Multi-target arrangement method of leakage detection sensors in water supply network based on leakage probability
Through a multi-objective optimization method based on leakage probability, the sensor layout of the water supply pipeline network is optimized, and the problems of insufficient leakage detection performance and information redundancy in the existing technology are solved, and priority detection of leakage events with greater impact are achieved and Minimize impact.
Patent Information
- Application Number
- CN202111072949.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-14
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-09-14
AI Technical Summary
The prior art failed to effectively consider the change in the leakage probability of each node in the water supply pipeline leakage detection, which made it difficult to detect all leakage events under the limited number of sensors, especially ignoring that it may lead to larger leakage. and failing to fully consider the impact of hydraulic model modeling and sensor measurement errors, resulting in insufficient information redundancy and detection performance.
Using a multi-objective optimization method based on leakage probability, simulating the impact of leakage in each node, calculating the leakage probability of each node, using NSGA-II and PROMETHEE for Pareto front-line solutions to optimize the sensor layout scheme , to ensure priority detection of leakage events with greater impact, and to consider the impact of hydraulic model and sensor measurement errors.
When the number of sensors is limited, priority is given to detecting leakage events that have a greater impact on the water supply pipeline network, reducing information redundancy, improving detection performance, ensuring the minimum impact of leakage in each node of the pipeline network, providing a more reasonable sensor optimization layout.
Smart Images

Figure CN113887020B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an optimized arrangement of sensors for a city water supply network, and in particular to a multi-target arrangement method of leakage detection sensors for a water supply network based on leakage probability. Background Art
[0002] Water supply network leakage will cause a large amount of water waste and have a significant impact on residents' lives, so leakage needs to be detected quickly. Based on the hydraulic model of the network and the sensor network, leakage can be effectively detected and located, but the leakage detection performance is closely related to the number and location of sensors. Considering the installation and maintenance costs of sensors, it is impossible to arrange sensors at every node in the network. Therefore, the optimal arrangement of sensors has been widely studied.
[0003] The optimal placement of sensors involves the selection of sensor locations and quantities, ensuring that the detection performance of the sensor network is optimal when the number of sensors is limited. The optimal placement of sensors is a multi-objective optimization problem. In addition to the number of sensors, the detection time, the number of leakage event detections, the coverage of the sensor network, etc. must also be considered. Most methods usually use NSGA-II for solution, which selects a set of Pareto front solutions based on various optimization objectives. Since the Pareto front solution usually contains a large number of solutions, even if these solutions are all optimal, only one of them can be selected, and it is difficult for decision makers to make a choice. Multi-criteria decision analysis can sort the Pareto front solutions according to conflicting objective functions, thereby assisting decision makers in obtaining the optimal solution. It has been widely used in the optimal placement of sensors, such as PROMETHEE, TOPSIS, and ELECTRE. PROMETHEE has the characteristics of being easy to use, easy to understand, and strong in handling uncertainty. It can select the optimal solution from the Pareto front solution according to the preference function. The present invention uses PROMETHEE to sort and cluster the Pareto front solutions.
[0004] In addition to improving the detection performance of the sensor network, it is also necessary to avoid information redundancy in the sensor network as much as possible. The sensor location is usually selected at the node that is more sensitive to the pressure change in the pipeline network, which is called maximizing the sensitivity of the sensor network. However, some high-sensitivity points in the pipeline network are usually distributed in a relatively concentrated area, which leads to information redundancy in the sensor network. In order to obtain the optimal sensor layout plan, it is necessary to ensure sufficient spatial coverage and maximum information of the sensor network at the same time, that is, to provide the most information through the least number of sensors. Some methods introduce information entropy, aiming to select sensor locations that can provide high information value. Although these methods use information entropy to select the location of the sensor, the influence of hydraulic modeling and sensor measurement errors is not considered in the process of optimizing the sensor layout. In fact, the spatial distribution of modeling and measurement errors has an important influence on the selection of sensor locations.
[0005] The inventors of the present application have found through research in the process of realizing the present invention that most of the above methods assume that each node of the water supply network has the same importance, that is, the change of the leakage probability of each node is not considered. This makes it difficult to complete the detection of all leakage events when the number of sensors is limited, especially the nodes that may cause larger leakage need to be given priority. In order to achieve the priority detection of leakage events with larger leakage, some methods arrange sensors based on leakage risk, and leakage events with larger leakage have a higher probability of leakage. Although this achieves the distinction between different leakage events, it only considers the single goal of smaller leakage. In fact, in addition to causing a large amount of water loss, leakage will also affect other nodes and pipelines in the water supply network, including node pressure drop, insufficient water supply, pipeline pressure drop and flow rate increase. The node pressure drop will affect the water supply, while the pipeline pressure drop and flow rate increase will affect the water quality of the network. In the pipeline network, the impact of leakage on the water supply network is different when each node leaks. Obviously, when the sensor network selects leakage events, leakage events with higher impact on the pipeline network should have a higher probability (weight) to be detected first. Summary of the invention
[0006] In view of the above-mentioned problems existing in the prior art, the present invention provides a multi-target layout method for leakage detection sensors in a water supply network based on leakage probability, which takes into account the changes in leakage at each node and obtains the optimal layout scheme of various leakage probability sensors based on the changes in leakage probability, so as to obtain the optimal layout scheme of sensors with the least impact of leakage on the water supply network.
[0007] To achieve the above purpose, the present invention adopts the following steps:
[0008] A multi-target arrangement method of leakage detection sensors for a water supply network based on leakage probability comprises the following steps:
[0009] Step 1, determine the impact of leakage at each node of the pipeline network on the pipeline network, and obtain the leakage probability of each node of the pipeline network;
[0010] Step 2: Simulate various leakage situations to obtain a leakage detection matrix;
[0011] Step 3: Use the multi-objective optimization model NSAG-II and the leakage detection matrix to solve the Pareto front solutions under different leakage probabilities, and use PROMETHEE to sort and cluster the Pareto front solutions to obtain the optimal solutions under various leakage probabilities;
[0012] Step 4: Compare and analyze various solutions based on the impact of leakage at each node on the pipeline network.
[0013] Furthermore, the step 1 specifically includes:
[0014] Step 1.1: Simulate the leakage at each node of the pipeline network to obtain the impact of leakage at each node on the pipeline network:
[0015] Step 1.2: According to the impact of leakage at each node of the pipeline network on the pipeline network, the leakage probability of each node of the pipeline network is obtained.
[0016] Furthermore, the step 1.1 is specifically as follows:
[0017] Assuming that leakage occurs at the network node, the leakage flow is added to each leakage node as the additional water demand of the node, as shown in formula (1):
[0018]
[0019] Where Dnew n,i represents the node water demand at time i when node n leaks, Din n,i The node water demand at time i when node n has no leakage, γ is the pressure index, c is the leakage coefficient, and p is the pressure;
[0020] After simulating the leakage at each node of the pipeline network according to formula (1), the normal working condition of the pipeline network is simulated to obtain the influence of leakage at each node of the pipeline network on the pressure of each node of the pipeline network, the pressure of each pipeline of the pipeline network and the flow rate, as shown in formula (2-4):
[0021]
[0022]
[0023]
[0024] In the formula, γ1 t (j,i),γ2 t (p,i) and γ3 t (p,i) respectively represent the impact of leakage at node i on the pressure of node j, pressure of pipeline p and flow rate at time t, i,j=1,…,N n ,N n is the total number of network nodes, p=1,…,P, P is the total number of network pipes. If leakage occurs at node i, the pressure drop at node j (pipeline p) exceeds 20%, then γ1 t (j,i)=1(γ2 t (p,i)=1), otherwise, γ1 t (j,i)=0(γ2 t (p,i)=0), if the flow velocity at pipe p increases by more than 20%, then γ3t (p,i)=1, otherwise γ3 t (p,i)=0, t=Δt, 2Δt,…, AΔt, Δt is the hydraulic simulation step, and AΔt is the total simulation time.
