Modeling and Analysis Method for Urban Water Supply Network Optimization

Abstract

The present invention discloses a modeling and analysis method for optimizing urban water supply networks, belonging to the field of smart water technology. The method includes: establishing a comprehensive pipe network database, collecting hydraulic parameters through Internet of Things devices and integrating them with GIS topology data; constructing a dynamic hydraulic simulation model, coupling real-time data from pressure sensors with the EPANET model, and correcting the friction coefficient of pipe sections; designing a hybrid optimization algorithm architecture, using a dual-objective genetic algorithm for the renovation of old pipe networks and a constrained particle swarm algorithm for the planning of new pipe networks; performing multi-stage optimization calculations, combining geographic constraints to generate a feasible solution space for pipe diameters and iteratively outputting solutions that meet pressure thresholds; generating a pipe network optimization configuration plan, and outputting a project implementation list through spatial overlay analysis and valve control logic. This method solves the modeling distortion problem caused by multi-source data silos, overcomes the defect that static models are difficult to adapt to dynamic water demand, and significantly improves the economic efficiency and engineering feasibility of pipe network optimization plans.

Description

Technical Field

[0001] The present invention relates to the field of smart water technology, and in particular to a modeling and analysis method for optimizing urban water supply networks. Background Art

[0002] As critical infrastructure, the operational efficiency of urban water supply networks directly impacts livelihood security and resource sustainability. In the renovation of older urban water supply networks, leakage rates due to aging pipes are generally high, necessitating precise identification of high-loss sections and optimized replacement strategies. In the planning of new urban areas, a balance must be struck between construction costs and water supply reliability. Existing technologies primarily rely on manual experience to design renovation solutions, making them inadequate for the complex optimization requirements of large-scale pipe networks.

[0003] Existing technologies present numerous challenges, including deficiencies in data integration. Traditional modeling relies on static GIS data and manual inspection records, lacking an effective mechanism for integrating real-time IoT data (pressure, flow) with spatial data. In urban smart water projects, discrepancies between the SCADA system and the GIS coordinate system prevent many monitoring points from matching topological nodes, resulting in distorted hydraulic model parameters. Existing models lack adaptability, with mainstream hydraulic models using fixed friction coefficients that fail to account for dynamic changes caused by pipeline aging. Model optimization efficiency for pipe networks is low, and the renovation of older pipe networks requires processing discrete decision variables (such as pipe diameter combinations). Traditional enumeration methods require computational time exceeding 72 hours for projects with over 1,000 pipe sections.

[0004] Existing technologies have numerous limitations. At the data level, multi-source, heterogeneous data lacks unified temporal and spatial benchmarks, leading to insufficient utilization of real-time monitoring data. At the model level, static parameter settings fail to reflect dynamic operating conditions such as valve regulation and water usage fluctuations. At the collaborative optimization level, single-objective optimization algorithms struggle to coordinate economic efficiency (renovation costs) with reliability (pressure stability), resulting in poor solution feasibility. These shortcomings make it difficult for existing methods to support high-precision and efficient pipeline network optimization decisions. There is an urgent need to develop technical solutions that integrate the entire data, model, and optimization chain. Summary of the Invention

[0005] The present invention overcomes the modeling distortion problem caused by multi-source data islands in the existing technology, overcomes the defect that static models are difficult to adapt to dynamic water demand, and significantly improves the economy and engineering feasibility of the pipeline network optimization solution.

[0006] In order to achieve the above object, the present invention adopts the following scheme:

[0007] The modeling and analysis method for optimizing urban water supply networks includes the following steps:

[0008] S1. Acquire multi-source heterogeneous data, including: collecting hydraulic parameters of pressure sensors and flow meters in the target pipe network through IoT devices, and simultaneously acquiring pipe network topology data, pipe material attribute data, and user water consumption history data from the geographic information system; aligning the acquired multi-source heterogeneous data according to a unified spatiotemporal coordinate system and storing them in a relational database to establish a comprehensive pipe network database;

[0009] S2. Build a node-segment topology relationship matrix based on the network topology data. Couple the real-time monitoring data from the pressure sensors with the EPANET hydraulic calculation model. Use the nonlinear least squares method to correct the segment friction coefficient. Generate a three-dimensional network hydraulic model that includes pipe diameters, elevations, and valve status, and construct a dynamic hydraulic simulation model.

[0010] S3. To meet the needs of old pipeline network renovation, a fitness function with dual objectives of pipeline network renovation cost and leakage reduction was established, and a genetic algorithm with an elite retention strategy was used to prioritize pipe segment replacement. To meet the needs of new pipeline network planning, a particle swarm optimization algorithm was established with construction cost and hydraulic stability as constraints, resulting in a hybrid optimization algorithm architecture.

[0011] S4. Perform multi-stage optimization calculations, inputting pipe material property data from the integrated pipe network database into the dynamic hydraulic simulation model for initial hydraulic balance calculations. Based on the geographical constraints of the old pipe network reconstruction area, a feasible solution space for pipe diameters is generated. Using a hybrid optimization algorithm architecture, iterative calculations are performed to determine a pipe diameter combination that meets the node pressure threshold.

[0012] S5. Perform spatial overlay analysis on the optimized pipe diameter combination plan and the road direction data in the geographic information system, and combine it with the valve position data to generate an engineering implementation list containing pipe section replacement coordinates, new pipeline paths, and valve control instructions to generate an optimized pipeline network configuration plan.

[0013] Preferably, the establishment of the integrated pipe network database in step S1 specifically includes the following steps:

[0014] An IoT pressure sensor is installed at the intersection of the target pipe network, and an ultrasonic flow meter is installed at the user's water inlet. The collected hydraulic parameters are transmitted to the edge computing gateway in real time via the LoRaWAN communication protocol.

[0015] The pipe network topology data stored in the geographic information system is converted into a vector layer in the WGS84 coordinate system, the address information in the user water consumption history data is geocoded, and a spatial mapping relationship with the pipe network topology nodes is established;

[0016] For the real-time hydraulic parameter data stream collected by IoT devices, a cubic spline interpolation algorithm is used to reconstruct data segments with missing timestamps. This data is then spatially overlaid with the pipe network topology data in the geographic information system to generate a three-dimensional spatiotemporal data cube containing the spatial coordinates of the pipe network nodes, pipe segment elevation data, and the corresponding time series hydraulic parameters.

[0017] A calculation model for the theoretical value of pipeline pressure is established. When the deviation between the actual sensor value and the theoretical value exceeds the safety threshold corresponding to the pipeline pressure level, the data anomaly mark is automatically triggered, and the quartile method is used to clean the abnormal data, retaining the valid data within the upper and lower quartiles.

[0018] The registered network topology data is stored in a PostGIS spatial database, and the time series hydraulic parameter data is stored in a temporal database. The two databases are then linked and indexed by establishing a unique identification code for the network nodes, which is generated by combining the network number, administrative division code, and spatial coordinate hash value.

[0019] Set a weekly incremental update cycle for geographic information system data, set a sliding average filter processing for the real-time data stream of the Internet of Things with a 5-minute time window, and automatically trigger the re-registration calculation of the pipe network topology data and hydraulic parameters when the user water consumption record data is updated.

[0020] Preferably, the following method is used to generate the pipe network optimization configuration scheme in step S5:

[0021] The pipe segment replacement coordinates in the optimized pipe diameter combination scheme were converted into a Geographic Information System (GIS) Shapefile format. A buffer zone analysis was performed with the road centerline vector layer, setting a construction safety distance of 1.2 meters outside the road red line. Pipeline paths with an overlap of more than 30% with existing underground pipeline corridors were eliminated. A node-valve association matrix was established based on the valve location data in the pipeline network topology data. A breadth-first search algorithm was used to determine the control influence range of each valve, generating a control instruction tree diagram containing the valve opening and closing sequence and timing.

[0022] Extract pipe attribute data from the integrated pipe network database and, combined with the operating radius parameters of the hoisting machinery at the construction site, perform curvature radius verification on pipe sections with a diameter ≥ 600mm. Insert transition elbow nodes when the turning angle exceeds the minimum turning radius of the excavator. Associate the newly created pipeline path after spatial overlay analysis with the construction feasibility verification results, sort them according to the unique identification code of the pipe network nodes, and output a structured table containing the pipe section number, starting point coordinates, end point coordinates, pipe diameter specifications, valve control sequence, and construction machinery model.

[0023] When the road direction data changes, the spatial overlay analysis is recalculated, and the coordinate data of the affected pipe sections is automatically updated through version number comparison. A constraint satisfaction algorithm is used to verify the spatiotemporal conflicts between pipe section replacement and new paths in the project implementation list. When two construction tasks are detected at the same coordinate point, the execution time of non-urgent tasks is automatically postponed and a conflict warning report is generated. The project implementation list and the valve control instruction tree diagram are merged and encoded to generate an XML format document, and data integrity is ensured through digital signatures.

[0024] Preferably, in step S2, the dynamic hydraulic simulation model is constructed using the following method:

[0025] The coordinates of the pipe segment endpoints are extracted based on the pipe network topology data of the geographic information system, and a 0-1 association matrix is ​​constructed with the pipe network nodes as row vectors and the adjacent pipe segments as column vectors. The flow direction of the pipe segment is represented by the positive and negative values ​​determined by the flow meter data.

[0026] The pressure data monitored in real time by the pressure sensor is divided into data slices at a 15-minute time granularity. The data is then injected into the hydraulic calculation model through the EPANET Toolkit API interface to establish a dynamic mapping relationship between the measured pressure values ​​and the calculated values.

[0027] To address the issue of error in setting the initial value of the friction coefficient of a pipe segment, the Levenberg-Marquardt algorithm was used as an iterative optimizer for the nonlinear least squares method. The root mean square error between the measured pressure data slice and the EPANET calculated value was used as the objective function to simultaneously optimize the friction coefficient combination of three adjacent pipe segments to complete the friction coefficient correction.

