A water supply pipe burst monitoring method based on physical sensor cooperative gain

Abstract

The present invention discloses a method for monitoring pipe bursts in water supply networks based on the cooperative gain of physical sensors. The method first uses the ratio of the pressure drop detected by the corresponding physical sensor at the time of a pipe burst to the standard deviation of the pressure detected by the corresponding physical sensor under normal operating conditions as the pressure monitoring point sensitivity. This pressure monitoring point sensitivity accurately reflects the sensitivity of the physical sensor to the pressure detected during a pipe burst. Based on this pressure monitoring point sensitivity, pipe sections with similar sensitivities can be clustered, thereby improving the efficiency of subsequent genetic algorithm optimization. The physical sensor combination corresponding to the virtual sensor obtained by optimizing the constraints and objective function of the genetic algorithm provided by the present invention can significantly reduce the coverage blind spots of the pipe network and have a higher fault tolerance rate. In other words, this physical sensor combination can better exchange information, thereby achieving more accurate pipe burst monitoring.

Description

Technical Field

[0001] The present invention belongs to the field of urban water supply network monitoring, and in particular relates to a water supply network pipe burst monitoring method based on physical sensor cooperative gain. Background Art

[0002] Pipe bursts and leaks are a major threat to the safety and reliability of water supply. They not only waste precious clean water resources but also lead to a series of problems, including water pollution, ground subsidence, and insufficient water pressure. These problems not only cause economic losses to water companies but also severely impact residents' daily lives. Developing more accurate pipe burst and leak monitoring technology is a key step in responding to national leakage control policies and improving pipe network operations and service capabilities.

[0003] With the widespread use of sensors, real-time pipe network monitoring platforms based on Supervisory Control and Data Acquisition (SCADA) have gradually become an essential foundation for supporting the digital management and intelligent operation of water supply networks. However, due to the massive scale of urban water supply networks and the limited cost of hardware deployment, monitoring equipment can only be deployed at a limited number of nodes. This results in significant imbalances in the spatial distribution of monitoring system performance, with numerous blind spots in areas with sparse sensor deployment. Therefore, how to maximize the monitoring system's detection accuracy and monitoring coverage for pipe bursts, given the limited monitoring equipment available, is a key research topic.

[0004] Current methods for detecting pipe bursts based on real-time monitoring can be broadly categorized into two types: data-driven approaches and hydraulic model-based approaches. Data-driven approaches typically do not rely on hydraulic models, but instead directly mine information from historical monitoring data to predict and determine system status. While this approach avoids the need for precise calibration of the hydraulic model, it is susceptible to interference from missing and erroneous data and has difficulty obtaining data labels. Consequently, this approach is less suitable for real-world pipe networks with limited labeled data. Consequently, hydraulic model-based approaches are often the preferred approach for pipe burst monitoring and identification. These methods typically determine the network status by comparing measured values ​​with the model's alarm thresholds.

[0005] Current hydraulic model-based methods typically monitor pipe bursts by independently setting alarm thresholds for each pressure sensor. This independent decision-making monitoring approach allows the pipe network to be further divided into Boolean physical coverage zones centered around each sensor. Under this zone concept, every node in the pipe network must be within the sensing range of at least one sensor to achieve complete coverage of the area.

[0006] However, as mentioned above, in practical applications, due to the limited density of sensor deployment, coverage blind spots are easily present in pipe networks. This is one of the drawbacks of independent sensor decision-making for pipe bursts. Furthermore, independent sensor decision-making also results in a low tolerance for errors in pipe burst determination, making it impossible to exchange information with other monitoring points for more accurate judgments. These drawbacks of independent sensor decision-making have received insufficient attention in previous pipe burst monitoring research. Summary of the Invention

[0007] The present invention provides a pipe burst monitoring method for a water supply network based on the cooperative gain of physical sensors, which can improve the pipe burst identification accuracy while expanding the pipe burst monitoring coverage.

[0008] A specific embodiment of the present invention provides a water supply network pipe burst monitoring method based on physical sensor cooperative gain, comprising:

[0009] A pipe burst simulation was performed on each pipe section of the water supply network to obtain a pressure monitoring point sensitivity sequence for each pipe section. The pressure monitoring point sensitivity is the ratio of the pressure drop value detected by the corresponding physical sensor during the pipe burst to the pressure standard deviation of the corresponding physical sensor monitoring value under normal operating conditions. Based on the normalized pressure monitoring point sensitivity sequence of each pipe section, the K-means clustering algorithm was used to cluster each pipe section to obtain a cluster center pipe section set.

