Water supply pipe network burst detection method based on wavelet decomposition to extract high-frequency components of pressure

Abstract

The application provides a water supply pipe network burst detection method based on wavelet decomposition to extract high-frequency components of pressure, comprising the following steps: step S1, generating an original pressure monitoring value matrix; step S2, separating high and low frequencies of the original pressure monitoring value matrix by using discrete wavelet transform to generate a high-frequency disturbance value detection column vector at the current time; step S3, detecting outliers in the high-frequency disturbance value detection column vector by using a COF algorithm; step S4, screening the detected outliers, marking outliers meeting the characteristics as abnormal points, and correcting the pressure values corresponding to the abnormal points after the detection of the day is completed; step S5, repeating the above steps, and issuing a burst warning when the number of continuous abnormal time of abnormal points meeting the burst characteristics exceeds the time threshold. The application is beneficial to more accurately reflecting the instantaneous change of the pressure in the water supply pipe network, ensuring the real-time detection, and is suitable for various types of water supply pipe networks.

Description

Technical Field

[0001] This invention relates to a method for detecting burst pipes in water supply networks, specifically a method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components. Background Technology

[0002] Due to the cumulative effects of corrosion, structural fatigue, and external environmental factors related to ground subsidence or third-party influences (such as surface loading), pressurized pipelines gradually age over time, leading to leakage. Pipeline leakage manifests in various forms, with pipe bursts being the most destructive. These bursts can cause significant water loss in a short period, potentially damaging the urban environment, polluting drinking water, and causing adverse social impacts. The "Notice on Strengthening Leakage Control in Public Water Supply Networks," issued in 2022, aims to establish a precise, efficient, safe, and long-term leakage control model for water supply networks, requiring improved information-based and intelligent management of leakage. Therefore, conducting research on rapid detection of water supply network bursts is crucial for timely repair, reducing network leakage, and ensuring water supply safety.

[0003] With the development of computer and communication technologies, Supervisory Control and Data Acquisition (SCADA) systems have been widely applied in the urban water supply industry. SCADA systems can collect data such as flow and pressure in the pipeline network, enabling real-time and comprehensive monitoring of the network's operational status. This provides strong data support for pipeline burst detection, and burst detection can be achieved through data mining, machine learning, and hydraulic simulation. Currently, numerous studies on SCADA data-driven methods for detecting bursts in water supply networks have been conducted both domestically and internationally. However, existing technologies still have the following shortcomings: For pressure data, although slow leaks, valve operations, and other maintenance operations can detect pressure changes within minutes or hours, the pressure transients caused by burst events often occur within seconds. The 5-15 minute pressure acquisition frequency used in most existing studies may lead to delayed burst detection results and low sensitivity. Existing technologies are typically based on networks with zoned metering, only enabling burst detection at the zoned level. However, due to high renovation costs, most areas in my country have not implemented zoned metering and still use a centralized monitoring model covering the entire pipeline network. Given the large scale and complex topology of these networks, zoned metering in existing technologies is difficult to apply. Summary of the Invention

[0004] This invention is made to solve the above-mentioned problems, and aims to provide a method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components. To this end, the following technical solution is provided.

[0005] This invention provides a method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components. The method includes the following steps: Step S1, sampling and preprocessing real-time high-frequency pressure data from a single monitoring point to generate an original pressure monitoring value matrix.

[0006] Step S2: Use discrete wavelet transform to separate the high and low frequencies of the original pressure monitoring value matrix, retain the high frequency part, and generate the high frequency disturbance value detection column vector at the current moment.

[0007] Step S3: Use the COF algorithm to detect outliers in the column vector by detecting high-frequency disturbance values;

[0008] Step S4: Based on the transient network pressure change characteristics of the pipe burst condition, screen outliers detected, mark outliers that meet the characteristics as anomalies, and correct the pressure values ​​corresponding to the anomalies after the end of the day's testing.

[0009] Step S5: Repeat steps S1-S4. When the number of consecutive abnormal moments of abnormal points that meet the characteristics of pipe bursting exceeds the time threshold, a pipe bursting alarm is issued.

[0010] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention may also have the following feature: wherein step S1 includes the following sub-steps:

[0011] Step S1-1: For the 64Hz high-frequency pressure monitoring data, sampling is performed every 10s. The sampled value is the average value of the pressure monitoring data within that second. The corresponding time is t1=(i,j), which represents the j-th time of the i-th day in the original pressure monitoring dataset.

