A method, system, and medium for leak detection in water supply networks that combines time-frequency domain analysis.

Abstract

This invention discloses a method, system, and medium for detecting leaks in water supply networks by combining time-frequency domain analysis. It relates to the field of network leakage detection technology and includes the following steps: acquiring historical normal water supply pressure data for each target pressure measuring point; performing normalization and frame-segmentation processing to obtain multiple frame window water supply pressure data sequences and obtain window pressure time-domain signals; learning local pressure characteristics of the window; obtaining key information of the data at each frequency; reconstructing frequency domain signal features and time domain signal features to obtain a fitted window pressure water supply data sequence; obtaining a reconstruction loss set; and using the reconstruction loss set to discriminate network leakage events. In this invention, the influence of time-domain and frequency-domain signals on water supply pressure is considered, maximizing the preservation of potentially important features of the water supply network in leakage, thereby solving the problem of low accuracy in leak detection in water supply networks and adapting to different network topologies.

Description

Technical Field

[0001] This invention relates to the field of pipeline leakage detection technology, and in particular to a method, system and medium for detecting leakage in water supply pipelines that combines time and frequency domain analysis. Background Technology

[0002] During normal operation of a pipeline system, pipe ruptures caused by various reasons can lead to changes in pressure and flow. When a pipe leaks, fluid can no longer flow from one end of the rupture to the other at the agreed pressure. At the same time, the flow rate at the node will decrease from the agreed amount of water. The reduced flow rate is the amount of water that is leaked. Pipeline leaks usually refer to leaks that occur gradually in the pipeline due to some reason (such as pipeline aging, valve damage, or other defects). The amount of leakage gradually increases over time until it is finally discovered and measures are taken to stop it. Pipeline leaks cause significant damage to the operation of the city.

[0003] The normal operation of pipeline systems is a crucial infrastructure pillar for rapid urban development. These systems are typically laid underground along buildings, and in older urban areas, the pipeline systems often have decades of history. Their condition is susceptible to various external influences, which can disrupt normal life. Therefore, effective prediction of pipeline pressure is essential. Timely detection of pressure changes and proactive intervention in emergencies can significantly reduce economic losses caused by abnormal pipeline pressure. Effective detection of abnormal leaks in water distribution networks (WDNs) ensures municipal water supply safety and improves urban operational services.

[0004] In existing technologies, leak detection methods for water distribution networks (WDNs) mainly rely on human experience. This places a high demand on skilled professionals, and even highly experienced professionals cannot accurately detect various emergencies. Leak detection methods for water supply networks primarily use historical data analogy. This method often uses historical data features too crudely, failing to detect subtle changes in the data, resulting in low accuracy in leak detection and prediction. Therefore, it is necessary to propose a water supply network leak detection method, system, and medium that incorporates time-frequency domain analysis to solve or at least partially solve the aforementioned technical problems. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, and medium for detecting leaks in water supply networks that combines time-frequency domain analysis, aiming to solve the technical problem of inaccurate detection results in existing water supply network leak detection methods.

[0006] To achieve the above objectives, this invention provides a method for detecting leaks in water supply networks that combines time-frequency domain analysis, comprising the following steps:

[0007] S100: Obtain historical pressure data of normal water supply at each target pressure measurement point within the target time period under each pipeline operation scenario.

[0008] S200, normalize the historical pressure data of normal water supply at the measuring point, divide the normalized historical pressure data of normal water supply at the measuring point into frames according to the preset frame length, obtain multiple frame window water supply pressure data sequences and obtain the window pressure time domain signal. Among them, a frame window water supply pressure data sequence includes multiple node pressure data arranged in time sequence, a preset time step corresponds to one node pressure data, and adjacent frame window water supply pressure data sequences have an overlap length.

[0009] S300 uses an attention model to learn local pressure features of a window based on the time-domain signal of window pressure.

[0010] S400 transforms the time-domain pressure signals of each window to obtain the frequency-domain pressure signals of the windows;

[0011] The window pressure frequency domain signal is divided into high-frequency data, mid-frequency data and low-frequency data according to the frequency. Among them, the mid-frequency data is the frequency data in the range of 40 to 90 Hz.

