An optimization method for water quality regional monitoring points in a water supply network

Abstract

An embodiment of the present invention provides a method for optimizing water quality regional monitoring points in a water supply network, including: constructing and calibrating a hydraulic water quality model based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network; using a structural depth clustering algorithm to combine the node time-series water quality change curves simulated by the model with the topological structure of the water supply network to divide all nodes in the water supply network into quality regions; comprehensively considering the node coverage rate and node water age to score the nodes in each quality region, and sorting the nodes in each quality region from high to low according to the scores; selecting a node combination with the top scores and node spacing meeting the set requirements from each quality region to generate a candidate monitoring point combination; comprehensively considering the multi-index information entropy and the coverage optimization target, and combining with the node spacing constraint to determine a monitoring point layout plan that meets the information gain and coverage requirements. In this way, the layout of monitoring points can be effectively optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of water supply networks, and in particular, to a method for optimizing water quality regional monitoring points in a water supply network. Background Art

[0002] As an important infrastructure for ensuring clean drinking water, the water supply network undertakes the core task of providing a safe water source for residents, enterprises, and industries. However, due to its distributed characteristics and vulnerability to pollution, the water supply network faces various potential risks such as exogenous pollution and endogenous pollution. These pollutions will spread through the network, affecting water quality safety and causing serious economic losses. Especially in the context of network aging and management deficiencies, it becomes increasingly difficult to ensure the water quality safety and stable operation of the water supply system.

[0003] With the development of water quality monitoring technology, the water quality monitoring system of modern water supply networks has been able to achieve real-time monitoring of the water quality status of the network to a certain extent. By collecting data on conventional water quality indicators such as residual chlorine, water age, and turbidity in real time, the monitoring system can help water supply companies grasp the dynamic changes of the water quality in the network, and indirectly judge whether there are potential abnormal situations by analyzing the abnormal fluctuations of these conventional indicators, providing data support for further investigation and rapid response. Especially the high-frequency data collection and multi-dimensional data fusion analysis can more accurately identify water quality fluctuations and improve the comprehensiveness and accuracy of the evaluation.

[0004] However, the existing water quality monitoring technology still has the defect of insufficient optimization of the monitoring point layout. Specifically, the existing monitoring point layout often relies on experience or simplified rules, lacking in-depth analysis of the complex topological structure of the network, resulting in uneven distribution of monitoring points and inability to fully cover the key areas of the network. Especially in large-scale and complex network systems, there are often blind spots. Therefore, although the existing technology has improved the monitoring ability of the water supply network to a certain extent, the monitoring point layout still needs to be further optimized. Summary of the Invention

[0005] In a first aspect, an embodiment of the present invention provides a method for optimizing water quality regional monitoring points in a water supply network, the method comprising:

[0006] Based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, a hydraulic water quality model is constructed and calibrated to minimize the error between the simulated values output by the hydraulic water quality model and the monitoring values at the positions of the monitoring points, wherein the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data;

[0007] Using the structural depth clustering algorithm, combine the node time-series water quality change curves simulated by the calibrated hydraulic water quality model with the water supply network topology to divide all nodes in the water supply network into quality regions, ensuring the consistency of the water quality change patterns of the nodes within the same quality region;

[0008] Comprehensively consider the node coverage rate and node water age to score the nodes in each quality region, and sort the nodes in each quality region from high to low according to the scores;

[0009] Through the greedy algorithm, select the node combinations with the top scores and node spacing meeting the set requirements from each quality region to generate candidate monitoring point combinations;

[0010] Comprehensively consider the multi-index information entropy and coverage optimization objectives, combine the node spacing constraints, and process the candidate monitoring point combinations through the greedy algorithm to determine the monitoring point layout plan that meets the information gain and coverage requirements.

[0011] In some implementable ways of the first aspect, based on the monitoring data of the existing monitoring points in the water supply network and the topology of the water supply network, construct and calibrate a hydraulic water quality model to minimize the error between the simulated values output by the hydraulic water quality model and the monitoring values of the monitoring points, including:

[0012] Based on the topology of the water supply network, construct a hydraulic model and calibrate it using hydraulic monitoring data. On the basis of the hydraulic model, based on the topology of the water supply network, construct water quality models corresponding to different water quality indicators and calibrate them using water quality monitoring data, and obtain the node time-series water quality change curves simulated by the water quality models; specifically, on the basis of the hydraulic model, set the source water quality pattern of the water plant effluent to simulate the water quality change entering the water supply network; set the initial main body water residual chlorine decay coefficient of all pipes of the same water plant effluent to the same value, and set different initial pipe wall residual chlorine decay coefficients for pipes of different pipe materials; through the genetic algorithm, obtain the main body water decay coefficients and pipe wall residual chlorine decay coefficients of all pipes to minimize the difference between the simulated residual chlorine concentration values calculated by the water quality model and the actual monitoring values of the corresponding monitoring points.

[0013] In some implementable ways of the first aspect, using the structural depth clustering algorithm, combine the node time-series water quality change curves simulated by the calibrated hydraulic water quality model with the water supply network topology to divide all nodes in the water supply network into quality regions, ensuring the consistency of the water quality change patterns of the nodes within the same quality region, including:

[0014] Generate an adjacency matrix based on the water supply network topology, construct a graph structure reflecting the physical connection relationship of nodes, and at the same time, based on the node time-series water quality change curves, construct a node feature matrix to describe the time-series characteristics and dynamic changes of nodes in the graph structure;

[0015] Pre-train an autoencoder to compress the node feature matrix, enhance the separability of features, and minimize the error between the original feature matrix and its reconstructed value through the reconstruction loss to generate an effective data representation;

[0016] Layer by layer, transfer the feature representations output by the autoencoder to a Graph Neural Network (GCN) module, perform graph convolution in combination with the adjacency matrix, fuse the structural information and temporal features of the nodes, and generate multi-level graph-aware node representations;

[0017] Use the Student's t-distribution to calculate the soft assignment probability Q between the nodes and the cluster centers, generate the target distribution P, and minimize the difference between the target distribution P and the soft assignment probability Q through the KL divergence, that is, the self-supervised clustering loss and the difference between the target distribution P and the classification distribution Z of the GCN module, that is, the classification loss of the GCN module Use the dual self-supervised mechanism to uniformly guide the joint learning;

[0018] Integrate the reconstruction loss the self-supervised clustering loss and the classification loss of the GCN module for joint optimization, and finally determine the clustering result of each node through the classification probability distribution output by the GCN module;

[0019] Adopt the elbow method to determine the optimal number of partitions. Specifically, by calculating the sum of squared errors of the node temporal feature curves under different numbers of partitions, find the inflection point of the error change trend, and use the number of partitions corresponding to the inflection point as the optimal number of partitions.

