A method and system for predicting risks in water supply pipelines

By constructing a three-field coupled control equation set and a deep learning model for multi-source monitoring data of water supply pipelines, combined with a graph attention network, the problem of missing physical mechanisms and fragmented multi-scale coupling relationships in pipeline risk prediction in existing technologies has been solved, enabling accurate assessment of pipeline risks and scientific operation and maintenance prioritization.

CN123117799APending Publication Date: 2026-11-17HONGJI JUNYE ENVIRONMENTAL PROTECTION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611320561.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing water supply pipeline risk prediction technologies suffer from a lack of physical mechanisms, fragmented multi-scale and multi-physical field coupling relationships, the ability to assess risk only for a single pipe, insufficient generalization, difficulty in quantifying failure probabilities and pipeline network cascading risks, and an inability to scientifically guide operation and maintenance prioritization.

Method used

By collecting multi-source monitoring data, a three-field coupled control equation set of electrochemical corrosion, elasticity, and fatigue damage is constructed. Combined with a deep learning model, multi-scale feature vectors are fused, and graph attention networks are used to simulate pipeline risk propagation and quantify single-pipe and system-level risks.

Benefits of technology

It improves the physical interpretability of pipeline risk prediction, captures early corrosion signals, reduces the deviation between prediction results and actual conditions, quantifies failure probability, and provides scientific operation and maintenance priority ranking.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117799A_ABST
    Figure CN123117799A_ABST
Patent Text Reader

Abstract

This invention discloses a risk prediction method and system for water supply pipelines. The method collects multi-source monitoring data on three types of water supply pipelines: microscopic corrosion, pipeline geometry, and hydraulic operation. It constructs a three-field coupled control equation set based on multiple mechanical theories; builds a physical constraint deep learning model, incorporating the residuals of the equation set into the loss function to constrain the model output; extracts three types of scale feature vectors from the monitoring data, designs a fusion operator combined with multi-head attention to obtain unified comprehensive features; constructs a limit state function based on the relationship between features and wall thickness evolution, and calculates the failure probability of a single pipe using the first-order second-moment method; constructs a graph attention network based on the pipeline network topology, simulates risk propagation to obtain the overall risk distribution of the pipeline network; finally, it classifies risk levels, calculates maintenance priorities, and outputs risk assessment and operation and maintenance scheduling schemes. This invention solves the problems of existing technologies, such as only being able to assess single-pipe risks independently, insufficient generalization, and difficulty in quantifying failure probabilities and pipeline network cascading risks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for water supply networks, specifically to a method and system for predicting risks in water supply pipelines. Background Technology

[0002] Water supply pipelines are a core lifeline infrastructure of cities, and the stable operation of the pipeline network directly guarantees the water supply for residents' lives and industrial production. With the continuous advancement of urbanization in China, the coverage of urban water supply networks has expanded year by year. Many aging pipelines, some over 10 years old, commonly suffer from corrosion, wall thickness reduction, and fatigue cracking. The average annual leakage rate of urban water supply networks nationwide exceeds 15%, and the leakage rate in older urban areas can reach 30%. Water waste and frequent pipe bursts and water outages are driving the widespread adoption of intelligent risk prediction technology in the industry. Current mainstream risk prediction solutions are based on pure data-driven machine learning algorithms, widely employing models such as random forests, support vector machines, basic neural networks, ensemble learning, and graph neural networks. They rely on smart water IoT to collect monitoring data such as pipeline pressure, flow rate, water quality, and basic pipeline attributes to complete pipeline risk level classification and failure probability estimation. This technology has been implemented in the daily operation and maintenance, leakage control, and pipeline renewal planning scenarios of water companies across the country. Some technical solutions, combined with digital twin and hydraulic simulation tools, train models using historical fault samples to achieve short-term prediction of pipeline faults, significantly reducing the cost of manual inspections and becoming the mainstream approach for pipeline risk assessment at present.

[0003] However, existing pure data-driven prediction schemes still suffer from five major, insurmountable flaws that hinder accurate, full-cycle, and systematic assessment of pipeline risks. First, they lack physical mechanisms, merely transforming pipeline risk into a data classification and fitting problem without incorporating underlying mechanisms of pipeline degradation such as corrosion electrochemical reactions, pipe wall stress, and material fatigue. The models are essentially "black boxes," with poor interpretability and significantly reduced generalization ability to new operating conditions outside the training set, making predictions prone to violating fundamental material and hydraulic principles. Second, they lack cross-scale correlation, failing to establish quantitative evolution relationships between microscopic corrosion current, mesoscopic wall thinning, and macroscopic hydraulic pressure fluctuations. Microscopic early corrosion anomalies cannot be transmitted to macroscopic risk assessment, resulting in the loss of early warning windows for pipelines. Third, they decouple multiple physics fields, modeling the corrosion field, stress field, and fatigue damage field in isolation, neglecting the relationship between corrosion thinning of the pipe wall and stress. The positive coupling feedback of "force concentration → accelerated fatigue cracking → further penetration of corrosive media" leads to a significant deviation between the risk assessment results and the actual deterioration state of the pipeline; fourth, the quantification of failure risk lacks reliable theoretical support, relying heavily on manually set empirical thresholds to classify risk levels, lacking analytical calculations of failure probability based on structural reliability theory, and failing to output quantitative indicators with confidence intervals, resulting in a lack of scientific quantitative basis for operation and maintenance resource scheduling; fifth, the lack of network-level risk spatiotemporal propagation modeling, calculating the risk of individual pipelines only, failing to simulate the chain reaction effect of failures based on the network topology, making it difficult to identify key weak nodes in the network, assess the system-level impact of pipeline bursts causing water outages, and support global emergency scheduling and preventive maintenance priority ranking.

[0004] Therefore, there is an urgent need for a risk prediction method for water supply pipelines to solve the problems of existing technologies, such as the inability to assess the risk of a single pipeline, insufficient generalization, and difficulty in quantifying the probability of failure and the risk of pipeline network cascading. Summary of the Invention

[0005] To address these issues, this invention provides a method and system for predicting risks in water supply pipelines, which solves problems such as the lack of a mechanism in existing pipeline risk prediction technologies, the fragmentation of multi-scale and multi-physical field coupling relationships, the inability to assess risks of a single pipe alone, insufficient generalization, difficulty in quantifying failure probabilities and pipeline network cascading risks, and the inability to scientifically guide operation and maintenance sequencing.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting risks in water supply pipelines, comprising:

[0007] Multi-source monitoring data of water supply pipelines are collected; the multi-source monitoring data includes three types of basic monitoring data: micro-scale electrochemical corrosion parameters, meso-scale pipeline geometric parameters, and macro-scale hydraulic operation parameters.

[0008] Based on the multi-source monitoring data, and according to electrochemical corrosion theory, elasticity, and fatigue damage mechanics, a set of three-field coupled control equations for corrosion field, stress field, and fatigue field is constructed.

