Method and device for determining heat supply network leakage risk information

Abstract

This invention provides a method and apparatus for determining leakage risk information in heating pipeline networks, which can be applied to the field of smart heating technology. The method includes: denoising initial pressure information to obtain high-frequency pressure disturbance information; obtaining multiple candidate pressure residuals characterizing the risk of leakage pressure waves based on the high-frequency pressure disturbance information and smoothed disturbance information obtained after smoothing the high-frequency pressure disturbance information; identifying multiple features extracted from multiple windows adjacent to the candidate times to obtain the probability value of the leakage pressure wave; obtaining the start time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window with a probability value greater than a probability threshold and the second pressure amplitude of the second sub-window in the target window; and, in the case of multiple target sensors, determining the number of risky heat exchange stations that meet the leakage conditions and the corresponding leakage risk level based on the start time, leakage triggering information, and center time.

Description

Technical Field

[0001] This invention relates to the field of smart heating technology, specifically to a method and apparatus for determining leakage risk information in heating pipeline networks. Background Technology

[0002] Among related technologies, methods for detecting leaks in heating pipe networks mainly include abnormal water replenishment judgment, pressure threshold alarm, thermal balance analysis, acoustic leak detection, infrared thermometry, tracer detection, and hydraulic model inversion.

[0003] However, in related technologies, low-frequency pressure fluctuations and normal hydraulic regulation can easily mask transient pressure waves caused by leakage; pressure threshold alarms are difficult to distinguish between pressure disturbances caused by leakage, pump frequency adjustment, and valve operation; the judgment of anomalies at a single measuring point lacks reliability and is easily affected by local operating conditions; the time difference relationship between the arrival times of pressure waves from multiple heat exchange stations is not fully utilized; and the leakage location results often only provide a single-point judgment, lacking consideration of the suspected range and confidence level. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method and apparatus for determining leakage risk information of heating pipeline networks.

[0005] According to a first aspect of the present invention, a method for determining leakage risk information of a heating pipeline network is provided, comprising: denoising initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations to obtain high-frequency pressure disturbance information characterizing the leakage risk of each heat exchange station; obtaining multiple candidate pressure residuals characterizing the existence of leakage pressure wave risk based on the high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information, each candidate pressure residual having a corresponding candidate time; identifying multiple features extracted from multiple windows adjacent to the candidate time to obtain the probability value of the leakage pressure wave; obtaining the start time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than a probability threshold and the second pressure amplitude of the second sub-window in the target window, the width of the second sub-window being greater than the width of the first sub-window; and, in the case of multiple target sensors, determining the number of risky heat exchange stations that meet the leakage conditions and the leakage risk level corresponding to the number based on the start time, the leakage triggering information of multiple target sensors, and the center time obtained by clustering the leakage triggering times of each target sensor.

[0006] A second aspect of the present invention provides a device for determining leakage risk information of a heating pipeline network, comprising: a denoising module for denoising initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations to obtain high-frequency pressure disturbance information characterizing the leakage risk of each heat exchange station; a residual determination module for obtaining multiple candidate pressure residuals characterizing the presence of leakage pressure wave risk based on the high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information, each candidate pressure residual having a corresponding candidate time; and an identification module for extracting multiple features obtained from multiple windows adjacent to the candidate times. The system performs identification to obtain the probability value of the leakage pressure wave; the time determination module is used to determine the start time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than the probability threshold and the second pressure amplitude of the second sub-window in the target window, where the width of the second sub-window is greater than the width of the first sub-window; the level determination module is used to determine the number of risk heat exchange stations that meet the leakage conditions and the corresponding leakage risk level when there are multiple target sensors, based on the start time, the leakage triggering information of multiple target sensors and the center time obtained by clustering the leakage triggering times of each target sensor.

