AI-based water supply network abnormal spatio-temporal migration early warning system

Abstract

The application discloses an AI-based water supply pipe network abnormal space-time migration early warning system, relates to the technical field of water supply pipe network operation and maintenance early warning, and comprises a multi-source non-intervention data acquisition module, an impedance abnormal symbiotic feature initialization module, a dynamic symbiotic evolution AI guiding module, a double-cycle traceability verification module and a symbiotic early warning and avoidance guiding output module which are sequentially connected to form a closed-loop collaborative mechanism.In the application, the whole-chain technology of multi-source non-intervention data acquisition, impedance abnormal symbiotic feature field construction, CEG-GNN model driven symbiotic evolution guidance, double-cycle traceability verification and early warning avoidance collaborative output is utilized to solve the problems of the lack of global dynamic implicit impedance perception, the lack of essential connection between abnormal migration and impedance, the poor direct measurement adaptability, and the insufficient early warning advancement and depth of existing systems, realize 4-8 hour early warning of water supply pipe network abnormalities, accurately output multi-dimensional information such as abnormal areas and time, and output avoidance schemes, and adapt to large-scale pipe network non-intervention normalized operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of water supply network operation and maintenance early warning technology, specifically to an AI-based early warning system for the spatiotemporal migration of abnormal water supply networks. Background Technology

[0002] Water supply networks are core infrastructure ensuring the normal operation of cities, and their operational stability is directly related to public safety and resource utilization efficiency. With the development of artificial intelligence technology, AI-based early warning systems for the spatiotemporal migration of anomalies in water supply networks have become a key technological direction for improving the operation and maintenance level of networks. The core objective is to use AI models to mine the spatiotemporal characteristics of monitoring data such as flow and pressure, and to achieve early identification of anomalies such as leaks and pressure surges, as well as prediction of migration trends.

[0003] Existing early warning systems of this type mostly focus on the extraction and identification of abnormal signals themselves, such as capturing the spatiotemporal distribution patterns of anomalies through models like recurrent neural networks and convolutional neural networks, and then outputting early warning information. However, existing systems have several problems: First, their ability to sense the core hydraulic parameters of the pipeline network, namely hydraulic impedance, is severely insufficient. They can only obtain local and static impedance data through direct measurement methods such as pipeline inspection robots, and cannot achieve global and dynamic perception of implicit impedance distribution. Second, they have not established a correlation between the spatiotemporal migration trajectory of anomalies and hydraulic impedance, ignoring the objective law that the migration of anomalies is essentially the movement of water in a dynamic impedance field. This results in early warnings only identifying the surface level of anomalies that have occurred, and failing to predict potential anomaly areas from the root cause. Third, they rely on directly measured impedance data to assist in early warning, but direct measurement has drawbacks such as high cost, low efficiency, and interference with pipeline network operation, making it difficult to adapt to the routine operation and maintenance needs of large-scale pipeline networks.

[0004] In summary, existing AI-based early warning systems for spatiotemporal migration anomalies in water supply networks lack effective perception of the implicit distribution of hydraulic impedance and the logic for correlating it with abnormal migration trajectories, resulting in insufficient depth and foresight in early warning. This makes them unable to meet the practical needs of precise operation and maintenance and proactive risk prevention in modern urban pipe networks. Therefore, this paper proposes an AI-based early warning system for spatiotemporal migration anomalies in water supply networks to overcome these problems. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based early warning system for the spatiotemporal migration of abnormal water supply networks, in order to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical issues, this invention provides an AI-based early warning system for spatiotemporal migration anomalies in water supply networks. This system comprises a multi-source non-interventional data acquisition module, an impedance anomaly symbiotic feature initialization module, a dynamic symbiotic evolution AI guidance module, a dual-loop source tracing verification module, and a symbiotic early warning and avoidance guidance output module, all connected in series to form a closed-loop collaborative mechanism. Through the core mechanism of impedance anomaly symbiotic evolution guidance, it simultaneously completes global impedance latent perception, anomaly migration root cause correlation, and non-interventional operation and maintenance adaptation.

[0007] Furthermore, the multi-source non-interventional data acquisition module adopts environmental pipeline network coupled disturbance data acquisition. The acquisition content includes: real-time flow data, real-time pressure data, node water temperature data and structured topology data obtained through existing pipeline network operation and maintenance equipment; soil moisture distribution data and surface vibration data collected through non-contact environmental disturbance sensing units, as well as precipitation data obtained by connecting to urban meteorological stations; and a historical symbiotic sample library constructed by extracting abnormal event data from the pipeline network's historical operation and maintenance database for the past 5 years. The sample includes abnormal full-cycle pipeline network data, environmental coupled disturbance data and abnormality type and handling result information.

