Pipe network on-line monitoring system

Abstract

The invention discloses a pipe network on-line monitoring system, which relates to the field of urban capital construction and comprises a sensing deployment module, a signal acquisition module, a signal processing module, a pipe network modeling module, a leakage positioning module, an edge calculation module, an energy scheduling module, an anomaly prediction module, a deposition evaluation module, an early warning visual module and a center management module. According to the method, the recognition capability of the system on a hidden and complex pipeline structure is improved, accurate and efficient three-dimensional modeling can be carried out, the path deviation resistance capability is improved, and the more comprehensive and robust leakage sensing capability is realized; the method is advantaged in that communication burden and response time delay are substantially reduced, sensitivity and foresight of abnormal evolution trend identification are improved, overall identification accuracy of the system is improved, a global optimal risk solution can be efficiently output, a high-risk area is visually displayed, and rapid pre-judgment and decision making of operation and maintenance personnel are facilitated.

Description

Technical Field

[0001] The present invention relates to the field of urban infrastructure, and in particular to an online pipe network monitoring system. Background Art

[0002] With the rapid development of urban infrastructure and the continuous expansion of underground pipeline systems, pipeline networks for various types, such as water supply, drainage, gas, and heat, are becoming increasingly dense and complex. As crucial "invisible infrastructure" for urban operations, the stable operation of underground pipeline networks is directly related to public safety and municipal security. In recent years, leaks caused by pipeline aging, accidental construction damage, and abnormal pressure have frequently occurred, posing a serious threat to urban safety and resource efficiency. Although various monitoring technologies, such as fiber optic sensing, vibration detection, and pressure analysis, have been introduced to pipeline network monitoring, traditional manual inspections and distributed monitoring methods face bottlenecks such as insufficient coverage, poor real-time performance, and delayed emergency response. In particular, in trenchless areas, deep buried areas, or areas with dense branch lines, existing methods struggle to accurately identify topological changes and hidden risk sources, leading to frequent leaks, siltation, corrosion, and even secondary disasters. To achieve refined and intelligent management of underground pipeline networks, an online pipeline network monitoring system is urgently needed.

[0003] After searching, Chinese patent number CN112268229A discloses an online monitoring system for pipeline networks. Although this invention realizes real-time online monitoring of the operating status of pipelines in the pipeline network and uses relays to achieve reliable long-distance wireless communication, avoiding the problems of large construction volume and high maintenance costs caused by communication cables, the system has poor recognition ability for hidden and complex pipeline structures, cannot perform accurate and efficient three-dimensional modeling, and has reduced its ability to resist path deviation. In addition, the sensitivity and foresight of existing pipeline network online monitoring systems in identifying abnormal evolution trends are low, the overall recognition accuracy of the system is low, and it is unable to efficiently output the global optimal risk solution, which is not conducive to the rapid prediction and decision-making of operation and maintenance personnel. To this end, we propose a pipeline network online monitoring system. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects in the prior art and provide an online monitoring system for a pipe network.

[0005] The present invention proposes an online monitoring system for a pipe network, which includes: a sensing deployment module, a signal acquisition module, a signal processing module, a pipe network modeling module, a leakage location module, an edge computing module, an energy scheduling module, an anomaly prediction module, a siltation assessment module, an early warning visualization module, and a central management module; The sensing deployment module is used to install low-power electromagnetic markers and various sensors at fixed points on the outer wall of the pipeline to collect pipeline network electromagnetic field data and various sensor data in real time; The signal acquisition module uses a mobile inspection robot to dynamically move and receive electromagnetic field data and various sensor data collected by the perception deployment module in real time; The signal processing module is used to pre-process the electromagnetic disturbance data and various sensor data collected by the signal acquisition module; The pipe network modeling module constructs a three-dimensional pipe network topology map based on the inspection robot's movement route and the processed electromagnetic disturbance data, and updates the connection relationship between pipeline nodes and branches in real time; The leakage location module evaluates the spatiotemporal correlation between sensor data based on the three-dimensional pipe network topology map and various sensor data, and performs double-blind location of the leakage point; The edge computing module is used to deploy edge terminal devices in the pipe well, receive various processed sensor data in real time, and perform predictive analysis on the edge sensor data; The energy scheduling module monitors the energy status of each edge terminal device in real time and automatically switches to the working mode of each edge terminal device; The anomaly prediction module uses the edge and center collaborative processing mechanism to predict the abnormal development trend of the pipeline; The siltation assessment module calculates the siltation probability distribution and assesses the siltation risk of the pipeline network based on the three-dimensional pipeline network topology map and various sensor data; The early warning visual module constructs a multi-level risk cloud map based on topological information, leakage point location and sedimentation risk data, and displays early warnings and classifies risk levels for each pipe network area; The central management module is used to coordinate the operating status of each module and provide a remote operation and maintenance interface and strategy optimization suggestions.

[0006] As a further solution of the present invention, the sensors deployed by the perception deployment module specifically include acceleration sensors, acoustic emission sensors, infrasound sensors, temperature sensors, flow sensors, turbidity sensors, pH sensors, conductivity sensors, magnetoresistance sensors, voltage sensors, current sensors and humidity sensors.

