Urban Pipeline Network Leakage Detection and Diagnosis System

Abstract

The present invention relates to the technical field of pipeline leakage detection, specifically to an urban pipeline leakage detection and diagnosis system, including a pipeline modeling unit, a status monitoring unit, a pipe area division unit, a first leakage detection unit, a dynamic display unit, and a second leakage detection unit. Specifically: the pipeline modeling unit is used to perform topological modeling on the pipeline network in the target area to obtain a pipeline topology map; the status monitoring unit is used to perform status monitoring on the pipeline network in the target area according to a preset sampling time based on a pre-set water pressure sensor and a vibration sensor. Based on the collected weighted pipeline topology map sequence, the present invention uses a clustering algorithm to divide the pipeline network and generate sub-pipeline networks of different modes. By classifying pipeline networks of different modes, the pipeline network status of different areas or sub-networks can be analyzed in a targeted manner, and the division of the pipeline network helps to perform leak detection more efficiently, and can give priority to areas that are most likely to have water leakage problems.

Description

Technical Field

[0001] The invention relates to the technical field of pipe network leakage detection, and in particular to a city pipe network leakage detection and diagnosis system. Background Art

[0002] Urban pipeline leak detection refers to the process of detecting and diagnosing water (or leakage) in underground pipeline systems such as urban water supply, drainage, gas, and heating. The purpose of this process is to promptly discover and locate leaks or anomalies of water, gas, heat and other substances in the pipeline network so as to take measures to repair or prevent them, reduce resource waste, pipeline damage, environmental pollution and other problems, and ensure the normal operation of the urban pipeline network.

[0003] Traditional systems fail to effectively model the topology of the pipe network, and usually rely on simple pipe network diagrams or manually maintained data. In this case, the structure of the pipe network may not fully match the actual state, and it is impossible to accurately describe the connection relationship and node information of the pipe network, which leads to an inability to fully understand the overall picture of the pipe network during leak detection, thus affecting the accuracy of diagnosis; and traditional systems cannot monitor the dynamic state of the pipe network in real time, and usually rely on regular manual inspections or breakpoint monitoring. This method may miss some potential signs of leaks, especially in the early stages of leaks or in difficult-to-detect pipe locations, and the system is difficult to respond quickly; and the analysis of the pipe network in traditional systems is usually simple, making it difficult to perform detailed regional divisions. Each part of the pipe network may be treated equally, and leaks or abnormal problems often occur in specific areas or patterns. Due to the lack of reasonable regional divisions, traditional systems may waste a lot of resources in areas that do not require key monitoring, reducing leak detection efficiency. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the disadvantages of the above-mentioned prior art and provide a leak detection and diagnosis system for urban pipe networks.

[0005] The technical solution adopted to solve the above technical problems is: an urban pipe network leak detection and diagnosis system, including a pipe network modeling unit, a state monitoring unit, a pipe area division unit, a first leak detection unit, a dynamic display unit and a second leak detection unit, specifically:

[0006] The pipeline network modeling unit is used to perform topological modeling on the pipeline network in the target area to obtain a pipeline topology map, wherein the pipeline topology map includes a node set and a connection edge set, wherein the nodes in the node set correspond to the pipeline nodes of the pipeline network in the target area, wherein the pipeline nodes correspond to water users, and the connection edges in the connection edge set correspond to the pipelines between the water users in the target area, and the pipeline topology map is transmitted to the status monitoring unit;

[0007] The status monitoring unit is used to receive the pipeline topology map transmitted by the pipeline network modeling unit, and perform status monitoring on the pipeline network in the target area according to the preset sampling time based on the pre-set water pressure sensor and vibration sensor to obtain the status data sequence corresponding to each pipeline node in the pipeline network, wherein the status data sequence includes a water pressure signal sequence and a vibration signal sequence, and the weight of each pipeline node in the pipeline topology map is assigned based on the status data sequence corresponding to each pipeline node in the pipeline network to obtain a weighted pipeline topology map sequence, and the weighted pipeline topology map sequence is transmitted to the pipe area division unit and the dynamic display unit.

[0008] Preferably, the pipe area division unit is used to receive the weighted pipeline topology map sequence transmitted by the status monitoring unit, and perform pipe area clustering on the weighted pipeline topology map sequence based on a clustering algorithm to divide the pipeline network within the target area into sub-pipeline networks of different modes, and transmit the sub-pipeline networks of different modes to the first leak detection unit.

[0009] Preferably, the first leak detection unit is used to receive sub-pipeline networks of different modes transmitted by the pipe area division unit, and compare the weights of the nodes of the sub-pipeline networks of different modes with the preset weight thresholds of the sub-pipeline networks of different modes. If the weights of the nodes of the sub-pipeline networks of different modes are greater than the preset weight threshold, node abnormality information is generated and the node abnormality information is transmitted to the dynamic display unit and the second detection unit.

