Water supply network water hammer control method and system based on multifunctional module fusion

Abstract

The invention discloses a water supply pipe network water hammer control method and system based on multifunctional module fusion, and relates to the technical field of data identification. A time sequence diagram neural network prediction and traceability module which realizes water hammer risk prediction and propagation path traceability based on a causal constraint graph neural network model; the reinforcement learning intervention decision module is used for generating an active intervention strategy through a reinforcement learning agent, forming closed-loop control, abstracting a water supply pipe network into a graph structure, and modeling in combination with a time sequence, so that a propagation path of pressure waves can be comprehensively reflected, a blind area of traditional local modeling is overcome, and the comprehensiveness and accuracy of water hammer event detection are improved; and a causal analysis result is input as an adjacent matrix, so that the interference of irrelevant edges on prediction in topology is effectively eliminated, and the learning efficiency and causal traceability of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data recognition, and particularly to a water hammer control method and system for water supply pipe networks based on the integration of multifunctional modules. Background Art

[0002] Water hammer is a pressure fluctuation phenomenon caused by sudden changes in fluid inertia, which widely exists in water supply pipe network systems. Its suddenness and instantaneous high-pressure characteristics pose a serious threat to the safety of pipeline systems. If not disposed of in time, it may lead to pipeline rupture, damage to pumping station equipment, and even large-scale water supply interruption accidents.

[0003] The commonly used water hammer early warning methods in the prior art mainly include the following categories: Early warning method based on pressure threshold: The system sets a fixed or dynamic threshold, such as triggering an alarm when exceeding 1.5 times the working pressure. However, in complex pipe networks, the false alarm rate can reach more than 30%, such as frequent false alarms due to turbulent interference during pump start and stop.

[0004] Detection method based on local time series modeling: Models such as long short-term memory network (LSTM) and convolutional neural network (CNN) are used to model the single-point pressure time series data. Although it can identify abnormal fluctuations, the average prediction lag is 5 - 10 seconds, and due to the lack of consideration of the pipe network topology structure, the false negative rate exceeds 20% in large-diameter pipe networks with a diameter above DN800.

[0005] Water hammer analysis method based on simulation and rule inference: The software such as EPANET is used to simulate the pipe network response, and engineering rules are combined to judge risks. However, in real-time scenarios, it takes 2 - 5 minutes from data collection to generating intervention instructions, which cannot meet the emergency response requirements of the "golden 10 seconds" for water hammer events.

[0006] In summary, the existing methods generally have the following problems: (1) Lack of modeling of the propagation path and spatial causal chain of pressure waves in the pipe network, resulting in the inability to achieve effective traceability; (2) Unable to provide targeted intervention strategies, and most can only give passive alarms and are difficult to actively prevent; (3) Ignoring the influence of the pipe network topology structure on the water hammer propagation mode, the model interpretability and generalization ability are poor.

[0007] Therefore, a water hammer control method and system for water supply pipe networks based on the integration of multifunctional modules are proposed to solve the above problems. Summary of the Invention

[0008] In view of this, the technical problem to be solved by the present invention is to propose a water hammer control method and system for water supply pipe networks based on the integration of multifunctional modules to solve the problems mentioned in the background art.

[0009] To achieve the above object, the present invention provides the following technical solutions: A water hammer control method for a water supply network based on the integration of multifunctional modules includes: S1, digital modeling of the pipe network: Based on the geographic information system (GIS) and design drawings, construct a pipe network topology map, deploy sensors and record pipe attributes; S2, real-time data collection and preprocessing: The sensors collect pipe pressure data at 0.5 - 2 Hz, fill in missing values through cubic spline interpolation, and eliminate outliers by combining the three-standard-deviation principle (3σ) with EPANET software simulation; S3, generation of causal diagrams and model input: Perform causal tests (Granger) on the preprocessed data (the lag order k = 2 - 5, and the time window is set to T = 30 - 90 seconds), generate a sparse causal adjacency matrix, and use it as the input of the graph neural network (TGNN); S4, water hammer risk prediction and traceability: The graph neural network (TGNN) model outputs the probability of water hammer occurrence and the causal propagation chain through the architecture of "graph convolutional neural network (GCN) + gated recurrent unit (GRU) + attention mechanism", and locates the trigger point and propagation path; S5, reinforcement learning and intelligent intervention: Build a pipe network hydraulic simulation environment through the EPANET platform. In the digital twin environment of the EPANET platform, the deep Q network (Deep Q Network) agent generates intervention actions according to the pressure distribution and predicted location, takes the minimum pressure fluctuation as the reward target, and executes the control strategy in real time through the data acquisition and monitoring system (SCADA) to form a closed-loop control.

[0010] Preferably, the sensor described in S1 is a pressure sensor, and is deployed upstream and downstream of (bifurcation points), valves. The deployment spacing of the pressure sensors ≤ 500 meters, and the recorded pipe attributes include material, length, and pipe diameter.