[0012] Step S1-2: Extract high-frequency pressure monitoring data from m days prior to the time to be detected. The number of samples taken from each monitoring point per day is n. Then, the current time to be detected, t2 = (m, n), represents the nth time on the mth day in the original pressure monitoring dataset.

[0013] Steps S1-3: Preprocess the sampled 10s raw pressure monitoring dataset;

[0014] Steps S1-4 involve dividing the preprocessed raw pressure monitoring dataset into m row vectors of length n, and storing them in chronological order into rows of the raw pressure monitoring value matrix P. For a single monitoring point, the corresponding raw pressure monitoring value matrix P is:

[0015]

[0016] Among them, the element P in the original pressure monitoring value matrix P i,jThis represents the pressure sampling value at time j on day i. The m-th row of P represents the pressure detection value of the monitoring point one day before the time to be detected t2=(m,n).

[0017] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention may also have the following feature: wherein steps 1-3 include the following sub-steps:

[0018] Step S1-3-1: If there are missing historical pressure monitoring data for a period of less than 5 minutes, then perform linear interpolation on the pressure monitoring data within that period to fill in the missing values. If there are missing historical pressure monitoring data for a period of more than 5 minutes, then use the pressure data from the same time the previous day to replace it.

[0019] Step S1-3-2: Using the pressure order of magnitude under normal operating conditions as a reference, set high and low thresholds to remove obviously abnormal pressure monitoring data from the original pressure monitoring dataset.

[0020] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention may also have the following feature: wherein step S2 includes the following sub-steps:

[0021] Step S2-1: Perform multi-level discrete wavelet transform on the original pressure monitoring value matrix P. The decomposition level k is taken as 1 / 2 of the maximum decomposition level of the matrix. One approximation coefficient f and k detail coefficients wi are obtained through k-level wavelet decomposition.

[0022] Step S2-2: Retain the approximation coefficient f and set the detail coefficients wi at each level to 0, then perform k-level wavelet reconstruction to obtain the reconstructed low-frequency pressure value matrix P. 1 ;

[0023] Steps S2-3: Compare the original pressure monitoring value matrix P with the reconstructed low-frequency pressure value matrix P 1 The difference is the high-frequency disturbance value matrix H of that monitoring point.

[0024]

[0025] Among them, the element h in H i,j This represents the high-frequency pressure disturbance value at time j on day i. The m-th row of H represents the high-frequency pressure disturbance value of the monitoring point one day before the time to be detected t2=(m,n).

[0026] Step S2-4: The last column of the high-frequency disturbance value matrix H is the high-frequency disturbance value detection column vector H at the current detection time. d ,

[0027]

[0028] Among them, the high-frequency disturbance value detection column vector H d element h in i h represents the high-frequency disturbance component extracted from the original pressure monitoring value at the same moment on the i-th historical day. m This represents the high-frequency disturbance pressure value at the moment to be detected.

[0029] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by the present invention may also have the following feature: wherein, in step S2-1 multi-level discrete wavelet transform, the wavelet basis function adopts "db4".

[0030] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention may also have the following feature: wherein step S3 includes the following sub-steps:

[0031] Step S3-1, the high-frequency disturbance value detection column vector H d Input COF algorithm;

[0032] Step S3-2: Calculate the high-frequency disturbance detection column vector H d Each detected sample h i Chain distance:

[0033]

[0034] Where m is H d The length of the value is the high-frequency disturbance pressure value extracted simultaneously m days before the time to be detected, dist(e) k ) for detection sample h i The Euclidean distance between the (k-1)th neighbor and the kth neighbor;

[0035] Step S3-3: Calculate the high-frequency disturbance detection column vector H. d Each detected sample h i COF value:

[0036]

[0037] Step S3-4, each detected sample h i The COF values ​​are arranged from largest to smallest absolute value, and the one with the largest absolute value is the detected sample h. i Outliers in;

[0038] Step S3-5, if the high-frequency disturbance value detection column vector H d The last detected sample h i If the COF value is among the selected outliers, proceed to step S4 for anomaly feature identification; otherwise, return to step S1 for the next burst tube detection.