[0009] A deep learning model with physical constraints is constructed, and the residuals of the three-field coupled control equations are added as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law.

[0010] Three types of scale feature vectors—microscale, mesoscale, and macroscale—are extracted from the multi-source monitoring data. A cross-scale feature fusion operator is designed based on the input dimension requirements of the deep learning model. Based on the cross-scale feature fusion operator, the three types of scale feature vectors are fused through a multi-head attention mechanism to generate a comprehensive feature vector with a unified dimension.

[0011] Based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations, a pipeline limit state function is constructed; based on the pipeline limit state function, an analytical expression for the failure probability of a single pipe segment considering multi-parameter uncertainties is derived through the first second-order moment method; the failure probability of each pipe segment is obtained by solving the analytical expression for the failure probability of a single pipe segment.

[0012] The actual topology of the water supply network is abstracted into a graph structure and a graph attention network is constructed. Based on the failure probability of each pipe segment, the chain propagation and evolution of risk along the pipe segments and nodes is simulated through the graph attention network, and the system-level overall risk distribution covering the entire network is obtained by solving the problem.

[0013] Based on the failure probability of each pipeline segment and the overall risk distribution at the system level, the pipeline risk level is classified and the maintenance priority index of each pipeline segment is calculated; the maintenance priority is sorted according to the maintenance priority index, and standardized pipeline risk assessment results and batch operation and maintenance scheduling schemes are output.

[0014] As a preferred embodiment of a risk prediction method for water supply pipelines, the expression of the three-field coupled control equations is as follows:

[0015]

[0016] In the formula, For pipe wall thickness; For time; is the molar mass of iron; The number of electrons transferred; It is Faraday's constant; The density of iron; For exchange current density; The anode transfer coefficient; The cathode transfer coefficient; This is an overpotential; It is the gas constant; Absolute temperature; It is circumferential stress; This is axial stress; The outer diameter of the pipe; This refers to the internal pressure of the pipeline. Spatial coordinates; This represents the cumulative value of fatigue damage. , These are material fatigue parameters; This is a function of the pressure cycle frequency.

[0017] As a preferred embodiment of a risk prediction method for water supply pipelines, the expression for the total loss function of the deep learning model is as follows:

[0018]

[0019]

[0020]

[0021] In the formula, This is the total loss function; For data-driven loss terms; This is the residual loss term in the partial differential equation; This refers to the boundary condition loss term; This is the initial condition loss term; , , These are the weighting coefficients; This represents the number of training samples; For the first Predicted wall thickness for each sample; For the first The true wall thickness of each sample; For the first The probability of prediction failure for each sample; For the first The true failure probability of a sample; and To balance the weights; Number of configuration points; This represents the corrosion current density predicted by the neural network.

[0022] As a preferred scheme for risk prediction of water supply pipelines, in the process of extracting feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data, electrochemical feature vectors are extracted from the microscale, geometric feature vectors are extracted from the mesoscale, and hydraulic feature vectors are extracted from the macroscale.

[0023] The expression for the electrochemical characteristic vector is:

[0024]

[0025] In the formula, These are electrochemical characteristic vectors; This is the corrosion potential; Polarization resistor; Chloride ion concentration; Dissolved oxygen concentration; pH of the aqueous solution;

[0026] The expression for the geometric feature vector is:

[0027]

[0028] In the formula, These are geometric eigenvectors; This refers to the length of the pipe section. The pipe inclination angle; This is the weld quality coefficient; This is the corrosion protection coating condition coefficient;

[0029] The expression for the hydraulic feature vector is:

[0030]

[0031] In the formula, This is a hydraulic feature vector; For traffic; For flow rate; For pressure drop; This refers to the frequency of pressure fluctuations. The standard deviation of pressure;

[0032] The expression for the comprehensive feature vector is:

[0033]

[0034] In the formula, This is a comprehensive feature vector; For the number of attention heads; This is the output weight matrix.

[0035] As a preferred embodiment of a risk prediction method for water supply pipelines, the expression for the pipeline limit state function is:

[0036]

[0037] In the formula, This is the limit state function of the pipeline; Minimum wall thickness allowed; This is the initial wall thickness; A vector of random variables;

[0038] The expression for the failure probability is:

[0039]

[0040] In the formula, This represents the probability of failure. The number of Monte Carlo samples; For indicator functions; For the first A vector of random variables from the sampled samples.

[0041] As a preferred embodiment of a risk prediction method for water supply pipelines, the expression for the overall system-level risk distribution is:

[0042]

[0043]

[0044] In the formula, This is the system-level overall risk distribution matrix; The number of network layers; For the first Risk state matrix of the layer; For the number of propagation steps; For the first Learnable parameters for each step; For activation functions; It is an adjacency matrix; It is a degree matrix.

[0045] As a preferred embodiment of a risk prediction method for water supply pipelines, the pipeline risk level is:

[0046]

[0047] In the formula, This represents the probability of failure. , , For the preset threshold, The range of values ​​is , The range of values ​​is , The range of values ​​is ;

[0048] The formula for calculating the maintenance priority index is as follows:

[0049]

[0050] In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .

[0051] This invention also provides a water supply pipeline risk prediction system, employing the above-mentioned water supply pipeline risk prediction method, including:

[0052] The data acquisition module is used to acquire multi-source monitoring data of the water supply pipeline; the multi-source monitoring data includes three types of basic monitoring data: micro-scale electrochemical corrosion parameters, meso-scale pipeline geometric parameters, and macro-scale hydraulic operation parameters.

[0053] The three-field coupled control equations construction module is used to construct a three-field coupled control equation set of corrosion field, stress field and fatigue field based on the multi-source monitoring data and according to electrochemical corrosion theory, elasticity and fatigue damage mechanics.

[0054] The deep learning model building module is used to build a physically constrained deep learning model and adds the residuals of the three-field coupled control equations as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law.

[0055] The multi-scale feature vector extraction and fusion module is used to extract feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data. Based on the input dimension requirements of the deep learning model, a cross-scale feature fusion operator is designed. Based on the cross-scale feature fusion operator, the three scale feature vectors are fused through a multi-head attention mechanism to generate a comprehensive feature vector of a unified dimension.

[0056] The failure probability calculation module is used to construct the pipeline limit state function based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations; based on the pipeline limit state function, it derives the analytical expression for the failure probability of a single pipe segment considering multiple parameter uncertainties through the first second-order moment method; and obtains the failure probability of each pipe segment by solving the analytical expression for the failure probability of a single pipe segment.

[0057] The system-level overall risk distribution acquisition module is used to abstract the actual topology of the water supply network into a graph structure and construct a graph attention network; based on the failure probability of each pipe segment, the graph attention network is used to simulate the chain propagation and evolution process of risk along the pipe segments and nodes of the network, and solve to obtain the system-level overall risk distribution covering the entire network.