[0020] In some realizable ways of the first aspect, pre-train an autoencoder to compress the node feature matrix, enhance the separability of features, and minimize the error between the original feature matrix and its reconstructed value through the reconstruction loss to generate an effective data representation, including:

[0021] Design the encoder structure to map the input features to a low-dimensional space and generate a dimensionality-reduced feature representation:

[0022] ;

[0023] wherein, is the original input feature matrix, is the feature representation output by the encoder, is the feature representation of the th layer of the encoder, is the number of layers of the encoder, is the The weight matrix and bias of the layer, is the activation function;

[0024] Design the decoder structure to reconstruct the input features and generate the reconstruction value through inverse mapping:

[0025] ;

[0026] Among them, is the feature representation output by the encoder, is the feature representation of the th layer of the decoder, is the number of decoder layers, is the th layer of the weight matrix and bias of the decoder, is the activation function;

[0027] Calculate the reconstruction loss to measure the difference between the original feature matrix and the reconstruction matrix:

[0028] ;

[0029] Among them, is the reconstruction loss, is the number of nodes, is the Euclidean norm, is the Frobenius norm, is the original feature vector of the th node, is the reconstructed feature vector of the th node, is the original feature matrix of all nodes, is the reconstructed feature matrix of all nodes;

[0030] By minimizing the reconstruction loss Train the autoencoder to learn the dimensionality-reduced node feature representation.

[0031] In some realizable ways of the first aspect, the feature representations output by the autoencoder are passed layer by layer to the graph neural network GCN module, and graph convolution is performed in combination with the adjacency matrix to fuse the structural information and temporal features of the nodes to generate multi-level graph-aware node representations, including:

[0032] Design the GCN module through the graph structure formed by the adjacency matrix, and continuously update the feature representation of the nodes through information propagation between nodes:

[0033] ;

[0034] Among them, is the The node feature representation of the layer is the activation function is the degree normalization matrix is the adjacency matrix plus self-loops is the node feature representation of the layer of the GCN module is the layer weight matrix of the GCN module is the number of layers of the GCN module. The first layer of the input to the GCN module , where is the original feature matrix of the input

[0035] Adopt a transfer operator to fuse the two feature representations of the autoencoder feature and the GCN structure information layer by layer:

[0036] ;

[0037] Among them is the feature representation of the fused node is the hyperparameter for weighing the two representations

[0038] Input the fused feature representation into the next layer of GCN:

[0039] ;

[0040] The last layer of the GCN module generates the probability distribution of each node belonging to each cluster center through Softmax:

[0041] ;

[0042] Among them is the probability that node i belongs to cluster center j is the feature representation of the last layer of the GCN module is the weight matrix of the last layer of the GCN module

[0043] In some realizable ways of the first aspect, use the Student's t-distribution to calculate the soft assignment probability Q between the node and the cluster center, generate the target distribution P, and minimize the difference between the target distribution P and the soft assignment probability Q through the KL divergence, that is, the self-supervised clustering loss , and the difference between the target distribution P and the classification distribution Z of the GCN module, that is, the classification loss of the GCN module , use a double self-supervised mechanism to uniformly guide the joint learning, including:

[0044] Calculate the similarity between each node and the cluster center through the Student's t-distribution to obtain the soft assignment probability Q:

[0045] ;

[0046] wherein, represents the soft assignment probability that node i is assigned to cluster center j, represents the feature representation of the i-th node generated by the autoencoder, represents the j-th cluster center vector, represents the remaining cluster center vectors, and v represents the degrees of freedom of the Student's t-distribution, represents the soft assignment probability from the node to the cluster center;

[0047] Generate the target distribution P based on the soft assignment probability Q to increase the weight of high-confidence assignments:

[0048] ;

[0049] wherein, is the cluster probability in the target distribution, is the soft assignment probability assigned to the remaining cluster centers, is the soft assignment frequency of the j-th cluster center, is the normalization term representing all cluster centers, is the target distribution that strengthens the cluster assignment of high confidence;

[0050] Minimize the difference between the target distribution P and the soft assignment probability Q through KL divergence to ensure that the features learned by the autoencoder are close to the cluster centers, thereby optimizing the cluster representation:

[0051] ;

[0052] wherein, is the difference between the target distribution P and the soft assignment probability Q, represents the number of nodes, represents the number of clusters;

[0053] Supervise the classification distribution Z output by the GCN module through the target distribution P and minimize the KL divergence between the two:

[0054] ;

[0055] wherein, is the difference between the target distribution P and the classification distribution Z of the GCN module represents the number of nodes, represents the number of clusters.

[0056] In some realizable ways of the first aspect, the nodes in each quality area are scored by comprehensively considering the node coverage rate and the node water age, and the nodes in each quality area are sorted from high to low according to the scores, including:

[0057] Based on the influence range of the direct connection pipes of the nodes and their secondary nodes, calculate the coverage rate of each node in each quality area, and convert it into a coverage rate score to reflect the coverage ability of the node to the water supply network;

[0058] Based on the water age index, calculate the water age of each node in each quality area, and convert it into a water age score to reflect the relative position of the node in the water supply network;

[0059] Combine the coverage rate score and the water age score, and perform weighted calculation according to the preset corresponding weights to obtain the comprehensive score of the node, and sort the nodes in each quality area from high to low according to the comprehensive score.

[0060] In some realizable ways of the first aspect, based on the influence range of the direct connection pipes of the nodes and their secondary nodes, calculate the coverage rate of each node in each quality area, and convert it into a coverage rate score, including:

[0061] For each node, calculate the length of the pipes directly connected to it, and use the other nodes connected by these pipes as secondary nodes, calculate the length of the pipes connected by the secondary nodes, and then calculate the sum of the length of the pipes directly connected to the node and the length of the pipes connected by the secondary nodes, and use this as the total coverage length of the node. Calculate the ratio of the total coverage length of the node to the total length of the pipes in the water supply network to obtain the coverage rate of the node;

[0062] Standardize the coverage rate of all nodes in the water supply network, and map the coverage rate to the interval [0,1] as the coverage rate score;

[0063] Based on the water age index, calculate the water age of each node in each quality area, and convert it into a water age score, including:

[0064] For each node, calculate the average water age of all time steps based on its water age time series characteristic data;

[0065] Standardize the average water age of all nodes in the water supply network, and map the average water age to the interval [0,1] as the water age score.

