A monitoring method and device for burst of urban water supply pipe network

Abstract

The application discloses a kind of for city water supply pipe network pipe burst monitoring method, this method is by time sliding window, monitoring data and benchmark data are divided into time sliding window multi-time data, time sliding window monitoring data and corresponding sliding window benchmark data are calculated, and the distance feature and shape feature of distinguishing pipe burst and normal event are calculated, according to the grouping result of sensor, the feature of unit grouping is recalculated, isolated forest pipe burst monitoring model is established, and monitoring result is obtained as the monitoring basis of final pipe burst event.The application also provides a kind of monitoring device.The method of the application can effectively improve the accuracy of pipe burst monitoring.

Description

Technical Field

[0001] This invention belongs to the field of municipal engineering and urban water supply network, and particularly relates to a method and device for monitoring pipe bursts in urban water supply networks. Background Technology

[0002] Pipe bursts in water supply networks can cause significant waste of water and energy, and can also lead to public safety issues. For example, when a pipe bursts, a large amount of water can flow out of the pipe and draw bacteria and pollutants into the pipe section, posing a risk to water safety.

[0003] Pipe burst monitoring methods can be divided into two types: hardware-based and software-based. Hardware-based methods mainly use equipment such as hydrophones and noise recorders for monitoring. These methods require a large number of human resources and have limited sound wave propagation distance. They require a large amount of equipment in a single area and are generally difficult to apply to metering zones or pipe networks throughout a city.

[0004] Software-based methods can effectively address large-scale metering zones or pipe network applications. These include model-based and data-driven approaches. Model-based methods typically use correlation analysis between model-simulated data and real-time monitoring data for pipe burst monitoring, relying entirely on the accuracy of model calibration, resulting in low accuracy. Data-driven methods generally include methods based on supervised learning algorithms and methods based on unsupervised learning algorithms. Supervised learning algorithms require a large amount of monitoring data with and without burst labels as training samples. Currently, water companies have not provided a large amount of labeled monitoring data, leading to low accuracy.

[0005] Methods based on unsupervised learning algorithms do not rely on the accuracy of a large amount of monitoring data labels and have good prospects for future applications in pipeline burst monitoring.

[0006] Patent document CN110043808B discloses a model-based pipe burst monitoring method. It uses pressure data simulated by a hydraulic model, sets a pipe burst monitoring threshold, and achieves pipe burst monitoring by comparing the similarity between real-time monitoring data and model-predicted data with the threshold. This method can detect actual pipe burst events, but it requires an accurate model, and its reliability depends entirely on the real-time accuracy of the model, thus demanding high accuracy from the model.

[0007] Patent document CN115931055A discloses a rural water supply operation diagnosis method based on big data analysis. The method includes: S1: using the flow and pressure data from the previous T days as the original flow and pressure dataset, removing abnormal data from the original dataset to obtain a filtered flow and pressure dataset; S2: supplementing the filtered flow and pressure dataset using a SARIMA model to obtain a dataset to be analyzed, dividing the dataset into datasets for different time periods, and further dividing each dataset into a flow dataset and a pressure dataset; S3: calculating the pressure standard deviation of each pressure dataset, and using this standard deviation to determine whether the pressure monitoring value at the corresponding time is abnormal; S4: calculating the flow estimate of each flow dataset using an improved and optimized Kalman filter model, and using this estimated flow value to determine whether the flow monitoring value at the corresponding time is abnormal; S5: if both the pressure monitoring value and the flow monitoring value are abnormal, the monitoring point data is determined to be abnormal, indicating a pipe burst. This method determines whether a pipe burst event has occurred based on pressure changes at each node, but it requires a large amount of data to train the model and also needs to process outlier data, making the whole process quite time-consuming. Summary of the Invention

[0008] The purpose of this invention is to provide a monitoring method for pipe bursts in water supply networks, which can effectively improve the monitoring efficiency and accuracy of pipe burst events.

[0009] To achieve the above objectives, the present invention provides a method for monitoring burst pipes in urban water supply networks, comprising the following steps:

[0010] Historical monitoring data from each sensor is acquired to construct a baseline pressure database.

[0011] The sensor's monitoring data is collected and compared with the corresponding benchmark pressure data in the benchmark pressure database. The data is then divided using a time sliding window length to obtain the corresponding time sliding window monitoring data and time sliding window benchmark data.

[0012] Similarity calculations are performed based on the time-sliding window monitoring data and the time-sliding window baseline data to obtain feature values ​​used to distinguish between pipe burst events and normal events.

[0013] Based on the location information of each sensor, a sensor grouping scheme is constructed.

