A method for detecting and locating leaks in a pipeline network based on flow analysis.

Abstract

This invention discloses a method for detecting and locating leaks in a pipeline network based on flow analysis. The method comprises the following steps: S1. Constructing a flow balance model of the pipeline network; S2. Based on the flow balance model constructed in step S1, monitoring the flow status of each node in the pipeline network in real time, and initially locating the leak point when a leak is detected; S3. Using flow gradient analysis and reverse hydraulics for refined location, within the suspected leak area identified in step S2, achieving meter-level accuracy in locating the leak point through flow gradient analysis and a reverse hydraulics iterative model; S4. Deploying multiple sensor nodes on the leaking pipeline determined in step S3, collecting detection data, determining the leak location based on relevant location fusion algorithms, and then combining this data with the leak point determined by flow gradient analysis and reverse hydraulics in step S3 for a comprehensive decision, ultimately determining the accurate location of the pipeline leak. This invention overcomes the problems of low accuracy and weak anti-interference in traditional methods, achieving accurate leak detection.

Description

Technical Field

[0001] This invention relates to the field of pipeline leakage detection, and in particular to a method for detecting and locating leakage points in a pipeline network based on flow analysis. Background Technology

[0002] Currently, water supply in most regions is delivered via pipelines. However, pipeline leaks and their effective management have always been a global concern. Pipeline leaks not only waste water resources but also directly threaten regional water supply security and ecological balance. Once a localized leak occurs, quickly pinpointing the leak location is of paramount importance. In recent years, with the development of IoT technology, pipeline leak detection and location technologies have also advanced rapidly, evolving from traditional auditory and visual inspection methods to data-driven machine learning techniques.

[0003] Traditional leak detection and location techniques include: visual inspection, which locates pipe leaks by observing water accumulation on the ground; acoustic detection, which uses a listening rod or other acoustic sensors to capture sound signals generated by changes in water pressure within the pipe, requiring specialized knowledge and experience to determine the leak location, and is susceptible to noise affecting accuracy; flow meter detection, which calculates changes in fluid flow within the pipe to determine if a leak exists, but its leak location accuracy is not high; ground-penetrating radar (GPR) technology utilizes the interaction between electromagnetic waves and underground media, transmitting high-frequency electromagnetic waves and receiving reflected signals to detect leaks; tracer gas detection involves injecting easily detectable and diffuse tracer gas into the pipe, and the leak is located by the gas diffusing to the leak point, but its real-time performance is poor; pressure gradient detection measures pressure changes along the pipe to detect leaks, but it cannot pinpoint the exact leak location; and infrared thermal imaging detects the infrared radiation energy emitted by objects, achieving real-time surface temperature detection in a non-contact manner, but is susceptible to background temperature differences and is prone to misidentification of leaks. Fiber optic technology cleverly deploys fiber optic sensors around or inside pipes for monitoring, but its high cost and sensitivity to environmental factors affect the accuracy of leak detection. Furthermore, these single detection technologies limit the accuracy and real-time performance of leak detection in complex environments.

[0004] In recent years, data-driven detection technologies such as Support Vector Machines (SVM), Anonymous Neural Networks (ANN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN) have emerged. These technologies train on large amounts of historical data to predict and identify leaks. However, SVM requires a large number of training samples to achieve good performance and struggles with scarce or imbalanced data. ANNs are unique in their strong nonlinear modeling capabilities, able to capture and learn complex nonlinear relationships. The network can automatically extract features from data and has strong generalization ability, making it suitable for processing various types of data. However, ANNs' generalization ability is limited when dealing with scarce or imbalanced data. CNNs demonstrate strong application potential in handling complex leak detection scenarios, especially when using multi-sensor data, but the training process requires significant data volume and computational resources. RNNs highlight their unique advantages in leak detection. They excel at processing sequential data, extracting temporal features and contextual information, and can meet different processing needs by adjusting the network structure (e.g., using LSTM or GRU units). However, these data-driven methods, which rely on in-depth mining and analysis of large amounts of data to detect pipe leaks, suffer from problems such as poor model interpretability and low reliability, thus limiting the accurate location of leaks.

[0005] In summary, existing leak detection methods for water supply networks still suffer from poor positioning accuracy and numerous interferences, leaving room for improvement. A detection method capable of accurately detecting leaks in water supply pipelines is needed. Summary of the Invention

[0006] This invention proposes a method for detecting and locating leaks in a pipe network based on flow analysis, comprising the following steps:

[0007] S1. Construct a flow balance model for the pipeline network;

[0008] S2. Based on the flow balance model constructed in step S1, monitor the flow status of each node in the pipeline network in real time. When a pipeline leak is detected, perform preliminary location of the leak area.

[0009] S3. Flow gradient analysis and reverse hydraulic fine-tuning: In the suspected leak area identified in step S2, flow gradient analysis and reverse hydraulic iterative model are used to locate the leak point with meter-level accuracy.

[0010] S4. Multiple sensor nodes are deployed on the leaking pipe identified in step S3 to collect detection data. The location of the leak is determined based on relevant positioning fusion algorithms. The leak is then combined with the leak identified in step S3 by flow gradient analysis and reverse hydraulic analysis to make a comprehensive decision and finally determine the accurate location of the pipe leak.

[0011] Further, step S1 specifically includes:

[0012] S1.1 Data Acquisition: High-precision flow meters and pressure sensors are installed at key nodes of the pipeline water supply network. The large water supply area is divided into several independent metering areas. Flow meters are set at the inlet and outlet of each area, and pressure sensors are deployed at the same locations as the flow meters.

[0013] S1.2 Data Preprocessing: Wavelet denoising is applied to the raw flow / pressure data, moving average filtering is performed on the pressure signal, and anomaly detection and cleaning are performed on the collected data;

[0014] S1.3 Constructing a hydraulic model of the water supply network: Based on the network topology modeling, establish steady-state hydraulic models and transient models.

[0015] Furthermore, in step S1.2, during data cleaning, gross error data is removed, and the isolated forest algorithm is used to identify complex anomaly patterns in the remaining data.

[0016] In step S1.3, during transient modeling, a dynamic pressure monitoring and compensation strategy and a pressure wave propagation frequency domain filtering strategy are adopted. In the dynamic pressure monitoring and compensation strategy, when the detected pressure change rate exceeds a threshold, the steady-state model is dynamically corrected using the following formula: Q comp =Q base +α·ΔP· ;

[0017] Among them, Q comp Q is the actual flow rate corrected for pressure changes. base It is the theoretical flow rate under conditions of no pressure fluctuations and external disturbances, where ΔP is the pressure change, Δt is the time step, and α is the pipe material elasticity correction coefficient. In the pressure wave propagation frequency domain filtering strategy, the original flow signal collected by the flow meter is decomposed by wavelet packet decomposition to separate the high-frequency fluctuation component caused by the pressure wave propagation. After removing the high-frequency component, the reconstructed signal is used as the input data of the steady-state model.

