validity threshold was set to 0.08 (above this value is considered a valid feature). The redundant feature removal rule was "for the same node, between the flow deviation rate and the relative flow deviation, retain the one with the higher correlation." Node B's comprehensive feature value of 0.101 is higher than the validity threshold, and its correlation with the leakage condition is 0.75 (higher than the correlation threshold), so it is retained as a core feature. The correlation between the "flow deviation rate" and "relative flow deviation" of its neighboring node J129 reaches 0.92, so the relative flow deviation feature is removed according to the rule. The final leakage identification feature set retains only the highly correlated and non-redundant core features.

[0055] Based on the core feature extraction criteria for leakage and the core parameters and initial information for feature selection for leakage identification feature calculation, key features in the core data extracted by classification are marked and fused. The constructed leakage identification feature set is then input into a leakage identification model based on a graph neural network to generate preliminary leakage identification results for the water supply network, including leakage location, leakage rate, and leakage type. Based on the core feature extraction criteria, the core data extracted by classification is traversed and analyzed to mark key features highly correlated with leakage conditions, including abnormal deviation numerical features, typical spatial distribution features of leakage, and temporal abrupt change features. Simultaneously, according to the core parameters for feature calculation, the marked key features are quantitatively extracted to obtain a key feature set with quantified values. Flow deviation features in the network that exceed the threshold, exhibit clustered spatial distribution, or show sudden temporal changes are marked. The quantified values ​​of each marked feature are calculated according to the set quantification parameters to form a quantified set of key features.

[0056] Combining the analysis strategy of pipeline hydraulic characteristics and multi-parameter coupling correlation, the extracted key features are subjected to multi-dimensional fusion calculations to achieve deep fusion of numerical features, spatial features, and temporal features. Simultaneously, redundant and weakly correlated features are removed based on initial information from feature selection. Finally, a standardized leakage identification feature set with a regular structure, complete information, and no redundancy is constructed. The feature dimensions and format of the feature set match the input requirements of the graph neural network leakage identification model. The quantitative features of flow deviation, pressure deviation, and node hydraulic residuals at the same node are coupled and fused to generate multi-parameter coupled features. The spatial distribution features of different nodes are regionally fused to generate regional spatial features, while redundant features generated during the fusion process are removed, forming the final leakage identification feature set.

[0057] The leakage identification model used in this section is a water supply network leakage identification model based on graph neural networks (GNN). The model uses the network topology as the graph structure, with network nodes as graph nodes and pipe connections as graph edges. It perfectly adapts to the topological characteristics of the water supply network, realizing spatial correlation analysis and reasoning of leakage features. The specific model application and operation logic are as follows: The standardized leakage identification feature set is used as the core input of the model. At the same time, the pipeline node association matching parameters and the node spatial data of the pipeline topology are used as auxiliary inputs. The model input data are all encapsulated in the graph data format of graph neural network. The leakage features of each pipeline node are used as the feature attributes of the graph node, and the pipeline connection relationship is used as the association weight of the graph edge.

[0058] The graph structure of the graph neural network corresponds one-to-one with the physical topology of the water supply network. The model performs convolution operations on the leakage features of adjacent nodes through graph convolutional layers, and explores the hydraulic correlation features between network nodes and the spatial propagation law of leakage. It fits the actual business logic of the flow and pressure deviations after leakage in the water supply network and is transmitted along the pipeline, so as to achieve accurate identification of leakage conditions.

[0059] After receiving the input data, the model performs deep extraction and fusion of leakage identification features through multi-level operations of graph convolutional layers, pooling layers, and fully connected layers. Finally, the output layer outputs the leakage probability and related parameters of each pipeline node, completing the leakage identification inference process.

[0060] The inference output of the graph neural network leakage identification model is analyzed and organized. The specific location of the leakage is determined based on the leakage probability threshold output by the model. The leakage rate is calculated based on the deviation value of the hydraulic parameters output by the model. The leakage type (such as point leakage, pipe section breakage leakage, etc.) is determined based on the network characteristics and deviation change pattern of the leakage location. Finally, a preliminary identification result of water supply network leakage containing three core information: leakage location, leakage rate, and leakage type is generated. Each piece of information is accompanied by the confidence value of the model inference, providing data support for subsequent quality control judgment.

[0061] S104, based on the confidence output of the identification model, performs multiple inferences for the same monitoring period to obtain multiple sets of confidence data, calculates the monitoring point level uncertainty by standard deviation method or coefficient of variation method, combines the suspected leakage area mask to construct the sensitive ring zone of the pipeline network, aggregates the monitoring point level uncertainty in the ring zone area to obtain the regional level score, and completes the quality control conclusion judgment by setting the threshold based on the historical leakage verification set.