[0028] The corrected friction coefficient was substituted into the EPANET model for transient simulation. The flow velocity and head loss data of each pipe section at different valve openings were extracted. The hydraulic state tensor of the pipe network was constructed with the spatial coordinates as the XY axis and the time series as the Z axis to generate a three-dimensional hydraulic characteristic field.

[0029] During the daily low water consumption period, the pressure regulating valve of the target pipeline network was closed, and the actual pressure distribution data was obtained through the SCADA system. The Kolmogorov-Smirnov test was performed against the simulation results of the three-dimensional hydraulic characteristic field. When the P value was less than 0.05, the friction coefficient was re-adjusted.

[0030] Preferably, the process of collaborative correction of the friction coefficient includes:

[0031] A dynamic water consumption pattern library was constructed. A long-short-term memory neural network was used to train historical user water consumption data to generate a water consumption prediction model that incorporates weekday, holiday, and seasonal variations. The model then outputs a water consumption curve for each node over the next 24 hours. The water consumption prediction curve was discretized into node water demand boundary conditions at 15-minute intervals. A dynamic weighting factor was introduced into the iterative process of the Levenberg-Marquardt algorithm to prioritize matching the measured pressure data slices during peak water consumption periods.

[0032] A spatial constraint on the friction coefficient is established, and the friction coefficient variation range of pipe sections made of different materials is determined based on the pipe property data. When the optimized friction coefficient exceeds the expected friction coefficient range of the pipe, an abnormal pipe state warning is triggered. A dual-threshold convergence criterion is set. When the root mean square error change of three consecutive iterations is less than 0.01m and the friction coefficient change is less than 0.0001, the optimization is terminated and the optimal friction coefficient combination is retained. The final corrected friction coefficient is written into the basic input file of the EPANET model, and the pipe property data in the integrated pipe network database is updated at the same time, and the version number and checksum are recorded.

[0033] Preferably, performing the multi-stage optimization calculation in step S4 includes the following steps:

[0034] Road width data, underground integrated pipeline corridor distribution data, and existing building coordinate data for the old pipeline network renovation area were extracted to establish a two-dimensional rasterized constraint matrix containing pipeline segment depth limits, horizontal offset tolerances, and construction restricted areas, resulting in a geographic constraint matrix. Based on the pipe material attribute data in the integrated pipeline network database, a set of pipe diameter specifications that met construction standards was screened. A Cartesian product operation was performed on the geographic constraint matrix, eliminating pipe diameter-path combinations with a spacing of less than 1.5 meters from underground power pipelines to construct a feasible solution space for pipe diameters.

[0035] Implementing hybrid algorithm parallel computing, embedding the velocity update formula of the particle swarm optimization algorithm into the genetic algorithm with elite retention strategy, dividing the feasible solution space of pipe diameter into four subspaces, calculating the initial fitness value of each subspace respectively, and retaining the solution set with the top 10% fitness value;

[0036] A pressure threshold check is performed, and the node pressure simulation values ​​output by the dynamic hydraulic simulation model are compared node by node with the minimum service pressure values ​​in the urban water supply specifications. When more than 5% of the nodes fail to meet the standards within three consecutive iteration cycles, the feasible solution space for pipe diameters is expanded. The pipe diameter combination schemes that pass the pressure threshold check are arranged in ascending order according to the renovation cost, and the top three candidate schemes and their corresponding hydraulic stability indicators are output to obtain the optimized solution set.

[0037] Preferably, constructing the feasible solution space of pipe diameters includes the following steps:

[0038] A multi-objective conflict resolution model was established, using Pareto frontier analysis to address the trade-off between renovation cost and hydraulic stability in pipe diameter selection. The effective solution screening condition was set at a rate where the renovation cost increase did not exceed 1.5 times the hydraulic stability improvement rate. A random forest classifier was trained using historical engineering data to predict the constructibility probability of each pipe diameter-path combination, automatically eliminating any combination with a predicted value below 0.7.

[0039] A dynamic solution space index is generated. The feasible solution space of pipe diameters is coded into a B+ tree index structure according to administrative divisions. The solution space data is retrieved and updated in combination with the unique identification codes of the pipe network nodes. The principal component analysis method is used to extract the key feature vectors of the pipe diameter combination scheme, reducing the solution space dimension from the original number of pipe sections n to the log2(n) level. 5% of feasible solutions are randomly selected for on-site inspection and verification. When the verification failure rate exceeds 20%, the geographic constraint matrix is ​​regenerated.

[0040] Preferably, the implementation of hybrid algorithm parallel computing includes the following specific steps:

[0041] A distributed computing architecture was constructed, using the MPI parallel computing framework to allocate the four subspaces to different computing nodes. Each node independently ran the hybrid optimization algorithm and synchronized the optimal solution data every 20 minutes. The crossover probability of the genetic algorithm was dynamically adjusted based on the degree of discreteness of the subspace solution set, reducing the crossover probability from 0.8 to 0.6 when the solution set standard deviation exceeded a set threshold.

[0042] The rate of change of the fitness value of each subspace is monitored. If the rate of change is less than 0.1% within three consecutive calculation cycles, the iterative calculation of the subspace is terminated early. The optimal solutions of the four subspaces are non-dominated sorted, and the congestion comparison operator is used to select evenly distributed Pareto optimal solutions to form the final optimization solution set. When the solution set of a subspace is empty due to a geographical constraint conflict, the buried depth limit of the pipe section in that area is relaxed by 10% and the calculation process is restarted.

[0043] Preferably, constructing the dynamic hydraulic simulation model in step S2 includes the following specific steps:

[0044] A dynamic data fusion interface was established to convert the pipe network topology data from the geographic information system into a GraphML format with topology verification. Pipeline connection relationships were stored in an adjacency table, and node elevation data was interpolated and completed using a digital elevation model. Real-time data coupling was implemented to perform Kalman filtering on the raw pressure data stream collected by the pressure sensor to eliminate transient fluctuations caused by water hammer, generating smoothed pressure time series data. Hydraulic model boundary conditions were then injected through the EPANET INP file interface.

[0045] Perform multi-parameter joint correction, adopt nonlinear least square method with constraints, use pipe segment friction coefficient and node foundation water demand as synchronous optimization variables, and set the physical constraint range of the initial value of pipeline friction coefficient;

[0046] A three-dimensional hydraulic response surface was constructed. The transient analysis module of EPANET was used to simulate the step-wise change of valve opening from 10% to 90%. The gradient matrix of the flow velocity in each pipe section as it changes with valve state was recorded, generating a three-dimensional hydraulic feature space consisting of pipe diameter, elevation, and valve opening.

[0047] During the stable water consumption period late at night every day, 50% of the regulating valves in the target area are closed, and the actual pressure distribution data is collected through the SCADA system. Dynamic time regularization is performed with the simulation value of the three-dimensional hydraulic characteristic space, and when the average absolute error exceeds 0.05MPa, the iterative correction of the friction coefficient is triggered.

[0048] Preferably, performing multi-parameter joint calibration comprises the following steps:

[0049] A dynamic decomposition model of water consumption patterns was constructed. The variational mode decomposition algorithm was used to decompose the user water consumption history data into six intrinsic mode function components, and the time-frequency characteristics of the daily cycle component and the random fluctuation component were extracted.

[0050] A time-varying weight factor was introduced into the objective function of the nonlinear least squares method, increasing the weight of the measured data for the period corresponding to the daily cycle component to 2.5 times that of the random fluctuation component. Based on the service life data in the pipe property database, an upper limit of 0.022 was set for the optimization of the friction coefficient for cast iron pipe sections over 20 years old, and a lower limit of 0.009 was set for polyethylene pipe sections replaced within 5 years. The MPI parallel computing framework was used to divide the pipe network into four hydraulic zones, with each zone independently running the Levenberg-Marquardt optimizer, achieving global convergence through the exchange of boundary node pressure data. A parameter credibility report was generated, recording the correlation coefficient between the change in the friction coefficient and the pressure error reduction rate in each iteration. When the absolute value of the correlation coefficient was less than 0.7, the pipe section was marked as a potential abnormal section and its spatial coordinates were output.

[0051] The present invention has at least the following beneficial effects: (1) Through the multi-source spatiotemporal data fusion mechanism, the data island problem is solved, the modeling data integrity is significantly improved, the dynamic hydraulic model is coupled with the measured data and parameter correction, and the friction coefficient calibration error is reduced; (2) the dual mechanism of data quality verification and model verification is combined to eliminate the interference of abnormal data on the optimization results and improve the stability of the scheme; (3) spatial overlay analysis and construction parameter integration are used to reduce the scheme change rate during the project implementation stage; (4) dynamic weight factors are coordinated with the parallel optimization framework to greatly compress the time consumption of large-scale pipeline network calculations; (5) solution space compression and conflict adaptive repair are used to improve the convergence speed of the optimization algorithm while ensuring the quality of the scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a principle flow chart of the modeling and analysis method for optimizing urban water supply networks provided by the present invention. DETAILED DESCRIPTION

[0053] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0054] like Figure 1 As shown, the modeling and analysis method for optimizing urban water supply networks provided by the present invention includes the following steps:

[0055] S1. Acquire multi-source heterogeneous data, including: collecting hydraulic parameters of pressure sensors and flow meters in the target pipe network through IoT devices, and simultaneously acquiring pipe network topology data, pipe material attribute data, and user water consumption history data in the geographic information system; aligning the acquired multi-source heterogeneous data according to a unified spatiotemporal coordinate system and storing them in a relational database to establish a comprehensive pipe network database.