[0018] Further, in step S2, the difference between the inlet flow and the outlet flow of each partition is calculated based on the flow balance method, and the calculation formula is as follows: Q in Q represents the flow rate value of the inlet flow meter. out This indicates the flow rate value of the outlet flow meter. This is the normal water consumption; when When this occurs, it is determined that the partition has a leak, where ε is a preset value.

[0019] Further, step S3 specifically includes: performing analysis based on the flow gradient to calculate the flow gradient between adjacent flow meters, using the following formula: Where ∇Q represents the flow gradient, Q i Let Q be the instantaneous flow rate of the i-th flow meter. i+1 Let L be the instantaneous flow rate of the downstream adjacent flow meter, and L be the pipe length between the two flow meters. If the flow gradient value of a certain pipe section is significantly greater than that of its upstream pipe section, the leak is determined to be located within that pipe section near the upstream flow meter. When the dynamic pressure change rate is greater than the threshold, the formula Q is used. comp =Q base +α·ΔP·

[0020] Real-time calibration of flow rate values, based on the corrected flow rate Q comp A reverse hydraulic inversion model was established to solve for the location of the leak.

[0021] Furthermore, the reverse hydraulic inversion model for determining the leak location includes the following steps:

[0022] Assume the leak location x and the leak flow rate Q. leak As the parameter to be solved, a flow-leak correlation equation is established. For a pipe section containing a leak, the corrected flow rate monitored by the upstream and downstream flow meters satisfies the formula: Q comp , upstream =Q comp , downstream +Q leak ; wherein, the Q comp , upstream Correcting the flow rate upstream of the leak point; the Q comp , downstream To correct the flow rate downstream of the leak, Q leak For leakage flow;

[0023] With the objective of minimizing the residual between simulated flow and corrected flow, an objective function is constructed. Where x is the location of the leak point; Q sim,i (x) represents the simulated flow rate of the i-th flow meter when the leak point is assumed to be at location x; Q real,i The corrected actual flow rate of the flow meter; n is the flow count of the target pipe section and the upstream and downstream related pipe sections;

[0024] The nonlinear least squares optimization algorithm is used for iterative solution until the termination condition of residual accuracy meeting the standard or position change convergence is met.

[0025] Furthermore, in step S4, the relevant positioning fusion algorithm includes multi-sensor node deployment and data acquisition, cross-correlation algorithm positioning, and multi-node correlation algorithm fusion positioning.

[0026] Furthermore, the cross-correlation algorithm for localization includes:

[0027] When sensor A and sensor B are located on opposite sides of the leak point, the sampled leak vibration signals are defined as follows: , ,but , Having similarity, if Lag Time is Then approximately: Where A is the cross-correlation coefficient, at this time , The cross-correlation function is:

[0028] ,

[0029] When R ab When the maximum value is reached, the corresponding τ is , This is the time difference between the leakage vibration wave reaching the two probes. If the sound velocity of the medium in the pipe is V, then the pipe length L between the leakage point and node B is (MV). T) / 2, at this point the leak point location between the two points is completed.

[0030] Furthermore, the multi-node related algorithm fusion positioning algorithm includes cluster head selection, clock synchronization algorithm, and multi-sensor data fusion positioning algorithm; wherein,

[0031] The cluster head is the node located at the center of the cluster and has the most remaining energy;

[0032] The clock synchronization algorithm includes: the cluster head node broadcasts a data packet carrying a reference timestamp beacon to each node in the cluster; each node receives and records the reference timestamp and its local time; the adjacent nodes of the cluster head sequentially transmit an integrated data packet carrying the reference time and its own local time until the last node, which averages all local times and transmits them back to each node in reverse order; each node updates its local time; the operation is repeated until the clock synchronization error is less than the threshold.

[0033] The multi-sensor data fusion positioning algorithm includes:

[0034] Under normal circumstances where the water supply pipeline within the cluster has no branches, the nodes are grouped, paired, and numbered according to the intensity of the leakage sound signal picked up by the nodes. This ensures that the location of the leakage point falls between the two nodes performing the cross-correlation algorithm. The number reflects the strength of the signal received by the paired node. The location of the leakage point determined by all paired nodes is calculated according to the cross-correlation algorithm positioning method. An improved weighted average algorithm is used to obtain the final location of the leakage point from multiple positioning results. The weighting factor is inversely proportional to the distance between the two paired nodes. Furthermore, the numbering is sorted and the abnormal weighting factor is removed during each fusion to eliminate paired nodes affected by sudden noise interference.

[0035] In the special case where there are branches in the pipeline within the cluster, for multiple group pairing results formed according to the node signal strength, if the deviation of the positioning results of different groups is within the meter range, the average value is taken as the leak point location; if the deviation between the two points is outside the meter range, the leak point is determined by comprehensive decision-making.

[0036] Furthermore, the comprehensive decision-making in step S4 includes the following steps: S4.4.1 Data Preparation and Structuring: Collect the positioning results obtained from the reverse hydraulic method and the multi-node data correlation positioning fusion algorithm, and establish an information matrix containing the location information of each candidate point, the source method, and the corresponding confidence / accuracy index;

[0037] S4.4.2 Preliminary Fusion and Overlap Analysis: Overlay all candidate points onto the pipeline network GIS map, identify highly spatially clustered areas, and filter points that both methods point to or points with extremely high confidence / accuracy. S4.4.3 Introduce field constraints: Quantify the field factors that affect the actual location of the leak, including pipe material and aging degree, interface / valve location, burial depth and soil load, soil properties and corrosive environment, historical leak records, external disturbances and pipeline topology.

[0038] S4.4.4 Establish a comprehensive decision-making model, design a scoring system, and calculate a comprehensive score for each candidate point;

[0039] S4.4.5 Sorting and Preliminary Decision-Making: Sorting by comprehensive score to determine the most likely leaks or multiple high-risk areas.