[0007] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0008] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0009] A fifth aspect of the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0010] According to embodiments of the present invention, high-frequency pressure disturbance information is extracted by denoising, effectively filtering out the masking of leakage transient pressure waves by low-frequency pressure fluctuations and normal hydraulic regulation. Based on residual analysis and multi-window feature identification between high-frequency disturbance information and smoothed disturbance information, the problem of traditional single pressure thresholds being unable to distinguish leakage, pump frequency regulation and valve action disturbances is avoided. Furthermore, the start time is determined by comparing the amplitudes of double-width sub-windows within the target window, and combined with the start time of multiple sensors, leakage trigger information and cluster center time, the reliability of single-measuring point anomaly judgment is improved. The time difference relationship between multiple measuring points is fully utilized to quantify the number of risk heat exchange stations and the corresponding leakage risk level, providing a basis for risk range and confidence level for the location results. Attached Figure Description

[0011] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0012] Figure 1 An application scenario diagram of the heating network leakage risk information determination method and apparatus according to an embodiment of the present invention is shown;

[0013] Figure 2 A flowchart of a method for determining leakage risk information in heating pipe networks according to an embodiment of the present invention is shown;

[0014] Figure 3 A structural block diagram of a device for determining leakage risk information in heating pipe networks according to an embodiment of the present invention is shown. Detailed Implementation

[0015] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0016] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0017] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0018] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0019] Urban centralized heating systems are characterized by large-scale pipeline networks, numerous pipe sections, a high proportion of underground installations, complex operating conditions, and significant seasonal load fluctuations. As the pipeline network ages, factors such as pipe corrosion, weld defects, compensator failure, valve seal damage, construction disturbances, and foundation settlement can all cause leakage. Leakage not only leads to increased water replenishment on the primary side (heat source side), heat loss, and increased operating energy consumption, but can also cause heating imbalances, localized undersupply, road flooding, equipment damage, and even safety accidents.

[0020] In related technologies, methods for detecting leaks in heating pipe networks mainly include abnormal water supply judgment, pressure threshold alarm, thermal balance analysis, acoustic leak detection, infrared thermometry, tracer detection, and hydraulic model inversion. However, acoustic leak detection relies on empirical information, resulting in low investigation efficiency and difficulty in covering concealed pipe sections; water supply and pressure threshold methods are limited to identifying obvious operational anomalies and are difficult to accurately locate leak points; infrared thermometry and tracer methods are greatly limited by site conditions, pipe network burial depth, and construction conditions; although traditional hydraulic model inversion can use operational data for diagnosis, it has high requirements for model accuracy, boundary conditions, and operational stability, and its location accuracy is limited in scenarios of sudden leaks or high-frequency transient disturbances.

[0021] When a leak occurs in a heating network, transient pressure disturbances are generated near the leak point and propagate in multiple directions along the pipeline as pressure waves. Theoretically, if high-frequency pressure sensors are installed at multiple heat exchange stations or key nodes, the location of the leak can be deduced from the propagation time difference of the pressure waves. However, in actual heating operation, pump frequency adjustments, valve opening changes, pressure fluctuations on the heat source side, local hydraulic regulation, sensor noise, and changes in user-side heat consumption can all cause pressure fluctuations of varying degrees. These pressure fluctuations include both low-frequency slow changes and high-frequency disturbances not caused by leaks. If only simple thresholds or single-station anomaly criteria are used, false alarms, missed alarms, or location errors are likely to occur.

[0022] Therefore, the key to locating leaks in heating pipe networks lies not only in detecting pressure fluctuations, but also in distinguishing the sources of these fluctuations, identifying the pressure waveforms truly caused by pipe leaks or sudden ruptures, and using the differences in the arrival times of the first wave among multiple heat exchange stations for reverse tracing. Traditional technologies lack a systematic processing flow for high-frequency pressure signals from heating pipe networks, especially lacking a complete technical solution that unifies low-frequency disturbance removal, pressure spike extraction, leak waveform identification, multi-station consistency judgment, and reverse location of pipe network topology.