[0025] Furthermore, four leakage probabilities are obtained in step 1.2: a first leakage probability (P1) is obtained based on the amount of node pressure drop, a second leakage probability (P2) is obtained based on the impact on water demand, a third leakage probability (P3) is obtained based on the pipeline pressure drop, and a fourth leakage probability (P4) is obtained based on the pipeline velocity increase, and a combined leakage probability (P5) is obtained based on the four leakage probabilities.
[0026] Furthermore, the four leakage probabilities and the combined leakage probability are obtained by the following steps:
[0027] The first leakage probability P1 is obtained based on the number of pressure drop nodes:
[0028]
[0029] Where P1(i) represents the probability of leakage at the pipeline network node i based on the number of pressure drop nodes, j = 1, 2, ..., N n ;
[0030] The second leakage probability P2 is obtained based on the impact on water demand:
[0031]
[0032] Where P2(i) represents the probability of leakage at the network node i based on the water demand, and D jt represents the water demand of node j at time t, Represents the total water demand of the water supply network during the entire simulation period;
[0033] The third leakage probability P3 is obtained based on the pipeline pressure drop:
[0034]
[0035] Where P3(i) represents the probability of leakage at the pipeline node i based on the pipeline pressure drop; L P is the length of pipeline p, p=1,2,…,P; the more pipeline pressure drops caused by leakage at node i, and the longer these pipelines are, the higher the probability of leakage at node i;
[0036] The fourth leakage probability P4 is obtained based on the increase in pipeline flow rate:
[0037]
[0038] Where P4(i) represents the probability of leakage at the network node i based on the increase in pipeline velocity. The more the number of pipeline velocities increase when leakage occurs at node i, and the longer these pipelines are, the higher the probability of leakage at node i.
[0039] Based on the joint leakage probability, the fifth leakage probability P5 is obtained:
[0040] P5(i)=ω p1 ·P1(i)+ω p2 ·P2(i)+ω p3 ·P3(i)+ω p4 ·P4(i) (9)
[0041] Where P5(i) represents the joint leakage probability obtained by comprehensively considering the four influences, ω1, ω2, ω3 and ω4 are the weights corresponding to the four leakage probabilities, satisfying ω1+ω2+ω3+ω4=1.
[0042] Furthermore, the step 2 specifically includes:
[0043] Step 2.1: Based on EPANET, simulate the leakage of each node in the pipeline network and obtain the simulated pressure measurement value y of each node in the pipeline network when leakage occurs:
[0044]
[0045] In the formula, y ij Represents the pressure value at node i when leakage occurs at node j, i,j=1,2,…,N n ;
[0046] Step 2.2: Simulate the normal working conditions of the pipeline network to obtain the model prediction values of each node in the pipeline network g i represents the pressure value at node i when there is no leakage in the pipe network predicted by the hydraulic model, and compares it with the measured value to obtain the leakage detection matrix R:
[0047]
[0048] In the formula, g i -y ij Indicates the predicted value g according to the hydraulic model after leakage occurs at node j i The pressure residual obtained by the sensor measurement value at node i is indicated by the indicator α to indicate whether leakage is detected. i -y ij When ≥γ, a leakage alarm is generated, that is, α=1, otherwise, α=0.
[0049] Furthermore, the step 3 specifically includes:
[0050] Step 3.1: Use the NSGA-II multi-objective optimization method to obtain the Pareto front solution of the sensor layout scheme under various leakage probabilities;
[0051] Step 3.2: Considering the four optimization objectives of sensor deployment: minimizing detection time, maximizing the number of leakage event detections, minimizing the number of sensors, and maximizing the joint information entropy of the sensor network, a multi-criteria decision analysis method is used to cluster the Pareto front solutions and sort the solutions within the clusters to obtain the optimal solution under various leakage probabilities.
[0052] Furthermore, step 3.1 specifically includes:
[0053] Step 3.1.1: Based on a randomly selected set of pipe network nodes as a sensor layout scheme, four optimization objective function values are calculated. The calculation of the four objective function values is shown in steps (1)-(4):
[0054] (1) According to the selected scheme, the detection time of all leakage events of the scheme is obtained;
[0055] The detection time is the total time that the sensor network can detect the leakage event, and the smaller the value, the better;
[0056]
[0057] Where F1(s) represents the detection time of sensor arrangement scheme s, γ(i,t) represents the shortest time for scheme s to detect leakage at node i, γ(i,t)=Δt′,2Δt′,…,BΔt′, Δt′ is the sampling time of the sensor, BΔt′ is the longest detection time of a leakage event, if scheme s cannot detect the leakage at node i, then γ(i,t)=BΔt′, Pv(i) represents the probability of leakage at node i, v=1,2,3,4,5;
[0058] (2) According to the selected scheme, the number of leakage events detected by the scheme is obtained;
[0059] Only when leakage is detected can appropriate measures be taken to reduce the damage caused. Therefore, the more leakage events are detected, the better:
[0060]
[0061] Where F2(s) represents the total number of leakage events that can be detected by sensor arrangement scheme s;
[0062] (3) According to the selected scheme, the number of sensors of the scheme is obtained;
[0063] The number of sensors is mainly related to the installation and maintenance costs of the sensor network. The number of sensors should be kept as small as possible:
[0064] F3(s)=min(s n ); (14)
[0065] Where, F3(s) represents the total number of sensors in sensor layout scheme s;
[0066] (4) According to the selected scheme, the joint information entropy of the scheme is obtained;
[0067] When comparing the model predictions g with the measured values y, the model uncertainty (u modelling ) and measurement uncertainty (u measurement ) and the true value are expressed as follows:
[0068] g+u modelling =reality=y+u measurement ; (15)
[0069] Transforming equation (15) yields the difference between the model prediction value and the measured value and the combined uncertainty (u combined ):
[0070] gy=u measurement -u modelling =u combined ; (16)
[0071] The pressure changes caused by leakage at different nodes of the pipeline network are different, that is, the pressure residuals of the same sensor are different when leakage occurs at different nodes. Considering the influence of modeling and measurement errors, a single pressure residual is divided into intervals with a width of W. Different leakage events are distributed in different intervals. W depends on the modeling error and measurement error, that is:
[0072] W=|u modelling |+|u measurement |; (17)
[0073] For a group of leakage events, a single sensor s1 in the sensor arrangement scheme s divides the pressure residuals of these leakage events into intervals according to W, and obtains N1 intervals; then, another sensor s2 in the sensor arrangement scheme s is combined to continue to divide the pressure residuals of these leakage events into intervals according to W, and the N1 intervals are further refined to obtain N2 intervals, and the leakage events are continuously divided into intervals in combination with the remaining sensors in the scheme s;
[0074]
[0075] Where F4(s) represents the joint information entropy of sensor layout scheme s, p(k) = Nk / N T represents the probability of leakage in the kth interval, k = 1, 2, ..., K, K represents the total number of intervals obtained by scheme s for the leakage event, N k represents the number of leakage events in the kth interval, N T Indicates the total number of leakage events;
[0076] Step 3.1.2: Select a new set of sensor nodes as the sensor optimization layout evolution scheme, and calculate the new objective function value based on (1)-(4) in step 3.1.1 as the next generation of evolution;
[0077] Step 3.1.3: Repeat steps 3.1.1 and 3.1.2 until no better sensor layout solution than the existing one can be obtained, and output the optimal sensor layout solution and its objective function value in the form of a Pareto front solution.