[0010] Preferably, the second leak detection unit is used to receive the node abnormality information transmitted by the first leak detection unit, and obtain the sub-pipeline networks of different modes corresponding to the node abnormality information, match the sub-pipeline networks of different modes with a preset pattern model table to obtain the leak detection models corresponding to the sub-pipeline networks of different modes, perform leak detection on the sub-pipeline networks of different modes based on the pre-trained leak detection models to obtain the leak detection labels corresponding to the sub-pipeline networks of different modes, and transmit the leak detection labels corresponding to the sub-pipeline networks of different modes to the dynamic display unit.

[0011] Preferably, the dynamic display unit is used to receive the weighted pipeline topology map sequence transmitted by the status monitoring unit, the node abnormality information transmitted by the first leak detection unit, and the leak detection labels corresponding to the sub-pipeline networks of different modes transmitted by the second leak detection unit, and display the weighted pipeline topology map sequence in time series, display the node abnormality information in the corresponding weighted pipeline topology map, and mark the sub-pipeline networks in the weighted pipeline topology map based on the leak detection labels corresponding to the sub-pipeline networks of different modes.

[0012] Preferably, the weighted pipeline topology map sequence is clustered by pipeline area based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, including:

[0013] The weights of the nodes in the weighted pipeline topology sequence are used as the input variable matrix, wherein the expression of the input variable matrix is ​​as follows:

[0014]

[0015] Where D represents the input variable matrix, represents the feature vector of the nth node in the weighted pipeline topology sequence, x nm Represents the mth feature of the nth node in the weighted pipeline topology sequence;

[0016] The cluster center points are calculated based on the input variable matrix, wherein the calculation formula of the cluster center points is as follows:

[0017]

[0018] in, represents the center point of the lth iteration of the i-th mode, α represents the power index, Represents the membership degree of the i-th mode at the l-th iteration.

[0019] Preferably, the weighted pipeline topology map sequence is clustered by pipeline area based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, and further includes:

[0020] The membership is updated based on the cluster center point to obtain an updated membership, wherein the calculation formula of the updated membership is as follows:

[0021]

[0022] Where c represents the number of different modes, Represents the distance from node j to the center of cluster i in the weighted pipeline topology sequence;

[0023] It is determined whether the error between the current iteration membership and the previous iteration is less than a preset error threshold, or whether the maximum number of iterations has been reached. If so, the iteration is stopped to divide the pipeline network in the target area into sub-pipeline networks of different modes.

[0024] Preferably, the leak detection model includes a pressure leak detection module, a vibration leak detection module and a label mapping module, wherein the pressure leak detection module adopts a support vector machine, and obtains the pressure leak detection label corresponding to the sub-pipeline network based on the pressure signal of the node in the sub-pipeline network, the vibration leak detection module adopts a graph neural network, and obtains the vibration leak detection label corresponding to the sub-pipeline network based on the vibration signal of the node in the sub-pipeline network, and the label mapping module adopts a neural network, and performs label mapping based on the pressure leak detection label and vibration leak detection label corresponding to the sub-pipeline network to obtain the leak detection label corresponding to the sub-pipeline network.

[0025] Preferably, obtaining the pressure leak detection label corresponding to the sub-pipeline network based on the pressure signal of the node in the sub-pipeline network includes:

[0026] Performing multiple random walks on each node in the sub-pipeline network, each walk starts from a random node and selects a neighboring node to continue walking until the number of walk steps reaches a predetermined length, so as to obtain a walk path sequence;

[0027] By training the walking path sequence, an embedding vector of each node is learned, wherein the embedding vector includes a hidden feature vector, wherein the hidden layer feature represents the position and semantics of the node in the sub-pipeline network;

[0028] Based on the embedding vector, elevation and pressure of each node in the sub-pipeline network as a feature vector;

[0029] The feature vector is classified based on a pre-trained support vector machine model to obtain a pressure leak detection label corresponding to the sub-pipeline network.

[0030] Preferably, obtaining a vibration leak detection tag corresponding to the sub-pipeline network based on a vibration signal of a node in the sub-pipeline network includes:

[0031] The vibration signal is divided based on sliding window sampling to obtain a sub-vibration signal sequence, wherein the expression of the sliding window sampling is as follows:

[0032]

[0033] Where n represents the number of sub-vibration signals after sliding window sampling, floor represents the rounding function, L represents the length of the vibration signal, d represents the sliding window length, and s represents the step size of each sliding window;

[0034] The sub-vibration signal sequence is converted into a corresponding frequency spectrum based on fast Fourier transform, wherein the expression of the fast Fourier transform is as follows:

[0035]

[0036] in, represents the spectrum of the ith sub-vibration signal of the node in the sub-pipeline network, FFT represents the fast Fourier transform function, Represents the frequency spectrum of the vibration signal, y n Indicates the fault type. Check whether the nth sub-vibration signal is greater than the preset spectrum threshold. If it is greater than the preset spectrum threshold, then y n =1, otherwise, y n =0;

[0037] The frequency domain of the sub-vibration signal is taken as the node, the spectrum mean is taken as the node value, a visual graph is constructed, the spectrum is taken as the node feature of the visual graph, and the fault type is taken as the node label of the visual graph.