[0023] In view of this, embodiments of the present invention provide a method and apparatus for determining leakage risk information of a heating pipeline network. The method includes: denoising initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations to obtain high-frequency pressure disturbance information characterizing the leakage risk of each heat exchange station; obtaining multiple candidate pressure residuals characterizing the existence of leakage pressure wave risk based on the high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information, each candidate pressure residual having a corresponding candidate time; identifying multiple features extracted from multiple windows adjacent to the candidate time to obtain the probability value of the leakage pressure wave; obtaining the starting time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than the probability threshold and the second pressure amplitude of the second sub-window in the target window, the width of the second sub-window being greater than the width of the first sub-window; and determining the number of risky heat exchange stations that meet the leakage conditions and the leakage risk level corresponding to the number of stations based on the starting time, the leakage triggering information of multiple target sensors, and the center time obtained by clustering the leakage triggering times of each target sensor, when there are multiple target sensors.

[0024] According to embodiments of the present invention, high-frequency pressure disturbance information is extracted by denoising, effectively filtering out the masking of leakage transient pressure waves by low-frequency pressure fluctuations and normal hydraulic regulation. Thus, based on residual analysis and multi-window feature identification between high-frequency disturbance information and smoothed disturbance information, the problem of traditional single pressure thresholds being unable to distinguish between leakage, pump frequency regulation, and valve action disturbances is avoided. The starting time is determined by comparing the amplitudes of dual-width sub-windows within the target window, and combined with the starting time of multiple sensors, leakage trigger information, and cluster center time, the reliability of single-measuring-point anomaly judgment is improved. The time difference relationship of multiple measuring points is fully utilized to quantify the number of risky heat exchange stations and the corresponding leakage risk level, providing a basis for the risk range and confidence level of the location results.

[0025] Figure 1 The diagram illustrates an application scenario of the method and apparatus for determining leakage risk information in heating pipe networks according to an embodiment of the present invention.

[0026] like Figure 1 As shown, application scenario 100 according to this embodiment may include a detection device 101, a network 102, and a server 103. The network 102 is used as a medium to provide a communication link between the detection device 101 and the server 103. The network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0027] The detection device 101 can be a sensor installed on the primary side supply / return water pipelines of each heat exchange station in the heating network, including sensors that trigger leakage events. Each detection device 101 is bound to a unique heat exchange station node. For example, a threaded high-frequency pressure transmitter can be used, which can be installed directly using the existing pipeline pressure measurement interface of the heat exchange station without the need for additional pipeline excavation.

[0028] Server 103 can be a processor (such as an edge computing / cloud computing unit) deployed in a heating dispatch center. It has a built-in pipeline topology database (storing topology parameters such as nodes, pipe segments, lengths, and wave speeds) and a leakage diagnosis algorithm module. It is the logical execution entity that performs the entire process of noise reduction, feature recognition, time-based picking, and risk assessment.

[0029] It should be noted that the method for determining the leakage risk information of the heating network provided in this embodiment of the invention can generally be executed by server 103. Correspondingly, the device for determining the leakage risk information of the heating network provided in this embodiment of the invention can generally be installed in server 103. The method for determining the leakage risk information of the heating network provided in this embodiment of the invention can also be executed by a server or server cluster that is different from server 103 but can communicate with detection device 101 and / or server 103. Correspondingly, the device for determining the leakage risk information of the heating network provided in this embodiment of the invention can also be installed in a server or server cluster that is different from server 103 but can communicate with detection device 101 and / or server 103.

[0030] It should be understood that Figure 1 The number of testing devices, networks, and servers shown is merely illustrative. Any number of testing devices, networks, and servers can be included depending on implementation needs.

[0031] Figure 2 A flowchart of a method for determining leakage risk information in heating pipe networks according to an embodiment of the present invention is shown.

[0032] like Figure 2 As shown, the method for determining the leakage risk information of the heating pipeline network in this embodiment includes operations S210 to S250.

[0033] In operation S210, the initial pressure information obtained by combining multiple pressure disturbance information acquired by the sensors of multiple heat exchange stations is denoised to obtain high-frequency pressure disturbance information that characterizes the pipeline leakage risk of each heat exchange station.