[0012] By using an attention mechanism, feature learning is performed on high-frequency, mid-frequency, and low-frequency data respectively to obtain key information of the data at each frequency.

[0013] By combining key information from data at various frequencies, a complete reconstructed frequency domain signal feature can be obtained.

[0014] S500 uses inverse Fourier transform to convert the reconstructed frequency domain signal features into reconstructed time domain signal features;

[0015] S600 uses weighted allocation to combine local pressure features of the window and reconstruct time-domain signal features to generate a new fitted window pressure water supply data sequence.

[0016] S700, obtain the window reconstruction loss of each fitted window pressure water supply data sequence and the corresponding frame window water supply pressure data sequence, and all window reconstruction losses constitute the reconstruction loss set.

[0017] S800 uses the reconstructed loss set to identify pipeline leakage events.

[0018] Furthermore, "using the reconstructed loss set to identify pipeline leakage events" specifically includes:

[0019] Obtain historical water supply pressure data of the pipeline network to be tested; process the historical water supply pressure data to be tested to obtain the corresponding original water supply pressure data sequence and the reconstructed water supply pressure data sequence to be tested; if the reconstruction loss value of the reconstructed water supply pressure data sequence and the original water supply pressure data sequence is greater than the peak value of the reconstruction loss set, it is determined that the pipeline network under test has leakage.

[0020] Furthermore, low-frequency data refers to frequency data between 0 and 40 Hz; high-frequency data refers to frequency data between 90 and 128 Hz.

[0021] Furthermore, the weight allocation value is 0.5.

[0022] Furthermore, the target duration is one year, the preset frame length is one week, and the preset time step is thirty minutes.

[0023] Furthermore, the WNTR tool was used to generate pressure data under different pipeline operation scenarios. One pressure data node was generated every half hour, and a pressure dataset with a total duration of one year was generated. The pressure dataset under each pipeline operation scenario was normalized to the [0,1] interval, and the pressure data nodes in the pressure dataset were divided into frames according to a fixed one-week time length. There are overlapping parts between frames. A frame window water supply pressure data sequence includes 336 node pressure data arranged in time sequence.

[0024] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for detecting leaks in a water supply network by combining time-frequency domain analysis.

[0025] The present invention also provides a water supply network leakage detection system combining time-frequency domain analysis, including a processing device, which is used to implement the steps of the above-mentioned water supply network leakage detection method combining time-frequency domain analysis.

[0026] Compared with existing technologies, the water supply network leakage detection method combining time-frequency domain analysis provided by this invention has the following beneficial effects:

[0027] The present invention provides a water supply network leakage detection method combining time and frequency domain analysis. Based on the time local dependency of multiple frame window water supply pressure data sequences, it obtains the time domain and frequency domain features of historical normal water supply pressure data at the measuring point. Specifically, it improves the leakage detection performance of the water supply network by learning high-frequency, medium-frequency, and low-frequency signal features respectively. By weighting and combining the local pressure features of the combined window and reconstructing the time domain signal features, a new fitting window pressure water supply data sequence is generated. Considering the influence of time and frequency domain signals on water supply pressure, it retains the most important features of the water supply network in leakage to the greatest extent, thereby solving the problem of low leakage detection accuracy in water supply networks. It can effectively distinguish leakage points and is adaptable to different network topologies. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0029] Figure 1 The flowchart is a process for the water supply network leakage detection method combining time-frequency domain analysis of the present invention.

[0030] Figure 2 This is a schematic diagram illustrating the principle of the water supply network leakage detection method combining time-frequency domain analysis of the present invention.