[0008] Furthermore, the impedance anomaly co-occurrence feature initialization module uses an impedance anomaly co-occurrence feature field to construct and performs the following steps: The collected data is spatiotemporally aligned using spatial coordinates and timestamps as dual references, and cleaned using an attention mechanism to form a pipeline network environment coupled data matrix; the implicit impedance disturbance factor is calculated based on the coupling principle of fluid mechanics and environmental mechanics. Using the pipeline network topology as the spatial framework, a three-dimensional impedance anomaly symbiotic feature field is constructed, which includes flow-pressure coupling values, implicit impedance disturbance factors, and environmental disturbance parameters.

[0009] Furthermore, the dynamic symbiotic evolution AI guidance module adopts the CEG-GNN model, which is a dual-branch single-guidance architecture, including an impedance field evolution branch, an abnormal migration evolution branch, and a symbiotic guidance unit; the impedance field evolution branch uses a latent impedance perturbation factor. The initial input is used to simulate the dynamic evolution of the impedance field through a 3-layer graph convolutional layer; the abnormal migration evolution branch uses the flow-pressure coupling value as the initial input and simulates the abnormal migration trajectory through a 3-layer graph convolutional layer; the symbiotic guiding unit constructs a guiding function based on the migration law of water flow along the path of minimum impedance, and converts the impedance field evolution output into attention weights to adjust the direction of abnormal migration evolution.

[0010] Furthermore, the co-evolutionary guidance rules of the CEG-GNN model include: an impedance constraint rule, where anomalous migration evolution conforms to the direction of impedance field gradient reduction; and an environmental perturbation enhancement rule, where... When the value is greater than 0.5, the learning rate of impedance field evolution in the corresponding region is increased and the probability of abnormal migration evolution is reduced; historical co-occurrence matching rules are used to match the current and historical evolution states in real time and adjust the evolution strategy according to the similarity.

[0011] Furthermore, the dual-loop source tracing verification module adopts an impedance source tracing and migration source tracing dual-loop verification mechanism: the first loop obtains the implicit impedance perturbation factor sequence by reverse-engineering the impedance field evolution process, and performs dynamic correlation coefficient verification with the output sequence of the initialization module. A correlation coefficient ≥ 0.85 is considered reliable; the second loop calculates the coupling matching degree between the abnormal migration path and the impedance field gradient direction. A matching degree ≥ 88% is considered reliable; if both loops pass, the evolution result is output; otherwise, the closed-loop optimization process is initiated.

[0012] Furthermore, the symbiotic early warning and avoidance guidance output module outputs multi-dimensional early warning information, including the precise coordinates of potential abnormal areas, the time window of abnormal occurrence, the probability of abnormal type, and a three-level risk level based on impedance evolution rate and abnormal migration speed; at the same time, it identifies high impedance barrier areas, generates an abnormal migration avoidance guidance scheme that includes pipeline valve adjustment strategies and temporary pressurization and depressurization points, and displays trajectory comparisons through a visual interface.

[0013] Furthermore, the symbiotic early warning and avoidance guidance output module adopts a multi-terminal collaborative push method to simultaneously push early warning information and avoidance guidance plans to the pipeline operation and maintenance management platform, the mobile application of operation and maintenance personnel, and the emergency command center.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. Achieve non-interventional pipeline network operation and maintenance adaptation: By acquiring multi-source non-interventional data, the high cost, low efficiency and interference with pipeline network operation caused by direct impedance measurement are avoided, and the network is adapted to the routine operation and maintenance needs of large-scale pipeline networks.

[0015] 2. Improve the accuracy of global impedance latent sensing: By constructing the CEG-GNN model with the characteristic field of impedance anomaly coexistence, the global dynamic latent impedance distribution can be captured without direct measurement, breaking the limitations of traditional local and static impedance sensing.

[0016] 3. Enhance the foresight and depth of anomaly early warning: Establish the intrinsic relationship between anomaly migration and impedance field through a symbiotic evolution guidance mechanism, predict potential anomaly areas from the root cause, solve the problem that the existing system can only identify anomalies that have already occurred on the surface, and achieve early warning 4-8 hours in advance.

[0017] 4. Ensure the accuracy of evolution results and early warning information: The dual-cycle tracing and verification mechanism ensures that the evolution results conform to physical laws through mutual verification of impedance tracing and migration tracing. The coordinate error of potential abnormal areas in the early warning information is ≤5 meters and the time window error is ≤30 minutes.