[0010] A genetic algorithm is used to select the optimal virtual sensor for each cluster center pipe section, and multiple optimal virtual sensors corresponding to the cluster center pipe section set are deduplicated to obtain a physical sensor cooperative gain system that can achieve improved pipe burst monitoring.

[0011] The objective function of the genetic algorithm is to maximize the virtual sensor monitoring sensitivity, which is the ratio of the sum of the pressure drop values ​​detected by multiple physical sensors fused by the virtual sensor during a pipe burst to the pressure standard deviation of the virtual sensor monitoring value under normal working conditions. The virtual sensor is composed of multiple physical sensors.

[0012] The constraint condition of the genetic algorithm is that the maximum value of the monitoring sensitivity of the virtual sensor of each cluster center pipe segment is greater than the maximum value of the sensitivity of each physical sensor monitoring point.

[0013] Preferably, the sensitivity sequence S of the pressure monitoring points when the jth pipe section bursts is j for:

[0014] S j =[s j1 ,s j2 ,…s ji …,s jN ]

[0015]

[0016] Where N is the number of pressure monitoring points, s ji is the sensitivity of the i-th pressure monitoring point when the j-th pipe section bursts, The occurrence intensity is The pressure drop value detected at the i-th pressure monitoring point when the pipe bursts, σ i is the pressure standard deviation of the physical sensor monitoring value at the i-th pressure monitoring point under normal working conditions.

[0017] Preferably, the objective function of the genetic algorithm is:

[0018]

[0019]

[0020]

[0021] in, The strength of the pipe segment j is When the pipe bursts, the gth virtual sensor v g The pressure drop value, is the gth virtual sensor v g The burst pressure drop value of the fused mth physical sensor, K is the gth virtual sensor v g The number of physical sensors fused, is the virtual sensor v g The standard deviation of pressure, H gm is the virtual sensor v g The pressure monitoring value of the mth physical sensor is fused, Refers to the virtual sensor v g The covariance of the pressure monitoring values ​​between the fused m1-th physical sensor and the m2-th physical sensor. When m1=m2, the covariance is the variance of the physical sensor.

[0022] Preferably, the constraints of the genetic algorithm are:

[0023] Subject to:△E j >0

[0024] △E j =E vj -E wj

[0025]

[0026]

[0027] Among them, △Ej E is the sensitivity improvement of the virtual sensor compared to the physical sensor when a pipe burst occurs in pipe section j, wj is the maximum sensitivity of the pressure monitoring point of the physical sensor corresponding to the pipe section j when the pipe bursts, E vj is the maximum sensitivity of the virtual sensor pressure monitoring when pipe section j bursts, N is the number of physical sensors, 2 N -N-1 is the number of virtual sensors generated by the collaboration, The strength of the burst pipe.

[0028] Preferably, a genetic algorithm is used to select the optimal virtual sensor for each cluster center pipe segment, including:

[0029] S1. Setting initial values ​​of parameters of the genetic algorithm, including the number of physical sensors, population size, number of iterations, crossover probability, and mutation probability, constructing a population based on virtual sensors with different gene combinations, and treating each virtual sensor as an individual;

[0030] S2. Combining the physical sensors in the cluster center segment to construct multiple virtual sensors, and constructing an initial parent population based on the multiple virtual sensors;

[0031] S3. In the initial parent population, replace the individuals that do not meet the constraints so that all individuals in the initial parent population meet the constraints;

[0032] S4. Use the objective function to evaluate the fitness of individuals in the initial parent population that meet the constraints, and retain the individuals with the best fitness;

[0033] S5. Perform crossover and mutation operations on the initial parent population that meets the constraints to obtain the offspring population, and degenerate the individuals in the offspring population that do not meet the constraints into the original individuals;

[0034] S6. After merging the parent population and the offspring population that meet the constraints, the fitness of the individuals is evaluated by the objective function, and the individuals with the best fitness are retained. At the same time, the top X high-fitness individuals of no more than the population size X are selected according to the fitness ranking to form a new parent population.

[0035] S7. Repeat steps S3-S6 until the number of iterations is reached, evaluate the retained individual with the best fitness again through the objective function, and use the individual with the highest fitness as the optimal virtual sensor of the cluster center segment.

[0036] Preferably, the number of iterations of the genetic algorithm is 50-200, the crossover probability is 0.2-0.5, and the mutation probability is 0.2-0.3.