[0040] This invention achieves precise location by combining flow analysis with multiple technologies. High-precision flow meters and pressure sensors are installed at key nodes in the pipeline network to collect and preprocess data (synchronization, noise reduction, and cleaning). Steady-state and transient hydraulic models are constructed to eliminate interference such as water hammer. Based on flow balance, the flow difference between the inlet and outlet of each zone is calculated; exceeding a threshold indicates a suspected leak. Time-series methods such as minimum nighttime flow and cluster analysis are used to verify and pinpoint the actual leak area. Then, the flow gradient of the pipe section is calculated to determine the leaking pipe. A reverse hydraulic model is built based on the corrected flow rate, and through iteration, the residual difference between the simulated and actual flow rates is minimized, locating the leak point to the meter level. Finally, multiple sensors are deployed on the target pipeline, and a cross-correlation algorithm is used for initial location. After optimization such as cluster head selection and clock synchronization, the results of reverse hydraulic analysis and multi-node location are combined with on-site constraints (such as pipe material and historical records) for comprehensive decision-making, achieving a final location accuracy of approximately 0.3 meters. This scheme overcomes the problems of low accuracy and weak anti-interference in traditional methods through multi-stage progressive location and the integration of multiple technologies and on-site factors, achieving precise leak detection. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, as follows: Figure 1 As shown, the accompanying drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the pipeline leak detection and distance location method based on flow analysis in this invention.

[0043] Figure 2 This is a schematic diagram illustrating the localization principle of the cross-correlation algorithm in this invention;

[0044] Figure 3 This is a schematic diagram illustrating the principle of precise leak location using multi-sensor node data fusion in this invention.

[0045] Figure 4 This is a schematic diagram of a branched pipe network structure in this invention. Detailed Implementation

[0046] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0047] This invention discloses a method for detecting and locating leaks in a pipe network based on flow analysis, combined with the attached... Figure 1 As shown,

[0048] The core of leak location based on flow data lies in analyzing abnormal changes in flow distribution within the water supply network. When a pipeline leaks, the flow upstream of the leak point increases (due to additional water flow from the leak), while the flow downstream decreases (because some water is lost from the leak). By constructing a flow balance model, performing flow gradient analysis and reverse hydraulic analysis, and employing relevant fusion algorithms, the location of the leak point is comprehensively determined. Its main methods include:

[0049] Step S1: Construct a flow balance model for the pipeline network.

[0050] Specifically, step S1 includes:

[0051] S1.1 Data Acquisition

[0052] High-precision flow meters should be installed at key nodes in the pipeline water supply network (such as pumping stations, zone inlets / outlets, and main pipeline intersections), with electromagnetic or ultrasonic flow meters being the preferred choice. The selection principles are as follows:

[0053] Electromagnetic flowmeter: Suitable for stable measurement of full-pipe flow and conductive liquids, with pipe diameter range of DN50~DN2000. During installation, ensure that the straight pipe sections before and after the pipe are ≥5 times the pipe diameter.

[0054] Ultrasonic flow meter: suitable for large pipe diameter (DN300 and above) or non-full pipe flow, supports clamp-on installation, but is sensitive to fluid turbidity and requires regular calibration.

[0055] Furthermore, for large-area water supply areas, the pipe network is divided into several independent metering zones (DMAs), and flow meters (of the same type to reduce system errors) are installed at the inlet and outlet of each zone to monitor flow data in real time. Pressure sensors are simultaneously deployed and installed at the same locations as the flow meters, with a sampling frequency ≥1Hz, for transient process analysis and leak verification.

[0056] S1.2 Data Preprocessing

[0057] PTP (Precision Time Protocol) or GPS clock synchronization is used to ensure that the timestamp error of all flow meters and pressure sensors is <0.1 seconds (to meet the requirements of subsequent flow gradient analysis). Wavelet denoising is applied to the raw flow / pressure data to remove high-frequency noise (>1kHz) and low-frequency drift (<200Hz). The pressure signal needs to be additionally filtered by moving average (window width 5 seconds) to eliminate instantaneous fluctuation interference.

[0058] Data cleaning involves anomaly detection of the collected data. In the first stage, gross errors are eliminated using the 3σ principle (data within ±3 standard deviations is retained). In the second stage, the Isolation Forest algorithm is used to identify complex anomaly patterns (such as sensor drift and intermittent interference) on the remaining data.

[0059] S1.3 Constructing a hydraulic model of the water supply network

[0060] First, based on pipeline topology modeling, a water supply pipeline topology model is established using professional hydraulic modeling software such as CFD. Specific steps include: labeling parameters such as pipe length, diameter, material, and valve location; inputting node coordinates and connection relationships to generate a water supply pipeline diagram (Graph structure); inputting pipe parameters, such as length L, diameter D, material C, and valve location and status; and labeling node parameters.

[0061] Secondly, steady-state hydraulic modeling is performed, including: constructing a steady-state model based on the principles of mass conservation and energy conservation, and calculating the flow distribution at each node under normal operating conditions.

[0062] For each node i, the following needs to be satisfied:

[0063] (In case of leakage)

[0064] Among them, Q inQ represents the flow of traffic into node i. out Q represents the flow of traffic leaving node i. leak This represents the amount of water leakage at node i.

[0065] Finally, transient modeling is required. Due to the water hammer effect and pressure wave propagation caused by transient processes (such as valve opening and closing, pump station start-up and shutdown) in actual pipeline operation, the steady-state model calculation results are uncertain. Therefore, dynamic pressure monitoring compensation strategies and pressure wave propagation frequency domain filtering strategies are needed to reduce their impact.

[0066] For dynamic pressure monitoring and compensation strategies, pressure sensors need to be installed at key nodes to monitor pressure fluctuations in real time. When the pressure change rate |∣∂P / ∂t∣>ε is detected... p When εp = 0.1 MPa / s, a significant transient process is considered to have occurred. In this case, the steady-state model is dynamically modified.

[0067] Q comp =Q base +α·ΔP·

[0068] Among them, Q comp Q represents the actual flow rate after correction for pressure changes. base This represents the theoretical flow rate under conditions of no pressure fluctuations and no external disturbances. ΔP is the pressure change, Δt is the time step, and α is the pipe elasticity correction coefficient.

[0069] For frequency domain filtering of pressure wave propagation, wavelet packet decomposition is required on the raw flow signal collected by the flow meter to separate the high-frequency fluctuation components caused by pressure wave propagation. After removing the high-frequency components, the reconstructed signal is used as input data for the steady-state model to eliminate the interference from pressure wave propagation.

[0070] Step S2: Based on the flow balance model constructed in step S1, the flow status of each node in the pipeline network is detected in real time. When a pipeline leak is detected, the leak point is initially located.

[0071] In step S2, the difference between the inlet and outlet flow rates of each zone is first calculated based on the flow balance method, and the presence of leakage in each pipe section is calculated according to the following formula:

[0072] Among them, Q in Q represents the flow rate value of the inlet flow meter. out This indicates the flow rate value of the outlet flow meter. This represents normal water consumption (which can be estimated using historical data).