[0018] 5. Enable coordinated early warning and avoidance: Output multi-dimensional early warning information such as three-level risk level and probability of abnormal type, and simultaneously generate avoidance guidance schemes such as valve adjustment, pressurization and depressurization. Combined with visualized trajectory comparison and multi-terminal collaborative push, improve the accuracy of pipeline network operation and maintenance and the efficiency of emergency response. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the AI-based early warning system for spatiotemporal migration of water supply network anomalies according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 The present invention provides a technical solution: See Figure 1 As shown, an embodiment of an AI-based early warning system for the spatiotemporal migration of anomalies in water supply networks is presented: I. System Overall Architecture: The core architecture adopts a full-chain technical solution guided by impedance anomaly symbiotic evolution, including a multi-source non-interventional data acquisition module, an impedance anomaly symbiotic feature initialization module, a dynamic symbiotic evolution AI guidance module, a dual-loop source tracing verification module, and a symbiotic early warning and avoidance guidance output module. These modules are sequentially connected to form a closed-loop collaborative mechanism of data input, symbiotic initialization, evolution guidance, source tracing verification, and early warning output. This completely eliminates the independent impedance measurement module and passive correlation module of existing systems. Through the core mechanism of actively guiding the symbiotic evolution of impedance field and anomaly migration, it simultaneously achieves three goals: global impedance implicit perception, anomaly migration root cause correlation, and non-interventional operation and maintenance adaptation.

[0022] Core technologies and functions of each module: ① Multi-source non-interventional data acquisition module: Employs environmental pipeline network coupled disturbance data acquisition to collect multi-dimensional data naturally generated during pipeline network operation without any active measurement operations, providing basic data support for symbiotic evolution; ② Impedance anomaly symbiotic feature initialization module: Constructs an impedance anomaly symbiotic feature field to transform multi-source data into an initial feature field for impedance anomaly symbiotic evolution, breaking the traditional mode of extracting features individually; ③ Dynamic symbiotic evolution AI guidance module: Constructs a symbiotic evolution guidance graph neural network to actively guide the symbiotic process of implicit impedance field evolution and anomaly migration evolution, and explores the intrinsic relationship between the two in the evolution process; ④ Dual-loop source tracing verification module: Employs a dual-loop verification mechanism of impedance tracing and migration tracing, ensuring the reliability and accuracy of symbiotic evolution results through dual-loop verification of evolution trajectory tracing and impedance distribution tracing; ⑤ Symbiotic early warning and avoidance guidance output module: Employs a collaborative output of early warning and avoidance, outputting anomaly advance warning information based on the verified symbiotic evolution results, and generating anomaly migration avoidance guidance schemes.

[0023] II. Core Modules: 1. Multi-source non-interventional data acquisition module: This module employs environmental pipeline network coupled disturbance data acquisition, breaking through the conventional approach of only collecting parameters of the pipeline network itself. Through a non-contact and non-interventional acquisition method, it achieves comprehensive acquisition of pipeline network and environmental coupled data, as detailed below: 1.1 Basic Pipeline Network Data Acquisition: Utilizing existing conventional maintenance equipment in the pipeline network without adding any new equipment, real-time flow data, real-time pressure data, and node water temperature data are collected from each monitoring node; structured topology data is retrieved through the pipeline network maintenance system. Specifically, the sampling frequency for real-time flow data is set to once per minute, with an acquisition accuracy controlled to ±0.01 cubic meters per second; the sampling frequency for real-time pressure data is set to once per minute, with an acquisition accuracy controlled to ±0.01 megapascals; the structured topology data includes information such as pipe identification, length, pipe diameter, material, connection node coordinates, and laying depth, and the data format is uniformly structured data tables.

[0024] 1.2 Environmental Coupling Disturbance Data Acquisition: An environmental disturbance sensing unit is added. This unit adopts a non-contact design, avoiding contact with the pipeline network itself to prevent interference with its operation. It collects environmental disturbance data around the pipeline network, specifically including soil moisture distribution data, surface vibration data, and precipitation data. Soil moisture distribution data is collected using distributed fiber optic sensing technology, with a sampling interval of 10 meters, and data transmission is achieved via fiber optic communication. Surface vibration data is collected using wireless vibration sensors deployed along the pipeline network's green belt at 50-meter intervals, and the data is transmitted to the system via a wireless local area network. Precipitation data is obtained through a real-time data interface connected to the city's meteorological station, with an update frequency of once per hour. This type of data is a novel data type not collected by existing technologies, used to capture the implicit disturbances of environmental factors on the pipeline network impedance, providing a comprehensive coupled data foundation for symbiotic evolution.

[0025] 1.3 Construction of Historical Symbiotic Sample Data: Anomaly event data from the past five years were extracted from the historical pipeline operation and maintenance database to construct a historical symbiotic sample library. Each sample includes pipeline data and environmental coupled disturbance data covering the entire cycle from anomaly occurrence to migration, along with corresponding anomaly type and handling results. The time range from anomaly occurrence to migration is defined as the period from the first detection of the anomaly to its complete handling, with data time intervals consistent with real-time data acquisition. Anomaly types are categorized into leakage and pressure surges, and handling results include handling measures, handling duration, and post-handling pipeline operating parameters. This sample library abandons the conventional approach of simply extracting anomaly trajectory labels, providing complete sample support for guiding symbiotic evolution.