[0058] The risk assessment and operation and maintenance scheduling scheme acquisition module is used to classify pipeline risk levels and calculate the maintenance priority index of each pipeline segment based on the failure probability of each pipeline segment and the overall risk distribution of the system level; sort the maintenance priorities according to the maintenance priority index, and output standardized pipeline risk assessment results and batch operation and maintenance scheduling schemes.

[0059] As a preferred solution for a water supply pipeline risk prediction system, the expression of the three-field coupled control equations in the three-field coupled control equations construction module is as follows:

[0060]

[0061] In the formula, For pipe wall thickness; For time; is the molar mass of iron; The number of electrons transferred; It is Faraday's constant; The density of iron; For exchange current density; The anode transfer coefficient; The cathode transfer coefficient; This is an overpotential; It is the gas constant; Absolute temperature; It is circumferential stress; This is axial stress; The outer diameter of the pipe; This refers to the internal pressure of the pipeline. Spatial coordinates; This represents the cumulative value of fatigue damage. , These are material fatigue parameters; This is a function of the pressure cycle frequency.

[0062] As a preferred solution for a water supply pipeline risk prediction system, in the deep learning model construction module, the expression for the total loss function of the deep learning model is as follows:

[0063]

[0064]

[0065]

[0066] In the formula, This is the total loss function; For data-driven loss terms; This is the residual loss term in the partial differential equation; This refers to the boundary condition loss term; This is the initial condition loss term; , , These are the weighting coefficients; This represents the number of training samples; For the first Predicted wall thickness for each sample; For the first The true wall thickness of each sample; For the first The probability of prediction failure for each sample; For the first The true failure probability of a sample; and To balance the weights; Number of configuration points; This represents the corrosion current density predicted by the neural network.

[0067] As a preferred solution for a water supply pipeline risk prediction system, the multi-scale feature vector extraction and fusion module extracts feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data, respectively, extracting electrochemical feature vectors from the microscale, geometric feature vectors from the mesoscale, and hydraulic feature vectors from the macroscale.

[0068] The expression for the electrochemical characteristic vector is:

[0069]

[0070] In the formula, These are electrochemical characteristic vectors; This is the corrosion potential; Polarization resistor; Chloride ion concentration; Dissolved oxygen concentration; pH of the aqueous solution;

[0071] The expression for the geometric feature vector is:

[0072]

[0073] In the formula, These are geometric eigenvectors; This refers to the length of the pipe section. The pipe inclination angle; This is the weld quality coefficient; This is the corrosion protection coating condition coefficient;

[0074] The expression for the hydraulic feature vector is:

[0075]

[0076] In the formula, This is a hydraulic feature vector; For traffic; For flow rate; For pressure drop; This refers to the frequency of pressure fluctuations. The standard deviation of pressure;

[0077] The expression for the comprehensive feature vector is:

[0078]

[0079] In the formula, This is a comprehensive feature vector; For the number of attention heads; This is the output weight matrix.

[0080] As a preferred embodiment of a water supply pipeline risk prediction system, the expression for the pipeline limit state function in the failure probability calculation module is as follows:

[0081]

[0082] In the formula, This is the limit state function of the pipeline; Minimum wall thickness allowed; This is the initial wall thickness; A vector of random variables;

[0083] The expression for the failure probability is:

[0084]

[0085] In the formula, This represents the probability of failure. The number of Monte Carlo samples; For indicator functions; For the first A vector of random variables from the sampled samples.

[0086] As a preferred embodiment of a water supply pipeline risk prediction system, the system-level overall risk distribution acquisition module contains the following expression for the system-level overall risk distribution:

[0087]

[0088]

[0089] In the formula, This is the system-level overall risk distribution matrix; The number of network layers; For the first Risk state matrix of the layer; For the number of propagation steps; For the first Learnable parameters for each step; For activation functions; It is an adjacency matrix; It is a degree matrix.

[0090] As a preferred solution for a water supply pipeline risk prediction system, the pipeline risk level in the risk assessment and operation and maintenance scheduling scheme acquisition module is:

[0091]

[0092] In the formula, This represents the probability of failure. , , For the preset threshold, The range of values ​​is , The range of values ​​is , The range of values ​​is ;

[0093] The formula for calculating the maintenance priority index is as follows:

[0094]

[0095] In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .

[0096] The present invention has the following advantages:

[0097] First, this invention combines electrochemical and mechanical theories with deep learning, relying on physical equations to constrain the model output, thus overcoming the limitations of a pure data black box and improving the model's interpretability and adaptability to unfamiliar working conditions.

[0098] Second, this invention connects micro, meso, and macro parameters and integrates multi-scale information by relying on feature fusion operators, which can capture early subtle corrosion signals in pipelines and extend the pipeline risk warning cycle.

[0099] Third, this invention establishes a set of three coupled equations to restore the positive feedback relationship between corrosion, stress, and fatigue, which closely matches the actual deterioration process of pipelines and reduces the deviation between the predicted results and the actual state of the pipeline.

[0100] Fourth, this invention uses reliability theory to calculate failure probability and combines it with graph networks to characterize the risk propagation law of pipeline network, which can quantify the risk of individual pipes and the overall pipeline network, providing an objective basis for the priority division of operation and maintenance.

[0101] Fifth, this invention sets up a model adaptive weight adjustment mechanism, which can automatically balance data fitting and physical constraint strength, adapt to pipeline prediction needs under different water quality and operating conditions, and has a wider range of applications. Attached Figure Description

[0102] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0103] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0104] Figure 1 This is a flowchart illustrating a water supply pipeline risk prediction method provided in Embodiment 1 of the present invention;

[0105] Figure 2 This is a schematic diagram of the physical constraint deep learning model architecture in a water supply pipeline risk prediction method provided in Embodiment 1 of the present invention;

[0106] Figure 3 This is a schematic diagram of the cross-scale feature fusion operator in a water supply pipeline risk prediction method provided in Embodiment 1 of the present invention;

[0107] Figure 4 This is a schematic diagram of a pipeline risk propagation model based on graph attention network in a water supply pipeline risk prediction method provided in Embodiment 1 of the present invention;

[0108] Figure 5 This is a schematic diagram of the architecture of a water supply pipeline risk prediction system provided in Embodiment 2 of the present invention. Detailed Implementation

[0109] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0110] Example 1

[0111] See Figure 1 Embodiment 1 of the present invention provides a method for predicting risks in water supply pipelines, comprising the following steps:

[0112] S1. Acquire multi-source monitoring data of the water supply pipeline by collecting data; the multi-source monitoring data includes three types of basic monitoring data: micro-scale electrochemical corrosion parameters, meso-scale pipeline geometric parameters, and macro-scale hydraulic operation parameters.

[0113] S2. Based on the multi-source monitoring data, and according to electrochemical corrosion theory, elasticity and fatigue damage mechanics, a set of three-field coupled control equations for corrosion field, stress field and fatigue field is constructed.