[0034] In embodiments of the present invention, the pressure disturbance information can be the raw pressure signal acquired in real time by the detection device, including various information such as normal operating pressure information, leakage disturbance information, operational disturbance information, and sensor noise. The initial pressure information can be a multi-channel pressure dataset formed by aligning the pressure disturbance information according to timestamps. Denoising can include rolling median detrending, high-pass / band-pass filtering, or first-order differential processing. The high-frequency pressure disturbance information can be the denoised signal, including high-frequency transient pressure wave components such as leakage and sudden pressure relief.

[0035] In operation S220, based on high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing high-frequency pressure disturbance information, multiple candidate pressure residuals characterizing the risk of leakage pressure waves are obtained. Each candidate pressure residual has a corresponding candidate time.

[0036] In embodiments of the present invention, the smoothed perturbation information can be obtained by performing moving mean smoothing or moving median smoothing on high-frequency pressure perturbation information. The candidate pressure residual can be anomaly residual points obtained after filtering using an adaptive threshold of Median Absolute Deviation (MAD) and robust outlier detection (e.g., Hampel anomaly detection). The candidate time can be the sampling time corresponding to the candidate pressure residual.

[0037] In operation S230, multiple features extracted from multiple windows adjacent to the candidate time are identified to obtain the probability value of the leakage pressure wave.

[0038] In embodiments of the present invention, the multiple windows can be local waveform windows cropped around the candidate time points. The multiple features can include amplitude features, energy features, time features, morphological features, and anomaly scores, etc. The probability value of the leakage pressure wave can be obtained by identifying multiple features using a recognition model to obtain a score recognition result.

[0039] In operation S240, based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than the probability threshold, and the second pressure amplitude of the second sub-window in the target window, the starting time of the pipeline leakage pressure wave arriving at the target sensor is obtained. The width of the second sub-window is greater than the width of the first sub-window.

[0040] In embodiments of the present invention, the target window can be a candidate window whose probability value is greater than a probability threshold, and is confirmed as a suspected leakage waveform window. The first sub-window and the second sub-window can be two sub-intervals of different durations divided within the target window, with the first sub-window having a smaller width and the second sub-window having a larger width. The first pressure amplitude and the second pressure amplitude can be pressure amplitude statistics (e.g., mean, maximum value) within the first sub-window and pressure amplitude statistics within the second sub-window, respectively.

[0041] The initial arrival time can be the first arrival time obtained after initial triggering using the Short-Term Average Ratio (STA / LTA) and processing using the Akaike Information Criterion (AIC). The STA / LTA is used to determine whether a sudden change in the signal has occurred by calculating the ratio of the signal mean within the short-term window to the signal mean within the long-term window. The Akaike Information Criterion can be used to determine the moment when the signal statistical characteristics change abruptly by assuming that the background noise and the event signal are two different stationary random processes and finding the local minimum location of the AIC function; this moment is the initial arrival time. The target sensor can be a sensor with a probability value greater than a threshold that successfully picks up the initial arrival time, i.e., the sensor that triggers the missed event.

[0042] When operating S250, if there are multiple target sensors, the number of risk heat exchange stations that meet the leakage conditions and the corresponding leakage risk level are determined based on the start time, leakage trigger information of multiple target sensors, and the center time obtained by clustering the leakage trigger times of each target sensor.

[0043] In embodiments of the present invention, the leakage triggering information can be a set of associated data such as the target sensor's identifier, start time, probability value, and anomaly score. The center time can be the result of a sliding window clustering, i.e., the cluster center of multiple start times (such as the mean or median). The number can be determined by the number of target sensors that trigger leakage events within the time window. The leakage risk level can be a level based on the number of risky heat exchange stations (e.g., Level 1 is high risk, Level 2 is medium risk).

[0044] For example, using high-frequency pressure sensors deployed on the primary side of multiple heat exchange stations as detection devices, and based on the nodes, pipe sections, lengths, pipe diameters, and sensor installation locations of the heating network, a process for determining or diagnosing leakage risk information in the heating network can be constructed, which includes high-frequency pressure acquisition, low-frequency disturbance removal, extraction of moving smooth difference peaks, leakage waveform identification, multi-station consistency judgment, pressure wave back propagation location, and risk result output.