[0039] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention may also have the following feature: wherein step S4 includes the following sub-steps:

[0040] Step S4-1: Calculate the high-frequency disturbance value detection column vector H d The average of the first (m-1) elements excluding the current time. and standard deviation σ:

[0041]

[0042]

[0043] Step S4-2: Determine whether outliers are abnormal using the 3σ principle, and label the detectionResult according to the characteristics of the burst pipe pressure drop and other abnormal pressure characteristics:

[0044]

[0045] Where detectionResult=0 indicates that the pressure is not abnormal at the current detection time, detectionResult=1 indicates that the pressure is abnormally decreasing, and detectionResult=2 indicates that the pressure is abnormally increasing.

[0046] Step S4-3: According to the characteristics of the pressure drop in the burst pipe, when detectionResult = 1, proceed to step S5 to judge the number of consecutive abnormal moments; otherwise, return to step S1 to detect the burst pipe at the next moment.

[0047] Step S4-4: After the detection is completed on the same day, the original pressure monitoring value corresponding to the detection time when detectionResult=1 or 2 is corrected, and the average pressure data at the same time in the previous m days is used as the correction value.

[0048] The water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by the present invention may also have the following features: wherein step S5 includes the following sub-steps: step S5-1, the number of consecutive occurrences of detectionResult=1 at the time to be detected is recorded as n, once detectionResult=0 or 2 occurs, it is considered that no burst has occurred, n is set to 0, and step S1 is entered to detect the burst at the next time.

[0049] Step S5-2: Set a continuous anomaly judgment time threshold. If n exceeds the number of moments corresponding to the time threshold, it is determined that a pipe burst has occurred at that moment, and a pipe burst alarm is issued.

[0050] The role and effect of invention

[0051] The method for detecting pipe bursts in water supply networks based on wavelet decomposition to extract high-frequency pressure components according to the present invention includes the following steps: Step S1, sampling and preprocessing real-time high-frequency pressure data of a single monitoring point to generate an original pressure monitoring value matrix; Step S2, using discrete wavelet transform to separate high and low frequencies in the original pressure monitoring value matrix, retaining the high-frequency part, and generating a high-frequency disturbance value detection column vector at the current moment; Step S3, using the COF algorithm to detect outliers in the high-frequency disturbance value detection column vector; Step S4, based on the transient network pressure change characteristics of the pipe burst condition, filtering out detected outliers, marking outliers that meet the characteristics as anomalies, and correcting the pressure values ​​corresponding to the anomalies after the end of the day's detection; Step S5, repeating steps S1-S4, and issuing a pipe burst alarm when the number of consecutive abnormal moments of anomalies that meet the pipe burst characteristics exceeds a time threshold.

[0052] Therefore, the water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention samples and analyzes high-frequency pressure data, which helps to more accurately reflect the instantaneous pressure changes in the water supply network, sensitively capture the pressure transients caused by burst events, ensure real-time detection, and save valuable time for subsequent repairs. This invention extracts the high-frequency pressure disturbance components caused by bursts, and can specifically separate the data changes caused by bursts, avoiding the impact of pressure fluctuations caused by normal changes in water demand on the pressure drop of bursts, ensuring the accuracy of detection results. Moreover, this invention can achieve real-time detection of bursts in large and complex pipe networks using only pressure monitoring data, making it more suitable for situations where there are relatively many pressure monitoring points and large-scale pipe networks in my country's water supply networks, and where zoned metering is not implemented. At the same time, the pressure sensor is not limited by factors such as pipe diameter and material, and is applicable to various types of water supply networks. Attached Figure Description

[0053] Figure 1 This is a flowchart of a water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components, as described in an embodiment of the present invention. Figure 1 ;

[0054] Figure 2 This is a flowchart of a water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components, as described in an embodiment of the present invention. Figure 2 ;

[0055] Figure 3 This is a flowchart of step S1 in an embodiment of the present invention;

[0056] Figure 4 This is a flowchart of the sub-steps of steps S1-3 in an embodiment of the present invention;

[0057] Figure 5 This is a flowchart of step S2 in an embodiment of the present invention;

[0058] Figure 6 This is a flowchart of step S3 in an embodiment of the present invention;

[0059] Figure 7 This is a flowchart of step S4 in an embodiment of the present invention;

[0060] Figure 8 This is a flowchart of step S5 in an embodiment of the present invention;

[0061] Figure 9 This is the original pressure monitoring data curve of a monitoring point in an embodiment of the present invention; and

[0062] Figure 10 These are high-frequency pressure component curves and low-frequency pressure component curves at a certain monitoring point in an embodiment of the present invention. Detailed Implementation

[0063] To make the technical means, creative features, objectives and effects of this invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components.

