Water supply network leakage area detection method and system based on sparse representation classification

Abstract

The invention provides a sparse representation classification water supply pipe network leakage area detection method and system, and relates to the technical field of water supply pipe network leakage detection, and the method specifically comprises the steps: obtaining the pipe network information of urban water supply pipes, and calculating the node similarity between the nodes of the pipe network, merging by adopting a mutually adjacent micro-cluster merging strategy based on the similarity to obtain an urban water supply pipe partition network; features of abnormal pressure fluctuation in the urban water supply pipe partition network are extracted based on a K-SVD dictionary learning algorithm, and an over-complete dictionary and a multi-label linear classifier are obtained; and obtaining to-be-detected pressure data based on the urban water supply pipe partition network, inputting the to-be-detected pressure data into a multi-label linear classifier for classification to obtain classification labels after the to-be-detected pressure data is inquired and obtained based on the over-complete dictionary, calculating the membership degree of the classification labels based on a membership function to locate a leakage area, and outputting a leakage area detection report. According to the method, the problem that the pressure sensitivity matrix of each node in the urban water supply network is difficult to solve is solved, and the leakage area is detected and positioned.

Description

Technical Field

[0001] The present invention relates to the technical field of water supply pipe network leakage detection, and in particular to a water supply pipe network leakage area detection method and system based on sparse representation classification. Background Art

[0002] With the acceleration of urbanization, the requirements for infrastructure construction are getting higher and higher. Among them, the urban water supply network has become an important infrastructure for the normal operation and life of the city. When the water supply network leaks, it often has an adverse impact on the economy, environment and life. Therefore, the leakage problem of the water supply network has attracted widespread attention. In order to reduce the occurrence of leakage accidents, pressure control technology is used to control the flow of water pipes, increase the water supply flow during the peak period of residential water use, and reduce the water supply flow during the low-peak period of residential water use. Although this has reduced the occurrence of leakage incidents to a certain extent and improved the quality of water supply services, it makes the pressure of the water pipe show complex characteristics, which brings difficulties to the pressure feature capture and leakage location of the water supply network.

[0003] In order to deal with the leakage problem, the existing technology usually deploys detection sites in the water supply network, but the deployed sites are relatively sparse, and only the fluctuation information of the pressure at the detection point can be obtained, which makes the leakage location problem a mathematically underdetermined problem. The existing technology solves it by obtaining the pressure sensitivity matrix of each node in the water supply network, but in practice it is difficult for the water supply network to meet this condition.

[0004] Therefore, it is urgent to develop a solution to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a water supply network leakage area detection method and system based on sparse representation classification, so as to improve the problem of difficulty in solving the pressure sensitivity matrix of each node in the urban water supply network, and then quickly detect and locate the leakage area.

[0006] The present invention provides a sparse representation classification water supply network leakage area detection method and system, which adopts the following technical solutions: In the first aspect, a water supply network leakage area detection method based on sparse representation classification specifically includes: Obtaining the network information of the urban water supply pipe, calculating the node similarity between the nodes of the pipe network, and merging the water supply pipe network by adopting the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the urban water supply pipe partition network; Based on the K-SVD dictionary learning algorithm, the characteristics of abnormal pressure fluctuations in the urban water supply pipe partition network are extracted to obtain an overcomplete dictionary and a multi-label linear classifier; The pressure data to be detected is obtained based on the urban water supply pipe zoning network, and its sparse representation is obtained based on the over-complete dictionary query. The sparse representation is input into a multi-label linear classifier for classification to obtain a classification label, and the membership degree of the classification label is calculated based on the membership function to locate the leakage area, and a leakage area detection report is output.