[0066] In some realizable ways of the first aspect, through the greedy algorithm, select a combination of nodes with the top scores and node spacing meeting the set requirements from each quality area to generate a candidate monitoring point combination, including:

[0067] Select the node with the highest comprehensive score from each quality area as the initial candidate node set, providing the initial input for the greedy algorithm;

[0068] Calculate the straight-line distance between the current candidate node and the selected nodes, and determine whether it is greater than the preset distance threshold;

[0069] If the straight-line distance between the current candidate node and the selected nodes is greater than the preset distance threshold, add it to the list of selected nodes; otherwise, skip the current candidate node and continue to select nodes in order according to the comprehensive score until the node screening of all quality areas is completed;

[0070] Output the candidate monitoring point combination according to the list of selected nodes.

[0071] In some implementable ways of the first aspect, by integrating the multi-index information entropy and the coverage optimization objective, combining the node spacing constraint, and processing the candidate monitoring point combination through the greedy algorithm, a monitoring point layout plan that meets the requirements of information gain and coverage range is determined, including:

[0072] For each candidate monitoring point combination of each index, evaluate the information gain of the candidate points on all indexes respectively, and take the sum of multiple information entropies as the total information gain of the candidate points;

[0073] Set the node spacing constraint. On the basis of meeting this constraint, comprehensively consider the coverage rate and the change of information gain, and through the greedy algorithm, select the candidate monitoring point combination that maximizes the multi-index comprehensive information entropy gain and meets the coverage rate requirement from all candidate monitoring point combinations, so as to obtain the monitoring point layout plan.

[0074] In the second aspect, an embodiment of the present invention provides a device for optimizing water quality area monitoring points of a water supply network, and the device includes:

[0075] A construction module, configured to construct and correct a hydraulic water quality model based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, so that the simulation value output by the hydraulic water quality model has the smallest error with the monitoring value of the monitoring points at the positions of the monitoring points, where the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data;

[0076] A division module, configured to use the structural depth clustering algorithm to combine the node time-series water quality change curve simulated by the corrected hydraulic water quality model with the topological structure of the water supply network to divide all nodes in the water supply network into quality areas, ensuring the consistency of the water quality change patterns of the nodes in the same quality area;

[0077] A scoring module, configured to comprehensively consider the node coverage rate and the node water age to score the nodes in each quality area, and sort the nodes in each quality area from high to low according to the scores;

[0078] A selection module, configured to select, by means of a greedy algorithm, a node combination with a top score and a node spacing meeting a set requirement from each quality region, and generate a candidate monitoring point combination.

[0079] A determination module, configured to comprehensively consider the multi-index information entropy and the coverage optimization objective, combine the node spacing constraint, and process the candidate monitoring point combination by means of a greedy algorithm to determine a monitoring point layout scheme meeting the information gain and coverage requirements.

[0080] In a third aspect, an embodiment of the present invention provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method as described above.

[0081] In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method as described above.

[0082] Compared with the prior art, the present invention has at least the following technical effects:

[0083] (1) The present invention adopts a structural depth clustering technology, extracts the time series features of nodes under specific water quality indicators through an autoencoder, embeds them into a graph convolutional network, and updates the model parameters through a double-layer self-supervised mechanism. Finally, all nodes are divided into quality regions by combining node features and topological information. In this way, it is possible to effectively identify regions with similar water quality changes in the water supply network and optimize the layout of monitoring points, so as to achieve more accurate water quality monitoring.

[0084] (2) The present invention determines the optimal number of partitions under water quality indicators through the elbow method, and comprehensively considers the node coverage rate and distribution, and optimizes the candidate monitoring point combination under water quality indicators. While ensuring that the quality region can reflect the operation characteristics of the water supply network, it pays attention to the coverage range of nodes and the attention to the end (sensitive points) of the water supply network, and proposes an optimal candidate monitoring point scheme. This strategy can better represent the water quality characteristics of the water supply network and ensure the comprehensiveness and representativeness of monitoring data.

[0085] (3) The present invention combines the quality zoning results of multiple water quality indicators and its optimized set of candidate monitoring points, uses information theory technology to minimize information redundancy and maximize information correlation, selects the most reasonable node combination that meets this condition, and further optimizes the distribution of monitoring points. By constructing a linkage mechanism for multiple monitoring points, it is possible to achieve data supplementation and cross-verification between nodes, improve the accuracy and reliability of monitoring data. At the same time, by combining the monitoring point data with the hydraulic model, the limited monitoring information can be accurately extended to the entire water supply network, comprehensively grasp the water quality status of the water supply network, and enhance the water quality guarantee ability.

[0086] It should be understood that the content described in the Summary of the Invention section is not intended to limit the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present invention will become more apparent. The drawings are used to better understand the present invention and do not constitute a limitation to the present invention. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0088] Figure 1 is a flowchart of a method for optimizing the water quality regional monitoring points of a water supply network provided by an embodiment of the present invention;

[0089] Figure 2 is a schematic diagram of the balance relationship curve between the number of monitoring points and the gain effect provided by an embodiment of the present invention;

[0090] Figure 3 is a schematic diagram of the multi-index monitoring point layout plan and the comprehensive optimization result provided by an embodiment of the present invention;

[0091] Figure 4 is a structural diagram of a device for optimizing the water quality regional monitoring points of a water supply network provided by an embodiment of the present invention;

[0092] Figure 5 is a structural diagram of an exemplary electronic device capable of implementing the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0093] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0094] In addition, the term "and / or" in the present invention merely describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally represents an "or" relationship between the front and rear associated objects.

[0095] To solve the technical problems in the background art, embodiments of the present invention provide a method, device, equipment, and storage medium for optimizing water quality regional monitoring points in a water supply network. The following will, in conjunction with the accompanying drawings, elaborate on a method, device, equipment, and storage medium for optimizing water quality regional monitoring points in a water supply network provided by embodiments of the present invention through specific embodiments.

[0096] Figure 1 It is a flowchart of a method for optimizing water quality regional monitoring points in a water supply network provided by an embodiment of the present invention. As Figure 1 shown, the method 100 for optimizing water quality regional monitoring points in a water supply network may include:

[0097] S110. Based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, construct and calibrate a hydraulic water quality model to minimize the error between the simulated values output by the hydraulic water quality model and the monitoring values at the monitoring points. Among them, the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data.

[0098] S120. Adopt a structural depth clustering algorithm to combine the node time-series water quality change curves simulated by the calibrated hydraulic water quality model with the topological structure of the water supply network to divide all nodes in the water supply network into quality regions, ensuring the consistency of the water quality change patterns of the nodes within the same quality region.