[0026] It should be noted here that: incorporating environmental pipeline network coupling disturbance data into the water supply network early warning data collection scope, and adopting a non-contact and non-interventional collection method, not only avoids the defects of directly measuring impedance, but also provides data on the driving factors of impedance dynamic changes that are currently lacking in existing technologies for subsequent co-evolution.

[0027] 2. Impedance anomaly co-occurrence characteristic initialization module: This module employs the construction of a co-occurring feature field of impedance anomalies, breaking away from the conventional approach of extracting impedance features and anomaly features separately. It achieves initial coupling between impedance and anomalies through three consecutive steps: data coupling alignment, calculation of implicit impedance perturbation factors, and construction of the co-occurring feature field. Specifically: Step 1: Data Coupling and Alignment: Using spatial coordinates and timestamps as dual benchmarks, the basic pipeline network data and environmental coupled disturbance data are spatiotemporally aligned to form a pipeline network-environment coupled data matrix. Spatial coordinates are based on the latitude and longitude coordinates of each monitoring node in the pipeline network, and timestamps are based on Beijing time, uniformly accurate to the minute. The row dimension of the pipeline network-environment coupled data matrix is ​​the timestamp, and the column dimension is the monitoring node. Matrix elements are the flow rate, pressure, water temperature, soil moisture, vibration amplitude, and precipitation data of that node at the corresponding timestamp. After alignment, an attention-based data cleaning algorithm is used to process the data, removing outliers and filling missing values. The criteria for identifying outliers are that the data deviates from the coupling relationship threshold of its corresponding spatiotemporal unit by more than 15%, and the coupling relationship is obtained by fitting a linear regression model. Missing values ​​are filled by inferring the coupling relationship of three adjacent spatiotemporal units, with the filling error controlled within 3%.

[0028] Step 2: Calculation of Latent Impedance Disturbance Factor: Based on the coupling principle of fluid mechanics and environmental mechanics, a correlation formula between environmental disturbance and impedance change is derived. This formula is used to calculate the latent impedance disturbance factor α, which directly reflects the degree of latent influence of environmental disturbance on local impedance. It can indirectly capture dynamic impedance changes without measuring impedance. The formula is as follows: ; In the formula: Latent impedance disturbance factor, with a value ranging from 0 to 1. The larger the value, the more significant the impact of environmental disturbances on the pipeline impedance in this area; The soil moisture influence weight, ranging from 0.3 to 0.6, is obtained through training with sample data from a historical co-occurrence sample library. Specifically, the training process involves linearly fitting the changes in soil moisture in historical samples with the corresponding changes in impedance; the resulting regression coefficients are then used. The possible values ​​of ; The change in soil moisture is calculated by subtracting the soil moisture value from the previous moment from the current soil moisture value. The soil moisture value is obtained through the environmental disturbance sensing unit. The vibration impact weight, ranging from 0.2 to 0.4, is also obtained through training with sample data from the historical co-occurrence sample library. The training process involves linearly fitting the changes in surface vibration amplitude and the corresponding changes in impedance in the historical samples, and the resulting regression coefficients are... The possible values ​​of ; The change in surface vibration amplitude is calculated by subtracting the surface vibration amplitude from the previous moment from the current moment. The surface vibration amplitude is collected by the wireless vibration sensor in the environmental disturbance sensing unit. Pipe material correction factor, determined based on pipe material, for metal pipes. The value is set to 1.2, the γ value of the plastic pipe is set to 0.8, and the pipe material information is obtained from the structured topology data.

[0029] Step 3: Construction of the Symbiotic Feature Field: Using the pipeline network topology as the spatial framework, a three-dimensional impedance anomaly symbiotic feature field is constructed. The spatial dimension of this feature field consists of the coordinates of each node in the pipeline network, consistent with the actual pipeline network topology; the temporal dimension consists of continuous timestamps, with time intervals consistent with the data acquisition frequency; the feature dimensions include flow-pressure coupling values ​​and implicit impedance disturbance factors. The environmental disturbance parameters include soil moisture and vibration amplitude, where the flow-pressure coupling value is the product of real-time flow data and real-time pressure data. With the factors determined, each characteristic value reflects the coupling relationship between impedance, anomaly, and environment, completely breaking the limitation of isolated characteristics in existing technologies.

[0030] It should be noted here that the concept and construction method of impedance anomaly co-occurrence characteristic field are proposed. The dynamic changes of impedance are indirectly captured by the correlation formula between environmental disturbance and impedance change. The initial coupling of the impedance implicit characteristics can be achieved without direct measurement of impedance.