[0078] Furthermore, step 3.2 specifically includes:
[0079] Step 3.2.1: Calculate the PROMETHEE method for all solutions Sort them and get the highest ranked solution and the lowest ranked solution calculate and Difference:
[0080]
[0081] If d1≥d threshold , d threshold represents the minimum difference among the solutions in the solution set, then solution a optimal and a worst To be divided into two different clusters C1 and C2, otherwise, stop dividing the cluster and output the optimal solution a optimal and its corresponding objective function value;
[0082] Step 3.2.2: Compare other solutions within the cluster and and Difference:
[0083]
[0084]
[0085] If d2≤d3, plan a is divided into cluster C1; otherwise, plan a is divided into cluster C2;
[0086] Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until all clusters are no longer divided into new clusters and all clusters are obtained.
[0087] Furthermore, in step 4, various schemes are compared and analyzed based on the impact of leakage at each node on the pipeline network, and the impact of leakage at each node of the pipeline network on the node pressure drop, water demand, pipeline pressure drop and pipeline flow rate increase under various sensor layout schemes is obtained. The smaller these impacts are, the better the sensor layout scheme is.
[0088] When optimizing the arrangement of sensors, the present invention obtains the leakage probability of each node of the pipeline network according to the impact of leakage on the pipeline network after leakage occurs at each node of the pipeline network. When comparing and screening the sensor arrangement schemes, the change of the leakage probability of each node of the pipeline network is considered, so as to ensure that the selected sensor arrangement scheme can give priority to detecting leakage events that cause serious leakage consequences, so that the impact on the pipeline network when leakage occurs at each node of the pipeline network is smaller. Through calculations in the embodiment, it is found that considering the change of the node leakage probability can obtain a more reasonable sensor optimization arrangement scheme when the number of sensors is limited, and the impact on the pipeline network is smaller when the sensor leaks. When the existing method arranges the sensors, the change of the leakage probability of each leakage event is not considered. When the number of sensors is limited, the sensor arrangement scheme obtained may ignore some leakage events that cause serious leakage consequences. The method proposed by the present invention distinguishes the leakage events according to the consequences of the leakage events, and gives priority to detecting the leakage events that cause serious leakage consequences, avoiding possible omissions, thereby ensuring that the impact on the pipeline network caused by leakage at each node of the pipeline network is minimized when the number of sensors is limited. BRIEF DESCRIPTION OF THE DRAWINGS
[0089] Figure 1 It is a flow chart of one embodiment of a multi-objective optimization arrangement method of a water supply network leakage detection sensor network based on leakage probability of the present invention;
[0090] Figure 2 Schematic diagram of clustering partitioning for multi-criteria decision analysis;
[0091] Figure 3 This is a schematic diagram of the example pipe network C-town;
[0092] Figure 4 is the leakage probability distribution of each node in the pipeline network based on the leakage probability P1;
[0093] Figure 5 is the leakage probability distribution of each node in the pipeline network based on the leakage probability P2;
[0094] Figure 6is the leakage probability distribution of each node in the pipeline network based on the leakage probability P3;
[0095] Figure 7 The leakage probability distribution of each node in the pipeline network is obtained based on the leakage probability P4;
[0096] Figure 8 Pareto front solution clustering based on various leakage probabilities in case 1: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5;
[0097] Fig. 9 Pareto front solution clustering based on various leakage probabilities in case 2: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5;
[0098] Fig.10 Pareto front de-clustering based on various contamination probabilities in case 3: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5;
[0099] Fig.11 The leakage event detection rate, detection time, number of sensors and joint information entropy distribution of the non-optimal solution: (a) Case 1; (b) Case 2; (c) Case 3;
[0100] Fig.12 The impact of each sensor optimization layout scheme on the pipeline network when leakage occurs at each node of the pipeline network under situation 1: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5;
[0101] Fig.13 The impact of each sensor optimization layout scheme on the pipeline network when leakage occurs at each node of the pipeline network under situation 2: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5;
[0102] Fig.14 The impact of the optimized layout of each sensor on the pipeline network when leakage occurs at each node in the pipeline network under situation 3: (a) P1; (b) P2; (c) P3; (d) P4; (e) P5. DETAILED DESCRIPTION
[0103] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0104] See also Figure 1 The embodiment of the present invention proposes a multi-target arrangement method of leakage detection sensors for a water supply network based on leakage probability, which specifically includes the following steps:
[0105] Step 1, determine the impact of leakage at each node of the pipeline network on the pipeline network, and obtain the leakage probability of each node of the pipeline network;
[0106] Step 2: Simulate various leakage situations to obtain a leakage detection matrix;
[0107] Step 3: Use the NSGA-II multi-objective optimization method and the leakage detection matrix to obtain the Pareto front solutions under different leakage probabilities, and use PROMETHEE to sort and cluster the Pareto front solutions to obtain the optimal solutions under various leakage probabilities;
[0108] Step 4: Compare and analyze various schemes based on the impact of leakage at each node on the pipeline network, and obtain the impact of leakage at each node of the pipeline network on node pressure drop, water demand, pipeline pressure drop and pipeline flow rate increase under various sensor layout schemes. The smaller these impacts are, the better the sensor layout scheme is.
[0109] Further, the step 1 specifically includes:
[0110] Step 1.1: Simulate the leakage at each node of the pipeline network to obtain the impact of leakage at each node on the pipeline network.
[0111] Since leakage at each node of the pipeline network has different impacts on different parts of the pipeline network, leakage at different nodes has different "importance" during leakage detection. Due to the limited number of sensors, it is usually impossible to detect all events. Therefore, when optimizing the arrangement of sensors, it is necessary to give priority to detecting leakage events with a higher degree of "importance", but it is difficult to quantify this "importance". The present invention uses leakage probability to represent this "importance".
[0112] Pipeline network leakage can occur in various forms, such as through circumferential cracks, longitudinal cracks and circular holes, etc. The orifice plate formula is usually used to simulate circular hole leakage. The present invention assumes that the leakage occurs at the pipeline network node, and adds the leakage flow rate as the additional water demand of the node to each leakage node, as shown in formula (1):
[0113]
[0114] Where Dnew n,i Din represents the node water demand (L / s) at time i when node n leaks. n,iThe node water demand (L / s) at time i when there is no leakage at node n, γ is the pressure index, c is the leakage coefficient, and p is the pressure (Pa).
