Method for monitoring leakage of urban water supply pipe network
By building a hydraulic model and intelligent monitoring model of the water supply pipeline network, the problem of timely detection and positioning of leakage in urban water supply pipeline networks is solved, and the leakage is quickly positioned and treated, reducing water resource waste and management costs.
Patent Information
- Application Number
- CN202410885644.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-03
- Publication Date
- 2024-11-22
AI Technical Summary
The existing technology is difficult to timely detect and locate the leakage of urban water supply pipelines, resulting in increased water resource waste and management costs.
By building a hydraulic model of the water supply pipeline network, selecting pressure monitoring points, simulating leakage conditions, recording pressure waveform data, and training intelligent monitoring models for real-time monitoring and positioning leakage.
It realizes timely discovery and rapid positioning of pipeline leakage, reduces water resource waste and management costs, and improves leakage treatment efficiency.
Smart Images

Figure CN118998624A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of urban water supply pipe network monitoring, and in particular relates to a method for monitoring leakage of an urban water supply pipe network. Background Art
[0002] The water supply network is one of the main infrastructures of the city. However, in actual operation, the water supply network has leakage and burst problems, which affect the effective use of water resources and the stability of supply. In terms of water pressure distribution, some areas in the network have periodic pressure shortages or overpressure problems. This unbalanced water supply phenomenon not only leads to waste of water resources, but also increases the energy consumption and management costs of the network. In addition, due to aging of the network and poor construction quality, water supply pipelines often leak, which also causes a lot of water waste.
[0003] Moreover, compared with pipe burst, pipe burst can usually be discovered in time and handled accordingly because it is visually observable. However, pipeline leakage is different from pipe burst in terms of its visual observability. When leakage occurs, even if it is an obvious leakage, it is necessary to detect the pipeline in a certain area to determine the final leakage node or pipe section; when hidden leakage occurs, it is even more difficult to determine the location of the leakage using the current conventional physical detection methods.
[0004] In summary, it is difficult to detect leakage problems in the water supply network in a timely manner, and it is difficult to locate the specific leakage location after the leakage is discovered. This will not only lead to waste of water resources, but also cause the leakage to not be dealt with in a timely manner, which is prone to escalation of problems and further waste of manpower and material resources.
[0005] Therefore, how to timely discover the leakage of the pipeline network and quickly and effectively locate the leakage position has become a problem that needs to be solved urgently. Summary of the invention
[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for monitoring leakage in a city water supply network, which can timely detect leakage in the network and quickly and effectively locate the leakage position.
[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0008] A method for monitoring leakage of a city water supply network comprises the following steps:
[0009] S1. Obtain the water supply network data of the area to be monitored and construct a corresponding water supply network hydraulic model to achieve mapping of the actual network topology structure;
[0010] S2. Based on the hydraulic model of the water supply network, pressure monitoring points are selected in each node of the water supply network through a multi-objective optimization method to obtain a pressure monitoring plan; wherein the optimization objectives include monitoring range, water volume covered and monitoring cost;
[0011] S3. Based on the hydraulic model of the water supply network, the hydraulic state of leakage is simulated to simulate various leakage conditions of each node; and the 24-hour pressure waveform data of each pressure monitoring point under each leakage condition is recorded as the leakage data set;
[0012] S4. Use the leakage data set to train the preset intelligent monitoring model;
[0013] S5. In the area to be tested of the actual pipe network, pressure monitoring points are arranged according to the pressure monitoring scheme obtained in S2, and real-time data of the pressure monitoring points are collected and input into the intelligent monitoring model trained in S4 to monitor water leakage in the pipe network.
[0014] Compared with the prior art, the present invention has the following beneficial effects:
[0015] Using this method, the corresponding hydraulic model of the water supply network is constructed through the actual water supply network data, and the mapping of the actual network topology can be realized. On this basis, the pressure monitoring points are selected through the multi-objective optimization method to obtain effective water pressure monitoring points. Through the water pressure monitoring data of these water pressure monitoring points, the overall water pressure of the water supply network can be understood.
[0016] When a leak occurs in the pipe network, the pressure at the monitoring point will change to a certain extent. The pressure change at the monitoring point will be different for different node leaks or different amounts of leaks at the same node. In other words, as long as there is a leak (regardless of the degree or type), it will be reflected in the water pressure monitoring data at the water pressure monitoring point, and different leakage situations will correspond to different detection numbers.
[0017] Therefore, this method simulates the hydraulic state of leakage based on the hydraulic model of the water supply network, simulates various leakage conditions of each node, and records the 24-hour pressure waveform data of each pressure monitoring point under each leakage condition to obtain a leakage data set. The intelligent monitoring model is trained with this leakage data set. In this way, through the trained intelligent monitoring model, as long as the water pressure data is input, it can quickly understand whether there is leakage, as well as the specific leakage location and leakage condition.
[0018] Then, the pressure monitoring points obtained by S2 are arranged in the actual pipe network to collect actual pressure data. The actual pressure data can be used by the intelligent monitoring model to determine whether there is leakage, and to quickly analyze the specific leakage location and leakage conditions when building damage occurs.
[0019] In summary, this method can timely discover the leakage of the pipeline network and quickly and effectively locate the leakage position.
[0020] Preferably, in S2, when the pressure monitoring points are selected by the multi-objective optimization method, the optimization objectives are converted into three objective functions: minimum unmonitored range, minimum number of monitoring points, and minimum uncovered water volume; and the pressure monitoring points are selected by the NSGA-Ⅱ algorithm.
[0021] In this way, the overall monitoring cost can be controlled while ensuring the effectiveness of monitoring.
[0022] Preferably, in S2, when selecting the pressure monitoring points by the multi-objective optimization method, the constraints include: the number of monitoring points 1≤k≤10; the monitoring distance of a single monitoring point ≤10 pipe segments; p,q ≥λ, where r p,q represents the sensitivity of node p to the water pressure change of node q, and λ is the preset sensitivity threshold.
[0023] In this way, the effectiveness of the selected monitoring points can be guaranteed.
[0024] Preferably, r p,q The calculation process includes:
[0025] First, the sensitivity coefficient of each node to the change of each node traffic is calculated by node traffic expansion, and the sensitivity coefficient matrix [X] is obtained. n×n ; where n is the number of nodes; x p,q for [X] n×n The element in the pth row and qth column represents the sensitivity coefficient of node q to the flow change of node p;
[0026] After that, the sensitivity coefficient matrix [X] n×n Perform standardization to obtain the standardized sensitivity coefficient matrix [X′] n×n ;
[0027]
[0028] In the formula, x p,q ′ is [X′] n×n The element at the pth row and qth column in ; Represents [X] n×n The average value of the elements in the qth column of S q Represents [X] n×n The standard deviation of the elements in the qth column of ;
[0029] Then, [X′] is obtained by the following formula n×nPerform extreme differentiation and map the data to the interval [0,1] to obtain the matrix [X″] after extreme differentiation n×n :
[0030]
[0031] In the formula, x p,q ″ represents the matrix [X″] n×n The element in row p and column q in p,q ′ represents [X′] n×n The element in row p and column q in qmin ′ represents the minimum element in the qth column of the normalized matrix; x qmax ′ represents the maximum element in the qth column of the normalized matrix;
[0032] Finally, r is calculated by p,q :
[0033]
[0034] And get the node similarity matrix [R] n×n ; r p,q for [R] n×n The element at the p-th row and q-th column in .