[0031] 3. Dynamic Symbiotic Evolution AI Guidance Module: This module is built using the CEG-GNN model, breaking through the existing model architectures of recurrent neural networks, convolutional neural networks, and conventional graph neural networks. Through a novel dual-branch, single-guided architecture design, it actively guides the co-existing process of latent impedance field evolution and anomalous migration evolution, simultaneously realizing global impedance perception and anomalous migration correlation during the evolution, as detailed below: Step 1: Model Architecture Design: The CEG-GNN model adopts a dual-branch, single-guided architecture, comprising two parallel and mutually guiding evolutionary branches and a symbiotic guiding unit. The two evolutionary branches are the impedance field evolution branch and the anomalous migration evolution branch, respectively. The symbiotic guiding unit is the core control unit of the model, constructing a guiding function based on the physical law of water flow migrating along the path of minimum impedance, and transmitting the evolutionary information of the two branches in real time to achieve mutual guidance. Specifically, the impedance field evolution branch utilizes the implicit impedance perturbation factor in the impedance anomalous symbiotic characteristic field. Using the impedance field evolution as the initial input, the dynamic evolution process of the implicit impedance field is simulated through three graph convolutional layers, with each graph convolutional layer having 64 output channels. The abnormal migration evolution branch uses the flow-pressure coupling value in the impedance anomaly symbiotic feature field as the initial input, and simulates the dynamic trajectory of the abnormal migration through three graph convolutional layers, with each graph convolutional layer having 64 output channels. The guidance function of the symbiotic guidance unit adopts an attention weight allocation mechanism, which converts the impedance value output by the impedance field evolution branch into attention weights, and adjusts the output features of the abnormal migration evolution branch accordingly. This enables the impedance field evolution to guide the abnormal migration evolution, i.e., the attention weights corresponding to the high impedance regions output by the impedance field evolution branch are reduced, thus inhibiting the abnormal migration evolution branch from evolving into those regions.

[0032] Step 2, Symbiotic Evolution Guidance Rules: Three guiding rules are established to regulate the evolutionary direction of the two evolutionary branches and ensure that the evolutionary process conforms to physical laws. Rule 1 is the impedance constraint rule: the evolutionary direction of anomalous migration must follow the gradient direction of the impedance field evolution; that is, the direction of impedance reduction is the preferred evolutionary direction of anomalous migration. The impedance gradient direction is calculated from the impedance value output by the impedance field evolution branch, specifically the direction of the impedance difference between adjacent nodes. Rule 2 is the environmental disturbance enhancement rule: when the implicit impedance disturbance factor... When the learning rate is greater than 0.5, the impedance field evolution rate in the corresponding region is strengthened by increasing the learning rate of the impedance field evolution branch, while the attention weight in that region is reduced to decrease the probability of abnormal migration to that region. The learning rate is increased from 0.001 to 0.002. Rule 3 is the historical co-occurrence matching rule, which matches the current evolution state with the evolution state in the historical co-occurrence sample library in real time and calculates the feature similarity between the two. When the similarity is greater than 80%, the historical evolution trend is followed. When the similarity is less than 50%, the adaptive adjustment mechanism is activated to adjust the attention weight allocation ratio of the co-occurrence guidance unit.

[0033] Step 3: Model Training and Evolution Simulation: Input the impedance anomaly co-occurrence feature field data into the CEG-GNN model. Use the impedance evolution anomaly migration co-occurrence trajectories in the historical co-occurrence sample library as supervision labels. Calculate the error between the model prediction results and the supervision labels using the co-occurrence loss function. Update the model parameters using an adaptive momentum optimization algorithm until the loss function converges. The co-occurrence loss function is used to comprehensively measure the impedance field evolution error and the anomaly migration evolution error, ensuring the synchronization of the evolution of the two branches. The function expression is as follows: ; In the formula: The coexistence loss function value measures the overall error between the model-predicted coexistence evolution trajectory and the historical true coexistence trajectory. The smaller the value, the higher the model's prediction accuracy; Weighting coefficients, ranging from 0.4 to 0.6, are used to balance the weights of impedance field evolution error and anomalous migration evolution error. They are determined through cross-validation, specifically by selecting different weights... The model is trained using the given values, and then tested on the validation set. The model with the smallest validation set error is selected. The value is taken as the final value; Impedance field evolution error is calculated using mean square error, which is the average of the sum of squares of the differences between the impedance values ​​at corresponding nodes and timestamps of the impedance field evolution trajectory predicted by the model and the historical true impedance field evolution trajectory. The abnormal migration and evolution error is calculated using mean squared error, which is the average of the sum of squares of the differences between the position coordinates of the corresponding nodes and timestamps of the abnormal migration and evolution trajectory predicted by the model and the historical actual abnormal migration and evolution trajectory.

[0034] During model training, the momentum parameter of the adaptive momentum optimization algorithm was set to 0.9, the weight decay coefficient was set to 0.0001, and the convergence threshold of the loss function was set to 0.0008. After training, the model outputs the dynamic evolution trajectory of the impedance field and the anomaly migration evolution trajectory over the next 4 to 8 hours based on the real-time input impedance anomaly co-occurrence feature field, realizing global dynamic impedance sensing and anomaly migration correlation.