[0062] In one implementation, based on the quality control requirements for leakage identification and the uncertainty calculation objective, the confidence output of the identification model, multi-round inference data for the same monitoring period, and spatial correlation information of monitoring points are structurally collected and standardized to generate basic quality control data containing confidence values, inference result fluctuations, and topological correlations of monitoring points. According to data attributes and quality control analysis requirements, three types of core data are collected in a targeted and structured manner: confidence values ​​corresponding to each monitoring point and each leakage identification result are collected from the identification model output; for multiple sets of data obtained from multiple inferences performed within the same monitoring period, the full inference results and confidence fluctuation data are collected; and topological correlation information such as node connection relationships, network partitions, and spatial adjacency relationships of monitoring points are collected from the pipeline network topology information, retaining the original spatiotemporal identifiers and correlation attributes of the data during the collection process. Using the monitoring point code as a unique identifier, the confidence level of the model's identification results of the leakage location, rate, and type of each monitoring point is collected. At the same time, the real-time change data of the confidence level of each monitoring point is collected during 10 inference processes within the same monitoring period. The upstream and downstream connection nodes and the pipeline network zoning information corresponding to each monitoring point are collected simultaneously to form a set of original quality control data with identification.

[0063] The collected raw quality control data undergoes standardized processing, including format unification, dimension alignment, and information completion. Confidence scores are uniformly mapped to fixed numerical ranges. Fluctuation data from multiple rounds of inference are normalized according to unified rules. Spatial correlation information of monitoring points is transformed into standardized topological correlation codes. Simultaneously, monitoring point codes, timestamps, and other correlation identifiers for various data types are completed to achieve accurate matching of the three types of data. All confidence scores output by the identification models are standardized to the 0-1 range. Extreme values, mean values, and other statistics are calculated for confidence score fluctuation data from multiple rounds of inference and recorded in a fixed format. Spatial adjacency relationships of monitoring points are transformed into topological correlation matrix codes, ensuring that each set of confidence score data and fluctuation data can be accurately bound to the spatial correlation information of the corresponding monitoring point.

[0064] The three types of data that have completed structured collection and standardized processing are integrated and packaged to generate standardized quality control basic data. This data includes three core dimensions: the confidence level of leakage identification at each monitoring point, the fluctuation of inference results during the same monitoring period, and the topological correlation of each monitoring point. The data in each dimension are correlated with the monitoring point code as the core identifier, which can directly support subsequent uncertainty calculation and regional aggregation analysis.

[0065] Based on the uncertainty analysis characteristics of water supply network leakage identification, calculation rules for monitoring data are designed, clarifying the applicable scenarios of the standard deviation method and the coefficient of variation method, the calculation threshold of uncertainty at the monitoring point level, and generating a specification for quantifying network uncertainty. For the primary water supply network in the central urban area (node ​​codes J001-J010), the confidence scores for leakage identification at each monitoring point are all in the range of 0.6-0.9 (uniform units), and conform to a normal distribution after normality testing (P-value > 0.05). The standard deviation method is selected to calculate uncertainty. For example, the 10-fold inference confidence score for node J003 is [0.78, 0.80, 0.79, 0.81, 0.82, 0.77, 0.80, 0.81, 0.79, 0.80], and the calculated standard deviation is 0.015, which is used as the uncertainty value for that node.

[0066] For the secondary water supply network in the suburbs (node ​​codes J128-J135), there are implicit differences in the dimensions of the confidence scores at each monitoring point (confidence scores for some nodes are concentrated in the range of 0.5-0.6, while those for others are in the range of 0.8-0.9, showing a large degree of dispersion). Therefore, the coefficient of variation method is selected. For example, the standard deviation of the confidence scores for node J128 after 10 inferences is 0.03, with a mean of 0.6, and the coefficient of variation is 0.03 / 0.6 = 0.05; the standard deviation of node J130 is 0.02, with a mean of 0.8, and the coefficient of variation is 0.02 / 0.8 = 0.025. The coefficient of variation eliminates the influence of dimensions, enabling cross-node uncertainty comparison. The method switches from the standard deviation method to the coefficient of variation method when the p-value for the normality test is ≤0.05 or the coefficient of variation of the mean confidence scores for each node is >0.2. If the p-value for the normality test of the confidence scores for a certain period in the primary network is 0.03 (not conforming to a normal distribution), the method automatically switches to the coefficient of variation method.