[0007] As a further solution of the present invention, the specific steps of the pipe network modeling module to construct a three-dimensional pipe network topology model are as follows: S1.1: The mobile inspection robot continuously collects electromagnetic field data while moving along the pipeline. Each set of collected electromagnetic field data is subjected to spatiotemporal normalization to eliminate deviations caused by different acquisition speeds and paths. The waveform gradient method is used to extract electromagnetic field mutation points, i.e., electromagnetic field distortion points caused by the presence of electromagnetic markers or structural changes. S1.2: Project each extracted electromagnetic field mutation point into the three-dimensional space coordinate system, and establish the spatial electromagnetic field mutation point set V={v1,v2,v3,...,v n}, where v n Represents the nth electromagnetic field mutation point, and constructs an undirected graph G = (V, E) based on the electromagnetic field mutation point set, where V represents each node in the undirected graph G, that is, each electromagnetic field mutation point, and E represents an edge to be evaluated between each pair of electromagnetic field mutation points at the beginning; S1.3: In an undirected graph consisting of electromagnetic field mutation points, multiple groups of virtual individuals are initialized. The starting node of each group of virtual individuals is randomly selected. Each virtual individual excludes the adjacent nodes that have been visited based on its current position and calculates the local attraction of the corresponding edges of the remaining adjacent nodes to form the current driving path set. S1.4: Calculate the pheromone concentrations of the edges between nodes in the undirected graph based on the three-dimensional Euclidean distances between each electromagnetic field mutation point and the correlation coefficients of the magnetic field perturbation curves. Then, calculate the probability of each virtual individual moving from the current node to each adjacent node based on the local attractiveness and pheromone concentration of each edge in the set of travel paths. Based on the selection probabilities of each adjacent node, determine the next movement destination using a roulette wheel method. S1.5: Each virtual individual jumps to a node in turn until no more nodes are available. Path construction ends, and the legitimacy of the paths constructed by each virtual individual is evaluated using a path legitimacy cost function. After each round of path construction, the quality of each path is evaluated. Based on the quality of the path traversed by the virtual individual, the pheromone concentration of each edge in the path is updated. At the same time, the local attractiveness of each edge is updated based on the historical jump frequency or perturbation matching degree. S1.6: If the undirected graph does not undergo structural changes after multiple rounds of iterations, the path is perturbed and the positions of each virtual individual are reset. The path is reconstructed, the pheromone concentration is updated, and the local attractiveness is updated again. After multiple rounds of iterations, the pheromone concentration change value of the path converges to the preset threshold, and the iteration is stopped. S1.7: Record the paths constructed by each virtual individual after the iteration, filter out the edges in each path that are below the preset pheromone threshold, and use the processed paths as a connected subgraph, that is, the global optimal path branch. Based on each connected subgraph, form the final topological structure, identify the three-dimensional coordinates of the electromagnetic field mutation points on each path, and then generate smooth pipe sections through the fitting algorithm to construct a complete three-dimensional pipe network topology map.

[0008] As a further solution of the present invention, the specific calculation formula of the waveform gradient method described in S1.1 is as follows: Where, Represents the disturbance intensity of the kth sampling point, when If it is higher than the preset threshold, the current sampling point is judged to be an electromagnetic field mutation point; represents the electromagnetic intensity gradient at the kth sampling point; represents the electromagnetic intensity at the kth sampling point; Represents the spatial coordinate axis where the inspection robot is located; The specific calculation formula for the selection probability described in S1.4 is as follows: Where, Represents that at time t, the virtual individual from node Move to Node probability; Represents the current path Upper pheromone concentration; represents the heuristic factor of the path, and ; as well as represent the parameters controlling the pheromone concentration and the heuristic factor weight respectively; Representative Path Upper pheromone concentration; Representative Path The inspiration factor; Representative Node The set of candidate neighbors.

[0009] As a further solution of the present invention, the leakage location module evaluates the spatiotemporal correlation between the sensor data and performs double-blind location of the leakage point in the following specific steps: S2.1: The leak location module extracts the pipeline vibration, infrasound, and temperature signals from the preprocessed sensor data. It samples each received set of physical signals within a fixed time window to synchronize the signal times and integrates the physical signals to construct a standardized sample set. S2.2: Based on the standardized sample set, set time series samples of each physical signal at each node location to construct a unified feature tensor. Then, based on the latest 3D pipeline network topology map, map each sensor sampling location to a pipeline topology node. Then, through discrete modeling, establish a signal distribution model for each type of signal at each node in different time windows. S2.3: Calculate the spatial distances between different nodes on each signal data channel. Based on the calculated spatial distances, obtain the joint signal difference scores for each node pair and construct an inter-node symmetric matrix. Normalize the inter-node symmetric matrix and construct an anomaly propagation structure diagram based on a preset threshold. Based on the anomaly propagation structure diagram, perform a comprehensive normalization construction on the signal mutation slope, peak amplitude, and temperature of each node to generate a comprehensive signal anomaly amplitude index corresponding to each node. S2.4: Calculate the anomaly intensity value of each node in the anomaly propagation diagram in the three-dimensional pipe network topology diagram based on the comprehensive anomaly amplitude index of the signal of each node, arrange the anomaly intensity values in descending order, and select the corresponding number of high-intensity nodes from high to low according to the preset ratio as the core anomaly candidate set.

[0010] As a further solution of the present invention, the specific calculation formula of the spatial distance in S2.3 is as follows: Where, Represents the spatial distance between node i and node j on signal channel m, where signal channel m includes the physical signal channels of pipeline vibration acceleration, infrasound signal amplitude, and temperature gradient sequence; Represents the set of all signal transmission matrices that meet the preset marginal constraints; represents the probability distribution of node i on signal channel m; represents the probability distribution of node j on signal channel m; represents the matching strategy for transmitting the quality from the u-th time sampling point of node i to the v-th time sampling point of node j; Represents the signal value of node i at the u-th time sampling point on signal channel m; Represents the signal value of node j at the uth time sampling point on signal channel m; p is used to control the distance measurement form. is the Manhattan distance, When is the Euclidean distance; The specific form of the abnormal propagation structure diagram described in S2.3 is as follows: Where, Indicates whether there is a connection between node and node j in the abnormal propagation structure graph, 1 represents the existence of a connection, and 0 represents the absence of a connection. Node i and node j are consistent with the node i and node j in the spatial distance calculation formula; M represents the symmetric matrix between nodes; Represents the abnormality threshold.