[0014] Based on the feature values, the corresponding set of training feature values ​​is calculated using the sensor grouping scheme as the unit.

[0015] A monitoring model is constructed using the isolated forest algorithm, and the monitoring model is trained based on the training feature value set to obtain an isolated forest pipe burst monitoring model for judging pipe burst events.

[0016] The collected monitoring data is input into the isolated forest pipe burst monitoring model to determine whether a pipe burst event has occurred in the area where the monitoring data is located.

[0017] This invention utilizes monitoring data from various sensors to calculate distance and shape feature values ​​within a time sliding window, using sensor groups as units, to establish an isolated forest burst pipe monitoring model. By leveraging the synergistic effect of multiple time points and multiple sensors, the efficiency and accuracy of burst pipe monitoring are improved.

[0018] Specifically, the construction process of the benchmark pressure database is as follows:

[0019] Step 1.1: Obtain historical monitoring data from the sensor at the same time over multiple days to form the historical monitoring dataset corresponding to the first time point.

[0020] The DBSCAN algorithm is used to cluster the historical monitoring dataset, and the data cluster with the highest density in the clustering results is selected. The data centroid of the data cluster is calculated as the reference data of the sensor at the same time. The above operation is repeated until the reference data of the sensor at each time is obtained.

[0021] Step 1.2: Repeat step 1.1 until the reference data of all sensors at each time moment is obtained, so as to build the corresponding reference pressure database.

[0022] Specifically, the expression for the time-sliding window monitoring data is as follows:

[0023]

[0024]

[0025] in, It is the monitoring data of sensor k at time j on day i. It is the time-sliding window monitoring data vector of sensor k at time j on day i. is the data matrix monitored by sensor k in a time sliding window on day i, TSWL is the length of the time sliding window, and m is the number of moments in the whole day.

[0026] Specifically, the expression for the time sliding window reference data is as follows:

[0027]

[0028]

[0029] in, It is the reference pressure data of sensor k at time j. It is the reference pressure data vector of sensor k in the j-th time sliding window. It is the reference pressure data matrix of sensor k.

[0030] Specifically, the feature values ​​include distance feature values ​​and shape feature values ​​of the sliding window data.

[0031] Specifically, the expression for the distance feature value is as follows:

[0032]

[0033] in, This represents the distance characteristic value of sensor k at the j-th time sliding window on day i.

[0034] Specifically, the expression for the shape feature value is as follows:

[0035]

[0036]

[0037] in, This represents the shape feature value of sensor k at time j on day i. This represents the ratio of the monitoring data of sensor k at the j-th time sliding window on day i to the reference data of sensor k at the j-th time sliding window on day i.

[0038] Specifically, the process of constructing the grouping scheme is as follows:

[0039] Step 2.1: Set the number of sensor groups n group and the average number of sensors per group n p .

[0040] Step 2.2: Obtain the position information of the sensors, calculate the distance between any two sensors, and construct the sensor distance matrix C. d .

[0041] Step 2.3: Obtain the sensor distance matrix C d The maximum distance d in a,b The sensors a and b corresponding to the maximum distance are assigned to sensor group a and sensor group b, respectively.

[0042] Step 2.4: Sort the distances between sensor a and the other sensors in ascending order, and select the first n... p —One sensor is assigned to sensor group a. Similarly, sensor group b is assigned in the same way as group a.

[0043] Step 2.5: Obtain the sensors from group a and group b in step 2.3. Repeat steps 2.2 and 2.4 for the remaining sensors until each sensor is assigned to a specific group. The sensor grouping scheme is now complete.

[0044] The present invention also provides a monitoring device, including a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor. The computer memory uses the aforementioned isolated forest pipe burst monitoring model. When the computer processor executes the computer program, it performs the following steps: inputting monitoring data from sensors in each area of ​​the pipeline network into the isolated forest pipe burst monitoring model to output whether a pipe burst event exists and its location.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] Feature values ​​are calculated using sliding time window monitoring data and preliminary prediction baseline data. The average value of the feature values ​​is calculated for each sensor group. The monitoring model is then trained to monitor pipe burst events. By utilizing temporal and spatial information, the features of pipe burst event and normal event monitoring data are amplified in multiple dimensions, thereby improving the efficiency and accuracy of pipe burst monitoring. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating a method for monitoring burst pipes in urban water supply networks, as provided in this embodiment.

[0048] Figure 2 This is a schematic diagram of the water supply network in this embodiment;

[0049] Figure 3 This is a schematic diagram of the preliminary prediction results based on the clustering algorithm in this embodiment;

[0050] Figure 4 This is a graph showing the change in distance characteristics over a one-week period, including the pipe burst event, in this embodiment.