[0067] The threshold for valid data volume in multi-round inference is set to ≥8 times. If a monitoring point only obtains 6 valid inference data due to temporary equipment failure, uncertainty calculation will not be performed temporarily until supplementary data is obtained. The confidence fluctuation judgment threshold is set to ±0.15 (based on the average confidence of 10 inferences). The average confidence of 10 inferences for node J132 is 0.75, with one data point at 0.58 (deviation from the average by 0.17, exceeding the threshold). This data point is judged as abnormal data and removed. Uncertainty is calculated based on the remaining 9 valid data points. The effective threshold for the final uncertainty value is set to 0-0.2. If the uncertainty value calculated by a node is 0.23 (exceeding the threshold), the node's recognition result is judged to have low credibility, and multi-round inference needs to be repeated.

[0068] First, the validity of the multi-round inference data is screened (outliers are removed) -- the consistency of data distribution and dimensions is checked -- the calculation method is selected according to rules -- the uncertainty at the monitoring point level is calculated -- and the results are verified to be within the valid threshold. Only inference data from the same monitoring period, without equipment failure or human intervention, are selected to ensure data homogeneity and reliability. The standard deviation method formula is as follows: (in, (where μ is the single confidence level, n is the mean, and σ is the number of valid data points). The formula for the coefficient of variation method is CV = σ / μ. Data is recorded in the structure of "monitoring point code - monitoring period - data volume - calculation method - uncertainty value - result validity", for example, "J003-2024-05-2010:00-10 times-standard deviation method-0.015-valid", achieving standardized traceability throughout the entire process.

[0069] Based on the regional aggregation requirements and scoring logic of leakage quality control, a multi-level computational mechanism is set up, including suspected area mask expansion, construction of sensitive ring zones in the pipeline network, and aggregation of uncertainties within the ring zones. Taking the suspected leakage areas of the suburban eastern district of the secondary water supply network (including nodes B / J128, J129, and J130) as the target, the mask expansion rules are set as follows: based on the original mask boundary (within 500 meters around node B), the expansion extends outward along the upstream and downstream connection direction of the pipeline network without crossing the boundary of the pipeline network zone; the step size is set to 200 meters, and the maximum expansion range does not exceed 1.5 kilometers to avoid including areas without hydraulic correlation; after expansion, the original mask only covers node B and one surrounding monitoring point, while after expansion, all upstream and downstream related nodes J129 and J130 and the connecting pipeline sections are included, forming an expanded mask with 3 core monitoring points and a total range of 1.2 kilometers, ensuring that the related areas along the leakage transmission path are covered.

[0070] Based on the expanded mask, the rules for constructing sensitive ring zones were set and applied: taking the core location of the suspected leakage area (node ​​B coordinates 30.72°N, 114.38°E) as the center, it was divided into three levels of ring zones: Level 1 ring zone (hydraulic distance ≤ 300 meters), Level 2 ring zone (300 meters < hydraulic distance ≤ 800 meters), and Level 3 ring zone (800 meters < hydraulic distance ≤ 1200 meters); only monitoring points with complete monitoring data and direct hydraulic connection to the core area were included in the corresponding ring zone; the ring zone composition results are as follows: Level 1 ring zone includes node B, Level 2 ring zone includes J129, and Level 3 ring zone includes J130. The spatial boundaries of each ring zone are defined by the hydraulic distance threshold, clearly distinguishing monitoring areas with different correlation intensities.

[0071] Aggregation rules were set and the above ring zones were calculated: a weighted average method was used, with the weight of the monitoring points in the first-level ring zone set to 0.6 (core correlation), the second-level ring zone set to 0.3 (moderate correlation), and the third-level ring zone set to 0.1 (weak correlation). The weights were calibrated based on hydraulic conduction efficiency. The uncertainty value of node B was 0.05 (calculated using the coefficient of variation method), J129 was 0.025, and J130 was 0.03. The aggregation value of the first-level ring zone = 0.05 × 0.6 = 0.03, the second-level ring zone = 0.025 × 0.3 = 0.0075, and the third-level ring zone = 0.03 × 0.1 = 0.003. The total aggregation value of the region = 0.03 + 0.0075 + 0.003 = 0.0405.

[0072] The integrated multi-level computing mechanism first generates an expansion mask according to the expansion rule, then divides sensitive zones based on hydraulic distance, aggregates the uncertainties of each zone according to weight, and outputs the total aggregated value of the region. The expansion step size is 200 meters, the maximum range is 1.5 kilometers, the hydraulic distance thresholds for the zones are 300 meters / 800 meters / 1200 meters, and the weights are 0.6 / 0.3 / 0.1. The result judgment criteria are as follows: the total aggregated value of the region needs to be compared with historical data to ensure that it is within a reasonable fluctuation range, so as to provide a standardized basis for subsequent conversion into regional quality control scores.