[0038] Preferably, obtaining the vibration leak detection tag corresponding to the sub-pipeline network based on the vibration signal of the node in the sub-pipeline network also includes:

[0039] The attention coefficient between each node is calculated based on the node features of the visible graph, wherein the attention coefficient is the edge weight of the visible graph, and the calculation formula of the attention coefficient is as follows:

[0040]

[0041] Among them, α ij represents the weight of the edge connecting the i-th node and the j-th node in the visible graph, ReLU represents the activation function, a represents the learnable parameter vector, and W represents the graph filter parameter matrix;

[0042] The embedding features between the nodes are calculated based on the attention coefficients and node features between the nodes in the visibility graph, wherein the calculation formula of the embedding features is as follows:

[0043]

[0044] Among them, z i Represents the embedded features of the i-th node in the visibility graph;

[0045] The embedded features between each node in the visible graph are passed through the convolutional layer of the embedded self-attention mechanism and the stacked graph attention to obtain the aggregated features;

[0046] The aggregated features are classified based on the fully connected layer and softmax to obtain the vibration leak detection label corresponding to each node in the sub-pipeline network.

[0047] Preferably, label mapping is performed based on the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network to obtain the leak detection label corresponding to the sub-pipeline network, including:

[0048] Based on a preset label code set, encoding and assigning values ​​to the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network to obtain a label code sequence;

[0049] The label coding sequence is matched with a preset label sequence table to obtain a leak detection label corresponding to the sub-pipeline network.

[0050] The beneficial effects of the present invention are as follows: (1) The present invention generates a detailed pipe network topology map by topological modeling of the pipe network in the target area. This topology map includes pipe network nodes (related to water users) and connection edges (information on the interconnection of pipes). Through this modeling, the system can accurately understand the structure of the entire pipe network and provide basic data for subsequent water leakage detection and diagnosis. In addition, through water pressure and vibration sensors, the system can collect real-time status data of the pipe network, including water pressure and vibration signal sequences. These data can reflect the health status of the pipe network, especially in the early stages of abnormality detection. (2) The present invention divides the pipe network into sub-pipeline networks of different modes by using a clustering algorithm based on the collected weighted pipe topology map sequence. By classifying pipe networks of different modes, the pipe network status of different areas or sub-networks can be analyzed in a targeted manner, and the division of the pipe network helps to perform leak detection more efficiently. , because the system can give priority to areas most likely to have water leakage problems, thereby improving detection efficiency; (3) The present invention can identify abnormal nodes and generate abnormal information by comparing the node weights with the preset threshold through the first leak detection unit. This method can timely identify nodes that may have water leakage, thereby greatly improving the accuracy and response speed of water leakage detection, and the second leak detection unit matches the node abnormal information with the preset pattern model table to generate specific leak detection labels according to different pipe network patterns. This pattern matching-based leak detection method can accurately identify the specific location and type of water leakage, and the weighted pipeline topology map, node abnormal information and leak detection labels can be visualized through the dynamic display unit. This dynamic display can not only show the operating status of the pipe network, but also display abnormal information and leak detection results in real time, helping staff to quickly locate problems and take measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0052] Figure numerals: 1. Pipe network modeling unit; 2. Status monitoring unit; 3. Pipe area division unit; 4. First leak detection unit; 5. Dynamic display unit; 6. Second leak detection unit. DETAILED DESCRIPTION

[0053] Embodiment 1, as Figure 1 As shown, the urban pipe network leak detection and diagnosis system proposed by the present invention includes a pipe network modeling unit 1, a state monitoring unit 2, a pipe area division unit 3, a first leak detection unit 4, a dynamic display unit 5 and a second leak detection unit 6. Specifically:

[0054] The pipeline network modeling unit 1 is used to perform topological modeling on the pipeline network in the target area to obtain a pipeline topology map, wherein the pipeline topology map includes a node set and a connection edge set, wherein the nodes in the node set correspond to the pipeline nodes of the pipeline network in the target area, wherein the pipeline nodes correspond to the water users, and the connection edges in the connection edge set correspond to the pipelines between the water users in the target area, and the pipeline topology map is transmitted to the state monitoring unit 2;

[0055] The status monitoring unit 2 is used to receive the pipeline topology map transmitted by the pipeline network modeling unit 1, and perform status monitoring on the pipeline network in the target area according to the preset sampling time based on the pre-set water pressure sensor and vibration sensor to obtain a status data sequence corresponding to each pipeline node in the pipeline network, wherein the status data sequence includes a water pressure signal sequence and a vibration signal sequence, and the weight of each pipeline node in the pipeline topology map is assigned based on the status data sequence corresponding to each pipeline node in the pipeline network to obtain a weighted pipeline topology map sequence, and the weighted pipeline topology map sequence is transmitted to the pipe area division unit 3 and the dynamic display unit 5.