[0035] It should be noted here that the CEG-GNN model achieves active co-evolutionary guidance of impedance field and anomalous migration, rather than the passive correlation of existing technologies, thus breaking the separation between impedance, anomalous, and environment from a mechanistic perspective.

[0036] 4. Dual-loop traceability verification module: This module employs a dual-loop verification mechanism involving impedance sourcing and migration sourcing, breaking away from the conventional approach of single numerical comparison verification. Through mutual verification and closed-loop optimization of the two loops—impedance sourcing verification and migration sourcing verification—the reliability of the co-evolution results is ensured, as detailed below: Step 1: First Cycle Impedance Source Tracing Verification: Based on the impedance field evolution trajectory output by the CEG-GNN model, trace the implicit impedance perturbation factor of each spatiotemporal unit in reverse. The evolutionary process of the impedance field is achieved by reversing the calculation process of the impedance field evolution branch, that is, by calculating the impedance of each spatiotemporal unit in reverse based on the output impedance value. Factor sequence. This is obtained by tracing the source. The output of the factor sequence and impedance anomaly co-occurrence feature initialization module Consistency verification of the factor sequences was performed using the dynamic correlation coefficient as the verification metric. The dynamic correlation coefficient was calculated using a sliding window method with a window size set to 30 minutes, and two values ​​were calculated. The correlation coefficient of the factor sequence within each window. When the dynamic correlation coefficient is greater than or equal to 0.85, the impedance field evolution trajectory is considered reliable; when the dynamic correlation coefficient is less than 0.85, the source tracing results are fed back to the impedance anomaly co-occurrence feature initialization module for correction. The influence weight of soil moisture in value calculation And the influence of vibration on weight The value of is used to regenerate the impedance anomaly co-occurrence feature field and input it into the CEG-GNN model for re-evolution.

[0037] Step 2, Second Cycle Migration Tracing Verification: Based on the anomalous migration evolution trajectory output by the CEG-GNN model, trace its coupling matching degree with the impedance field evolution trajectory in reverse. The coupling matching degree is calculated as the ratio of the overlap length between the anomalous migration path and the impedance field gradient direction to the total length of the anomalous migration path. The total length of the anomalous migration path is the sum of the straight-line distances between adjacent nodes in the anomalous migration evolution trajectory; the overlap length is the sum of the lengths of line segments in the anomalous migration path aligned with the impedance field gradient direction. When the coupling matching degree is greater than or equal to 88%, it indicates that the anomalous migration mainly evolves along the impedance reduction direction, conforming to the physical laws of water flow, and the anomalous migration evolution trajectory is deemed credible. When the coupling matching degree is less than 88%, the verification results are fed back to the CEG-GNN model, the attention weight allocation ratio of the symbiotic guiding unit is adjusted, and the evolution simulation is repeated.

[0038] Step 3 Closed-loop optimization: The final evolution result is output only when both impedance sourcing verification and migration sourcing verification pass. If either loop fails, the closed-loop optimization process of initialization correction, re-evolution, and re-verification is started, repeating steps 1 to 2 until both loop verifications pass.

[0039] It should be noted here that the dual-loop tracing verification mechanism verifies the results by tracing impedance and migration in reverse rather than by comparing them in the forward direction. This ensures the physical rationality and accuracy of the evolution results and further strengthens the intrinsic relationship between impedance and anomalous migration.

[0040] 5. Symbiotic Early Warning and Avoidance Guidance Output Module: This module adopts a collaborative output of early warning and avoidance, breaking through the conventional function of simply outputting early warning information. It achieves collaborative output of early warning and avoidance guidance through proactive early warning information output, anomaly migration avoidance guidance scheme output, and multi-terminal collaborative push, as detailed below: Advanced early warning information output: Based on the evolution results verified by the double-loop system, multi-dimensional early warning information is output. Specifically, the precise coordinates of potential anomaly areas are determined by the coordinates of the starting node of the anomaly migration evolution trajectory, with a coordinate error controlled within 5 meters; the precise time window for anomaly occurrence is determined by the time the anomaly first arrives at the potential anomaly area in the anomaly migration evolution trajectory, with a time error controlled within 30 minutes; the anomaly type probability is obtained by matching the current evolutionary features with the similarity features of two types of anomaly samples in the historical co-occurrence sample library, and the anomaly type with the highest similarity is the predicted type, while the probability value of this type is also output; the risk level is divided based on impedance evolution rate and anomaly migration speed. Specifically, Level 1 (extremely high risk) corresponds to an impedance evolution rate greater than 0.05 per hour and an anomaly migration speed greater than 100 meters per hour; Level 2 (high risk) corresponds to an impedance evolution rate between 0.02 and 0.05 per hour and an anomaly migration speed between 50 and 100 meters per hour; and Level 3 (general risk) corresponds to an impedance evolution rate less than 0.02 per hour and an anomaly migration speed less than 50 meters per hour.