[0056] The IoT pressure sensors (e.g., piezoresistive sensors) deployed at key nodes of the target pipe network (e.g., pipe junctions, pump station outlets) and ultrasonic flow meters at the user’s water inlet end collect pipeline pressure values ​​(unit: MPa) and instantaneous flow values ​​(unit: m³) in real time at a sampling frequency of 1-5 minutes. 3 / h). Simultaneously extract vector-formatted pipeline network topology data (including pipe segment start / end coordinates and connection relationships), pipe attribute data (such as pipe material type, inner diameter, roughness coefficient, and service life), and user water consumption history data (daily / hourly water consumption series) from the geographic information system. This multi-source heterogeneous data is then unified into a temporal and spatial benchmark: GIS topology data is converted into point and line layers in the WGS84 coordinate system; text addresses in water consumption records are geocoded; and real-time IoT data streams are linked to spatial data through timestamp alignment. Finally, the aligned data is stored in a relational database (such as PostgreSQL), forming a comprehensive data table containing spatial coordinate fields, timestamp fields, hydraulic parameter fields, and pipe attribute fields.

[0057] S2. Based on the pipeline network topology data, a node-pipeline segment topology relationship matrix is ​​established. The real-time monitoring data of the pressure sensor is coupled with the EPANET hydraulic calculation model. The pipe segment friction coefficient is corrected using the nonlinear least squares method to generate a three-dimensional pipeline network hydraulic model that includes pipe diameter, elevation, and valve status, and construct a dynamic hydraulic simulation model.

[0058] Based on the pipeline network topology data, a node-link topology relationship matrix is ​​established. This matrix is ​​a sparse matrix, with row indices corresponding to pipeline node numbers and column indices corresponding to pipeline link numbers. A matrix element value of 1 indicates a connection between a node and a link, while a value of -1 indicates the flow is in the opposite direction. Real-time pressure data streams monitored by pressure sensors (sliced ​​at 15-minute intervals) are dynamically injected into the model boundary conditions via the EPANET hydraulic calculation engine's programming interface (such as the Toolkit API). To address the issue of error in the specification of link friction coefficients (such as the C value in the Hazen-Williams equation) in traditional models, a nonlinear least squares method is used for collaborative correction. With the root mean square error (RMS) between the measured pressure slices and the EPANET-calculated values ​​as the optimization objective, the Levenberg-Marquardt algorithm is used to iteratively adjust the friction coefficient combinations of adjacent links (optimizing 3-5 related links at a time). These corrected parameters are then fed into the EPANET transient simulation module, which outputs a hydraulic model with three-dimensional features: spatial dimensions (node ​​coordinates on the X and Y axes), attribute dimensions (Z-axis elevation and pipe diameter), and dynamic dimensions (flow velocity and pressure distribution at different valve openings).

[0059] S3. To meet the needs of old pipeline network renovation, a fitness function with pipeline network renovation cost and leakage water reduction as dual objectives is established, and a genetic algorithm with an elite retention strategy is used to prioritize pipe section replacement. To meet the needs of new pipeline network planning, a particle swarm optimization algorithm with construction cost and hydraulic stability as constraints is established, resulting in a hybrid optimization algorithm architecture.

[0060] For the old pipeline network renovation scenario, a dual-objective fitness function was constructed: Objective 1 is the cost of replacing a pipe segment (including pipe material and construction costs), and Objective 2 is the reduction in water leakage (based on simulated leakage values ​​calculated by the hydraulic model). A genetic algorithm with an elite retention strategy was used for optimization: the population was initialized with the pipeline segment combinations to be renovated, two-point crossover was used for crossover, and mutation was performed to randomly replace pipe diameters with a probability of 0.1-0.3. The top 10% elite solutions in each generation were retained. For the new pipeline network planning scenario, a particle swarm optimization algorithm was developed with construction cost (total pipe material price + construction cost) and hydraulic stability (variance of node pressure fluctuations) as constraints. The particle position vector represents the pipe diameter combination, and the velocity update formula incorporates an inertia weighting factor (ranging from 0.4 to 0.9). Both algorithms share the results of the dynamic hydraulic model. High-priority pipeline segments (e.g., those with a leakage risk greater than 5%) identified by the old pipeline network optimization output serve as constraints for the initial solution space of the new pipeline network planning.

[0061] S4. Perform multi-stage optimization calculations, input pipe property data from the integrated pipe network database into the dynamic hydraulic simulation model for initial hydraulic balance calculations, generate a feasible solution space for pipe diameters based on the geographical constraints of the old pipe network renovation area, and iteratively calculate through a hybrid optimization algorithm architecture to obtain a pipe diameter combination solution that meets the node pressure threshold.

[0062] First, pipe material property data (such as pipe diameter and friction coefficient) from the integrated pipe network database was input into the dynamic hydraulic model to perform an initial hydraulic balance calculation (simulating the flow and pressure distribution in the original state). A feasible solution space for pipe diameters was generated based on the geographic constraints of the existing pipe network renovation area. This approach involved extracting road width data (required to be ≥4 meters for construction), underground integrated pipe corridor locations (horizontal spacing from power lines ≥1.5 meters), and building coordinates (construction restricted zone radius ≥3 meters) to form a two-dimensional raster constraint matrix. Pipe diameters that met the GB / T 50268 standard (e.g., ductile iron pipes DN100-DN600) were selected, and a Cartesian product operation was performed to eliminate pipe diameter-path combinations that violated the geographic constraints. Iterative calculations are performed using a hybrid optimization algorithm architecture: a genetic algorithm performs selection, crossover, and mutation within the feasible solution space, while a particle swarm optimization optimizes the subspace solution set (dividing the solution into 4-6 subregions). Each iteration uses a dynamic hydraulic model to verify that node pressures meet urban water supply regulations (e.g., minimum service pressure ≥ 0.28 MPa). If more than 5% of node pressures fail to meet the standard over three consecutive iterations, the feasible solution space for pipe diameters is expanded (e.g., adding a DN700 option). The final output is a list of candidate solutions (the top three) ranked in ascending order of renovation cost, along with hydraulic stability indicators (e.g., pressure differential ≤ 0.15 MPa).

[0063] S5. Perform spatial overlay analysis on the optimized pipe diameter combination plan and the road direction data in the geographic information system, and combine it with the valve position data to generate an engineering implementation list containing pipe section replacement coordinates, new pipeline paths, and valve control instructions to generate an optimized pipeline network configuration plan.

[0064] The optimized pipe diameter combination (including the start / end coordinates of the pipe segments) was spatially overlaid with the road centerline vector layer in the geographic information system. A buffer zone was established, with a construction safety distance of 1.0-1.5 meters outside the road red line. Pipeline routes with an overlap of more than 30% with existing underground pipeline corridors (such as thermal pipelines) were automatically eliminated. A node-valve association matrix (identifying the influence range of valve control) was established using valve location data (extracted from the pipeline network topology data). A breadth-first search algorithm was used to generate a valve control instruction tree (for example, closing the upstream valve before the downstream valve in the event of a pipe burst). Construction feasibility parameters were associated: For pipe segments with a diameter ≥600 mm, the path curvature was verified based on the minimum turning radius of the lifting machinery (e.g., ≥8 meters for excavators). Transition elbow node coordinates were automatically inserted when the turning angle exceeded 45°. Finally, a structured engineering implementation list is generated: sorted by the unique identification code of the pipeline network node, including the fields "pipe section number - starting point coordinates - end point coordinates - pipe diameter specification - valve operation sequence - construction machinery model" (for example, a DN400 pipe section uses a 200-ton crane), output in XML format and digitally signed to ensure data integrity.

[0065] This method solves the parameter distortion problem caused by data silos in traditional modeling through a multi-source spatiotemporal data fusion mechanism. It dynamically couples measured data with hydraulic models to significantly improve the calibration accuracy of key parameters such as the friction coefficient. A hybrid optimization algorithm collaboratively processes new and old pipeline network scenarios, reducing renovation costs while ensuring water supply reliability. Spatial analysis combined with construction constraint verification significantly reduces the risk of solution changes during the project implementation phase.

[0066] In another technical solution, the establishment of the integrated pipe network database in step S1 specifically includes the following steps:

[0067] An IoT pressure sensor is installed at the intersection of the target pipe network, and an ultrasonic flow meter is installed at the user's water inlet. The collected hydraulic parameters are transmitted to the edge computing gateway in real time via the LoRaWAN communication protocol.

[0068] The pipe network topology data stored in the geographic information system is converted into a vector layer in the WGS84 coordinate system, the address information in the user water consumption history data is geocoded, and a spatial mapping relationship with the pipe network topology nodes is established;

[0069] For the real-time hydraulic parameter data stream collected by IoT devices, a cubic spline interpolation algorithm is used to reconstruct data segments with missing timestamps. This data is then spatially overlaid with the pipe network topology data in the geographic information system to generate a three-dimensional spatiotemporal data cube containing the spatial coordinates of the pipe network nodes, pipe segment elevation data, and the corresponding time series hydraulic parameters.

[0070] A calculation model for the theoretical value of pipeline pressure is established. When the deviation between the actual sensor value and the theoretical value exceeds the safety threshold corresponding to the pipeline pressure level, the data anomaly mark is automatically triggered, and the quartile method is used to clean the abnormal data, retaining the valid data within the upper and lower quartiles.

[0071] The registered network topology data is stored in a PostGIS spatial database, and the time series hydraulic parameter data is stored in a temporal database. The two databases are then linked and indexed by establishing a unique identification code for the network nodes, which is generated by combining the network number, administrative division code, and spatial coordinate hash value.

[0072] Set a weekly incremental update cycle for geographic information system data, set a sliding average filter processing for the real-time data stream of the Internet of Things with a 5-minute time window, and automatically trigger the re-registration calculation of the pipe network topology data and hydraulic parameters when the user water consumption record data is updated.