[0051] Figure 5 This is a diagram showing the shape feature changes over a period of one week, including the pipe burst event, in this embodiment. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. Below...

[0053] When the description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0054] This embodiment demonstrates pipe burst detection based on a metering zone of a city's water supply network. The zone contains 132 water demand nodes and 153 pipe segments, with a total of 10 sensors deployed (corresponding to monitoring points 0-9). The node IDs are as follows: J204, J39, J40, J366, J408, J417, J7, J179, J161, J156.

[0055] like Figure 1 As shown, a method for monitoring burst pipes in urban water supply networks includes the following steps:

[0056] Historical monitoring data from each sensor is acquired to construct a baseline pressure database.

[0057] More specifically, the process of constructing the benchmark pressure database is as follows:

[0058] Step 1.1: Obtain historical monitoring data from the sensor at the same time over multiple days to form the historical monitoring dataset corresponding to the first time point.

[0059] like Figure 2 The diagram shown is a schematic of the water supply network, with the black dots representing the monitoring points.

[0060] The DBSCAN algorithm is used to cluster the historical monitoring dataset, and the data cluster with the highest density in the clustering results is selected. The data centroid of the data cluster is calculated as the reference data of the sensor at the same time. The above operation is repeated until the reference data of the sensor at each time is obtained.

[0061] Step 1.2: Repeat step 1.1 until the reference data of all sensors at each time moment is obtained, so as to build the corresponding reference pressure database.

[0062] In this embodiment, the monitoring data time step is 15 minutes, and the total number of moments in a day is m = 96. A two-dimensional historical monitoring dataset is constructed by sequentially collecting historical monitoring data from the same time period over 400 days, including two consecutive moments, to obtain historical monitoring datasets for each sensor and time period (a total of 10 × 48 datasets). Baseline data is calculated based on the clustering results of each dataset, such as... Figure 3 As shown, the data from sensor 0 at times 1 and 2 has been clustered. Black in the figure represents predicted burst points, and gray represents predicted normal points. Table 1 shows the baseline data for each time point of the sensor in this embodiment (96 rows total, partial excerpt):

[0063] Table 1. Reference pressure data (m) for each sensor at each time step.

[0064]

[0065] The sensor's monitoring data is collected and compared with the corresponding benchmark pressure data in the benchmark pressure database. The data is then divided using a time sliding window length to obtain the corresponding time sliding window monitoring data and time sliding window benchmark data.

[0066] More specifically, the expression for the time-sliding window monitoring data is as follows:

[0067]

[0068]

[0069] in, It is the monitoring data of sensor k at time j on day i. It is the time-sliding window monitoring data vector of sensor k at time j on day i. is the data matrix monitored by sensor k in a time sliding window on day i, TSWL is the length of the time sliding window, and m is the number of moments in the whole day.

[0070] The expression for the time sliding window baseline data is as follows:

[0071]

[0072]

[0073] in, It is the reference pressure data of sensor k at time j. It is the reference pressure data vector of sensor k in the j-th time sliding window. It is the reference pressure data matrix of sensor k.

[0074] In this embodiment, the monitoring data from the 1600-day sensor is divided into monitoring data and baseline data from step 1 using a certain time sliding window length (TSWL=4), resulting in 153,597 time sliding window monitoring data and 93 time sliding window baseline data. The monitoring data from the first day of sensor 1, including the three time sliding window monitoring data and the time sliding window data matrix for the first day, are taken as follows:

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] The baseline data for the first day, including the three time sliding windows and the baseline data matrix for the first day, are presented below:

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] When monitoring and analyzing pipe bursts, the characteristics of the monitoring data can be calculated by comparing the monitoring data with the baseline data, which can, to some extent, distinguish between pipe burst events and normal events.

[0087] Similarity calculations are performed based on the time-sliding window monitoring data and the time-sliding window baseline data to obtain feature values ​​used to distinguish between pipe burst events and normal events.

[0088] More specifically, the features include distance features of the sliding window data and shape features of the sliding window data.

[0089] Distance feature values ​​of dynamic window data:

[0090]

[0091] in, This represents the distance characteristic value of sensor k at the j-th time sliding window on day i.

[0092] Shape characteristics of sliding window data:

[0093]

[0094]

[0095] in, This represents the shape feature value of sensor k at time j on day i. This represents the ratio of the monitoring data of sensor k at the j-th time sliding window on day i to the reference data of sensor k at the j-th time sliding window on day i.