[0073] when (Threshold, usually ε = max(Q)) inQ out If the partition is found to have a leak, proceed to step S3 for verification; otherwise, it is considered normal.

[0074] Furthermore, after the flow balance method initially identifies the suspected leak zones, time-series analysis is needed to verify the persistence of the leak in order to eliminate temporary interferences (such as sudden water usage or sensor fluctuations) and ensure the reliability of the detection results. For example, any of the following auxiliary methods can be used:

[0075] The minimum flow rate method at night: This method highlights leakage characteristics during periods of low water usage, selecting 2-4 AM as the analysis period. This is because this is a low-water period, when normal residential / industrial water consumption is at its lowest and most stable (baseline flow is stable). If a continuous leak exists, the leak point will continuously generate additional flow (unrelated to normal water usage), causing the actual flow rate during this period to be significantly higher than the normal baseline. This method focuses on the "stable window with minimal interference," amplifying the flow anomaly caused by the leak and reducing the interference of daily water usage fluctuations on detection.

[0076] DBSCAN (Density Clustering Algorithm) Analysis: Outliers from temporary special events are excluded. Nighttime flow data may contain non-leaking temporary anomalies (such as fire emergency water supply, pipe flushing, instantaneous sensor malfunctions, etc.). These events manifest as short-term, large flow fluctuations (outliers), but are not continuous leaks. DBSCAN identifies "dense areas" (normal flow fluctuation ranges) and "sparse outliers" (special events) in the data. After excluding outliers, the remaining flow anomalies are more likely to be stable deviations caused by continuous leaks, thus improving data purity.

[0077] Historical baseline comparison: Leak stability is verified through continuous deviation. The current nighttime flow rate after treatment is compared with the historical baseline (e.g., the average normal nighttime flow rate for the same period over the past 30 days). A threshold of "10% above the baseline for 3 consecutive days" is set as the alarm threshold. The historical baseline represents the normal flow rate level when there is no leak; a leak will cause the current flow rate to consistently exceed the baseline. The "3 consecutive days" setting avoids misjudgments due to occasional daily fluctuations (such as short-term water usage changes caused by abnormal weather), ensuring that the anomaly is continuous. The "10% threshold" is set based on the ratio of pipeline leakage to normal water usage, balancing sensitivity and false alarm rate.

[0078] CUSUM and Bayesian Change Point Detection: Accurately Identifying Anomalies in Traffic Trends. CUSUM (Cumulative Sum Algorithm): By accumulating the deviation of traffic from the baseline, when the cumulative value exceeds a preset threshold, a "change point" (a sudden change in traffic trend) is identified. It is suitable for detecting continuous, small increases in traffic caused by leaks (such as slow leaks), and compared to single-data comparisons, it is better at capturing long-term trend anomalies. Bayesian Change Point Detection calculates the posterior probability of "change points" (the points in time when a leak occurs or continues) in the traffic sequence based on a probabilistic model. By quantifying uncertainty, it accurately identifies the turning point from a normal to an abnormal state in traffic, verifying whether the leak persists and its start time.

[0079] These time-series analysis methods verify the persistence of traffic anomalies through the logic of "focusing on low-interference periods → excluding temporary anomalies → comparing with historical baselines → detecting trend changes," ultimately confirming whether there is a real leak in the suspicious partition, providing a reliable basis for subsequent refined positioning, and reducing the false judgment rate.

[0080] Step S3, flow gradient analysis and reverse hydraulic fine-tuning location: After identifying suspected leak areas in the coarse location stage, the fine-tuning stage achieves meter-level accuracy in location through flow gradient analysis and a reverse hydraulic iterative model. This step further narrows down the leak location area after initially locating the area with leakage using the flow balance method in step S2, improving the location accuracy from the regional level to the meter level. This provides specific coordinates for subsequent precise repair and solves the problem of coarse location by the single flow balance method.

[0081] Specifically, step S3 includes:

[0082] S3.1 Based on flow gradient analysis, dense flow meters (spacing ≤ 1 km) were arranged along the pipeline in the suspected leakage zone, and the time synchronization error of these flow meters was ≤ 0.1 seconds.

[0083] First, calculate the flow gradient between adjacent flow meters: ;

[0084] The flow gradient between adjacent flow meters is calculated using a formula, where Q represents the flow gradient. i Let Q be the instantaneous flow rate of the i-th flow meter. i+1 Let L be the instantaneous flow rate of the downstream adjacent flow meter, and L be the pipe length between the two flow meters. If the flow gradient value of a certain pipe section is significantly greater than that of its upstream pipe section, the leak is determined to be located within that pipe section near the upstream flow meter.

[0085] Then, for real-time pressure data, when the dynamic pressure (the rate of change of pressure over time) |∣∂P|∂t∣>0.1MPa / s, it indicates that the pipeline network is transitioning from a steady state to an unstable state. At this point, proceed to step S3.2.

[0086] S3.2, Using the formula: Q comp =Q base +α·ΔP·

[0087] Real-time calibration of flow rate values, based on the corrected flow rate Q comp A reverse hydraulic inversion model was established to solve for the location of the leak.

[0088] S3.3 Reverse hydraulic calculation: The reverse hydraulic inversion model is based on the premise that "pipeline leakage will cause local flow anomalies," and establishes mathematical relationships through the following assumptions and constraints:

[0089] Step 1: Assuming the leak is located within a certain pipe section, calculate the leak location x (distance from the start of the pipe section, in meters) and the leakage flow rate Q. leak As a parameter to be solved, the presence of a leak will change the hydraulic characteristics of the pipe section, manifested as an increase in flow upstream of the leak, a decrease in flow downstream, and an abnormal pressure decay pattern along the pipe.

[0090] The second step is to establish the flow-leakage correlation equation based on the corrected flow Q. comp (Dynamic pressure interference has been eliminated), construct the correlation equation between the leak and the flow rate. For pipe sections containing leaks, the corrected flow rates monitored by upstream and downstream flow meters must satisfy: Q comp , upstream =Q comp , downstream +Q leak ; where Q comp , upstream To correct the flow upstream of the leak, Q comp , downstream To correct the flow rate downstream of the leak, Q leak This is the leakage flow rate (related to the size of the leak and the pipeline pressure).