[0037] Preferably, multiple optimal virtual sensors corresponding to the cluster center pipe segment set are deduplicated to obtain a physical sensor cooperative gain system capable of improving pipe burst monitoring, including:

[0038] A genetic algorithm is used to screen out the optimal virtual sensor of each cluster center pipe section in the cluster center pipe section set to obtain multiple optimal virtual sensors, and duplicate virtual sensors in the multiple optimal virtual sensors are removed to obtain a physical sensor cooperative gain system.

[0039] Preferably, the parameters of the K-means clustering algorithm include the pipe network hydraulic model, the location of the pressure monitoring point, the number of clusters, and the simulated pipe burst intensity.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention, for the first time, uses the ratio of the pressure drop value detected by the corresponding physical sensor when a pipe bursts to the pressure standard deviation of the corresponding physical sensor monitoring value under normal operating conditions as the pressure monitoring sensitivity for pipe bursts. This monitoring sensitivity can accurately reflect the pressure response sensitivity of the physical sensor in pipe burst monitoring. Based on this pressure monitoring sensitivity, pipe sections with similar sensor pipe burst response characteristics can be clustered, thereby improving the optimization efficiency of the subsequent genetic algorithm.

[0042] The physical sensor combination corresponding to the virtual sensor obtained by optimizing the constraints and objective function of the genetic algorithm provided by the present invention can significantly reduce the coverage blind area of ​​the pipeline network and have a higher fault tolerance rate. That is, the physical sensor combination can better obtain more accurate pipe burst monitoring judgment through information fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a water supply network pipe burst cooperative gain monitoring system based on pressure monitoring points provided by a specific embodiment of the present invention;

[0044] Figure 2 A schematic diagram of the pipe network provided in Example 1;

[0045] Figure 3 Provides a comparison chart of confidence information coverage of the physical system and the gain system at different simulation times for Example 1;

[0046] Figure 4 A comparison chart of the number of unmonitorable pipe sections and the minimum coverage flow rate of the physical system and the gain system at different simulation times provided in Example 1;

[0047] Figure 5 This is a comparison chart of the false alarm rate and missed alarm rate of the physical system and the gain system at different simulation times provided in Example 1. DETAILED DESCRIPTION

[0048] The purpose of this invention is to develop a method for monitoring pipe bursts in water supply networks based on sensor cooperative gain, which improves the range and accuracy of pipe burst monitoring. The following describes a specific embodiment of the invention in further detail, with reference to the accompanying drawings. The invention is particularly applicable to pipe burst monitoring based on pressure monitoring points in urban water distribution systems.

[0049] In the face of the problems of existing pipe burst monitoring models that are prone to blind spots and low fault tolerance, the specific embodiment of the present invention uses the maximum value of the ratio of the pressure drop value detected by multiple physical sensors fused by the virtual sensor during a pipe burst to the pressure standard deviation of the virtual sensor monitoring value under normal operating conditions as the objective function of the genetic algorithm optimization. It achieves the goal of providing a suitable physical sensor cooperative combination for each pipe section, improving the fault tolerance rate and reducing the monitoring blind spots, thereby achieving the purpose of accurately monitoring pipe bursts.

[0050] A specific embodiment of the present invention provides a water supply network pipe burst monitoring method based on physical sensor cooperative gain, comprising:

[0051] S1. Construct the initial parameter values ​​of the K-means clustering algorithm based on pipe burst sensitivity based on different water supply systems: The initial parameter values ​​of the K-means clustering algorithm include the pipe network hydraulic model, the location of existing pressure monitoring points, the number of cluster centers, and the simulated pipe burst intensity. The M pipe segment sequence [p1, p2, ..., p j ,…,p M ].

[0052] S2. Traverse all pipe sections and perform pipe burst simulation under simulated pipe burst intensity to obtain the pressure monitoring point sensitivity sequence for pipe burst of each pipe section, and normalize the sequence to obtain the normalized sensitivity sequence for pressure monitoring point for pipe burst of each pipe section. The normalized sensitivity matrix is ​​constructed from the sensitivity sequence of the pressure monitoring points of each pipe section that has burst.