[0073] Deploy IoT pressure sensors (e.g., piezoresistive sensors with optional accuracy of 0.5%FS or 1.0%FS) at the intersection of pipes in the target pipe network (e.g., tees, elbows, and other hydraulically sensitive points), and install ultrasonic flow meters (with an accuracy of 1.0%R) at the user's water inlet (e.g., water meter wells). The sensors transmit pressure values ​​(in MPa) and flow values ​​(in m) via the LoRaWAN communication protocol (transmission distance 2-5 km, power consumption 10 mW). 3 Data is collected and transmitted in real time to an edge computing gateway (such as the Huawei AR502H) at a rate of 100 / h. The gateway adds a timestamp (accurate to the millisecond) and device ID to the raw data, encapsulates it in JSON format, and uploads it to the cloud platform via a 4G network. This process addresses the timeliness issues of traditional manual data collection and provides a high-frequency data foundation for dynamic modeling (the sampling interval is typically 5 minutes, adjustable from 1 to 15 minutes).

[0074] Pipeline network topology data (Shapefile format) in the Geographic Information System (GIS) was converted into a point-line layer in the WGS84 coordinate system. Node elevations were interpolated using a Digital Elevation Model (DEM) (with an accuracy of ±0.1m). Text addresses in user water usage records (e.g., "XX Road, XX No.") were geocoded to generate latitude and longitude coordinates and mapped to the nearest pipeline node. For missing values ​​in the IoT data stream (e.g., signal interruptions), a cubic spline interpolation algorithm was used to reconstruct the complete time series (e.g., to fill in three consecutive missing data points). During the data quality verification phase, theoretical pressure values ​​were calculated based on pipeline hydraulic equations (e.g., the Bernoulli equation). Any deviation from the measured value exceeding the pipeline pressure safety threshold (e.g., ≥0.8 MPa for PVC pipes and ≥1.2 MPa for cast iron pipes) was flagged as an anomaly and cleaned using the quartile method (data outside the range of Q1 - 1.5 IQR to Q3 + 1.5 IQR) was eliminated.

[0075] The registered spatial data is stored in a PostGIS database (with fields including node ID, coordinates, and elevation), while time series data is stored in a time series database (such as InfluxDB, which stores pressure / flow values ​​at a 5-minute granularity). Cross-database indexing is achieved through the unique identifier of the pipeline network node (consisting of 16 characters: a 6-digit administrative division code + a 4-digit pipeline network number + a 6-digit coordinate hash value). The data update mechanism includes weekly incremental updates of GIS data (only updated areas are synchronized), a sliding average filter (with a 5-minute window, adjustable to 3-10 minutes) for noise reduction on the IoT data stream, and automatic re-registration of topological data with real-time data when user water consumption is updated (e.g., topological changes caused by the addition of new user nodes).

[0076] Achieve high-precision spatiotemporal alignment of multi-source data and eliminate heterogeneous data fusion errors; the dynamic data quality monitoring mechanism significantly improves the reliability of modeling data; and the multimodal storage structure supports rapid retrieval and analysis of large-scale pipeline network data.

[0077] In another technical solution, the generation of the pipe network optimization configuration solution in step S5 uses the following method:

[0078] The pipe segment replacement coordinates in the optimized pipe diameter combination scheme were converted into a Geographic Information System (GIS) Shapefile format. A buffer zone analysis was performed with the road centerline vector layer, setting a construction safety distance of 1.2 meters outside the road red line. Pipeline paths with an overlap of more than 30% with existing underground pipeline corridors were eliminated. A node-valve association matrix was established based on the valve location data in the pipeline network topology data. A breadth-first search algorithm was used to determine the control influence range of each valve, generating a control instruction tree diagram containing the valve opening and closing sequence and timing.

[0079] Extract pipe attribute data from the integrated pipe network database and, combined with the operating radius parameters of the hoisting machinery at the construction site, perform curvature radius verification on pipe sections with a diameter ≥ 600mm. Insert transition elbow nodes when the turning angle exceeds the minimum turning radius of the excavator. Associate the newly created pipeline path after spatial overlay analysis with the construction feasibility verification results, sort them according to the unique identification code of the pipe network nodes, and output a structured table containing the pipe section number, starting point coordinates, end point coordinates, pipe diameter specifications, valve control sequence, and construction machinery model.

[0080] When the road direction data changes, the spatial overlay analysis is recalculated, and the coordinate data of the affected pipe sections is automatically updated through version number comparison. A constraint satisfaction algorithm is used to verify the spatiotemporal conflicts between pipe section replacement and new paths in the project implementation list. When two construction tasks are detected at the same coordinate point, the execution time of non-urgent tasks is automatically postponed and a conflict warning report is generated. The project implementation list and the valve control instruction tree diagram are merged and encoded to generate an XML format document, and data integrity is ensured through digital signatures.

[0081] The coordinates of the pipe segments (WGS84 latitude and longitude) in the optimized pipe diameter combination scheme are converted into a GIS Shapefile format and buffered against the road centerline layer (the buffer radius is set to 1.2 meters, adjustable to 1.0-1.5 meters depending on the road grade). The system automatically calculates the spatial overlap between the newly constructed pipeline route and existing underground pipeline corridors (such as gas pipelines) (based on vector surface intersection analysis). If the overlap area exceeds 30% (the threshold can be set to 20%-40%), the route is eliminated. A node-valve association matrix (a matrix element of 1 indicates that the valve controls the node) is constructed using valve location data. A breadth-first search (BFS) algorithm is used to traverse the pipeline network topology backward from the burst point to generate a valve closure sequence tree (for example, prioritizing the closure of upstream main pipeline valves and delaying the closure of downstream branch pipeline valves).

[0082] Pipe material properties (e.g., DN600 ductile iron pipe weighs 12.5 tons / m) are extracted from the integrated pipeline network database and combined with construction site parameters (e.g., a 200-ton crane operating radius ≥ 8 meters) to perform path curvature verification. The system calculates the turning angle of the pipe segment (calculating the curvature radius using three coordinates). When the turning radius is less than the minimum turning radius of the machine (e.g., an excavator with a minimum turning radius of 6 meters), an elbow node is automatically inserted (adding coordinate points to ensure the angle is ≤ 30°). A structured project list is generated: a table is generated by sorting the unique identifiers of the pipeline network nodes, with fields including starting coordinates (longitude, latitude), end coordinates, pipe diameter (e.g., DN400), valve operation sequence (e.g., "V001 valve closes at T+5 minutes"), and construction machine model (e.g., "XCMG 200-ton crane").

[0083] When GIS road data changes (such as the addition of a new median strip), the system locates the affected pipe sections through version number comparison (spatial topology analysis) and automatically triggers a coordinate update (recalculating the path if the offset exceeds 0.5 meters). A constraint satisfaction algorithm (CSP) is used to detect construction spatiotemporal conflicts: construction tasks are defined as variables (e.g., "Pipe section A replacement: coordinates (X1, Y1), time T1-T2"), with the constraint that only one task is allowed at the same coordinate point within 24 hours. When a conflict is detected (e.g., coordinate point P is assigned both excavation and lifting tasks during time T1), the system automatically postpones non-urgent tasks (the delay can be set to 12-48 hours) and generates an early warning report. The final solution is encoded as an XML document (including a digital signature checksum) that complies with the ISO 55000 standard.

[0084] Achieve deep integration of spatial constraints and construction parameters, significantly reducing project implementation risks; a dynamic conflict detection mechanism effectively avoids construction resource scheduling conflicts; and standardized output formats significantly improve cross-departmental collaboration efficiency.

[0085] In another technical solution, in step S2, the dynamic hydraulic simulation model is constructed using the following method:

[0086] The coordinates of the pipe segment endpoints are extracted based on the pipe network topology data of the geographic information system, and a 0-1 association matrix is ​​constructed with the pipe network nodes as row vectors and the adjacent pipe segments as column vectors. The flow direction of the pipe segment is represented by the positive and negative values ​​determined by the flow meter data.

[0087] The pressure data monitored in real time by the pressure sensor is divided into data slices at a 15-minute time granularity. The data is then injected into the hydraulic calculation model through the EPANET Toolkit API interface to establish a dynamic mapping relationship between the measured pressure values ​​and the calculated values.

[0088] To address the issue of error in setting the initial value of the friction coefficient of a pipe segment, the Levenberg-Marquardt algorithm was used as an iterative optimizer for the nonlinear least squares method. The root mean square error between the measured pressure data slice and the EPANET calculated value was used as the objective function to simultaneously optimize the friction coefficient combination of three adjacent pipe segments to complete the friction coefficient correction.

[0089] The corrected friction coefficient was substituted into the EPANET model for transient simulation. The flow velocity and head loss data of each pipe section at different valve openings were extracted. The hydraulic state tensor of the pipe network was constructed with the spatial coordinates as the XY axis and the time series as the Z axis to generate a three-dimensional hydraulic characteristic field.

[0090] During the daily low water consumption period, the pressure regulating valve of the target pipeline network was closed, and the actual pressure distribution data was obtained through the SCADA system. The Kolmogorov-Smirnov test was performed against the simulation results of the three-dimensional hydraulic characteristic field. When the P value was less than 0.05, the friction coefficient was re-adjusted.

[0091] Based on the pipeline network topology data from the geographic information system, the coordinates (latitude and longitude) of the pipeline segment endpoints are extracted to construct a node-segment topology relationship matrix. The row index is the pipeline node (e.g., nodes N001-N500), and the column index is the adjacent pipeline segment (e.g., pipeline segments P100-P300). A matrix element value of 1 indicates a forward connection, and -1 indicates a reverse connection (flow direction is determined by the sign of the flow meter data). Raw pressure data from the real-time pressure sensor (which may contain water hammer noise) is sliced ​​into data slices at a fixed time granularity (typically 15 minutes, adjustable from 5 to 30 minutes). Hydraulic model boundary conditions (e.g., node water demand) are then injected through the EPANET Toolkit API. This process establishes a dynamic mapping between measured and calculated pressures: when the measured pressure at a node consistently exceeds the simulated value by more than 0.05 MPa, the associated pipeline segment is automatically marked as an area for friction coefficient optimization.

