A water supply network leakage location method based on monitoring system and spatial clustering

Abstract

The present invention relates to the technical field of urban water supply pipe networks, and provides a method for locating leakage in a water supply pipe network based on a monitoring system and spatial clustering. The present invention performs similarity evaluation on available pressure measurement values ​​on certain selected internal nodes of the pipe network from SCADA and reference values ​​from a simulated leakage scenario database to obtain candidate leakage scenarios, then uses spatial clustering to construct candidate leakage areas, and finally quantifies the possibility that the candidate leakage areas contain real leakage nodes and gives detection priorities for the candidate areas. The present invention comprises the following steps: (1) constructing a dominant sensor sequence; (2) calculating leakage location indicators; (3) scoring candidate leakage scenarios; (4) constructing candidate leakage areas; and (5) locating leakage areas. The method utilizes the temporal correlation between monitoring data of a certain pressure sensor at multiple moments and the spatial correlation between monitoring data of multiple sensors to achieve deep mining of data spatiotemporal information.

Description

Technical Field

[0001] The present invention relates to the technical field of urban water supply pipe networks, and in particular to a method for locating leakage in a water supply pipe network based on a monitoring system and spatial clustering. Background Art

[0002] The urban water supply network system provides guarantee for urban development and residents' life. It is one of the primary infrastructures of the city and needs to ensure its safe, efficient and economical operation. In view of the serious leakage of urban water supply network and frequent pipe burst accidents, which cause a large amount of high-quality water resources to be wasted and affect the normal social production and living order, water supply enterprises need to find the leakage point as quickly as possible and then carry out repair operations to avoid further deterioration of the situation. The method of directly locating the leakage point using equipment such as noise recorders has the disadvantages of high cost, low efficiency and high requirements for the working environment. The leakage location method based on model and data drive is an important means to improve the operation and management level of my country's water supply industry and reduce the leakage of the network.

[0003] In view of this, the present invention aims at improving the leakage positioning accuracy and time efficiency, and proposes a water supply network leakage positioning method based on a monitoring system and spatial clustering with a dominant sensor sequence as the core. Summary of the invention

[0004] One object of the present invention is to provide a method for locating water supply network leakage based on a monitoring system and spatial clustering. To achieve the above object, the technical solution adopted by the present invention is:

[0005] A method for locating water supply network leakage based on a monitoring system and spatial clustering is provided, which is characterized by comprising the following steps:

[0006] (1) Constructing a dominant sensor sequence;

[0007] (2) Calculate leakage location indicators;

[0008] (3) Scoring of candidate leakage scenarios;

[0009] (4) construct candidate leakage areas;

[0010] (5) Locating the leakage area.

[0011] Further, the step (1) is specifically as follows:

[0012] (1.1) Prediction hydraulic model:

[0013] Establish a node water demand prediction model to predict the node water demand in the future DT time periods (t = {1, 2..., T}) in real time; use the predicted water demand at time t as the input of the pipe network model to establish the hydraulic model at time t;

[0014] (1.2) Construct the sensitivity weight matrix:

[0015] Using the hydraulic model at time t, all pressure sensors arranged in the water supply network are traversed to perform sensitivity analysis on all water-using nodes in the network; the sensor sensitivity matrix at time t is obtained. The sensitivity matrices at different times are weighted averaged to finally obtain the sensor sensitivity weight matrix within the prediction time range DT.

[0016]

[0017] The sensor sensitivity matrix is the sensor node S k The node pressure at is node N k Changes in water demand at Indicates that at the network node N k When leakage occurs at sensor S k Effect of pressure change, n s is the total number of sensors and n n is the total number of nodes in the network;

[0018] (1.3) Reconstruct the sensor sequence elements:

[0019] matrix The nth column represents the occurrence of n The leakage loss at the pipe network is n s The sensitivity of the sensors is calculated by sorting the elements in the nth column from large to small according to their sensitivity values. Based on the rule that the sum of the sensitivity values ​​corresponding to the first i elements of the sorted column vector and the total sensitivity value satisfy the set ratio, the sequence elements are reconstructed to obtain node N. n The dominant sensor sequence of node N n The following conditions must be met:

[0020]

[0021] (1.4) Construct leakage pressure residual matrix:

[0022] Based on the hydraulic model at time t in (1.1), the pressure value at the pressure sensor location in the pipe network is calculated to obtain the pressure prediction vector at time t.