[0035] In this way, the pressure monitoring correlation between the nodes can be understood, so as to select a node combination that can effectively monitor the water supply network as a whole as a pressure monitoring solution.
[0036] Preferably, 0.8≤λ<1.
[0037] The effectiveness of the selected detection points can be guaranteed.
[0038] Preferably, in S3, a random function is also used to divide the data in the leakage data set into a training set and a test set; the training set is used to train the intelligent monitoring model, and the test set is used to test the training effect of the intelligent monitoring model; the data set in the leakage data is also preprocessed, and the preprocessing includes data normalization.
[0039] Through such processing, the effectiveness of the intelligent monitoring model training can be guaranteed, thereby ensuring the effectiveness of actual monitoring.
[0040] Preferably, in S4, the intelligent monitoring model is an intelligent model based on a random forest algorithm, a radial basis function network in a neural network, a convolutional neural network in a deep learning algorithm, or a long short-term memory neural network.
[0041] These intelligent models / algorithms are relatively mature and can ensure the effectiveness of monitoring.
[0042] Preferably, the intelligent monitoring model is an intelligent model based on a random forest algorithm.
[0043] Experiments have shown that compared with other intelligent models, the intelligent model based on the random forest algorithm has a very high accuracy in locating leakage points. Deviations will only occur when the leakage amount is small, and the deviation points are all near the actual leakage points, which can still greatly narrow the scope of investigation. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to make the purpose, technical solution and advantages of the invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings, in which:
[0045] Figure 1 A flowchart of the method is shown in FIG.
[0046] Figure 2 Schematic diagram of the time-varying coefficient of the Z region in the embodiment;
[0047] Figure 3 is a topological diagram of the water supply network in the Z area in the embodiment;
[0048] Figure 4 Schematic diagram of the instantaneous working condition of the water supply network in zone Z in the embodiment;
[0049] Figure 5 is an example diagram of the NSGA-Ⅱ algorithm flow in the embodiment;
[0050] Figure 6 It is a schematic diagram of the NSGA-Ⅱ algorithm result in the Z region in the embodiment;
[0051] Figure 7 Schematic diagram of the location of monitoring points in the Z region in the embodiment;
[0052] Figure 8 Schematic diagram of the leakage repair point of the Z zone pipe network in the embodiment;
[0053] Fig. 9 Schematic diagram of the locations of leakage nodes and false alarm nodes in the embodiment. DETAILED DESCRIPTION
[0054] The following is a further detailed description through specific implementation methods:
[0055] Example:
[0056] like Figure 1 As shown, this embodiment discloses a method for monitoring leakage of a city water supply network, comprising the following steps:
[0057] S1. Obtain the water supply network data of the area to be monitored and construct a corresponding water supply network hydraulic model to achieve mapping of the actual network topology.
[0058] The establishment of the hydraulic model of the water supply network requires the collection and organization of geographical data, water quality data, pipe materials and diameters of the network, and the use of modeling software and algorithms to calculate the pressure, flow, speed and other parameters in the network to simulate and optimize the operation of the water supply network, thereby laying the foundation for the subsequent pressure monitoring point site selection and leakage location. In specific implementation, the hydraulic model of the water supply network can be established through the Epanet2.2 software. The establishment of the hydraulic model of the water supply network mainly relies on the flow characteristics of the fluid and the topological structure of the network. The flow of the fluid in the pipeline obeys the mass conservation equation and the momentum conservation equation. The mass conservation equation describes the flow change of the fluid in the pipeline, while the momentum conservation equation describes the pressure change of the fluid in the pipeline. These two equations are the basis for establishing the hydraulic model of the water supply network. The topological structure includes components such as pipelines, valves, pumping stations, water tanks, and the connection relationship between them. By establishing the topological structure of the network, the relationship between each node and pipe section can be determined, providing a basis for subsequent hydraulic analysis.
[0059] In order to better understand this method, an example of hydraulic modeling of a water supply network in a specific area (Z area) is used for illustration.
[0060] The collection of basic information is provided by the online operation and maintenance system and GIS system of the water plant dispatching center. Combined with the field survey, the CAD drawings of the pipeline network construction, the hydraulic data of the water plant and pump station, the historical data of water supply, the records of pipeline network leakage operation and maintenance, etc. are sorted out. The general situation is as follows:
[0061] Pump station and infrastructure information
[0062] The water used in this area is supplied by a booster pump station. The booster pump station has a water supply transfer and lifting capacity of 20,000 tons / day. The booster pump station has a high-level water tank with a static head of 24m. The outlet elevation is 350m and the total outlet head is about 81m. The water plant pump station has a total of 3 horizontal centrifugal pumps, two in use and one in reserve. The basic information of the equipment is shown in Table 1.
[0063] Table 1 Basic information of water pump unit
[0064]
[0065] Water volume and water supply curve
[0066] The water supply area of Zone Z is approximately 50km 2 , obtained water consumption data from the water plant dispatching center, and collected and analyzed the daily water supply flow for nearly 3 years. The highest day was August 19, 2022, and the highest instantaneous water flow was 1425.05m 3 / h. In view of the more obvious gradient of water consumption on the highest day, which has a more significant effect on subsequent pressure monitoring, the water consumption curve on the highest day is used as the daily water consumption curve of the city, and the water consumption at the highest time on the highest day is the outlet flow of the pump station in the hydraulic model. Figure 2 shown.
[0067] Hydraulic foundation information of the pipe network
[0068] In addition to reflecting the topological structure of the pipeline network, the CAD drawings of the pipeline network construction also include basic information such as pipe length, pipe diameter, node coordinates, and major auxiliary structures. Combined with the node elevation provided by the GIS system, the basic hydraulic information of the 228 nodes in the area in addition to the node water demand is determined. The conventional method to determine the node water demand is to analyze the historical data of user water meters provided by the water company. It needs to be based on a large amount of historical data on user water consumption. This method is closer to the water use situation of the actual pipeline network, but the data is not only large in quantity but also difficult to collect. After investigation and analysis, the water use in this area is basically residential water. Except for a few nodes with less water use, the water use of other nodes is relatively balanced. Therefore, the node water demand is determined using the hydraulic leveling software Guanlide. The water plant outlet flow adopts the highest daily maximum of 1425.05m 3 / h, which is converted to 398.85L / S. The flow distribution method is to distribute according to the pipe length. Some flow data from the water plant dispatching center is used for calibration to finally determine the water demand data of each node. Some node and pipe section information is shown in Table 2.
[0069] Table 2 Information of some nodes and pipe sections
[0070]
[0071] Pipeline network topology establishment
[0072] Establishing a pipe network topology model is essentially a process of mapping the actual pipe network. The pipe network construction CAD drawings contained in the collected pipe network data are compared with the existing pipe network information, and the corresponding pipe sections are added and deleted. That is, new trunk pipes with a diameter greater than 100 mm are supplemented, and branch pipes with a diameter less than 100 mm are omitted. The locations of the nodes and main structures (pump stations, pools, water towers, etc.) of the pipe network are also basic information for establishing the topological structure. Save the optimized CAD drawings as dxf files, and then use a format converter to convert the dxf files into inp files that can be recognized by hydraulic modeling software, and import them into Epanet2.2 to achieve the preliminary establishment of the water supply network topology structure. Verify the connection relationship between nodes and main structures in the model to achieve the final establishment of the pipe network topology. Figure 3 As shown, this is the topological structure of the water supply network in area Z.