[0056] In the present invention, pipeline topology: This is a graphical representation method used to represent the nodes in the pipeline network and the connections between them. It consists of two main parts: Node set: Each node represents a location or pipeline interface in the pipeline network. Usually, these nodes correspond to the "pipeline nodes" in the actual pipeline network. These nodes generally represent actual facilities or access points, such as water pumping stations, water diversion points, valves, etc. Connection edge set: The connection edges represent the pipeline connections between the nodes in the pipeline network. They represent physical pipelines, connecting different water pipes and water pumping stations, etc.; The water pressure sensor is used to measure the water pressure in the pipeline. The water pressure is a key indicator of the pipeline operation status, which can reflect the strength of the water flow, whether the pipeline is blocked or damaged, etc.; The vibration sensor is used to detect the vibration of the pipeline or equipment. The vibration signal can reflect the health of the pipeline, such as whether there is leakage, rupture, etc.

[0057] In an optional embodiment, the pipeline area division unit 3 is used to receive the weighted pipeline topology map sequence transmitted by the status monitoring unit 2, and perform pipeline area clustering on the weighted pipeline topology map sequence based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, and transmit the sub-pipeline networks of different modes to the first leak detection unit 4.

[0058] It should be noted that the clustering algorithm is an unsupervised learning algorithm used to divide the data set into several groups (clusters). The data objects within each cluster have a high similarity, while the objects between different clusters have a low similarity. The purpose of the clustering algorithm is to assign samples or objects to different categories based on the characteristics of the data, so that objects in the same category have similar attributes, while objects in different categories are quite different; pipe area clustering refers to the use of a clustering algorithm to divide the pipe network area into multiple sub-pipeline networks. These sub-pipeline networks (or pipe areas) are aggregated based on the topological structure and status data (such as water pressure, vibration, etc.) of the pipe network. Each pipe area represents a relatively independent area, which usually has different operating modes in pipe network management; mode refers to the operating status or behavior mode of each sub-pipeline network in the pipe network system.

[0059] In an optional embodiment, the first leak detection unit 4 is used to receive sub-pipeline networks of different modes transmitted by the pipe area division unit 3, and compare the weights of the nodes of the sub-pipeline networks of different modes with the preset weight thresholds of the sub-pipeline networks of different modes. If the weights of the nodes of the sub-pipeline networks of different modes are greater than the preset weight threshold, node abnormality information is generated and transmitted to the dynamic display unit 5 and the second detection unit 6.

[0060] It should be noted that node abnormal information refers to the abnormal state recorded by the system when it is found that the weight of a node is greater than a preset threshold during the comparison process.

[0061] In an optional embodiment, the second leak detection unit 6 is used to receive the node abnormality information transmitted by the first leak detection unit 4, and obtain the sub-pipeline networks of different modes corresponding to the node abnormality information, match the sub-pipeline networks of different modes with the preset pattern model table to obtain the leak detection models corresponding to the sub-pipeline networks of different modes, perform leak detection on the sub-pipeline networks of different modes based on the pre-trained leak detection models to obtain the leak detection labels corresponding to the sub-pipeline networks of different modes, and transmit the leak detection labels corresponding to the sub-pipeline networks of different modes to the dynamic display unit 5.

[0062] It should be noted that the preset mode model table is usually established based on historical data, engineering experience or simulation results, and records the leak detection model of the pipeline network under different modes; based on the matching results, the system can select a leak detection model suitable for the mode. The leak detection model is trained through machine learning, statistical analysis and other technologies, and is used to detect possible leakage problems in the pipeline network.

[0063] In an optional embodiment, the dynamic display unit 5 is used to receive the weighted pipeline topology map sequence transmitted by the status monitoring unit 2, the node abnormality information transmitted by the first leak detection unit 4, and the leak detection labels corresponding to the sub-pipeline networks of different modes transmitted by the second leak detection unit 6, and display the weighted pipeline topology map sequence in time series, display the node abnormality information in the corresponding weighted pipeline topology map, and mark the sub-pipeline networks in the weighted pipeline topology map based on the leak detection labels corresponding to the sub-pipeline networks of different modes.