[0011] As a further solution of the present invention, the specific steps of the edge computing module performing predictive analysis on edge sensor data are as follows: S3.1: Each deployed edge terminal device receives sensor data on flow rate, pressure, vibration, electromagnetic field intensity, temperature, and turbidity collected by sensors within its respective pipe network area. It then divides the sensor data into multiple sensor data sets of consistent length, based on a sliding time window preset by the edge terminal device. S3.2: Normalize each sensor dataset using the local mean difference method, set a soft threshold, and filter out high-frequency random disturbances using Daubechies wavelet decomposition. Then, load the lightweight LSTM network model on each edge device and initialize the hidden and memory states of the lightweight LSTM network model to 0 or the previous historical state. S3.3: Each sensor dataset is fed into the lightweight LSTM network model in order from oldest to newest according to the corresponding time. The lightweight LSTM network model forward propagates each sensor dataset received, processes each sensor data in each sensor dataset layer by layer, and updates the hidden state corresponding to each time step until all sensor datasets at the last time step are processed. S3.4: Input the hidden state of the last time step into the linear regression layer to predict the value at the next moment. Calculate the prediction error between the predicted value and the actual value at the next moment. Based on the current prediction error, calculate the short-term anomaly deviation rate corresponding to each type of sensor data as the anomaly score. The specific calculation formula for the short-term anomaly deviation rate is as follows: Where, Represents short-term abnormal deviation; Represents the predicted Sensor value at the moment; Representing reality The sensor value at the moment; Represents the minimum constant that prevents division by zero; S3.5: Based on the anomaly scores of the combined sensor data, a scoring dataset is constructed, and event discrimination is performed through a lightweight classifier. If any anomaly score is higher than the preset anomaly threshold of the corresponding sensor, the pipe network area under the jurisdiction of the edge terminal device is determined to be in a "critical" or "abnormal" state, and the timestamp, channel number, and prediction error are uploaded to the central management module for secondary confirmation.

[0012] As a further solution of the present invention, the specific steps of the abnormality prediction module predicting the abnormal development trend of the pipeline are as follows: S4.1: The central management module regularly obtains prediction values, error information and various sensor data from each edge terminal device to build a collaborative time series matrix ,in represents the collaborative time series of the nth node, represents the multi-factor observation point obtained by fusion calculation of predicted value, error information and various sensor data, n represents the node number, T represents the time length, and then the trend interpolation method is used to fill the missing values in each time step; S4.2: The anomaly prediction module sets sliding window lengths at different levels, extracts multi-scale sequences from the constructed collaborative time series matrix, calculates the local trend differential increments for each scale sequence, analyzes trend changes based on the calculated trend differential increments, and then fits the changing trends at different time steps through a fixed-length sliding time window to generate a trend evolution curve. The local trend differential increments at multiple scales are then weighted and fused to generate trend risk scores at different time steps. S4.3: If the trend risk score is higher than the preset risk threshold, it is considered a trend anomaly and the corresponding network node area is marked as a potential risk segment. At the same time, the nodes are combined in pairs according to different combinations to establish different node pairs. The trend risk series of the two groups of nodes in different node pairs over the past N time steps are extracted to construct multiple groups of N×N grid graphs. Each point in the grid graph represents the local Euclidean distance between the two nodes at different time steps. S4.4: Start from the starting point of each grid graph and move toward the end point, and dynamically plan the cumulative minimum cost path to generate the DTW distance of each node pair. Collect the DTW distances of all node pairs and generate a symmetric similarity matrix. Then, perform density clustering on the generated symmetric similarity matrix to construct multiple trend-coordinated node clusters. The cluster center or the node where the trend changes first is used as the potential source of anomaly propagation. Then, construct an anomaly propagation graph based on the time delay gradient to restore the risk diffusion path, and output the risk value for multiple time steps in the future through the Transformer prediction model.

[0013] As a further solution of the present invention, the specific calculation formula of the local trend differential increment described in S4.2 is as follows: Where, represent Time Node In the The trend increment on the scale; represents the The sliding window length of the scale; represent Time Node The scale sequence value of ; represent Time Node The scale sequence value of A random positive number.

[0014] As a further solution of the present invention, the specific steps of the siltation assessment module for assessing the risk of pipe network siltation are as follows: S5.1: The sedimentation assessment module collects flow, turbidity, and pH sensor data from each sensor node, then time-aligns the collected sensor data. Based on the processed sensor data, it calculates the flow information entropy change rate, turbidity gradient mean, and pH drift value for each pipe network area, three sets of sedimentation assessment indicators. S5.2: Based on historical training or expert experience, set weight coefficients for the three calculated results: the rate of change of flow information entropy, the mean value of the turbidity gradient, and the pH drift value. Integrate the three sets of sedimentation assessment indicators to establish a comprehensive sedimentation risk cost function. S5.3: Set each pipe segment in the pipe network as a set of binary variables ,in Representative If there is siltation in the pipe section, it means there is no siltation in the pipe section. , L represents the total number of monitoring sections in the pipeline. The overall state vector is constructed based on the binary variables of each section. Then, based on the overall state vector, the comprehensive sedimentation risk cost function is converted into the QUBO cost function to establish the corresponding QUBO matrix; S5.4: Input the QUBO matrix into the quantum annealer and set the annealing temperature, number of iterations, and number of measurement samples. Initialize the binary variables of each pipe segment to the corresponding quantum superposition state. Then map the QUBO cost function to the Hamiltonian of the constructed problem to obtain the corresponding target Hamiltonian. Based on the initialized sets of quantum superposition states, initialize the driving Hamiltonian corresponding to the current quantum annealer. S5.5: By controlling the time evolution of parameters, the initial driving Hamiltonian is transformed into the target Hamiltonian. The Hamiltonian is then driven to gradually approach the ground state of the target Hamiltonian, while the corresponding cost function value is minimized. A "quantum measurement" is performed on this state to obtain a specific binary solution vector. The problem Hamiltonian construction, driving Hamiltonian initialization, and driving Hamiltonian evolution are then repeated until the preset number of iterations is reached. S5.6: Count the occurrences of each pipe segment The number of times the siltation occurs is counted and the probability of occurrence is calculated. At the same time, based on the calculated siltation probability values, a pipe section probability vector is constructed and the “siltation risk probability map” of each pipe section is output.