[0115] After simulating the leakage at each node of the pipeline network according to formula (1), the normal working condition of the pipeline network is simulated to obtain the influence of leakage at each node of the pipeline network on the pressure of each node of the pipeline network, the pressure of each pipeline of the pipeline network and the flow rate, as shown in formula (2-4):
[0116]
[0117]
[0118]
[0119] In the formula, γ1 t (j,i),γ2 t (p,i) and γ3 t (p,i) respectively represent the impact of leakage at node i on the pressure of node j, pressure of pipeline p and flow rate at time t, i,j=1,…,N n ,N n is the total number of network nodes, p=1,…,P, P is the total number of network pipes. If leakage occurs at node i, the pressure drop at node j (pipeline p) exceeds 20%, then γ1 t (j,i)=1(γ2 t (p,i)=1), otherwise, γ1 t (j,i)=0(γ2 t (p,i)=0), if the flow velocity at pipe p increases by more than 20%, then γ3 t (p,i)=1, otherwise γ3 t (p,i)=0, t=Δt, 2Δt,…, AΔt, Δt is the hydraulic simulation step, and AΔt is the total simulation time.
[0120] Step 1.2: According to the impact of leakage on the pipeline network at each node of the pipeline network, the leakage probability of each node of the pipeline network is obtained. In this embodiment, four leakage probabilities are obtained according to various impacts: the first leakage probability (P1) is obtained based on the amount of node pressure drop, the second leakage probability (P2) is obtained based on the impact on water demand, the third leakage probability (P3) is obtained based on the pipeline pressure drop, and the fourth leakage probability (P4) is obtained based on the increase in pipeline velocity, and a joint leakage probability (P5) is obtained based on the four leakage probabilities. The specific description is as follows:
[0121] 1.2.1 Obtaining the first leakage probability P1 based on the number of pressure drop nodes
[0122] The pressure drop of the pipe network node will inevitably affect the water supply of the node. The more nodes with pressure drop, the greater the impact. Therefore, the first leakage probability P1 is:
[0123]
[0124] Where P1(i) represents the probability of leakage at the pipeline network node i based on the number of pressure drop nodes, j = 1, 2, ..., N n ;
[0125] 1.2.2 Obtaining the second leakage probability P2 based on the impact on water demand
[0126] The impact of the pressure drop at the network node on water supply is not only related to the number of nodes with pressure drop, but also to the water demand of these nodes. Obviously, the pressure drop at the nodes with greater water demand will cause greater harm, because these nodes usually correspond to more people. Therefore, the second leakage probability P2 is:
[0127]
[0128] Where P2(i) represents the probability of leakage at the network node i based on the water demand, and D jt represents the water demand of node j at time t, Represents the total water demand of the water supply network during the entire simulation period.
[0129] 1.2.3 The third leakage probability P3 is obtained based on the pipeline pressure drop
[0130] In addition to affecting the nodes, leakage in the pipeline network will also affect the pipeline. When leakage occurs, it may cause transient state of the pipeline network, which may cause a significant drop in pressure in some pipelines or even negative pressure, which may cause water quality problems. Therefore, the third leakage probability P3 is:
[0131]
[0132] Where P3(i) represents the probability of leakage at the pipeline node i based on the pipeline pressure drop; L P is the length of pipeline p, p = 1, 2, ..., P. The more pipeline pressure drops caused by leakage at node i, and the longer these pipelines are, the higher the probability of leakage at node i.
[0133] 1.2.4 The fourth leakage probability P4 is obtained based on the increase of pipeline flow velocity
[0134] Due to leakage, the flow rate of some pipes in the water supply network will increase significantly, which will also cause water quality problems in the water supply network. Therefore, the fourth leakage probability P4 is:
[0135]
[0136] Where P4(i) represents the probability of leakage at the network node i based on the increase in pipeline velocity. The more the number of pipeline velocities increase when node i leaks, and the longer these pipelines are, the higher the probability of leakage at node i.
[0137] 1.2.5 The fifth leakage probability P5 is obtained based on the joint leakage probability
[0138] The above leakage probabilities only consider the impact of one aspect when leakage occurs. The present invention combines these four leakage probabilities to obtain a combined leakage probability, namely the fifth leakage probability P5:
[0139] P5(i)=ω p1 ·P1(i)+ω p2 ·P2(i)+ω p3 ·P3(i)+ω p4 ·P4(i) (9)
[0140] Wherein, P5(i) represents the joint leakage probability obtained by comprehensively considering the four influences, ω1, ω2, ω3 and ω4 are the weights corresponding to the four leakage probabilities, respectively, satisfying ω1+ω2+ω3+ω4=1, and the weights can change according to the importance of the four probabilities. The present invention adopts the same weights in the example.
[0141] Combine the leakage probability P5 with the four leakage probabilities P1-P4 mentioned above, and substitute P1-P5 when solving the multi-objective optimization function in step 3. The importance of each leakage event detected by each sensor layout scheme is represented by the change of the leakage probability of each node. When a leakage event with a higher leakage probability is detected, the corresponding optimization objective function value is higher, thereby ensuring that the leakage event with a high leakage probability is detected first.
[0142] When the sensor network detects leakage in the pipeline network, it usually compares the predicted value g of the pipeline network hydraulic model with the measured value y of the sensor. If the pressure drop of a certain sensor exceeds the threshold γ, a leakage alarm will be generated. When selecting the sensor location, it is necessary not only to consider the node location that is sensitive to the pressure change of each node in the pipeline network, but also to consider the impact of the spatial distribution of hydraulic modeling and sensor measurement errors on the selection of sensor locations. Therefore, when optimizing the layout of sensors, it is necessary to determine the changes in pressure at each node of the pipeline network when leakage occurs at each node of the pipeline network and the spatial distribution of modeling and measurement errors.
[0143] Further, the step 2 specifically includes:
[0144] Step 2.1: Based on EPANET, simulate the leakage of each node in the pipeline network and obtain the simulated pressure measurement value y of each node in the pipeline network when leakage occurs:
[0145]
[0146] In the formula, y ij Represents the pressure value at node i when leakage occurs at node j, i,j=1,2,…,N n .
[0147] Step 2.2: Simulate the normal working conditions of the pipeline network to obtain the model prediction values of each node in the pipeline network g i represents the pressure value at node i when there is no leakage in the pipe network predicted by the hydraulic model, and compares it with the measured value to obtain the leakage detection matrix R:
[0148]
[0149] In the formula, g i -y ij Indicates the predicted value g according to the hydraulic model after leakage occurs at node j i The pressure residual obtained by the sensor measurement value at node i. The indicator α is usually used to indicate whether leakage is detected. Considering the measurement error of the sensor and the fluctuation of the actual node water demand, when g i -y ij ≥γ, a leakage alarm will be generated, that is, α=1, otherwise, α=0. Obviously, sensors can only be deployed at some nodes of the pipeline network, and the leakage detection matrix D(s) of the sensor deployment scheme s is a submatrix of R. Only when a certain node of the pipeline network is selected by scheme s, the corresponding row in R will be added to D(s). For example, if a sensor is deployed at node i in scheme s, the i-th row in R will be added to D(s).