[0092] To address the initial error in the friction coefficient of a pipe segment (e.g., the default value for cast iron pipes is 0.018±0.005), a nonlinear optimization algorithm (Levenberg-Marquardt algorithm) was used. Three adjacent pipe segments were used as optimization units (to reduce the risk of local optima), the objective function was the root mean square error (RMSE) between the measured pressure slice and the EPANET-calculated value, and the iteration step size was set to 0.001-0.005. The corrected friction coefficient was then applied to the EPANET transient module to simulate flow conditions under step-variable valve openings (10% / 30% / 50% / 70% / 90%). Flow velocity and head loss data for each pipe segment were extracted, and a three-dimensional hydraulic characteristic field was constructed: the X and Y axes represent spatial coordinates (in meters), and the Z axis represents time series (in hours). Attribute values ​​consist of a pressure / flow matrix (e.g., 100 nodes × 24 hours × 5 valve states).

[0093] During the daily low water consumption period (e.g., 02:00-04:00), 50% of the pressure regulating valves (not selected from the main pipelines) in the target pipeline network are closed. Actual pressure distribution data is collected through the SCADA system. The measured data is then compared with the simulated values ​​of the three-dimensional hydraulic characteristic field through the Kolmogorov-Smirnov test (KS test). The cumulative probability difference of the pressure distribution function is calculated. A P value of less than 0.05 (the significance threshold can be set between 0.01 and 0.1) indicates that the model is inaccurate. This triggers the iterative correction process: Returning to the friction coefficient optimization step, the parameters of the top 10% of pipeline sections with the largest differences in the KS test are adjusted until the P value meets the target or the maximum number of iterations (e.g., 20) is reached.

[0094] The dynamic data coupling mechanism significantly improves the real-time adaptability of the hydraulic model, the multi-parameter collaborative correction effectively reduces the friction coefficient calibration error, and the statistical test-driven iteration mechanism ensures the long-term reliability of the model.

[0095] The process of coordinated correction of friction coefficient includes:

[0096] A dynamic water consumption pattern library was constructed. A long-short-term memory neural network was used to train historical user water consumption data to generate a water consumption prediction model that incorporates weekday, holiday, and seasonal variations. The model then outputs a water consumption curve for each node over the next 24 hours. The water consumption prediction curve was discretized into node water demand boundary conditions at 15-minute intervals. A dynamic weighting factor was introduced into the iterative process of the Levenberg-Marquardt algorithm to prioritize matching the measured pressure data slices during peak water consumption periods.

[0097] A spatial constraint on the friction coefficient is established, and the friction coefficient variation range of pipe sections made of different materials is determined based on the pipe property data. When the optimized friction coefficient exceeds the expected friction coefficient range of the pipe, an abnormal pipe state warning is triggered. A dual-threshold convergence criterion is set. When the root mean square error change of three consecutive iterations is less than 0.01m and the friction coefficient change is less than 0.0001, the optimization is terminated and the optimal friction coefficient combination is retained. The final corrected friction coefficient is written into the basic input file of the EPANET model, and the pipe property data in the integrated pipe network database is updated at the same time, and the version number and checksum are recorded.

[0098] Constructing a dynamic water consumption pattern library: A long short-term memory (LSTM) neural network is trained on historical user water consumption data (spanning ≥ 1 year). Input features include date type (weekday / holiday), temperature (obtained from the Meteorological Bureau API), and season identifier (spring / summer / autumn / winter). The library outputs a 24-hour water consumption curve for each node (with a 15-minute temporal resolution), discretizing it into EPANET node water demand boundary conditions (in L / s). A dynamic weighting factor is introduced into the LM algorithm iterations: a weight of 2.0-3.0 (adjustable) is assigned to the measured pressure data slices during peak water consumption periods (e.g., 07:00-09:00), while the weight is reset to 1.0 during off-peak periods. This ensures that the optimization process more closely reflects actual load fluctuations.

[0099] Set the physical constraint range of the friction coefficient based on the pipe property data:

[0100] Polyvinyl chloride (PVC) pipes: 0.008-0.012 (lower value for new pipes); cast iron pipes: 0.015-0.020 (service life < 15 years); the upper limit for old cast iron pipes is relaxed to 0.022;

[0101] When the optimized value exceeds the limit, an alarm for pipe segment anomalies (such as internal wall scaling) is automatically triggered. The MPI parallel framework is used to divide the pipe network into four hydraulic zones (grouped by topological connectivity), with each zone independently running the LM optimizer (the number of computation nodes equals the number of zones). A boundary data exchange mechanism synchronizes pressure data at zone boundary nodes every five iterations (for example, forced synchronization when the node pressure difference exceeds 0.03 MPa) to avoid local optimization conflicts.

[0102] Set the dual-threshold convergence criterion:

[0103] Primary threshold: RMSE change over three consecutive iterations < 0.01m; Secondary threshold: Friction coefficient change < 0.0001;

[0104] Optimization terminates when any threshold is met. A parameter credibility report is output: This reports the correlation coefficient (Pearson coefficient) between the change in friction coefficient and the pressure error drop rate for each pipe segment. When |r| < 0.7, the segment is marked as a "potentially abnormal segment" (possibly containing an unmonitored leak). Finally, the corrected parameters are written to the EPANET INP basic file (ASCII format). The pipe material attribute table in the integrated pipe network database is simultaneously updated with the new fields "Optimization Version Number Vxx" and "MD5 Checksum" (e.g., e99a18c4).

[0105] In another technical solution, the multi-stage optimization calculation in step S4 includes the following steps:

[0106] Road width data, underground integrated pipeline corridor distribution data, and existing building coordinate data for the old pipeline network renovation area were extracted to establish a two-dimensional rasterized constraint matrix containing pipeline segment depth limits, horizontal offset tolerances, and construction restricted areas, resulting in a geographic constraint matrix. Based on the pipe material attribute data in the integrated pipeline network database, a set of pipe diameter specifications that met construction standards was screened. A Cartesian product operation was performed on the geographic constraint matrix, eliminating pipe diameter-path combinations with a spacing of less than 1.5 meters from underground power pipelines to construct a feasible solution space for pipe diameters.

[0107] Implementing hybrid algorithm parallel computing, embedding the velocity update formula of the particle swarm optimization algorithm into the genetic algorithm with elite retention strategy, dividing the feasible solution space of pipe diameter into four subspaces, calculating the initial fitness value of each subspace respectively, and retaining the solution set with the top 10% fitness value;

[0108] A pressure threshold check is performed, and the node pressure simulation values ​​output by the dynamic hydraulic simulation model are compared node by node with the minimum service pressure values ​​in the urban water supply specifications. When more than 5% of the nodes fail to meet the standards within three consecutive iteration cycles, the feasible solution space for pipe diameters is expanded. The pipe diameter combination schemes that pass the pressure threshold check are arranged in ascending order according to the renovation cost, and the top three candidate schemes and their corresponding hydraulic stability indicators are output to obtain the optimized solution set.

[0109] Road width data for the old pipeline reconstruction area (the default requirement is ≥4 meters to accommodate construction machinery, which can be relaxed to 3.5-5.0 meters), underground integrated pipeline corridor distribution data (such as heat / power pipeline coordinates), and existing building coordinates (such as the restricted area radius of historical protected buildings ≥5 meters) are extracted to construct a two-dimensional grid constraint matrix: the area is divided into 1m×1m grids (resolution adjustable from 0.5-2m), with grid values ​​marked as 0 (construction area) or 1 (restricted area). Based on the pipe material properties in the integrated pipeline database, a set of pipe diameter specifications that meet the GB / T 50268 standard (such as ductile iron pipes DN100-DN600, a total of 12 specifications) is selected. The original solution space (pipe diameter specification × path coordinate combination) is generated through the Cartesian product, and high-risk combinations with horizontal distances from underground power pipelines less than 1.5 meters (the safety threshold can be set to 1.0-2.0 meters) are automatically eliminated. Finally, a feasible solution space of pipe diameters (approximately 10 4 magnitude solution).

[0110] The velocity update formula of particle swarm optimization (PSO) is embedded in the genetic algorithm with elite retention strategy, including:

[0111] Genetic algorithm operation: initial population size 500, two-point crossover probability 0.7-0.9, mutation probability 0.1-0.3;

[0112] PSO fusion mechanism: 10% of the optimal solution of each generation is injected into the particle swarm, and the particle velocity update formula introduces an inertia weight of 0.6-0.9;

[0113] The feasible solution space is divided into four subspaces (based on network topology connectivity), and fitness values ​​(including renovation costs and leakage reduction) are calculated independently for each subspace. The top 10% of solutions (approximately 50 solutions) are retained in each iteration, while the remaining solutions are eliminated. A dynamic hydraulic model verifies node pressures: EPANET is used to simulate the flow conditions corresponding to each solution, and node pressures are compared to the minimum value specified in the CJJ 207 specification (e.g., 0.28 MPa). If more than 5% of nodes fail to meet the standard, the solution space is expanded (e.g., adding a DN700 specification).

[0114] The solution set that passes the pressure check is sorted in ascending order of renovation cost, and the top three candidate solutions are output. Hydraulic stability indicators calculated include: pressure range (highest and lowest node pressure difference ≤ 0.15 MPa), pressure fluctuation variance (24-hour simulation value variance ≤ 0.01), and bottleneck section identification (sections with flow velocities > 2.5 m / s account for < 5%). The final solution is accompanied by a construction risk assessment report (e.g., path curvature analysis of a DN600 section requiring a 200-ton crane).

[0115] Through the deep integration of geographical constraints and regulatory standards, illegal solutions can be avoided at the source; hybrid algorithms are used to collaboratively improve the optimization efficiency of complex pipe networks, and multi-dimensional stability indicators ensure the reliability of the water supply system.