[0015] Beneficial effects of the present invention: The present invention collects electromagnetic disturbance data through a mobile inspection robot, extracts electromagnetic field mutation points and projects them into three-dimensional space, constructs a node set and an undirected graph structure, initializes multiple groups of virtual individuals, and randomly selects and sets the starting node of each group of virtual individuals. Each virtual individual excludes the visited adjacent nodes based on the current position, and calculates the local attraction of the corresponding edges of the remaining adjacent nodes to form a current driving path set. According to the mutual correlation coefficient of the three-dimensional space Euclidean distance between each electromagnetic field mutation point and the magnetic field disturbance curve, the pheromone concentration of the edges between each node in the undirected graph is calculated. Then, according to the local attraction and pheromone concentration of each edge in the driving path set, the distance of each virtual individual from the current node to each node is calculated. The selection probability of adjacent nodes is based on the selection probability of each adjacent node. The roulette method is used to determine the next moving target to construct the optimal path set and reconstruct the three-dimensional topological structure. The system further integrates multi-physical sensor signals such as vibration, infrasound and temperature to construct a spatiotemporal feature tensor and map it to the topological map nodes. By calculating the signal distribution differences between nodes, an abnormal propagation map is constructed, the signal mutation slope and temperature anomaly amplitude are extracted, and high-intensity abnormal nodes are identified. Ultimately, topology recognition and double-blind positioning of leakage sources are achieved, which improves the system's ability to recognize hidden and complex pipeline structures, enables accurate and efficient three-dimensional modeling, improves the ability to resist path deviation, and achieves more comprehensive and robust leakage perception capabilities.

[0016] The present invention deploys multiple edge terminal devices, collects and pre-processes various sensor data in the pipe network area, uses a lightweight LSTM model to predict signal trends and judge abnormal trends, sets sliding window lengths of different levels for the abnormal prediction module, extracts multi-scale sequences from the constructed collaborative time series matrix, calculates the local trend differential increment of each scale sequence, analyzes trend changes based on the calculated trend differential increment, and then fits the changing trends of different time steps through a sliding time window of a fixed length to generate a trend evolution curve, and then performs weighted fusion on the local trend differential increments of multiple scales to generate trend risk scores at different time steps, and then uses DTW dynamic time to calculate the trend risk score. Regularization and density cluster analysis are performed to identify trend collaborative node clusters and abnormal source points, and future risk evolution is predicted through the Transformer model. The sedimentation assessment module integrates flow entropy, turbidity gradient and pH drift, and then constructs a QUBO risk cost function. The quantum annealing machine is input to solve the minimum cost path of each pipe section, the sedimentation probability is calculated, and a sedimentation risk probability map is generated to achieve intelligent risk identification and early warning, significantly reduce communication burden and response delay, enhance the sensitivity and foresight of abnormal evolution trend identification, improve the overall recognition accuracy of the system, and can efficiently output the global optimal risk solution and intuitively display high-risk areas, which helps operation and maintenance personnel to quickly predict and make decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a framework diagram of a pipeline network online monitoring system. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0021] Example 1: This embodiment of the present invention provides a pipe network online monitoring system. Figure 1 , Figure 1 This is a framework diagram of a pipeline network online monitoring system provided by an embodiment of the present invention. The system includes: a sensing deployment module, a signal acquisition module, a signal processing module, a pipeline network modeling module, a leak location module, an edge computing module, an energy scheduling module, an anomaly prediction module, a siltation assessment module, an early warning visualization module, and a central management module.

[0022] The perception deployment module is used to install low-power electromagnetic markers and various sensors at fixed points on the outer wall of the pipeline to collect the electromagnetic field data of the pipeline network and various sensor data in real time; the signal acquisition module uses a mobile inspection robot to dynamically move and receive the electromagnetic field data and various sensor data collected by the perception deployment module in real time; the signal processing module is used to pre-process the electromagnetic disturbance data and various sensor data collected by the signal acquisition module.

[0023] The pipeline network modeling module constructs a three-dimensional pipeline network topology map based on the inspection robot's movement route and processed electromagnetic disturbance data, and updates the connection relationship between pipeline nodes and branches in real time.

[0024] Specifically, the mobile inspection robot continuously collects electromagnetic field data while moving along the pipeline, and performs spatiotemporal normalization on each set of collected electromagnetic field data to eliminate the deviation caused by different acquisition speeds and paths. The waveform gradient method is used to extract electromagnetic field mutation points, that is, electromagnetic field distortion points caused by the presence of electromagnetic markers or structural changes. Each extracted electromagnetic field mutation point is projected into a three-dimensional spatial coordinate system, and a spatial electromagnetic field mutation point set V = {v1, v2, v3, ..., v n}, where v nRepresents the nth electromagnetic field mutation point, and constructs an undirected graph G=(V,E) based on the set of electromagnetic field mutation points, where V represents each node in the undirected graph G, that is, each electromagnetic field mutation point, and E represents an edge to be evaluated between each pair of electromagnetic field mutation points at the initial stage. In the undirected graph composed of electromagnetic field mutation points, multiple groups of virtual individuals are initialized, and the starting node of each group of virtual individuals is set by randomly selecting. Each virtual individual excludes the visited adjacent nodes based on the current position, and calculates the local attraction of the corresponding edges of the remaining adjacent nodes to form the current driving path set. According to the mutual correlation coefficient between the three-dimensional Euclidean distance between each electromagnetic field mutation point and the magnetic field perturbation curve, the pheromone concentration of the edges between each node in the undirected graph is calculated. Then, according to the local attraction and pheromone concentration of each edge in the driving path set, the selection probability of each virtual individual moving from the current node to each adjacent node is calculated. Based on the selection probability of each adjacent node, the next moving target is determined by the roulette method, and each virtual individual jumps to the node in turn until there is no choice. When the virtual individuals are in the node state, the path construction is terminated and the legitimacy of the paths constructed by each virtual individual is evaluated by the path legitimacy cost function. After each round of path construction, the quality of each path is evaluated and the pheromone concentration of each edge in the path is updated according to the quality of the path traversed by the virtual individual. At the same time, the local attraction of each edge is updated according to the historical jump frequency or perturbation matching degree. If the undirected graph does not change in multiple rounds, the path is perturbed and the position of each virtual individual is reset. The path construction, pheromone concentration update and local attraction update are repeated again until the pheromone concentration change value of the path converges to the preset threshold after multiple rounds of iteration. The iteration is stopped and the paths constructed by each virtual individual after the iteration is recorded. The edges in each path below the preset pheromone threshold are screened out and the processed paths are regarded as a connected subgraph, i.e., the global optimal path branch. The final topology is formed based on each connected subgraph and the three-dimensional coordinates of the electromagnetic field mutation point on each path are identified. Then, a smooth pipe segment is generated by the fitting algorithm to construct a complete three-dimensional pipe network topology.