[0099] S130. Comprehensively consider the node coverage rate and node water age to score the nodes in each quality region, and sort the nodes in each quality region from high to low according to the scores.

[0100] S140. Through a greedy algorithm, select a combination of nodes with the top scores and node spacing meeting the set requirements from each quality region to generate a candidate monitoring point combination.

[0101] S150. Comprehensively consider the multi-index information entropy and coverage optimization objectives, combine the node spacing constraints, and process the candidate monitoring point combination through a greedy algorithm to determine a monitoring point layout plan that meets the information gain and coverage requirements.

[0102] For the convenience of further understanding, the above content will be elaborated below in combination with specific embodiments.

[0103] Based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, a hydraulic and water quality model is constructed and calibrated to minimize the error between the simulated values output by the hydraulic and water quality model and the monitoring values at the monitoring points. Among them, the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data.

[0104] (1.1) Collect the positions, types (for example, hydraulic monitoring points for monitoring pressure, flow, etc.; water quality monitoring points for monitoring residual chlorine, turbidity, etc.) of existing monitoring points in the water supply network, as well as hydraulic and water quality monitoring data. Optionally, a hydraulic and water quality online monitoring system or an artificial inspection method at a certain frequency can be used to detect and collect hydraulic and water quality data.

[0105] (1.2) Based on the topological structure of the water supply network, a hydraulic model is constructed and calibrated using hydraulic monitoring data. On the basis of the hydraulic model, based on the topological structure of the water supply network, water quality models corresponding to different water quality indicators are constructed and calibrated using water quality monitoring data, and the time-series water quality change curves of each node simulated by the model are obtained.

[0106] Furthermore, on the basis of the hydraulic model, the water source water quality pattern of the water plant's effluent is set to simulate the water quality change entering the water supply network.

[0107] Set the initial main body water residual chlorine decay coefficient of all pipelines from the same water plant's effluent to the same value, and set different initial pipe wall residual chlorine decay coefficients for pipelines of different pipe materials.

[0108] Through the genetic algorithm, the main body water decay coefficient and the pipe wall residual chlorine decay coefficient of all pipelines are obtained to minimize the gap between the simulated residual chlorine concentration value calculated by the water quality model and the actual monitoring value of the corresponding monitoring point.

[0109] (2) Using the structural depth clustering algorithm, combine the time-series water quality change curves of the nodes simulated by the calibrated hydraulic and water quality model with the topological structure of the water supply network to divide the quality regions of all nodes in the water supply network to ensure the consistency of the water quality change patterns of the nodes within the same quality region.

[0110] (2.1) Generate an adjacency matrix based on the topological structure of the water supply network to construct a graph structure reflecting the physical connection relationship of the nodes. At the same time, based on the time-series water quality change curves of the nodes, construct a node feature matrix to describe the time-series characteristics and dynamic changes of the nodes in the graph structure.

[0111] (2.2) Pre-train an autoencoder to compress the node feature matrix and enhance the separability of the features. Minimize the error between the original feature matrix and its reconstructed value through the reconstruction loss to generate an effective data representation.

[0112] Furthermore, design an encoder structure to map the input features into a low-dimensional space and generate a reduced-dimensional feature representation:

[0113] ;

[0114] where, is the original input feature matrix, is the feature representation output by the encoder, is the feature representation of the th layer of the encoder, is the number of layers of the encoder, is the weight matrix and bias of the th layer of the encoder, is the activation function.

[0115] Design a decoder structure to reconstruct the input features and generate a reconstruction value through inverse mapping:

[0116] ;

[0117] where, is the feature representation output by the encoder, is the feature representation of the th layer of the decoder, is the number of layers of the decoder, is the weight matrix and bias of the th layer of the decoder, is the activation function.

[0118] Calculate the reconstruction loss to measure the difference between the original feature matrix and the reconstruction matrix:<mmath>

[0119] ;

[0120] where, is the reconstruction loss,<mmath> is the number of nodes,<mmath> is the Euclidean<mmath> norm,<mmath> is the Frobenius norm,<mmath> is the original feature vector of the <mmath> th node,<mmath> is the reconstructed feature vector of the <mmath> th node,<mmath> is the original feature matrix of all nodes,<mmath> is the reconstructed feature matrix of all nodes.

[0121] Train the autoencoder by minimizing the reconstruction loss<mmath> so that it learns the reduced-dimensional node feature representation.

[0122] (2.3) The feature representation output by the autoencoder is passed layer by layer to the graph neural network (GCN) module, and graph convolution is performed in combination with the adjacency matrix to fuse the structural information and temporal features of the nodes to generate a multi-level graph-aware node representation.

[0123] Furthermore, the GCN module is designed based on the graph structure composed of the adjacency matrix, and the feature representation of the nodes is continuously updated through information propagation between nodes:

[0124] ;

[0125] in, It is the GCN module The node feature representation of the layer, is the activation function, is the normalized degree matrix, Add self-loops to the adjacency matrix, It is the GCN module The node feature representation of the layer, It is the GCN module The weight matrix of the layer, is the number of GCN module layers, the first layer of GCN module input ,in, is the original feature matrix of the input.

[0126] The transfer operator is used to fuse the two feature representations of autoencoder features and GCN structure information layer by layer:

[0127] ;

[0128] in, is the feature representation of the fused node, is a hyperparameter that trades off between the two representations.

[0129] The fused features are represented as Input to the next layer of GCN:

[0130] .

[0131] The last layer of the GCN module generates the probability distribution of each node belonging to each cluster center through Softmax:

[0132] ;

[0133] in, is the probability that node i belongs to cluster center j, It is the feature representation of the last layer of the GCN module. is the weight matrix of the last layer of the GCN module.

[0134] (2.4) Calculate the soft assignment probability Q between the nodes and the cluster centers using the Student's t-distribution, generate the target distribution P, and minimize the difference between the target distribution P and the soft assignment probability Q through the KL divergence, which is also the self-supervised clustering loss , and the difference between the target distribution P and the classification distribution Z of the GCN module, which is also the classification loss of the GCN module , and use the dual self-supervised mechanism to uniformly guide the joint learning.

[0135] Further, calculate the similarity between each node and the cluster center through the Student's t-distribution to obtain the soft assignment probability Q:

[0136] ;

[0137] where represents the soft assignment probability that node i is assigned to cluster center j, represents the feature representation of the i-th node generated by the autoencoder, represents the j-th cluster center vector, represents the remaining cluster center vectors, v represents the degrees of freedom of the Student's t-distribution, represents the soft assignment probability from the node to the cluster center.