[0011] Preferably, for filling in missing values through cubic spline interpolation in S2, manual verification is performed when the correlation coefficient between adjacent nodes < 0.7, and outliers with a deviation > 20% are determined as wild values.

[0012] Preferably, the lag order of the causal test (Granger) described in S3 is set to k = 2 - 5, and the time window is set to T = 30 - 90 seconds.

[0013] Preferably, the S5 specifically includes: S5.1, building a digital twin environment: Build a pipe network hydraulic simulation environment on the EPANET platform, simulate the real-time pressure distribution and intervention effect, and provide a training scenario for reinforcement learning; S5.2, reinforcement learning strategy training: The Deep Q Network agent aims to minimize pressure fluctuations, learns valve opening and pump speed adjustment strategies, and combines safety constraints to ensure that operations meet engineering boundary conditions; S5.3, intervention strategy execution and feedback: Control instructions are sent to edge devices, including valves and pump stations, with a delay range of 10~500ms through the data acquisition and monitoring system (SCADA), to adjust the status of the pipeline network in real time and form a closed-loop control.

[0014] A water hammer control system for a water supply network based on the fusion of multifunctional modules, comprising: Pressure propagation map construction module: collects raw pressure data from sensors and constructs the raw data into a sparse relationship graph, collects data based on the pressure bus, calculates pipeline interconnection information, and sends the pipeline interconnection results to the time series graph neural network prediction and tracing module for risk prediction and propagation path tracing; Time-series graph neural network prediction and tracing module: Build a graph neural network model based on causal constraints, calculate the results in the pressure propagation map construction module, realize water hammer risk prediction and propagation path tracing, and pass the prediction results to the reinforcement learning intervention decision module.

[0015] Reinforcement learning intervention decision module: Analyze the output results of the timing graph neural network prediction and traceability module, generate active intervention strategies through reinforcement learning agents, and send them to pump stations and valves through the control bus. After the strategy is executed, re-read the latest pressure data, and make learning adjustments based on the pressure data to form a self-closed "prediction → control → re-learning" cycle.

[0016] Preferably, the pressure propagation map construction module includes: Pipeline network digital modeling module: Based on the geographic information system (GIS) and design drawings, build the pipeline network topology map, deploy sensors and record pipeline attributes; Real-time data acquisition and preprocessing module: The sensor collects pipeline pressure data at 0.5-2Hz, fills missing values through cubic spline interpolation, and removes outliers by combining the triple standard deviation principle (3σ) with EPANET software simulation; Causal graph generation and model input module: performs causal testing (Granger) on the preprocessed data (lag order k=2~5, time window set to T=30~90 seconds) to generate a sparse causal adjacency matrix as the input of the graph neural network (TGNN).

[0017] Preferably, the temporal graph neural network prediction and tracing module adopts a causal-constrained graph neural network (TGNN) model, including: Input layer: Multidimensional time series, including pressure, first derivative, fluctuation intensity plus causal adjacency matrix; Feature extraction layer: Composed of two layers of graph convolutional networks (GCNs), used to capture the spatial features of the pipe network topology; Time series modeling layer: Includes one layer of gated recurrent unit (GRU), used to extract the time dependence of the pressure sequence; Attention layer: Introduces a multi-head attention mechanism (4 heads) inside the model to enhance the perception ability of key nodes and key paths, realizes the learning of the pressure propagation path, and enhances the weight allocation of key nodes and paths; The attention layer is configured to weight the features of key nodes and key paths inside the model to enhance the weight allocation of key nodes and key paths.

[0018] Output layer: The probability of water hammer occurrence (0 - 1) plus the causal propagation chain vector (traceability path) not only includes whether a water hammer event will occur, but also can trace back the corresponding causal propagation chain, that is, from the trigger point to the fluctuation propagation path, assisting maintenance personnel in accurate traceability positioning.

[0019] Preferably, the reinforcement learning intervention decision-making module includes: Digital twin environment construction module: Build a pipe network hydraulic simulation environment on the EPANET platform, simulate the real-time pressure distribution and intervention effects, and provide training scenarios for reinforcement learning; Reinforcement learning strategy training module: The DQN agent of the deep Q network aims to minimize pressure fluctuations, learns valve opening and pump speed adjustment strategies, and combines safety constraints to ensure that operations meet engineering boundary conditions; Its input end is connected to the output end of the time series graph neural network prediction and traceability module, and is used to generate intervention strategies according to the water hammer risk probability and propagation path information; Intervention strategy execution and feedback module: Includes an execution module and a feedback acquisition sub-module; The input end of the execution module is connected to the output end of the reinforcement learning strategy training module, and is used to convert the intervention strategy into control instructions for valves or pump stations and send them down through the control bus. The feedback acquisition sub-module has two output channels: (1) Send the real-time pressure sensor readings after execution to the time series graph neural network prediction and traceability module in the form of the original multidimensional time series, update its sliding time window, and use it as the input tensor for the next prediction cycle; (2) Calculate the reward signal based on the pressure data and output it to the reinforcement learning strategy training module for adjusting the intervention strategy; Through the cycle of "prediction - strategy generation - instruction issuance - data and reward feedback - re-prediction", a closed-loop adaptive water hammer control for the water supply pipe network is formed.