[0091] The third step is to construct an objective function with the goal of minimizing the residual between the simulated flow and the corrected flow:

[0092] The objective function is: ;

[0093] Where x is the location of the leak point, Q sim,i (x) represents the simulated flow rate of the i-th flow meter when the leak point is assumed to be at location x. real,i The corrected actual flow rate of the flow meter (i.e., Q) comp (n) represents the number of flow meters for the target pipe section and its upstream and downstream related pipe sections. The flow meters must cover the key nodes of the target pipe section and its upstream and downstream related pipe sections, including at least two upstream and two downstream of the leak point, to ensure that changes in flow gradient can be captured.

[0094] The objective function is physically represented by the sum of squared residuals between the simulated flow rate and the corrected flow rate. The smaller the value, the closer the assumed leak location x is to the actual leak location. By minimizing this value, accurate leak location inversion can be achieved. The initial location with the smallest residual is selected as the starting point for iteration, and a nonlinear least squares optimization algorithm is used for iterative solution. The leak location inversion process, with the objective of minimizing the residual between the simulated and corrected flow rates, continuously adjusts the leak location parameters to gradually reduce the residual between the simulated and corrected flow rates, eventually converging to the optimal solution. The specific steps are as follows:

[0095] S3.3.1 Gradient Calculation: For the current leak location, calculate the difference (residual) between the simulated flow rate and the corrected flow rate at each flowmeter, clarifying the deviation of each sensor. Through hydraulic model parameter perturbation experiments, simulate the change in simulated flow rate at each flowmeter when the leak location changes slightly (e.g., ±0.1 meters), thereby determining "the change in simulated flow rate for every 1-meter change in leak location," i.e., the sensitivity coefficient. Multiply the residuals by the sensitivity coefficients and sum them to obtain the gradient value of the objective function with respect to the current leak location. The positive or negative direction of the gradient directly reflects "which direction the location adjustment will reduce the residual." If the gradient is positive, the leak location needs to be moved to the left (upstream) to reduce the residual; if the gradient is negative, it needs to be moved to the right (downstream).

[0096] S3.3.2 Adaptive Step Size Adjustment: The step size is the adjustment range of the leak point location in each iteration. Its size directly affects the iteration efficiency and convergence stability, and needs to be dynamically adapted according to the residual changes. The initial step size setting is based on the pipe length and flowmeter spacing to determine the initial adjustment range (e.g., 5% of the target pipe section length), ensuring that the initial iteration can quickly approach the optimal solution. After each iteration, compare the residual changes before and after the adjustment. If the residual decreases significantly (e.g., more than 10%), it indicates that the current step size direction is correct and the range is reasonable. The next iteration increases the step size to accelerate convergence. If the residual increases or decreases only slightly (e.g., less than 1%), it indicates that the step size is too large and may cause oscillations. The next iteration decreases the step size to stably approach the optimal solution. If the residual fluctuates frequently, the minimum step size protection mechanism is triggered, forcibly limiting the step size to within a threshold meter to avoid over-adjustment.

[0097] S3.3.3, Position Update: Iteratively adjust along the gradient descent direction. Update the leak position based on the gradient direction and the current step size: If the gradient is positive, move the leak position upstream along the pipeline by the current step size; if the gradient is negative, move the leak position downstream along the pipeline by the current step size. After each position update, it is necessary to check whether the new position exceeds the target pipe section range (e.g., it must not exceed the monitoring range of the upstream and downstream flow meters). If it exceeds, it will be automatically truncated to the pipe section boundary to ensure physical rationality.

[0098] S3.3.4. The iteration terminates when either of the following two termination conditions is met, thus ensuring the reliability of the inversion results: First, the residual accuracy meets the standard, the relative deviation between the simulated flow and the corrected flow of all flowmeters is ≤0.5% (i.e., the flow deviation of a single sensor does not exceed 0.5% of the actual flow), and the overall residual sum of squares is stable at a minimum value, with the change in three consecutive iterations ≤0.001; the position change converges. Second, the adjustment of the leak point position between two adjacent iterations is ≤0.1 meters, indicating that the position is close to the actual leak point, and continuing the iteration has no significant meaning for improving accuracy.

[0099] Step S4: After determining the leaking pipe through flow gradient and hydraulic inversion in step S3, the present invention further proposes step S4 to collect detection data from multiple sensor nodes deployed on the determined leaking pipe, determine the location of the leak point based on relevant positioning fusion algorithms, and then make a comprehensive decision based on flow gradient analysis and reverse hydraulic determination to finally determine the accurate location of the pipe leak.

[0100] Specifically, step S4 includes the following key technologies:

[0101] (1) Deployment of multiple sensor nodes and data acquisition

[0102] To enhance the reliability and accuracy of leak detection systems, this invention utilizes sensor nodes deployed along water supply pipelines. It calculates data from multiple sensor nodes located at identified leak points along the pipeline. This multi-sensor node data acquisition overcomes the limitations of single sensors, and cross-validation of multi-node data reduces misjudgments caused by single sensor failures or localized noise. It also provides spatially distributed physical signals (vibration, acoustics) for subsequent cross-correlation algorithms, allowing for the inference of leak location based on signal propagation characteristics, thus compensating for the insufficient capture of microscopic physical features by flow / hydraulic models.

[0103] (2) Principle of cross-correlation algorithm for localization

[0104] like Figure 2 As shown, under normal circumstances: when sensor A and sensor B are located on opposite sides of the leak point, if nodes A and B are controlled by a synchronous clock, the sampled leak vibration signals will be respectively... , ,but , Having similarity, if Lag Time is Then approximately: , where A is the cross-correlation coefficient. At this time... , The cross-correlation function is:

[0105]

[0106] When R ab When the maximum value is reached, the corresponding τ is ,Right now Figure 2 In the diagram, T represents the time difference between the leakage vibration wave and the two probes. If the sound velocity of the medium in the pipe is V, then the pipe length L between the leakage point and node B is (MV). T) / 2, at this point the leak location between the two points is completed, which is basically consistent with the location in step S3. Utilizing the propagation law of physical signals (vibration waves), the correlation between "signal time difference and spatial distance" is directly established to achieve preliminary leak location. This, together with the "flow rate-hydraulic inversion" in step S3, forms a two-dimensional verification of "hydraulic data + physical signal," providing basic location results for subsequent multi-node fusion algorithms as input for refined optimization.

[0107] (3) The above is the ideal situation for locating water leaks, but the following problems still exist: 1. When the synchronous clock sampling of nodes A and B is out of sync, it will cause the time difference T value to deviate, which will directly affect the positioning accuracy; 2. If the water leak point is located outside the two nodes or exactly at the location of sensor A or B, then the formula L = (MV T) / 2=(MM) / 2=0, which caused the positioning method to fail; 3. When the sound wave signal at the leak point is weak and there is strong noise in the surrounding environment (greater than 80dB), it may lead to the noise source being misjudged as the leak point.