[0073] The boundary conditions of the water supply network are the restrictions and input parameters set in the hydraulic model simulation and analysis, which are used to simulate the actual situation and constrain the network model. The restrictions include water inlet conditions, water outlet conditions, valve and pump station operating conditions, and pipe section connection conditions. Reasonable setting of boundary conditions is crucial for accurately simulating the operation of the water supply system and predicting performance, and needs to be determined according to actual conditions. The only water inlet of the pipe network in this area is the booster pump station. The corresponding node number in the hydraulic model is J1. During the simulation, all pipe sections are in the normal open state. According to the data content, the pump characteristic curve and water demand mode are constrained accordingly. The head loss formula adopts the Heize-Williams formula, the flow unit adopts L / S, the total duration is 24 hours, the hydraulic step is 1h, and the remaining parameters are constrained according to the default characteristics of Epanet2.2.
[0074] Model verification and optimization
[0075] According to the modeling process, 228 nodes and 246 pipe sections in the Z area network were successfully operated with a hydraulic step length of 1 hour. Figure 4 Shown is the simulated operating status of the water supply network in Zone Z at 10 a.m. on the highest day.
[0076] Taking 1h as the hydraulic step and 24h as the time scale, after simulation, it was found that the pressure of some nodes was high in 24h, which corresponded to the red and yellow nodes in the figure, while the pressure of some nodes was low during the peak water use period, which corresponded to the blue and green nodes on the way. After multiple runs and analyses, the pressure distribution status verified the overall trend of the actual pipe network, but the accuracy and reliability of the model still need further calibration and optimization.
[0077] Model optimization is essentially a verification of the model, which means macroscopically verifying the correctness of the initial hydraulic data, and checking the pipe length, pipe diameter, elevation, node water demand and other data in turn to see if they are correct, so that the optimized model is more in line with the actual pipe network from a macroscopic level. After verification and optimization, the accuracy and reliability of the model meet the requirements, the pipe materials of some pipe sections and the basic water demand of some nodes are adjusted, and a pipe diameter error is corrected. After multiple successful runs, the final Z-area water supply network hydraulic model is obtained.
[0078] S2. Based on the hydraulic model of the water supply network, pressure monitoring points are selected at each node of the water supply network through a multi-objective optimization method to obtain a pressure monitoring plan; wherein the optimization objectives include monitoring range, water coverage and monitoring cost.
[0079] The multi-objective optimization algorithm can effectively solve the problem that the traditional single-objective optimization algorithm only considers one objective function, which leads to the inability to fully utilize the limited monitoring points. The factors that affect the layout of pressure monitoring points are often contradictory, including the number of monitoring points, monitoring range, water volume covered, monitoring cost, burst pipe repair benefits and other factors. The multi-objective problem can be converted into a single-objective problem through linear weighting method, maximum and minimum method, target programming method and other methods. It can also use particle swarm optimization, genetic algorithm, non-dominated sorting genetic algorithm and other multi-objective optimization methods to find a balance point between multiple objective functions, that is, no single solution can achieve the optimal solution for all objectives, but a set of solutions is found, which are optimal or close to the optimal solution for different objective functions. Several methods for converting multi-objective problems into single-objective problems often have strong subjectivity in assigning weights or confrontational comparisons. The direct use of multi-objective optimization algorithms can better control the relationship and influence between different objectives and obtain more scientific solutions.
[0080] In the specific implementation, the pressure monitoring points are selected by NSGA-Ⅱ algorithm. When the pressure monitoring points are selected by multi-objective optimization method, the optimization objectives are converted into three objective functions: minimum unmonitored range, minimum number of monitoring points, and minimum uncovered water volume.
[0081] The basic principle of heuristic search of NSGA-II algorithm (fast non-inferior solution sorting genetic algorithm) is similar to that of traditional genetic algorithm, which also searches and evolves through selection, crossover and mutation. Its core idea is fast non-inferior solution sorting and crowding sorting. Different from the maintenance method of MOPSO algorithm that searches and optimizes by retaining the non-inferior solution set, NSGA-II algorithm uses non-inferior solution sorting to sort individuals and divides individuals into different levels or layers. At the same time, it maintains the diversity of the population by calculating the crowding distance to ensure that the individuals are evenly distributed in the target space. Compared with traditional genetic algorithms, it is faster in optimization speed, and has better solution diversity and convergence. However, compared with MOPSO's high-dimensional processing capabilities of particle parallel search and information exchange, it may face the dimensionality disaster of the search space in high-dimensional problems. However, when optimizing the site selection of pressure monitoring points, the objective functions are monitoring range, water volume coverage, and monitoring cost. With only three objective functions, there is no challenge of high latitude. Although the above algorithms are applicable, the solution set of NSGA is more diverse and balanced, and it will not fall into the local optimal solution. The parameter adjustment is simpler and the operation is easier to implement. Therefore, the NSGA-Ⅱ algorithm is used for multi-objective optimization of pressure monitoring points.
[0082] The basic steps of the NSGA-Ⅱ algorithm add fast non-inferior solution sorting and congestion sorting on the basis of the genetic algorithm. Through fast non-inferior solution sorting, the dominance relationship between individuals is compared to determine their levels or hierarchies, so that individuals in each level are not dominated by individuals in other levels. Its basic steps include:
[0083] (1) Initialization: Set the dominance counts and dominated sets of all individuals in the population to 0.
[0084] (2) Dominance relationship judgment: For any two individuals A and B in the population, the dominance relationship between them is determined by comparing their objective function values:
[0085] ① If A dominates B (i.e., A is not inferior to B in all objective functions and is superior to B in at least one objective function), then increase B’s dominance count by 1 and add A to B’s dominated set.
[0086] ②If B dominates A, increase A’s dominance count by 1 and add B to A’s dominated set.
[0087] ③If A and B do not dominate each other (that is, there is neither a dominating relationship nor a dominated relationship between them), do nothing.
[0088] (3) Determine the first-level non-inferior solutions: Find all individuals whose dominance count is 0 and set them as the first-level non-inferior solutions.
[0089] (4) Recursively construct other levels: Starting from the first level of non-inferior solutions, determine the non-inferior solutions of higher levels layer by layer. For each individual in the non-inferior solution set of the current level, reduce the dominance count of the individuals in their dominated set by 1. If the dominance count of an individual becomes 0, add it to the non-inferior solution set of the next level, until all individuals are assigned to a certain level.
[0090] By quickly sorting non-inferior solutions, NSGA-II can sort the individuals in the population according to the level of non-inferior solutions, ensuring that individuals in each level are not dominated by individuals in other levels. In this way, NSGA-II can maintain the diversity of the population during the evolution process, retain multiple non-inferior solution sets, and thus provide better multi-objective optimization results.