[0053] The sensitivity sequence S of the pressure monitoring point (physical sensor) of the j-th pipe section burst provided by the specific embodiment of the present invention is j for:

[0054] S j =[s j1 ,s j2 ,…s ji …,s jN ]

[0055]

[0056] Where N is the number of pressure monitoring points, s ji is the sensitivity of the i-th pressure monitoring point when the j-th pipe section bursts, The occurrence intensity is The pressure drop value detected by the physical sensor at the i-th pressure monitoring point when the pipe bursts, σ i is the pressure standard deviation of the monitoring value at the i-th pressure monitoring point under normal operating conditions. The pressure monitoring point sensitivity provided by the present invention can better reflect the sensitivity of the physical sensor in detecting pipe bursts. The reason is that a relatively small standard deviation σ indicates that the pressure fluctuation of the sensor under normal operating conditions is small, and the pressure is relatively stable under normal operating conditions. The pressure drop after a pipe burst will have a more obvious deviation compared to the normal operating conditions, and the sensor is more sensitive to pipe burst events. If the standard deviation σ of the sensor under normal operating conditions is relatively large, it means that the pressure value under normal operating conditions fluctuates greatly, and the pressure drop signal after a pipe burst is easily submerged in the normal fluctuation, and the sensor is less sensitive to pipe burst events.

[0057] The normalized sensitivity matrix provided by the specific embodiment of the present invention for

[0058]

[0059] Where N is the number of pressure monitoring points and M is the number of pipe sections.

[0060] S3, based on the normalized sensitivity matrix, use the K-means clustering algorithm to cluster each pipe segment to obtain the cluster center pipe segment set, that is, the T root cluster center pipe segment sequence [p1, p2, ..., p l ,…,p T ], the present invention clusters each pipe section based on the normalized sensitivity matrix, and realizes that the pipe sections with similar sensor burst responses are divided into one cluster, which is beneficial to the optimization of the ordered genetic algorithm and improves the optimization efficiency.

[0061] S4. In a specific embodiment of the present invention, a genetic algorithm is used to select the optimal virtual sensor for each cluster center pipe segment, including:

[0062] S41. Initial values ​​of the parameters of the genetic algorithm are set. The parameters of the genetic algorithm include the number of physical sensors, population size, number of iterations, objective function, constraints, crossover probability, and mutation probability. A population of virtual sensors based on different gene combinations, i.e., cooperative combinations of physical sensors, is constructed. Each virtual sensor is treated as an individual. In this specific embodiment of the present invention, the genes referred to are physical sensors.

[0063] The objective function provided by the specific embodiment of the present invention is the maximum value of the virtual sensor's pipe burst monitoring sensitivity. The virtual sensor's pressure monitoring sensitivity is the ratio of the pressure drop value detected by multiple physical sensors fused by the virtual sensor during a pipe burst to the pressure standard deviation of the virtual sensor monitoring value under normal operating conditions. The virtual sensor provided by the specific embodiment of the present invention is obtained by the cooperation of multiple physical sensors. The purpose of constructing the objective function in the present invention is to find the optimal combination of physical sensors corresponding to the cluster center pipe section, so that when a pipe burst occurs in the cluster center pipe section, the virtual sensor fused by the optimal combination of physical sensors has a higher pipe burst sensitivity, that is, it has fewer blind spots and a higher fault tolerance rate.

[0064] In a specific embodiment, the objective function of the genetic algorithm is:

[0065]

[0066]

[0067]

[0068] in, The strength of the pipe segment j is When the pipe bursts, the gth virtual sensor v g The pressure drop value, is the gth virtual sensor v g The burst pressure drop value of the fused mth physical sensor, K is the gth virtual sensor v g The number of physical sensors fused, is the virtual sensor v g The standard deviation of pressure, H gm is the virtual sensor v g The pressure monitoring value of the mth physical sensor is fused, Refers to the virtual sensor v g The covariance of the pressure monitoring values ​​between the fused m1-th physical sensor and the m2-th physical sensor. When m1=m2, the covariance is the variance of the physical sensor.

[0069] The constraint condition provided by the specific embodiment of the present invention is that the maximum value of the sensitivity of the virtual sensor for pipe burst monitoring of each cluster center pipe segment is greater than the maximum value of the sensitivity of each physical monitoring point.

[0070] The constraints of the genetic algorithm provided by the specific embodiment of the present invention are:

[0071] Subject to:△E j >0

[0072] △E j =Evj -E wj

[0073]

[0074]

[0075] Among them, △E j E is the sensitivity improvement of the virtual sensor relative to the physical sensor when a pipe burst occurs in pipe segment j, wj is the maximum sensitivity of the pressure monitoring point of the physical sensor corresponding to the pipe section j when the pipe bursts, E vj is the maximum sensitivity of the virtual sensor pressure monitoring when pipe section j bursts, N is the number of physical sensors, 2 N -N-1 is the number of virtual sensors generated by the collaboration, The strength of the burst pipe.