[0116] Constructing the feasible solution space of pipe diameter includes the following steps:

[0117] A multi-objective conflict resolution model was established, using Pareto frontier analysis to address the trade-off between renovation cost and hydraulic stability in pipe diameter selection. The effective solution screening condition was set at a rate where the renovation cost increase did not exceed 1.5 times the hydraulic stability improvement rate. A random forest classifier was trained using historical engineering data to predict the constructibility probability of each pipe diameter-path combination, automatically eliminating any combination with a predicted value below 0.7.

[0118] A dynamic solution space index is generated. The feasible solution space of pipe diameters is coded into a B+ tree index structure according to administrative divisions. The solution space data is retrieved and updated in combination with the unique identification codes of the pipe network nodes. The principal component analysis method is used to extract the key feature vectors of the pipe diameter combination scheme, reducing the solution space dimension from the original number of pipe sections n to the log2(n) level. 5% of feasible solutions are randomly selected for on-site inspection and verification. When the verification failure rate exceeds 20%, the geographic constraint matrix is ​​regenerated.

[0119] A Pareto frontier analysis model was developed: the horizontal axis represents the pipeline network renovation cost (10,000 yuan), and the vertical axis represents the hydraulic stability index (a dimensionless score of 0-1). A non-dominated sorting algorithm was used to select the frontier solution set (approximately 20% of the total number of solutions). The effective solution selection criteria were set as follows: if a solution improves stability by ΔS% and increases renovation costs by ΔC%, then ΔC% / ΔS% must be ≤ 1.5 (with an adjustable threshold of 1.2-2.0). A random forest classifier was trained using a historical engineering database (containing over 500 cases). Input features included pipe diameter, soil type, and road grade; the output was the probability of constructability (0-1). Combinations with a predicted probability < 0.7 (with a threshold of 0.6-0.8) were automatically excluded (e.g., the risk of laying a DN500 pipe on a soft soil roadbed).

[0120] Principal Component Analysis (PCA) dimensionality reduction was performed: five principal components (cumulative contribution >85%) of the pipe diameter combination scheme were extracted, reducing the solution space dimensionality from the original number of pipe segments n (e.g., 200 dimensions) to the log2(n) level (approximately 8 dimensions). A B+ tree index structure was established: the administrative division code (e.g., 110101) was used as the root node key, and leaf nodes stored the unique identifier (16-bit code) of the pipe network node and the corresponding solution space pointer. The index update mechanism automatically expanded the solution space branches under the administrative division when a user node was added (e.g., adding a DN400 pipe segment combination).

[0121] A random sample of 50 feasible solutions (approximately 50 solutions) undergoes on-site inspection. Verification includes the actual location of underground obstacles (such as unmarked optical cables) and temporary construction conditions (such as soil bearing capacity during the rainy season). When the verification failure rate exceeds 20% (the threshold can be set between 15% and 25%), the geographic constraint matrix is ​​regenerated. This involves adjusting the restricted area radius (for example, expanding the building restricted area from 3 meters to 4 meters) and adding newly discovered pipeline corridor data. Verification results feed back into the machine learning model, incrementally updating the random forest classifier.

[0122] The intelligent screening mechanism can significantly reduce the implementation risk of the solution, the dimensionality reduction index can improve the efficiency of large-scale solution space retrieval, and the on-site feedback closed loop can optimize the accuracy of geographic constraints.

[0123] Implementing hybrid algorithm parallel computing includes the following specific steps:

[0124] A distributed computing architecture was constructed, using the MPI parallel computing framework to allocate the four subspaces to different computing nodes. Each node independently ran the hybrid optimization algorithm and synchronized the optimal solution data every 20 minutes. The crossover probability of the genetic algorithm was dynamically adjusted based on the degree of discreteness of the subspace solution set, reducing the crossover probability from 0.8 to 0.6 when the solution set standard deviation exceeded a set threshold.

[0125] The rate of change of the fitness value of each subspace is monitored. If the rate of change is less than 0.1% within three consecutive calculation cycles, the iterative calculation of the subspace is terminated early. The optimal solutions of the four subspaces are non-dominated sorted, and the congestion comparison operator is used to select evenly distributed Pareto optimal solutions to form the final optimization solution set. When the solution set of a subspace is empty due to a geographical constraint conflict, the buried depth limit of the pipe section in that area is relaxed by 10% and the calculation process is restarted.

[0126] Use the MPI parallel framework (such as OpenMPI) to allocate the four subspaces to independent computing nodes (each node is configured with 8-core CPU + 32GB memory). Inter-node data synchronization mechanism:

[0127] The optimal solutions of each subspace are exchanged every 20 minutes (adjustable 10-30 minutes), and the differences in solutions are compared through hash value comparison (such as SHA-256 checksum). The difference solution set triggers global non-dominated sorting reorganization; the genetic algorithm parameters are dynamically adjusted: the standard deviation σ of the subspace solution set is monitored. When σ>0.15 (the threshold can be set to 0.1-0.2), the crossover probability is reduced from 0.8 to 0.6, and the mutation probability is increased from 0.2 to 0.3 to enhance population diversity.

[0128] Monitor the subspace fitness value change rate δ (δ=|f t -f t-1 | / f t-1 ): When δ < 0.1% (the threshold can be set to 0.05%-0.2%) for three consecutive calculation cycles (each cycle is 20 minutes), the subspace calculation is terminated early. The Pareto optimal solutions (about 200) output by the four subspaces are merged:

[0129] The non-dominated sorting hierarchy (Rank 1 is the optimal solution set) uses the crowding distance comparison operator to select 50 evenly distributed solutions to form the final solution set. When the solution set of a subspace is empty due to a geographical constraint conflict (for example, the entire area is a construction restricted area), the buried depth limit of the pipe section is automatically relaxed by 10% (for example, the original depth limit of 2.0 meters is increased to 2.2 meters) and the calculation is restarted.

[0130] Establish an iterative repair mechanism: When an empty solution subspace is detected, the system automatically performs the following actions: locate the conflicting area (e.g., coordinate range X1-Y1), relax the constraint parameters (burial depth / horizontal offset + 10%), and record the number of relaxations (if more than three times, it will be marked as a "high conflict area"); finally, output a set of optimized solutions with a conflict resolution log (e.g., "relax the burial depth of area A to 2.2 meters") to guide project implementation priorities.

[0131] Dynamic parameter adjustment can effectively avoid local optimal traps, the early stopping mechanism significantly saves computing resources, and conflict adaptive repair improves the feasibility of the solution.

[0132] In another technical solution, the construction of the dynamic hydraulic simulation model in step S2 includes the following specific steps:

[0133] A dynamic data fusion interface was established to convert the pipe network topology data from the geographic information system into a GraphML format with topology verification. Pipeline connection relationships were stored in an adjacency table, and node elevation data was interpolated and completed using a digital elevation model. Real-time data coupling was implemented to perform Kalman filtering on the raw pressure data stream collected by the pressure sensor to eliminate transient fluctuations caused by water hammer, generating smoothed pressure time series data. Hydraulic model boundary conditions were then injected through the EPANET INP file interface.

[0134] Perform multi-parameter joint correction, adopt nonlinear least square method with constraints, use pipe segment friction coefficient and node foundation water demand as synchronous optimization variables, and set the physical constraint range of the initial value of pipeline friction coefficient;

[0135] A three-dimensional hydraulic response surface was constructed. The transient analysis module of EPANET was used to simulate the step-wise change of valve opening from 10% to 90%. The gradient matrix of the flow velocity in each pipe section as it changes with valve state was recorded, generating a three-dimensional hydraulic feature space consisting of pipe diameter, elevation, and valve opening.

[0136] During the stable water consumption period late at night every day, 50% of the regulating valves in the target area are closed, and the actual pressure distribution data is collected through the SCADA system. Dynamic time regularization is performed with the simulation value of the three-dimensional hydraulic characteristic space, and when the average absolute error exceeds 0.05MPa, the iterative correction of the friction coefficient is triggered.

[0137] The pipeline network topology data from the geographic information system was converted to GraphML format (an XML-based graph data format). Topological validation was performed to ensure the integrity of pipe segment connections. Adjacency tables were used to store node connections (e.g., node A connects to pipe segments P1 and P2). Node elevation data was interpolated from a digital elevation model (DEM) (with an accuracy of ±0.5 meters). Kalman filtering was applied to the raw pressure data stream collected by the pressure sensor to reduce noise. A state equation (including pressure values ​​and their first-order derivatives) was established, and the observation equation incorporated pressure readings from the SCADA system. The filter gain was set to an adjustable range of 0.6-0.9 to effectively eliminate transient fluctuations caused by water hammer (e.g., sudden pressure changes >0.15 MPa / s). The smoothed pressure time series data (with a sampling interval of 15 minutes) was injected into the hydraulic model boundary conditions via the EPANET INP file interface, enabling dynamic coupling of real-time data with the static model.

[0138] Perform multi-parameter joint correction: use the pipe friction coefficient and node foundation water demand as synchronous optimization variables, and adopt the constrained nonlinear least squares method. Set the physical constraint range:

[0139] The friction coefficient of cast iron pipes was 0.016-0.020 (initial value: 0.018±0.002); the friction coefficient of polyethylene pipes was 0.010-0.012 (initial value: 0.011±0.001). The EPANET transient analysis module was used to simulate step-wise changes in valve opening (10%, 30%, 50%, 70%, and 90%), recording the gradient matrix of flow velocity in each pipe section as it changes with valve state (e.g., the flow velocity of a DN400 pipe changes from 1.2 m / s at 50% opening to 1.8 m / s at 70% opening). A three-dimensional hydraulic response surface was constructed: the X-axis represents pipe diameter (DN100-DN600), the Y-axis represents elevation (0-50 meters), and the Z-axis represents valve opening (10%-90%). The surface values ​​represent the pressure distribution characteristics (in MPa).