[0091] Crowding sorting measures the distribution density of individuals by calculating their crowding in the objective function space, achieving diversity maintenance and guidance for selecting operations, helping the algorithm to balance diversity and convergence in multi-objective optimization problems and find a better approximate optimal solution set. Individuals with high crowding indicate that there are fewer individuals around them and they have relatively unique positions. Individuals with high crowding are preferred when selecting operations to preserve diversity and avoid individuals gathering in a certain area. The basic steps include sorting individuals by objective function, calculating the crowding of individuals on each objective function, accumulating crowding values, and sorting in descending order.
[0092] Define the population size, maximum number of iterations, crossover coefficient, mutation rate and other parameters according to the problem, use algorithm statements to define the objective function and fitness function, and then embed the parameters and functions into the main program of the algorithm to start the algorithm running. The basic process is as follows: Figure 5 shown.
[0093] When selecting pressure monitoring points by multi-objective optimization method, the constraints include: number of monitoring points 1≤k≤10; monitoring distance of a single monitoring point ≤10 pipe segments; r p,q ≥λ, where r p,q represents the sensitivity of node p to the water pressure change of node q, and λ is the preset sensitivity threshold.
[0094] r p,q The calculation process includes:
[0095] 1) By expanding the node traffic, the sensitivity coefficient of each node to the change of each node traffic is calculated to obtain the sensitivity coefficient matrix [X] n×n ; Where n is the number of nodes; x p,q for [X] n×n The element in the p-th row and q-th column represents the sensitivity coefficient of node q to the traffic change of node p.
[0096] The water pressure and flow between the nodes of the pipeline network affect each other. When the water pressure and flow of a node change, the water pressure and flow of the nodes around the node will also be affected. However, due to the properties and topological structure of each node, the degree of influence is different. In order to quantify the sensitivity of mutual perception between nodes, a node sensitivity matrix is constructed.
[0097] Assume that there are n nodes in the pipe network, p and q are two of the n nodes, the pressure of node p during normal operation is Hp, when the flow changes such as pipe burst and leakage occur at node q, the change flow is ΔQq, and the corresponding pressure change is ΔHq. At this time, the pressure at node p changes due to the flow change at node q, and the change is ΔHp. The ratio of the pressure change ΔHp at node p to the pressure change ΔHq at node q is the influence coefficient of node q on node p.
[0098] The size of the flow change of a certain node will directly affect the perception sensitivity of other nodes. Considering that the ultimate goal is to locate the leakage, the sensitivity coefficient of each node to the flow change of this node is calculated by node flow diffusion, and an n×n node sensitivity coefficient matrix [X] is obtained. n×n .
[0099] 2) When the flow rate of each node is diffused in turn, due to the huge difference in water demand between different nodes, the difference in water demand between nodes will lead to the difference in order of magnitude of the data in the sensitivity coefficient matrix. In order to eliminate this difference, the sensitivity coefficient matrix is standardized to ensure the rationality of the sensitivity matrix result, and the sensitivity coefficient matrix [X′] after standardized processing is obtained. n×n ; Sensitivity coefficient matrix [X] n×n The standardization process is as follows:
[0100]
[0101] In the formula, x p,q ′ is [X′] n×n The element at the pth row and qth column in ; Represents [X] n×n The average value of the elements in the qth column of S q Represents [X] n×n The standard deviation of the elements in the qth column of .
[0102] 3) Standardized sensitivity matrix [X′] n×n The huge difference in magnitude has been eliminated, but the values are positive and negative, which still lacks a certain degree of comparability. In order to achieve this comparability, [X′] is calculated by the following formula: n×n Perform extreme differentiation and map the data to the interval [0,1] to obtain the matrix [X″] after extreme differentiation n×n :
[0103]
[0104] In the formula, x p,q ″ represents the matrix [X″] n×nThe element in row p and column q in p,q ′ represents [X′] n×n The element in row p and column q in qmin ′ represents the minimum element in the qth column of the normalized matrix; x qmax ′ represents the maximum element in the qth column of the normalized matrix;
[0105] 4) Sensitivity matrix [X″] after extreme difference processing n×n Calibrate and find the similarity coefficient between any two nodes. Calculate r by the following formula p,q :
[0106]
[0107] And get the node similarity matrix [R] n×n ; r p,q for [R] n×n The element at the p-th row and q-th column in .
[0108] About Constraints
[0109] The data in the fuzzy similarity matrix reflects the similarity between two nodes, that is, the sensitivity of water pressure changes between nodes, r p,q It represents the sensitivity of node p to the water pressure change of node q. A sensitivity threshold λ is introduced. Only data greater than this threshold is valid. Its expression is as follows:
[0110] r p,q >λ; wherein, 0.8≤λ<1, and λ is 0.8 in this embodiment.
[0111] When the sensitivity is greater than the threshold λ, node p is considered to be sensitive to the water pressure change of node q. Conversely, when the sensitivity is less than or equal to the threshold λ, node p is considered to be insensitive to the water pressure change of node q. From this, the pressure sensitivity range of each node can be known, which can be used as one of the constraints for optimizing site selection.
[0112] In urban water supply networks, the water flow paths between nodes are usually non-unique, that is, there are multiple paths that can connect the same node to the target node. The non-uniqueness of node water flow paths is caused by the combined effect of multiple factors. The complex ring or mesh topology of the water supply network allows multiple paths to connect between nodes. The existence of parallel pipes, branches and intersections of pipes increase the diversity of water flow paths. Water can reach the target node through different branches and intersections. The adjustment and operation of valves and control facilities can change the path of water flow, further increasing the uncertainty of node water flow paths. Therefore, the Dijkstra algorithm in graph theory is used to solve this difference. By calculating the shortest path between any two nodes, the shortest path matrix [D] is obtained. n×n . Introducing distance threshold Data smaller than this threshold is valid, and its expression is as follows:
[0113]
[0114] In specific implementation, the distance threshold There are 10 pipe sections.
[0115] In actual pipe networks, pressure monitoring points are usually arranged in key locations of water supply networks to cover major pipelines, nodes and important areas. Because the monitoring distance of pressure monitoring points is limited, the shortest distance between nodes is used as a constraint. If the shortest distance between nodes exceeds a certain limit, even if the sensitivity of the node meets the condition, the monitoring capability of the node is considered to be invalid.
[0116] Assuming that there are k monitoring points in total, according to the scale and number of nodes of the actual water supply network, combined with the experience of engineering deployment, it can be known that the number of k is limited. In order to ensure the convergence of the algorithm and meet the actual application, at least 1 monitoring point is installed and at most 10 monitoring points are installed, that is:
[0117] 1≤k≤10
[0118] The set of nodes selected as monitoring points is M, then the set of monitoring points is M = (M 1 ,M 2 ,...M k ),use Representative monitoring point M p The set of nodes covered that satisfy the constraints, r o,j Represents the similarity coefficient between node p and each node, d o,j Represents the shortest distance between node p and each node, and J represents the total set (that is, the union of the node sets that meet the constraints covered by all monitoring points), which is expressed as follows:
[0119]
[0120] About the objective function
[0121] Before running the multi-objective optimization algorithm, it is necessary to determine the objective function and quantify the objective function. When optimizing the site selection of pressure monitoring points, factors such as coverage, water volume, leakage perception, and monitoring cost are usually considered as single objectives. This method uses the three objectives of monitoring range, water volume, and monitoring cost as optimization indicators for multi-objective optimization to solve the monitoring range and water volume benefits under economic constraints.