[0073] The process integrates and executes basic quality control data, pipeline uncertainty quantification specifications, and multi-level calculation mechanisms. Quality control results are determined through threshold comparison, generating a quality control conclusion that includes monitoring point-level uncertainty values, regional quality control scores, and leakage identification. Basic quality control data is input into the pipeline uncertainty quantification specifications. The corresponding calculation method (standard deviation method or coefficient of variation method) is selected according to the applicable scenarios defined in the specifications. After removing outlier data, uncertainty calculations are performed on the multi-round inference confidence data of each monitoring point during the same monitoring period to obtain the monitoring point-level uncertainty value for each monitoring point. All calculation processes strictly follow the standardized operating procedures and calculation thresholds to ensure the accuracy of the calculation results. For monitoring points that meet the applicable scenarios of the standard deviation method, the standard deviation of their multi-round inference confidence data is calculated, and this standard deviation is used as the uncertainty value for that monitoring point. For monitoring points that meet the applicable scenarios of the coefficient of variation method, the coefficient of variation of their confidence data is calculated, and this coefficient of variation is used as the uncertainty value for that monitoring point. All calculation results are bound to the monitoring point code.

[0074] Based on the calculated uncertainty values ​​at the monitoring point level, a full-process calculation is carried out according to a multi-level operation mechanism. First, the mask of the suspected leakage area is expanded, then a multi-level network sensitive ring is constructed, and finally, according to the uncertainty aggregation rules within the ring, the uncertainty values ​​at the monitoring point level in each ring are weighted and aggregated. At the same time, combined with the quality control evaluation criteria for leakage identification, the aggregated regional uncertainty values ​​are converted into standardized regional quality control scores. The score is negatively correlated with the regional quality control level, that is, the higher the score, the better the quality control level of the regional leakage identification results.

[0075] Based on massive amounts of historical leakage verification data and quality control analysis results, and combined with the actual business needs of leakage monitoring, multi-level thresholds for leakage identification quality control judgment are set, including a Pass threshold, a Needs-Inspection threshold, and a Fail threshold. Each threshold corresponds to a different regional-level quality control score range. The aggregated and calculated regional-level quality control scores are compared one by one with the preset thresholds to determine the quality control level of each suspected leakage area. A regional-level quality control score ≥80 is set as the Pass level, 60-79 as the Needs-Inspection level, and <60 as the Fail level. The actual quality control score of each area is compared with this threshold range to determine the corresponding quality control level.

[0076] The monitoring point-level uncertainty values ​​of each monitoring point, the regional-level quality control scores of each suspected leakage area, and the quality control level judgment results of each area are integrated to generate a complete leakage identification quality control conclusion judgment result. This result includes three core parts: a set of uncertainty values ​​at the monitoring point level of the entire pipeline network (including uncertainty values ​​and identifiers of each monitoring point), a quality control score table for each suspected leakage area (including quality control scores and ring-zone level information for each area), and a leakage identification quality control conclusion judgment table (including quality control level, judgment basis, and threshold comparison results for each area). All result data are standardized in a unified format, which can directly support subsequent result visualization and structured report output.

[0077] S105 integrates basic pipeline data, monitoring data-hydraulic model pairs, leakage identification results, monitoring point-level uncertainty heatmaps, and regional-level scores, and displays them in a multi-dimensional overlay visualization on the terminal monitoring platform. It also outputs a structured report that includes data calibration logs, scores for each region, quality control judgment results, and priority inspection suggestions.

[0078] In one implementation, based on the visualization requirements of pipeline network monitoring, the computing power of the terminal platform, and the intuitiveness requirements of data display, the system integrates the type attributes, dimensional correlation logic, and display boundaries of multi-source pipeline network data, clarifying the core basis and constraints for visualization construction. It comprehensively reviews pipeline network basic data, monitoring data and hydraulic model pairing results, leakage identification results, monitoring point-level uncertainty heatmaps, regional scores, and other information, distinguishing different attributes such as topological data, numerical data, and image data, and sorting out the inherent logic such as spatiotemporal correlation and causal correlation to evaluate the core information value of each data item. The system is categorized and integrated according to visualization priority and business relevance, selecting key data dimensions that can support operation and maintenance decisions. Pipeline network basic topology data, core identification results such as leakage location and leakage rate, and uncertainty heatmaps are listed as high-priority data, while detailed intermediate data during calibration are listed as low-priority data. Following the principle of prioritizing the display of core data and retrieving auxiliary data as needed, the overall data display logic is constructed.

[0079] Based on the actual business objectives of water supply network leakage monitoring, the core basis for visualization construction is clearly defined. Focusing on three main directions—accurate presentation of leakage locations, intuitive differentiation of risk levels, and convenient support for operation and maintenance—the core dimensions, granularity, and related display logic of visualization are determined, enabling the visualization results to directly support operation and maintenance decisions. With the goal of accurately presenting leakage locations, the network topology display is refined to the secondary pipeline segment level, and specific node coordinates and leakage types are marked in the leakage identification result layer. With the goal of intuitively differentiating risk levels, uncertainty heatmaps are displayed using color gradients corresponding to the risk levels.