[0020] Compared with the prior art, the present invention provides a water hammer control method and system based on the integration of multifunctional modules, having the following beneficial effects: (1) Combining graph structure and time series modeling to improve prediction accuracy and positioning ability; Abstracting the water supply network as a graph structure and combining time series modeling can comprehensively reflect the propagation path of pressure waves, overcome the blind spots of traditional local modeling, and improve the comprehensiveness and accuracy of water hammer event detection.

[0021] (2) Introducing a causal graph to guide the modeling of the time series graph neural network (TGNN) to improve the interpretability and stability of the model; Taking the causal analysis results as the input of the adjacency matrix can effectively eliminate the interference of irrelevant edges in the topology on the prediction, and improve the learning efficiency and causal traceability ability of the model.

[0022] (3) Implementing a closed-loop control of prediction - traceability - intervention; Through the reinforcement learning module, the system can not only achieve passive detection, but also actively adjust the operation state of the water supply network, thereby reducing water hammer damage and forming an intelligent closed-loop control system.

[0023] (4) Having engineering deployability and generality; The system structure is clear and the modules are distinct. It can be deployed on the existing water supply monitoring platform, adapt to water supply networks of various scales and complexities, and has a broad promotion prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the steps of a water hammer control method for a water supply network based on the integration of multifunctional modules according to the present invention; Figure 2 It is a schematic diagram of the overall process of a water hammer early warning and intelligent intervention system according to the present invention; Figure 3 It is a schematic diagram of the causal graph construction process according to the present invention; Figure 4 It is a schematic diagram of the structure of the time series graph neural network (TGNN) according to the present invention; Figure 5 It is a schematic diagram of the reinforcement learning intervention training process according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] The present invention is further described in detail below based on the accompanying drawings and embodiments.

[0027] Example 1, please refer to Figures 1 to 5 As shown: In order to solve the problems mentioned in the technical solution, the embodiment of the present application provides a water hammer control system for a water supply network based on the fusion of multifunctional modules, including: 1. Pressure propagation map construction module, responsible for pipe network topology modeling and causal graph generation; Among them, the pressure propagation map construction module includes: (1) Pipeline network digital modeling module: Based on the geographic information system (GIS) and design drawings, the pipeline network topology map is constructed, sensors are deployed and pipeline attributes are recorded; (2) Real-time data acquisition and preprocessing module: The sensor collects pipeline pressure data at 0.5-2 Hz, fills missing values through cubic spline interpolation, and removes outliers by combining the triple standard deviation principle (3σ) with EPANET software simulation; (3) Causal graph generation and model input module: Perform causal test (Granger) on the preprocessed data (lag order k = 2~5, time window set to T = 30~90 seconds) to generate a sparse causal adjacency matrix as the input of the graph neural network (TGNN).

[0028] 2. Time series graph neural network prediction and traceability module, based on the causal constraint graph neural network (TGNN) model to achieve water hammer risk prediction and propagation path traceability; Among them, the time series graph neural network prediction and traceability module adopts the causal constraint graph neural network (TGNN) model, including: (1) Input layer: multidimensional time series, including pressure, first-order derivative, fluctuation intensity and causal adjacency matrix; (2) Feature extraction layer: It consists of two layers of graph convolutional networks (GCNs) to capture the spatial features of the pipeline network topology; (3) Temporal modeling layer: including a layer of gated recurrent unit (GRU) to extract the temporal dependency of the pressure sequence; (4) Attention layer: A multi-head attention mechanism (4 heads) is introduced into the model to enhance the perception of key nodes and key paths, realize the learning of pressure propagation paths, and enhance the weight distribution of key nodes or paths; (5) Output layer: The probability of water hammer occurrence (0-1) plus the causal propagation chain vector (tracing path) not only includes whether a water hammer event will occur, but also can trace back the corresponding causal propagation chain, that is, from the trigger point to the wave propagation path, to assist operation and maintenance personnel in accurate tracing and positioning.

[0029] 3. Reinforcement Learning Intervention Decision-making Module: Analyze the output results from the time-series graph neural network prediction and tracing module, generate proactive intervention strategies through a reinforcement learning agent, and send them to the pump station and valves via the control bus. After the strategy execution is completed, read the latest pressure data again and perform learning and adjustment based on the pressure data to form a self-closed-loop "prediction → control → re-learning" cycle.