[0114] S3. Construct a deep learning model with physical constraints, and add the residuals of the three-field coupled control equations as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law.

[0115] S4. Extract feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data. Design a cross-scale feature fusion operator based on the input dimension requirements of the deep learning model. Based on the cross-scale feature fusion operator, fuse the three scale feature vectors through a multi-head attention mechanism to generate a comprehensive feature vector of a unified dimension.

[0116] S5. Based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations, construct the pipeline limit state function; based on the pipeline limit state function, derive the analytical expression for the failure probability of a single pipe segment considering multiple parameter uncertainties through the first second-order moment method; obtain the failure probability of each pipe segment by solving the analytical expression for the failure probability of a single pipe segment.

[0117] S6. Abstract the actual topology of the water supply network into a graph structure and construct a graph attention network; based on the failure probability of each pipe segment, simulate the chain propagation and evolution of risk along the pipe segments and nodes through the graph attention network, and solve to obtain the system-level overall risk distribution covering the entire network.

[0118] S7. Based on the failure probability of each pipeline segment and the overall risk distribution at the system level, classify the pipeline risk level and calculate the maintenance priority index for each pipeline segment; sort the maintenance priorities according to the maintenance priority index, and output the standardized pipeline risk assessment results and the batch operation and maintenance scheduling plan.

[0119] In this embodiment, in step S1, multi-source monitoring data of the water supply pipeline is acquired by collecting data; the multi-source monitoring data includes three types of basic monitoring data: microscale electrochemical corrosion parameters, mesoscale pipeline geometric parameters, and macroscale hydraulic operation parameters.

[0120] Specifically, multi-source heterogeneous monitoring data is collected through various sensors deployed in the water supply network.

[0121] The microscale data includes: corrosion current density. (Measured via an electrochemical workstation, range 0-100 μA / cm²), corrosion potential (Reference electrode method, accuracy ±1 mV), polarization resistance (Linear polarization method, range 0-10 kΩ·cm²), water quality parameters pH value (glass electrode method, accuracy ±0.01), chloride ion concentration (Ion-selective electrode method, range 0-1000 mg / L), temperature (PT100 resistance temperature detector, accuracy ±0.1℃), dissolved oxygen (Fluorescence method, range 0-20 mg / L).

[0122] Mesoscale data includes: pipe wall thickness (Ultrasonic thickness gauge, accuracy ±0.1 mm), pipe outer diameter (Measured with vernier calipers, accuracy ±0.5 mm), pipe section length (Obtained from GIS system, accuracy ±0.1 m), pipe inclination angle (Tilt sensor, accuracy ±0.1°), weld quality coefficient (Based on radiographic testing rating, with values ​​ranging from 0.6 to 1.0), Anti-corrosion coating condition coefficient (Based on electric spark detection, the value is 0.5-1.0).

[0123] Macro-scale data includes: pipeline internal pressure (Pressure transmitter, range 0-2 MPa, accuracy ±0.25%), flow rate (Electromagnetic flowmeter, accuracy ±0.5%), flow velocity (Calculated from flow rate and pipe diameter), pressure drop (Differential pressure transmitter, accuracy ±0.1%), pressure fluctuation frequency (Obtained from FFT analysis), pressure standard deviation (Obtained through statistical calculations).

[0124] All sensor data is aggregated to the edge computing node via an industrial IoT gateway, preprocessed, and then uploaded to the cloud data center. The data sampling frequency is set according to the parameter characteristics: electrochemical parameters once per minute, geometric parameters once per day (or periodic inspection), and hydraulic parameters once per second.

[0125] In this embodiment, in step S2, based on the multi-source monitoring data, a set of three-field coupled control equations for corrosion field, stress field, and fatigue field is constructed according to electrochemical corrosion theory, elasticity, and fatigue damage mechanics.

[0126] Specifically, the relationship between corrosion current density and electrode potential is established based on the Butler-Volmer equation:

[0127]

[0128] In the formula, This represents the corrosion current density, expressed in A / m². The exchange current density is expressed in A / m². The anode transfer coefficient; The cathode transfer coefficient; The number of electrons transferred; This is the Faraday constant, with a value of 96485 C / mol; This is an overpotential, measured in volts (V). The gas constant is 8.314 J / (mol·K); This is absolute temperature, measured in Kelvin (K).

[0129] The relationship between corrosion current density and wall thickness reduction rate is established based on Faraday's law:

[0130]

[0131] In the formula, The pipe wall thickness is expressed in meters (m). Time, in seconds; The molar mass of iron is 0.0558 kg / mol. The density of iron is 7870 kg / m³.

[0132] Based on the theory of thin-walled cylinders in elasticity, the stress distribution equation for the pipeline is established:

[0133]

[0134] In the formula, This is the circumferential stress, expressed in Pa. This represents axial stress, measured in Pa. Radial stress, in Pa; This refers to the internal pressure of the pipeline, expressed in Pa. This represents the outer diameter of the pipe, in meters (m).

[0135] A fatigue damage evolution equation is established based on Miner's linear cumulative damage theory:

[0136]

[0137] In the formula, This represents the cumulative value of fatigue damage. For the first The number of cycles in the stress cycle; For the first Fatigue life at stress level 1, in cycles; For the first Level stress amplitude, in Pa; and These are material fatigue parameters, determined experimentally.

[0138] By combining the above equations, a three-field coupled control equation system is established, with the pipe wall thickness as the reference value. For coupling variables:

[0139]

[0140] In the formula, Spatial coordinates; The pressure cycle frequency is a function of the pipe wall thickness. As a coupling variable, it simultaneously affects stress distribution and fatigue damage accumulation rate.

[0141] This set of equations reveals the coupling mechanism between corrosion, stress, and fatigue: corrosion reduces wall thickness → reduced wall thickness increases stress → increased stress accelerates fatigue damage → fatigue cracks promote the penetration of corrosive media → corrosion is further accelerated. This positive feedback mechanism is the core driving force of pipeline failure evolution.

[0142] In this embodiment, in step S3, a deep learning model with physical constraints is constructed, and the residuals of the three-field coupled control equations are added as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law.

[0143] Specifically, such as Figure 2 As shown, a neural network with an encoder-decoder architecture is constructed. The encoder extracts feature representations from multi-source monitoring data, and the decoder outputs the predicted values ​​of the remaining pipe wall thickness and failure probability. The residuals of the three-field coupled control equations are added as regularization terms to the loss function.

[0144] The neural network architecture is designed as follows:

[0145] The encoder section receives 19-dimensional input vectors: 7-dimensional microscopic features, 6-dimensional mesoscopic features, and 6-dimensional macroscopic features. The first hidden layer is a fully connected layer with 128 neurons, using the ReLU activation function. The second hidden layer is a fully connected layer with 64 neurons, also using the ReLU activation function.