[0138] Generate the target distribution P based on the soft assignment probability Q to increase the weight of high-confidence assignments:

[0139] ;

[0140] where is the cluster probability in the target distribution, is the soft assignment probability assigned to the remaining cluster centers, is the soft assignment frequency of the j-th cluster center, is the normalization term representing all cluster centers, is the target distribution that strengthens the high-confidence cluster assignments.

[0141] Minimize the difference between the target distribution P and the soft assignment probability Q through the KL divergence to ensure that the features learned by the autoencoder are close to the cluster centers, thereby optimizing the cluster representation:

[0142] ;

[0143] where is the difference between the target distribution P and the soft assignment probability Q, represents the number of nodes, represents the number of clusters.

[0144] Supervise the classification distribution \(Z\) output by the GCN module through the target distribution \(P\), and minimize the KL divergence between the two:

[0145] ;

[0146] where is the difference between the target distribution \(P\) and the classification distribution \(Z\) of the GCN module represents the number of nodes, represents the number of clusters.

[0147] (2.5) Comprehensive reconstruction loss , self-supervised clustering loss and the classification loss of the GCN module are jointly optimized, and finally the clustering results of each node are determined through the classification probability distribution output by the GCN module.

[0148] (2.6) Use the elbow method to determine the optimal number of partitions. Specifically, by calculating the sum of squared errors of the node time series feature curves under different numbers of partitions, find the inflection point of the error change trend, and use the number of partitions corresponding to the inflection point as the optimal number of partitions, so as to balance the partition accuracy and the model complexity and improve the rationality of the quality area division.

[0149] (3) Comprehensively consider the node coverage rate and the node water age to score the nodes in each quality area, and sort the nodes in each quality area from high to low according to the scores.

[0150] (3.1) Based on the directly connected pipes of the node and the influence range of its secondary nodes, calculate the coverage rate of each node in each quality area, and convert it into a coverage rate score to reflect the coverage ability of the node to the water supply network.

[0151] Further, for each node, calculate the length of the pipes directly connected to it, and use the other nodes connected by these pipes as secondary nodes, calculate the length of the pipes connected by the secondary nodes, and then calculate the sum of the length of the pipes directly connected to the node and the length of the pipes connected by the secondary nodes, and use this as the total coverage length of the node. Calculate the ratio of the total coverage length of the node to the total length of the pipes in the water supply network to obtain the coverage rate of the node.

[0152] Standardize the coverage rates of all nodes in the water supply network, map the coverage rates to the interval [0, 1] as the coverage rate scores, and ensure that the coverage ranges of different nodes can be compared on a unified scale to reflect the relative importance of the node's coverage ability to the pipe network. Among them, the formula used for standardization is as follows:

[0153] ;

[0154] where is the node coverage rate score, is the node coverage rate, is the minimum node coverage rate, is the maximum node coverage rate.

[0155] (3.2) Based on the water age index, calculate the water age of each node in each quality area. Nodes with longer water ages are usually close to the end of the pipe network and may have higher water quality risks. Then convert it into a water age score to reflect the relative position of the node in the water supply network and its importance.

[0156] For each node, calculate the average water age of all time steps based on its water age time series characteristic data to characterize the overall level of the water retention time of the node;

[0157] Standardize the average water age of all nodes in the water supply network, map the average water age to the interval [0,1] as the water age score, eliminate the absolute scale difference of the water age values of different nodes, ensure that the scores are compared within a unified range, and thus reflect the relative position of the node and its potential impact on water quality safety. Among them, the formula used for standardization is as follows:

[0158] ;

[0159] Among them, is the node water age score, is the average water age of the node, is the minimum average water age of the node, is the maximum average water age of the node.

[0160] (3.3) Combine the coverage rate score and the water age score, perform weighted calculation according to the preset corresponding weights to obtain the comprehensive score of the node, and sort the nodes in each quality area from high to low according to the comprehensive score.

[0161] (4) Through the greedy algorithm, select the node combination with the top scores and the node spacing meeting the set requirements from each quality area to generate the candidate monitoring point combination.

[0162] (4.1) Select the node with the highest comprehensive score from each quality area as the initial candidate node set, providing the initial input for the greedy algorithm;

[0163] (4.2) Based on the comprehensive score ranking, and at the same time meeting the physical distance constraint between nodes, calculate the straight-line distance between the current candidate node and the selected nodes, and judge whether it is greater than the preset distance threshold;

[0164] (4.3) If the straight-line distance between the current candidate node and the selected nodes is greater than the preset distance threshold, add it to the list of selected nodes; otherwise, skip the current candidate node and continue to select nodes in order according to the comprehensive score ranking until the node screening for all quality regions is completed;

[0165] (4.4)Output the candidate monitoring point combinations according to the list of selected nodes.

[0166] (5)Integrate the multi-index information entropy and the coverage optimization objective, combine the node spacing constraint, and process the candidate monitoring point combinations through a greedy algorithm to determine the monitoring point layout plan that meets the requirements of information gain and coverage.

[0167] (5.1)Evaluate the information gain of each candidate point in all indicators for each candidate monitoring point combination of each indicator, and take the sum of multiple information entropies as the total information gain of the candidate point;

[0168] (5.2)Set the node spacing constraint. On the basis of meeting this constraint, comprehensively consider the coverage rate and the change of information gain, and through the greedy algorithm, select the candidate monitoring point combination that maximizes the multi-index comprehensive information entropy gain and meets the coverage rate requirement from all candidate monitoring point combinations to obtain the monitoring point layout plan.

[0169] As an example, the implementation process of the optimized method for water quality regional monitoring points of water supply networks proposed by the present invention is further elaborated below in combination with Case Pipe Network 1 in the BWSN competition.

[0170] (1)Based on the existing monitoring point data and the pipe network topology structure, construct and calibrate the hydraulic and water quality model to minimize the error between the simulated value and the monitored value at the monitoring point position, thereby improving the accuracy of the model. Use the EPANET water quality solver for water quality modeling and simulate the water age and residual chlorine concentration of the nodes. In this case, the monitoring data for calibration is generated by EPANET software simulation, and the measurement error follows a Gaussian distribution with a mean of 0. The extraction frequency of water quality simulation data is once every 30 minutes. Starting from the simulation start time, collect the water quality data of 126 nodes within 96 consecutive hours. Store the processed data in the characteristic matrices of water age and residual chlorine respectively. There are a total of 193 time steps of water quality data, and the dimension of each characteristic matrix is 126 rows and 193 columns for subsequent analysis and modeling.