[0108] Based on the above, this invention further proposes to improve the accuracy of leak detection by fusing multiple node-related algorithms. Specifically, this includes three key methods: cluster head selection, clock synchronization algorithm, and multi-sensor data fusion positioning algorithm.

[0109] For cluster head selection, this invention forms a cluster from candidate areas identified by traffic flow localization. Due to their different locations, nodes within a cluster perceive different data when detecting the same leak point. The cluster head is determined by considering factors such as the physical location, spatial distance, and remaining energy of ordinary nodes and the cluster head node. In this algorithm, each node calculates its own location and remaining energy and transmits the results to a designated node. This designated node then synthesizes and judges the results, selecting the node at the center of the cluster with the most remaining energy as the cluster head. The cluster head information is then broadcast to all nodes within the cluster. Cluster head selection significantly reduces data transmission redundancy. By centrally processing data through the cluster head, algorithm efficiency is improved. Prioritizing nodes with sufficient energy and central locations as core nodes ensures data reliability and coverage, avoiding errors caused by weak signals from edge nodes.

[0110] The clock synchronization algorithm works as follows: First, the cluster head node broadcasts a data packet carrying a reference timestamp beacon to all nodes in the cluster. Upon receiving the data packet, each node immediately records the reference timestamp and its local time. Then, its neighboring node 1 sends an integrated data packet carrying the reference time and its own local time to its next neighboring node 2. When node 2 receives the integrated data packet from node 1, it adds its own local time and reassembles it into a new data packet, which is then sent to the next neighboring node 3. This process continues until node n, at which point node n has obtained the local clock information of all nodes in the cluster (including the cluster head). Finally, node n averages all local times and then transmits this average value to all nodes in the cluster in the reverse order, updating their local times accordingly. This process is repeated until the clock synchronization error is less than a threshold. This effectively solves the problem of "time difference T calculation deviation" in the cross-correlation algorithm, ensuring that the sampling time of multiple nodes is aligned and avoiding positioning errors caused by clock asynchrony.

[0111] Multi-sensor node data fusion algorithms, such as Figure 3 As shown, under normal circumstances, when there are no branches in the water supply pipeline within the cluster, the received leakage vibration wave signal intensity can be directly grouped, paired, and sorted, and then cross-correlation and fusion localization algorithms can be performed. However, because the intensity of the leakage vibration sound wave at the pipeline leak point decreases with increasing distance between the measuring node and the sound source, based on this characteristic, the nodes within the cluster can be paired, sorted, and numbered according to the strength of the leakage sound signal picked up by each node. This ensures that the leak point location falls between the two nodes used in the cross-correlation algorithm, and the number reflects the strength of the signal received by the paired node. The multi-sensor node data fusion algorithm can solve the problems of "leak point exceeding the range of two nodes" or "noise interference": through cross-validation of multiple groups of nodes, abnormal results are eliminated, ensuring that the localization result falls within a reasonable range. The weighted averaging strategy highlights the contribution of highly reliable node pairs, reduces the influence of weak signals or noise, and thus improves the robustness of the algorithm.

[0112] Using the cross-correlation positioning method in step (2) of step S4, the location of the leak point determined by all paired nodes is calculated. An improved weighted average algorithm is then applied to multiple sets of positioning results to obtain the final location of the leak point.

[0113] As the principle of signal attenuation shows, the farther the sensor nodes are from the leak point in the pipe, the greater the signal attenuation, the less obvious the leakage information carried in the signal, and the lower the reliability of the positioning result calculated by the paired nodes. Therefore, a weighting factor (i.e., weight) inversely proportional to the distance between the paired nodes is introduced as a coefficient of the positioning result of the paired nodes. This weighted average fusion algorithm can further improve the final positioning accuracy.

[0114] Within each section, various random bursts of noise interfere with or even submerge the sound waves from the leaking pipe. Nodes affected by this noise cause the traditional weighted average's weighting factors to become ineffective, leading to greater errors. Therefore, an improved weighted average algorithm can avoid the impact of these bursts of noise. The solution involves reviewing the sorted numbers during each fusion process to check for any abnormal weighting factors. If any are found, they are removed, ensuring that paired nodes affected by bursts of noise do not participate in data fusion, thus improving the algorithm's robustness.

[0115] Special case: When there are branches in the pipeline within the cluster, this is the most difficult situation to determine, such as... Figure 4 As shown, nodes 1 to 6 are identified. Assuming point A is the leak point, there are three groups based on the node signal strength: Group 1: 2-5, 1-6; Group 2: 3-5, 4-6; Group 3: 2-3, 1-4. According to the cross-correlation algorithm, the leak point determined by Groups 1 and 2 is A, while Group 3 is incorrectly identified as B. If the distance between point B and point A is within the meter range, we consider the leak point to be at (A+B) / 2, and the leak point can be accurately found within the specified construction area. If the distance between point B and point A is outside the meter range, then step (4) is to comprehensively determine the leak point.

[0116] (4) Implementation plan for comprehensive decision-making loopholes

[0117] Based on the location results obtained from "flow gradient analysis and reverse hydraulic method" and "multi-node data correlation and location fusion algorithm", and combined with the objective conditions on site, the most reasonable and probable actual leak point location is selected to guide precise excavation and repair. The core strategy employs data fusion, weighted analysis, and on-site verification.

[0118] Specific steps of the solution:

[0119] S4.4.1 Data Preparation and Structure

[0120] Collect the location results and clearly record the coordinates / pipe segment identifiers of the N-1 most likely leak points obtained from the "flow gradient analysis and reverse hydraulic method" and their corresponding confidence levels (such as calculation error range and simulation consistency). Simultaneously record the coordinates / pipe segment identifiers of the M-1 most likely leak points obtained from the "multi-node data correlation location fusion algorithm" and their corresponding location accuracy indicators (such as correlation coefficient peak value and location error radius). An information matrix can be established, creating a table containing the location information (coordinates, pipe segment ID, distance from key nodes), source method, and the confidence / accuracy indicators provided by that method for each candidate point. Standardize and structure the results of the two location methods to provide a unified data foundation for subsequent fusion analysis, avoiding analytical biases caused by inconsistent data formats, and clarifying the reliability basis of each candidate point.

[0121] S4.4.2 Preliminary Fusion and Overlap Analysis

[0122] All candidate points (N+M in total) located by the two methods are overlaid on the pipeline network GIS map. Highly clustered areas (i.e., multiple points very close together) are identified through spatial clustering. Priority is given to points where both methods point to the same location, or where one method has extremely high confidence / accuracy (e.g., points with a location error radius much smaller than the pipe spacing). This quickly narrows down the potential leak locations spatially. "Multi-method cross-validation" reduces the limitations of a single method (e.g., reverse hydraulic methods may be affected by pipe topology, and multi-node algorithms may be affected by noise). Screening for clustered areas and high-confidence points can initially eliminate obviously unreasonable candidate points, reducing the complexity of subsequent analysis.