[0146] Decoder section: The third hidden layer is a fully connected layer with 64 neurons, using the ReLU activation function. The fourth hidden layer is a fully connected layer with 32 neurons, also using the ReLU activation function. The output layer has 2 neurons, outputting the predicted wall thickness. and failure probability A linear activation function is used.

[0147] Define the physical constraint loss function:

[0148]

[0149] In the formula, This is the total loss function; For data-driven loss terms; This is the residual loss term in the partial differential equation; This refers to the boundary condition loss term; This is the initial condition loss term; , , These are weighting coefficients, with values ​​ranging from [0.01, 1.0].

[0150] The data-driven loss term is defined as follows:

[0151]

[0152] In the formula, This represents the number of training samples; For the first Predicted wall thickness for each sample; For the first The true wall thickness of each sample; For the first The probability of prediction failure for each sample; For the first The true failure probability of a sample; and To balance the weights, all values ​​are set to 0.5.

[0153] The residual loss term in a partial differential equation is defined as:

[0154]

[0155] In the formula, Number of configuration points; This represents the corrosion current density predicted by the neural network.

[0156] Residual loss term in partial differential equations Select 1000 collocation points within the computational domain and calculate the sum of squares of the equation residuals. Boundary condition loss term. Physical boundary conditions constraining both ends and the inner and outer surfaces of the pipe. Initial condition loss term. constraint The initial wall thickness distribution at time t.

[0157] In this embodiment, the model training uses the Adam optimizer with a learning rate of 0.001, a batch size of 64, and 5000 training epochs. The training dataset contains 10,000 samples from 5 years of monitoring data from a city's water supply network. The validation set comprises 20% and is used for early stopping to prevent overfitting.

[0158] In one possible embodiment, the adaptive adjustment mechanism for the weight coefficients of the physical constraint loss function is further optimized. (Weight coefficients) , , It is no longer fixed, but dynamically adjusted according to the gradient during training:

[0159]

[0160] In the formula, To adjust the step size. When the PDE residual gradient is much larger than the data loss gradient, the step size is automatically reduced. Avoid training instability; conversely, increase [the risk of instability]. Strengthen physical constraints.

[0161] This adaptive mechanism enables the model to automatically balance data fit and physical consistency at different training stages, improving convergence speed and final accuracy.

[0162] In this embodiment, in step S4, feature vectors of three scales—microscale, mesoscale, and macroscale—are extracted from the multi-source monitoring data. Based on the input dimension requirements of the deep learning model, a cross-scale feature fusion operator is designed. Based on the cross-scale feature fusion operator, the three scale feature vectors are fused through a multi-head attention mechanism to generate a comprehensive feature vector of a unified dimension.

[0163] Specifically, such as Figure 3 As shown, firstly, feature vectors are extracted from three scales:

[0164] Extracting electrochemical feature vectors at the microscale:

[0165]

[0166] In the formula, These are electrochemical characteristic vectors; This is the corrosion potential, measured in volts (V). This is the polarization resistance, measured in Ω·m². This represents chloride ion concentration, expressed in mg / L. Dissolved oxygen concentration, in mg / L; The pH value is the pH of the aqueous solution.

[0167] Extracting geometric feature vectors from the mesoscopic scale:

[0168]

[0169] In the formula, These are geometric eigenvectors; This refers to the length of the pipe section, in meters (m). The pipe inclination angle is expressed in rad. This is the weld quality coefficient; This is the corrosion protection coating condition coefficient.

[0170] Extracting hydraulic feature vectors from a macroscopic scale:

[0171]

[0172] In the formula, This is a hydraulic feature vector; The flow rate is expressed in m³ / s. The velocity is expressed in m / s. Pressure drop, unit: Pa; The pressure fluctuation frequency is expressed in Hz. The pressure standard deviation is expressed in Pa.

[0173] A linear transformation maps features from three scales to a unified dimension:

[0174]

[0175] In the formula, , , This is a learnable weight matrix.

[0176] Through three learnable weight matrices , , (All dimensions are) This maps features to the query, key, and value space.

[0177] Then, calculate the multi-head attention weights:

[0178]

[0179] In the formula, is the dimension of the key vector.

[0180] The advantage of multi-head attention mechanisms is that they can simultaneously focus on different combinations of features at different scales. For example, some heads may focus on the correlation between microscopic corrosion parameters and macroscopic pressure fluctuations, while others may focus on the correlation between mesoscopic geometric parameters and microscopic water quality.

[0181] The cross-scale fusion feature vector is:

[0182]

[0183] In the formula, This is a comprehensive feature vector; For the number of attention heads; This is the output weight matrix.

[0184] In this embodiment, in step S5, a pipeline limit state function is constructed based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations; based on the pipeline limit state function, an analytical expression for the failure probability of a single pipe segment considering multi-parameter uncertainties is derived using the first second-order moment method; and the failure probability of each pipe segment is obtained by solving the analytical expression for the failure probability of a single pipe segment.

[0185] Specifically, define the limit state function:

[0186]

[0187] In the formula, This is the limit state function of the pipeline; Minimum wall thickness allowed; This is the initial wall thickness; Let be a vector of random variables.

[0188] The failure probability is defined as:

[0189]

[0190] In the formula, Let be the joint probability density function of the random variables.

[0191] The failure probability is approximated using the first second-order moment method (FOSM):

[0192]

[0193] In the formula, The cumulative distribution function of the standard normal distribution; This is a reliability indicator.

[0194] The formula for calculating the reliability index is:

[0195]

[0196] In the formula, The mean of the limit state function; The standard deviation of the limit state function; For expectation operators; This is the variance operator.

[0197] Considering the uncertainty of random variables, Monte Carlo integration is used:

[0198]

[0199] In the formula, This represents the probability of failure. The number of Monte Carlo samples; This is an indicator function that takes the value 1 when the condition is met and 0 otherwise. For the first A vector of random variables from the sampled samples.

[0200] In this embodiment, the advantage of the Monte Carlo method is that it can handle randomly distributed variables and nonlinear limit state functions, but its computational cost is high. In practical applications, FOSM can be used to quickly screen high-risk pipe sections first, and then Monte Carlo can be used to accurately calculate the risks of high-risk pipe sections.

[0201] In this embodiment, in step S6, the actual topology of the water supply network is abstracted into a graph structure and a graph attention network is constructed. Based on the failure probability of each pipe segment, the chain propagation and evolution process of risk along the pipe segments and nodes is simulated through the graph attention network to obtain the system-level overall risk distribution covering the entire network.

[0202] Specifically, such as Figure 4 As shown in the figure, the water supply network is abstracted into a diagram. ,in This is a set of nodes, representing pipeline junctions, valves, and pumping stations. Let be the set of edges, representing pipe segments.

[0203] Constructing the node feature matrix ,in The number of nodes; For feature dimensions.

[0204] Constructing an adjacency matrix :

[0205]

[0206] Calculate the degree matrix :

[0207]

[0208] Define the graph attention layer:

[0209]

[0210] In the formula, For nodes Updated feature representation; For nodes The set of neighbors; The weight matrix is ​​a learnable weight matrix; This represents the attention coefficient.