[0025] It should be further explained that the specific calculation formula of the waveform gradient method is as follows: Where, Represents the disturbance intensity of the kth sampling point, when If it is higher than the preset threshold, the current sampling point is judged to be an electromagnetic field mutation point; represents the electromagnetic intensity gradient at the kth sampling point; represents the electromagnetic intensity at the kth sampling point; Represents the spatial coordinate axis where the inspection robot is located; The specific calculation formula for selection probability is as follows: Where, Represents that at time t, the virtual individual from node Move to Node probability; Represents the current path Upper pheromone concentration; represents the heuristic factor of the path, and ; as well as represent the parameters controlling the pheromone concentration and the heuristic factor weight respectively; Representative Path Upper pheromone concentration; Representative Path The inspiration factor; Representative Node The set of candidate neighbors.

[0026] The leakage location module evaluates the spatiotemporal correlation between sensor data based on the three-dimensional pipe network topology map and various sensor data, and performs double-blind location of the leakage point.

[0027] Specifically, the leakage location module extracts the pipeline vibration, infrasound wave signal and temperature physical signal groups from the pre-processed sensor data, and samples each group of received physical signals within a fixed time window to synchronize the signal time, integrate each group of physical signals to build a standardized sample set, and according to the standardized sample set, set the time series sample of each physical signal at each node position to build a unified feature tensor. Then, according to the latest three-dimensional pipeline network topology map, map each sensor sampling position to the pipeline topology node, and then establish a signal distribution model for each type of signal at each node in different time windows through discrete modeling, and calculate the spatial distance between different nodes on each signal data channel. According to the calculated spatial distance, the joint signal difference score of each node pair is obtained, and a symmetric matrix between nodes is constructed. The symmetric matrix between nodes is then normalized. According to the preset threshold, an abnormal propagation structure diagram is constructed. Based on the abnormal propagation structure diagram, the signal mutation slope, peak amplitude and temperature of each node are comprehensively normalized to generate the signal comprehensive abnormal amplitude index corresponding to each node. According to the signal comprehensive abnormal amplitude index of each node, the abnormal intensity value of each node in the three-dimensional pipe network topology diagram in the abnormal propagation diagram is calculated, and the abnormal intensity values are arranged in descending order. According to the preset ratio, the corresponding number of high-intensity nodes are selected from high to low as the core abnormality candidate set.

[0028] In this embodiment, the specific calculation formula of the spatial distance is as follows: Where, Represents the spatial distance between node i and node j on signal channel m, where signal channel m includes the physical signal channels of pipeline vibration acceleration, infrasound signal amplitude, and temperature gradient sequence; Represents the set of all signal transmission matrices that meet the preset marginal constraints; represents the probability distribution of node i on signal channel m; represents the probability distribution of node j on signal channel m; represents the matching strategy for transmitting the quality from the u-th time sampling point of node i to the v-th time sampling point of node j; Represents the signal value of node i at the u-th time sampling point on signal channel m; Represents the signal value of node j at the uth time sampling point on signal channel m; p is used to control the distance measurement form. is the Manhattan distance, When is the Euclidean distance; The specific form of the abnormal propagation structure diagram described in S2.3 is as follows: Where, Indicates whether there is a connection between node and node j in the abnormal propagation structure graph, 1 represents the existence of a connection, and 0 represents the absence of a connection. Node i and node j are consistent with the node i and node j in the spatial distance calculation formula; M represents the symmetric matrix between nodes; Represents the abnormality threshold.

[0029] Example 2: This embodiment of the present invention provides a pipe network online monitoring system. Figure 1 , Figure 1 This is a framework diagram of a pipeline network online monitoring system provided by an embodiment of the present invention. The system includes: a sensing deployment module, a signal acquisition module, a signal processing module, a pipeline network modeling module, a leak location module, an edge computing module, an energy scheduling module, an anomaly prediction module, a siltation assessment module, an early warning visualization module, and a central management module.

[0030] The edge computing module is used to deploy edge terminal devices in the pipe well, receive various processed sensor data in real time, and perform predictive analysis on the edge sensor data.

[0031] Specifically, each deployed edge terminal device receives various sensor data such as flow rate, pressure, vibration, electromagnetic field strength, temperature and turbidity collected by sensors in the pipe network area under its jurisdiction, and divides various sensor data into multiple groups of sensor data sets with the same length according to the sliding time window preset by the edge terminal device. Each group of sensor data sets is normalized by the local mean difference method, and then the soft threshold is set. The high-frequency random disturbance is filtered out through Daubechies wavelet decomposition, and then the lightweight LSTM network model of each edge terminal device is loaded, and the hidden state and memory state of the lightweight LSTM network model are initialized to 0 or the previous historical state. Each group of sensor data sets are input into the lightweight LSTM network model in sequence from old to new according to the corresponding time. The lightweight LSTM network model will receive each group of sensor data sets. The sensor data set is forward propagated, and each sensor data in each sensor data set is processed layer by layer, and the hidden state corresponding to each time step is updated until each group of sensor data sets of the last time step is processed. The hidden state of the last time step is input into the linear regression layer to predict the value at the next moment, and the prediction error between the predicted moment value and the actual moment value is calculated. Based on the current prediction error, the short-term abnormal deviation rate corresponding to each type of sensor data is calculated as the abnormality score. According to the abnormal scores of the combined various types of sensor data, a scoring data set is constructed, and event discrimination is performed through a lightweight classifier. If any abnormal score is higher than the preset abnormal threshold of the corresponding sensor, it is determined that the pipe network area under the jurisdiction of the edge terminal device is in a "critical" or "abnormal" state, and the timestamp, channel number and prediction error are uploaded to the central management module for secondary confirmation.

[0032] It should be noted that the specific calculation formula for the short-term abnormal deviation rate is as follows: Where, Represents short-term abnormal deviation; Represents the predicted Sensor value at the moment; Representing reality The sensor value at the moment; Represents the smallest constant that prevents division by zero.