[0076] S42. Combining the physical sensors in the cluster center pipe section, that is, randomly generating a water supply network pressure monitoring point cooperation plan to construct multiple virtual sensors, and constructing an initial parent population based on the multiple virtual sensors.

[0077] The initial parent population V with a population size of Z provided in a specific embodiment of the present invention is:

[0078]

[0079] S43. According to the constraint conditions, replace the individuals in the initial parent population of step S42 that do not meet the constraint conditions, so that all individuals in the initial parent population meet the constraint conditions.

[0080] S44. Use the objective function to evaluate the fitness of individuals in the initial parent population that meet the constraints, and retain the individuals with the best fitness.

[0081] S45. Perform crossover and mutation operations on the initial parent population that meets the constraints to obtain the offspring population, and degenerate the individuals in the offspring population that do not meet the constraints to the original individuals.

[0082] S46. After merging the parent population and the offspring population that meet the constraints, the fitness of the individuals is evaluated through the objective function, and the individuals with the best fitness are retained. At the same time, the top Z high-fitness individuals of no more than the population size Z are selected according to the fitness ranking to form a new parent population.

[0083] S47. Repeat steps S43-S46 until the number of iterations is reached, evaluate the retained individual with the best fitness again through the objective function, use the individual with the highest fitness as the optimal virtual sensor of the cluster center segment, and add the optimal virtual sensor to the gain system.

[0084] S48. Deduplication of multiple optimal virtual sensors corresponding to the cluster center pipe section set is performed to obtain a physical sensor cooperative gain system that can achieve improved pipe burst monitoring: a genetic algorithm is used to screen out the optimal virtual sensor of each cluster center pipe section in the cluster center pipe section set to obtain an optimal virtual sensor set, and duplicate virtual sensors in the optimal virtual sensor set are removed to obtain a physical sensor cooperative gain system.

[0085] The specific embodiment of the present invention provides a specific process for constructing a cooperative gain objective function and constraint conditions, including:

[0086] For a pipe burst monitoring system consisting of N pressure monitoring points, s i ∈S(i=1,...,N). For a monitoring point s i , whose measured value is H i , approximately obey Normal distribution. If a pipe segment j has a strength burst pipe, pressure monitoring point i A pressure drop signal will be received Therefore, based on the assumptions of unburst pipes and burst pipes, the reading composition of the pressure monitoring point can be obtained:

[0087]

[0088]

[0089]

[0090] in, and They represent the two hypotheses of unexploded pipe and exploded pipe respectively, H i is the monitoring point s i The pressure measurement value, is the pressure value when the pipe is not burst, is the pressure drop caused by the burst pipe, is the monitoring point s i pressure alarm threshold.

[0091] according to (unexploded pipe) assumption, monitoring point s i The false alarm rate can be expressed as:

[0092]

[0093] Where F(x) is the cumulative probability distribution function of the standard normal distribution.

[0094] According to the Neyman-Pearson criterion, the alarm threshold can be determined by ensuring that the false alarm rate does not exceed the specified value, and the monitoring point s is set. i The upper limit of false positive rate is but:

[0095]

[0096] According to formula (4) (6), the monitoring point s can be obtained i Alarm threshold

[0097]

[0098] When there is a burst pipe in the system, (burst pipe) hypothesis, monitoring point s i The underreporting rate of pipe burst events is:

[0099]

[0100] In the specific embodiment of the present invention, the measured values ​​of the physical sensors are fused by the summation rule in the value fusion principle to generate the virtual sensor monitoring value. g1 、w g2 ...w gk Reading H g1 、H g2 ...H gk Add and get the virtual sensor v g Monitoring value:

[0101]

[0102] Where k is the cooperation degree of the virtual sensor, k≤N, w gm It means that the virtual sensor v g The mth physical sensor associated.

[0103] The virtual sensor v provided by the specific embodiment of the present invention g The alarm model is as follows:

[0104]

[0105]

[0106]

[0107]

[0108] According to formula (8), it is easy to get:

[0109]

[0110]

[0111]

[0112] in, is the virtual sensor v g The mean pressure, is the virtual sensor v g The standard deviation of pressure.