[0007] The beneficial effects of the sparse representation classification water supply network leakage area detection method provided by the present invention are: in view of the problem that it is difficult to obtain the pressure sensitivity matrix of each node in the urban water supply network, a node similarity measurement criterion with poor energy redundancy is designed, the similarity between nodes is calculated by using the density peak clustering algorithm, and the micro-cluster merging strategy of mutual neighbors is designed to merge the leakage areas with similar characteristics in the urban water supply network; the discriminant coding error and the linear classification error are introduced into dictionary learning as regular terms together with the reconstruction error to ensure that the sparse representation corresponding to different leakage areas has excellent classification characteristics, accurately capture the characteristics of abnormal pressure fluctuations of water supply network nodes, and use the least squares orthogonal matching pursuit algorithm and singular decomposition to improve the speed of learning an over-complete dictionary; finally, the sparse representation of the abnormal pressure is obtained by using dictionary learning and the classification label of the abnormal point is obtained by a multi-label linear classifier, and the leakage area positioning of the urban water supply network is completed by using the membership function.

[0008] Optionally, the process of calculating the node similarity between nodes of the pipe network includes: Calculating energy redundancy differences between nodes of the pipe network, and constructing a node similarity measurement criterion based on the energy redundancy differences; Calculating node similarity between nodes of the pipe network based on the node similarity measurement criterion; Among them, the node similarity measurement criteria include: judging the energy redundancy difference between water supply network nodes based on the density peak clustering algorithm, and the energy redundancy difference is equal to the sum of the position potential energy difference between the nodes and the resistance energy consumption caused by the head loss.

[0009] Optionally, based on the node similarity, a mutually adjacent micro-cluster merging strategy is adopted to merge the water supply network to obtain a city water supply network, including: Based on the node similarity, the nodes of the water supply network are merged adjacent to each other to obtain micro-clusters; calculating a first mutual proximity between a node and a micro-cluster, and calculating a second mutual proximity between micro-clusters; Based on the first mutual proximity and the second mutual proximity, microclusters exceeding a preset threshold are merged in sequence to obtain a city water supply pipe partition network.

[0010] Optionally, when microclusters exceeding a preset threshold are sequentially merged to obtain a city water pipe partition network, if two microclusters belong to different clusters or are semantically irrelevant, the second mutual proximity is set to 0 to avoid merging.

[0011] Optionally, and / or, the formula for calculating the energy redundancy difference between nodes is as follows: ; ; ; In the formula, Represents the position potential energy difference between nodes; It represents the resistance energy consumption caused by head loss; is the weight of water; For Node and The flow rate of the pipe section between and Indicates the nodes at both ends of the pipe segment and The elevation of is the resistance coefficient of the pipe section; Poor energy redundancy; And / or, the water supply network node similarity calculation formula is as follows: ; ; In the formula, Indicates node similarity; For all pipe segments The maximum value of The value is poor energy redundancy; is the set of water demand nodes; is a collection of water source nodes; Representation Node and The similarity matrix between is the identity matrix; And / or, the first mutual proximity calculation formula between the node and the micro-cluster is as follows: ; ; ; In the formula, is the mutual proximity between nodes and micro-clusters, is the proximity of the micro-cluster to the node; is the proximity of the node to the micro-cluster; represents microclusters; , and They are all nodes; and is the weighted proximity; And / or, the second mutual proximity between micro-clusters is calculated as follows: ; In the formula, Represents micro cluster To Micro Cluster The degree of proximity between them; Represents micro cluster To Micro Cluster The degree of proximity between is the second mutual proximity.

[0012] Optionally, the features of abnormal pressure fluctuations in the urban water supply pipe partition network are extracted based on a K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier, including: The discriminative sparse coding error and linear classification error are introduced as regular terms into the objective function of dictionary learning; The objective function is solved based on the least squares orthogonal matching pursuit algorithm, and a sparse representation of the abnormal pressure fluctuation of the urban water supply pipe partition network is obtained; After sparse coding, the dictionary is updated based on the singular value decomposition algorithm; Repeat sparse coding and dictionary updating until the dictionary and sparse representation no longer change, and obtain an overcomplete dictionary and multi-label linear classifier.

[0013] Optionally, obtaining a sparse representation based on an overcomplete dictionary query, and inputting the sparse representation into a multi-label linear classifier for classification to obtain a classification label, including: When abnormal pressure fluctuations are detected in the urban water supply pipe partition network, the pressure data to be detected is obtained, and the sparse representation of the corresponding pressure fluctuations is obtained by querying the pressure data to be detected based on the overcomplete dictionary; The sparse representation is classified based on a multi-label linear classifier to obtain the classification label of the sparse representation.