[0096] In this embodiment, based on 1600 days of monitoring data, the similarity calculation was performed on the monitoring data of 153597 time sliding windows and the baseline data of the corresponding time sliding windows (TSWL=4), and the distance feature values ​​and shape feature values ​​of all sensors were obtained as shown in Tables 2 and 3 (a total of 115197 rows, partial excerpt).

[0097] Table 2. Distance characteristic values ​​of the sensor

[0098]

[0099] Table 3. Numerical values ​​of sensor shape characteristics

[0100]

[0101] Based on the location information of each sensor, a sensor grouping scheme is constructed.

[0102] More specifically, the process of constructing the grouping scheme is as follows:

[0103] Step 2.1: Set the number of sensor groups n group and the average number of sensors per group n p .

[0104] Step 2.2: Obtain the position information of the sensors, calculate the distance between any two sensors, and construct the sensor distance matrix C. d .

[0105] Step 2.3: Obtain the sensor distance matrix C d The maximum distance d in a,b The sensors a and b corresponding to the maximum distance are assigned to sensor group a and sensor group b, respectively.

[0106] Step 2.4: Sort the distances between sensor a and the other sensors in ascending order, and select the first n... p —One sensor is assigned to sensor group a. Similarly, sensor group b is assigned in the same way as group a.

[0107] Step 2.5: Obtain the sensors from group a and group b in step 2.3. Repeat steps 2.2 and 2.4 for the remaining sensors until each sensor is assigned to a specific group. The sensor grouping scheme is now complete.

[0108] In this embodiment, the initial sensor matrix C d As shown below:

[0109]

[0110] In C d The distance between sensor 0 and sensor 9 is the largest. They are initially divided into group 0 and group 9. The distances between sensors 1 to 8 and sensors 0 and 9 are shown in Table 4.

[0111] Table 4 shows the distances (m) between some sensors.

[0112] Sensor 1 807 2835 Sensor 2 800 2888 Sensor 3 951 2507 Sensor 4 1772 1991 Sensor 5 2290 1208 Sensor 6 2678 886 Sensor 7 2399 1473 Sensor 8 3030 671

[0113] In this example, the quantity n group =3 and the average number of sensors per group n p =3. After sorting the distances of sensor 0 to sensors 1 to 8 from smallest to largest, the two smallest distances are sensor 1 and 2. Therefore, sensor group 0 includes sensors 0, 1 and 2. Similarly, sensor group 9 includes sensors 6, 7, 8 and 9. The third group of sensors is sensors 4, 5 and 6, resulting in the final sensor grouping scheme.

[0114] Based on the feature values, the corresponding set of training feature values ​​is calculated using the sensor grouping scheme as the unit.

[0115] More specifically, the distance feature values ​​and shape feature values ​​in this embodiment, based on sensor groups, are shown in Tables 5 and 6:

[0116] Table 5 Distance characteristics of sensor groups

[0117]

[0118] Table 6 Shape characteristics of sensor groups

[0119]

[0120] Take a burst pipe incident as an example, such as Figure 4 and Figure 5 The figure shows the changes in distance and shape feature values ​​of three sensor groups containing two pipe burst events over a period of one week. When a pipe burst event occurs, the two features suddenly increase and remain so for a period of time, which can distinguish between leakage events and normal events.

[0121] A monitoring model is constructed using the isolated forest algorithm, and the monitoring model is trained based on the training feature value set to obtain an isolated forest pipe burst monitoring model for judging pipe burst events.

[0122] The collected monitoring data is input into the isolated forest pipe burst monitoring model to determine whether a pipe burst event has occurred in the area where the monitoring data is located.

[0123] More specifically, in this embodiment, if the monitoring data from three consecutive time points is determined to be a burst pipe by a trained isolated forest burst pipe recognition model, then the event to be monitored is determined to be a burst pipe event.

[0124] In this embodiment, by combining time information from multiple moments based on a time sliding window with spatial information from multiple sensor monitoring data, the detection rate of pipe burst events reaches 90%, while the monitoring effect based on single-moment data is only 65%, and the monitoring effect based on single sensor data is only 63%, which greatly improves the monitoring accuracy.

[0125] This embodiment also provides a monitoring device, including a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor. The computer memory uses the isolated forest burst pipe monitoring model described in the above embodiment.

[0126] When the computer processor executes the computer program, it performs the following steps: inputting the monitoring data of sensors in each area of ​​the pipeline network into the isolated forest pipe burst monitoring model, so as to output whether a pipe burst event exists and the area where it is located.