[0033] The energy scheduling module monitors the energy status of each edge terminal device in real time and automatically switches to the working mode of each edge terminal device; the abnormality prediction module uses the collaborative processing mechanism of the edge and center ends to predict the abnormal development trend of the pipeline.

[0034] Specifically, the central management module regularly obtains prediction values, error information and various sensor data from each edge terminal device to build a collaborative time series matrix ,in represents the collaborative time series of the nth node, Represents a multi-factor observation point obtained by fusion calculation of predicted values, error information and various sensor data, n represents the node number, T represents the time length, and then the trend interpolation method is used to fill in the missing values in each time step. The abnormal prediction module sets different levels of sliding window lengths, extracts multi-scale sequences from the constructed collaborative time series matrix, calculates the local trend differential increment of each scale sequence, analyzes the trend change based on the calculated trend differential increment, and then fits the changing trend of different time steps through the set fixed-length sliding time window to generate a trend evolution curve, and then performs weighted fusion on the local trend differential increments of multiple scales to generate trend risk scores for different time steps. If the trend risk score is higher than the preset risk threshold, it is regarded as a trend anomaly, and the corresponding pipeline node area is marked as a potential risk segment. At the same time, according to different combination methods The nodes are combined in pairs to establish different node pairs. The trend risk sequences of the two groups of nodes in different node pairs in the past N time steps are extracted to construct multiple groups of N×N grid graphs. Each point in the grid graph represents the local Euclidean distance between the two nodes at different time steps. Starting from the starting point of each grid graph and moving to the end point, the cumulative minimum cost path is dynamically planned to generate the DTW distance of each node pair. The DTW distances of all node pairs are collected and a symmetric similarity matrix is generated. The generated symmetric similarity matrix is then density clustered to construct multiple node clusters with coordinated trends. The cluster center or the node where the trend changes first is used as the potential source point of anomaly propagation. The anomaly propagation graph is then constructed based on the time delay gradient to restore the risk diffusion path, and the risk values for multiple future time steps are output through the Transformer prediction model.

[0035] It should be further explained that the specific calculation formula for the local trend differential increment is as follows: Where, represent Time Node In the The trend increment on the scale; represents the The sliding window length of the scale; represent Time Node The scale sequence value of ; represent Time Node The scale sequence value of A random positive number.

[0036] The sedimentation assessment module calculates the sedimentation probability distribution and assesses the sedimentation risk of the pipeline network based on the three-dimensional pipeline network topology map and various sensor data.

[0037] Specifically, the sedimentation assessment module collects various sensor data such as flow, turbidity and pH value from each sensor node, and then time-aligns the collected sensor data. After that, based on the processed sensor data, the flow information entropy change rate, turbidity change gradient mean and pH value drift value of each pipe network area are calculated. The three sets of sedimentation assessment indicators are set based on historical training or expert experience. The weight coefficients of the three sets of calculation results of flow information entropy change rate, turbidity change gradient mean and pH value drift value are integrated, and the three sets of sedimentation assessment indicators are established to establish a comprehensive sedimentation risk cost function. Each pipe section in the pipe network is set as a set of binary variables. ,in Representative If there is siltation in the pipe section, it means there is no siltation in the pipe section. , L represents the total number of monitoring sections in the pipeline. The overall state vector is constructed based on the binary variables of each section. Then, based on the overall state vector, the comprehensive sedimentation risk cost function is converted into the QUBO cost function to establish the corresponding QUBO matrix. The QUBO matrix is input into the quantum annealing machine, and the annealing temperature, number of iterations and number of measurement samples are set. The binary variables of each section are initialized to the corresponding quantum superposition state. Then, the QUBO cost function is mapped to the constructed problem Hamiltonian to obtain the corresponding target Hamiltonian. Based on the initialized sets of quantum superposition states, the driving Hamiltonian corresponding to the current quantum annealing machine is initialized. By controlling the evolution of parameters over time, the initial driving Hamiltonian is transformed into the target Hamiltonian. Then, the driving Hamiltonian gradually approaches the ground state of the target Hamiltonian, and the corresponding cost function value is minimized. A "quantum measurement" is performed on this state to obtain a specific binary solution vector. The problem Hamiltonian construction, driving Hamiltonian initialization and driving Hamiltonian evolution are repeated until the preset number of iterations is reached. The occurrence of each section is counted. The number of times the siltation occurs is counted and the probability of occurrence is calculated. At the same time, based on the calculated siltation probability values, a pipe section probability vector is constructed and the “siltation risk probability map” of each pipe section is output.

[0038] The early warning visual module constructs a multi-level risk cloud map based on topological information, leakage point location, and sedimentation risk data, and displays early warnings and classifies risk levels for each pipeline network area; the central management module is used to coordinate the operating status of each module and provide a remote operation and maintenance interface and strategy optimization suggestions.

[0039] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A pipe network online monitoring system, characterized in that: include: Perception deployment module, signal acquisition module, signal processing module, pipe network modeling module, leak location module, edge computing module, energy scheduling module, anomaly prediction module, sedimentation assessment module, early warning visualization module, and central management module; The sensing deployment module is used to install low-power electromagnetic markers and various sensors at fixed points on the outer wall of the pipeline to collect pipeline network electromagnetic field data and various sensor data in real time; The signal acquisition module uses a mobile inspection robot to dynamically move and receive electromagnetic field data and various sensor data collected by the perception deployment module in real time; The signal processing module is used to pre-process the electromagnetic disturbance data and various sensor data collected by the signal acquisition module; The pipe network modeling module constructs a three-dimensional pipe network topology map based on the inspection robot's movement route and the processed electromagnetic disturbance data, and updates the connection relationship between pipeline nodes and branches in real time; The leakage location module evaluates the spatiotemporal correlation between sensor data based on the three-dimensional pipe network topology map and various sensor data, and performs double-blind location of the leakage point; The edge computing module is used to deploy edge terminal devices in the pipe well, receive various processed sensor data in real time, and perform predictive analysis on the edge sensor data; The energy scheduling module monitors the energy status of each edge terminal device in real time and automatically switches to the working mode of each edge terminal device; The anomaly prediction module uses the edge and center collaborative processing mechanism to predict the abnormal development trend of the pipeline; The siltation assessment module calculates the siltation probability distribution and assesses the siltation risk of the pipeline network based on the three-dimensional pipeline network topology map and various sensor data; The early warning visual module constructs a multi-level risk cloud map based on topological information, leakage point location and sedimentation risk data, and displays early warnings and classifies risk levels for each pipe network area; The central management module is used to coordinate the operating status of each module and provide a remote operation and maintenance interface and strategy optimization suggestions.