[0064] <Example>

[0065] Figure 1 This is a flowchart of a water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components, as described in an embodiment of the present invention. Figure 1 ; Figure 2 This is a flowchart of a water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components, as described in an embodiment of the present invention. Figure 2 .

[0066] like Figure 1 and Figure 2 As shown, the water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by the present invention includes the following steps:

[0067] Step S1: Sample and preprocess the real-time high-frequency pressure data of a single monitoring point to generate an original pressure monitoring value matrix.

[0068] Figure 3 This is a flowchart of step S1 in an embodiment of the present invention.

[0069] like Figure 3 As shown, step S1 includes the following sub-steps:

[0070] Step S1 includes the following sub-steps:

[0071] Step S1-1: For the 64Hz high-frequency pressure monitoring data, sampling is performed every 10s. The sampled value is the average value of the pressure monitoring data within that second. The corresponding time is t1=(i,j), which represents the j-th time of the i-th day in the original pressure monitoring dataset.

[0072] Step S1-2: Extract high-frequency pressure monitoring data from m days prior to the time to be detected. The number of samples taken from each monitoring point per day is n. Then, the current time to be detected, t2 = (m, n), represents the nth time on the mth day in the original pressure monitoring dataset.

[0073] Steps S1-3: Preprocess the sampled 10s raw pressure monitoring dataset;

[0074] Figure 4 This is a flowchart of the sub-steps of steps S1-3 in an embodiment of the present invention.

[0075] like Figure 4 As shown, steps S1-3 include the following sub-steps:

[0076] Step S1-3-1: If there are missing historical pressure monitoring data for a period of less than 5 minutes, then perform linear interpolation on the pressure monitoring data within that period to fill in the missing values. If there are missing historical pressure monitoring data for a period of more than 5 minutes, then use the pressure data from the same time the previous day to replace it.

[0077] Step S1-3-2: Using the pressure order of magnitude under normal operating conditions as a reference, set high and low thresholds to remove obviously abnormal pressure monitoring data from the original pressure monitoring dataset.

[0078] Steps S1-4 involve dividing the preprocessed raw pressure monitoring dataset into m row vectors of length n, and storing them in chronological order into rows of the raw pressure monitoring value matrix P. For a single monitoring point, the corresponding raw pressure monitoring value matrix P is:

[0079]

[0080] Among them, the element P in the original pressure monitoring value matrix P i,j This represents the pressure sampling value at time j on day i. The m-th row of P represents the pressure detection value of the monitoring point one day before the time to be detected t2=(m,n).

[0081] After data preprocessing, discrete wavelet transform is used to separate high and low frequencies in the original pressure detection dataset, and the high-frequency components in the pressure detection data are extracted, which are the specific information of the pressure transient of the burst pipe.

[0082] Figure 5 This is a flowchart of step S2 in an embodiment of the present invention.

[0083] like Figure 5 As shown, step S2 includes the following sub-steps:

[0084] Step S2-1: Perform a multi-level discrete wavelet transform on the original pressure monitoring value matrix P. The decomposition level k is taken as half of the maximum decomposition level of the matrix. Through k-level wavelet decomposition, one approximation coefficient f and k detail coefficients wi are obtained respectively. In this embodiment, the wavelet basis function used in the multi-level discrete wavelet transform is "db4".

[0085] Step S2-2: Retain the approximation coefficient f and set the detail coefficients wi at each level to 0, then perform k-level wavelet reconstruction to obtain the reconstructed low-frequency pressure value matrix P. 1 .

[0086] Steps S2-3: Compare the original pressure monitoring value matrix P with the reconstructed low-frequency pressure value matrix P 1 The difference is the high-frequency disturbance value matrix H of that monitoring point.

[0087]

[0088] Among them, the element h in H i,j This represents the high-frequency pressure disturbance value at time j on day i. The m-th row of H represents the high-frequency pressure disturbance value of the monitoring point one day before the time to be detected t2=(m,n).