[0030] Among them, the reinforcement learning intervention decision-making module includes: (1) Digital Twin Environment Construction Module: Build a pipe network hydraulic simulation environment on the EPANET platform to simulate real-time pressure distribution and intervention effects, providing a training scenario for reinforcement learning; (2) Reinforcement Learning Strategy Training Module: The DQN agent of the deep Q-network aims to minimize pressure fluctuations, learn valve opening and pump speed adjustment strategies, and ensure that operations comply with engineering boundary conditions in combination with safety constraints; its input end is connected to the output end of the time-series graph neural network prediction and tracing module, and is used to generate intervention strategies based on water hammer risk probability and propagation path information; (3) Intervention Strategy Execution and Feedback Module: Includes an execution module and a feedback acquisition sub-module; the input end of the execution module is connected to the output end of the reinforcement learning strategy training module, and is used to convert the intervention strategy into control instructions for valves or pump stations and send them via the control bus. The feedback acquisition sub-module has two output channels: Send the real-time pressure sensor readings after execution to the time-series graph neural network prediction and tracing module in the form of the original multi-dimensional time series, update its sliding time window, and use it as the input tensor for the next prediction cycle; Calculate the reward signal based on the pressure data and output it to the reinforcement learning strategy training module for adjusting the intervention strategy; through the cycle of "prediction - strategy generation - instruction issuance - data and reward feedback - re-prediction", a closed-loop adaptive water hammer control for the water supply pipe network is formed.

[0031] Among them, the proactive intervention of the present invention uses a discrete-action deep Q-network (DQN), and its input-output is defined as follows.

[0032] Input state vector: Real-time valve opening: θ1…θ m , m adjustable valves, each represented as a percentage (%); Pressure observation: Instantaneous pressure P1…P of n sensors n (MPa); Pressure change rate: The corresponding first derivative dP1 / d t …dP n / d t (MPa·s -1 ) Therefore, the overall input variables are approximately = [θ1…θm , P1…P n , dP1 / d t …dP n / d t T Output action vector: For each valve, give {x1, x2, x3... x n}} with a total of n relative opening degrees; For each variable frequency pump, give {y1, y2, y3... y n}} with a total of n frequencies.

[0033] Where θ represents the valve opening; P represents the instantaneous pressure; dP1 / d t represents the corresponding first derivative, dP n / d t represents the corresponding nth derivative, x represents the opening, and y represents the frequency.

[0034] The DQN network outputs the Q values of all action combinations, selects the maximum value from them, and executes the corresponding valve-pump speed adjustment command.

[0035] Reward (internal calculation during the training phase), which is divided into two goals: The first calculation index encourages reducing the pressure deviation, and the second calculation index punishes excessive operations. As long as the network makes the network pressure as close as possible to the design value and the actions are not too frequent, it will receive positive incentives; otherwise, it will be punished. After repeated learning in a large number of simulation conditions, the model will learn to directly output the above-mentioned valve and water pump adjustment commands within milliseconds, and these commands are the only results of the reinforcement learning module of the present invention for external use.

[0036] Example 2, please refer to Figures 1 to 5 as shown: A water hammer control method for a water supply network based on multi-functional module fusion in this solution is as follows: Step 1, digital modeling of the water supply network: Based on the geographic information system (GIS) and design drawings, construct a network topology map, deploy sensors and record pipeline attributes; Abstract the water supply network as a heterogeneous graph G=(V, E), deploy pressure sensors upstream and downstream of bifurcation points and valves, the deployment spacing of pressure sensors ≤ 500 meters, and the recorded pipeline attributes include material, length, and pipe diameter.

[0037] Example: Node deployment: Nodes V={v1, v2,..., v n}} correspond to pressure sensors (deployed upstream and downstream of bifurcation points and valves, spacing ≤ 500 meters); Edge attributes: Edges E={e ij}} represent pipelines, and edge attributes include length L ij (m), pipe diameter D ij ​(mm), roughness (mm), historical leakage times C ij .

[0038] Among them, V represents the node list: number the physical locations on the pipeline where sensors can be installed or control can be executed - such as bifurcations, upstream and downstream of valves, and the inlet of pumping stations - one by one as v1, v2... v n , and this long string of numbers is called the node list V.

[0039] (1) E represents the pipe segment list: as long as there is a section of actual pipe directly connected between the numbers v i and v j , we record an edge e ij , and write the information such as the length, diameter, and material of that section of pipe on this edge; all such e ij together are E.

[0040] (2) The subscripts i and j correspond to the starting node vi and the ending node vj respectively; therefore, eij is "the specific pipe section flowing from node i to node j", which uniquely identifies the physical pipe section between these two points and its attributes.

[0041] Step 2, real-time data collection and preprocessing: The sensor collects pipeline pressure data at 0.5 - 2 Hz, fills in the missing values through cubic spline interpolation, and eliminates the outliers by combining the three - standard - deviation principle (3σ) with the EPANET software simulation; Among them, fill in the missing values through cubic spline interpolation, and conduct manual verification when the correlation coefficient between adjacent nodes < 0.7. When the outlier deviation > 20%, it is determined as a wild value.