[0171] (2) By adopting the structural depth clustering method, the time-series water quality change curves of nodes simulated by the calibrated water quality model are combined with the topological structure of the water supply network to divide all nodes into quality regions. By comprehensively considering node characteristics and topological relationships, the consistency of the water quality change patterns of nodes within the same region is ensured. The elbow method is used to analyze the sum of squared errors of the time-series characteristic curves of nodes under different numbers of partitions, determine the inflection point of the error change trend, and use the number of partitions corresponding to the inflection point as the optimal quality partition number. Under the water age index, through calculation and verification, the optimal number of partitions for this water supply network is determined to be 9; under the residual chlorine index, the optimal number of partitions is 10. Finally, the partition results are combined with the actual situation of the pipe network, and economic costs and partition effects are comprehensively considered to optimize the partition results, providing a basic basis for the layout of subsequent monitoring points.

[0172] (3) Nodes within each quality region are scored. The scoring criteria comprehensively consider the coverage rate and node water age, and the nodes within the region are sorted according to their scores. The coverage rate score is obtained by calculating the ratio of the length of the pipes directly connected to the node and the pipes within the influence range of its secondary nodes to the total length of the pipe network. The higher the coverage rate, the wider the monitoring range of the node. The water age score is calculated based on the average water age of the node. The higher the water age, it usually indicates that the node is closer to the end of the pipe network and can better monitor the water quality changes in sensitive areas. Finally, through the weighted synthesis of the coverage rate and water age scores, the comprehensive score of each node is generated. The higher the comprehensive score of the node, the more widely it can monitor the water quality changes in the pipe network and pay attention to the water quality characteristics at the end of the pipe network.

[0173] (4) Through the greedy algorithm, a combination of nodes with higher scores and node spacing meeting the set requirements is selected from each quality region to generate a candidate monitoring point scheme. When arranging monitoring points, to ensure the coverage effect between monitoring points, the physical distance between any two monitoring points needs to be set to not less than 1 km. Specifically, the nodes are sorted from high to low according to their scores, and the node with the highest score is selected in turn, and it is checked whether the shortest path distance between this node and the selected nodes meets the constraint conditions. On the basis of meeting the distance constraint, node combinations with higher scores are preferentially selected to ensure that the monitoring point layout can widely cover the pipe network and maximize the effectiveness and rationality of the monitoring scheme.

[0174] (5) Integrate the information entropy and coverage rate optimization objectives of multiple indicators, combine with the node spacing constraint, and finally determine the layout plan of monitoring points that meets the requirements of information gain and coverage through the greedy algorithm. In specific implementation, based on the quality zoning results of the two indicators of water age and residual chlorine, first extract the intersection nodes of the candidate sets of monitoring points corresponding to the two indicators, and give priority to including them in the final layout plan. Subsequently, perform weighted calculations on the information gain and coverage rate gain of the remaining candidate nodes, and sort them according to the weighted scores. During the node selection process, it is necessary to simultaneously meet the physical constraint condition that the node spacing is not less than 1 km. On the premise of meeting the constraint conditions, combined with the cost-effectiveness trade-off relationship of the layout of monitoring points, gradually select nodes from the candidate nodes with higher scores and add them to the final plan. Finally, according to the balance relationship between the layout number and the gain effect (such as Figure 2 shown), select 7 monitoring points, and the specific layout plan is as Figure 3 shown.

[0175] In summary, the present invention has at least the following technical effects:

[0176] (1) The present invention adopts the structural depth clustering technology, extracts the time series features of nodes under specific water quality indicators through the autoencoder, and embeds them into the graph convolutional network. At the same time, the model parameters are updated through a double-layer self-supervised mechanism. Finally, the quality regions of all nodes are divided by combining the node features and topological information. In this way, it can effectively identify the regions with similar water quality changes in the water supply network and optimize the layout of monitoring points, so as to achieve more accurate water quality monitoring.

[0177] (2) The present invention determines the optimal number of partitions under the water quality indicators through the elbow method, and comprehensively considers the node coverage rate and distribution, and optimizes the candidate monitoring point combinations under the water quality indicators. While ensuring that the quality regions can reflect the operation characteristics of the water supply network, it pays attention to the coverage range of nodes and the attention to the end points (sensitive points) of the water supply network, and proposes the optimal candidate monitoring point plan. This strategy can better represent the water quality characteristics of the water supply network, and at the same time ensure the comprehensiveness and representativeness of the monitoring data.

[0178] (3) The present invention combines the quality zoning results of multiple water quality indicators and their optimized candidate monitoring point sets, uses information theory technology to minimize information redundancy and maximize information correlation, selects the most reasonable node combination that meets this condition, and further optimizes the distribution of monitoring points. By constructing a linkage mechanism for multiple monitoring points, it can realize the supplement and cross-validation of data between nodes, improve the accuracy and reliability of monitoring data. At the same time, combined with the monitoring point data and the hydraulic model, accurately promote the limited monitoring information to the entire water supply network, comprehensively master the water quality status of the water supply network, and improve the water quality guarantee ability.

[0179] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0180] The above is the introduction of the method embodiments. The following further illustrates the solution of the present invention through device embodiments.

[0181] Figure 4 The following is a structural diagram of an optimized device for water quality regional monitoring points of a water supply network provided by an embodiment of the present invention. As Figure 4 shown, the optimized device 400 for water quality regional monitoring points of a water supply network may include:

[0182] A construction module 410, configured to construct and correct a hydraulic water quality model based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, so that the simulation value output by the hydraulic water quality model is minimized with respect to the monitoring value at the monitoring point. Among them, the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data.

[0183] A division module 420, configured to use a structural depth clustering algorithm to combine the node time-series water quality change curve simulated by the corrected hydraulic water quality model with the topological structure of the water supply network to divide all nodes in the water supply network into quality regions, ensuring the consistency of the water quality change patterns of the nodes in the same quality region.

[0184] A scoring module 430, configured to comprehensively consider the node coverage rate and node water age to score the nodes in each quality region, and sort the nodes in each quality region from high to low according to the scores.

[0185] A selection module 440, configured to select, through a greedy algorithm, a node combination with a top score ranking and a node spacing meeting the set requirements from each quality region to generate a candidate monitoring point combination.

[0186] A determination module 450, configured to comprehensively consider the multi-index information entropy and the coverage optimization target, combine the node spacing constraint, and process the candidate monitoring point combination through a greedy algorithm to determine a monitoring point layout plan that meets the information gain and coverage range requirements.