[0127] This invention uses sliding time window monitoring data and preliminary prediction baseline data to calculate distance and shape features, calculates the average value of feature values ​​on a sensor group basis, trains a monitoring model to monitor pipe burst events, and utilizes time and space information to amplify the features of pipe burst event and normal event monitoring data in multiple dimensions, thereby improving the efficiency and accuracy of pipe burst monitoring.

Claims

1. A method for monitoring burst pipes in urban water supply networks, characterized in that, Includes the following steps: Historical monitoring data from each sensor is acquired to construct a baseline pressure database; The process of constructing the benchmark pressure database is as follows: Step 1.1: Obtain historical monitoring data from the sensor at the same time over multiple days to form the historical monitoring dataset corresponding to the first time point; The DBSCAN algorithm is used to cluster the historical monitoring dataset, and the data cluster with the highest density in the clustering results is selected. The data centroid of the data cluster is calculated as the reference data of the sensor at the same time. The above operation is repeated until the reference data of the sensor at each time is obtained. Step 1.2: Repeat step 1.1 until the reference data of all sensors at each time moment is obtained, so as to build the corresponding reference pressure database; The sensor's monitoring data is collected and compared with the corresponding benchmark pressure data in the benchmark pressure database. The data is then divided using a time sliding window length to obtain the corresponding time sliding window monitoring data and time sliding window benchmark data. Similarity calculations are performed based on the time sliding window monitoring data and the time sliding window baseline data to obtain feature values ​​used to distinguish between pipe burst events and normal events. Based on the location information of each sensor, a sensor grouping scheme is constructed; The process of constructing the grouping scheme is as follows: Step 2.1, Set the number of sensor groups. and the average number of sensors per group ; Step 2.2: Obtain the position information of the sensors, calculate the distance between any two sensors, and construct a sensor distance matrix. ; Step 2.3: Obtain the sensor distance matrix. Maximum distance in The sensor corresponding to the maximum distance and Distributed to sensors respectively Groups and sensors Within the group; Step 2.4, Place the sensor Sort the distances to the other sensors from smallest to largest, and then... The sensors are divided into sensor categories. Within the group, similarly, sensors Grouping methods and Same group; Step 2.5, obtain from Step 2.3 Groups and For the grouped sensors, repeat steps 2.2 and 2.4 for the remaining sensors until each sensor is assigned to a specific group, thus completing the sensor grouping scheme; calculate the corresponding set of training feature values ​​based on the feature values, taking the groups in the sensor grouping scheme as the unit; A monitoring model is constructed using the isolated forest algorithm, and the monitoring model is trained based on the training feature set to obtain an isolated forest pipe burst monitoring model for judging pipe burst events. The collected monitoring data is input into the isolated forest pipe burst monitoring model to determine whether a pipe burst event has occurred in the area where the monitoring data is located.

2. The monitoring method for burst pipes in urban water supply networks according to claim 1, characterized in that, The expression for the time-sliding window monitoring data is as follows: ;in, It is a sensor In the Heavenly Monitoring data at each moment, It is a sensor In the Heavenly A time-sliding window monitoring data vector at each moment It is a sensor In the A daily sliding window monitoring data matrix It is the length of the time sliding window. It represents the total number of moments throughout the day.

3. The monitoring method for burst pipes in urban water supply networks according to claim 1, characterized in that, The expression for the time sliding window reference data is as follows: ;in It is a sensor In the The baseline pressure data at each moment, It is a sensor In the A vector of baseline pressure data for a time-sliding window; It is a sensor The baseline pressure data matrix.

4. The monitoring method for burst pipes in urban water supply networks according to claim 1, characterized in that, The feature values ​​include the distance feature value of the sliding window data and the shape feature value of the sliding window data.

5. The monitoring method for burst pipes in urban water supply networks according to claim 4, characterized in that, The expression for the distance feature value is as follows: ;in, Indicates sensor In the Heavenly Distance feature values ​​for each time sliding window.

6. The monitoring method for burst pipes in urban water supply networks according to claim 4, characterized in that, The expression for the shape feature value is as follows: ； in, Indicates sensor In the Heavenly The shape feature values ​​of a time sliding window Indicates that sensor k is at the 1st dpole. Heavenly Monitoring data and sensors in a time sliding window In the Heavenly The ratio of the baseline data for each time sliding window.

7. A monitoring device, comprising a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor, characterized in that, The computer memory uses the isolated forest pipe burst monitoring model as described in claim 1; when the computer processor executes the computer program, it performs the following steps: inputting the monitoring data of the sensors in each area of ​​the pipeline network into the isolated forest pipe burst monitoring model, so as to output whether a pipe burst event exists and the area where it is located.

Images