2. A pipe network online monitoring system according to claim 1, characterized in that: The specific steps of constructing a three-dimensional pipe network topology model by the pipe network modeling module are as follows: S1.1: The mobile inspection robot continuously collects electromagnetic field data while moving along the pipeline. Each set of collected electromagnetic field data is subjected to spatiotemporal normalization to eliminate deviations caused by different acquisition speeds and paths. The waveform gradient method is used to extract electromagnetic field mutation points, i.e., electromagnetic field distortion points caused by the presence of electromagnetic markers or structural changes. S1.2: Project each extracted electromagnetic field mutation point into the three-dimensional space coordinate system, and establish the spatial electromagnetic field mutation point set V={v1,v2,v3,...,v n }, where v n Represents the nth electromagnetic field mutation point, and constructs an undirected graph G = (V, E) based on the electromagnetic field mutation point set, where V represents each node in the undirected graph G, that is, each electromagnetic field mutation point, and E represents an edge to be evaluated between each pair of electromagnetic field mutation points at the beginning; S1.3: In an undirected graph consisting of electromagnetic field mutation points, multiple groups of virtual individuals are initialized. The starting node of each group of virtual individuals is randomly selected. Each virtual individual excludes the adjacent nodes that have been visited based on its current position and calculates the local attraction of the corresponding edges of the remaining adjacent nodes to form the current driving path set. S1.4: Calculate the pheromone concentrations of the edges between nodes in the undirected graph based on the three-dimensional Euclidean distances between each electromagnetic field mutation point and the correlation coefficients of the magnetic field perturbation curves. Then, calculate the probability of each virtual individual moving from the current node to each adjacent node based on the local attractiveness and pheromone concentration of each edge in the set of travel paths. Based on the selection probabilities of each adjacent node, determine the next movement destination using a roulette wheel method. S1.5: Each virtual individual jumps to a node in turn until no more nodes are available. Path construction ends, and the legitimacy of the paths constructed by each virtual individual is evaluated using a path legitimacy cost function. After each round of path construction, the quality of each path is evaluated. Based on the quality of the path traversed by the virtual individual, the pheromone concentration of each edge in the path is updated. At the same time, the local attractiveness of each edge is updated based on the historical jump frequency or perturbation matching degree. S1.6: If the undirected graph does not undergo structural changes after multiple rounds of iterations, the path is perturbed and the positions of each virtual individual are reset. The path is reconstructed, the pheromone concentration is updated, and the local attractiveness is updated again. After multiple rounds of iterations, the pheromone concentration change value of the path converges to the preset threshold, and the iteration is stopped. S1.7: Record the paths constructed by each virtual individual after the iteration, filter out the edges in each path that are below the preset pheromone threshold, and use the processed paths as a connected subgraph, that is, the global optimal path branch. Based on each connected subgraph, form the final topological structure, identify the three-dimensional coordinates of the electromagnetic field mutation points on each path, and then generate smooth pipe sections through the fitting algorithm to construct a complete three-dimensional pipe network topology map.

3. A pipe network online monitoring system according to claim 2, characterized in that: The specific calculation formula of the waveform gradient method described in S1.1 is as follows: Where, Represents the disturbance intensity of the kth sampling point, when If it is higher than the preset threshold, the current sampling point is judged to be an electromagnetic field mutation point; represents the electromagnetic intensity gradient at the kth sampling point; represents the electromagnetic intensity at the kth sampling point; Represents the spatial coordinate axis where the inspection robot is located; The specific calculation formula for the selection probability described in S1.4 is as follows: Where, Represents that at time t, the virtual individual from node Move to Node probability; Represents the current path Upper pheromone concentration; represents the heuristic factor of the path, and ; as well as represent the parameters controlling the pheromone concentration and the heuristic factor weight respectively; Representative Path Upper pheromone concentration; Representative Path The inspiration factor; Representative Node The set of candidate neighbors.

4. A pipe network online monitoring system according to claim 2, characterized in that: The specific steps of the leakage location module to evaluate the spatiotemporal correlation between the sensor data and perform double-blind location of the leakage point are as follows: S2.1: The leak location module extracts the pipeline vibration, infrasound, and temperature signals from the preprocessed sensor data. It samples each received set of physical signals within a fixed time window to synchronize the signal times and integrates the physical signals to construct a standardized sample set. S2.2: Based on the standardized sample set, set time series samples of each physical signal at each node location to construct a unified feature tensor. Then, based on the latest 3D pipeline network topology map, map each sensor sampling location to a pipeline topology node. Then, through discrete modeling, establish a signal distribution model for each type of signal at each node in different time windows. S2.3: Calculate the spatial distances between different nodes on each signal data channel. Based on the calculated spatial distances, obtain the joint signal difference scores for each node pair and construct an inter-node symmetric matrix. Normalize the inter-node symmetric matrix and construct an anomaly propagation structure diagram based on a preset threshold. Based on the anomaly propagation structure diagram, perform a comprehensive normalization construction on the signal mutation slope, peak amplitude, and temperature of each node to generate a comprehensive signal anomaly amplitude index corresponding to each node. S2.4: Calculate the anomaly intensity value of each node in the anomaly propagation diagram in the three-dimensional pipe network topology diagram based on the comprehensive anomaly amplitude index of the signal at each node. Arrange the anomaly intensity values in descending order and select the corresponding number of high-intensity nodes from high to low according to a preset ratio as the core anomaly candidate set. S2.5: A clustering algorithm based on connectivity constraints is used to divide the core anomaly candidate set into multiple anomaly clusters. Each anomaly cluster is considered to be a region with leakage propagation response. The inverse ratio of the anomaly intensity value is then used as the weight of the corresponding edge in the corresponding anomaly cluster to construct a core anomaly subgraph corresponding to each anomaly cluster. The Dijkstra algorithm is used to calculate all shortest paths between nodes in the core anomaly subgraph. S2.6: Based on all the shortest paths between the nodes in each core anomaly subgraph, the potential leakage points of the corresponding anomaly cluster are calculated. The identified potential leakage points are then mapped to the three-dimensional pipe network topology map. The upstream signal transmission path is then traced back based on the connection path. If the signal anomaly amplitude in the upstream path shows "radial enhancement", it is judged that there is a diffusion effect upstream. Combined with the pipe diameter and flow direction information, redundant potential leakage points are screened out and a potential leakage point set is output at the same time.