[0122] Monitoring range
[0123] By r p,q >λ, The monitoring range of each node can be obtained. When there are multiple monitoring points in the pipeline network, the total monitoring range of the pipeline network is not the direct addition of the monitoring ranges of each monitoring point, because the monitoring ranges of the monitoring points may overlap. Therefore, it is expressed by the following formula.
[0124]
[0125] Among them, x i is the node number node, j i For x i Node monitoring range, It is the union of nodes covered by all monitoring points, that is, the monitoring range.
[0126] Covering water
[0127] Each node has a corresponding node water demand. After the monitoring range is determined, the coverage water volume is equal to the sum of the water demand of all nodes within the monitoring range. For example, if the monitoring range is [1, 2, 3, 5, 111, 194, 227], the coverage water volume is the sum of the water demand of nodes 1, 2, 3, 5, 111, 194, and 227, which is expressed as the following formula.
[0128]
[0129] Among them, y i Also the node number, s i for y i The water demand of the node, is the sum of the water demand of all nodes within the monitoring range, that is, the coverage water volume. Taking the monitoring range [1, 2, 3, 5, 111, 194, 227] as an example, its corresponding y 1 ,y 2 ,y 3 ,y 5 ,y 11i1 ,y 194 ,y227 Take 1, the rest y i Take 0 and multiply it by the corresponding node water demand s i , and obtain the total covering water volume S.
[0130] Monitoring cost (number of monitoring points)
[0131] From the expressions of the two objectives of detection range and water volume coverage, it can be seen that the more monitoring points there are, the larger the monitoring range and water volume covered. This is obviously inconsistent with the basic logic of multi-objective optimization and the actual engineering needs. In actual engineering applications, it is often necessary to consider the constraints of economic conditions and use as few monitoring points as possible to achieve overall monitoring of the pipeline network. Therefore, the monitoring cost is introduced as one of the objective functions of multi-objective optimization. However, in actual engineering, it is often complicated to analyze the cost of monitoring points in detail. In simple terms, it can be divided into construction and installation costs, equipment and material procurement, maintenance technical service fees, etc. Although the cost structure is relatively complex, in actual engineering, construction units often use the form of total price or unit price lump sum, that is, the cost caused by the construction difficulty between monitoring points is not considered, and the price of raw materials purchased by the same manufacturer is relatively uniform, so the number of monitoring points is equivalent to the cost. The expression is as follows.
[0132]
[0133] Among them, k is a constraint condition, which constrains the number of monitoring points within a certain range and needs to be reasonably set according to the size of the actual pipeline network and to ensure the efficiency of the algorithm operation.
[0134] In summary, the objective function form of the three-objective optimization is established. Since the NSGA-Ⅱ algorithm is essentially a maximum or minimum optimization, the three objectives are set as: Z 1 The number of monitoring points is the least; Z 2 The unmonitored range is the smallest, that is, the proportion of the number of unmonitored nodes to the total number of nodes; Z 3 The amount of uncovered water is the smallest and the expression is as follows.
[0135]
[0136] For ease of understanding, the above method is used to select monitoring points in the Z area.
[0137] The water supply network in District Z serves an area of about 50 square kilometers, which is divided into high, medium and low water supply areas. The scope of this method is the high-area network. The water use types are residential and commercial, and there is no industrial water use. The total population of the area is about 87,000, and their water is all transferred and supplied by a booster pump station. The booster pump station has a water supply transfer and lifting capacity of 20,000 tons / day. The basic information of the remaining pump stations has been introduced in detail when the hydraulic model was built in Section 2.3 of this article. Since there are no pressure monitoring points in this area, the water supply pressure is unbalanced and pipes frequently burst in the network. In order to improve its water supply status, the scientific nature of the pressure monitoring point site selection method is also tested.
[0138] The hydraulic model of the water supply network in Zone Z has been established in S1. According to the steps of building the pressure monitoring point site selection model, the diffuser coefficients are set for the 228 nodes of the network in turn. After many tests, the diffusion coefficient of each node is 3L / s / m 1 / 2 ,The nodes of the water supply network in the hydraulic model use the same urban daily water consumption curve. When analyzing the pressure change, the average pressure is taken as the initial water pressure. The data are shown in Table 3.
[0139] Table 3 Initial water pressure at some nodes
[0140]
[0141] After diffusing 228 nodes in turn, the corresponding pressure change matrix is obtained. The pressure change information of some nodes is shown in Table 4.
[0142] Table 4 Pressure variation matrix of some nodes
[0143]
[0144] The second column of data in the table represents the water pressure of each node at 0 o'clock on the highest day, including nodes 2, 3, 4, 51, 86, 114, 141, 216, 227 and 228. The third column represents the setting of diffusion coefficient 3L / s / m at node 2. 1 / 2 The water pressure change data of some nodes are obtained in the same way to obtain a 228×228 pressure change matrix.
[0145] The pressure change matrix can be used to calculate the pressure change Δh of each node in the pipeline network when the diffusion is calculated in turn at each node, and then the pressure difference matrix of the node water pressure change can be obtained, as shown in Table 5.
[0146] Table 5 Pressure difference matrix of some nodes
[0147]
[0148] It can be seen from Table 5 that the same diffusion coefficient has different effects on the pressure of different nodes. This is because the diffusion coefficient is related to the initial pressure of the node. Different initial pressures of the nodes will result in different pressure changes. It is necessary to standardize and extreme the node sensitivity matrix to eliminate the order of magnitude differences in the diffusion flow between different nodes, so that the pressure change degree is more comparable.
[0149] First, the sensitivity value of each node is calculated according to the pressure difference matrix to obtain the node sensitivity matrix, as shown in Table 6.
[0150] Table 6 Pressure sensitivity matrix of some nodes
[0151]
[0152] The node sensitivity matrix is then standardized and extremeized, and the results are shown in Tables 7 and 8.
[0153] Table 7 Partial node standardization-pressure sensitivity matrix
[0154]
[0155] Table 8: Partial node extreme difference-pressure sensitivity matrix
[0156]
[0157] After standardization and extreme differentiation, the node sensitivity matrix has eliminated the quantitative differences and compressed them all into the [0,1] interval. Finally, the Euclidean distance method is used for calibration, and the fuzzy similarity coefficient of the node is calculated to obtain the final node fuzzy similarity matrix, as shown in Table 9.
[0158] Table 9 Fuzzy similarity matrix of some nodes
[0159]
[0160] The second column of data in the table represents the similarity between the water pressure of node 2 and other nodes. The elements on the diagonal are all 1, which means that the fuzzy similarity coefficient of each node with itself is 1, that is, the water pressure is completely correlated. After repeated calculations and verifications in the complete node fuzzy similarity matrix and combined with engineering experience, in order to improve the accuracy of subsequent leakage positioning, when the node fuzzy coefficient reaches above 0.8, it is considered that the water pressure between nodes is correlated, that is, the sensitivity threshold λ in the constraint condition is taken as 0.8.
[0161] Another constraint is that the shortest path between nodes should be less than the distance threshold The normal hydraulic operation is simulated in Epanet2.2, and the Dijkstra algorithm is used to count the shortest paths of water flow between nodes to form a distance matrix. Part of the matrix is shown in Table 10.