[0031] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0032] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0034] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0035] Please refer to Figure 1 and Figure 2 This invention provides a method for detecting leaks in a water supply network combining time-frequency domain analysis, comprising the following steps: S100, acquiring historical normal water supply pressure data of each target pressure measuring point within a target time period under each pipeline operation scenario; S200, normalizing the historical normal water supply pressure data of the measuring points, dividing the normalized historical normal water supply pressure data of the measuring points into frames according to a preset frame length, acquiring multiple frame window water supply pressure data sequences and obtaining window pressure time-domain signals, wherein one frame window water supply pressure data sequence includes multiple node pressure data arranged in time sequence, one preset time step corresponds to one node pressure data, and adjacent frame window water supply pressure data sequences have an overlap length; S300, learning local pressure features of the window based on the window pressure time-domain signals using an attention model; S400, transforming each window pressure time-domain signal to obtain window pressure frequency-domain signals; according to The frequency domain signal of the window pressure is divided into high-frequency data, mid-frequency data, and low-frequency data based on frequency. Mid-frequency data refers to data with frequencies ranging from 40 to 90 Hz. An attention mechanism is used to learn features from the high-frequency, mid-frequency, and low-frequency data to obtain key information for each frequency. This key information is then combined to obtain complete reconstructed frequency domain signal features. In step S500, an inverse Fourier transform is used to convert the reconstructed frequency domain signal features into reconstructed time domain signal features. In step S600, weighted allocation is used to combine the local pressure features of the window and the reconstructed time domain signal features, generating a new fitted window pressure water supply data sequence. In step S700, the window reconstruction loss for each fitted window pressure water supply data sequence and the corresponding frame window water supply pressure data sequence is obtained, and all window reconstruction losses constitute a reconstruction loss set. In step S800, the reconstruction loss set is used to identify pipeline leakage events.

[0036] The present invention provides a water supply network leakage detection method combining time and frequency domain analysis. Based on the time local dependency of multiple frame window water supply pressure data sequences, it obtains the time domain and frequency domain features of historical normal water supply pressure data at the measuring point. Specifically, it improves the leakage detection performance of the water supply network by learning high-frequency, medium-frequency, and low-frequency signal features respectively. By weighting and combining the local pressure features of the combined window and reconstructing the time domain signal features, a new fitting window pressure water supply data sequence is generated. Considering the influence of time and frequency domain signals on water supply pressure, it retains the most important features of the water supply network in leakage to the greatest extent, thereby solving the problem of low leakage detection accuracy in water supply networks. It can effectively distinguish leakage points and is adaptable to different network topologies.

[0037] Understandably, the target duration can be a historical one-year duration, a historical six-month duration, or other durations; the preset frame length can be a one-week duration, a two-week duration, or other durations, and the preset frame length is less than the target duration; the preset time step duration can be a half-hour duration, a one-hour duration, or other durations. The selection in the solution of this invention is based on the actual situation. Optionally, the overlap length is one preset time step.

[0038] Understandably, in the solution of the present invention, the historical water supply pressure data of each target pressure measuring point is normalized and then processed in frames to form multiple frame window water supply pressure data sequences. The local pressure characteristics of the window can be learned by training a model on the multiple frame window water supply pressure data sequences and after the model converges. At the same time, after performing Fourier transform on each frame window water supply pressure data sequence, a corresponding window pressure frequency domain signal is formed. The key information of the data at each frequency can be obtained by training a model on the multiple window pressure frequency domain signals and after the model converges.

[0039] Understandably, water supply pressure data refers to the data obtained by pressure sensors installed at detection nodes on the pipeline. The pressure sensors are used to monitor the pressure signals inside the pipe and send them to the data platform via IoT terminals.

[0040] Further, step S800 specifically includes: acquiring historical water supply pressure data of the pressure measuring points of the pipeline network to be tested; processing the historical water supply pressure data to obtain the corresponding original water supply pressure data sequence and the reconstructed water supply pressure data sequence; if the reconstruction loss value of the reconstructed water supply pressure data sequence and the original water supply pressure data sequence is greater than the peak value of the reconstruction loss set, then it is determined that the pipeline network to be tested has leakage; if the reconstruction loss value of the reconstructed water supply pressure data sequence and the original water supply pressure data sequence is not greater than the peak value of the reconstruction loss set, then it is determined that the pipeline network to be tested has no leakage. Understandably, in the solution of this invention, the historical pressure data of the water supply to be tested in the pipeline network can be obtained using a one-week time window, or a two-week time window or other time lengths. The data is then processed with a preset frame length and a preset time step to obtain the corresponding reconstructed water supply pressure data sequence. Then, based on attention models, frequency division processing, attention mechanisms, etc., the original water supply pressure data sequence and the reconstructed water supply pressure data sequence are obtained. Finally, the reconstruction loss value of the reconstructed water supply pressure data sequence and the original water supply pressure data sequence is obtained. If the reconstruction loss value is greater than the peak value of the reconstruction loss set, it is determined that the pipeline network under test has a leak; otherwise, it is determined that the pipeline network under test has no leak.