[0045] First, high-frequency pressure data acquisition and pipeline topology modeling can be performed. High-frequency pressure sensors are installed at multiple heat exchange stations or key nodes of the heating network to collect primary side pressure signals in real time. The sampling frequency of the sensors can be determined according to the network scale, pipe segment length, and pressure wave propagation velocity, preferably a sampling frequency that can capture the arrival time of the first wave of leakage transient pressure waves. The acquired data includes at least the sensor number, heat exchange station name, installation node number, sampling time, and pressure value. Pipeline topology data is read, including node sets, pipe segment sets, pipe segment start and end points, pipe segment length, pipe diameter, and pressure wave propagation velocity parameters, to obtain multiple pressure disturbance information, which are then combined to obtain the initial pressure information.

[0046] Secondly, the initial pressure information is denoised. This involves removing low-frequency pressure disturbances and extracting high-frequency pressure waves. High-pass filtering, band-pass filtering, rolling median detrending, moving average detrending, or combinations thereof can be used to remove low-frequency pressure changes. Filtering is used to suppress low-frequency fluctuations caused by normal operating pressure, slow changes in heat source side pressure, stable pump operation, small-range valve adjustments, and slow load changes on the user side, while retaining high-frequency pressure disturbance information (high-frequency transient pressure waves) caused by pipeline leakage, rupture, or sudden pressure relief.

[0047] After obtaining high-frequency pressure disturbance information, the candidate pressure residuals are obtained by moving smoothing difference. The filtered high-frequency pressure disturbance signal is then subjected to moving smoothing to form a local smoothed baseline. The difference between the filtered signal and the local smoothed baseline is used to obtain the pressure residual signal. Candidate pressure residuals at different time points are identified from the pressure residual signal through adaptive thresholding, anomaly detection, local extremum judgment, and short-term energy surge detection.

[0048] Multiple features (e.g., amplitude, energy, time, morphology, and anomaly score) are extracted from multiple windows adjacent to the candidate time points. These extracted features are then input into a waveform recognizer. The waveform recognizer can employ rule-based thresholding or a machine learning classification model. After successful recognition, the starting time point is further accurately identified using short-to-long window ratio, gradient abrupt changes, AIC information criterion, or combinations thereof.

[0049] The suspected leakage pressure wave events identified by each heat exchange station are sorted and clustered according to their start and arrival times. If the number of heat exchange stations that trigger the event reaches the minimum number of trigger stations within a preset time window, it is determined that a suspected leakage event at the pipeline level has occurred; if only a single heat exchange station or a few local measuring points trigger the event, it is determined to be a local disturbance, sensor noise, or a non-leakage event.

[0050] For example, high-frequency pressure sensors (sampling frequency 200Hz) are deployed in the primary side water supply pipelines of 8 heat exchange stations to collect pressure disturbance information. After aligning with the timestamps, the initial pressure information is obtained. The rolling median with a window length of 2s is used to detrend the pressure and filter out the slowly varying components of the heat source pressure to obtain high-frequency pressure disturbance information. The high-frequency pressure disturbance information is smoothed by the moving average with a window length of 500ms to obtain smoothed disturbance information. The difference between the two is used to obtain the pressure residual. Based on the MAD adaptive threshold, 6 candidate pressure residuals are selected, corresponding to 6 candidate time points.

[0051] For each candidate moment, a 0.5s window is extracted before and after it, and multiple features are extracted (e.g., amplitude, STA / LTA ratio, duration, anomaly assessment value, gradient strength feature). A rule-based recognizer can be used as the identification model, setting multiple thresholds. When all conditions are met, the probability of judgment is 1 (leakage probability threshold = 0.5), resulting in three target windows. Within each target window, a first sub-window of 20ms and a second sub-window of 200ms are divided, and the first and second pressure amplitudes are calculated respectively. When the STA / LTA ratio is greater than 2.5, the initial trigger point is determined. The AIC information criterion is then used for refinement to obtain the start time of the three target sensors. Sliding window clustering (window half-width = 0.1s) is performed on the three start times to obtain the center time. The number of target sensors within the window is counted as 3, satisfying the leakage condition, and the sensor is classified as a level 2 medium-risk sensor, corresponding to three risk heat exchange stations.