[0023]

[0024] in is the pressure prediction data of the i-th sensor at time t; n s is the number of pressure sensors; ns Row and column vectors;

[0025] (1.5) Construct the pressure matrix for simulating leakage scenarios:

[0026] The predicted value of water demand at a node at time t is increased by 5, 10, 15, ... 350 m3 / h in sequence to simulate the possible leakage scenario of the node. The above process traverses all nodes in the pipe network, calculates the pressure value at the pressure sensor position under the leakage scenario, and obtains the pressure matrix of the simulated leakage scenario at time t.

[0027]

[0028] in In the kth leakage scenario, the simulated value of the monitoring data of the i-th pressure sensor; is the pressure vector of the monitoring point under the kth leakage scenario; n is the number of leakage scenarios, n s ×n matrix;

[0029] (1.6) The leakage pressure matrix Each column vector of Subtract to obtain the leakage pressure residual matrix

[0030]

[0031] in is the leakage pressure residual matrix under the kth leakage scenario, n s ×n matrix;

[0032] (1.7) Obtain the real-time pressure residual vector at time t

[0033] When the SCADA system uploads the pressure data at the tth moment (t = {1, 2..., k}), the real-time pressure vector p(t) at the tth moment is obtained.

[0034]

[0035] where p i (t) is the i-th pressure sensor data transmitted by the SCADA system at time t; p(t) is n s Row and column vectors;

[0036] The real-time pressure vector p(t) and the predicted pressure vector Subtract to obtain the real-time pressure residual vector r(t),

[0037] Where r(t) is ns Row and column vector.

[0038] Further, the step (2) is specifically as follows:

[0039] (2.1) Selecting some similarity metrics as leakage location indicators;

[0040] (2.2) Leakage pressure residual matrix Each column vector corresponds to a node N n , determine node N n Whether it is a real leakage node, the node N obtained in (1.3) is used n The dominant sensor sequence is the real-time pressure residual vector r(t) at time t and the leakage pressure residual matrix at time t The membership node N in n The column vector of is reconstructed and preprocessed;

[0041] Among them, node N n The number of dominant sensor sequence utilizations is equal to the number of times it occurs at node N n The total number of simulated leakage scenarios.

[0042] Further, the step (3) is specifically as follows:

[0043] (3.1) The real-time pressure residual vector r′(t) after sequence reconstruction is combined with the column vector of the leakage pressure residual matrix Perform correlation analysis, calculate leakage location indicators, and obtain the similarity vector of the corresponding indicators at time t Assuming that the system performs leakage location analysis by scanning the pressure sensor data covering a total of j time periods within the time window, a real-time location analysis similarity matrix S is constructed. Indicator ,

[0044]

[0045] Where S Indicator is an n×j matrix, is the leakage location index value of the real scene and the kth simulated leakage scene at time t, x is the initial time of the monitoring data sequence in the scanning time window, and x+j is the corresponding time at the end of the monitoring data sequence in the scanning time window;

[0046] (3.2) Based on the time series weighting, the leakage real-time location analysis similarity matrix is ​​converted into a leakage real-time location analysis vector.

[0047]

[0048] Where P't is the time series weight coefficient at time t, S' Indicator is the real-time positioning analysis vector at time t, S′ Indicatoris an n×1 column vector;

[0049] (3.3) Sum the scores of each leakage location index to obtain the candidate scene score vector C Indicator ,

[0050]

[0051] in is the location analysis scoring vector of the i-th leakage location indicator, and SumIn is the total number of selected leakage location indicators.

[0052] Further, the step (4) is specifically as follows:

[0053] The candidate scene score vector C sumscore The simulated leakage scenarios corresponding to the elements above the scoring threshold λ are selected as candidate leakage scenarios. The setting rule of the scoring threshold λ is that after sorting the elements in the scoring vector from large to small, the ratio of the number of elements greater than the threshold to the total elements n is equal to V3; the total number of candidate scenarios nc = n*V3; each candidate scenario contains two characteristic parameters: leakage location and leakage intensity. The above strategy allows scenarios with the same location but different leakage intensity to be selected as candidate scenarios at the same time. The different leakage locations are counted to construct the candidate node set {C node}; Through the clustering algorithm, the set {C node} are spatially clustered to construct several candidate leakage areas, and the cluster center of each area is used as the candidate leakage area center.

[0054] Further, the step (5) is specifically as follows:

[0055] To quantify the detection priority of leakage candidate areas, the following three indicators should be calculated separately:

[0056] (5.1) Indicates the sensor sequence:

[0057] Based on the pressure monitoring data at time t covered by the scanning time window, the pressure residual vector r(t) is obtained, and then based on the average weight, the pressure residual vector in the time domain is time-series normalized to form a normalized pressure residual vector After arranging the elements in the vector from large to small in terms of value, the sensors corresponding to the element values ​​before the sorting are taken to form the indicator sensor sequence.