[0041] Output of the anomaly migration avoidance guidance scheme: Based on the impedance field evolution trajectory, high-impedance barrier regions on the anomaly migration path are identified. The criteria for identifying high-impedance barrier regions are areas with an impedance value greater than 0.8 and an impedance evolution rate greater than 0.03 per hour. An avoidance guidance scheme is generated for these regions, specifically including suggested pipeline valve adjustment strategies and temporary pressurization / depressurization points. The pipeline valve adjustment strategy involves opening backup valves around the high-impedance barrier region and closing some valves along the anomaly migration path to strengthen the blocking effect of the high-impedance barrier; the valve opening is set to 80%. Temporary pressurization / depressurization points are selected in the upstream region of the anomaly migration path, with pressurization pressure set at 0.3 MPa and depressurization pressure set at 0.1 MPa, guiding the anomaly migration towards a non-core area through pressurization and depressurization. Simultaneously, a visual interface compares the anomaly migration trajectories before and after avoidance, using a pipeline network electronic map as the base map, with red solid lines and green dashed lines marking the anomaly migration trajectories before and after avoidance, respectively.

[0042] Multi-terminal collaborative push: Early warning information and avoidance guidance plans are simultaneously pushed to the pipeline network operation and maintenance management platform, the mobile application for operation and maintenance personnel, and the emergency command center. The pipeline network operation and maintenance management platform uses wired network transmission with a data transmission rate greater than 10 megabits per second; the mobile application for operation and maintenance personnel uses 5G network transmission with a push latency controlled within 5 seconds; and the emergency command center receives data through a dedicated data interface. The pushed content includes early warning information data tables, avoidance guidance plan documents, and visual trajectory comparison charts. It also supports one-click navigation to potentially abnormal areas and avoidance operation points, achieving full-process collaboration in early warning, response, and avoidance.

[0043] It should be noted here that the goal is to achieve coordinated output of early warning and avoidance guidance, breaking through the limitation of existing early warning systems that only inform of risks without providing solutions.

[0044] III. System Workflow: This system achieves early warning and avoidance guidance for abnormal spatiotemporal migration throughout the entire process through the following five consecutive steps, ensuring that all modules work together and solve all problems synchronously. The workflow is as follows: Step 1: Multi-source non-interventional data acquisition: Start the multi-source non-interventional data acquisition module to collect basic pipeline data and environmental coupled disturbance data, and extract abnormal event data from the historical operation and maintenance database to build a historical symbiotic sample library; the entire acquisition process does not involve any active measurement operations, avoiding the defects of direct measurement.

[0045] Step 2, Impedance Anomaly Co-occurrence Feature Initialization: The impedance anomaly co-occurrence feature initialization module is activated. The data acquired in Step 1 is coupled and aligned. An attention-based data cleaning algorithm is used to remove outlier data and fill in missing values. The implicit impedance perturbation factor is calculated based on the coupling principle of fluid mechanics and environmental mechanics. A three-dimensional impedance anomaly symbiotic feature field is constructed using the pipeline network topology as the spatial skeleton to achieve the initial coupling of impedance latent features.

[0046] Step 3, AI-guided dynamic symbiotic evolution: Start the AI-guided dynamic symbiotic evolution module, input the impedance anomaly symbiotic feature field data constructed in Step 2 into the CEG-GNN model, train the model based on the symbiotic trajectory in the historical symbiotic sample library, and optimize the model parameters through the symbiotic loss function; after training, the model outputs the dynamic evolution trajectory of the impedance field and the anomaly migration evolution trajectory for the next 4 to 8 hours, and simultaneously realizes global impedance perception and anomaly migration correlation.

[0047] Step 4, Dual-loop source tracing verification: Start the dual-loop source tracing verification module and perform impedance source tracing verification and migration source tracing verification based on the evolution trajectory output in Step 3; if both loop verifications pass, proceed to Step 5; if any loop fails, the verification result is fed back to the corresponding module for parameter correction, and Steps 2 to 4 are re-executed until both loop verifications pass.

[0048] Step 5, Symbiotic Early Warning and Avoidance Guidance Output: Activate the symbiotic early warning and avoidance guidance output module. Based on the evolution results verified in Step 4, output multi-dimensional advanced early warning information and abnormal migration avoidance guidance schemes. Simultaneously push relevant information to the pipeline operation and maintenance management platform, the mobile application of operation and maintenance personnel, and the emergency command center to complete the early warning and response guidance.