5. A pipe network online monitoring system according to claim 4, characterized in that: The specific calculation formula for the spatial distance mentioned in S2.3 is as follows: Where, Represents the spatial distance between node i and node j on signal channel m, where signal channel m includes the physical signal channels of pipeline vibration acceleration, infrasound signal amplitude, and temperature gradient sequence; Represents the set of all signal transmission matrices that meet the preset marginal constraints; represents the probability distribution of node i on signal channel m; represents the probability distribution of node j on signal channel m; represents the matching strategy for transmitting the quality from the u-th time sampling point of node i to the v-th time sampling point of node j; Represents the signal value of node i at the u-th time sampling point on signal channel m; Represents the signal value of node j at the uth time sampling point on signal channel m; p is used to control the distance measurement form. is the Manhattan distance, When is the Euclidean distance; The specific form of the abnormal propagation structure diagram described in S2.3 is as follows: Where, Indicates whether there is a connection between node i and node j in the abnormal propagation structure graph. 1 represents the existence of a connection, and 0 represents the absence of a connection. Node i and node j are consistent with the node i and node j in the spatial distance calculation formula. M represents the symmetric matrix between nodes; Represents the abnormality threshold.

6. The pipe network online monitoring system according to claim 1, characterized in that: The specific steps of the edge computing module for predictive analysis of edge sensor data are as follows: S3.1: Each deployed edge terminal device receives sensor data on flow rate, pressure, vibration, electromagnetic field intensity, temperature, and turbidity collected by sensors within its respective pipe network area. It then divides the sensor data into multiple sensor data sets of consistent length, based on a sliding time window preset by the edge terminal device. S3.2: Normalize each sensor dataset using the local mean difference method, set a soft threshold, and filter out high-frequency random disturbances using Daubechies wavelet decomposition. Then, load the lightweight LSTM network model on each edge device and initialize the hidden and memory states of the lightweight LSTM network model to 0 or the previous historical state. S3.3: Each sensor dataset is fed into the lightweight LSTM network model in order from oldest to newest according to the corresponding time. The lightweight LSTM network model forward propagates each sensor dataset received, processes each sensor data in each sensor dataset layer by layer, and updates the hidden state corresponding to each time step until all sensor datasets at the last time step are processed. S3.4: Input the hidden state of the last time step into the linear regression layer to predict the value at the next moment. Calculate the prediction error between the predicted value and the actual value at the next moment. Based on the current prediction error, calculate the short-term anomaly deviation rate corresponding to each type of sensor data as the anomaly score. The specific calculation formula for the short-term anomaly deviation rate is as follows: Where, Represents short-term abnormal deviation; Represents the predicted Sensor value at the moment; Representing reality The sensor value at the moment; Represents the minimum constant that prevents division by zero; S3.5: Based on the anomaly scores of the combined sensor data, a scoring dataset is constructed and event identification is performed using a lightweight classifier. If any anomaly score is higher than the preset anomaly threshold of the corresponding sensor, the pipe network area under the jurisdiction of the edge terminal device is determined to be in a "critical" or "abnormal" state, and the timestamp, channel number, and prediction error are uploaded to the central management module for secondary confirmation.

7. A pipe network online monitoring system according to claim 6, characterized in that: The specific steps of the abnormal prediction module to predict the abnormal development trend of the pipeline are as follows: S4.1: The central management module regularly obtains prediction values, error information and various sensor data from each edge terminal device to build a collaborative time series matrix ,in represents the collaborative time series of the nth node, represents the multi-factor observation point obtained by fusion calculation of predicted value, error information and various sensor data, n represents the node number, T represents the time length, and then the trend interpolation method is used to fill the missing values in each time step; S4.2: The anomaly prediction module sets sliding window lengths at different levels, extracts multi-scale sequences from the constructed collaborative time series matrix, calculates the local trend differential increments for each scale sequence, analyzes trend changes based on the calculated trend differential increments, and then fits the changing trends at different time steps through a fixed-length sliding time window to generate a trend evolution curve. The local trend differential increments at multiple scales are then weighted and fused to generate trend risk scores at different time steps. S4.3: If the trend risk score is higher than the preset risk threshold, it is considered a trend anomaly and the corresponding network node area is marked as a potential risk segment. At the same time, the nodes are combined in pairs according to different combinations to establish different node pairs. The trend risk series of the two groups of nodes in different node pairs over the past N time steps are extracted to construct multiple groups of N×N grid graphs. Each point in the grid graph represents the local Euclidean distance between the two nodes at different time steps. S4.4: Start from the starting point of each grid graph and move toward the end point, and dynamically plan the cumulative minimum cost path to generate the DTW distance of each node pair. Collect the DTW distances of all node pairs and generate a symmetric similarity matrix. Then, perform density clustering on the generated symmetric similarity matrix to construct multiple trend-coordinated node clusters. The cluster center or the node where the trend changes first is used as the potential source of anomaly propagation. Then, construct an anomaly propagation graph based on the time delay gradient to restore the risk diffusion path, and output the risk value for multiple time steps in the future through the Transformer prediction model.

8. The pipe network online monitoring system according to claim 7, characterized in that: The specific calculation formula for the local trend differential increment mentioned in S4.2 is as follows: Where, represent Time Node In the The trend increment on the scale; represents the The sliding window length of the scale; represent Time Node The scale sequence value of ; represent Time Node The scale sequence value of A random positive number.

Images