[0058]

[0059] Where j (j≤STWL) is the total number of time series monitoring data contained in the scanning time window of this leakage location analysis, and the maximum value is equal to the scanning time window length (STWL);

[0060] (5.2) Candidate region integral:

[0061] All nodes in the candidate node set {Cnode} correspond to a total of n i candidate leakage scenarios, each node corresponds to several candidate scenarios with the same leakage location but different leakage intensity. Based on the candidate scenario scoring vector C sumscor e, node n i The corresponding candidate scene scores are reconstructed into node n i The candidate scene score vector Compute Node n i Total score of SN i ,

[0062]

[0063] in is an N ni row and column vectors, Represents a column vector The kth row element of ; each candidate region contains several elements in the candidate node set {Cnode}. Assuming that after spatial clustering, m candidate regions and m candidate region centers are formed, and each candidate region contains a node set {Nj} (j = 1, 2..., m), then the candidate region integral SN j Equal to belong to {N j All nodes n of} i ∈{N j Total score Sn i Sum; Assume {N j The total number of elements in} is x;

[0064]

[0065] (5.3) Cumulative frequency of candidate regions:

[0066] Each candidate region contains a set of nodes {N j}(j=1,2...,m), calculate the j}Node n i (n i ∈{N j The ratio of the number of times it appears in the candidate scenes to the total number of candidate scenes P i , as shown in the following formula, the candidate region cumulative frequency SP is obtained by summing j (j=1,2...,m)

[0067]

[0068] One or more of the above technical solutions in the embodiments of the present invention have at least the following technical effects or advantages:

[0069] The method for locating leakage in a water supply network based on a monitoring system and spatial clustering is of great significance for improving the comprehensive performance of the leakage locating system. Many sensors are arranged in the water supply network. Due to different installation positions, the feedback information brought to different sensors by a leakage in the network is a coupling of different spatial information and time information. Specifically, the installation geographical location of the sensor, the intricate connection pipeline (topology) relationship between the sensor and the leakage, the size and location of the leakage, the geographical distance between the leakage and the sensor and other factors will affect the feedback information collected by the sensor. In the present invention, the dominant sensor sequence of each candidate leakage node is composed of a specific number of sensors arranged in the network and sensitive to the pressure change on the node. The dominant sensor sequence is conducive to the extraction of positive feedback information, reduces the misleading of the negative feedback information on the leakage location analysis, and can effectively improve the leakage location accuracy. In addition, in order to reduce the leakage amount, the shorter the time required from triggering the leakage monitoring system alarm to the staff finding the real leakage point within the estimated range given by the leakage location algorithm, the better. Since the large-scale pipe network model has tens of thousands of nodes, the leakage area location result may contain multiple nodes, resulting in an increase in the workload of detecting leakage and a longer time. The index for quantifying the detection priority of candidate areas proposed by the present invention effectively improves the time efficiency of detecting leakage. In summary, the present invention realizes the deep mining of spatiotemporal information of monitoring data and improves the accuracy and time efficiency of leakage location. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flow chart of the method for locating leakage in a water supply network in the present invention;

[0071] Figure 2 Provide a distribution map of the water supply network and sensor layout for City J;

[0072] Figure 3 is the sensor sensitivity weight matrix diagram;

[0073] Figure 4 Schematic diagram of candidate leakage areas for leakage location.

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work. DETAILED DESCRIPTION

[0075] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0076] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0077] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0078] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0079] Assuming that a leak occurs at a certain node, multiple pressure sensors will respond simultaneously, resulting in synchronous pressure changes. Considering that the pipeline leak is not repaired immediately, the sensor response will continue for a period of time. In this case, the pressure data obtained at different times at a certain monitoring point are temporally autocorrelated, and the pressure data from different sensors are usually spatially correlated. Using the spatiotemporal correlation of multiple sensors can significantly reduce the uncertainty of a single sensor, thereby improving the robustness of the leak location algorithm.

[0080] The time of leakage, leakage flow, leakage location, sensor location and the topological structure of the pipe network will affect the response of the sensor, resulting in differences in the characterization of the spatiotemporal correlation between sensor measurements. The present invention combines the spatiotemporal information of the pressure sensor with spatial clustering to achieve leakage area positioning; (1) based on the spatial correlation between multiple sensors, a dominant sensor sequence is developed for each node; (2) the temporal correlation between monitoring data of multiple time steps is utilized; (3) the possibility of candidate leakage areas containing real leakage nodes is quantified and the detection priority of candidate areas is given.