[0162] Table 10 Partial node distance matrix
[0163]
[0164] The second column of the table shows the shortest flow path from node 1 to each node. By analyzing the table, we can find that the distance threshold It has a certain impact on the monitoring range of the node. In order to meet the constraint that the number of monitoring points is within 10 to achieve the monitoring of the entire pipeline network and ensure the scientific nature of the monitoring range, after many tests, the distance threshold is determined. Take 10.
[0165] After repeated verification and validation in the algorithm model, λ = 0.8 was determined after multiple optimizations. (The farthest monitoring distance of a single monitoring point is the distance of 10 pipe segments), 1≤k≤10 three constraints, use matlab platform to call NSGA-Ⅱ algorithm and Epanet matrix data, write the algorithm code of target function target1, target2, target3, and then set the main parameters of the algorithm in turn, set the initial population size to 50, mutation rate to 0.4, crossover ratio to 0.1, and maximum number of iterations to 1000. Since the algorithm is different from the uniqueness of the results of mathematical methods in optimization, the Pareto frontier output after each operation is different. After several algorithm optimizations, the algorithm results are compared, and the optimization solution set with a higher frequency is retained as the optimization solution set for pressure monitoring point location. The final algorithm results are as follows: Figure 6 As shown, it can be seen that the algorithm cannot achieve the optimal solution in terms of the number of monitoring points, unmonitored range, and uncovered water volume. Further analysis is needed to select the pressure monitoring point layout scheme in the solution set.
[0166] The scattered points on the Pareto front contain one or more monitoring point layout schemes. The scheme with a higher node appearance frequency is selected to obtain the monitoring point layout schemes with different numbers as shown in Table 11.
[0167] Table 11 Monitoring program set
[0168]
[0169] The algorithm shows that the schemes with the largest monitoring range and the largest water volume coverage under different numbers are different. The results of the algorithm are converted into a scatter plot of monitoring range and water volume coverage. It can be seen that with the increase in the number of monitoring points, the monitoring range and water volume coverage are increasing, but the growth rate is gradually slowing down. By analyzing the monitoring range and water volume coverage growth rate curve, it can be found that after the number of monitoring points increases to 6, the monitoring range growth rate is only 0.44% for each additional monitoring point, and the monitoring range benefit is not high; when the number of monitoring points increases to 5 and 6, the monitoring range growth rate is 0.88%, but the coverage water volume growth rate is 1.98% when the number of monitoring points is 6, which is higher than the coverage water volume growth rate of 0.61% when the number of monitoring points is 5, and when the number of monitoring points is greater than 6, the corresponding coverage water volume growth rate decreases sharply, indicating that the coverage water volume benefit of increasing the monitoring points is not high; therefore, the algorithm scheme with 6 monitoring points, 98.68% monitoring coverage and 237.071L / S coverage water volume is finally adopted as the pressure monitoring point site selection optimization scheme.
[0170] According to the final pressure monitoring point location optimization plan, the corresponding results when the number of corresponding monitoring points in the algorithm is 6 are obtained, and the nodes finally selected as monitoring points are determined to be: node 2, node 51, node 86, node 114, node 141, node 216, and the monitoring points are numbered 1-6 in sequence. The monitoring point numbers are shown in Table 12, and the positions in the pipe network are shown in Figure 7 shown.
[0171] Table 12 Node corresponding monitoring point number
[0172]
[0173] S3. Based on the hydraulic model of the water supply network, a leakage hydraulic state simulation is performed to simulate various leakage conditions of each node; and the 24-hour pressure waveform data of each pressure monitoring point under each leakage condition is recorded as a leakage data set; and the data in the leakage data set is divided into a training set and a test set by a random function; the training set is used to train the intelligent monitoring model, and the test set is used to test the training effect of the intelligent monitoring model; the data set in the leakage data is also preprocessed, and the preprocessing includes data normalization.
[0174] S4. Use the leakage data set to train the preset intelligent monitoring model.
[0175] Among them, the intelligent monitoring model is an intelligent model based on a random forest algorithm, a radial basis function network in a neural network, a convolutional neural network in a deep learning algorithm, or a long short-term memory neural network. In this embodiment, the intelligent monitoring model is an intelligent model based on a random forest algorithm. Experiments have shown that compared with other intelligent models, the use of an intelligent model based on a random forest algorithm has a very high accuracy rate in locating leakage points. Deviations only occur when the leakage amount is small, and the deviation points are all near the actual leakage point, which can still greatly reduce the scope of investigation.
[0176] For a better understanding, the Z region is still taken as an example for explanation.
[0177] For the actual pipe network, leakage may occur at any position of any pipe in the pipe network. However, compared with the middle of the pipe, the node is more prone to loose interfaces, aging of seals and other problems, which may lead to water leakage. At the same time, the hydraulic model software is also easier to realize the node leakage condition, so only the leakage location research is carried out for the node leakage. The hydraulic model of the water supply network in the Z area established by S1 is used to simulate the hydraulic state of leakage, and the node ejector is used to simulate the leakage condition. After repeated simulation and adjustment, the leakage coefficient of each node is [0,4], with a gradient of 0.05. A single node has a total of 80 leakage flow rates. The relationship between the leakage coefficient and the leakage flow rate is shown in the following formula:
[0178] q=Cp γ ;
[0179] Wherein, q is the leakage flow rate L / S; p is the node pressure m; C is the leakage coefficient; γ is the pressure index, and for the nozzle, γ is 0.5.
[0180] The leakage simulation was carried out for 228 nodes of the pipe network in turn, with a total of 228*80=18240 leakage conditions. Taking node 150 with a leakage coefficient of 0.5 as an example, its original water demand is 1.53L / s, the node pressure is 41.86m, and its leakage flow is calculated as 3.23L / s according to the above formula. The node water demand after leakage is 4.76L / s.
[0181] Theoretically, as long as the pressure data of all nodes under each working condition are known, a one-to-one correspondence can be established. This can be achieved in the hydraulic model, but it has no practical application value, because in the actual pipe network, only the pressure value of the pressure monitoring point is known. Therefore, in the model, only the pressure of the pressure monitoring point is obtained. The same pressure monitoring points as the actual water supply pipe network in Zone Z are used to record the 24-hour pressure waveform data of 6 pressure monitoring points under each leakage condition. A total of 228*80*6*24=2626560 pressure data constitute the total pressure data set. The leakage data is divided into training set and test set by random function in the algorithm. According to the ratio of 7:3, 70% of the data is randomly selected for training and 30% of the data is used for testing.
[0182] In the leakage data set, the pressure data corresponding to each working condition contains leakage information. If these pressure data are analyzed manually, it will take geometric times of time and effort. Moreover, when new leakage locations and leakage flows appear, the timeliness and accuracy of manual analysis are obviously unable to meet the basic requirements of leakage location. Therefore, machine learning and deep learning methods are used to learn leakage characteristics, and then the regression of leakage pressure data is realized through prediction.
[0183] In principle, machine learning is to let machine learning algorithms learn patterns and rules from data in order to make predictions or decisions on new, unseen data. In the classification and regression problem, the ensemble learning algorithm and neural network algorithm in machine learning algorithms have their own excellent performance. This method uses the random forest (RF) in the ensemble algorithm, the radial basis function network (RBF) in the neural network, and the convolutional neural network (CNN) and long short-term memory neural network (LSTM) in the deep learning algorithm.