[0014] Optionally, calculating the membership degree of the classification label based on the membership function to locate the leakage area includes: The membership of the classification labels is calculated based on the membership function and sorted in descending order to obtain a membership set, and the previous The area corresponding to the membership degree is located, and the leakage area is outputted. The leakage area detection report; wherein, is a natural number greater than 1; The membership function is used to calculate the membership of the classification label as follows: ; In the formula, Indicates abnormal pressure fluctuations By Region The degree of membership caused by leakage in For Region Abnormal internal pressure fluctuations The classification label of is the total number of regions, The region number.

[0015] In the second aspect, a sparse representation classification water supply network leakage area detection system specifically includes: The pipe network partitioning module is used to obtain the pipe network information of the urban water supply pipe, calculate the node similarity between the nodes of the pipe network, and merge the water supply pipe network by adopting the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the urban water supply pipe partition network; A feature extraction module, which extracts the features of abnormal pressure fluctuations in the urban water supply pipe partition network based on a K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier; The leakage location module obtains the pressure data to be detected based on the urban water supply pipe zoning network, obtains a sparse representation based on an over-complete dictionary query, inputs the sparse representation into a multi-label linear classifier to obtain a classification label, calculates the membership degree of the classification label based on the membership function to locate the leakage area, and outputs a leakage area detection report.

[0016] Optionally, the multi-label linear classifier in the feature extraction module includes: an input layer, a linear transformation layer, and an output layer; Input layer, used to receive label information and pass it to the linear transformation layer; The linear transformation layer consists of multiple linear classifiers, which are used to calculate the linear combination of input features and corresponding weights and add bias terms to obtain the predicted score or probability of each label; The output layer determines the label set to which the input belongs based on the predicted score or probability of the linear transformation layer and outputs the result.

[0017] The beneficial effects of the second aspect can refer to the description of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The overall flow chart of a water supply network leakage area detection method based on sparse representation classification provided by the present invention; Figure 2 A schematic flow chart of the steps of a method for detecting leakage areas in a water supply network using sparse representation classification provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, 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. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0020] The embodiment of the present invention provides a method for detecting leakage areas in a water supply network based on sparse representation classification. Figure 1 , including the following steps: S1. Obtain the network information of the city water supply pipe, calculate the node similarity between the nodes of the pipe network, and merge the water supply pipe network by using the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the city water supply pipe partition network; S2. Extracting the characteristics of abnormal pressure fluctuations in the urban water supply pipe partition network based on the K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier; S3. Obtain the pressure data to be detected based on the urban water supply pipe zoning network, and obtain its sparse representation based on the over-complete dictionary query, input the sparse representation into a multi-label linear classifier for classification to obtain a classification label, calculate the membership degree of the classification label based on the membership function to locate the leakage area, and output the leakage area detection report.

[0021] In other embodiments, see Figure 2 , calculate the energy redundancy difference between the nodes of the pipe network, merge the micro-cluster based on the mutual proximity degree to obtain the urban water supply pipe partition network, initialize the dictionary, obtain the sparse coefficient representation based on the least squares orthogonal matching pursuit algorithm, and update the dictionary atoms based on the K-SVD algorithm until all atoms are updated and the maximum number of iterations is reached to obtain an over-complete dictionary and a multi-label linear classifier. When the abnormal pressure fluctuation of the urban water supply pipe partition network is monitored, the sparse coefficient representation of the pressure anomaly is obtained based on the over-complete dictionary, and the classification label is obtained through the multi-label linear classifier. The leakage area is located through the membership function and the leakage area report is output.

[0022] In some embodiments, when executing step S1, the following steps are included: S1-1, obtain the network information of the city water supply pipe, and calculate the node similarity between the nodes of the network; S1-2. Based on the node similarity, the water supply network is merged by using a mutually adjacent micro-cluster merging strategy to obtain a city water supply network.