[0064] Embodiment 2, the urban pipe network leak detection and diagnosis system proposed by the present invention, compared with embodiment 1, this embodiment also includes: clustering the weighted pipe topology map sequence based on the clustering algorithm to divide the pipe network in the target area into sub-pipeline networks of different modes, including:

[0065] The weights of the nodes in the weighted pipeline topology sequence are used as the input variable matrix, where the expression of the input variable matrix is ​​as follows:

[0066]

[0067] Where D represents the input variable matrix, represents the feature vector of the nth node in the weighted pipeline topology sequence, x nm Represents the mth feature of the nth node in the weighted pipeline topology sequence;

[0068] The cluster center points are calculated based on the input variable matrix, where the calculation formula of the cluster center points is as follows:

[0069]

[0070] in, represents the center point of the lth iteration of the i-th mode, α represents the power index, Represents the membership degree of the i-th mode at the l-th iteration.

[0071] In an optional embodiment, the weighted pipeline topology graph sequence is clustered into pipeline areas based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, and further includes:

[0072] The membership is updated based on the cluster center to obtain the updated membership, where the calculation formula for the updated membership is as follows:

[0073]

[0074] Where C represents the number of different modes, Represents the distance from node j to the center of cluster i in the weighted pipeline topology sequence;

[0075] It is determined whether the error between the current iteration membership and the previous iteration is less than the preset error threshold, or whether the maximum number of iterations has been reached. If it is less than the preset error threshold, or the maximum number of iterations has been reached, the iteration is stopped to divide the pipeline network in the target area into sub-pipeline networks of different modes.

[0076] In an optional embodiment, the leak detection model includes a pressure leak detection module, a vibration leak detection module and a label mapping module, wherein the pressure leak detection module adopts a support vector machine, and obtains the pressure leak detection label corresponding to the sub-pipeline network based on the pressure signal of the node in the sub-pipeline network, the vibration leak detection module adopts a graph neural network, and obtains the vibration leak detection label corresponding to the sub-pipeline network based on the vibration signal of the node in the sub-pipeline network, and the label mapping module adopts a neural network, and performs label mapping based on the pressure leak detection label and vibration leak detection label corresponding to the sub-pipeline network to obtain the leak detection label corresponding to the sub-pipeline network.

[0077] In an optional embodiment, obtaining a pressure leak detection tag corresponding to a sub-pipeline network based on a pressure signal of a node in the sub-pipeline network includes:

[0078] Perform multiple random walks on each node in the sub-pipeline network. Each walk starts from a random node and selects a neighboring node to continue walking until the number of walk steps reaches a predetermined length to obtain a walk path sequence.

[0079] By training the walking path sequence, the embedding vector of each node is learned, wherein the embedding vector includes a hidden feature vector, wherein the hidden layer feature represents the position and semantics of the node in the sub-pipeline network;

[0080] Based on the embedding vector, elevation and pressure of each node in the sub-pipeline network as feature vector;

[0081] The feature vectors are classified based on the pre-trained support vector machine model to obtain the pressure leak detection labels corresponding to the sub-pipeline network.

[0082] It should be noted that random walk is a commonly used method in graph data analysis, which can simulate the relationship and mutual influence between nodes. In the sub-pipeline network, nodes represent certain locations in the pipeline network (such as pipeline intersections, valves, sensors, etc.), and edges represent the connections between these nodes (such as the connection relationship between pipelines); through multiple random walks, multiple path sequences are generated, each path sequence reflects the relationship between a node and its neighbors, and reveals the structure and position characteristics of the node in the network; by training the walk path sequence, the embedding vector of each node is learned. The embedding vector is a way to convert the high-dimensional information of the node (such as structural and semantic features) into a low-dimensional vector representation, so that the position of each node in the vector space can capture the relationship and similarity between the nodes; training of embedding vectors: This training is usually done through graphs. Embedding algorithm implementation, common algorithms include DeepWalk, Node2Vec, etc. These algorithms generate an embedding vector for each node by learning the path sequence, with the goal of making similar nodes closer in the embedding space. During the training process, the network topology (the connection relationship between nodes) and the high-order relationship of the nodes (captured by random walks) will be encoded into the embedding vector; elevation refers to the geographical location information of the node, usually expressed as the elevation (altitude, etc.) of the node location. In pipe network analysis, elevation may affect the pressure distribution of water flow; support vector machine (SVM) is a commonly used supervised learning classification algorithm that can effectively handle high-dimensional data and nonlinear classification problems. The task of the SVM model is to classify based on the feature vector of each node, so as to determine whether the node has abnormal pressure or leakage problems.