[0081] In view of this, the present invention aims to improve the accuracy and time efficiency of leakage location, and proposes a method for locating leakage in a water supply network based on a monitoring system and spatial clustering, with a dominant sensor sequence as the core. This method uses the temporal correlation between the monitoring data of a certain pressure sensor at multiple moments and the spatial correlation between the monitoring data of multiple sensors to achieve deep mining of the spatiotemporal information of the monitoring data, and improve the accuracy and time efficiency of leakage location.

[0082] See also Figure 1 The embodiment of the present application provides a method for locating water supply network leakage based on monitoring system and spatial clustering. The innovation is easy to understand. The following is a further detailed description of the implementation of the present invention in combination with the accompanying drawings and examples (taking DT=3 as an example). The process of the method for locating water supply network leakage in the present invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0083] 1. Build a dominant sensor sequence.

[0084] like Figure 1 As shown in Figure 1, City J has 3 water sources, 491 water demand nodes, 640 pipe sections, and a total length of 433.52 km. The water output of the water plant is known, and a total of 20 pressure sensors are arranged (n s =20).

[0085] (1) Establish a node water demand prediction model to predict the node water demand in the next three time periods (t = {i, i + 1, i + 2}) in real time; use the predicted water demand at time t as the input of the pipe network model to establish the hydraulic model at time t.

[0086] Table 1 Predicted node water demand at time t = 1

[0087] Node Number 1 2 … 240 … 490 491 Water requirement (CMH) 6.83 7.31 … 42.52 … 23.07 36.71

[0088] (2) Using the hydraulic model at time t, traverse all pressure sensors arranged in the water supply network and perform sensitivity analysis on all water-using nodes in the network; obtain the sensor sensitivity matrix at time t The sensitivity matrices at different times are weighted averaged to finally obtain the sensor sensitivity weight matrix within the prediction time range DT. like Figure 2 The weight value is represented by the column height, and the larger the weight, the lighter the color. Nodes with a smaller topological distance or pipeline distance to the pressure monitoring point usually have a higher weight value. The hydraulic correlation between the sensor and the node is effectively quantified. When the node index corresponding to the X-axis and Y-axis coordinates is the same, the weight value peaks, that is, the sensor most related to the node is the sensor arranged at the node. As shown in Table 2, the sensor index "1" corresponds to the node index "368", and the weight value is 6.269 when the coordinates are (368,1). The dominant sensor sequence of the node with index "368" is constructed based on the first column of Table 2.

[0089] Table 2 Partial weight peaks of sensitivity matrix

[0090]

[0091] (3) Sort the elements in the nth column from largest to smallest according to their sensitivity values. Based on the rule that the sum of the sensitivity values ​​corresponding to the first i elements of the sorted column vector and the total sensitivity value satisfy the set ratio, reconstruct the sequence elements to obtain node N. n The dominant sensor sequence is shown in Table 3, where node N n The following conditions must be met:

[0092]

[0093] Table 3. The serial numbers of the leading sensors corresponding to all nodes in the pipe network.

[0094]

[0095] (4) Using the hydraulic model at time t, EPANET software is used to simulate the pressure value at the pressure sensor location as the pressure prediction vector at time t. The elements of the vector are sorted according to the EPANET node index based on the node number corresponding to the pressure sensor.

[0096]

[0097] The pressure prediction vector is a 20-row and 20-column vector. Here is an example of the pressure prediction vector at time t=1:

[0098]

[0099] (5) Simulate the leakage pressure matrix under multiple scenarios at time t

[0100] The water demand for each node increases from 5m3 / h to 350m3 / h, with each increase of m

[0101] / h. Therefore, each node simulates 70 leakage scenarios. The above process traverses all 491 nodes, simulating a total of n = 34370 (491 × 70) leakage scenarios. n s ×n (20×34370) matrix. Here is an example of the leakage pressure matrix at time t=1 (the first 10 columns):

[0102]

[0103] (6) Simulate time t to obtain the leakage pressure residual matrix

[0104] For the first moment, Subtract each column of Calculate the residual matrix of the un-leakage pressure (20×34370). Here is given The first 10 columns of

[0105]

[0106] (7) Obtain the real-time pressure residual vector r(t).