[0049] Summarize: This embodiment utilizes a core mechanism guided by the symbiotic evolution of impedance anomalies. It designs a complete technical solution encompassing multi-source non-interventional data acquisition, construction of impedance anomaly symbiotic feature fields, a CEG-GNN model, dual-loop source tracing verification, and collaborative output for early warning and avoidance. This addresses the three core issues of existing technologies simultaneously and at their root. Furthermore, it produces effects such as anomaly migration avoidance guidance, pipeline sub-health source tracing, and multi-scenario adaptive early warning.

Claims

1. An AI-based early warning system for the spatiotemporal migration of anomalies in water supply networks, characterized in that: It includes a multi-source non-interventional data acquisition module that forms a closed-loop collaborative mechanism in sequence, an impedance anomaly symbiotic feature initialization module, a dynamic symbiotic evolution AI guidance module, a dual-loop source tracing verification module, and a symbiotic early warning and avoidance guidance output module. Through the core mechanism of impedance anomaly symbiotic evolution guidance, it simultaneously completes global impedance implicit perception, anomaly migration root cause association, and non-interventional operation and maintenance adaptation.

2. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The multi-source non-interventional data acquisition module adopts environmental pipeline network coupled disturbance data acquisition. The acquisition content includes: real-time flow data, real-time pressure data, node water temperature data and structured topology data obtained through existing pipeline network operation and maintenance equipment; soil moisture distribution data and surface vibration data collected through non-contact environmental disturbance sensing units, as well as precipitation data obtained by connecting to urban meteorological stations; and a historical symbiotic sample library constructed by extracting abnormal event data from the pipeline network's historical operation and maintenance database for the past 5 years. The sample includes abnormal full-cycle pipeline network data, environmental coupled disturbance data and abnormality type and handling result information.

3. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The impedance anomaly co-occurrence feature initialization module constructs an impedance anomaly co-occurrence feature field and performs the following steps: The collected data is spatiotemporally aligned using spatial coordinates and timestamps as dual references, and cleaned using an attention mechanism to form a pipeline network environment coupled data matrix; the implicit impedance disturbance factor is calculated based on the coupling principle of fluid mechanics and environmental mechanics. Using the pipeline network topology as the spatial framework, a three-dimensional impedance anomaly symbiotic feature field is constructed, which includes flow-pressure coupling values, implicit impedance disturbance factors, and environmental disturbance parameters.

4. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The dynamic symbiotic evolution AI guidance module adopts the CEG-GNN model, which is a dual-branch single-guidance architecture, including an impedance field evolution branch, an abnormal migration evolution branch, and a symbiotic guidance unit; the impedance field evolution branch uses a hidden impedance perturbation factor. Using the initial input, the dynamic evolution of the impedance field is simulated through three layers of graph convolution. The abnormal migration evolution branch takes the flow-pressure coupling value as the initial input and simulates the abnormal migration trajectory through a 3-layer graph convolutional layer; the symbiotic guidance unit constructs a guidance function based on the migration law of water flow along the path of minimum impedance, and converts the impedance field evolution output into attention weights to adjust the direction of abnormal migration evolution.

5. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 4, characterized in that: The co-evolutionary guidance rules of the CEG-GNN model include: impedance constraint rules, where anomalous migration evolution conforms to the direction of impedance field gradient reduction; and environmental perturbation enhancement rules, where... When the value is greater than 0.5, the learning rate of impedance field evolution in the corresponding region is increased and the probability of abnormal migration evolution is reduced; historical co-occurrence matching rules are used to match the current and historical evolution states in real time and adjust the evolution strategy according to the similarity.

6. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The dual-loop source tracing verification module adopts an impedance source tracing and migration source tracing dual-loop verification mechanism: the first loop obtains the implicit impedance perturbation factor sequence by reversely tracing the impedance field evolution process, and performs dynamic correlation coefficient verification with the output sequence of the initialization module. If the correlation coefficient is ≥0.85, it is considered reliable. The second loop calculates the coupling matching degree between the abnormal migration path and the impedance field gradient direction. A matching degree of ≥88% is considered reliable. If both loops pass, the evolution result is output; otherwise, the closed-loop optimization process is started.

7. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The symbiotic early warning and avoidance guidance output module outputs multi-dimensional early warning information, including the precise coordinates of potential abnormal areas, the time window of abnormal occurrence, the probability of abnormal type, and a three-level risk level based on impedance evolution rate and abnormal migration speed. At the same time, it identifies high impedance barrier areas, generates an abnormal migration avoidance guidance scheme that includes pipeline valve adjustment strategies and temporary pressurization and depressurization points, and displays trajectory comparisons through a visual interface.

8. The AI-based early warning system for spatiotemporal migration of water supply network anomalies as described in claim 1, characterized in that: The symbiotic early warning and avoidance guidance output module adopts a multi-terminal collaborative push method to simultaneously push early warning information and avoidance guidance plans to the pipeline operation and maintenance management platform, the mobile application of operation and maintenance personnel, and the emergency command center.

Images