[0123] S4.4.3 Introducing on-site constraints (weighting factors)

[0124] Key constraint variables are identifying and quantifying on-site factors that may affect the actual location of leaks, such as: 1. Pipe material and aging degree: cast iron pipes are prone to leaks at joints, while PVC pipes are prone to leaks at damaged points; 2. Weak points in older pipes (such as severely corroded areas) have higher weight; 3. Joint / valve location: leaks are more frequent near joints or valves, and candidate points close to joints / valve should be given bonus points; 4. Burial depth and overburden load: areas that are too shallow or too deep, or areas with heavy loads (roads, buildings) above, have a higher risk of leakage; 5. Soil properties and corrosive environment: areas with loose soil or highly corrosive soil have a higher risk of leakage; 6. Historical leakage records: whether there are repair records nearby, and easily leaking pipe sections have higher weight; 7. External disturbances: whether there is recent construction nearby (excavation, piling, etc.), and potential third-party damage points should be given special attention; 8. Pipeline topology: stress concentration points such as elbows, tees, and reducers are more prone to damage. By combining the algorithm results with the actual conditions of the engineering site and giving higher weight to high-risk scenarios (such as old pipeline interfaces), the evaluation of candidate points can be made closer to the actual leakage patterns, reducing the contradiction that "the algorithm can accurately locate leaks but there are no leaks on site".

[0125] S4.4.4 Establish a comprehensive decision-making model:

[0126] Design a scoring system to quantitatively evaluate each candidate point, with weights for positioning accuracy (Wt), on-site conditions (Wc), and expert experience (We).

[0127] Calculate a composite score for each candidate point: Composite Score = Score t *Wt + Score c * Wc + Score e*We. By weightedly integrating "theoretical positioning accuracy," "actual on-site risks," and "expert experience," a quantitative balance for multi-dimensional decision-making is achieved. The weight allocation ensures the core role of algorithm accuracy (Wt has the highest weight) while also taking into account engineering practice principles (Wc) and tacit knowledge that is difficult to quantify (We), making the final result more practical.

[0128] It should be noted that if multiple candidate points are clustered in a certain area (even if they come from different methods), the overall score of that area (or its representative points) should be improved.

[0129] S4.4.5, Sorting and Preliminary Decision:

[0130] All candidate points are ranked according to their overall scores. The difference between the highest and second-highest score points is determined. If the difference is significant (e.g., much greater than the model error or the positioning accuracy itself), the highest score point can be preliminarily identified as the "most likely leak point." If the scores of the top few points are very close, several highly probable areas are considered to exist. When preparing the validation plan, these areas should be covered, and priorities should be clearly defined to avoid decision ambiguity due to too many candidate points. The determination of significant differences can quickly pinpoint a single leak point, while close scores indicate the need for further validation. This balances decision-making efficiency and accuracy, providing clear target areas for subsequent field validation.

[0131] Furthermore, non-invasive on-site verification can be employed, and methods such as sound correlation, gas tracer analysis, and infrared thermal imaging can be used to assist decision-making under specific conditions.

[0132] S4.4.1 to S4.4.6, through the logic of data standardization → spatial filtering → on-site weighting → quantitative decision-making → on-site verification, achieved a deep integration of theoretical algorithm positioning with actual engineering scenarios, ultimately improving the leak point positioning accuracy to about 0.3 meters, providing a reliable basis for precise excavation and repair, and solving the problems of single-method positioning being easily interfered with and disconnected from reality.

[0133] In summary, this invention achieves precise location by combining flow analysis with multiple technologies. High-precision flow meters and pressure sensors are installed at key nodes in the pipeline network to collect and preprocess data (synchronization, noise reduction, and cleaning). Steady-state and transient hydraulic models are constructed to eliminate interference such as water hammer. Based on flow balance, the flow difference between the inlet and outlet of each zone is calculated; exceeding a threshold indicates a suspected leak. Time-series methods such as minimum nighttime flow and cluster analysis are used to verify and pinpoint the actual leak area. Then, the flow gradient of the pipe section is calculated to determine the leaking pipe. A reverse hydraulic model is built based on the corrected flow rate, and through iteration, the residual difference between the simulated and actual flow rates is minimized, locating the leak point to the meter level. Finally, multiple sensors are deployed on the target pipeline, and a cross-correlation algorithm is used for initial location. After optimization such as cluster head selection and clock synchronization, the results of reverse hydraulic analysis and multi-node location are combined with on-site constraints (such as pipe material and historical records) for comprehensive decision-making, achieving a final location accuracy of approximately 0.3 meters. This scheme overcomes the problems of low accuracy and weak anti-interference in traditional methods through multi-stage progressive location and the integration of multiple technologies and on-site factors, achieving precise leak detection.