[0080] Based on the computing power limits of the terminal platform, such as data processing speed, concurrent display capabilities, and compatibility with display devices (e.g., PC / mobile compatibility and data security requirements), visualization constraints are defined. These constraints include the maximum number of data display levels, the maximum number of overlay layers, the response latency threshold for interactive operations, and the display permissions for sensitive data. This ensures the stability and feasibility of the visualization solution in practical applications. According to the terminal platform's computing power, the number of overlay layers is set to no more than 5 to avoid display lag due to excessive layers; the response latency for interactive operations is set to no more than 1 second to ensure smooth operation for maintenance personnel; and for sensitive data such as the core topology of the pipeline network, hierarchical viewing permissions are set, allowing only authorized personnel to view the complete information.

[0081] To balance the need for multi-dimensional data overlay and display with operational and maintenance decision-making efficiency, a visualization hierarchy design was developed, clearly defining the criteria for the core display layer and related configuration items. The core display layer includes the pipeline network basic topology layer, the leakage identification result layer, and the uncertainty hotspot layer. Related configuration items include data overlay methods, layer transparency, and interactive query permissions.

[0082] The three core display layers have independent definition standards and functional positioning, each capable of fulfilling specific data display needs while supporting flexible overlay usage. The pipeline network basic topology layer, based on pipeline network topology information, defines the display scope, clarifies core display elements and identification rules, and its core function is to provide a spatial benchmark for the pipeline network, offering spatial positioning support for other layers. This layer displays, by default, the primary and secondary pipeline sections and key monitoring points of the entire pipeline network, labeling pipeline section codes, monitoring point numbers, and geographic coordinates, providing spatial reference for leak location marking and uncertain heatmap overlay.

[0083] The leakage identification results layer focuses on the preliminary leakage identification results, clearly displaying the content, labeling rules, and information update frequency. Its core function is to intuitively present key leakage information. This layer uses red dots to mark the location of point leaks, orange lines to mark the extent of pipe section damage and leakage, directly labels the leakage rate value, and synchronously updates the leakage information according to the monitoring data update frequency.

[0084] The uncertainty heatmap is based on the uncertainty values ​​at the monitoring point level. It establishes rules for heatmap generation, display range, and transparency standards. Its core function is to intuitively distinguish the reliability level of leakage identification results. A color mapping relationship is used according to the uncertainty value range: green corresponds to values ​​between 0 and 0.3, representing high reliability; yellow corresponds to values ​​between 0.3 and 0.7, representing medium reliability; and red corresponds to values ​​between 0.7 and 1, representing low reliability. The heatmap only covers suspected leakage areas and surrounding related pipelines, ensuring key information is clearly highlighted.

[0085] The three related configuration items also have clearly defined standards and parameter ranges, adaptable to diverse display scenarios and operational needs. Layer overlay includes three core modes: transparent overlay, toggle display, and layered hiding. Transparent overlay is suitable for simultaneously viewing the pipeline topology and leakage location; toggle display is suitable for comparing leakage changes over different time periods; and layered hiding is suitable for focusing on details of a single layer. The system defaults to overlaying a basic pipeline topology layer and a leakage identification result layer, while also supporting custom layer combinations.

[0086] The transparency of each core display layer can be adjusted between 30% and 100%. The system sets the default transparency to 70% for the basic pipeline topology layer and 90% for the leakage identification result layer, balancing the visibility of the topology baseline with the highlighting effect of leakage information. The default transparency of the uncertainty heatmap is 60%, which does not obscure the leakage markings and pipeline topology details after being overlaid.

[0087] Based on the data security management and operation and maintenance division of labor requirements, interactive query hierarchical permissions are set. Ordinary operation and maintenance personnel have viewing permissions, can browse visualization results and perform basic queries; technical management personnel have editing permissions, can adjust layer-related configurations; administrators have export permissions, can export visualization data and report content, and permissions can be dynamically adjusted.

[0088] Based on the actual operation and maintenance characteristics of water supply network leakage monitoring, visualization optimization rules were formulated, including on-demand scaling of the topology layer, precise labeling of the result layer, and color gradient adaptation of the heatmap layer, to improve the display effect and practical adaptability. The topology layer supports multi-level switching of global scaling, partition scaling, and node scaling. After scaling, the elements can be displayed adaptively. The global view displays the primary pipeline and core monitoring points, while the node view displays the detailed parameters of the corresponding nodes. Double-clicking a leakage node can quickly zoom in to the surrounding area, meeting different needs for global inspection planning and on-site inspection positioning.