[0089] Step S2-4: The last column of the high-frequency disturbance value matrix H is the high-frequency disturbance value detection column vector H at the current detection time. d ,

[0090]

[0091] Among them, the high-frequency disturbance value detection column vector H d element h in i h represents the high-frequency disturbance component extracted from the original pressure monitoring value at the same moment on the i-th historical day. m This represents the high-frequency disturbance pressure value at the moment of detection.

[0092] Pressure data from different dates show similar trends. Therefore, extracting data from the same time on different dates in historical data as outlier detection targets can prevent the daily changes in pressure monitoring values ​​in the pipeline network from masking the pressure fluctuations caused by pipe bursts.

[0093] Figure 6 This is a flowchart of step S3 in an embodiment of the present invention.

[0094] like Figure 6 As shown, step S3 includes the following sub-steps:

[0095] Step S3-1, the high-frequency disturbance value detection column vector H d Input the COF algorithm.

[0096] Step S3-2: Calculate the high-frequency disturbance detection column vector H d Each detected sample h i Chain distance:

[0097]

[0098] Where m is H d The length of the value is the high-frequency disturbance pressure value extracted simultaneously m days before the time to be detected, dist(e) k ) for detection sample h i The Euclidean distance between the (k-1)th neighbor sample and the kth neighbor sample.

[0099] Step S3-3: Calculate the high-frequency disturbance detection column vector H. d Each detected sample h i COF value:

[0100]

[0101] Step S3-4, each detected sample h i The COF values ​​are arranged from largest to smallest absolute value, and the one with the largest absolute value is the detected sample h. i Outliers in the data.

[0102] Step S3-5, if the high-frequency disturbance value detection column vector H d The last detected sample h i If the COF value is among the selected outliers, proceed to step S4 for anomaly feature identification; otherwise, return to step S1 for the next burst tube detection.

[0103] To avoid pressure fluctuations caused by normal water demand changes in the pipeline network being mistaken for pipe bursts and affecting the detection effect, the detected abnormal states can be further screened based on the pipeline network pressure response characteristics under pipe burst events. That is, step S4 is a further abnormal feature identification of the detection time that was judged as an outlier in step S3.

[0104] Figure 7 This is a flowchart of step S4 in an embodiment of the present invention.

[0105] like Figure 7 As shown, step S4 includes the following sub-steps:

[0106] Step S4-1: Calculate the high-frequency disturbance value detection column vector H d The average of the first (m-1) elements excluding the current time. and standard deviation σ:

[0107]

[0108]

[0109] Step S4-2: Determine whether outliers are abnormal using the 3σ principle, and label the detectionResult according to the characteristics of the burst pipe pressure drop and other abnormal pressure characteristics:

[0110]

[0111] In this context, detectionResult=0 indicates that the pressure is not abnormal at the current detection time, detectionResult=1 indicates that the pressure is abnormally decreasing, and detectionResult=2 indicates that the pressure is abnormally increasing.

[0112] Step S4-3: According to the characteristics of the pressure drop in the burst pipe, when detectionResult = 1, proceed to step S5 to determine the number of consecutive abnormal moments; otherwise, return to step S1 to detect the burst pipe at the next moment.

[0113] Step S4-4: After the detection is completed on the same day, the original pressure monitoring value corresponding to the detection time when detectionResult=1 or 2 is corrected, and the average pressure data at the same time in the previous m days is used as the correction value.

[0114] To avoid misjudging occasional abnormal pressure fluctuations as pipe bursts and interfering with subsequent manual maintenance, a pipe burst is only determined when abnormal pressure markers appear consecutively and accumulate to a certain number of times.

[0115] Figure 8 This is a flowchart of step S5 in an embodiment of the present invention.

[0116] like Figure 8 As shown, step S5 includes the following sub-steps:

[0117] Step S5-1: Record the number of consecutive occurrences of detectionResult=1 at the time to be detected as n. Once detectionResult=0 or 2 occurs, it is considered that no tube burst has occurred, n is set to 0, and the process proceeds to step S1 to detect the tube burst at the next time.

[0118] Step S5-2: Set a continuous anomaly judgment time threshold. If n exceeds the number of moments corresponding to the time threshold, it is determined that a pipe burst has occurred at that moment, and a pipe burst alarm is issued.