[0083] In an optional embodiment, obtaining a vibration leak detection tag corresponding to a sub-pipeline network based on a vibration signal of a node in the sub-pipeline network includes:

[0084] The vibration signal is divided based on sliding window sampling to obtain a sub-vibration signal sequence, where the expression of sliding window sampling is as follows:

[0085]

[0086] Where n represents the number of sub-vibration signals after sliding window sampling, floor represents the rounding function, L represents the length of the vibration signal, d represents the sliding window length, and s represents the step size of each sliding window;

[0087] The sub-vibration signal sequence is converted into the corresponding spectrum based on the fast Fourier transform, where the expression of the fast Fourier transform is as follows:

[0088]

[0089] in, represents the spectrum of the ith sub-vibration signal of the node in the sub-pipeline network, FFT represents the fast Fourier transform function, Represents the frequency spectrum of the vibration signal, y n Indicates the fault type. Check whether the nth sub-vibration signal is greater than the preset spectrum threshold. If it is greater than the preset spectrum threshold, then y n =1, otherwise, y n =0;

[0090] The frequency domain of the sub-vibration signal is taken as the node, the spectrum mean is taken as the node value, a visual graph is constructed, the spectrum is taken as the node feature of the visual graph, and the fault type is taken as the node label of the visual graph.

[0091] It should be noted that vibration signals are usually generated by the physical movement of pipelines in the pipeline system. Abnormal vibration (such as increase or frequency change) may indicate problems in the pipeline network, such as leakage, cracks or other faults. The vibration leak detection label processes and analyzes these vibration signals to determine whether there is a problem with the pipeline and label it accordingly (such as "normal" or "abnormal"); sliding window sampling is a common method in time series analysis, which is used to extract local information from a continuous data stream. The purpose of sliding window sampling is to divide the vibration signal into shorter sub-signals (windows) with a fixed length in order to more accurately analyze the local characteristics of the signal; Fast Fourier Transform (FFT) is an efficient algorithm used to calculate the Fourier transform of a signal, converting the signal from the time domain to the frequency domain. In the frequency domain, the frequency components and their intensities of the signal can be analyzed through the spectrum to identify potential fault modes.

[0092] In an optional embodiment, obtaining a vibration leak detection tag corresponding to a sub-pipeline network based on a vibration signal of a node in the sub-pipeline network further includes:

[0093] The attention coefficient between each node is calculated based on the node features of the visible graph, where the attention coefficient is the edge weight of the visible graph. The calculation formula of the attention coefficient is as follows:

[0094]

[0095] Among them, α ij represents the weight of the edge connecting the i-th node and the j-th node in the visible graph, ReLU represents the activation function, a represents the learnable parameter vector, and W represents the graph filter parameter matrix;

[0096] The embedding features between each node are calculated based on the attention coefficient and node features between each node in the visibility graph, where the calculation formula of the embedding feature is as follows:

[0097]

[0098] Among them, z i Represents the embedded features of the i-th node in the visibility graph;

[0099] The embedded features between each node in the visible graph are passed through the convolutional layer of the embedded self-attention mechanism and the stacked graph attention to obtain the aggregated features;

[0100] The aggregated features are classified based on the fully connected layer and softmax to obtain the vibration leak detection label corresponding to each node in the sub-pipeline network.

[0101] It should be noted that the attention coefficient is the edge weight calculated by an attention mechanism, which is used to quantify the importance between different nodes in the graph; Graph filter parameter matrix: Graph filter is an operator or matrix used to represent graph structure information in graph neural networks, which helps the model propagate information in the graph structure; Stacked graph attention convolution layer refers to stacking multiple graph attention layers in sequence, and enhancing the expressiveness of the model through multi-layer processing; The fully connected layer (also called the linear layer) is a basic neural network layer, each input node is connected to each output node, weighted summation is performed and output is obtained through the activation function. In graph neural networks, it is used to map node features to different spaces, such as mapping node embedding features to classification label space.

[0102] In an optional embodiment, label mapping is performed based on the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network to obtain the leak detection label corresponding to the sub-pipeline network, including:

[0103] Based on a preset label coding set, the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network are coded and assigned values ​​to obtain a label coding sequence;

[0104] The label coding sequence is matched with a preset label sequence table to obtain the leak detection label corresponding to the sub-pipeline network.

[0105] It should be noted that the label encoding set refers to a pre-set encoding rule or set, which defines a specific encoding for each possible label. These encodings are usually in the form of numbers or vectors to simplify processing and calculation. The purpose of the label encoding set is to convert discrete category labels (such as "normal" or "abnormal") into numerical forms that computers can understand and operate; the label encoding sequence is a sequence obtained by converting the original discrete label (such as "normal" or "abnormal") into a series of numbers or vectors according to the label encoding set. For example, if the vibration and pressure detection results of a pipeline are "normal" and "abnormal" respectively, then their corresponding codes in the label encoding set may be "0" and "1", which forms a label encoding sequence, such as [0,1]; the label sequence table is a preset label set, which is usually used to encode all possible label combinations. For example, if there are vibration and pressure detection labels at multiple locations, they may be combined into a sequence according to the conditions of all locations, and a unique code is assigned to each combination. This sequence contains the codes of all possible label combinations.