[0187] It can be understood that Figure 4 each module / unit in the optimized device 400 for water quality regional monitoring points of a water supply network shown in Figure 1The functions of the various steps in the optimized method 100 for water quality regional monitoring points shown can achieve their corresponding technical effects. For the sake of brevity, they will not be elaborated here.

[0188] Figure 5 It is a structural diagram of an exemplary electronic device capable of implementing the embodiments of the present invention. The electronic device 500 is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 500 can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown in the present invention, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed in the present invention.

[0189] As Figure 5 shown, the electronic device 500 may include a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0190] Multiple components in the electronic device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, an optical disc, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the electronic device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0191] The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 executes the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 can be implemented as a computer program product, including a computer program, which is tangibly contained in a computer-readable medium, such as the storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of method 100 described above can be executed. Alternatively, in other embodiments, the computing unit 501 can be configured to execute method 100 in any other suitable manner (e.g., by means of firmware).

[0192] The various embodiments described above in the present invention can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0193] The program code for implementing the methods of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0194] In the context of the present invention, a computer-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the computer-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0195] It should be noted that the present invention also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute method 100 and achieve the corresponding technical effects achieved by the method of the embodiments of the present invention. For the sake of concise description, details are not repeated herein.

[0196] In addition, the present invention also provides a computer program product, which includes a computer program that implements method 100 when executed by a processor.

[0197] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved. The present invention places no restrictions herein.

[0198] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for optimizing water quality regional monitoring points in a water supply network, characterized in that, The method includes: Based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network, a hydraulic and water quality model is constructed and calibrated to minimize the error between the simulated values output by the hydraulic and water quality model and the monitoring values at the monitoring points. Among them, the monitoring points include hydraulic monitoring points and water quality monitoring points, and the monitoring data includes hydraulic monitoring data and water quality monitoring data; Using the structural depth clustering algorithm, the node time-series water quality change curves simulated by the calibrated hydraulic and water quality model are combined with the topological structure of the water supply network to divide all nodes in the water supply network into quality regions, ensuring that the water quality change patterns of nodes in the same quality region are consistent; Comprehensively considering the node coverage rate and node water age, score the nodes in each quality region, and sort the nodes in each quality region from high to low according to the scores; Through the greedy algorithm, select a combination of nodes with the highest scores and node spacing meeting the set requirements from each quality region to generate a candidate monitoring point combination; Combining the multi-index information entropy and the coverage optimization objective, and combining the node spacing constraint, process the candidate monitoring point combination through the greedy algorithm to determine the monitoring point layout plan that meets the information gain and coverage requirements; The construction and calibration of the hydraulic and water quality model based on the monitoring data of existing monitoring points in the water supply network and the topological structure of the water supply network to minimize the error between the simulated values output by the hydraulic and water quality model and the monitoring values at the monitoring points includes: Based on the topological structure of the water supply network, a hydraulic model is constructed and calibrated using hydraulic monitoring data. On the basis of the hydraulic model, water quality models corresponding to different water quality indicators are constructed based on the topological structure of the water supply network and calibrated using water quality monitoring data, and the time-series water quality change curves of each node simulated by the water quality model are obtained. Specifically, on the basis of the hydraulic model, the water source water quality pattern of the water plant effluent is set to simulate the water quality change entering the water supply network; the initial main body water chlorine residual decay coefficient of all pipes of the same water plant effluent is set to the same value, and different initial pipe wall chlorine residual decay coefficients are set for pipes of different materials; through the genetic algorithm, the main body water decay coefficient and pipe wall chlorine residual decay coefficient of all pipes are obtained to minimize the gap between the simulated value of the chlorine residual concentration calculated by the water quality model and the actual monitoring value of the corresponding monitoring point.

2. The method according to claim 1, wherein The use of the structural depth clustering algorithm to combine the node time-series water quality change curves simulated by the calibrated hydraulic and water quality model with the topological structure of the water supply network to divide all nodes in the water supply network into quality regions to ensure that the water quality change patterns of nodes in the same quality region are consistent includes: Based on the topological structure of the water supply network, an adjacency matrix is generated to construct a graph structure reflecting the physical connection relationship of nodes. At the same time, based on the node time-series water quality change curves, a node feature matrix is constructed to describe the time-series characteristics and dynamic changes of nodes in the graph structure; Pre-train an autoencoder to compress the node feature matrix, enhance the separability of features, and minimize the error between the original feature matrix and its reconstructed value through the reconstruction loss L res to generate an effective data representation; The feature representations output by the autoencoder are gradually passed to the graph neural network GCN module, and graph convolution is performed in combination with the adjacency matrix to fuse the structural information and time-series characteristics of the nodes to generate a multi-level graph-aware node representation; Calculate the soft assignment probability Q between the node and the cluster center using the Student's t-distribution, generate the target distribution P, and minimize the difference between the target distribution P and the soft assignment probability Q through the KL divergence, that is, the self-supervised clustering loss L clu and the difference between the target distribution P and the classification distribution Z of the GCN module, that is, the classification loss L of the GCN module gcn , and use the double self-supervised mechanism to uniformly guide the joint learning; Comprehensive reconstruction loss L res , self-supervised clustering loss l clu and the classification loss L of the GCN module gcn are jointly optimized, and finally the clustering result of each node is determined by the classification probability distribution output by the GCN module; The elbow method is used to determine the optimal number of partitions. Specifically, by calculating the sum of squared errors of the node time series feature curves under different numbers of partitions, the inflection point of the error change trend is found, and the number of partitions corresponding to the inflection point is used as the optimal number of partitions.

3. The method according to claim 2, wherein The pre-trained autoencoder compresses the node feature matrix to enhance the separability of features, and minimizes the error between the original feature matrix and its reconstruction value through the reconstruction loss L res to generate an effective data representation, including: Design the encoder structure to map the input features to a low-dimensional space and generate a reduced-dimensional feature representation: Among them, H (0) = X is the original input feature matrix, and H (l) is the feature representation output by the encoder. H (l-1) is the feature representation of the (l - 1)-th layer of the encoder. L is the number of layers of the encoder, is the weight matrix and bias of the l-th layer of the encoder, and φ is the activation function; Design the decoder structure to reconstruct the input features and generate a reconstructed value through inverse mapping: Among them, is the feature representation output by the decoder, is the feature representation of the (l-1)-th layer of the decoder, where L is the number of decoder layers, is the weight matrix and bias of the l-th layer of the decoder, and φ is the activation function; Calculate the reconstruction loss to measure the difference between the original feature matrix and the reconstructed matrix: Among them, L res is the reconstruction loss, N is the number of nodes, and ‖·‖ 2 is the Euclidean l2 norm, is the Frobenius norm, x i is the original feature vector of the i-th node, is the reconstructed feature vector of the i-th node, X is the original feature matrix of all nodes, is the reconstructed feature matrix of all nodes; By minimizing the reconstruction loss L res Train an autoencoder to learn a reduced-dimensional node feature representation.