[0150] Further, the step 3 specifically includes:
[0151] Step 3.1: Use the NSGA-II multi-objective optimization method to obtain the Pareto front solution of the sensor layout scheme under various leakage probabilities;
[0152] NSGA-II is an improved NSGA algorithm. NSGA-II obtains a set of Pareto front solutions based on various objective function trade-offs. The binary decision variable of the NSGA-II multi-objective optimization model is the selection of sensor locations. The length of the chromosome is the total number of pipe network nodes. If a sensor is arranged at node i, the value at node i is 1, otherwise its value is 0. First, a group of pipe network nodes are randomly selected as sensor arrangement nodes. According to these pipe network nodes, the corresponding rows are selected from the leakage detection matrix R and added to the leakage detection matrix D(s) of the sensor arrangement scheme S, and then the objective function value and crowding distance of the sensor arrangement scheme are calculated. Then, a group of new pipe network nodes are selected as to improve the objective function value and evolve to obtain the next generation. Finally, the pipe network nodes are continuously selected until no better solution than the existing generation can be obtained, and the sensor optimization arrangement scheme and its objective function value are output in the form of Pareto front solutions. The NSGA-II multi-objective optimization model proposed in the present invention considers four optimization objective functions, including: shortest detection time, maximum number of leakage event detections, minimum number of sensors, and maximum joint information entropy of the sensor network.
[0153] Step 3.1 specifically includes:
[0154] Step 3.1.1: Based on a randomly selected set of pipe network nodes as a sensor layout scheme, four optimization objective function values are calculated. The calculation of the four objective function values is shown in steps (1)-(4).
[0155] (1) According to the selected scheme, the detection time of all leakage events of the scheme is obtained;
[0156] The detection time is the total time that the sensor network can detect the leakage event, and the smaller the value, the better.
[0157]
[0158] Where F1(s) represents the detection time of sensor arrangement scheme s, γ(i,t) represents the shortest time for scheme s to detect leakage at node i, γ(i,t)=Δt′,2Δt′,…,BΔt′, Δt′ is the sampling time of the sensor, BΔt′ is the longest detection time of a leakage event, if scheme s cannot detect the leakage at node i, then γ(i,t)=BΔt′, Pv(i) represents the probability of leakage at node i, v=1,2,3,4,5.
[0159] (2) According to the selected scheme, the number of leakage events detected by the scheme is obtained;
[0160] Only when leakage is detected can appropriate measures be taken to reduce the harm it causes. Therefore, the more leakage events are detected, the better.
[0161]
[0162] Where F2(s) represents the total number of leakage events that can be detected by sensor arrangement scheme s.
[0163] (3) According to the selected scheme, the number of sensors of the scheme is obtained;
[0164] The number of sensors is mainly related to the installation and maintenance costs of the sensor network, and the number of sensors should be kept as small as possible.
[0165] F3(s)=min(s n ); (14)
[0166] Where, F3(s) represents the total number of sensors in sensor layout scheme s;
[0167] (4) According to the selected scheme, the joint information entropy of the scheme is obtained;
[0168] When comparing the model predictions g with the measured values y, the model uncertainty (u modelling ) and measurement uncertainty (u measurement ) and the true value are expressed as follows:
[0169] g+u modelling =reality=y+u measurement ; (15)
[0170] Transforming equation (15) yields the difference between the model prediction value and the measured value and the combined uncertainty (u combined ):
[0171] gy=u measurement -u modelling =u combined ; (16)
[0172] The pressure changes caused by leakage at different nodes in the pipeline network are different, that is, the pressure residuals of the same sensor are different when leakage occurs at different nodes. Considering the influence of modeling and measurement errors, a single pressure residual is divided into intervals with a width of W. Different leakage events are distributed in different intervals. W depends on the modeling error and measurement error, that is:
[0173] W=|u modelling |+|u measurement |; (17)
[0174] For a set of leakage events, a single sensor s1 in the sensor arrangement scheme s divides the pressure residuals of these leakage events into intervals according to W, and obtains N1 intervals. Then, another sensor s2 in the sensor arrangement scheme s is combined to continue to divide the pressure residuals of these leakage events into intervals according to W, and the N1 intervals are further refined to obtain N2 intervals, and the leakage events are continuously divided into intervals in combination with the remaining sensors in the scheme s. Obviously, the more intervals the sensor network divides and the fewer leakage events in a single interval, the better the sensor network. The joint information entropy is used in various methods to evaluate the interval division ability of the sensor, and the joint information entropy of the sensor network should be improved as much as possible.
[0175]
[0176] Where F4(s) represents the joint information entropy of sensor layout scheme s, p(k) = N k / N T represents the probability of leakage in the kth interval, k = 1, 2, ..., K, K represents the total number of intervals obtained by scheme s for the leakage event, N k represents the number of leakage events in the kth interval, N T Indicates the total number of leakage events.
[0177] Step 3.1.2: Select a new set of sensor nodes as the sensor optimization layout evolution scheme, and calculate the new objective function value based on (1)-(4) in step 3.1.1 as the next generation of evolution.
[0178] Step 3.1.3: Repeat steps 3.1.1 and 3.1.2 until no better sensor layout solution than the existing one can be obtained, and output the optimal sensor layout solution and its objective function value in the form of a Pareto front solution.
[0179] Step 3.2: Considering the four optimization objectives of sensor deployment: minimizing detection time, maximizing the number of leakage event detections, minimizing the number of sensors, and maximizing the joint information entropy of the sensor network, a multi-criteria decision analysis method (such as Figure 2 As shown in Figure 2, cluster the Pareto front solutions and sort the solutions within the clusters to obtain the optimal solutions under various leakage probabilities;
[0180] The multi-objective optimization model returns a trade-off curve containing a set of optimal points, called Pareto points. The multi-criteria decision making (MCDM) method can select the most ideal solution from the Pareto points according to the decision maker's priorities. PROMETHEE is a non-parametric ranking method that ranks alternative solutions according to the selected preferences (weights) based on the positive and negative preference flows of each solution in the value ranking relationship. This paper uses the PROMETHEE method to rank and cluster the Pareto front solutions and obtain the optimal solutions within various clusters. Step 3.2 is mainly divided into three steps:
[0181] Step 3.2.1: Calculate the PROMETHEE method for all solutions Sort them and get the highest ranked solution and the lowest ranked solution calculate and Difference:
[0182]
[0183] If d1≥d threshold (d threshold represents the minimum difference of each solution in the solution set), then solution a optimal and a worst To be divided into two different clusters C1 and C2, otherwise, stop dividing the cluster and output the optimal solution a optimal and its corresponding objective function value;
[0184] Step 3.2.2: Compare other solutions within the cluster and and Difference:
[0185]
[0186]
[0187] If d2≤d3, plan a is divided into cluster C1; otherwise, plan a is divided into cluster C2;
[0188] Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until all clusters are no longer divided into new clusters and the optimal solution within all clusters is obtained. k represents the total number of clusters obtained by division.
[0189] The present invention considers four criteria (F1, F2, F3 and F4) when ranking and clustering alternatives, as shown in equations (12)-(14) and (18). At the same time, three different situations are considered when clustering: (1) Situation 1, w1=w2=w3=w4=0.25; (2) Situation 2, w1=w2=w4=0.2, w3=0.4; (3) Situation 3, w1=w3=w4=0.2, w2=0.4.
[0190] The technical solution of the present invention is described in detail below with a specific example:
[0191] Step 1: Use the hydraulic model of the water supply network to simulate the sample network model C-town, determine the impact of leakage at each node of the network on the network, and obtain the leakage probability of each node of the network;
[0192] like Figure 3 As shown, the C-town example network has a total of 407 nodes, 443 pipelines, 1 constant head water source, 7 water tanks, 11 pumps and 5 valves. The simulation parameters are set as follows: total simulation time is 96h, hydraulic time step is 10min, water quality time step is 5min, mode time step is 10min. Sensors are arranged on the nodes of the network, and there are a total of 407 initial candidate locations. Leakage occurs at each node of the network, and the leakage simulation starts at 24 times a day, corresponding to 24 demand modes. Therefore, the total number of leakage events is 9768 (i.e., 24×407). Considering that the intrusion of pollutants should be detected in the shortest possible time, the present invention limits the maximum detection time of a single leakage event to 5h (BΔt′=5h), the sensor sampling time is 5min, and γ=W=1 mmHg.