[0113] The specific embodiment of the present invention is based on (unexploded pipe) assumption, virtual sensor v g The false alarm rate can be expressed as:

[0114]

[0115] To avoid additional false alarms, set the same false alarm rate threshold as a single physical sensor. The Neyman-Pearson criterion is used to determine the alarm threshold of the virtual sensor:

[0116]

[0117] When there is a burst pipe in the system, Based on the assumption that pipe burst occurs, the false alarm rate of the virtual sensor for pipe burst events is shown in formula (18):

[0118]

[0119] The goal is to construct virtual sensors that can reduce the missed alarm rate for pipe bursts without causing additional false alarms. However, due to the unique characteristics of water supply systems, not all physical sensors can achieve a gain effect through collaboration. Simple combinations can even lead to deterioration in monitoring performance. To achieve this goal, it is necessary to rationally design the combination of participating physical sensors and the degree of collaboration. The gain system design is formulated as an optimization problem.

[0120] Assume that the physical sensors constitute a set W = {w1,w2,…,w r ,…w N}, virtual sensors constitute a set where w r ,v g Represent the rth physical sensor and gth virtual sensor respectively. According to the false alarm rate formulas (8)(18) of physical sensors and virtual sensors, when pipeline j When the burst flow rate is large or small, any physical sensor w r With virtual sensor v g The decrease in the false negative rate between for:

[0121]

[0122] in, It is the ratio of the deviation degree of the abnormal value after the pipe burst to the normal standard deviation, reflecting the monitoring sensitivity of the sensor to the pipe burst event.

[0123] Maximize the reduction in false negative rate As a design goal to achieve gain system optimization. Considering that for a certain event The maximum sensitivity of a physical sensor is a constant, so maximizing It is equivalent to finding the virtual sensor with the greatest sensitivity.

[0124]

[0125]

[0126]

[0127] in, is the virtual sensor v g The occurrence strength of pipe segment j is The pressure drop value of the burst pipe, is the virtual sensor v g The burst pressure drop value of the fused mth physical sensor, K is the virtual sensor v g The number of physical sensors fused, is the virtual sensor v g The standard deviation of pressure, H gm is the virtual sensor v g The pressure monitoring value of the mth physical sensor is fused. On the basis of finding the most sensitive virtual sensor, the selected virtual sensor also needs to meet the constraint that the sensitivity improvement is greater than 0. For the physical sensor, when pipeline j occurs When the burst flow rate is large or small, the maximum sensitivity of the physical sensor is:

[0128]

[0129] The sensitivity gain of a virtual sensor can be expressed as the sensitivity of the virtual sensor minus the maximum sensitivity of the physical sensor:

[0130] △E j =E vj -E wj (24) To achieve virtual sensor gain, the constraints that need to be met are:

[0131] Subject to: △E j>0 (25)

[0132] △E j =E vj -E wj (26)

[0133]

[0134] Among them, △E j E is the sensitivity increase of the virtual sensor for pipe segment j, wj is the maximum burst sensitivity of the physical sensor with respect to pipe segment j, E vj is the maximum burst sensitivity of the virtual sensor with respect to pipe segment j, N is the number of physical sensors, 2 N -N-1 is the number of virtual sensors generated by the collaboration.

[0135] A specific embodiment of the present invention provides a pipe burst alarm model for physical sensors and virtual sensors. The present invention evaluates the reduction in the false alarm rate of the virtual sensor compared to the original optimal physical sensor. The proposed K-means clustering algorithm based on pipe burst sensitivity focuses on reducing the number of similar pipe burst response pipe sections to improve the optimization efficiency when applying the genetic algorithm in the subsequent traversal of the pipe sections. The proposed genetic algorithm avoids traversing exponential virtual sensor cooperation schemes when there are many physical sensors by setting the goal and constraints of maximum positive sensitivity improvement, and can obtain a relatively optimal solution under limited computing power. In general, the present invention not only improves the pipe burst identification accuracy of the water supply network pipe burst monitoring system through the cooperation of sensors, but also provides an efficient and convenient coupling algorithm for the design of the optimal sensor cooperation scheme.

[0136] The confidence information coverage provided by the present invention means: given a monitoring point s i Upper limit of false positive rate and false negative rate When the strength of pipe segment j is When the pipe bursts, the monitoring point s i The false positive rate is less than And the false negative rate is less than Then the pipe section j is called the monitoring point s i Coverage. Confidence is high in monitoring pipe bursts within the coverage area of ​​the monitoring points. The highly sensitive virtual sensors generated by the present invention, through collaboration with physical sensors, meet the upper limit on false alarm rates while achieving a lower false alarm rate than the original physical sensors. This approach is expected to achieve wider coverage and reduce coverage blind spots.