[0106] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. Urban pipe network leak detection and diagnosis system, characterized in that: The system comprises a pipe network modeling unit (1), a state monitoring unit (2), a pipe area division unit (3), a first leak detection unit (4), a dynamic display unit (5) and a second leak detection unit (6), specifically: The pipeline network modeling unit (1) is used to perform topological modeling on the pipeline network in the target area to obtain a pipeline topology map, wherein the pipeline topology map includes a node set and a connection edge set, wherein the nodes in the node set correspond to pipeline nodes of the pipeline network in the target area, wherein the pipeline nodes correspond to water users, and the connection edges in the connection edge set correspond to pipelines between various water users in the target area, and the pipeline topology map is transmitted to the state monitoring unit (2); The state monitoring unit (2) is used to receive the pipeline topology map transmitted by the pipeline network modeling unit (1), and perform state monitoring on the pipeline network in the target area according to a preset sampling time based on a pre-set water pressure sensor and a vibration sensor, so as to obtain a state data sequence corresponding to each pipeline node in the pipeline network, wherein the state data sequence includes a water pressure signal sequence and a vibration signal sequence, and based on the state data sequence corresponding to each pipeline node in the pipeline network, a weight is assigned to each pipeline node in the pipeline topology map to obtain a weighted pipeline topology map sequence, and the weighted pipeline topology map sequence is transmitted to the pipe area division unit (3) and the dynamic display unit (5).

2. The urban pipe network leak detection and diagnosis system according to claim 1 is characterized in that: The pipeline area division unit (3) is used to receive the weighted pipeline topology map sequence transmitted by the state monitoring unit (2), and perform pipeline area clustering on the weighted pipeline topology map sequence based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, and transmit the sub-pipeline networks of different modes to the first leak detection unit (4); The first leak detection unit (4) is used to receive the sub-pipeline networks of different modes transmitted by the pipe area division unit (3), and compare the weights of the nodes of the sub-pipeline networks of different modes with the preset weight thresholds of the sub-pipeline networks of different modes; if the weights of the nodes of the sub-pipeline networks of different modes are greater than the preset weight threshold, node abnormality information is generated, and the node abnormality information is transmitted to the dynamic display unit (5) and the second detection unit (6).

3. The urban pipe network leak detection and diagnosis system according to claim 2 is characterized in that: The second leak detection unit (6) is used to receive the node abnormality information transmitted by the first leak detection unit (4), and obtain the sub-pipeline networks of different modes corresponding to the node abnormality information, match the sub-pipeline networks of different modes with a preset mode model table to obtain the leak detection models corresponding to the sub-pipeline networks of different modes, perform leak detection on the sub-pipeline networks of different modes based on the pre-trained leak detection models to obtain the leak detection labels corresponding to the sub-pipeline networks of different modes, and transmit the leak detection labels corresponding to the sub-pipeline networks of different modes to the dynamic display unit (5); The dynamic display unit (5) is used to receive the weighted pipeline topology map sequence transmitted by the status monitoring unit (2), the node abnormality information transmitted by the first leak detection unit (4), and the leak detection labels corresponding to the sub-pipeline networks of different modes transmitted by the second leak detection unit (6), and display the weighted pipeline topology map sequence in time series, display the node abnormality information in the corresponding weighted pipeline topology map, and mark the sub-pipeline networks in the weighted pipeline topology map based on the leak detection labels corresponding to the sub-pipeline networks of different modes.

4. The urban pipe network leak detection and diagnosis system according to claim 2, characterized in that: The weighted pipeline topology sequence is clustered based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, including: The weights of the nodes in the weighted pipeline topology sequence are used as the input variable matrix, wherein the expression of the input variable matrix is ​​as follows: Where D represents the input variable matrix, represents the feature vector of the nth node in the weighted pipeline topology sequence, x nm Represents the mth feature of the nth node in the weighted pipeline topology sequence; The cluster center points are calculated based on the input variable matrix, wherein the calculation formula of the cluster center points is as follows: in, represents the center point of the lth iteration of the i-th mode, α represents the power index, Represents the membership degree of the i-th mode at the l-th iteration.