[0193] The step 1 specifically includes:
[0194] Step 1.1: Simulate the leakage at each node of the pipeline network to obtain the impact of leakage at each node on the pipeline network;
[0195] First, the leakage at each node of the pipeline network is simulated, and then the normal working condition of the pipeline network is simulated to obtain the impact of leakage at each node of the pipeline network on the pressure of each node of the pipeline network, the pressure of each pipeline of the pipeline network and the flow rate;
[0196]
[0197]
[0198]
[0199] Step 1.2: According to the impact of leakage at each node of the pipeline network on the pipeline network, the leakage probability of each node of the pipeline network is obtained.
[0200] Figure 4 , Figure 5 , Figure 6 and Figure 7 The leakage probability distributions of the example pipeline network based on leakage probabilities P1, P2, P3 and P4 are shown in the figure. As shown in the figure, this case study observed significant spatial variations in node leakage probability, with node leakage probabilities falling into three levels from high to low.
[0201] The impact of leakage on the network node is mainly reflected in the impact on water supply and water quality. The drop in node pressure will affect the node water supply, but due to the different water demands of each node, the leakage probability distribution based on the node pressure drop (P1) and water demand shortage (P2) is different. For example, node 4 ( Figure 4 ), the number of pressure drop nodes affected by it is relatively small, but the water demand of these nodes is large, so P1 is low and P2 is large. At the same time, the leakage of node 4 has a small impact on the pressure drop of the pipeline, but a large impact on the increase of pipeline flow rate.
[0202] 2. Simulate the leakage situation at each node of the pipeline network to obtain the leakage detection matrix;
[0203]
[0204] R is the leakage detection matrix obtained by simulating 407 nodes under 24 demand modes. As shown in formula (24), each row in the leakage detection matrix represents the detection of each leakage event in the pipeline network by the sensor at a certain node in the pipeline network, and each column in the leakage detection matrix represents the detection of a leakage event in the pipeline network by the sensors at each node in the pipeline network. For example, g i -y ij represents the detection status of the sensor at node i after leakage event j occurs. If g i -y ij ≥γ, then the sensor at node i can detect leakage event j; if g i -y ij<γ, then the sensor at node i cannot detect leakage event j. The leakage detection matrix contains the detection status of each leakage event in the pipeline network by sensors at all nodes of the pipeline network. For a sensor layout scheme, sensors are usually only arranged at some nodes of the pipeline network. Therefore, when calculating the optimization objective function value of each sensor layout scheme, it is necessary to first filter out the corresponding rows in the leakage detection matrix and add them to the leakage detection matrix of the sensor layout scheme. For example, assuming that scheme S arranges sensors at node 407, the 407th row in matrix R is added to the leakage detection matrix of scheme S, and then the optimization objective function values of scheme S are calculated based on the leakage detection matrix of scheme S and compared with the objective function values of other schemes to obtain the Pareto front solution.
[0205] Step 3: Use the NSGA-II method to solve the Pareto front solutions under various leakage probabilities, and use PROMETHEE to sort and cluster the Pareto front solutions to obtain the optimal solution.
[0206] The proposed multi-objective optimization layout method for sensors is used to obtain the Pareto front solutions under five leakage probabilities. The optimization objective functions are shown in Equations (12)-(14) and (18). Figure 8 , Fig. 9 , Fig.10 The Pareto front solutions are clustered under Case 1, Case 2, and Case 3. Each group of Pareto front solutions in each case has four clusters (high, medium-high, medium, and low). In the same case, the Pareto front solutions with different leakage probabilities have different distributions.
[0207] like Figure 8 As shown in the figure, 4 clusters are obtained based on case 1 under various leakage probabilities. The optimal solution (cluster 1) prefers solutions with fewer sensors. These solutions have shorter leakage detection time and are at the inflection point of joint information entropy. The recognition rate of leakage events is higher when the number of sensors is small. Cluster 2, cluster 3, and cluster 4 are distributed at both ends of cluster 1, and the number of Pareto front solutions in each cluster is not equal.
[0208] like Fig. 9 As shown in the figure, four clusters are obtained based on case 2 under various leakage probabilities. The optimal solution (cluster 1) is the solution with a small number of sensors. As the number of sensors increases, the Pareto front solution is divided into clusters 2, 3, and 4. Compared with the other three clusters, cluster 1 has a small number of sensors, a short pollution detection time, and a large joint information entropy. Although the number of leakage events detected will increase if the number of sensors is further increased, the joint information entropy will not change much.
[0209] like Fig.10As shown in the figure, the distribution of the Pareto front solution obtained based on case 3 is more chaotic compared with cases 1 and 2. When the number of leakage event detections is given priority, cluster 2 and cluster 4 are clearly distinguished. Cluster 2 is basically cluster 1 under case 2. This part of the solution has a small number of sensors and can obtain a large joint information entropy, but the number of leakage event detections is small.
[0210] In order to compare various schemes, the present invention calculates the detection time (P0), detection rate (P0), true detection time (P1-P5) and true detection rate (P1-P5) of each optimal scheme based on equal probability (P0) and the five leakage probabilities proposed, as well as the number of sensors and joint information entropy of each scheme, as shown in Fig.11 As shown in Figure 2. P0 does not consider the change of node leakage probability, and the probability of leakage at each node in the pipeline network is equal. Four solutions (C1, C2, C3 and C4) were obtained under three situations with five leakage probabilities, and a total of 60 solutions were obtained.
[0211] like Fig.11 As shown in the figure, the detection rates of various schemes are low at 1 hour, and the real detection rates of the same scheme under different leakage probabilities are also quite different. This difference decreases at 5 hours. The detection rate (P0) of each scheme is lower than the real detection rate (P1-P5). This is because when the number of leakage events detected is the same, the change in leakage probability usually pays more attention to leakage events that cause serious impacts. These events have a higher leakage probability, and thus a higher real detection rate. This difference is also reflected in the detection time. The higher the leakage probability of leakage events detected in a shorter time (1 hour), the lower the probability of leakage events detected in a longer time (5 hours), and the shorter the leakage detection time (lower impact). The leakage detection time from long to short is: P0, P3, P5, P1, P2 and P4. In fact, although various pollution probabilities pay more attention to the impact caused by a certain aspect of leakage, these impacts also have some internal connections. For example, the more leakage events that cause pressure drop, the more likely the number of pipeline pressure drops may be caused by the more nodes that cause pressure drop. However, when the number of sensors is small, the optimal sensor layout scheme based on different leakage probabilities will still tend to reduce the impact on one aspect of the pipeline network, such as Fig.11 Scheme 12 shown in a has the shortest detection time when the leakage probability of each node is P3, indicating that some high-probability (P3) leakage events are easier to detect in a shorter time.