[0211] The attention coefficient is calculated as follows:

[0212]

[0213] In the formula, A learnable attention vector; This indicates vector concatenation.

[0214] In this embodiment, the risk propagation process is modeled as follows:

[0215]

[0216] In the formula, For the first Risk state matrix of the layer; For the number of propagation steps; For the first Learnable parameters for each step; This is the activation function.

[0217] In this embodiment, the system-level risk distribution is as follows:

[0218]

[0219] In the formula, This is the system-level overall risk distribution matrix; This represents the number of network layers.

[0220] In this embodiment, in step S7, based on the failure probability of each pipeline segment and the overall risk distribution at the system level, the pipeline risk level is divided and the maintenance priority index of each pipeline segment is calculated; the maintenance priority is sorted according to the maintenance priority index, and the standardized pipeline risk assessment result and the batch operation and maintenance scheduling plan are output.

[0221] Specifically, define risk level classification criteria based on failure probability. Risk levels are determined by comparing the results with preset threshold ranges.

[0222]

[0223] In the formula, This represents the probability of failure. , , For the preset threshold, The range of values ​​is , The range of values ​​is , The range of values ​​is .

[0224] Calculate the maintenance priority index:

[0225]

[0226] In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .

[0227] according to The maintenance priority list is generated by sorting the pipe sections in descending order. The top 10 pipe sections are included in the maintenance plan for the next quarter, the pipe sections from top 11 to 50 are included in the annual maintenance plan, and the remaining pipe sections are continuously monitored.

[0228] In one possible embodiment, the invention can be extended to water supply networks with mixed pipe materials:

[0229] For pipes made of different materials such as ductile iron pipes, steel pipes, PE pipes, and PVC pipes, material-specific constitutive relations are established respectively:

[0230] Ductile iron pipe: adopts a linear elastic model, elastic modulus GPa, Poisson's ratio ;

[0231] Steel pipe: using an elastic-plastic model, yield strength MPa, hardening modulus GPa;

[0232] PE pipe: using a viscoelastic model, relaxation time s, long-term modulus GPa;

[0233] PVC pipe: uses a linear elastic model, elastic modulus GPa, Poisson's ratio .

[0234] The corrosion model is also adjusted according to the material: for metal pipes (ductile iron, steel), an electrochemical corrosion model is used. For non-metallic pipes (PE, PVC), an aging degradation model based on the Arrhenius equation is used. The material identification module automatically selects the corresponding constitutive relation and corrosion model, enabling unified risk assessment of multi-material pipe networks.

[0235] The application scenarios of this invention are as follows:

[0236] In the daily operation and maintenance of old urban water supply networks, this invention integrates multi-scale monitoring parameters and multi-physical field mechanisms to output pipeline risk classification and maintenance sequencing schemes, assisting water personnel in carrying out planned maintenance.

[0237] In the scenario of planning and evaluating pipeline networks in newly built urban areas, this invention can simulate the deterioration trend of pipelines under long-term hydraulic and corrosive effects, providing a risk reference for pipe material selection and pipeline routing.

[0238] In emergency dispatch scenarios involving burst water supply pipelines, this invention relies on graph attention networks to extrapolate the range of risk chain propagation, helping staff predict the affected areas and optimize the allocation of emergency response resources.

[0239] In the long-term monitoring and control of municipal underground pipelines, this invention captures early abnormal signals of micro-corrosion, identifies potentially deteriorated pipe sections in advance, and extends the preparation cycle for preventive maintenance of pipelines.

[0240] In the scenario of unified assessment of multi-material mixed pipeline networks, this invention is adapted to the corrosion and mechanical models corresponding to steel pipes, ductile iron, and plastic pipes, so as to realize integrated risk assessment of the entire pipeline network.

[0241] In the context of supporting analysis scenarios for water affairs digital twin platforms, this invention outputs failure probability data with uncertainty quantification, enriching the visualization content of pipeline safety status in digital twin systems.

[0242] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0243] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0244] Example 2

[0245] See Figure 5 Embodiment 2 of the present invention also provides a water supply pipeline risk prediction system, comprising:

[0246] Data acquisition module 001 is used to acquire multi-source monitoring data of water supply pipelines; the multi-source monitoring data includes three types of basic monitoring data: micro-scale electrochemical corrosion parameters, meso-scale pipeline geometric parameters, and macro-scale hydraulic operation parameters.

[0247] The three-field coupled control equations construction module 002 is used to construct a three-field coupled control equations of corrosion field, stress field and fatigue field based on the multi-source monitoring data and according to electrochemical corrosion theory, elasticity and fatigue damage mechanics.

[0248] The deep learning model building module 003 is used to build a physically constrained deep learning model and adds the residuals of the three-field coupled control equations as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law.

[0249] The multi-scale feature vector extraction and fusion module 004 is used to extract three types of scale feature vectors—microscale, mesoscale, and macroscale—from the multi-source monitoring data; design a cross-scale feature fusion operator according to the input dimension requirements of the deep learning model; and fuse the three types of scale feature vectors through a multi-head attention mechanism based on the cross-scale feature fusion operator to generate a comprehensive feature vector with a unified dimension.

[0250] The failure probability calculation module 005 is used to construct the pipeline limit state function based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations; based on the pipeline limit state function, it derives the analytical expression for the failure probability of a single pipe segment considering multiple parameter uncertainties through the first second moment method; and obtains the failure probability of each pipe segment by solving the analytical expression for the failure probability of a single pipe segment.

[0251] The system-level overall risk distribution acquisition module 006 is used to abstract the actual topology of the water supply network into a graph structure and construct a graph attention network; based on the failure probability of each pipe segment, the graph attention network is used to simulate the chain propagation and evolution process of risk along the pipe segments and nodes of the network, and solve to obtain the system-level overall risk distribution covering the entire network.

[0252] The risk assessment and operation and maintenance scheduling scheme acquisition module 007 is used to classify pipeline risk levels and calculate the maintenance priority index of each pipeline segment based on the failure probability of each pipeline segment and the overall risk distribution of the system level; sort the maintenance priorities according to the maintenance priority index, and output standardized pipeline risk assessment results and batch operation and maintenance scheduling schemes.

[0253] In this embodiment, in the three-field coupling control equations construction module 002, the expression of the three-field coupling control equations is:

[0254]

[0255] In the formula, For pipe wall thickness; For time; is the molar mass of iron; The number of electrons transferred; It is Faraday's constant; The density of iron; For exchange current density; The anode transfer coefficient; The cathode transfer coefficient; This is an overpotential; It is the gas constant; Absolute temperature; It is circumferential stress; This is axial stress; The outer diameter of the pipe; This refers to the internal pressure of the pipeline. Spatial coordinates; This represents the cumulative value of fatigue damage. , These are material fatigue parameters; This is a function of the pressure cycle frequency.