[0107] The real-time pressure vector p(t) is constructed by monitoring the data of multiple pressure sensors transmitted by the SCADA system. The real-time pressure vector p(t) is compared with the predicted pressure vector Subtract to obtain the real-time pressure residual vector

[0108] The real-time pressure vector and the real-time pressure residual vector are both 20-row and 20-column vectors. Here is an example of the real-time pressure vector and the real-time pressure residual vector at time t=1:

[0109]

[0110] 2. Calculate leakage location indicators

[0111] (1) Seven indicators, including Euclidean distance (Eu), Manhattan distance (Ci), Chebyshev distance (Ch), cosine similarity (Co), Pearson correlation coefficient (Pea), Spearman rank correlation coefficient (Spe) and Kendall τ correlation coefficient (Ken), were selected for positioning index calculation, and the average weight was used to accumulate the scores, that is, the proportion of each indicator was 1 / 7 (Eu:Ci:Ch:Spe:Ken:Co:Pea=1:1:1:1:1:1:1).

[0112] (2) Leakage pressure residual matrix Each column vector corresponds to a node N n , determine node N n Whether it is a real leakage node, the node N obtained in (1.3) is used n The dominant sensor sequence is the real-time pressure residual vector r(t) at time t and the leakage pressure residual matrix at time t The membership node N in n The column vector of the node N is reconstructed and preprocessed. n The number of dominant sensor sequence utilizations is equal to the number of times it occurs at node N n The total number of simulated leakage scenarios.

[0113] 3. Scoring of candidate leakage scenarios

[0114] (1) The real-time pressure residual vector r′(t) after sequence reconstruction is combined with the column vector of the leakage pressure residual matrix Perform correlation analysis, calculate leakage location indicators, and obtain the similarity vector of the corresponding indicators at time t

[0115]

[0116] Indicator=Eu,Ci,Ch,Spe,Ken,Co,Pea

[0117] Take the Pearson correlation coefficient in the leakage location indicator as an example:

[0118]

[0119] Correlation Vector is an n-row and column vector, n=34370, which is the number of leakage scenarios. Here is an example at time t=1:

[0120]

[0121] Assuming that the system performs leakage location analysis by scanning the pressure sensor data covering a total of DT = 3 time periods within the time window, a real-time location analysis similarity matrix S is constructed. Indicator ,

[0122]

[0123] Indicator=Eu,Ci,Ch,Spe,Ken,Co,Pea

[0124] (2) According to the time series weight vector, the leakage real-time location analysis similarity matrix is ​​converted into a leakage real-time location analysis vector.

[0125]

[0126] Indicator=Eu,Ci,Ch,Spe,Ken,Co,Pea

[0127] (3) Sum the scores of each leakage location indicator to obtain the candidate scene score vector C Indicator

[0128]

[0129] Indicator=Eu,Ci,Ch,Spe,Ken,Co,Pea

[0130] 4. Construct candidate leakage areas.

[0131] (1) The candidate scene scoring vector C sumscore The simulated leakage scenarios corresponding to the elements above the scoring threshold λ are selected as candidate leakage scenarios. The setting rule of the scoring threshold λ is that after sorting the elements in the scoring vector from large to small, the ratio of the number of elements greater than the threshold to the total elements n is equal to V3. The total number of candidate scenarios nc = n*V3. Each candidate scenario contains two characteristic parameters: leakage location and leakage intensity. The above strategy allows scenarios with the same location but different leakage intensity to be selected as candidate scenarios at the same time. Different leakage locations are counted to construct the candidate node set {C node}. Hierarchical clustering algorithm and k-means clustering algorithm are used to cluster the set {C node} to perform spatial clustering on the elements in the image, construct several candidate leakage areas, and the cluster center of each area is used as the candidate leakage area center. Figure 4 As shown in the figure, the candidate area cluster center is taken as the circle center, and the distance from the node farthest from the center to the center of the circle among all the candidate nodes belonging to the area is taken as the radius to construct the coverage circle of the candidate area. Three candidate leakage areas are constructed, where the number "1" indicates the first detection ranking, that is, the regional positioning analysis results believe that the leakage is most likely to occur in this area. If the real leakage node is located in the candidate area with the first detection priority ranking (the "1st" area), it indicates that the positioning accuracy of the leakage area is good, and the detection priority helps to shorten the time spent on detecting the leakage.

[0132] 5. Locate the leakage area.

[0133] To quantify the detection priority of leakage candidate areas, the following three indicators should be calculated separately:

[0134] (1) Indicator sensor sequence

[0135] Based on the pressure monitoring data at time t covered by the scanning time window, the pressure residual vector r(t) is obtained, and then based on the average weight, the pressure residual vector in the time domain is time-series normalized to form a normalized pressure residual vector After arranging the elements in the vector from large to small in terms of value, the sensors corresponding to the element values ​​before the sorting are taken to form the indicator sensor sequence.