[0184] RBF neural network is a single hidden layer neural network in the three-layer perception system. It consists of an input layer, a hidden layer and an output layer. The output of the input layer is the input of the hidden layer, and the input of the output layer is the output of the hidden layer. The weight product between the two determines the final output result. RBF neural network has excellent generalization ability. Its simple structure can effectively avoid complex calculations, thereby improving the efficiency and performance of the network. The research results show that RBF neural network can effectively approximate any nonlinear function under compact set and arbitrary precision conditions, thereby achieving the expected results.
[0185] The RF algorithm shows high accuracy and generalization ability when processing large-scale data sets and a large number of features. It is one of the most widely used machine learning algorithms. The implementation of the algorithm mainly involves three aspects: decision tree, Bootstraping, and Bagging.
[0186] CNN is a deep learning model widely used in computer vision and image processing. Its design is inspired by the working principle of the visual cortex in biology, especially the concept of receptive field and local connection in the visual cortex. CNN belongs to the feedforward neural network in the neural network. Its core idea is to construct nonlinear mapping through the connection weights between layers. Each neuron receives input from the neurons in the previous layer, and performs nonlinear transformation on the input through the activation function, and then passes the result to the neurons in the next layer. At the same time, each neuron is connected to all neurons in the previous layer. Each connection has an associated weight. These weights determine the propagation and weighting of the input signal in the network. During the training process, the weights are adjusted layer by layer according to the gradient of the loss function through the back-propagation algorithm, so that the output of the network can approach the expected target.
[0187] Long short-term memory neural network is a variant structure of recurrent neural network (RNN). Based on the fact that RNN only has short-term memory ability, LSTM has a special memory method when processing time series data. It has the ability to retain a large gradient memory and can also determine whether the memory gradient exceeds the set threshold, thereby improving the RNN gradient disappearance and gradient explosion problems.
[0188] Data processing method settings
[0189] The pressure data in the original leakage database has differences in gradient. In order to make the algorithm perform better, it is necessary to use two methods, data normalization and data tiling, in the algorithm. Data tiling is to simplify the input structure of deep learning models CNN and LSTM. The specific contents are as follows:
[0190] 1) Data normalization
[0191] Data normalization is an important data preprocessing technique that can eliminate the dimensional differences between different features and improve the accuracy of model training and evaluation. Commonly used data normalization methods are:
[0192] Min-max normalization: linearly maps the data to between the specified minimum and maximum values.
[0193]
[0194] Among them, x is the original data and y is the normalized data.
[0195] Standardization: Transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.
[0196]
[0197] Among them, x is the original data and z is the standardized data.
[0198] Decimal scaling normalization: Map the data to the range [-1, 1] by moving the decimal point.
[0199]
[0200] Here, x is the original data, and i is an integer whose maximum absolute value is less than 1.
[0201] For the pressure data, the mapminmax function is uniformly used to perform minimum and maximum normalization on the input data of the training set and the test set.
[0202] 2) Data Tiling
[0203] Data tiling is used to convert multidimensional data structures into one-dimensional data structures, simplifying the representation and processing of data to make it suitable for the input of neural network models. The tiling process mainly consists of the following two steps:
[0204] ① Multidimensional data conversion: The first step of data tiling is to convert multidimensional data into one-dimensional data. For two-dimensional data, flatten the data into one-dimensional vectors by row or column. For higher-dimensional data, appropriate methods can be used to convert it into a one-dimensional representation.
[0205] ② Index mapping: During the data flattening process, the index information of each element in the original data needs to be recorded so that the data structure can be restored. Usually, a mapping relationship is used to correspond the index in the one-dimensional data to the index in the original data.
[0206] Basic algorithm parameter settings
[0207] After multiple trials and adjustments, the basic parameter settings of the four algorithms are shown in Table 13.
[0208] Table 13 Basic parameters of positioning module
[0209]
[0210] Among them, the Adam gradient descent algorithm optimizes network parameters by minimizing the loss function, so that the network can better fit the training data and make predictions. The Adam gradient descent algorithm combines the characteristics of momentum gradient descent and adaptive learning rate, and can adaptively adjust the learning rate during the training process. In each iteration, the learning rate is adjusted by calculating the first-order moment estimate (mean) and second-order moment estimate (variance) of the gradient, and then the parameters are updated by maintaining the momentum and adaptive learning rate of each model parameter.
[0211] The inventors compared the four models, and the performance comparison in leakage location and leakage coefficient regression is shown in Table 14.
[0212] Table 14 Algorithm performance index comparison
[0213]
[0214] Among them, R 2 is the coefficient of determination. The closer its value is to 1, the higher the reference value of the model's prediction results. MAE is the mean absolute error; RMSE is the root mean square error.
[0215] In terms of leakage location performance, RF>CNN>LSTM>RBF. The excellent performance of the RF algorithm lies in its retention of features. The 144 pressure data and leakage coefficients of 6 monitoring points for 24 hours were analyzed and ranked by importance, which significantly improved its classification and regression capabilities. As a simple neural network algorithm, RBF only relies on nonlinear fitting capabilities when locating nodes, and has poor learning capabilities for features. CNN and LSTM extract features through multi-layer neural networks, which improves the positioning accuracy to a certain extent. However, due to the high similarity of pressure data in the leakage data set, the algorithm is prone to extract common features or fail to extract key features when extracting features, resulting in the overall performance of the two deep learning algorithms not being as good as RF.
[0216] In terms of leakage coefficient regression performance, RF>RBF>CNN>LSTM. Overall, the four algorithms have no obvious deviation values when regressing the leakage coefficient, indicating that the pressure change characteristics under different leakage amounts are easier to be learned by the algorithm. RBF has a higher fitting ability than CNN and LSTM when regressing the leakage coefficient, indicating that deep learning networks are not necessarily suitable for the current data set.
[0217] RF performs better than the other three algorithms in leakage location and leakage coefficient regression. Therefore, in this embodiment, the built intelligent monitoring model based on the RF algorithm is used to perform subsequent analysis on the leakage of the engineering example.
[0218] S5. In the area to be tested of the actual pipe network, pressure monitoring points are arranged according to the pressure monitoring scheme obtained in S2, and real-time data of the pressure monitoring points are collected and input into the intelligent monitoring model trained in S4 to monitor water leakage in the pipe network.
[0219] In order to better understand the effect of this method, an example is given based on the above content of the Z area and the intelligent monitoring model based on the RF algorithm.
[0220] In order to test its performance in actual application, the water supply network in Zone Z was analyzed after arranging 6 monitoring points. After the monitoring points were installed, the SCADA system in the area could realize online pressure monitoring, but the existence of leakage still depended on manual investigation. According to the leakage repair records of the dispatching center, it was found that in the actual operation of the network in the area, the locations with frequent leakage and pipe burst (large flow leakage) were concentrated in Figure 8 In the red area, according to statistics, before the pressure monitoring points were installed, the average annual frequency of pipe bursts and leakages in this area was as high as more than 100 times, and there was no way to investigate the undetected small flow leakage; after the pressure monitoring points were installed, the maintenance personnel conducted regular leakage maintenance inspections on the frequent pipe burst points. There were 21 leakages (excluding pipe bursts) with maintenance records. Taking the maintenance record on September 12, 2023 as an example, there was leakage in this area on that day, and the leakage location was Figure 8 In the leakage area on the right, there is no obvious water seepage on the ground, and the leakage amount is not recorded. Due to the lag in leakage inspection, the pressure data of 24 hours before the leakage record is retrieved from the SCADA system, and the pressure data of the monitoring point is input into the algorithm to verify the positioning accuracy of the algorithm and provide a certain reference value for the leakage amount.