[0256] In this embodiment, in the deep learning model construction module 003, the expression for the total loss function of the deep learning model is:

[0257]

[0258]

[0259]

[0260] In the formula, This is the total loss function; For data-driven loss terms; This is the residual loss term in the partial differential equation; This refers to the boundary condition loss term; This is the initial condition loss term; , , These are the weighting coefficients; This represents the number of training samples; For the first Predicted wall thickness for each sample; For the first The true wall thickness of each sample; For the first The probability of prediction failure for each sample; For the first The true failure probability of a sample; and To balance the weights; Number of configuration points; This represents the corrosion current density predicted by the neural network.

[0261] In this embodiment, the multi-scale feature vector extraction and fusion module 004 extracts feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data, respectively, extracting electrochemical feature vectors from the microscale, geometric feature vectors from the mesoscale, and hydraulic feature vectors from the macroscale.

[0262] The expression for the electrochemical characteristic vector is:

[0263]

[0264] In the formula, These are electrochemical characteristic vectors; This is the corrosion potential; Polarization resistor; Chloride ion concentration; Dissolved oxygen concentration; pH of the aqueous solution;

[0265] The expression for the geometric feature vector is:

[0266]

[0267] In the formula, These are geometric eigenvectors; This refers to the length of the pipe section. The pipe inclination angle; This is the weld quality coefficient; This is the corrosion protection coating condition coefficient;

[0268] The expression for the hydraulic feature vector is:

[0269]

[0270] In the formula, This is a hydraulic feature vector; For traffic; For flow rate; For pressure drop; This refers to the frequency of pressure fluctuations. The standard deviation of pressure;

[0271] The expression for the comprehensive feature vector is:

[0272]

[0273] In the formula, This is a comprehensive feature vector; For the number of attention heads; This is the output weight matrix.

[0274] In this embodiment, the expression for the pipeline limit state function in the failure probability calculation module 005 is:

[0275]

[0276] In the formula, This is the limit state function of the pipeline; Minimum wall thickness allowed; This is the initial wall thickness; A vector of random variables;

[0277] The expression for the failure probability is:

[0278]

[0279] In the formula, This represents the probability of failure. The number of Monte Carlo samples; For indicator functions; For the first A vector of random variables from the sampled samples.

[0280] In this embodiment, the expression for the system-level overall risk distribution in the system-level overall risk distribution acquisition module 006 is:

[0281]

[0282]

[0283] In the formula, This is the system-level overall risk distribution matrix; The number of network layers; For the first Risk state matrix of the layer; For the number of propagation steps; For the first Learnable parameters for each step; For activation functions; It is an adjacency matrix; It is a degree matrix.

[0284] In this embodiment, the pipeline risk level in the risk assessment and operation and maintenance scheduling scheme acquisition module 007 is:

[0285]

[0286] In the formula, This represents the probability of failure. , , For the preset threshold, The range of values ​​is , The range of values ​​is , The range of values ​​is ;

[0287] The formula for calculating the maintenance priority index is as follows:

[0288]

[0289] In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .

[0290] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0291] Example 3

[0292] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a water supply pipeline risk prediction method. The program code includes instructions for executing the water supply pipeline risk prediction method of Embodiment 1 or any possible implementation thereof.

[0293] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0294] Example 4

[0295] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0296] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute a water supply pipeline risk prediction method according to Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0297] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0298] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0299] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0300] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A water supply pipe risk prediction method characterized by, include: Multi-source monitoring data of water supply pipelines were collected; The multi-source monitoring data includes three types of basic monitoring data: microscale electrochemical corrosion parameters, mesoscale pipeline geometric parameters, and macroscale hydraulic operation parameters. Based on the multi-source monitoring data, and according to electrochemical corrosion theory, elasticity, and fatigue damage mechanics, a set of three-field coupled control equations for corrosion field, stress field, and fatigue field is constructed. A deep learning model with physical constraints is constructed, and the residuals of the three-field coupled control equations are added as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law. Three types of scale feature vectors—microscale, mesoscale, and macroscale—are extracted from the multi-source monitoring data. A cross-scale feature fusion operator is designed based on the input dimension requirements of the deep learning model. Based on the cross-scale feature fusion operator, the three types of scale feature vectors are fused through a multi-head attention mechanism to generate a comprehensive feature vector with a unified dimension. Based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations, a pipeline limit state function is constructed; based on the pipeline limit state function, an analytical expression for the failure probability of a single pipe segment considering multi-parameter uncertainties is derived through the first second-order moment method; the failure probability of each pipe segment is obtained by solving the analytical expression for the failure probability of a single pipe segment. The actual topology of the water supply network is abstracted into a graph structure and a graph attention network is constructed. Based on the failure probability of each pipe segment, the chain propagation and evolution of risk along the pipe segments and nodes is simulated through the graph attention network, and the system-level overall risk distribution covering the entire network is obtained by solving the problem. Based on the failure probability of each pipeline segment and the overall risk distribution at the system level, the pipeline risk level is classified and the maintenance priority index of each pipeline segment is calculated; the maintenance priority is sorted according to the maintenance priority index, and standardized pipeline risk assessment results and batch operation and maintenance scheduling schemes are output.

2. The water supply pipe risk prediction method according to claim 1, characterized by, The expression for the three-field coupled control equations is: wherein is the pipe wall thickness; is time; is the molar mass of iron; is the number of electron transfers; is the Faraday constant; is the density of iron; is the exchange current density; is the anodic transfer coefficient; is the cathodic transfer coefficient; is the overpotential; is the gas constant; is the absolute temperature; is the hoop stress; is the axial stress; is the pipe outer diameter; is the pipe internal pressure; is the spatial coordinate; is the fatigue damage accumulation value; , is the material fatigue parameter; is the pressure cycle frequency function.

3. A water supply pipe risk prediction method according to claim 2, characterized in that, The expression for the total loss function of the deep learning model is: wherein, is the total loss function; is the data-driven loss term; is the partial differential equation residual loss term; is the boundary condition loss term; is the initial condition loss term; , , is the weight coefficient; is the number of training samples; is the predicted wall thickness of the th sample; is the true wall thickness of the th sample; is the predicted failure probability of the th sample; is the true failure probability of the th sample; and is the balance weight; is the number of design points; is the corrosion current density predicted by the neural network.

4. The water supply pipe risk prediction method according to claim 3, characterized by, In the process of extracting feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data, electrochemical feature vectors are extracted from the microscale, geometric feature vectors are extracted from the mesoscale, and hydraulic feature vectors are extracted from the macroscale. The expression for the electrochemical characteristic vector is: wherein, is an electrochemical feature vector; is a corrosion potential; is a polarization resistance; is a chloride ion concentration; is a dissolved oxygen concentration; is a water solution pH; The expression for the geometric feature vector is: wherein, is a geometric feature vector; is a pipe segment length; is a pipe inclination angle; is a weld quality coefficient; is a coating condition coefficient; The expression for the hydraulic feature vector is: wherein is a hydraulic feature vector; is a flow rate; is a flow velocity; is a pressure drop; is a pressure fluctuation frequency; is a pressure standard deviation; The expression for the comprehensive feature vector is: In the formula, is a comprehensive feature vector; is the number of attention heads; is an output weight matrix.