[0136]

[0137] Where j (j≤STWL) is the total number of time series monitoring data contained in the scanning time window of this leakage location analysis, and the maximum value is equal to the scanning time window length (STWL).

[0138] Assuming that after spatial clustering, m candidate regions and m candidate region centers are formed, the distance D from the m candidate region centers to each sensor node in the indicator sensor sequence is calculated in turn. j -i (i=1,...,ng,j=1,2...,m), sum to get the candidate region distance score index SD j =∑D j -i (i=1,...,ng,j=1,2...,m), this study believes that the distance score SD j The lower the value, the higher the possibility that region j contains a real leaky node. j The size of defines the detection priority of the leakage candidate area.

[0139] (2) Candidate region points

[0140] Candidate node set {C node All nodes in} correspond to a total of n i candidate leakage scenarios, each node corresponds to several candidate scenarios with the same leakage location but different leakage intensity. Based on the candidate scenario scoring vector C sumscore , node n i The corresponding candidate scene scores are reconstructed into node n i The candidate scene score vector Compute Node n i Total score of S ni :

[0141]

[0142] in is an N ni row and column vectors, Represents a column vector The k-th row element of each candidate region contains a set of candidate nodes {C node}, assuming that after spatial clustering, m candidate regions and m candidate region centers are formed, and the node set contained in each candidate region is {N j}(j=1,2...,m), then the candidate region integral SNj Equal to belong to {N j All nodes n of} i ∈{N j Total score S ni Assume {N j}, then the total number of elements in is x.

[0143]

[0144]

[0145] The candidate region integration method considers that SN j The higher the value, the higher the possibility that the area contains real leakage nodes. j The size of defines the detection priority of the leakage candidate area.

[0146] (3) Cumulative frequency of candidate regions

[0147] Each candidate region contains a set of nodes {N j}(j=1,2...,m), calculate the j}Node n i (n i ∈{N j The ratio Pi of the number of times the candidate scene appears in the candidate scene to the total number of candidate scenes is as follows, and the sum is used to obtain the cumulative frequency SP of the candidate region: j (j=1,2...,m)

[0148]

[0149] The higher the cumulative frequency of the candidate area, the higher the possibility that the area contains the real leakage node. i The size of defines the detection priority of the leakage candidate area.

[0150] When using the candidate region integral method or candidate region cumulative frequency method to define the region detection priority, if the region evaluation (SN j and SP j ) is very close to each other, it is difficult to give the detection priority. In order to solve this problem, the screening weight Y is introduced. Each node contained in the candidate area corresponds to a sub-index (S ni and P i ), sort these node sub-indices from large to small, and take the sum of the top Y% of sub-indices after sorting to represent the candidate area. This can filter out non-dominant node information in the candidate area (the strength of node information is related to the level of sub-indices), extract the information of dominant nodes, and enhance the distinction between candidate areas to define detection priority.

[0151] The regional positioning accuracy is characterized by the geographical distance between the real leakage node and the candidate leakage area cluster center. Table 4 shows the improvement of positioning accuracy before and after using the dominant sensor sequence under different measurement uncertainties. The use of the dominant sensor sequence can improve the performance of the leakage positioning method. Considering different degrees of sensor measurement uncertainty, the use of the dominant sensor sequence (N sd =6) Compared with not using the dominant sensor sequence, the positioning accuracy is improved by about 40-60%.

[0152] Table 4 Difference in positioning accuracy before and after the use of the dominant sensor under different noise levels

[0153]

[0154] The regional positioning effect of the method is evaluated based on the ranking level of candidate leakage areas matched by the real leakage nodes. Given the matching rule: the distance between the real leakage node and all candidate nodes is calculated respectively, and the real leakage node is matched with the area to which the candidate node corresponding to the minimum distance belongs. As shown in Table 5a and Table 5b, different screening boundaries V3 are taken, and different indicators for quantifying the detection priority of leakage candidate areas are used to evaluate the regional positioning effect of leakage events in different leakage flow intervals.