[0140] During the daily peak water usage period (02:00-04:00), 50% of the regulating valves in the target area (preferably branch valves) were closed. Actual pressure distribution data was collected through the SCADA system (sampling density ≥ 3 points / km²). The dynamic time warping (DTW) algorithm was used to match the measured data with simulated values ​​in the three-dimensional hydraulic feature space. The pressure curve morphology was calculated for similarity. When the mean absolute error (MAE) exceeded 0.05 MPa (the threshold could be set between 0.03 and 0.08 MPa), the friction coefficient iterative correction process was automatically triggered. Correction priority was determined by error magnitude: pipe sections with MAE > 0.1 MPa were prioritized for optimization, and a new response surface was generated after each correction.

[0141] Dynamic filtering can significantly improve the availability of pressure data, multi-parameter joint correction can enhance the physical rationality of the model, and the response surface can intuitively represent the sensitive areas of valve regulation.

[0142] Performing a multi-parameter joint calibration involves the following steps:

[0143] A dynamic decomposition model of water consumption patterns was constructed. The variational mode decomposition algorithm was used to decompose the user water consumption history data into six intrinsic mode function components, and the time-frequency characteristics of the daily cycle component and the random fluctuation component were extracted.

[0144] A time-varying weight factor was introduced into the objective function of the nonlinear least squares method, increasing the weight of the measured data for the period corresponding to the daily cycle component to 2.5 times that of the random fluctuation component. Based on the service life data in the pipe property database, an upper limit of 0.022 was set for the optimization of the friction coefficient for cast iron pipe sections over 20 years old, and a lower limit of 0.009 was set for polyethylene pipe sections replaced within 5 years. The MPI parallel computing framework was used to divide the pipe network into four hydraulic zones, with each zone independently running the Levenberg-Marquardt optimizer, achieving global convergence through the exchange of boundary node pressure data. A parameter credibility report was generated, recording the correlation coefficient between the change in the friction coefficient and the pressure error reduction rate in each iteration. When the absolute value of the correlation coefficient was less than 0.7, the pipe section was marked as a potential abnormal section and its spatial coordinates were output.

[0145] Constructing a dynamic decomposition model of water consumption patterns: Using the variational mode decomposition (VMD) algorithm, historical user water consumption data (time series length ≥ 365 days) is decomposed into six intrinsic mode function (IMF) components (the number of components can be adjusted from 5 to 8). Key characteristic components are extracted: IMF1: a diurnal component (frequency 0.04-0.06 Hz); IMF2: a random fluctuation component (frequency > 0.1 Hz); and IMF3-6: seasonal / holiday components. The instantaneous frequency of each component is calculated using a Hilbert transform, and a time-frequency characteristic matrix (24 hours × frequency intensity) is constructed. A time-varying weighting factor is introduced into the nonlinear least squares objective function: measured data during the period dominated by the diurnal component (e.g., 07:00-09:00) is assigned a weight of 2.5 (adjustable from 2.0 to 3.0), while the weight for the random fluctuation period is set to 1.0.

[0146] Strengthened constraints based on service life data from the pipe property database: For cast iron pipes with a service life of more than 20 years, the optimized upper limit of the friction coefficient is 0.022 (the standard upper limit is 0.020); for polyethylene pipes with a service life of less than 5 years, the optimized lower limit is 0.009 (the standard lower limit is 0.010).

[0147] Using an MPI parallel framework (such as MPICH2), the network was divided into four hydraulic zones (grouped according to hydraulic similarity). Each zone was assigned an independent computing node to run the Levenberg-Marquardt optimizer. A boundary data synchronization mechanism was implemented: pressure data at zone boundary nodes were exchanged every 10 iterations (for example, synchronization was forced when the node pressure difference exceeded 0.02 MPa). Global convergence was achieved when the pressure difference at all zone boundaries was less than 0.01 MPa.

[0148] Generate a parameter credibility report: Record the correlation coefficient (Pearson coefficient) between the change in friction coefficient ΔC and the pressure error drop rate ΔE for each iteration. Set a credibility threshold: |r| ≥ 0.7: Optimization is effective (green mark); 0.5 ≤ |r| < 0.7: Optimization is questionable (yellow mark); |r| < 0.5: Optimization is invalid (red mark). When |r| < 0.7, the pipe segment is marked as potentially abnormal, and the spatial coordinates and anomaly type are output (e.g., "Coordinate (116.3E, 39.9N): |r| = 0.62, suspected inner wall corrosion"). Finally, generate an optimized parameter version file (including timestamp and digital signature) and automatically update it to the integrated pipe network database.

[0149] The time-varying weight mechanism is used to accurately capture the characteristics of water use patterns, differentiated constraints on age improve the adaptability of old pipeline networks, and credibility assessment enables early warning of hidden defects.

[0150] It should be noted that although the steps are described above in a specific order, this does not necessarily mean that the steps must be performed in this specific order. In fact, some of these steps can be performed concurrently or even in a different order, as long as the required functions can be achieved. The number of devices and processing scales described here are intended to simplify the description of the present invention. Applications, modifications, and variations of the present invention will be apparent to those skilled in the art.

[0151] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A modeling and analysis method for optimizing urban water supply networks, characterized in that: The following steps are involved: S1. Acquire multi-source heterogeneous data, including: collecting hydraulic parameters of pressure sensors and flow meters in the target pipe network through IoT devices, and simultaneously acquiring pipe network topology data, pipe material attribute data, and user water consumption history data from the geographic information system; aligning the acquired multi-source heterogeneous data according to a unified spatiotemporal coordinate system and storing them in a relational database to establish a comprehensive pipe network database; S2. Build a node-segment topology relationship matrix based on the network topology data. Couple the real-time monitoring data from the pressure sensors with the EPANET hydraulic calculation model. Use the nonlinear least squares method to correct the segment friction coefficient. Generate a three-dimensional network hydraulic model that includes pipe diameters, elevations, and valve status, and construct a dynamic hydraulic simulation model. S3. To meet the needs of old pipeline network renovation, a fitness function with dual objectives of pipeline network renovation cost and leakage reduction was established, and a genetic algorithm with an elite retention strategy was used to prioritize pipe segment replacement. To meet the needs of new pipeline network planning, a particle swarm optimization algorithm was established with construction cost and hydraulic stability as constraints, resulting in a hybrid optimization algorithm architecture. S4. Perform multi-stage optimization calculations, inputting pipe material property data from the integrated pipe network database into the dynamic hydraulic simulation model for initial hydraulic balance calculations. Based on the geographical constraints of the old pipe network reconstruction area, a feasible solution space for pipe diameters is generated. Using a hybrid optimization algorithm architecture, iterative calculations are performed to determine a pipe diameter combination that meets the node pressure threshold. S5. Perform spatial overlay analysis on the optimized pipe diameter combination plan and the road direction data in the geographic information system. Combined with the valve location data, generate an engineering implementation list containing pipe section replacement coordinates, new pipeline paths, and valve control instructions to generate an optimized pipe network configuration plan. The multi-stage optimization calculation performed in step S4 includes the following steps: Road width data, underground integrated pipeline corridor distribution data, and existing building coordinate data for the old pipeline network renovation area were extracted to establish a two-dimensional rasterized constraint matrix containing pipeline segment depth limits, horizontal offset tolerances, and construction restricted areas, resulting in a geographic constraint matrix. Based on the pipe material attribute data in the integrated pipeline network database, a set of pipe diameter specifications that met construction standards was screened. A Cartesian product operation was performed on the geographic constraint matrix, eliminating pipe diameter-path combinations with a spacing of less than 1.5 meters from underground power pipelines to construct a feasible solution space for pipe diameters. Implementing hybrid algorithm parallel computing, embedding the velocity update formula of the particle swarm optimization algorithm into the genetic algorithm with elite retention strategy, dividing the feasible solution space of pipe diameter into four subspaces, calculating the initial fitness value of each subspace respectively, and retaining the solution set with the top 10% fitness value; A pressure threshold check is performed, and the node pressure simulation values ​​output by the dynamic hydraulic simulation model are compared node by node with the minimum service pressure values ​​in the urban water supply specifications. When more than 5% of the nodes fail to meet the standards within three consecutive iteration cycles, the feasible solution space for pipe diameters is expanded. The pipe diameter combination schemes that pass the pressure threshold check are arranged in ascending order according to the renovation cost, and the top three candidate schemes and their corresponding hydraulic stability indicators are output to obtain the optimized solution set.

2. The modeling and analysis method for optimizing urban water supply network according to claim 1, characterized in that: The establishment of the integrated pipe network database in step S1 specifically includes the following steps: An IoT pressure sensor is installed at the intersection of the target pipe network, and an ultrasonic flow meter is installed at the user's water inlet. The collected hydraulic parameters are transmitted to the edge computing gateway in real time via the LoRaWAN communication protocol. The pipe network topology data stored in the geographic information system is converted into a vector layer in the WGS84 coordinate system, the address information in the user water consumption history data is geocoded, and a spatial mapping relationship with the pipe network topology nodes is established; For the real-time hydraulic parameter data stream collected by IoT devices, a cubic spline interpolation algorithm is used to reconstruct data segments with missing timestamps. This data is then spatially overlaid with the pipe network topology data in the geographic information system to generate a three-dimensional spatiotemporal data cube containing the spatial coordinates of the pipe network nodes, pipe segment elevation data, and the corresponding time series hydraulic parameters. A calculation model for the theoretical value of pipeline pressure is established. When the deviation between the actual sensor value and the theoretical value exceeds the safety threshold corresponding to the pipeline pressure level, the data anomaly mark is automatically triggered, and the quartile method is used to clean the abnormal data, retaining the valid data within the upper and lower quartiles. The registered network topology data is stored in a PostGIS spatial database, and the time series hydraulic parameter data is stored in a temporal database. The two databases are then linked and indexed by establishing a unique identification code for the network nodes, which is generated by combining the network number, administrative division code, and spatial coordinate hash value. Set a weekly incremental update cycle for geographic information system data, set a sliding average filter processing for the real-time data stream of the Internet of Things with a 5-minute time window, and automatically trigger the re-registration calculation of the pipe network topology data and hydraulic parameters when the user water consumption record data is updated.