[0089] The labeling of the results layer follows offset rules to avoid obscuring pipes and key elements. Multiple leaks in the same pipe section are labeled together, and clicking on a label expands to view details of each leak point. The interface displays the location and type of the leak by default, and when the mouse hovers over it, it displays complete information such as the leak rate, confidence level, and identification time, avoiding labeling chaos and ensuring information completeness.

[0090] The hotspot layer uses the RGB color space, with gradients divided according to the range of uncertain values, resulting in clear color contrasts that align with the cognitive habits of safety warnings. The system provides a color weakness adaptation mode, which replaces red with blue and yellow with purple after switching, ensuring that color-blind maintenance personnel can accurately identify gradient differences.

[0091] By integrating visualization constraints, hierarchical setting standards, and optimization rules, and simultaneously completing the structured compilation and format standardization of data calibration logs, regional quality control scores, quality control judgment results, and priority inspection suggestions, a water supply network leakage monitoring results output system is constructed, which includes multi-dimensional overlay visualization configuration data and standardized structured report templates. The visualization construction basis, constraints, hierarchical standards, and optimization rules are parameterized and structured to form visualization configuration data that can be directly imported into the terminal monitoring platform. Parameters such as the line width of the pipeline network basic topology layer, the icon size of the leakage identification result layer, and the color mapping table of the uncertainty heat layer are clearly defined, enabling rapid deployment of visualization functions.

[0092] The system systematically analyzes key data throughout the entire process, identifies core dimensions for structured reports, and covers data calibration logs, regional quality control scores, quality control judgment results, and priority inspection recommendations. It extracts key indicators such as the number of outliers removed and the average flow deviation rate, and generates an inspection priority ranking based on leakage rate, quality control level, and pipeline importance, ensuring that the report data is comprehensive, accurate, and consistent with the visualization results.

[0093] A unified report template is designed, dividing the report into sections such as monitoring overview, data calibration results, leakage identification results, quality control judgment results, inspection recommendations, and attachments. The report standardizes the presentation format of various data and the summary format of judgment results. The report is output in PDF format by default, and authorized personnel can export it in an editable Word format. The attachments also include supplementary materials such as data calibration logs and the original uncertainty heatmap.

[0094] By integrating multi-dimensional overlay visualization configuration data with standardized structured report templates, a complete output system for water supply network leakage monitoring results is formed. This system not only supports real-time multi-dimensional overlay visualization on the terminal monitoring platform to meet the dynamic viewing and interactive operation needs of operation and maintenance personnel, but also quickly outputs standardized structured reports to support multi-department collaborative decision-making and results archiving, achieving efficient delivery and practical transformation of monitoring results.

[0095] In one implementation, such as Figure 2 As shown, this application also provides a monitoring device for leakage in water supply networks, comprising: The multi-source pipeline network data acquisition module 201 is used to collect flow monitoring data, pressure sensing data and pipeline network topology information of water supply pipeline network, and read basic pipeline network parameters such as monitoring point location, data acquisition frequency, pipeline diameter / material in real time, and generate a standardized pipeline network monitoring basic dataset. The data preprocessing and model calibration module 202 is used to perform outlier removal and smoothing preprocessing on the raw flow / pressure data based on the pipeline network basic data. It constructs a hydraulic model mask based on the pipeline network topology, screens suspected leakage areas through regional hydraulic balance analysis, realizes bidirectional calibration of monitoring data and hydraulic model through flow-pressure coupling characteristics, introduces time series data trend analysis to compensate for dynamic load changes in the pipeline network, generates monitoring data-hydraulic model pairs, and outputs calibration quality control curves and log information. The leakage feature extraction and identification module 203 is used to extract the flow deviation, pressure deviation, and nodal hydraulic residual index of real and simulated data based on the monitoring data-hydraulic model pair, mine the time-series change trend, spatial distribution law and multi-parameter coupling correlation features of the deviation, construct the leakage identification feature set, input it into the leakage identification model based on graph neural network, and generate preliminary identification results of leakage location, leakage rate and leakage type. The leakage identification quality control judgment module 204 is used to perform multiple inferences on the same monitoring period based on the confidence output of the identification model to obtain multiple sets of confidence data. It calculates the monitoring point level uncertainty by standard deviation method or coefficient of variation method, combines the leakage suspected area mask to construct the pipeline network sensitive ring, aggregates the monitoring point level uncertainty in the ring area to obtain the regional level score, and completes the quality control conclusion judgment by setting the threshold according to the historical leakage verification set. The results visualization and report output module 205 is used to integrate basic pipeline data, monitoring data-hydraulic model pairs, leakage identification results, monitoring point-level uncertainty heatmaps and regional scores, and complete multi-dimensional overlay visualization on the terminal monitoring platform. At the same time, it outputs a structured report including data calibration logs, scores of each region, quality control judgment results and priority inspection suggestions.