[0041] Furthermore, low-frequency data refers to frequency data between 0 and 40 Hz; high-frequency data refers to frequency data between 90 and 128 Hz. In the scheme of this invention, mid-frequency data includes 40 Hz and 90 Hz frequency data, low-frequency data is frequency data lower than mid-frequency data, and high-frequency data is frequency data higher than mid-frequency data. The historical pressure data of normal water supply at each pipeline operation scenario is converted into a frequency domain signal through short-time Fourier transform. The data length of the frequency domain signal is consistent with the original pressure length. At the same time, the frequency domain signal is divided into high, medium, and low parts. The attention mechanism is used to learn features from these three frequency domain segments respectively. The learned frequency domain signals are recombined into a complete frequency domain signal. Finally, the frequency domain signal is converted back into a time domain signal through inverse Fourier transform. This solves the technical problem that high-frequency signals and low-frequency signals are greatly affected by environmental interference factors. If these frequency range data are directly input into the learning model, it will reduce the learning performance of the model and thus affect the accuracy of pressure data prediction.

[0042] Furthermore, a learnable weight parameter is used to weight the local features and frequency domain features respectively, and these weights are then input into a linear activation function to generate a new sequence. Optionally, the local features are multiplied by 'a', and the time domain transformed from the frequency domain is multiplied by (1-a) to generate a new sequence. The new sequence is compared with the original sequence to check the similarity. Since training is performed using leak-free data, the similarity will be high (and the loss value will be small) when there is no leakage. Preferably, the weight allocation value is 0.5.

[0043] In one specific embodiment of the present invention, the target duration is one year, the preset frame length is one week, and the duration of a preset time step is thirty minutes; a frame window water supply pressure data sequence includes 336 node pressure data arranged in time sequence.

[0044] Furthermore, the WNTR tool was used to generate pressure data under different pipeline operation scenarios. A pressure data node was generated every half hour, for a total of one year's worth of pressure data. The pressure data for each scenario was normalized to the [0,1] interval, and the pressure data was divided into frames according to a fixed frame length. There was overlap between frames, and the fixed frame length was set to one week. One frame window of water supply pressure data sequence included 336 (24 2 7=336) node pressure data arranged in time sequence, with an overlap length of one preset time step.

[0045] Please refer to this again. Figure 2 , Figure 2 A single curve represents the data of a pressure station on a pipeline (multiple pressure stations may be installed on the pipeline), which is a multivariate time series. The main concepts of this invention include: normalized pressure station data; applying an attention mechanism to learn the temporal changes of pressure data in a frame window water supply pressure data sequence (window pressure time domain signal); obtaining local pressure features of the window; dividing the obtained window pressure frequency domain signal into high-frequency, mid-frequency, and low-frequency data for feature learning to obtain key information of the data at each frequency; combining the key information of the data at each frequency to obtain complete reconstructed frequency domain signal features; obtaining reconstructed time domain signal features; and finally combining the local pressure features of the window and the reconstructed time domain signal features through weight allocation β.

[0046] This invention provides a method for detecting leaks in water supply networks that combines time-frequency domain analysis, the implementation of which is as follows:

[0047] The WNTR tool was used to collect historical pressure data of normal water supply at monitoring points under various pipeline operation scenarios (urban pipeline operation scenarios, such as city A pipeline operation scenario, city B pipeline operation scenario, etc.), and the historical pressure data of normal water supply at monitoring points was preprocessed and normalized to the [0,1] interval. Specifically, the WNTR tool was used to generate historical pressure data of normal water supply at monitoring points under different pipeline operation scenarios, with one pressure data node every half hour, generating a historical dataset of pressure data nodes for a total of one year. The pipeline networks used in the historical dataset are provided in the form of EPANET INP files. The model parameters (pipeline length, diameter, roughness) of each pipeline network are different to increase the uncertainty of the model data. This invention uses a pressure-driven simulation mode for each leakage scenario, where the amount of water delivered depends on the pressure. At the same time, the basic periodic component of demand is approximated based on the Fourier series (FS) of the actual historical water demand of the water supply enterprise.