[0023] Specifically, when executing step S1-1, the process of calculating the node similarity between nodes of the pipe network includes: Obtaining the network information of the city water supply pipe, calculating the energy redundancy difference between the nodes of the network, and constructing a node similarity measurement criterion based on the energy redundancy difference; Calculating node similarity between nodes of the pipe network based on the node similarity measurement criterion; Among them, the node similarity measurement criteria include: judging the energy redundancy difference between water supply network nodes based on the density peak clustering algorithm, and the energy redundancy difference is equal to the sum of the position potential energy difference between the nodes and the resistance energy consumption caused by the head loss.

[0024] Furthermore, the network information of urban water supply pipes includes: pipeline information (pipeline material, diameter, length, laying method, burial depth), node information (node ​​location, type, connection relationship), valve information (valve type, location, switch status), water plant water volume data, pump station operation data, pressure data (pressure values ​​of each node in the network), flow data (flow values ​​of each pipe section in the network) and water quality data (water quality indicators in the network, such as turbidity, residual chlorine, pH value, etc.).

[0025] In fact, the formula for calculating the energy redundancy difference between nodes is as follows: ; ; ; In the formula, It represents the position potential energy difference between nodes, reflecting the difference of storage position potential energy between nodes; It represents the resistance energy consumption caused by head loss, reflecting the difficulty of energy transmission between nodes; is the weight of water; For Node and The flow rate of the pipe section between and Indicates the nodes at both ends of the pipe segment and The elevation of is the resistance coefficient of the pipe section; The energy redundancy is poor.

[0026] In fact, the present invention proposes the concept of energy redundancy difference between nodes, and uses node redundancy difference as a density peak clustering algorithm to determine the similarity measurement criterion between nodes in the water supply network; the energy redundancy difference between any two nodes can effectively reflect the node's ability to resist faults. When the energy difference between different nodes in the water supply network is large, the system's ability to resist faults is poor; when the energy distribution between nodes is more balanced, the system's ability to resist faults is stronger; when the energy between nodes is easier to transmit, the system's ability to resist faults is stronger; the energy redundancy difference between nodes can reflect the node's ability to resist faults. For this reason, the energy redundancy difference between nodes is used as the similarity to measure the closeness between nodes. and The difference in the surplus energy between nodes is defined as the energy redundancy difference between nodes.

[0027] Furthermore, based on the energy redundancy difference between the nodes in the water supply network, the density peak clustering algorithm is used to calculate the similarity of the water supply network nodes. The calculation formula for the similarity of the water supply network nodes is as follows: ; ; In the formula, Representation Node and similarity; For all pipe segments The maximum value of The value is poor energy redundancy; is the set of water demand nodes; is a collection of water source nodes; Representation Node and The similarity matrix between is the identity matrix.

[0028] Specifically, when executing step S1-2, the water supply network is merged based on the node similarity by using the mutually adjacent micro-cluster merging strategy to obtain the urban water supply network, including: S1-2-1, based on the node similarity, the nodes of the water supply network are merged adjacent to each other to obtain micro-clusters; S1-2-2, calculating the first mutual proximity between the node and the micro-cluster, and calculating the second mutual proximity between the micro-clusters; S1-2-3. Based on the above step S1-2-2, the microclusters whose first mutual proximity and second mutual proximity exceed the preset threshold are merged in sequence to obtain the urban water supply pipe partition network.

[0029] In fact, when executing step S1-2-1, the computing node To Node The weighted proximity of is calculated, and micro-clusters are obtained by mutual proximity merging based on the weighted proximity. The weighted proximity calculation formula is as follows: ; ; In the formula, Representation Node of The set of nearest neighbors; Representation Node The number of neighbors of For Node and The node similarity of represents the set of water demand nodes, is the natural exponential function, Indicates other situations.

[0030] In fact, when executing step S1-2-2, the first mutual proximity calculation formula between the node and the micro-cluster is as follows: ; ; ; In the formula, is the first mutual proximity between the node and the micro-cluster, is the proximity of the micro-cluster to the node; is the proximity of the node to the micro-cluster; represents microclusters; , and They are all nodes; and is the weighted proximity; The second mutual proximity between micro-clusters is calculated as follows: ; In the formula, Represents micro cluster To Micro Cluster The degree of proximity between them; Represents micro cluster To Micro Cluster The degree of proximity between is the second mutual proximity.