[0212] Fig.11In (a), when clustering and sorting the Pareto front solutions, case 1 uses the same weight for the four optimization objective functions. The optimal solution (C1) obtained under various leakage probabilities has a small number of sensors, a short leakage detection time, a high number of leakage event detections, and a high joint information entropy. Other solutions either have too many sensors (solution 4), a long detection time (solution 18), or a low joint information entropy and detection rate (solution 12). Fig.11 In (b), scenario 2 prefers the solution with fewer sensors. For the optimal solution C1, the leakage detection performance of the solution based on probability P3 (Scheme 1) and P4 (Scheme 5) is significantly better than that based on P1 (Scheme 9), P2 (Scheme 13) and P5 (Scheme 17). This shows that when optimizing the arrangement of sensors, considering the change in leakage probability of each node based on the pipeline pressure drop (P3) and pipeline flow rate increase (P4) can obtain a more reasonable solution. Fig.11 (c) Scenario 3 prefers the schemes with more leakage events to detect, because the number of sensors is larger. Since a large number of sensors are deployed, the leakage detection rate of these schemes is higher and the leakage detection time is shorter. At the same time, compared with Scenario 2, the joint information entropy of these schemes obtained in Scenario 3 is higher.
[0213] In fact, there is little difference between the scheme with a small number of sensors and the scheme with a large number of sensors in terms of the number of leakage events detected and the leakage detection time. The main difference is reflected in the joint information entropy. When there are a small number of sensors, although most leakage events can be detected, they can only determine the occurrence of leakage, but not the specific location of the leakage. After deploying a large number of sensors, the approximate location of the leakage can be determined based on the pressure changes between each sensor. However, considering the installation and maintenance costs of sensors, usually only a small number of sensors are deployed, and the economic cost of leakage always exists.
[0214] Step 4: Compare various solutions based on the impact of leakage at each node on the pipeline network
[0215] In order to compare various schemes, the present invention calculates the impact of the optimal scheme under three situations at different times (1h, 3h and 5h) after the leakage of the pipeline network, respectively. Fig.12 , Fig.13 and Fig.14 shown.
[0216] like Fig.12 As shown in the figure, the impact of leakage on the pipeline (I3 and I4) is significantly greater than the impact on the node (I1 and I2). Previous studies have focused on the impact of leakage on water supply (water volume), while ignoring the impact of leakage on the pipeline (water quality), which is obviously unreasonable. At 1h, the impact of some schemes on the pipeline nodes is zero ( Fig.12(a), 12(b) and 12(d)), but at 3h and 5h, these schemes begin to affect the nodes. This is because some leakage events (undetected) have only a small impact on the network nodes (smaller pressure drop) in a short period of time after the leakage occurs. As the leakage time increases, these leakages begin to affect the network nodes (larger pressure drop) and affect their water supply.
[0217] like Fig.13 As shown in the figure, the impact of each scheme in Situation 2 on the pipeline network at the same time is significantly higher than that in Situation 1. This is because the number of sensors in Situation 2 is smaller. Although in practice, schemes with fewer sensors are usually selected due to the economic level of leakage, increasing the number of sensors will undoubtedly improve the leakage detection performance. Fig.11 The results in (b) are similar. The solutions based on leakage probabilities P3 and P4 have less impact on the water supply network under various leakage probabilities than the solutions obtained by other leakage probabilities (P1, P2, and P5). Therefore, when optimizing the arrangement of sensors, considering the change in leakage probability at each node of the network based on the impact of leakage on the pipeline (pipeline pressure drop and velocity increase), and adding this probability change in the optimization objective function may result in a more reasonable solution. In some methods, the number of sensors is a constant. In this case, the optimal solution can be obtained by reducing the optimization objective function to 3.
[0218] like Fig.14 As shown in Figure 3, the impact of each scheme on the pipeline network in Scenario 3 is relatively small, and the differences between the schemes are also small. This is because Scenario 3 prefers schemes with a larger number of leakage event detections, which can complete the detection of most leakage events, regardless of the high or low leakage probabilities of these events. When there are a large number of sensors, leakage events can not only be detected quickly, but also various leakage events can be distinguished. Considering the possibility of sensor failure, the scheme with a larger number of sensors has better robustness.
[0219] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A multi-target arrangement method of leakage detection sensors for water supply network based on leakage probability, characterized in that: The steps include: Step 1, determine the impact of leakage at each node of the pipeline network on the pipeline network, and obtain the leakage probability of each node of the pipeline network; Step 2: Simulate various leakage situations to obtain a leakage detection matrix; Step 3: Use the NSGA-II multi-objective optimization method and the leakage detection matrix to obtain the Pareto front solutions under different leakage probabilities, and use PROMETHEE to sort and cluster the Pareto front solutions to obtain the optimal solutions under various leakage probabilities; Step 4: Compare and analyze various solutions based on the impact of leakage at each node on the pipeline network; The step 1 specifically includes: Step 1.1: Simulate the leakage at each node of the pipeline network to obtain the impact of leakage at each node on the pipeline network: Step 1.2: According to the impact of leakage at each node of the pipeline network on the pipeline network, the leakage probability of each node of the pipeline network is obtained; In step 1.2, four leakage probabilities are obtained: a first leakage probability P1 is obtained based on the amount of node pressure drop, a second leakage probability P2 is obtained based on the affected water demand, a third leakage probability P3 is obtained based on the pipeline pressure drop, and a fourth leakage probability P4 is obtained based on the pipeline velocity increase, and a combined leakage probability P5 is obtained based on the four leakage probabilities; The four leakage probabilities and the combined leakage probability are obtained specifically through the following steps: The first leakage probability P1 is obtained based on the number of pressure drop nodes: Where P1(i) represents the probability of leakage at the pipeline network node i based on the number of pressure drop nodes, j = 1, 2, ..., N n ; The second leakage probability P2 is obtained based on the impact on water demand: Where P2(i) represents the probability of leakage at the network node i based on the water demand, and D jt represents the water demand of node j at time t, Represents the total water demand of the water supply network during the entire simulation period; The third leakage probability P3 is obtained based on the pipeline pressure drop: Where P3(i) represents the probability of leakage at the pipeline node i based on the pipeline pressure drop; L P is the length of pipeline p, p=1,2,…,P; the more pipeline pressure drops caused by leakage at node i, and the longer these pipelines are, the higher the probability of leakage at node i; The fourth leakage probability P4 is obtained based on the increase in pipeline flow rate: Where P4(i) represents the probability of leakage at the network node i based on the increase in pipeline velocity. The more the number of pipeline velocities increase when leakage occurs at node i, and the longer these pipelines are, the higher the probability of leakage at node i. Based on the joint leakage probability, the fifth leakage probability P5 is obtained: P5(i)=ω p1 ·P1(i)+ω p2 ·P2(i)+ω p3 ·P3(i)+ω p4 ·P4(i) (9) Where P5(i) represents the joint leakage probability obtained by comprehensively considering the four influences, ω1, ω2, ω3 and ω4 are the weights corresponding to the four leakage probabilities, satisfying ω1+ω2+ω3+ω4=1.