[0042] Step 3, causal graph generation and model input: Perform causal testing (Granger) on the preprocessed data (the lag order k = 2 - 5, and the time window is set to T = 30 - 90 seconds), generate a sparse causal adjacency matrix, and use it as the input of the graph neural network (TGNN); Among them, the lag order of the causal testing (Granger) is set to k = 2 - 5, and the time window is set to T = 30 - 90 seconds.

[0043] Perform causal testing (Granger) on the historical data (the time window T = 30 - 90 seconds, and the sliding step size Δt = 1 second), where the lag order k = 2 - 5, retain the significant causal edges with p - value ≤ 0.05 and F ij ≥ 2.5, calculate the causal influence value F ij between nodes, and generate an N×N causal influence matrix C.

[0044] Sparsification processing: Eliminate the edges with p - value > 0.05, and retain the significant causal edges F ij≥2.5, forming a sparse adjacency matrix A ∈ {0, 1} n × n .

[0045] This causal graph, as the sparse adjacency matrix of TGNN, can improve the model interpretability and reduce redundant noise edges in the topology.

[0046] Step 4, water hammer risk prediction and traceability: The graph neural network (TGNN) model outputs the probability of water hammer occurrence and the causal propagation chain through the architecture of "graph convolutional neural network (GCN) + gated recurrent unit (GRU) + attention mechanism", and locates the trigger point and propagation path. Among them, the operation method of this system adopts a TGNN model with causal constraints. The input layer of the graph neural network (TGNN) model: multi-dimensional time series (pressure, first derivative, fluctuation intensity) + causal adjacency matrix. Feature extraction layer: 2-layer graph convolutional network (GCN) for capturing the spatial features of the pipe network topology. Temporal modeling layer: 1-layer gated recurrent unit (GRU) for extracting the time dependence of the pressure sequence. Attention layer: multi-head attention mechanism (4 heads) for enhancing the weight assignment of key nodes and paths. Output layer: probability of water hammer occurrence (0 - 1) + causal propagation chain vector (traceability path).

[0047] This solution introduces an attention mechanism inside the model to enhance the perception ability of key nodes and key paths, realizes the learning of the pressure propagation path. At the same time, the output of the graph neural network (TGNN) model not only includes whether a water hammer event will occur, but also can trace back the corresponding causal propagation chain (i.e., from the trigger point to the fluctuation propagation path), assisting the operation and maintenance personnel for accurate traceability positioning.

[0048] Step 5, build a pipe network hydraulic simulation environment through the EPANET platform. In the digital twin environment of the EPANET platform, the deep Q network (Deep Q Network) agent generates intervention actions according to the pressure distribution and predicted location, with the minimum pressure fluctuation as the reward target, and executes the control strategy in real time through the data acquisition and monitoring system (SCADA) to form a closed-loop control. Specifically, it includes the following steps: (1) Digital twin environment construction: Build a pipe network hydraulic simulation environment on the EPANET platform to simulate the real-time pressure distribution and intervention effect, providing a training scenario for reinforcement learning. (2) Reinforcement learning strategy training: The deep Q network (Deep Q Network) agent aims to minimize the pressure fluctuation, learns strategies including valve opening and pump speed adjustment, and combines safety constraints to ensure that the operations meet the engineering boundary conditions. (3) Intervention strategy execution and feedback: Control instructions are sent to edge devices, including valves and pumping stations, through a data acquisition and monitoring system (SCADA) with a delay range of 10 - 500 ms to real-time regulate the status of the pipe network, achieving active control of water hammer risk and forming a closed-loop control.

[0049] Example 3; What is different about this solution compared to Example 2 is: In this embodiment, three typical scenarios of water supply pipe networks are selected, covering different pipe diameters, materials, and water hammer types. The sensor deployment follows the design principles of this solution (upstream and downstream of bifurcation points / valves, spacing ≤ 500 m): Table 1 is a statistical table of experimental scenarios and parameters:

[0050] Table 2 is a statistical table comparing the accuracy of precise prediction and traceability effects:

[0051] Table 3 is a statistical table of the effects of active intervention and closed-loop control;

[0052] The detection accuracy is improved by 15% - 20% compared to the traditional LSTM, and the traceability error is controlled within ≤ 2 node spacings (about 1000 meters), meeting the engineering positioning accuracy requirements. Through the causal adjacency matrix (graph structure) + temporal features (GRU) + attention mechanism, the spatial causal chain and temporal dependence of pressure wave propagation are captured, solving the problems of "local modeling blind spots" and "lack of traceability ability" of traditional methods.

[0053] The peak pressure is reduced by 38.9% - 40%, and the arrival time of the wave peak is delayed by 40% - 59%, far exceeding the traditional method (without active intervention ability), meeting the requirements of the "golden 10 seconds" for water hammer emergency. By training the DQN agent in the EPANET digital twin environment and combining the issuance of millisecond-level control instructions, a closed-loop of "prediction - strategy generation - execution" is achieved.