[0096] The computer-readable storage medium provided in the above embodiments of this application and the monitoring method for leakage in water supply networks provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0097] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating the monitoring method, system, electronic device, and readable storage medium for monitoring leakage in water supply networks are basically similar to the embodiments for monitoring leakage in water supply networks described above, and therefore the description is relatively simple. Relevant parts can be referred to in the descriptions of the embodiments for monitoring leakage in water supply networks described above.

Claims

1. A method for monitoring leakage in water supply networks, characterized in that, include: Collect flow monitoring data, pressure sensing data and pipeline topology information of water supply network, and read basic pipeline parameters including monitoring point location, data acquisition frequency, and pipeline diameter / material; Based on pipeline network data, the raw flow / pressure data is first preprocessed by outlier removal and smoothing. A hydraulic model mask is constructed based on the pipeline network topology. Preliminary screening of suspected leakage areas is completed through regional hydraulic balance analysis. Then, bidirectional calibration of the monitoring data and hydraulic model is achieved through flow-pressure coupling characteristics. Time-series data trend analysis is introduced to compensate for dynamic load changes in the pipeline network, generating monitoring data-hydraulic model pairs. Simultaneously, calibration quality control curves and log information are output, including: hydraulic analysis requirements based on water supply network leakage monitoring, preprocessing specifications for multi-source sensor data, and spatial analysis of the pipeline network topology. The system performs multi-dimensional preprocessing on the collected raw flow / pressure data, including outlier removal, data smoothing, and time-series completion, generating standardized flow datasets, standardized pressure datasets, and time-series correlated sensor data for hydraulic calculations. Based on the hydraulic modeling specifications for water supply networks, the system performs structured processing on the standardized flow datasets, standardized pressure datasets, and time-series correlated sensor data. Spatial mapping between monitoring points and hydraulic model nodes is achieved through network topology matching, generating a network hydraulic model mask, a regional hydraulic balance calculation template, and a flow-pressure coupling analysis framework. The analysis logic for screening leaks in water supply networks involves performing regional hydraulic verification, initial screening of suspected areas, and coupling feature calibration on the network hydraulic model mask, regional hydraulic balance calculation template, and flow-pressure coupling analysis framework. This completes the location of suspected leak areas and bidirectional matching of monitoring data and hydraulic models, compensating for deviations caused by dynamic load changes in the network, and generating monitoring data-hydraulic model pairs. The calibration accuracy and data consistency of these pairs are verified, and calibration quality control curves and data calibration log information are generated based on the quality control evaluation standards for network monitoring, thus realizing the integration of monitoring data and hydraulic models. The entire process of model calibration is recorded and quality control is evaluated. The flow-pressure coupling analysis framework uses the hydraulic calculation formula H=KQ², where H represents head loss, K represents the network resistance coefficient, and Q represents flow rate. A one-hour time-series data sliding window is used to calculate the theoretical pressure values ​​corresponding to different flow ranges, which are then compared with the actual monitored pressure values. Two-way calibration involves adjusting the pipe friction coefficient and local resistance coefficient in the hydraulic model based on the flow-pressure coupling variation characteristics of the actual monitoring data, ensuring that the simulated flow-pressure correlation is consistent with reality. Based on the monitoring data-hydraulic model pair, flow deviation, pressure deviation, and nodal hydraulic residual indices are extracted from real and simulated data. The temporal variation trend, spatial distribution pattern, and multi-parameter coupling correlation characteristics of the deviations are explored to construct a leakage identification feature set. This feature set is input into a graph neural network-based leakage identification model to generate preliminary identification results for leakage location, leakage rate, and leakage type. This includes: classifying and extracting data from the monitoring data-hydraulic model pair, the node spatial data of the pipeline topology, and the time-series data acquired by sensors, and performing dimensional alignment processing to generate a real monitoring dataset, a simulation calculation dataset, and pipeline node association matching parameters. Based on the accuracy target of leakage identification, the deviation calculation rules of the real monitoring dataset and the simulation calculation dataset are associated, combined with the pipeline hydraulic calculation logic. The system integrates numerical and spatiotemporal distribution information of flow deviation, pressure deviation, and node hydraulic residuals to generate a basis for extracting core leakage features. Based on the construction goals of the leakage identification feature set, the core leakage feature data, the mining rules for deviation time series changes, and the analysis strategy for multi-parameter coupling correlation are respectively mapped to feature quantification models and feature set construction standards to generate core parameters for leakage identification feature calculation and initial information for feature selection. Based on the basis for extracting core leakage features and the core parameters for leakage identification feature calculation and initial information for feature selection, the key features in the core data extracted by classification are labeled and fused. The completed leakage identification feature set is input into the leakage identification model based on graph neural network to generate preliminary leakage identification results of water supply network including leakage location, leakage rate, and leakage type. Based on the confidence output of the identification model, multiple inferences are performed for the same monitoring period to obtain multiple sets of confidence data. The uncertainty at the monitoring point level is calculated using the standard deviation method or the coefficient of variation method. A sensitive ring zone of the pipeline network is constructed using a mask of suspected leakage areas. The uncertainty at the monitoring point level is aggregated in the ring zone to obtain a regional-level quality control score. A threshold is set based on the historical leakage validation set to complete the quality control conclusion judgment. This includes: based on the quality control requirements for leakage identification and the uncertainty calculation objective, the confidence output of the identification model, multiple rounds of inference data for the same monitoring period, and spatial correlation information of monitoring points are structurally collected and standardized to generate basic quality control data containing confidence values, fluctuations in inference results, and topological correlations of monitoring points; based on the uncertainty analysis characteristics of water supply network leakage identification, the calculation rules for monitoring data are designed, clarifying the applicable scenarios of the standard deviation method and the coefficient of variation method, the calculation threshold for uncertainty at the monitoring point level, and generating pipeline network uncertainty... The system employs a deterministic quantification standard. Combining the regional aggregation requirements and score evaluation logic of leakage quality control, it establishes a multi-level computational mechanism for suspected area mask expansion, pipeline network sensitive ring construction, and uncertainty aggregation within the ring. It integrates and executes the quality control basic data, pipeline network uncertainty quantification standard, and multi-level computational mechanism, completing the quality control result judgment through threshold comparison, and generating a quality control conclusion judgment result containing monitoring point-level uncertainty values, regional-level quality control scores, and leakage identification. Specifically, the calculation of monitoring point-level uncertainty using the standard deviation method or the coefficient of variation method includes: when the leakage identification confidence data of each monitoring point conforms to a normal distribution after a normality test and the P-value is >0.05, the standard deviation method is used to calculate uncertainty; when there are implicit differences in the dimensions and large differences in the dispersion of the confidence data of each monitoring point, the coefficient of variation method is used. The system switches from the standard deviation method to the coefficient of variation method when the normality test P-value is ≤0.05 or the coefficient of variation of the mean confidence of each node is >0.