[0048] Extract the preprocessed pressure signal features; normalize the pressure data for each scene to the [0,1] interval, divide the data into frames according to a fixed frame length, with overlapping parts between frames, and set the fixed frame length to one cycle, i.e., 24. 2 7 = 336 time points, with an overlap length of one time step; during normalization processing,

[0049]

[0050] L is the frame window size. It is the pressure data under each frame window (the window data sequence of the nth pressure station at the tth time), and the data normalization expression is:

[0051]

[0052] min represents the minimum pressure value in the sequence, and max represents the maximum pressure value in the sequence. It is the smallest constant, which can avoid division by zero when the maximum and minimum values ​​are the same.

[0053] A new fitting window of pressure water supply data sequence is reconstructed using an attention model. Specifically, the reconstruction expression using the attention model is z' = T(z), where z is the original temporal feature pressure sequence input into the attention model, T is the encoder-decoder structure attention model, and z' is the new temporal feature pressure sequence generated after attention reconstruction. In this invention, the pressure data is first concatenated with the data encoding, instead of being summed as in most methods.

[0054]

[0055] pe is the location code corresponding to the pressure value of x1(t), specifically, pe1(t) is the location code corresponding to x1(t);

[0056]

[0057] The attention mechanism mainly calculates the dependence of a value on other stress values, and the final output always carries attention weights.

[0058] The preprocessed pressure signal frequency domain features are extracted. Historical pressure data from normal water supply at each pipeline operation scenario is processed to obtain a window pressure time domain signal, which is then converted to a frequency domain signal using a short-time Fourier transform. The data length of the frequency domain signal is consistent with the original pressure length. Simultaneously, the frequency domain signal is divided into high, medium, and low frequency segments. An attention mechanism is used to learn features from each of these three frequency segments. Finally, the learned frequency domain signals are recombined into a complete frequency domain signal. Finally, an inverse Fourier transform is used to convert the frequency domain signal back to the time domain signal. This invention argues that the highest frequency signal and the low frequency signal are generally more susceptible to environmental interference. Directly inputting data from these frequency segments into the learning model will reduce the model's learning performance, thereby affecting the accuracy of pressure data prediction.

[0059] Learnable weights are used to combine local features (window local pressure features) and periodic features (reconstructed time-domain signal features) to generate a new fitted window pressure water supply data sequence. Optionally, a learnable weight parameter is used to weight the local features and frequency domain features respectively, and these weights are input into a linear activation function to generate a new fitted window pressure water supply data sequence. Specifically, the weight parameter is 0.5 when combining the features.

[0060] Calculate the reconstruction loss between the reconstruction pressure and the original pressure; the expression for calculating the reconstruction loss is: loss = ||z' - z||2, which represents the 2-norm distance between the reconstruction pressure and the pressure data.

[0061] Reconstruction loss is used to identify pipeline leakage events. During the training phase, the model is trained using leak-free data to obtain a series of loss value sets, and the peak value of the loss value set is calculated. During the testing phase, the test loss value is compared with the peak value. If the test loss value is smaller than the peak value, the output is no leakage; otherwise, the output is leakage.

[0062] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the steps of the above-described method for detecting leaks in a water supply network incorporating time-frequency domain analysis.

[0063] The present invention also provides a water supply network leakage detection system combining time-frequency domain analysis, including a processing device, which is used to implement the steps of the above-mentioned water supply network leakage detection method combining time-frequency domain analysis.

[0064] The water supply network leakage detection method, system, and computer storage medium of the present invention, which combines time-frequency domain analysis, have the following advantages: Since only pressure data from pressure stations is considered for anomaly detection without requiring additional information about the network, it can adapt to different network structures; the operating pressure of the network varies in each scenario, therefore, normalizing the pressure data for each scenario effectively captures pressure data changes; considering both time-domain and frequency-domain features on water supply pressure, it preserves the possible characteristics of the network in leakage to the greatest extent; it can detect leaks in the water supply network, and by learning the pressure data features in the time and frequency domains, it can effectively distinguish leakage scenarios, improving the accuracy of water supply network leakage detection.