[0054] Table 4 is a statistical table of cross-scenario generalization ability:

[0055] The accuracy fluctuation under different scenarios is < 5%, significantly better than the traditional LSTM (fluctuation > 12%), and is suitable for pipe networks with pipe diameters of DN200 - DN2000 and mixed materials. Causality graph-driven modeling eliminates topological redundant edges, enabling the model to focus on the real pressure propagation path and reducing the interference of pipe diameter and material differences on prediction.

[0056] From the above tables, it can be seen that this solution has the following advantages compared to the prior art: (1)Precise prediction and tracing: By integrating pipeline network topology and time-series data (sampling frequency 0.5 - 2 Hz), the accuracy of water hammer detection is increased to over 90%, and the tracing error ≤ 2 node spacings.

[0057] (2)Active intervention and closed-loop control: Generate adjustment strategies 1 - 3 seconds before the water hammer peak arrives, reducing the pressure fluctuation amplitude by 38.9% - 40% and delaying the wave crest propagation speed by 40% - 59%.

[0058] (3)Cross-scenario generalization ability: Through causal graph-driven modeling, the prediction accuracy fluctuation of the model in pipeline networks with different pipe diameters (DN200 - DN2000) and materials (steel pipe / plastic pipe) is < 5%, significantly better than traditional methods (fluctuation > 12%).

[0059] Example 4; To further verify the performance of this solution in complex pipeline networks, a hybrid material pipeline network (DN200 - DN2000, 200 nodes) is selected for experiments. As Figure 3 shown, in the figure, the pressure time-series data is first preprocessed, including missing value filling and outlier removal. Subsequently, the data segments are divided through a time window sliding strategy (such as 30 - 90 second window, 1 second step), and the Granger causality test (lag 2 - 5 orders) is performed on each data segment. Calculate the causal influence value between nodes and generate a causal influence matrix; through sparsification processing, redundant edges are removed (retain significant causal edges with p value ≤ 0.05), and finally, a sparse causal adjacency matrix for the time-series graph neural network (TGNN) model is output.

[0060] As Figure 4 shown, the input layer receives pressure time-series data (including pressure value, first derivative, fluctuation intensity) and the causal adjacency matrix. After the preprocessing layer calculates the dynamic features, the spatial features of the pipeline network topology are extracted by the graph convolutional layer (GCN); then the GRU layer is used to capture the time-series dependence relationship, and the key nodes and paths are weighted by the attention layer. Finally, the output layer outputs the water hammer occurrence probability and the causal propagation chain to achieve risk prediction and tracing.

[0061] As Figure 5 shown, in the figure, the EPANET digital twin pipeline network (environment), DQN network module, and experience replay pool are represented by rectangular boxes, and the data flows such as state, action, and reward are marked by arrows, showing the reinforcement learning closed-loop training process: The environment outputs states such as pressure distribution to the DQN network module, and the module generates actions such as adjusting valve / pump speed according to the state and acts on the environment. The environment feedbacks rewards related to pressure fluctuations, and the module stores the interaction experience (including the next state) in the replay pool and uses it to update the network parameters. Through the experience replay mechanism and DQN algorithm, the strategy is optimized to minimize the pressure fluctuation and improve the intervention efficiency.

[0062] As shown in Table 5, in 50 simulated water hammer experiments on a hybrid material pipe network (DN200 - DN2000, 200 nodes, sensor spacing ≤ 500 meters, sampling frequency 1 Hz), the detection accuracy rate of this solution reached 93.2%, an increase of 14.7% compared with traditional LSTM (78.5%) and an increase of 11.1% compared with traditional graph convolutional network (GCN) (82.1%); the tracing error ≤ 1.5 nodes, only 50% of that of the traditional graph convolutional network (GCN) (≤ 3.0 nodes), proving that it overcomes the blind spots of traditional local modeling and realizes the comprehensive and accurate detection and positioning of water hammer events. The model training of this solution only takes 8 hours, a reduction of 20% - 33% compared with traditional LSTM (12 hours) and graph convolutional network (GCN) (10 hours); the data dependence is only 50,000 pieces, a reduction of 37.5% compared with traditional methods, and the cross-scenario accuracy rate fluctuation < 5% (Table 3), reflecting the improvement of causal constraints on learning efficiency and model stability.

[0063] Table 5 is a comparative statistical table of detection accuracy rate and tracing ability:

[0064] As shown in Table 6, in simulated water hammer events, this solution can generate intervention strategies 1 - 3 seconds before the pressure peak arrives, reducing the pressure peak by 41.7% and delaying the wave peak propagation speed by 38%. Compared with manual intervention (reducing 20% - 25% and delaying 10% - 15%), it is 1.7 - 2.5 times more improved, and the misoperation rate is only 0.2 times / month, far lower than 3 - 5 times / month of manual scheduling, achieving a breakthrough from passive alarm to active control.