[0031] Furthermore, the first mutual proximity between the node and the micro-cluster, that is, the mutual proximity between the water supply network node and a certain area, the mutual proximity between the node and the micro-cluster indicates the weight of the node belonging to the micro-cluster. The greater the mutual proximity between the node and the micro-cluster, the greater the possibility that the node and the area are divided into the same area; the second mutual proximity between micro-clusters and micro-clusters, that is, the mutual proximity between the water supply network area and the area, the second mutual proximity between micro-clusters and micro-clusters indicates the possibility that two areas belong to the same area.

[0032] In fact, when executing step S1-2-3 and merging microclusters exceeding the preset threshold in sequence to obtain the urban water supply pipe partition network, if two microclusters belong to different clusters or are semantically irrelevant, the second mutual proximity is set to 0 to avoid merging.

[0033] Furthermore, if the two microclusters should actually be together, the mutual proximity of the two microclusters is increased by a corresponding proportion; if the two microclusters should not actually be together, but are very close to each other, the mutual proximity of the two microclusters is set to 0 to avoid merging.

[0034] In some embodiments, when executing step S2, the characteristics of abnormal pressure fluctuations in the urban water supply pipe partition network are extracted based on the K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier, including: S2-1, initialize the dictionary, and introduce the discriminative sparse coding error and linear classification error as regular terms into the objective function of dictionary learning; S2-2, sparse coding, based on the least squares orthogonal matching pursuit algorithm to solve the objective function, to obtain the sparse representation of the abnormal pressure fluctuation of the urban water supply pipe partition network; S2-3, update the dictionary, after sparse coding, update the dictionary based on the singular value decomposition algorithm; S2-4, iterative training, repeating sparse coding and updating the dictionary until the dictionary and sparse representation no longer change, obtaining an overcomplete dictionary and multi-label linear classifier.

[0035] Specifically, when executing step S2-1, the pipe network leakage detection driven by the hydraulic model reconstructs the node leakage contribution rate vector according to the pressure fluctuation vector; Furthermore, the two satisfy the equation: ; In the formula, is the pressure fluctuation vector, is the node leakage contribution rate vector, The pressure-leakage sensitivity matrix is ​​adopted and the normalized leakage feature vector and its classification label multi-label are used as supervision information to adaptively learn the discriminative dictionary and linear label classifier.

[0036] Furthermore, discriminative sparse coding error and classification error are added as regularization terms in the objective function to balance the reconstruction and discrimination capabilities of the dictionary. The discriminative sparse coding error term makes the leakage feature vectors of the same leakage area have similar sparse representations, and the linear classification error term makes the multi-label linear classifier optimal.

[0037] Furthermore, the objective function of dictionary learning is: ; In the formula, and is a weight parameter used to adjust the relative contribution of the reconstruction error and the corresponding regularization term; The pressure-leakage sensitivity matrix corresponding to the quantification of the pipe network zoning results; The discriminative dictionary composed of atoms ensures the over-completeness of the atomic dictionary; is the sparse representation corresponding to the normalized leakage feature vector; It is the block diagonal matrix corresponding to the discriminative sparse coding, which is used to provide classification label information; is the linear transformation matrix; is a multi-label linear classifier.

[0038] Specifically, when executing step S2-2, given an overcomplete dictionary, a normalized leakage feature vector to be encoded, or an abnormal pressure fluctuation, the sparse representation Norm threshold, reconstruction error tolerance; The sparse representation with the smallest norm consists of two parts, namely the support set of the sparse representation and the non-zero values ​​on the support set.