[0065] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting leaks in water supply networks combining time-frequency domain analysis, characterized in that, Including the following steps: Acquire historical pressure data of each target pressure measuring point under each pipeline operation scenario within the target time period during normal water supply; The historical pressure data of normal water supply at the measuring point is normalized. The normalized historical pressure data of normal water supply at the measuring point is then divided into frames according to a preset frame length to obtain multiple frame window water supply pressure data sequences and obtain window pressure time domain signals. Among them, a frame window water supply pressure data sequence includes multiple node pressure data arranged in time sequence. A preset time step corresponds to one node pressure data. Adjacent frame window water supply pressure data sequences have an overlap length. The local pressure characteristics of the window are learned using an attenuation model based on the time-domain signal of window pressure. The time-domain signals of each window pressure are transformed to obtain the frequency-domain signals of the window pressure. Based on frequency, the frequency-domain signals of the window pressure are divided into high-frequency data, mid-frequency data, and low-frequency data, where the mid-frequency data is frequency data in the range of 40 to 90 Hz. An attention mechanism is used to learn features from the high-frequency, mid-frequency, and low-frequency data to obtain key information for each frequency. The key information from each frequency is combined to obtain the complete reconstructed frequency-domain signal features. The inverse Fourier transform is used to convert the reconstructed frequency domain signal features into reconstructed time domain signal features; We use weighted allocation to combine local pressure features of the window and reconstruct time-domain signal features to generate a new fitted window pressure water supply data sequence. Obtain the window reconstruction loss for each fitted window pressure water supply data sequence and the corresponding frame window water supply pressure data sequence; all window reconstruction losses constitute a reconstruction loss set. The reconstructed loss set is used to identify pipeline leakage events.

2. The water supply network leakage detection method combining time-frequency domain analysis according to claim 1, characterized in that, "Using the reconstructed loss set to identify pipeline leakage events" specifically includes: Obtain historical water supply pressure data of the pressure measuring points in the pipeline network under test; The historical water supply pressure data to be measured is processed to obtain the corresponding original water supply pressure data sequence and the reconstructed water supply pressure data sequence to be measured. If the reconstruction loss value between the reconstructed water supply pressure data sequence to be tested and the original water supply pressure data sequence to be tested is greater than the peak value of the reconstruction loss set, then it is determined that the pipeline network to be tested has leakage; if the reconstruction loss value between the reconstructed water supply pressure data sequence to be tested and the original water supply pressure data sequence to be tested is not greater than the peak value of the reconstruction loss set, then it is determined that the pipeline network to be tested has no leakage.

3. The water supply network leakage detection method combining time-frequency domain analysis according to claim 1, characterized in that, Low-frequency data refers to frequency data between 0 and 40 Hz; high-frequency data refers to frequency data between 90 and 128 Hz.

4. The water supply network leakage detection method combining time-frequency domain analysis according to claim 1, characterized in that, The weight allocation value is 0.

5.

5. The water supply network leakage detection method combining time-frequency domain analysis according to any one of claims 1 to 4, characterized in that, The target duration is one year, the preset frame length is one week, and the preset time step is thirty minutes.

6. The water supply network leakage detection method combining time-frequency domain analysis according to claim 5, characterized in that, The WNTR tool was used to generate pressure data for different pipeline operation scenarios. One pressure data node was generated every half hour, and a pressure dataset with a total duration of one year was generated. The pressure dataset for each pipeline operation scenario is normalized to the [0,1] interval, and the pressure data nodes in the pressure dataset are divided into frames according to a fixed one-week time length. There is an overlap between the frames. A water supply pressure data sequence of a frame window includes 336 node pressure data arranged in time sequence.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the water supply network leakage detection method combining time-frequency domain analysis as described in any one of claims 1 to 6.

8. A water supply network leakage detection system combining time-frequency domain analysis, characterized in that, The device includes a processing apparatus for implementing the steps of the water supply network leakage detection method combining time-frequency domain analysis as described in any one of claims 1 to 6.

Images