[0119] In this embodiment, the method for detecting water supply pipe bursts based on wavelet decomposition to extract high-frequency pressure components provided by the present invention is used to illustrate whether a pipe burst occurred at a monitoring point of a city's water supply network on January 15, 2022. According to the maintenance record of a monitoring point of a city's water supply network on January 15, 2022, the report was received at 16:20 on January 15, and the maintenance personnel cut off the water supply at 18:59 to carry out emergency repairs.

[0120] Figure 9 This is the original pressure monitoring data curve of a certain monitoring point in an embodiment of the present invention.

[0121] Step S1: Sample and preprocess the real-time high-frequency pressure data from a single monitoring point to generate an original pressure monitoring value matrix. The sampling frequency for this monitoring point is 64Hz. Pressure monitoring data for the four days preceding 00:00 on January 15, 2022, are collected. Figure 9 As shown, the raw high-frequency pressure data is sampled every 10 seconds, and the sampled value is the average value of the pressure monitoring data within that second. Pressure monitoring data above 100m and below 1m are removed, and short-term missing data are filled by linear interpolation. Long-term missing data are replaced with pressure data from the same time the previous day. The processed data is stored in the raw pressure monitoring value matrix P. There are 8640 pressure data points per day, so the raw pressure monitoring value matrix P has a dimension of 4 rows and 8640 columns.

[0122] Figure 10 These are high-frequency pressure component curves and low-frequency pressure component curves at a certain monitoring point in an embodiment of the present invention.

[0123] Step S2: The original pressure matrix is ​​separated into high and low frequencies using discrete wavelet transform, retaining the high-frequency component to generate a column vector for detecting high-frequency disturbance values ​​at the current moment. The maximum wavelet decomposition level of the pressure monitoring value matrix P is 12, therefore a 6-level wavelet decomposition is performed on the original pressure monitoring value matrix P, yielding one approximation coefficient and six detail coefficients. After setting all detail coefficients to 0, a 6-level wavelet reconstruction is performed. The original pressure monitoring value matrix P and the reconstructed low-frequency pressure value matrix P are then compared. 1 The difference is the high-frequency disturbance matrix H at that monitoring point. The high-frequency pressure component curves and low-frequency pressure component curves obtained after high-frequency and low-frequency separation are shown below. Figure 10 As shown, the upper part is the high-frequency part and the lower part is the low-frequency part. Obviously, the low-frequency pressure component curve represents the normal pressure fluctuation trend, while the high-frequency pressure component curve shows a more obvious drop anomaly at 13:39 than the original pressure curve, which can better reflect the specific information of the tube rupture.

[0124] Step S3: Use the COF algorithm to detect outliers in the pressure disturbance value column vector. Detect the high-frequency disturbance value column vector H at the current moment. dInput the COF algorithm to calculate the COF value at each detection time. The larger the absolute value, the higher the degree of outlier. If the absolute value of the COF value at the current detection time is the largest, it is determined to be an outlier, and proceed to step S4.

[0125] Step S4: Based on the transient network pressure change characteristics under pipe burst conditions, filter out the detected outliers and mark those that meet the characteristics as anomalies. Calculate the high-frequency disturbance detection column vector H. d The average and standard deviation of the first 4 days were used to qualitatively and quantitatively determine outliers based on the 3σ principle and the pressure drop characteristics of the burst pipe. Outliers that met the criteria were marked as detectionResult=1.

[0126] Step S5: Repeat steps S1 to S4. When the number of consecutive abnormal moments that meet the characteristics of a burst pipe exceeds the time threshold, a burst pipe alarm is issued. In this example, the time threshold is set to 4 minutes. At 13:39 on January 15th, the number of moments marked as detectionResult=1 has accumulated for more than 4 minutes, i.e., more than 24 times. Therefore, the output detection result is that a burst pipe has occurred, and a burst pipe alarm is issued at 13:39.

[0127] Therefore, in this embodiment, the water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by the present invention enables the alarm to be issued nearly 3 hours earlier than the time of receiving the report, and the detection result is determined to be correct, thus successfully shortening the time required from the occurrence of the burst to the detection of the burst.