[0137] In one embodiment, the water supply network provided in this embodiment consists of 2 reservoirs, 782 nodes and 905 pipelines. Figure 2 shown.

[0138] The main parameters of the genetic algorithm provided in this embodiment are set as follows: the number of pressure sensors is 8, the population size is 100, the number of iterations is 200, the crossover probability is 0.2, and the mutation probability is 0.2. Figure 1 The design process shown in the figure resulted in a gain system that included 13 virtual sensor combinations with the highest pipe burst sensitivity, as shown in Table 1, including three physical sensors and ten virtual sensors. This embodiment of the present invention applied the resulting gain system to pipe burst monitoring in this case study, comparing and evaluating the original monitoring system with the proposed gain monitoring system using four metrics: confidence information coverage (CIC), minimum coverage flow (MCF), number of unmonitored pipe sections, and alarm probability.

[0139] Table 1 shows the composition of the gain system sensor provided in Example 1.

[0140]

[0141] exist Under the burst flow rate, the pipe burst delay simulation is carried out for 24 hours a day, and the results are Figure 3 The changes in the confidence information coverage of the physical monitoring system and the gain system and the average flow rate of the pipe section at different times are shown. Figure 3 As can be seen, the proposed augmentation system provides a higher confidence information coverage of the pipeline network at any given time than the original physical monitoring system, with an average coverage increase of 25.34%. The confidence information coverage fluctuates over time, primarily due to the positive correlation between the burst pressure drop signal strength and the pipeline flow rate.

[0142] Further simulations of the changes in the average minimum coverage flow (MCF) and the number of unmonitored pipe sections at different times are shown in the following results: Figure 4 As shown. Figure 4 It can be seen that the minimum coverage flow of the gain system (blue five-pointed star) at each moment is significantly lower than that of the physical system (blue triangle), with an average reduction of up to 29.69m 3 / h. This demonstrates the remarkable effectiveness of the proposed sensor cooperative gain method in improving the accuracy of pipe burst identification. Furthermore, the gain system (red five-pointed star) significantly reduces the number of unmonitorable pipe sections compared to the physical monitoring system (red triangle). In particular, at 4°C, 21 pipe sections that were previously unmonitorable due to bursts were identified by the monitoring system thanks to the gain. This demonstrates that the proposed cooperative gain method can enhance the monitoring system's ability to detect pipe bursts in blind spots.

[0143] In order to verify the effectiveness of the cooperative gain method proposed in this invention in improving the alarm accuracy, the proposed calculation formula is used to calculate the false alarm rate and average missed alarm rate of the system before and after cooperation at different times. The results are as follows: Figure 5As shown in the figure, the false alarm rate of the gain system (red five-pointed star) is basically unchanged compared with the physical system (red triangle), with the maximum increase of only 0.16%. For the 24 simulation moments, the average missed alarm rate of the gain system (blue five-pointed star) for each pipe burst is lower than that of the physical system (blue triangle), and the maximum average missed alarm rate reduction is 11.81% at 4 degrees. This shows that the cooperative gain method proposed in the present invention can reduce the missed alarm rate of pipe bursts without worsening the false alarm rate, thereby improving the alarm accuracy of the monitoring system.

[0144] In summary, the sensor cooperative gain-based method proposed in the present invention can be more effectively used for pipe burst monitoring in water supply systems than traditional physical monitoring systems.

Claims

1. A water supply pipe burst monitoring method based on physical sensor cooperative gain, characterized in that: include: A pipe burst simulation was performed on each pipe section of the water supply network to obtain a pressure monitoring point sensitivity sequence for each pipe section. The pressure monitoring point sensitivity is the ratio of the pressure drop value detected by the corresponding physical sensor during the pipe burst to the pressure standard deviation of the corresponding physical sensor monitoring value under normal operating conditions. Based on the normalized pressure monitoring point sensitivity sequence of each pipe section, the K-means clustering algorithm was used to cluster each pipe section to obtain a cluster center pipe section set. A genetic algorithm is used to select the optimal virtual sensor for each cluster center pipe section, and multiple optimal virtual sensors corresponding to the cluster center pipe section set are deduplicated to obtain a physical sensor cooperative gain system that can achieve improved pipe burst monitoring. The objective function of the genetic algorithm is to maximize the virtual sensor monitoring sensitivity, which is the ratio of the sum of the pressure drop values ​​detected by multiple physical sensors fused by the virtual sensor during a pipe burst to the pressure standard deviation of the virtual sensor monitoring value under normal working conditions. The virtual sensor is composed of multiple physical sensors. The constraint condition of the genetic algorithm is that the maximum value of the monitoring sensitivity of the virtual sensor of each cluster center pipe segment is greater than the maximum value of the sensitivity of each physical sensor monitoring point.

2. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: The sensitivity sequence S of the pressure monitoring points when the j-th pipe section bursts j for: S j =[s j1 ,s j2 ,…s ji …,s jN ] Where N is the number of pressure monitoring points, s ji is the sensitivity of the i-th pressure monitoring point when the j-th pipe section bursts, The occurrence intensity is The pressure drop value detected at the i-th pressure monitoring point when the pipe bursts, σ i is the pressure standard deviation of the physical sensor monitoring value at the i-th pressure monitoring point under normal working conditions.

3. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: The objective function of the genetic algorithm is: in, The strength of the pipe segment j is When the pipe bursts, the gth virtual sensor v g The pressure drop value, is the gth virtual sensor v g The burst pressure drop value of the fused mth physical sensor, K is the gth virtual sensor v g The number of physical sensors fused, is the virtual sensor v g The standard deviation of pressure, H gm is the virtual sensor v g The pressure monitoring value of the mth physical sensor is fused, Refers to the virtual sensor v g The covariance of the pressure monitoring values ​​between the fused m1-th physical sensor and the m2-th physical sensor. When m1=m2, the covariance is the variance of the physical sensor.

4. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: The constraints of the genetic algorithm are: Subject to:△E j >0 △E j =And vj -AND wj Among them, △E j E is the sensitivity improvement of the virtual sensor compared to the physical sensor when a pipe burst occurs in pipe section j, wj is the maximum sensitivity of the pressure monitoring point of the physical sensor corresponding to the pipe section j when the pipe bursts, E vj is the maximum sensitivity of the virtual sensor pressure monitoring when pipe section j bursts, N is the number of physical sensors, 2 N -N-1 is the number of virtual sensors generated by the collaboration, The strength of the burst pipe.

5. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: Genetic algorithm is used to select the optimal virtual sensor for each cluster center pipe segment, including: S1. Setting initial values ​​of parameters of the genetic algorithm, including the number of physical sensors, population size, number of iterations, crossover probability, and mutation probability, constructing a population based on virtual sensors with different gene combinations, and treating each virtual sensor as an individual; S2. Combining the physical sensors in the cluster center segment to construct multiple virtual sensors, and constructing an initial parent population based on the multiple virtual sensors; S3. In the initial parent population, replace the individuals that do not meet the constraints so that all individuals in the initial parent population meet the constraints; S4. Use the objective function to evaluate the fitness of individuals in the initial parent population that meet the constraints, and retain the individuals with the best fitness; S5. Perform crossover and mutation operations on the initial parent population that meets the constraints to obtain the offspring population, and degenerate the individuals in the offspring population that do not meet the constraints into the original individuals; S6. After merging the parent population and the offspring population that meet the constraints, the fitness of the individuals is evaluated by the objective function, and the individuals with the best fitness are retained. At the same time, the top X high-fitness individuals of no more than the population size X are selected according to the fitness ranking to form a new parent population. S7. Repeat steps S3-S6 until the number of iterations is reached, evaluate the retained individual with the best fitness again through the objective function, and use the individual with the highest fitness as the optimal virtual sensor of the cluster center segment.

6. The water supply network burst monitoring method based on physical sensor cooperative gain according to claim 5 is characterized in that: The number of iterations of the genetic algorithm is 50-200 times, the crossover probability is 0.2-0.5, and the mutation probability is 0.2-0.

3.

7. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: The multiple optimal virtual sensors corresponding to the cluster center pipe section set are deduplicated to obtain a physical sensor cooperative gain system that can improve pipe burst monitoring, including: A genetic algorithm is used to screen out the optimal virtual sensor of each cluster center pipe section in the cluster center pipe section set to obtain multiple optimal virtual sensors, and duplicate virtual sensors in the multiple optimal virtual sensors are removed to obtain a physical sensor cooperative gain system.

8. The water supply pipe burst monitoring method based on physical sensor cooperative gain according to claim 1 is characterized in that: The parameters of the K-means clustering algorithm include the pipe network hydraulic model, the location of the pressure monitoring point, the number of clusters, and the simulated pipe burst intensity.

Images