5. A water supply pipe risk prediction method according to claim 4, characterized in that, The expression for the pipeline limit state function is: wherein is a pipeline limit state function; is a minimum allowable wall thickness; is an initial wall thickness; is a random variable vector; The expression for the failure probability is: wherein is the failure probability; is the number of Monte Carlo samples; is the indicator function; is the vector of random variables for the th sample.

6. A water supply pipe risk prediction method according to claim 5, wherein, The expression for the overall system-level risk distribution is: wherein, is the system-level overall risk distribution matrix; is the number of network layers; is the risk state matrix of the layer; is the number of propagation steps; is the learnable parameter of the step; is the activation function; is the adjacency matrix; is the degree matrix.

7. A water supply pipe risk prediction method according to claim 6, characterized in that, The risk level of the pipeline is: In the formula, is a failure probability; , , is a preset threshold value, The value range of is , The value range of is , The value range of is ; The formula for calculating the maintenance priority index is as follows: In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .

8. A water supply pipeline risk prediction system, employing the water supply pipeline risk prediction method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire multi-source monitoring data of the water supply pipeline. The multi-source monitoring data includes three types of basic monitoring data: microscale electrochemical corrosion parameters, mesoscale pipeline geometric parameters, and macroscale hydraulic operation parameters. The three-field coupled control equations construction module is used to construct a three-field coupled control equation set of corrosion field, stress field and fatigue field based on the multi-source monitoring data and according to electrochemical corrosion theory, elasticity and fatigue damage mechanics. The deep learning model building module is used to build a physically constrained deep learning model and adds the residuals of the three-field coupled control equations as regularization terms to the total loss function so that the output of the neural network satisfies the physical conservation law. The multi-scale feature vector extraction and fusion module is used to extract feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data. Based on the input dimension requirements of the deep learning model, a cross-scale feature fusion operator is designed. Based on the cross-scale feature fusion operator, the three scale feature vectors are fused through a multi-head attention mechanism to generate a comprehensive feature vector of a unified dimension. The failure probability calculation module is used to construct the pipeline limit state function based on the comprehensive feature vector and the wall thickness attenuation evolution relationship derived through the three-field coupled control equations; based on the pipeline limit state function, it derives the analytical expression for the failure probability of a single pipe segment considering multiple parameter uncertainties through the first second-order moment method; and obtains the failure probability of each pipe segment by solving the analytical expression for the failure probability of a single pipe segment. The system-level overall risk distribution acquisition module is used to abstract the actual topology of the water supply network into a graph structure and construct a graph attention network; based on the failure probability of each pipe segment, the graph attention network is used to simulate the chain propagation and evolution process of risk along the pipe segments and nodes of the network, and solve to obtain the system-level overall risk distribution covering the entire network. The risk assessment and operation and maintenance scheduling scheme acquisition module is used to classify pipeline risk levels and calculate the maintenance priority index of each pipeline segment based on the failure probability of each pipeline segment and the overall risk distribution of the system level; sort the maintenance priorities according to the maintenance priority index, and output standardized pipeline risk assessment results and batch operation and maintenance scheduling schemes.

9. A water supply pipeline risk prediction system according to claim 8, characterized in that, In the three-field coupled control equations construction module, the expression of the three-field coupled control equations is: In the formula, For pipe wall thickness; For time; is the molar mass of iron; The number of electrons transferred; It is Faraday's constant; The density of iron; For exchange current density; The anode transfer coefficient; The cathode transfer coefficient; This is an overpotential; It is the gas constant; Absolute temperature; It is circumferential stress; For axial stress; The outer diameter of the pipe; This refers to the internal pressure of the pipeline. Spatial coordinates; This represents the cumulative value of fatigue damage. , These are material fatigue parameters; This is a function of the pressure cycle frequency.

10. A water supply pipeline risk prediction system according to claim 9, characterized in that, In the deep learning model construction module, the expression for the total loss function of the deep learning model is: In the formula, This is the total loss function; For data-driven loss terms; This is the residual loss term in the partial differential equation; This refers to the boundary condition loss term; This is the initial condition loss term; , , These are the weighting coefficients; This represents the number of training samples; For the first Predicted wall thickness for each sample; For the first The true wall thickness of each sample; For the first The probability of prediction failure for each sample; For the first The true failure probability of a sample; and To balance the weights; Number of configuration points; The corrosion current density is predicted by the neural network. In the multi-scale feature vector extraction and fusion module, during the process of extracting feature vectors of three scales—microscale, mesoscale, and macroscale—from the multi-source monitoring data, electrochemical feature vectors are extracted from the microscale, geometric feature vectors are extracted from the mesoscale, and hydraulic feature vectors are extracted from the macroscale. The expression for the electrochemical characteristic vector is: In the formula, These are electrochemical characteristic vectors; This is the corrosion potential; Polarization resistor; Chloride ion concentration; Dissolved oxygen concentration; pH of the aqueous solution; The expression for the geometric feature vector is: In the formula, These are geometric eigenvectors; This refers to the length of the pipe section. The pipe inclination angle; This is the weld quality coefficient; This is the corrosion protection coating condition coefficient; The expression for the hydraulic feature vector is: In the formula, This represents the hydraulic feature vector; For traffic; For flow rate; For pressure drop; This refers to the frequency of pressure fluctuations. The standard deviation of pressure; The expression for the comprehensive feature vector is: In the formula, This is a comprehensive feature vector; For the number of attention heads; To output the weight matrix; In the failure probability calculation module, the expression for the pipeline limit state function is: In the formula, This is the limit state function of the pipeline; Minimum wall thickness allowed; This is the initial wall thickness; A vector of random variables; The expression for the failure probability is: In the formula, This represents the probability of failure. The number of Monte Carlo samples; For indicator functions; For the first A vector of random variables from the next sampling; In the system-level overall risk distribution acquisition module, the expression for the system-level overall risk distribution is: In the formula, This is the system-level overall risk distribution matrix; The number of network layers; For the first Risk state matrix of the layer; For the number of propagation steps; For the first Learnable parameters for each step; For activation functions; It is an adjacency matrix; It is a degree matrix; In the risk assessment and operation and maintenance scheduling scheme acquisition module, the pipeline risk level is: In the formula, This represents the probability of failure. , , For the preset threshold, The range of values ​​is , The range of values ​​is , The range of values ​​is ; The formula for calculating the maintenance priority index is as follows: In the formula, For the first Maintenance priority index for each pipe section; The severity of the consequences of failure; This represents the risk propagation impact coefficient. , , For the weighting coefficients, satisfying .