[0052] Understandably, by filtering out low-frequency components such as normal operating pressure and slow changes in heat sources through a noise reduction step, the problem of traditional pressure threshold alarm methods being easily masked by low-frequency fluctuations and having a high false alarm rate is solved. Through a two-stage screening process of "residual extraction and multi-feature recognition," leakage waves are distinguished from non-leakage disturbances such as pump frequency adjustment, valve action, and sensor noise, overcoming the deficiency of traditional single-station threshold judgment in being unable to distinguish the source of disturbance. Through multi-station consistency judgment (statistical analysis of the number of risk heat exchange stations), misjudgments caused by local interference at a single station (such as sensor failure or local hydraulic regulation) are avoided. Compared with single-point acoustic leakage listening methods, this improves the anti-interference capability of event recognition.

[0053] In one feasible embodiment, a high-pass filter can be used to obtain high-frequency pressure disturbance information, utilizing the lower cutoff frequency. Suppressing slow low-frequency variations. A bandpass filter can also be used at the lower cutoff frequency. With upper cutoff frequency The frequency bands associated with leakage transient pressure waves are retained between them.

[0054] To further highlight the abrupt change edges of the pressure wave, a first-order difference can be performed on the high-frequency pressure disturbance signal to obtain the pressure difference value. Time gradients can also be used. Characterizing the rate of pressure change between adjacent sampling points, the leakage pressure wave is sudden, so the time gradient will show obvious abrupt changes near the arrival of the first wave, thus determining the high-frequency pressure disturbance information.

[0055] According to embodiments of the present invention, high-frequency pressure disturbance information is extracted by denoising, effectively filtering out the masking of leakage transient pressure waves by low-frequency pressure fluctuations and normal hydraulic regulation. Thus, based on residual analysis and multi-window feature identification between high-frequency disturbance information and smoothed disturbance information, the problem of traditional single pressure thresholds being unable to distinguish between leakage, pump frequency regulation, and valve action disturbances is avoided. The starting time is determined by comparing the amplitudes of dual-width sub-windows within the target window, and combined with the starting time of multiple sensors, leakage trigger information, and cluster center time, the reliability of single-measuring-point anomaly judgment is improved. The time difference relationship of multiple measuring points is fully utilized to quantify the number of risky heat exchange stations and the corresponding leakage risk level, providing a basis for the risk range and confidence level of the location results.

[0056] As we understand it, the above text has already explained how to determine the leakage risk information of the heating network from the perspective of the overall technical solution. The following section will further explain how to obtain high-frequency pressure disturbance information.

[0057] According to an embodiment of the present invention, the initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations is denoised to obtain high-frequency pressure disturbance information characterizing the pipeline leakage risk of each heat exchange station. This includes: combining reference pressure change information characterizing the normal operating pressure of the pipeline network, pressure fluctuation information caused by leakage or pressure relief, pressure disturbance information caused by equipment adjustment, and random disturbance information caused by sensor noise from multiple pressure disturbance information to obtain initial pressure information; and subtracting the noise information caused by multiple reference pressure changes from the initial pressure information to obtain high-frequency pressure disturbance information.

[0058] In embodiments of the present invention, the reference pressure change information can be the normal operating pressure of the pipeline network and information on slowly changing pressures, such as slowly changing pressure on the heat source side, stable operating pressure of the pump set, slow fluctuations in user-side load, and pressure changes caused by small-scale conventional valve adjustments. Pressure fluctuation information can be transient pressure disturbances caused by pipeline leakage, sudden pressure relief, or rupture, characterized by strong abrupt changes, fast propagation speed, and slow attenuation. Pressure disturbance information can be pressure disturbances caused by pump set frequency regulation, large-scale valve opening and closing, and active adjustment on the heat source side, characterized by strong controllability and spectral characteristics different from leakage waves.