[0039] Furthermore, the support set is initialized to an empty set, and the pressure abnormal fluctuation is initially assigned the residual reconstruction error. In the next iteration, we select atoms from the overcomplete dictionary that are no longer in the support set and that can make the remaining reconstruction error The atom with the smallest norm is expanded into the support set to generate a new support set. After iterations, the remaining reconstruction error is expressed as follows: ; In the formula, is the residual reconstruction error, Abnormal pressure fluctuations, is the word segmentation dictionary composed of atoms corresponding to the support set in the overcomplete dictionary, is a sparse representation, is the number of iterations; Furthermore, in In the first iteration, candidate atoms are added to the word segmentation dictionary, and corresponding support points are added to the support set; In fact, to make the residual reconstruction error The norm is minimal, solving the least squares problem: ; Let the derivative of the above equation be zero to obtain the optimal solution of the least squares problem: ; According to the optimal solution, the remaining reconstruction error after adding the candidate atom is obtained; In the formula, Abnormal pressure fluctuations, is the word segmentation dictionary, is a non-zero value on the newly added support point, is the residual reconstruction error, is a sparse representation, is the matrix transpose, is a candidate atom; Specifically, when executing step S2-3, after sparse coding, K-SVD updates the dictionary through singular decomposition. In order to obtain the optimal solution of the super-complete dictionary and multi-label linear classifier parameters, the objective function of dictionary learning is equivalent to the following form: ; In the formula, , , For a new dictionary, is the block diagonal matrix corresponding to the discriminative sparse coding, is the pressure-leakage sensitivity matrix, are the multi-label linear classifier parameters, is a supercomplete dictionary, is the classification label of the normalized leakage feature vector, is the linear transformation matrix, and is the weight parameter.

[0040] In fact, the purpose of dictionary learning is to obtain a new dictionary, requiring each atom in the dictionary to play its own role, reducing the reconstruction error of the pressure-leakage sensitivity matrix, and updating the atoms in the new dictionary one by one in sequence: each time the update is performed, except for the head atom in the new dictionary, the other atoms are kept unchanged, and the head atom and sparse representation are updated at the same time to minimize the reconstruction error after the update. The update formula is as follows: ; In the formula, is the pressure-leakage sensitivity matrix, For a new dictionary, is the head atom, is the identity matrix, and is a sparse representation, and is the atomic number.

[0041] Furthermore, the singular value decomposition of the identity matrix is ​​applied, and the decomposition formula is as follows: ; in, , is an orthogonal matrix; The first column of is the solution of the head atom, The first column of is the product of the coefficient vector and the unit diagonal matrix, is a diagonal matrix, is the identity matrix.

[0042] Specifically, during the execution of step S2-4, steps S2-2 and S2-3 are repeated until the dictionary and the sparse representation no longer change, thereby obtaining an overcomplete dictionary and a multi-label linear classifier.

[0043] In some embodiments, when executing step S3, it specifically includes: S3-1, obtaining pressure data to be detected based on the urban water supply pipe zoning network, and obtaining its sparse representation based on the super-complete dictionary query, and inputting the sparse representation into a multi-label linear classifier for classification to obtain a classification label; S3-2. Calculate the membership degree of the classification label based on the membership function to locate the leakage area and output the leakage area detection report.

[0044] Specifically, when executing step S3-1, obtaining a sparse representation based on an overcomplete dictionary query, and inputting the sparse representation into a multi-label linear classifier for classification to obtain a classification label, the process includes: When abnormal pressure fluctuations are detected in the urban water supply pipe partition network, the pressure data to be detected is obtained, and the sparse representation of the corresponding pressure fluctuations is obtained by querying the pressure data to be detected based on the overcomplete dictionary; The sparse representation is classified based on a multi-label linear classifier to obtain the classification label of the sparse representation.

[0045] In fact, when abnormal pressure fluctuations are detected in the pipeline network, the sparse representation of the abnormal pressure fluctuations under the overcomplete atomic dictionary is obtained through K-SVD dictionary learning, and then the classification label of the abnormal pressure fluctuations is solved using a multi-label linear classifier.

[0046] Specifically, when executing step S3-2, the membership degree of the classification label is calculated based on the membership function to locate the leakage area, including: The membership of the classification labels is calculated based on the membership function and sorted in descending order to obtain a membership set, and the previous The area corresponding to the membership degree is located, and the leakage area is outputted. The leakage area detection report; wherein, is a natural number greater than 1.

[0047] Furthermore, the membership function calculates the classification label membership as follows: ; In the formula, Indicates abnormal pressure fluctuations caused by the area The degree of membership caused by leakage in For Region Classification label for abnormal internal pressure fluctuations, Abnormal pressure fluctuations, is the total number of regions, The region number.