[0065] Table 6 is a comparative statistical table of intervention response speed and effect:

[0066] As can be seen from the above tables, the specific advantages of this solution are as follows: (1) Precise and efficient: The detection accuracy rate is 93.2%, significantly higher than traditional models (LSTM / GCN is improved by about 11% - 15%); the tracing error ≤ 1.5 nodes, and the positioning accuracy far exceeds that of the graph convolutional network (GCN) (≤ 3 nodes) and manual (> 5 nodes).

[0067] (2) Fast and active intervention: Generate strategies 1 - 3 seconds before the pressure peak arrives, and the response speed reaches the millisecond level (manual requires more than 10 minutes); reduce the pressure peak by 41.7% (manual is only 20% - 25%) and delay the wave peak propagation by 38% (manual is 10% - 15%), with a performance improvement of 1.7 - 2.5 times.

[0068] (3) Low resource dependence: The training takes 8 hours (reduced by 33% - 50% compared with LSTM / GCN), and only 50,000 pieces of data are required (lower than traditional models).

[0069] (4) High reliability: The misoperation rate is 0.2 times per month, less than 1 / 15 of manual scheduling (3 - 5 times per month), completely avoiding the passive defects of traditional alarm systems.

[0070] In summary, as can be seen from Embodiment 1 to Embodiment 4, compared with the prior art, the present invention has the following significant advantages: (1) Combining graph structure and time - series modeling to improve prediction accuracy and positioning ability; Abstracting the water supply network as a graph structure and combining time - series modeling can comprehensively reflect the propagation path of pressure waves, overcome the blind spots of traditional local modeling, and improve the comprehensiveness and accuracy of water hammer event detection.

[0071] (2) Introducing causal graphs to guide the modeling of time - series graph neural networks (TGNNs) to improve model interpretability and stability; Taking the causal analysis results as the input of the adjacency matrix can effectively eliminate the interference of irrelevant edges in the topology on prediction, and improve the learning efficiency and causal traceability ability of the model.

[0072] (3) Implementing a closed - loop control of prediction - traceability - intervention; Through the reinforcement learning module, the system can not only achieve passive detection, but also actively adjust the operation state of the water supply network, thereby reducing water hammer damage and forming an intelligent closed - loop control system.

[0073] (4) Having engineering deployability and generality; The system structure is clear and the modules are distinct. It can be deployed on the existing water supply monitoring platform, adapt to water supply networks of various scales and complexities, and has broad promotion prospects.

[0074] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non - exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0075] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A water hammer control method for a water supply network based on the integration of multifunctional modules, characterized in that include: S1, digital modeling of pipeline network: based on geographic information system and design drawings, construct pipeline network topology map, deploy sensors and record pipeline attributes; S2, real-time data acquisition and preprocessing: The sensor collects pipeline pressure data at 0.5-2Hz, fills missing values through cubic spline interpolation, and removes outliers by triple standard deviation principle combined with EPANET software simulation; S3, causal graph generation and model input: perform causal tests on preprocessed data and generate a sparse causal adjacency matrix as the input of the graph neural network; S4, Water hammer risk prediction and traceability: The graph neural network model outputs the probability of water hammer occurrence and the causal propagation chain through the "graph convolutional neural network + gated recurrent unit + attention mechanism" architecture, and locates the trigger point and propagation path; S5, reinforcement learning and intelligent intervention: A pipeline network water conservancy simulation environment is built through the EPANET platform. In the digital twin environment of the EPANET platform, the DQN agent of the deep Q network generates intervention actions based on the pressure distribution and predicted position, with minimizing pressure fluctuations as the reward goal. The control strategy is executed in real time through the data acquisition and monitoring system to form a closed-loop control.

2. The water hammer control method for a water supply pipe network based on the integration of multifunctional modules according to claim 1, wherein The sensor described in S1 is a pressure sensor, and is deployed at the bifurcation point of the pipeline and upstream and downstream of the valve. The pressure sensor deployment spacing is ≤500 meters. The recorded pipeline properties include material, length and pipe diameter.

3. A water hammer control method for a water supply pipeline network based on the integration of multifunctional modules according to claim 1, characterized in that, As described in S2, missing values were filled by cubic spline interpolation, and manual verification was performed when the correlation coefficient of adjacent nodes was <0.

7. The abnormal value deviation >20% was determined to be an outlier.

4. A water hammer control method for a water supply network based on the integration of multifunctional modules according to claim 1, characterized in that, The lag order of the causal test described in S3 is set to k = 2~5, and the time window is set to T = 30~90 seconds.