[0155] The results show that: 1) The regional positioning method performs well within the given leakage flow range. About 97% of the leakage events have real leakage locations within the constructed candidate areas, and 3% of the leakage events are not within the candidate areas. In about 70% of the leakage events, the real leakage nodes are matched to the candidate areas with the first detection priority. The regional detection priority helps to shorten the time to detect the leakage. 2) The regional positioning method has good positioning accuracy within the given leakage flow range. Only the positioning results of the leakage events matched to the candidate areas with the first detection priority are counted. The cumulative probability of the geographical distance between the real leakage node and the candidate node is about 48% within 1000m, 81% within 2000m, and 94% within 3000m. 3) The tested V3 value has little effect on the comprehensive performance of regional positioning. The better V3 value can increase the probability of matching the candidate area with the first detection priority by about 6%, slightly improving the time efficiency of leakage detection. 4) The regional positioning effect is best when the detection priority index uses the indicator sensor sequence. A better detection priority index can increase the probability of matching the first detection priority area by about 34%, and the probability of matching the first and second detection priority areas by about 20%. However, for the five detection priority indicators tested, the proportion of leakage events effectively located in the candidate area exceeded 94%, and the overall performance of regional positioning was not much different. In addition, the larger the leakage flow of the leakage event, the better the leakage area positioning effect. Compared with small flow leakage events, the probability of matching the candidate area of ​​the first detection priority for large flow leakage events increased by about 2%-6%. And the higher the detection priority ranking of the candidate area, the more leakage events will match the candidate area.

[0156] Table 5a Statistical probability of leakage matching results in different leakage flow intervals

[0157]

[0158]

[0159] Table 5b Statistical probability of leakage matching results in different leakage flow intervals

[0160]

[0161] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for locating water supply network leakage based on monitoring system and spatial clustering, characterized in that: The steps include: (1) Constructing a dominant sensor sequence; (2) Calculate leakage location indicators; (3) Scoring of candidate leakage scenarios; (4) construct candidate leakage areas; (5) Location of leakage area; The step (1) is specifically as follows: (1.1) Prediction hydraulic model: Establish a node water demand prediction model to predict the node water demand in the future DT time periods (t = {1, 2..., T}) in real time; use the predicted water demand at time t as the input of the pipe network model to establish the hydraulic model at time t; (1.2) Construct the sensitivity weight matrix: Using the hydraulic model at time t, all pressure sensors arranged in the water supply network are traversed to perform sensitivity analysis on all water-using nodes in the network; the sensor sensitivity matrix at time t is obtained. The sensitivity matrices at different times are weighted averaged to finally obtain the sensor sensitivity weight matrix within the prediction time range DT. The sensor sensitivity matrix is the sensor node S k The node pressure at is node N k Changes in water demand at Indicates that at the network node N k When leakage occurs at sensor S k Effect of pressure change, n s is the total number of sensors and n n is the total number of nodes in the network; (1.3) Reconstruct the sensor sequence elements: matrix The nth column represents the occurrence of n The leakage loss at the pipe network is n s The sensitivity of the sensors is calculated by sorting the elements in the nth column from large to small according to their sensitivity values. Based on the rule that the sum of the sensitivity values ​​corresponding to the first i elements of the sorted column vector and the total sensitivity value satisfy the set ratio, the sequence elements are reconstructed to obtain node N. n The dominant sensor sequence of node N n The following conditions should be met: (1.4) Construct leakage pressure residual matrix: Based on the hydraulic model at time t in (1.1), the pressure value at the pressure sensor location in the pipe network is calculated to obtain the pressure prediction vector at time t. in is the pressure prediction data of the i-th sensor at time t; n s is the number of pressure sensors; n s Row and column vectors; (1.5) Construct the pressure matrix for simulating leakage scenarios: The predicted value of water demand at a node at time t is increased by 5, 10, 15, ... 350 m / h in sequence to simulate the possible leakage scenario of the node. The above process traverses all nodes in the pipe network, calculates the pressure value at the pressure sensor position under the leakage scenario, and obtains the pressure matrix of the simulated leakage scenario at time t. in In the kth leakage scenario, the simulated value of the monitoring data of the i-th pressure sensor; is the pressure vector of the monitoring point under the kth leakage scenario; n is the number of leakage scenarios, n s ×n matrix; (1.6) The leakage pressure matrix Each column vector of Subtract to obtain the leakage pressure residual matrix in is the leakage pressure residual matrix under the kth leakage scenario, n s ×n matrix; (1.7) Obtain the real-time pressure residual vector at time t When the SCADA system uploads the pressure data at the tth moment (t = {1, 2..., k}), the real-time pressure vector p(t) at the tth moment is obtained. where p i (t) is the i-th pressure sensor data transmitted by the SCADA system at time t; p(t) is n s Row and column vectors; The real-time pressure vector p(t) and the predicted pressure vector Subtract to obtain the real-time pressure residual vector r(t), Where r(t) is n s Row and column vector.