4. The method according to claim 3, characterized in that The feature representation output by the autoencoder is passed layer by layer to the graph neural network GCN module. Graph convolution is performed in combination with the adjacency matrix to fuse the structural information and time series features of the nodes and generate multi-level graph-aware node representations, including: Design the GCN module through the graph structure formed by the adjacency matrix, and continuously update the feature representation of the nodes through information propagation between nodes: Among them, Z (l) is the node feature representation of the l-th layer of the GCN module, φ is the activation function, is the normalized degree matrix, is the adjacency matrix plus the self-loop, Z (l-1) is the node feature representation of the (l - 1)-th layer of the GCN module, is the weight matrix of the (l - 1)-th layer of the GCN module, L is the number of layers of the GCN module, and the first layer of the input of the GCN module Among them, X is the original feature matrix of the input; Use a transfer operator to fuse the two feature representations of the autoencoder features and the GCN structural information layer by layer: Among them, is the feature representation of the fused node, and ∈ is the hyperparameter for weighing the two representations; Input the fused feature representation into the next layer of GCN: The last layer of the GCN module generates the probability distribution of each node belonging to each cluster center through Softmax: Among them, Z ij is the probability that node i belongs to cluster center j, and Z (L) is the feature representation of the last layer of the GCN module, and is the weight matrix of the last layer of the GCN module.

5. The method according to claim 4, characterized in that, Calculating the soft assignment probability Q between nodes and cluster centers using Student's t-distribution, generating the target distribution P, and minimizing the difference between the target distribution P and the soft assignment probability Q through KL divergence, that is, the self-supervised clustering loss L clu , and the difference between the target distribution P and the classification distribution Z of the GCN module, that is, the classification loss L of the GCN module gcn , using a dual self-supervised mechanism to uniformly guide joint learning, including: Calculate the similarity between each node and the cluster center through the Student's t-distribution to obtain the soft assignment probability Q: where q ij represents the soft assignment probability that node i is assigned to cluster center j, h i represents the feature representation of the i-th node generated by the autoencoder, μ j represents the j-th cluster center vector, μ j' represents the remaining cluster center vectors, v represents the degrees of freedom of the Student's t-distribution, Q = [q ij represents the soft assignment probability from nodes to cluster centers; Generate the target distribution P based on the soft assignment probability Q to increase the weight of high-confidence assignments: where p ij is the clustering probability in the target distribution, q ij' is the soft assignment probability assigned to the remaining cluster centers, f j = ∑ i q ij is the soft assignment frequency of the j-th cluster center, f j' is the normalization term representing all cluster centers, and P = [p ij is the target distribution that strengthens the clustering assignment of high confidence; Minimize the difference between the target distribution P and the soft assignment probability Q through KL divergence to ensure that the features learned by the autoencoder are close to the cluster center, thereby optimizing the cluster representation: where L clu is the difference between the target distribution P and the soft assignment probability Q, I represents the number of nodes, and J represents the number of clusters; Supervise the classification distribution Z output by the GCN module through the target distribution P and minimize the KL divergence between the two: Among them, L gcn is the difference between the target distribution P and the classification distribution Z of the GCN module, I represents the number of nodes, and J represents the number of clusters.

6. The method according to claim 1, characterized in that, The nodes in each quality region are scored by comprehensively considering the node coverage rate and node water age, and the nodes in each quality region are sorted from high to low according to the scores, including: Based on the influence range of the directly connected pipes of the nodes and their secondary nodes, calculate the coverage rate of each node in each quality region and convert it into a coverage rate score to reflect the coverage ability of the node to the water supply network; Based on the water age index, calculate the water age of each node in each quality region and convert it into a water age score to reflect the relative position of the node in the water supply network; Combine the coverage rate score and the water age score, perform weighted calculation according to the preset corresponding weights to obtain the comprehensive score of the node, and sort the nodes in each quality region from high to low according to the comprehensive score.

7. The method according to claim 6, wherein Based on the influence range of the directly connected pipes of the nodes and their secondary nodes, calculate the coverage rate of each node in each quality region and convert it into a coverage rate score, including: For each node, calculate the length of the pipes directly connected to it, and use the other nodes connected by these pipes as secondary nodes to calculate the length of the pipes connected by the secondary nodes. Then calculate the sum of the length of the pipes directly connected to the node and the length of the pipes connected by the secondary nodes as the total coverage length of the node. Calculate the ratio of the total coverage length of the node to the total length of the pipes in the water supply network to obtain the coverage rate of the node; Standardize the coverage rates of all nodes in the water supply network, map the coverage rate to the interval [0,1] as the coverage rate score; Based on the water age index, calculate the water age of each node in each quality area and convert it into a water age score, including: For each node, calculate the average water age of all time steps based on its water age time series characteristic data; Standardize the average water age of all nodes in the water supply network, map the average water age to the interval [0, 1], and use it as the water age score.

8. The method according to claim 1, characterized in that, By using the greedy algorithm, select a node combination with the top scores and node spacing meeting the set requirements from each quality area to generate a candidate monitoring point combination, including: Select the node with the highest comprehensive score from each quality area as the preliminary candidate node set, providing the initial input for the greedy algorithm; Calculate the straight-line distance between the current candidate node and the selected nodes, and determine whether it is greater than the preset distance threshold; If the straight-line distance between the current candidate node and the selected nodes is greater than the preset distance threshold, add it to the list of selected nodes; otherwise, skip the current candidate node and continue to select nodes in sequence according to the comprehensive score ranking until the node screening of all quality areas is completed; Output the candidate monitoring point combination according to the list of selected nodes.

9. The method according to claim 1, wherein Integrate the multi-index information entropy and coverage optimization objectives, combine the node spacing constraints, and process the candidate monitoring point combination through the greedy algorithm to determine the monitoring point layout plan that meets the requirements of information gain and coverage range, including: For each candidate monitoring point combination of each index, evaluate the information gain of the candidate points on all indexes respectively, and use the sum of multiple information entropies as the total information gain of the candidate points; Set the node spacing constraint. On the basis of meeting this constraint, comprehensively consider the coverage rate and the change of information gain, and through the greedy algorithm, select the candidate monitoring point combination that maximizes the multi-index comprehensive information entropy gain and meets the coverage rate requirement from all candidate monitoring point combinations to obtain the monitoring point layout plan.

Images