2. The method for arranging multiple targets of water supply network leakage detection sensors based on leakage probability as claimed in claim 1, characterized in that: The step 1.1 is specifically as follows: Assuming that leakage occurs at the network node, the leakage flow is added to each leakage node as the additional water demand of the node, as shown in formula (1): Where Dnew n,i represents the node water demand at time i when node n leaks, Din n,i The node water demand at time i when node n has no leakage, γ is the pressure index, c is the leakage coefficient, and p is the pressure; After simulating the leakage at each node of the pipeline network according to formula (1), the normal working condition of the pipeline network is simulated to obtain the influence of leakage at each node of the pipeline network on the pressure of each node of the pipeline network, the pressure of each pipeline of the pipeline network and the flow rate, as shown in formula (2-4): In the formula, γ1 t (j,i),γ2 t (p,i) and γ3 t (p,i) respectively represent the impact of leakage at node i on the pressure of node j, pressure of pipeline p and flow rate at time t, i,j=1,…,N n ,N n is the total number of network nodes, p=1,…,P, P is the total number of network pipes. If leakage occurs at node i, the pressure drop at node j (pipeline p) exceeds 20%, then γ1 t (j,i)=1(γ2 t (p,i)=1), otherwise, γ1 t (j,i)=0(γ2 t (p,i)=0), if the flow velocity at pipe p increases by more than 20%, then γ3 t (p,i)=1, otherwise γ3 t (p,i)=0, t=Δt, 2Δt,…, AΔt, Δt is the hydraulic simulation step, and AΔt is the total simulation time.
3. The multi-target arrangement method of water supply network leakage detection sensors based on leakage probability as claimed in claim 1, characterized in that: The step 2 specifically includes: Step 2.1: Based on EPANET, simulate the leakage of each node in the pipeline network and obtain the simulated pressure measurement value y of each node in the pipeline network when leakage occurs: In the formula, y ij Represents the pressure value at node i when leakage occurs at node j, i,j=1,2,…,N n ; Step 2.2: Simulate the normal working conditions of the pipeline network to obtain the model prediction values of each node in the pipeline network g i represents the pressure value at node i when there is no leakage in the pipe network predicted by the hydraulic model, and compares it with the measured value to obtain the leakage detection matrix R: In the formula, g i -y ij Indicates the predicted value g according to the hydraulic model after leakage occurs at node j i The pressure residual obtained by the sensor measurement value at node i is indicated by the indicator α to indicate whether leakage is detected. i -y ij When ≥γ, a leakage alarm is generated, that is, α=1, otherwise, α=0.
4. The multi-target arrangement method of water supply network leakage detection sensors based on leakage probability as claimed in claim 1, characterized in that: The step 3 specifically includes: Step 3.1: Use the NSGA-II multi-objective optimization method to obtain the Pareto front solution of the sensor layout scheme under various leakage probabilities; Step 3.2: Considering the four optimization objectives of sensor deployment: minimizing detection time, maximizing the number of leakage event detections, minimizing the number of sensors, and maximizing the joint information entropy of the sensor network, a multi-criteria decision analysis method is used to cluster the Pareto front solutions and sort the solutions within the clusters to obtain the optimal solution under various leakage probabilities.
5. The multi-target arrangement method of water supply network leakage detection sensors based on leakage probability as claimed in claim 4, characterized in that: Step 3.1 specifically includes: Step 3.1.1: Based on a randomly selected set of pipe network nodes as a sensor layout scheme, four optimization objective function values are calculated. The calculation of the four objective function values is shown in steps (1)-(4): (1) According to the selected scheme, the detection time of all leakage events of the scheme is obtained; The detection time is the total time that the sensor network can detect the leakage event, and the smaller the value, the better; Where F1(s) represents the detection time of sensor arrangement scheme s, γ(i,t) represents the shortest time for scheme s to detect leakage at node i, γ(i,t)=Δt',2Δt',…,BΔt', Δt' is the sampling time of the sensor, BΔt' is the longest detection time of a leakage event, if scheme s cannot detect the leakage at node i, then γ(i,t)=BΔt', Pv(i) represents the probability of leakage at node i, v=1,2,3,4,5; (2) According to the selected scheme, the number of leakage events detected by the scheme is obtained; Only when leakage is detected can appropriate measures be taken to reduce the damage caused. Therefore, the more leakage events are detected, the better: Where F2(s) represents the total number of leakage events that can be detected by sensor arrangement scheme s; (3) According to the selected scheme, the number of sensors of the scheme is obtained; The number of sensors is mainly related to the installation and maintenance costs of the sensor network. The number of sensors should be kept as small as possible: F3(s)=min(s n ); (14) Where, F3(s) represents the total number of sensors in sensor layout scheme s; (4) According to the selected scheme, the joint information entropy of the scheme is obtained; When comparing the model predictions g with the measured values y, the model uncertainty u modelling and measurement uncertainty u measurement The relationship with the true value is expressed as follows: g+u modelling =reality=y+u measurement ; (15) Transform equation (15) to obtain the difference between the model prediction value and the measured value and the combined uncertainty u combined : gy=u measurement -u modelling =u combined (16) The pressure changes caused by leakage at different nodes of the pipeline network are different, that is, the pressure residuals of the same sensor are different when leakage occurs at different nodes. Considering the influence of modeling and measurement errors, a single pressure residual is divided into intervals with a width of W. Different leakage events are distributed in different intervals. W depends on the modeling error and measurement error, that is: W=|u modelling |+|u measurement |; (17) For a group of leakage events, a single sensor s1 in the sensor arrangement scheme s divides the pressure residuals of these leakage events into intervals according to W, and obtains N1 intervals; then, another sensor s2 in the sensor arrangement scheme s is combined to continue to divide the pressure residuals of these leakage events into intervals according to W, and the N1 intervals are further refined to obtain N2 intervals, and the leakage events are continuously divided into intervals in combination with the remaining sensors in the scheme s; Where F4(s) represents the joint information entropy of sensor layout scheme s, p(k) = N k / N T represents the probability of leakage in the kth interval, k = 1, 2, ..., K, K represents the total number of intervals obtained by scheme s for the leakage event, N k represents the number of leakage events in the kth interval, N T Indicates the total number of leakage events; Step 3.1.2: Select a new set of sensor nodes as the sensor optimization layout evolution scheme, and calculate the new objective function value based on (1)-(4) in step 3.1.1 as the next generation of evolution; Step 3.1.3: Repeat steps 3.1.1 and 3.1.2 until no better sensor layout solution than the existing one can be obtained, and output the optimal sensor layout solution and its objective function value in the form of a Pareto front solution.
6. The method for arranging multiple targets of water supply network leakage detection sensors based on leakage probability as claimed in claim 4, characterized in that: Step 3.2 specifically includes: Step 3.2.1: Calculate the PROMETHEE method for all solutions Sort them and get the highest ranked solution and the lowest ranked solution calculate and Difference: If d1≥d threshold , d threshold represents the minimum difference among the solutions in the solution set, then solution a optimal and a worst To be divided into two different clusters C1 and C2, otherwise, stop dividing the cluster and output the optimal solution a optimal and its corresponding objective function value; Step 3.2.2: Compare other solutions within the cluster and and Difference: If d2≤d3, plan a is divided into cluster C1; otherwise, plan a is divided into cluster C2; Step 3.2.3: Repeat steps 3.2.1 and 3.2.2 until all clusters are no longer divided into new clusters and all clusters are obtained.
7. The method for arranging multiple targets of water supply network leakage detection sensors based on leakage probability according to claim 1, characterized in that: In step 4, various schemes are compared and analyzed based on the impact of leakage at each node on the pipeline network, and the impact of leakage at each node of the pipeline network on node pressure drop, water demand, pipeline pressure drop and pipeline flow rate increase under various sensor layout schemes is obtained. The smaller these impacts are, the better the sensor layout scheme is.