2. Integrating basic pipeline data, monitoring data-hydraulic model pairs, leakage identification results, monitoring point-level uncertainty heatmaps, and regional-level quality control scores, the system provides multi-dimensional overlay visualization on the terminal monitoring platform. It also outputs a structured report containing data calibration logs, regional-level quality control scores, quality control judgment results, and priority inspection recommendations.

2. The method as described in claim 1, characterized in that, Integrating basic pipeline network data, monitoring data-hydraulic model pairs, leakage identification results, monitoring point-level uncertainty heatmaps, and regional-level quality control scores, the system provides multi-dimensional overlay and visualization on the terminal monitoring platform. It also outputs a structured report including data calibration logs, regional-level quality control scores, quality control judgment results, and priority inspection recommendations. Based on the needs for visualization of pipeline monitoring, the computing power support of terminal platforms, and the requirements for intuitive data display, this paper integrates the type attributes, dimensional correlation logic, and display boundaries of multi-source pipeline data, and clarifies the core basis and constraints for visualization construction. Based on the need to balance the display of multi-dimensional data with the efficiency of operation and maintenance decision-making, the logic for setting the visualization hierarchy is designed, and the definition criteria of the core display layer and related configuration items are clarified. The core display layer includes the pipeline network basic topology layer, the leakage identification result layer, and the uncertainty hot layer. The related configuration items include the data overlay method, layer transparency, and interactive query permissions. Based on the operational characteristics of pipeline leakage monitoring, visualization optimization rules were set up, including on-demand scaling of the topology layer, precise labeling of the result layer, and color gradient adaptation of the heat layer, to ensure that the display effect is compatible with practical operation. The visualization constraints, hierarchical setting standards, and optimization rules are integrated and processed. Simultaneously, the data calibration logs, regional quality control scores, quality control judgment results, and priority inspection suggestions are structured and formatted to generate a water supply network leakage monitoring results output system that includes multi-dimensional overlay visualization configuration data and standardized structured report templates.

3. A monitoring device for leakage in water supply networks, characterized in that, The device is used to execute executable instructions to perform the monitoring method for leakage in water supply networks as described in any one of claims 1 to 2.

4. An electronic device, characterized in that, include: First processor; The processor also includes a memory for storing executable instructions of the first processor, wherein the first processor is configured to execute the monitoring method for leakage in a water supply network as described in any one of claims 1 to 2 by executing the executable instructions.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the monitoring method for leakage in water supply networks as described in any one of claims 1 to 2.

Images