2. A method for locating water supply network leakage based on monitoring system and spatial clustering according to claim 1, characterized in that: The step (2) is specifically as follows: (2.1) Selecting some similarity metrics as leakage location indicators; (2.2) Leakage pressure residual matrix Each column vector corresponds to a node N n , determine node N n Whether it is a real leakage node, the node N obtained in (1.3) is used n The dominant sensor sequence of the real-time pressure residual vector r(t) at time t and the leakage pressure residual matrix at time t The membership node N in n The column vector of is reconstructed and preprocessed; Among them, node N n The number of dominant sensor sequence utilizations is equal to the number of times it occurs at node N n The total number of simulated leakage scenarios.

3. The method for locating water supply network leakage based on monitoring system and spatial clustering according to claim 1, characterized in that: The step (3) is specifically as follows: (3.1) The real-time pressure residual vector r′(t) after sequence reconstruction is combined with the column vector of the leakage pressure residual matrix Perform correlation analysis, calculate leakage location indicators, and obtain the similarity vector of the corresponding indicators at time t Assuming that the system performs leakage location analysis by scanning the pressure sensor data covering a total of j time periods within the time window, a real-time location analysis similarity matrix S is constructed. Indicator , Where S Indicator is an n×j matrix, is the leakage location index value of the real scene and the kth simulated leakage scene at time t, x is the initial time of the monitoring data sequence in the scanning time window, and x+j is the corresponding time at the end of the monitoring data sequence in the scanning time window; (3.2) Based on the time series weighting, the leakage real-time location analysis similarity matrix is ​​converted into a leakage real-time location analysis vector. Where P't is the time series weight coefficient at time t, S' Indicator is the real-time positioning analysis vector at time t, S′ Indicator is an n×1 column vector; (3.3) Sum the scores of each leakage location index to obtain the candidate scene score vector C Indicator , in is the location analysis scoring vector of the i-th leakage location indicator, and SumIn is the total number of selected leakage location indicators.

4. A method for locating water supply network leakage based on monitoring system and spatial clustering according to any one of claim 1, characterized in that: The step (4) is specifically as follows: The candidate scene score vector C sumscore The simulated leakage scenarios corresponding to the elements above the scoring threshold λ are selected as candidate leakage scenarios. The setting rule of the scoring threshold λ is that after sorting the elements in the scoring vector from large to small, the ratio of the number of elements greater than the threshold to the total elements n is equal to V3; The total number of candidate scenes nc = n*V3; each candidate scene contains two characteristic parameters: leakage location and leakage intensity. The above strategy allows scenes with the same location but different leakage intensity to be selected as candidate scenes at the same time. The different leakage locations are counted to construct the candidate node set {C node }; Through the clustering algorithm, the set {C node } are spatially clustered to construct several candidate leakage areas, and the cluster center of each area is used as the candidate leakage area center.

5. The method for locating water supply network leakage based on monitoring system and spatial clustering according to claim 1, characterized in that: The step (5) is specifically as follows: To quantify the detection priority of leakage candidate areas, the following three indicators should be calculated separately: (5.1) Indicates the sensor sequence: Based on the pressure monitoring data at time t covered by the scanning time window, the pressure residual vector r(t) is obtained, and then based on the average weight, the pressure residual vector in the time domain is time-series normalized to form a normalized pressure residual vector After arranging the elements in the vector from large to small in terms of value, the sensors corresponding to the element values ​​before the sorting are taken to form the indicator sensor sequence. Where j (j≤STWL) is the total number of time series monitoring data contained in the scanning time window of this leakage location analysis, and the maximum value is equal to the scanning time window length (STWL); (5.2) Candidate region integral: All nodes in the candidate node set {Cnode} correspond to a total of n i candidate leakage scenarios, each node corresponds to several candidate scenarios with the same leakage location but different leakage intensity. Based on the candidate scenario scoring vector C sumscor e, node n i The corresponding candidate scene scores are reconstructed into node n i The candidate scene score vector Compute Node n i Total score of SN i , in is an N ni row and column vectors, Represents a column vector The kth row element of ; each candidate region contains several elements in the candidate node set {Cnode}. Assuming that after spatial clustering, m candidate regions and m candidate region centers are formed, and each candidate region contains a node set {Nj} (j = 1, 2..., m), then the candidate region integral SN j Equal to belong to {N j All nodes n of} i ∈{N j Total score Sn i Sum; Assume {N j The total number of elements in} is x; (5.3) Cumulative frequency of candidate regions: Each candidate region contains a set of nodes {N j }(j=1,2...,m), calculate the j }Node n i (n i ∈{N j The ratio of the number of times it appears in the candidate scenes to the total number of candidate scenes P i , as shown in the following formula, the candidate region cumulative frequency SP is obtained by summing j (j=1,2...,m)

Images