[0221] The leakage records were sorted out, and the leakage of the pipe section was mapped to the node closest to the leakage point. Among the 21 leakage records, 18 leakage records were screened as single leakage points. The leakage times of each actual leakage node are shown in Table 15.
[0222] Table 15 Leakage information record
[0223]
[0224] The pressure data of the six pressure monitoring points corresponding to the 18 leakage records 24 hours before the leakage was recorded are retrieved from the SCADA system. Taking the leakage record on September 12, 2023 as an example, the leakage area is Figure 8 In the leakage area on the right, the node number closest to the leakage point is 142. The 24-hour pressure data of the six pressure monitoring points on September 11, 2023, the day before the leakage was discovered, are retrieved from the SCADA system, as shown in Table 16.
[0225] Table 16 SCADA pressure data at node 142 leakage
[0226]
[0227] The pressure data corresponding to the 18 leakage records were exported in turn, and the 18 sets of data were input into the RF algorithm. The algorithm results are shown in Table 17.
[0228] Table 17 Leakage verification algorithm results
[0229]
[0230] It can be seen from Table 17 that, in the 18 leakage locations, 15 leakage locations were successfully located, and the positioning accuracy rate reached 83.33%. Nodes 142, 146, and 156 were located incorrectly once respectively. By comparing their leakage amounts, the leakage amounts were relatively small when they were located incorrectly. This shows that when the leakage amount is small, the pressure characteristics of the data cannot be accurately learned by the algorithm, resulting in positioning deviation.
[0231] By analyzing the locations of the false alarm nodes, it is found that they are basically located near the actual leakage points. The location relationship is as follows: Fig. 9 As shown, the error between the false alarm node and the actual leakage node is within two pipe sections, which reduces the scope of leakage detection for the actual leakage.
[0232] In summary, when this method is used to locate actual pipe network leakage, there is a certain error in locating smaller leakage, but the overall accuracy can reach 83.33%, indicating that the algorithm model has good performance and certain anti-noise ability. When the actual pipe network is running, the leakage area can be locked by the algorithm first, and then combined with physical means to greatly reduce the cost of manual leak detection. At the same time, the leakage output by the model can be used as a reference value to provide certain guidance for water supply personnel and water supply companies to grasp leakage information.
[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.
Claims
1. A method for monitoring leakage in a city water supply network, characterized in that: The following steps are involved: S1. Obtain the water supply network data of the area to be monitored and construct a corresponding water supply network hydraulic model to achieve mapping of the actual network topology structure; S2. Based on the hydraulic model of the water supply network, pressure monitoring points are selected in each node of the water supply network through a multi-objective optimization method to obtain a pressure monitoring plan; wherein the optimization objectives include monitoring range, water volume covered and monitoring cost; S3. Based on the hydraulic model of the water supply network, the hydraulic state of leakage is simulated to simulate various leakage conditions of each node; and the 24-hour pressure waveform data of each pressure monitoring point under each leakage condition is recorded as the leakage data set; S4. Use the leakage data set to train the preset intelligent monitoring model; S5. In the area to be tested of the actual pipe network, pressure monitoring points are arranged according to the pressure monitoring scheme obtained in S2, and real-time data of the pressure monitoring points are collected and input into the intelligent monitoring model trained in S4 to monitor water leakage in the pipe network.
2. The method for monitoring leakage of a city water supply network according to claim 1, characterized in that: In S2, when the pressure monitoring points are selected by the multi-objective optimization method, the optimization objectives are converted into three objective functions: the minimum unmonitored range, the minimum number of monitoring points, and the minimum uncovered water volume; and the pressure monitoring points are selected by the NSGA-Ⅱ algorithm.
3. The method for monitoring leakage of a city water supply network according to claim 1, characterized in that: In S2, when selecting pressure monitoring points by multi-objective optimization method, the constraints include: number of monitoring points 1≤k≤10; monitoring distance of a single monitoring point ≤10 pipe segments; r p,q ≥λ, where r p,q represents the sensitivity of node p to the water pressure change of node q, and λ is the preset sensitivity threshold.
4. The urban water supply network leakage monitoring method according to claim 3, characterized in that: r p,q The calculation process includes: First, the sensitivity coefficient of each node to the change of each node traffic is calculated by node traffic expansion, and the sensitivity coefficient matrix [X] is obtained. n×n ; Where n is the number of nodes; x p,q for [X] n×n The element in the pth row and qth column represents the sensitivity coefficient of node q to the flow change of node p; After that, the sensitivity coefficient matrix [X] n×n Perform standardization to obtain the standardized sensitivity coefficient matrix [X′] n×n ; In the formula, x o,q ′ is [X′] n×n The element at the pth row and qth column in ; Represents [X] n×n The average value of the elements in the qth column of S q Represents [X] n×n The standard deviation of the elements in the qth column of ; Then, [X′] is obtained by the following formula n×n Perform extreme differentiation and map the data to the interval [0,1] to obtain the matrix [X″] after extreme differentiation n×n : In the formula, x o,q ″ represents the matrix [X″] n×n The element in row p and column q in o,q ′ Represents [X′] n×n In the pth row, the qth The element of the column; x qmin represents the minimum element in the qth column of the normalized matrix; x qmax Represents the maximum element in the qth column of the normalized matrix; Finally, r is calculated by p,q : And get the node similarity matrix [R] n×n ; r p,q for [R] n×n The element at the p-th row and q-th column in .
5. The urban water supply network leakage monitoring method according to claim 3, characterized in that: 0.8≤λ<1.
6. The urban water supply network leakage monitoring method according to claim 1, characterized in that: In S3, a random function is also used to divide the data in the leakage data set into a training set and a test set; the training set is used to train the intelligent monitoring model, and the test set is used to test the training effect of the intelligent monitoring model; The data sets in the leaky data are also preprocessed, including data normalization.
7. The method for monitoring leakage of a city water supply network according to claim 1, characterized in that: In S4, the intelligent monitoring model is an intelligent model based on a random forest algorithm, a radial basis function network in a neural network, a convolutional neural network in a deep learning algorithm, or a long short-term memory neural network.
8. The method for monitoring leakage of a city water supply network according to claim 7, characterized in that: The intelligent monitoring model is an intelligent model based on the random forest algorithm.
Citation Information
Patent Citations
Optimum layout method aiming at pressure monitoring points of water supply pipe network
CN111119282A
Water supply pipe network hidden leakage recognition method based on dynamic process
CN111853555A
Accurate identification method based on deep neural network for pipe explosion pipeline of water supply network
CN112097126A
Cited By
Water supply network leakage point positioning method and system, storage medium and computer
CN119337753A
Urban pipe network monitoring optimization method and system based on agent map and medium
CN119623777A
Method and equipment for generating characteristic curve of centrifugal water pump and medium
CN119830740A
Method, device and system for detecting and positioning leakage of multiple pipelines of water supply network
CN120368231A
Intelligent city water supply pipe network emergency supervision Internet of Things large model system and method
CN120802783A