Claims

1. A method for detecting and locating leaks in a pipe network based on flow analysis, characterized in that, Includes the following steps: S1. Construct a flow balance model for the pipeline network; S2. Based on the flow balance model constructed in step S1, the flow status of each node in the pipeline network is detected in real time. When a pipeline leak is detected, the leak point is initially located. S3. Flow gradient analysis and reverse hydraulic fine-tuning: Within the suspected leak area identified in step S2, flow gradient analysis and a reverse hydraulic iterative model are used to achieve meter-level accuracy in locating the leak point. This includes: Based on the flow gradient analysis, the flow gradient between adjacent flow meters is calculated using the following formula: Where ∇Q represents the flow gradient, Q i Let Q be the instantaneous flow rate of the i-th flow meter. i+1 Let L be the instantaneous flow rate of the downstream adjacent flow meter, and L be the pipe section length between the two flow meters. If the flow gradient value of a certain pipe section is significantly greater than that of its upstream pipe section, then the leak point is determined to be located in that pipe section near the upstream flow meter. When the dynamic pressure change rate is greater than the threshold, the formula Q is used. comp =Q base +α·ΔP· Real-time calibration of flow rate values, based on the corrected flow rate Q comp To determine the location of the leak, a reverse hydraulic inversion model is established. This process includes the following steps: Assume the leak location x and the leak flow rate Q. leak As the parameter to be solved, a flow-leak correlation equation is established. For a pipe section containing a leak, the corrected flow rate monitored by the upstream and downstream flow meters satisfies the formula: Q comp , upstream =Q comp , downstream +Q leak ; wherein, Q comp , upstream Correcting the flow rate upstream of the leak point; the Q comp , downstream To correct the flow rate downstream of the leak, Q leak For leakage flow; With the objective of minimizing the residual between simulated flow and corrected flow, an objective function is constructed. Where x is the location of the leak point; Q sim,i (x) represents the simulated flow rate of the i-th flow meter when the leak point is assumed to be at location x. real,i The corrected actual flow rate of the flow meter is denoted as n, and the flow count of the target pipe section and the upstream and downstream related pipe sections is denoted as n. The nonlinear least squares optimization algorithm is used for iterative solution until the termination condition of residual accuracy meeting the standard or position change convergence is met is met. S4. Multiple sensor nodes are deployed on the leaking pipe determined in step S3 to collect detection data, and the location of the leak is determined according to the relevant positioning fusion algorithm. The relevant positioning fusion algorithm includes cross-correlation algorithm positioning and multi-node correlation algorithm fusion positioning. The cross-correlation algorithm for localization is as follows: When sensor A and sensor B are located on opposite sides of the leak point, the sampled leak vibration signals are defined as a(t) and b(t), respectively. Then, a(t) and b(t) have similarity; if a(t) lags b(t) by a certain time... Then approximately: Where A is the cross-correlation coefficient, the cross-correlation function of a(t) and b(t) is: ， When R ab When the maximum value is reached, the corresponding τ is , This is the time difference between the vibration wave from the leak reaching the two probes. If the sound velocity of the medium in the pipe is V, then the pipe length between the leak point and node B is L = (MV × T) / 2. At this point, the leak location between the two points is completed, and M is the distance between the two probes. The multi-node related algorithm fusion localization algorithm includes cluster head selection, clock synchronization algorithm, and multi-sensor data fusion localization algorithm; wherein, the cluster head is the node located at the center of the cluster and has the most remaining energy; The clock synchronization algorithm includes: the cluster head node broadcasts a data packet carrying a reference timestamp beacon to each node in the cluster; each node records the reference timestamp and its local time upon receiving the data packet; the adjacent nodes of the cluster head sequentially transmit an integrated data packet carrying the reference time and its own local time until the last node, which averages all local times and transmits them back to each node in reverse order; each node updates its local time; and the operation is repeated until the clock synchronization error is less than the threshold. The multi-sensor data fusion localization algorithm includes: under normal conditions where the water supply pipeline within the cluster has no branches, the leak sound signal intensity picked up by the nodes is grouped, paired, and sorted by number to ensure that the leak point location falls between the two nodes performing the cross-correlation algorithm. The number reflects the strength of the signal received by the paired node. The leak point location determined by all paired nodes is calculated according to the cross-correlation algorithm localization method. An improved weighted average algorithm is used to obtain the final location of the leak point from multiple localization results. The weighting factor is inversely proportional to the distance between the two paired nodes. During each fusion, the sorted numbers are sorted and abnormal weighting factors are removed to exclude paired nodes affected by sudden noise interference. Under special conditions where the pipeline within the cluster has branches, for multiple grouping and pairing results formed by node signal intensity, if the deviation of different group localization results is within the meter range, the average value is taken as the leak point location. If the deviation between the two points is outside the meter range, a comprehensive decision is made by combining the leak point determined by the flow gradient analysis and reverse hydraulics in step S3 to finally determine the accurate location of the pipeline leak.

2. The method according to claim 1, characterized in that, Step S1 specifically includes: S1.1 Data Acquisition: High-precision flow meters and pressure sensors are installed at key nodes of the pipeline water supply network. The large water supply area is divided into several independent metering areas. Flow meters are set at the inlet and outlet of each area, and pressure sensors are deployed at the same locations as the flow meters. S1.2 Data Preprocessing: Wavelet denoising is applied to the raw flow / pressure data, moving average filtering is performed on the pressure signal, and anomaly detection and cleaning are performed on the collected data; S1.3 Constructing a hydraulic model of the water supply network: Based on the network topology modeling, establish steady-state hydraulic models and transient models.

3. The method according to claim 2, characterized in that: In step S1.2, during data cleaning, gross error data is removed, and the isolated forest algorithm is used to identify complex anomaly patterns in the remaining data. In step S1.3, during transient modeling, a dynamic pressure monitoring and compensation strategy and a pressure wave propagation frequency domain filtering strategy are adopted. In the dynamic pressure monitoring and compensation strategy, when the pressure change rate is detected to be greater than a threshold, the steady-state model is dynamically corrected using the following formula: Q comp =Q base +α·ΔP· ; Among them, Q comp Q is the actual flow rate corrected for pressure changes. base It is the theoretical flow rate under conditions of no pressure fluctuations and external disturbances, where ΔP is the pressure change, Δt is the time step, α is the pipe elasticity correction coefficient, and L is the pipe length. In the frequency domain filtering strategy for pressure wave propagation, wavelet packet decomposition is performed on the raw flow signal collected by the flow meter to separate the high-frequency fluctuation component caused by pressure wave propagation. After removing the high-frequency component, the reconstructed signal is used as the input data for the steady-state model.

4. The method according to claim 1, characterized in that, In step S2, the difference between the inlet flow and the outlet flow of each zone is calculated based on the flow balance method. The calculation formula is as follows: ,in Q in This indicates the flow rate value of the inlet flow meter. Q out Q represents the flow rate value of the outlet flow meter. consumption This is the normal water consumption; when When this occurs, it is determined that the partition has a leak, where ε is a preset value.

5. The method according to claim 1, characterized in that, The comprehensive decision-making process in step S4 includes the following steps: S4.4.1 Data Preparation and Structuring: Collect the positioning results obtained from flow gradient analysis, reverse hydraulic method and multi-node data correlation positioning fusion algorithm, and establish an information matrix containing the location information of each candidate point, the source method and the corresponding confidence / accuracy index; S4.4.2 Preliminary Fusion and Overlap Analysis: Overlay all candidate points onto the pipeline network GIS map, identify highly spatially clustered areas, and filter points that both methods point to or points with extremely high confidence / accuracy. S4.4.3 Introduce field constraints: Quantify the field factors that affect the actual location of the leak, including pipe material and aging degree, interface / valve location, burial depth and soil load, soil properties and corrosive environment, historical leak records, external disturbances and pipeline topology. S4.4.4 Establish a comprehensive decision-making model, design a scoring system, and calculate a comprehensive score for each candidate point; S4.4.5 Sorting and Preliminary Decision-Making: Sorting by comprehensive score to determine the most likely leaks or multiple high-risk areas.

Images