[0128] The role and effect of the embodiments

[0129] The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components according to the present invention includes the following steps: Step S1, sampling and preprocessing real-time high-frequency pressure data from a single monitoring point to generate an original pressure monitoring value matrix. After data preprocessing, high and low frequencies are separated in the original pressure detection dataset using discrete wavelet transform to extract the high-frequency components in the pressure detection data, which are the specific information of burst pipe pressure transients. Step S2, high and low frequencies are separated in the original pressure monitoring value matrix using discrete wavelet transform, retaining the high-frequency part to generate a high-frequency disturbance value detection column vector at the current moment. Pressure data from different dates have similar trends. Extracting data from the same moment on different dates in historical data as outlier detection objects can avoid the daily changes in pressure monitoring values ​​in the network masking the pressure fluctuations caused by burst pipe pressure. Step S3, outliers in the high-frequency disturbance value detection column vector are detected using the COF algorithm. To avoid pressure fluctuations caused by normal water demand changes in the network being mistaken for burst pipes and affecting the detection effect, the detected abnormal states can be further screened based on the network pressure response characteristics under burst pipe events. Step S4 involves further anomaly feature identification of the detected outlier moments. Based on the transient pipeline pressure change characteristics under pipe rupture conditions, outliers are screened, and those matching the features are marked as anomalies. The pressure values ​​corresponding to the anomalies are then corrected after the day's detection is completed. Step S5 repeats steps S1-S4. When the number of consecutive anomaly moments matching the pipe rupture characteristics exceeds a time threshold, a pipe rupture alarm is issued.

[0130] Therefore, the water supply network burst detection method based on wavelet decomposition to extract high-frequency pressure components provided by this invention samples and analyzes high-frequency pressure data, which helps to more accurately reflect the instantaneous pressure changes in the water supply network, sensitively capture the pressure transients caused by burst events, ensure real-time detection, and save valuable time for subsequent repairs. This invention extracts the high-frequency pressure disturbance components caused by bursts, and can specifically separate the data changes caused by bursts, avoiding the impact of pressure fluctuations caused by normal changes in water demand on the pressure drop of bursts, ensuring the accuracy of detection results. Moreover, this invention can achieve real-time detection of bursts in large and complex pipe networks using only pressure monitoring data, making it more suitable for situations where there are relatively many pressure monitoring points and large-scale pipe networks in my country's water supply networks, and where zoned metering is not implemented. At the same time, the pressure sensor is not limited by factors such as pipe diameter and material, and is applicable to various types of water supply networks.

[0131] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components, characterized in that, Includes the following steps: Step S1: Sample and preprocess the real-time high-frequency pressure data of a single monitoring point to generate an original pressure monitoring value matrix; Step S2: Use discrete wavelet transform to separate the high and low frequencies of the original pressure monitoring value matrix, retain the high frequency part, and generate the high frequency disturbance value detection column vector at the current moment; Step S3: Use the COF algorithm to detect outliers in the high-frequency disturbance detection column vector; Step S4: Based on the transient network pressure change characteristics of the pipe burst condition, the detected outliers are screened, the outliers that meet the characteristics are marked as anomalies, and the pressure values ​​corresponding to the anomalies are corrected after the detection is completed on the same day. Step S5: Repeat steps S1-S4. When the number of consecutive abnormal moments at the abnormal points that meet the characteristics of a burst pipe exceeds a time threshold, a burst pipe alarm is issued. Step S1 includes the following sub-steps: Step S1-1: For the 64Hz high-frequency pressure monitoring data, sampling is performed every 10 seconds. The sampled value is the average value of the pressure monitoring data within those 10 seconds. The corresponding time is t1 = (i, j), which represents the j-th time of the i-th day in the original pressure monitoring dataset. Step S1-2: Extract the high-frequency pressure monitoring data from the m days prior to the time to be detected. The number of sampling data points per day is n. Then, the current time to be detected t2 = (m, n) represents the nth time on the mth day in the original pressure monitoring dataset. Steps S1-3: Preprocess the sampled 10-second raw pressure monitoring dataset; Steps S1-4 involve dividing the preprocessed original pressure monitoring dataset into m row vectors of length n, and storing them in chronological order in each row of the original pressure monitoring value matrix P. For a single monitoring point, the corresponding original pressure monitoring value matrix P is: ， Among them, the element P in the original pressure monitoring value matrix P i,j This represents the pressure sampling value at time j on day i. The m-th row of P represents the pressure detection value of this monitoring point one day before the time to be detected, t2 = (m, n). Step S2 includes the following sub-steps: Step S2-1: Perform multi-level discrete wavelet transform on the original pressure monitoring value matrix P. The decomposition level k is taken as 1 / 2 of the maximum decomposition level of the matrix. One approximation coefficient f and k detail coefficients wi are obtained through k-level wavelet decomposition. Step S2-2: Retain the approximation coefficient f and set the detail coefficients wi at each level to 0, then perform k-level wavelet reconstruction to obtain the reconstructed low-frequency pressure value matrix P. 1 ; Step S2-3, the original pressure monitoring value matrix P and the reconstructed low-frequency pressure value matrix P 1 The difference is the high-frequency disturbance value matrix H of that monitoring point. ， Among them, the element h in H i, j This represents the high-frequency pressure disturbance value at time j on day i. The m-th row of H represents the high-frequency pressure disturbance value of the monitoring point one day before the time to be detected t2 = (m, n). Step S2-4: The last column of the high-frequency disturbance value matrix H is the high-frequency disturbance value detection column vector at the current detection time. , ， Among them, the high-frequency disturbance value detection column vector element h in i h represents the high-frequency disturbance component extracted from the original pressure monitoring value at the same moment on the i-th historical day. m This represents the high-frequency disturbance pressure value at the moment of detection.