3. The modeling and analysis method for optimizing urban water supply network according to claim 1, characterized in that: The following method is used to generate the pipe network optimization configuration scheme in step S5: The pipe segment replacement coordinates in the optimized pipe diameter combination scheme were converted into a Geographic Information System (GIS) Shapefile format. A buffer zone analysis was performed with the road centerline vector layer, setting a construction safety distance of 1.2 meters outside the road red line. Pipeline paths with an overlap of more than 30% with existing underground pipeline corridors were eliminated. A node-valve association matrix was established based on the valve location data in the pipeline network topology data. A breadth-first search algorithm was used to determine the control influence range of each valve, generating a control instruction tree diagram containing the valve opening and closing sequence and timing. Extract pipe attribute data from the integrated pipe network database and, combined with the operating radius parameters of the hoisting machinery at the construction site, perform curvature radius verification on pipe sections with a diameter ≥ 600mm. Insert transition elbow nodes when the turning angle exceeds the minimum turning radius of the excavator. Associate the newly created pipeline path after spatial overlay analysis with the construction feasibility verification results, sort them according to the unique identification code of the pipe network nodes, and output a structured table containing the pipe section number, starting point coordinates, end point coordinates, pipe diameter specifications, valve control sequence, and construction machinery model. When the road direction data changes, the spatial overlay analysis is recalculated, and the coordinate data of the affected pipe sections is automatically updated through version number comparison. A constraint satisfaction algorithm is used to verify the spatiotemporal conflicts between pipe section replacement and new paths in the project implementation list. When two construction tasks are detected at the same coordinate point, the execution time of non-urgent tasks is automatically postponed and a conflict warning report is generated. The project implementation list and the valve control instruction tree diagram are merged and encoded to generate an XML format document, and data integrity is ensured through digital signatures.

4. The modeling and analysis method for optimizing urban water supply network according to claim 1, characterized in that: In step S2, the dynamic hydraulic simulation model is constructed using the following method: The coordinates of the pipe segment endpoints are extracted based on the pipe network topology data of the geographic information system, and a 0-1 association matrix is ​​constructed with the pipe network nodes as row vectors and the adjacent pipe segments as column vectors. The flow direction of the pipe segment is represented by the positive and negative values ​​determined by the flow meter data. The pressure data monitored in real time by the pressure sensor is divided into data slices at a 15-minute time granularity. The data is then injected into the hydraulic calculation model through the EPANET Toolkit API interface to establish a dynamic mapping relationship between the measured pressure values ​​and the calculated values. To address the issue of error in setting the initial value of the friction coefficient of a pipe segment, the Levenberg-Marquardt algorithm was used as an iterative optimizer for the nonlinear least squares method. The root mean square error between the measured pressure data slice and the EPANET calculated value was used as the objective function to simultaneously optimize the friction coefficient combination of three adjacent pipe segments to complete the friction coefficient correction. The corrected friction coefficient was substituted into the EPANET model for transient simulation. The flow velocity and head loss data of each pipe section at different valve openings were extracted. The hydraulic state tensor of the pipe network was constructed with the spatial coordinates as the XY axis and the time series as the Z axis to generate a three-dimensional hydraulic characteristic field. During the daily low water consumption period, the pressure regulating valve of the target pipeline network was closed, and the actual pressure distribution data was obtained through the SCADA system. The Kolmogorov-Smirnov test was performed against the simulation results of the three-dimensional hydraulic characteristic field. When the P value was less than 0.05, the friction coefficient was re-adjusted.

5. The modeling and analysis method for optimizing urban water supply network according to claim 4, characterized in that: The process of coordinated correction of friction coefficient includes: A dynamic water consumption pattern library was constructed. A long-short-term memory neural network was used to train historical user water consumption data to generate a water consumption prediction model that incorporates weekday, holiday, and seasonal variations. The model then outputs a water consumption curve for each node over the next 24 hours. The water consumption prediction curve was discretized into node water demand boundary conditions at 15-minute intervals. A dynamic weighting factor was introduced into the iterative process of the Levenberg-Marquardt algorithm to prioritize matching the measured pressure data slices during peak water consumption periods. A spatial constraint on the friction coefficient is established, and the friction coefficient variation range of pipe sections made of different materials is determined based on the pipe property data. When the optimized friction coefficient exceeds the expected friction coefficient range of the pipe, an abnormal pipe state warning is triggered. A dual-threshold convergence criterion is set. When the root mean square error change of three consecutive iterations is less than 0.01m and the friction coefficient change is less than 0.0001, the optimization is terminated and the optimal friction coefficient combination is retained. The final corrected friction coefficient is written into the basic input file of the EPANET model, and the pipe property data in the integrated pipe network database is updated at the same time, and the version number and checksum are recorded.

6. The modeling and analysis method for optimizing urban water supply network according to claim 1, characterized in that: Constructing the feasible solution space of pipe diameter includes the following steps: A multi-objective conflict resolution model was established, using Pareto frontier analysis to address the trade-off between renovation cost and hydraulic stability in pipe diameter selection. The effective solution screening condition was set at a rate where the renovation cost increase did not exceed 1.5 times the hydraulic stability improvement rate. A random forest classifier was trained using historical engineering data to predict the constructibility probability of each pipe diameter-path combination, automatically eliminating any combination with a predicted value below 0.

7. A dynamic solution space index is generated. The feasible solution space of pipe diameters is coded into a B+ tree index structure according to administrative divisions. The solution space data is retrieved and updated in combination with the unique identification codes of the pipe network nodes. The principal component analysis method is used to extract the key feature vectors of the pipe diameter combination scheme, reducing the solution space dimension from the original number of pipe sections n to the log2(n) level. 5% of feasible solutions are randomly selected for on-site inspection and verification. When the verification failure rate exceeds 20%, the geographic constraint matrix is ​​regenerated.

7. The modeling and analysis method for optimizing urban water supply network according to claim 6, characterized in that: Implementing hybrid algorithm parallel computing includes the following specific steps: A distributed computing architecture was constructed, using the MPI parallel computing framework to allocate the four subspaces to different computing nodes. Each node independently ran the hybrid optimization algorithm and synchronized the optimal solution data every 20 minutes. The crossover probability of the genetic algorithm was dynamically adjusted based on the degree of discreteness of the subspace solution set, reducing the crossover probability from 0.8 to 0.6 when the solution set standard deviation exceeded a set threshold. The rate of change of the fitness value of each subspace is monitored. If the rate of change is less than 0.1% within three consecutive calculation cycles, the iterative calculation of the subspace is terminated early. The optimal solutions of the four subspaces are non-dominated sorted, and the congestion comparison operator is used to select evenly distributed Pareto optimal solutions to form the final optimization solution set. When the solution set of a subspace is empty due to a geographical constraint conflict, the buried depth limit of the pipe section in that area is relaxed by 10% and the calculation process is restarted.

8. The modeling and analysis method for optimizing urban water supply network according to claim 1, characterized in that: The construction of the dynamic hydraulic simulation model in step S2 includes the following specific steps: A dynamic data fusion interface was established to convert the pipe network topology data from the geographic information system into a GraphML format with topology verification. Pipeline connection relationships were stored in an adjacency table, and node elevation data was interpolated and completed using a digital elevation model. Real-time data coupling was implemented to perform Kalman filtering on the raw pressure data stream collected by the pressure sensor to eliminate transient fluctuations caused by water hammer, generating smoothed pressure time series data. Hydraulic model boundary conditions were then injected through the EPANET INP file interface. Perform multi-parameter joint correction, adopt nonlinear least square method with constraints, use pipe segment friction coefficient and node foundation water demand as synchronous optimization variables, and set the physical constraint range of the initial value of pipeline friction coefficient; A three-dimensional hydraulic response surface was constructed. The transient analysis module of EPANET was used to simulate the step-wise change of valve opening from 10% to 90%. The gradient matrix of the flow velocity in each pipe section as it changes with valve state was recorded, generating a three-dimensional hydraulic feature space consisting of pipe diameter, elevation, and valve opening. During the stable water consumption period late at night every day, 50% of the regulating valves in the target area are closed, and the actual pressure distribution data is collected through the SCADA system. Dynamic time regularization is performed with the simulation value of the three-dimensional hydraulic characteristic space, and when the average absolute error exceeds 0.05MPa, the iterative correction of the friction coefficient is triggered.

9. The modeling and analysis method for optimizing urban water supply network according to claim 8, characterized in that: Performing a multi-parameter joint calibration involves the following steps: A dynamic decomposition model of water consumption patterns was constructed. The variational mode decomposition algorithm was used to decompose the user water consumption history data into six intrinsic mode function components, and the time-frequency characteristics of the daily cycle component and the random fluctuation component were extracted. A time-varying weight factor was introduced into the objective function of the nonlinear least squares method, increasing the weight of the measured data for the period corresponding to the daily cycle component to 2.5 times that of the random fluctuation component. Based on the service life data in the pipe property database, an upper limit of 0.022 was set for the optimization of the friction coefficient for cast iron pipe sections over 20 years old, and a lower limit of 0.009 was set for polyethylene pipe sections replaced within 5 years. The MPI parallel computing framework was used to divide the pipe network into four hydraulic zones, with each zone independently running the Levenberg-Marquardt optimizer, achieving global convergence through the exchange of boundary node pressure data. A parameter credibility report was generated, recording the correlation coefficient between the change in the friction coefficient and the pressure error reduction rate in each iteration. When the absolute value of the correlation coefficient was less than 0.7, the pipe section was marked as a potential abnormal section and its spatial coordinates were output.

Images