5. The urban pipe network leak detection and diagnosis system according to claim 4 is characterized in that: The weighted pipeline topology map sequence is clustered based on a clustering algorithm to divide the pipeline network in the target area into sub-pipeline networks of different modes, and further includes: The membership is updated based on the cluster center point to obtain an updated membership, wherein the calculation formula of the updated membership is as follows: Where c represents the number of different modes, Represents the distance from node j to the center of cluster i in the weighted pipeline topology sequence; It is determined whether the error between the current iteration membership and the previous iteration is less than a preset error threshold, or whether the maximum number of iterations has been reached. If so, the iteration is stopped to divide the pipeline network in the target area into sub-pipeline networks of different modes.

6. The urban pipe network leak detection and diagnosis system according to claim 3, characterized in that: The leakage detection model includes a pressure leakage detection module, a vibration leakage detection module and a label mapping module, wherein the pressure leakage detection module adopts a support vector machine, and obtains the pressure leakage detection label corresponding to the sub-pipeline network based on the pressure signal of the node in the sub-pipeline network, the vibration leakage detection module adopts a graph neural network, and obtains the vibration leakage detection label corresponding to the sub-pipeline network based on the vibration signal of the node in the sub-pipeline network, and the label mapping module adopts a neural network, and performs label mapping based on the pressure leakage detection label and vibration leakage detection label corresponding to the sub-pipeline network to obtain the leakage detection label corresponding to the sub-pipeline network.

7. The urban pipe network leak detection and diagnosis system according to claim 6, characterized in that: Acquiring a pressure leak detection tag corresponding to the sub-pipeline network based on a pressure signal of a node in the sub-pipeline network includes: Performing multiple random walks on each node in the sub-pipeline network, each walk starts from a random node and selects a neighboring node to continue walking until the number of walk steps reaches a predetermined length, so as to obtain a walk path sequence; By training the walking path sequence, an embedding vector of each node is learned, wherein the embedding vector includes a hidden feature vector, wherein the hidden layer feature represents the position and semantics of the node in the sub-pipeline network; Based on the embedding vector, elevation and pressure of each node in the sub-pipeline network as a feature vector; The feature vector is classified based on a pre-trained support vector machine model to obtain a pressure leak detection label corresponding to the sub-pipeline network.

8. The urban pipe network leak detection and diagnosis system according to claim 7, characterized in that: Acquiring a vibration leak detection tag corresponding to the sub-pipeline network based on a vibration signal of a node in the sub-pipeline network includes: The vibration signal is divided based on sliding window sampling to obtain a sub-vibration signal sequence, wherein the expression of the sliding window sampling is as follows: Where n represents the number of sub-vibration signals after sliding window sampling, floor represents the rounding function, L represents the length of the vibration signal, d represents the sliding window length, and s represents the step size of each sliding window; The sub-vibration signal sequence is converted into a corresponding frequency spectrum based on fast Fourier transform, wherein the expression of the fast Fourier transform is as follows: in, represents the spectrum of the ith sub-vibration signal of the node in the sub-pipeline network, FFT represents the fast Fourier transform function, Represents the frequency spectrum of the vibration signal, y n Indicates the fault type. Check whether the nth sub-vibration signal is greater than the preset spectrum threshold. If it is greater than the preset spectrum threshold, then y n =1, otherwise, y n =0; The frequency domain of the sub-vibration signal is taken as the node, the spectrum mean is taken as the node value, a visual graph is constructed, the spectrum is taken as the node feature of the visual graph, and the fault type is taken as the node label of the visual graph.

9. The urban pipe network leak detection and diagnosis system according to claim 8, characterized in that: Acquiring a vibration leak detection tag corresponding to the sub-pipeline network based on a vibration signal of a node in the sub-pipeline network, further comprising: The attention coefficient between each node is calculated based on the node features of the visible graph, wherein the attention coefficient is the edge weight of the visible graph, and the calculation formula of the attention coefficient is as follows: Among them, α ij represents the weight of the edge connecting the i-th node and the j-th node in the visible graph, ReLU represents the activation function, a represents the learnable parameter vector, and W represents the graph filter parameter matrix; The embedding features between the nodes are calculated based on the attention coefficients and node features between the nodes in the visibility graph, wherein the calculation formula of the embedding features is as follows: Among them, z i Represents the embedded features of the i-th node in the visibility graph; The embedded features between each node in the visible graph are passed through the convolutional layer of the embedded self-attention mechanism and the stacked graph attention to obtain the aggregated features; The aggregated features are classified based on the fully connected layer and softmax to obtain the vibration leak detection label corresponding to each node in the sub-pipeline network.

10. The urban pipe network leak detection and diagnosis system according to claim 9, characterized in that: Label mapping is performed based on the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network to obtain the leak detection label corresponding to the sub-pipeline network, including: Based on a preset label code set, encoding and assigning values ​​to the pressure leak detection label and the vibration leak detection label corresponding to the sub-pipeline network to obtain a label code sequence; The label coding sequence is matched with a preset label sequence table to obtain a leak detection label corresponding to the sub-pipeline network.

Images