[0048] In fact, when the abnormal pressure fluctuations monitored are caused by leakages occurring in multiple areas, the classification labels obtained by the sparse representation classification method can effectively reveal all suspicious areas where leakages occur.

[0049] The present invention also provides a water supply network leakage area detection system based on sparse representation classification, which specifically includes: The pipe network partitioning module is used to obtain the pipe network information of the urban water supply pipe, calculate the node similarity between the nodes of the pipe network, and merge the water supply pipe network by adopting the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the urban water supply pipe partition network; A feature extraction module, which extracts the features of abnormal pressure fluctuations in the urban water supply pipe partition network based on a K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier; The leakage location module obtains the pressure data to be detected based on the urban water supply pipe zoning network, obtains its sparse representation based on the over-complete dictionary query, inputs the sparse representation into the multi-label linear classifier to obtain the classification label, calculates the membership degree of the classification label based on the membership function to locate the leakage area, and outputs the leakage area detection report.

[0050] In some embodiments, the multi-label linear classifier in the feature extraction module includes: an input layer, a linear transformation layer, and an output layer; Input layer, used to receive label information and pass it to the linear transformation layer; The linear transformation layer consists of multiple linear classifiers, which are used to calculate the linear combination of input features and corresponding weights and add bias terms to obtain the predicted score or probability of each label; The output layer determines the label set to which the input belongs based on the predicted score or probability of the linear transformation layer and outputs the result.

[0051] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A water supply network leakage area detection method based on sparse representation classification, characterized in that: include: Obtaining the network information of the city water supply pipe, calculating the node similarity between the network nodes, and merging the water supply network by using the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the city water supply pipe partition network; Based on the K-SVD dictionary learning algorithm, the characteristics of abnormal pressure fluctuations in the urban water supply pipe partition network are extracted to obtain an overcomplete dictionary and a multi-label linear classifier; The pressure data to be detected is obtained based on the urban water supply pipe zoning network, and a sparse representation is obtained based on the over-complete dictionary query. The sparse representation is input into a multi-label linear classifier for classification to obtain a classification label, and the membership degree of the classification label is calculated based on the membership function to locate the leakage area, and a leakage area detection report is output.

2. According to the sparse representation classification method for detecting leakage areas in water supply pipe networks according to claim 1, it is characterized in that: The process of calculating the node similarity between pipe network nodes includes: Calculating energy redundancy differences between nodes of the pipe network, and constructing a node similarity measurement criterion based on the energy redundancy differences; Calculating node similarity between nodes of the pipe network based on the node similarity measurement criterion; Among them, the node similarity measurement criteria include: judging the energy redundancy difference between water supply network nodes based on the density peak clustering algorithm, and the energy redundancy difference is equal to the sum of the position potential energy difference between the nodes and the resistance energy consumption caused by the head loss.

3. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 2 is characterized in that: Based on the node similarity, the water supply network is merged by using the mutually adjacent micro-cluster merging strategy to obtain the urban water supply pipe partition network, including: Based on the node similarity, the nodes of the water supply network are merged adjacent to each other to obtain micro-clusters; calculating a first mutual proximity between a node and a micro-cluster, and calculating a second mutual proximity between micro-clusters; Based on the first mutual proximity and the second mutual proximity, microclusters exceeding a preset threshold are merged in sequence to obtain a city water supply pipe partition network.

4. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 3 is characterized in that: When the microclusters exceeding the preset threshold are merged in sequence to obtain the urban water supply pipe partition network, if two microclusters belong to different clusters or are semantically irrelevant, the second mutual proximity is set to 0 to avoid merging.

5. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 3 is characterized in that: And / or, the energy redundancy difference between nodes is calculated as follows: ； ； ； In the formula, Represents the position potential energy difference between nodes; It represents the resistance energy consumption caused by head loss; is the weight of water; For Node and The flow rate of the pipe section between and Indicates the nodes at both ends of the pipe segment and The elevation of is the resistance coefficient of the pipe section; Poor energy redundancy; And / or, the water supply network node similarity calculation formula is as follows: ； ； In the formula, Indicates node similarity; For all pipe segments The maximum value of The value is poor energy redundancy; is the set of water demand nodes; is a collection of water source nodes; Representation Node and The similarity matrix between is the identity matrix; And / or, the first mutual proximity calculation formula between the node and the micro-cluster is as follows: ； ； ； In the formula, is the mutual proximity between nodes and micro-clusters, is the proximity of the micro-cluster to the node; is the proximity of the node to the micro-cluster; represents microclusters; , and They are all nodes; and is the weighted proximity; And / or, the second mutual proximity between micro-clusters is calculated as follows: ； In the formula, Micro cluster To Micro Cluster The degree of proximity between them; Micro cluster To Micro Cluster The degree of proximity between is the second mutual proximity.

6. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 1, characterized in that: Based on the K-SVD dictionary learning algorithm, the characteristics of abnormal pressure fluctuations in the urban water supply pipe partition network are extracted to obtain an overcomplete dictionary and a multi-label linear classifier, including: The discriminative sparse coding error and linear classification error are introduced as regular terms into the objective function of dictionary learning; The objective function is solved based on the least squares orthogonal matching pursuit algorithm, and a sparse representation of the abnormal pressure fluctuation of the urban water supply pipe partition network is obtained; After sparse coding, the dictionary is updated based on the singular value decomposition algorithm; Repeat sparse coding and dictionary updating until the dictionary and sparse representation no longer change, and obtain an overcomplete dictionary and multi-label linear classifier.

7. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 1 is characterized in that: A sparse representation is obtained based on an overcomplete dictionary query, and a classification label is obtained by inputting the sparse representation into a multi-label linear classifier for classification, including: When abnormal pressure fluctuations are detected in the urban water supply pipe partition network, the pressure data to be detected is obtained, and the sparse representation of the corresponding pressure fluctuations is obtained by querying the pressure data to be detected based on the overcomplete dictionary; The sparse representation is classified based on a multi-label linear classifier to obtain the classification label of the sparse representation.

8. The method for detecting leakage areas in a water supply network based on sparse representation classification according to claim 1, characterized in that: The membership degree of the classification label is calculated based on the membership function to locate the leakage area, including: The membership of the classification labels is calculated based on the membership function and sorted in descending order to obtain a membership set, and the previous The area corresponding to the membership degree is located, and the leakage area is outputted. The leakage area detection report; wherein, is a natural number greater than 1; The membership function is used to calculate the membership of the classification label as follows: ； In the formula, Indicates abnormal pressure fluctuations By Region The degree of membership caused by leakage in For Region Abnormal internal pressure fluctuations The classification label of is the total number of regions, The region number.

9. A water supply network leakage area detection system based on sparse representation classification, characterized in that: Specifically include: The pipe network partitioning module is used to obtain the pipe network information of the urban water supply pipe, calculate the node similarity between the nodes of the pipe network, and merge the water supply pipe network by adopting the mutually adjacent micro-cluster merging strategy based on the node similarity to obtain the urban water supply pipe partition network; A feature extraction module, which extracts the features of abnormal pressure fluctuations in the urban water supply pipe partition network based on a K-SVD dictionary learning algorithm to obtain an overcomplete dictionary and a multi-label linear classifier; The leakage location module obtains the pressure data to be detected based on the urban water supply pipe zoning network, obtains a sparse representation based on an over-complete dictionary query, inputs the sparse representation into a multi-label linear classifier to obtain a classification label, calculates the membership degree of the classification label based on the membership function to locate the leakage area, and outputs a leakage area detection report.

10. A water supply network leakage area detection system based on sparse representation classification according to claim 9, characterized in that: The multi-label linear classifier in the feature extraction module includes: an input layer, a linear transformation layer, and an output layer; Input layer, used to receive label information and pass it to the linear transformation layer; The linear transformation layer consists of multiple linear classifiers, which are used to calculate the linear combination of input features and corresponding weights and add bias terms to obtain the predicted score or probability of each label; The output layer determines the label set to which the input belongs based on the predicted score or probability of the linear transformation layer and outputs the result.

Images