5. A water hammer control method for a water supply network based on the integration of multifunctional modules according to claim 1, characterized in that, The S5 specifically includes: S5.1, Digital Twin Environment Construction: Build a pipe network hydraulic simulation environment on the EPANET platform to simulate real-time pressure distribution and intervention effects, and provide training scenarios for reinforcement learning; S5.2, reinforcement learning strategy training: The DQN agent of the deep Q network aims to minimize pressure fluctuations, learns valve opening and pump speed adjustment strategies, and combines safety constraints to ensure that operations meet engineering boundary conditions; S5.3, intervention strategy execution and feedback: The control instructions are sent to edge devices, including valves and pump stations, with a delay range of 10 to 500 ms through the data acquisition and monitoring system to adjust the status of the pipeline network in real time and form a closed-loop control.

6. A water hammer control system for a water supply network based on the integration of multi-functional modules, applicable to a water hammer control method for a water supply network based on the integration of multi-functional modules according to any one of claims 1-5, characterized in that, include: Pressure propagation map construction module: collects raw pressure data from sensors and constructs the raw data into a sparse relationship graph, collects data based on the pressure bus, calculates pipeline interconnection information, and sends the pipeline interconnection results to the time series graph neural network prediction and tracing module for risk prediction and propagation path tracing; Time-series graph neural network prediction and tracing module: Build a graph neural network model based on causal constraints, calculate the results in the pressure propagation map construction module, realize water hammer risk prediction and propagation path tracing, and pass the prediction results to the reinforcement learning intervention decision module; Reinforcement learning intervention decision module: Analyze the output results of the timing graph neural network prediction and traceability module, generate active intervention strategies through reinforcement learning agents, and send them to pump stations and valves through the control bus. After the strategy is executed, re-read the latest pressure data, and make learning adjustments based on the pressure data to form a self-closed "prediction → control → re-learning" cycle.

7. A water hammer control system for a water supply network based on the integration of multifunctional modules according to claim 6, characterized in that, The pressure propagation map construction module includes: Pipeline network digital modeling module: Based on the geographic information system and design drawings, build the pipeline network topology map, deploy sensors and record pipeline attributes; Real-time data acquisition and preprocessing module: The sensor collects pipeline pressure data at 0.5-2Hz, fills in missing values through cubic spline interpolation, and removes abnormal values by combining the triple standard principle with EPANET software simulation; Causal graph generation and model input module: performs causal testing on preprocessed data and generates a sparse causal adjacency matrix as input to the graph neural network.

8. A water hammer control system for a water supply network based on the integration of multifunctional modules according to claim 6, characterized in that, The time series graph neural network prediction and traceability module adopts a graph neural network model with causal constraints, including: Input layer: multidimensional time series, including pressure, first-order derivative, fluctuation intensity plus causal adjacency matrix; Feature extraction layer: It consists of two layers of graph convolutional networks, which are used to capture the spatial characteristics of the pipeline network topology; Temporal modeling layer: includes a layer of gated recurrent units to extract the temporal dependencies of the pressure sequence; Attention layer: a multi-head attention mechanism is introduced into the model to enhance the perception of key nodes and key paths, realize the learning of pressure propagation paths, and enhance the weight distribution of key nodes and paths; the attention layer is configured to weight the features of key nodes and key paths within the model to enhance the weight distribution of key nodes and key paths; Output layer: water hammer occurrence probability plus causal propagation chain vector, that is, the traceability path not only includes whether a water hammer event will occur, but also can trace back the corresponding causal propagation chain, that is, from the trigger point to the wave propagation path, to assist operation and maintenance personnel in accurate traceability and positioning.

9. A water hammer control system for a water supply network based on the integration of multifunctional modules according to claim 6, characterized in that, The reinforcement learning intervention decision module includes: Digital twin environment building module: Build a pipe network hydraulic simulation environment on the EPANET platform to simulate real-time pressure distribution and intervention effects, and provide training scenarios for reinforcement learning; Reinforcement learning strategy training module: The DQN agent of the deep Q network aims to minimize pressure fluctuations, and learns strategies for regulating valve opening and pump speed, combining safety constraints to ensure that operations meet engineering boundary conditions. Its input is connected to the output of the time-series graph neural network prediction and tracing module to generate intervention strategies based on water hammer risk probability and propagation path information. Intervention strategy execution and feedback module: including an execution module and a feedback collection submodule; the input end of the execution module is connected to the output end of the reinforcement learning strategy training module, and is used to convert the intervention strategy into a control instruction for a valve or pump station and issue it via a control bus. The feedback collection submodule has two output paths: (1) The real-time pressure sensor readings after execution are sent to the time series graph neural network prediction and tracing module in the form of original multi-dimensional time series, and its sliding time window is updated as the input tensor of the next prediction cycle; (2)Calculate the reward signal based on the pressure data and output it to the reinforcement learning policy training module for adjusting the intervention strategy; through the cycle of "prediction - policy generation - instruction issuance - data and reward feedback - re - prediction", a closed - loop adaptive water hammer control for the water supply network is formed.

Images