2. The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components according to claim 1, characterized in that: in, Step S1-3 includes the following sub-steps: Step S1-3-1: If there are missing historical pressure monitoring data for a period of less than 5 minutes, then perform linear interpolation on the pressure monitoring data within that period to fill in the missing values. If there are missing historical pressure monitoring data for a period of more than 5 minutes, then use the pressure data from the same time the previous day to replace it. Step S1-3-2: Using the pressure order of magnitude under normal operating conditions as a reference, set high and low thresholds to remove obviously abnormal pressure monitoring data from the original pressure monitoring dataset.

3. The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components according to claim 1, characterized in that: in, In step S2-1, the wavelet basis function used in the multi-level discrete wavelet transform is "db4".

4. The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components as described in claim 1. Its features are: in, Step S3 includes the following sub-steps: Step S3-1, the high-frequency disturbance value detection column vector Input COF algorithm; Step S3-2: Calculate the high-frequency disturbance value detection column vector. Each detected sample h i Chain distance: ， Where m is The length of the value is calculated as follows: This refers to the extraction of high-frequency disturbance pressure values ​​from m days prior to the time of detection. To detect sample h i The Euclidean distance between the (k-1)th neighbor and the kth neighbor; Step S3-3: Calculate the high-frequency disturbance value detection column vector. Each detected sample h i COF value: ， Step S3-4, each detected sample h i The COF values ​​are arranged from largest to smallest absolute value, and the one with the largest absolute value is the detected sample h. i Outliers in; Step S3-5, if the high-frequency disturbance value detection column vector The last detected sample h i If the COF value belongs to the selected outlier, then proceed to step S4 for anomaly feature identification; Otherwise, return to step S1 to perform the next burst tube detection.

5. The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components, as described in claim 4. Its features are: in, Step S4 includes the following sub-steps: Step S4-1: Calculate the high-frequency disturbance value detection column vector. The average of the first (m-1) elements excluding the current time. and standard deviation σ: ， ， Step S4-2: Determine whether outliers are abnormal using the 3σ principle, and then analyze them according to the characteristics of pipe rupture pressure drop and other abnormal pressure characteristics. mark: ， in, This indicates that the pressure at the current detection time is not abnormal. This indicates an abnormal drop in pressure. This indicates an abnormal increase in pressure; Step S4-3, according to the characteristics of the pressure drop during pipe rupture, when When =1, proceed to step S5 to determine the number of consecutive abnormal moments; otherwise, return to step S1 to detect the burst pipe at the next moment. Step S4-4: After the testing is completed on the same day, for The original pressure monitoring value corresponding to the detection time = 1 or 2 is corrected, and the average pressure data at the same time in the previous m days is used as the correction value.

6. The method for detecting burst pipes in water supply networks based on wavelet decomposition to extract high-frequency pressure components according to claim 5, characterized in that: in, Step S5 includes the following sub-steps: Step S5-1, the time to be detected Let n be the number of consecutive occurrences. If n = 0 or 2, it is considered that no pipe burst has occurred. n is set to 0, and the process proceeds to step S1 to detect pipe burst at the next moment. Step S5-2: Set a continuous anomaly judgment time threshold. If n exceeds the number of moments corresponding to the time threshold, it is determined that a pipe burst has occurred at that moment, and a pipe burst alarm is issued.

Images