[0059] Random disturbance information can be random pressure fluctuations caused by sensor sampling jitter, electromagnetic interference, and transmission errors, characterized by irregularity and small amplitude. Noise information can be slowly varying components estimated from the reference pressure change information through methods such as rolling median and filtering. Denoising can be a process of removing low-frequency components from the initial pressure information.

[0060] During normal operation of the heating network, slow pressure changes are caused by heat source regulation, circulation pump operation, user-side load variations, and small-scale valve movements. To avoid interference from these low-frequency pressure changes in leakage detection, this invention removes the low-frequency baseline from the original pressure signal. For example, high-frequency pressure sensors are deployed at multiple heat exchange stations in the primary heating (source side) network. These sensors are preferably installed in locations within the primary supply or return water pipes where pressure changes are relatively stable, maintenance is convenient, and the network pressure wave propagation characteristics can be reflected.

[0061] No. The raw pressure signal from each pressure sensor consists of the sampling time and the pressure value. The total number of sampling points is denoted as M, and the total number of sensors is denoted as N. Sampling frequency. The time interval Δt between adjacent sampling points can be determined using the conventional reciprocal relationship.

[0062] The j-th sampling time It can be determined from the start time. The sampling sequence number j and the sampling time interval Δt are determined. The initial pressure information can be obtained by combining multiple pressure disturbance information acquired by sensors from multiple heat exchange stations, expressed as follows: , Indicates the first Initial pressure information from each sensor; This indicates the normal operating pressure of the pipeline network and the slowly changing pressure components. This indicates the pressure wave disturbance component caused by leakage, sudden pressure drop, or pipeline rupture. This indicates the pressure disturbance component caused by operational changes such as pump frequency adjustment, valve opening and closing, and heat source side regulation. This represents sensor noise, sampling jitter, and random interference components.

[0063] For example, a low-frequency pressure baseline can be constructed using the rolling median. This is used to describe the slowly changing portion of the original pressure, thus obtaining noise information. The rolling window length can be determined based on the sampling time interval Δt and the window half-width. The window can be configured to cover low-frequency variations without smoothing out the initial wave of leakage. The initial pressure information, also known as the high-frequency pressure disturbance signal, is obtained by subtracting the low-frequency pressure baseline from the original pressure signal, as shown in formula (1) below:

[0064] (1);

[0065] in, It can represent the first The sensor at the first High-frequency pressure disturbance signal at any given time; This represents the original pressure signal; This represents noise information, specifically the low-frequency pressure baseline. Noise reduction can effectively suppress slow pressure fluctuations in the heating system while preserving short-term pressure spikes caused by leakage or sudden pressure drops.

[0066] For example, a city's branched primary heating network has 12 heat exchange stations, and each heat exchange station's water supply pipeline is equipped with one high-frequency pressure sensor. The pressure disturbance information collected by each sensor includes four categories: reference pressure change information, pressure fluctuation information caused by leakage or depressurization, pressure disturbance information caused by equipment adjustment, and random disturbance information caused by sensor noise.

[0067] The pressure disturbance information from all 12 sensors was aligned according to the sampling time and combined to obtain the initial pressure information, covering the pressure time-series data of all 12 nodes in the entire pipeline network. A rolling median with a window length of 2 seconds was used to process the initial pressure information of each sensor to calculate the low-frequency pressure baseline for that sensor, which represents the noise information caused by various reference pressure changes. This baseline fully covers the slow-varying characteristics of the reference pressure changes and does not include the transient abrupt changes of leakage waves. Subtracting this low-frequency pressure baseline from the initial pressure information yields the high-frequency pressure disturbance information.

[0068] According to embodiments of the present invention, traditional techniques directly determine leakage based on the absolute value of the original pressure, without distinguishing the source of pressure fluctuations. This often leads to misjudging slow changes in heat source pressure or user load as leakage, or masking small-amplitude leakage transients within the normal operating pressure. Embodiments of the present invention, through four-component decomposition and targeted noise reduction, isolate only non-leakage reference pressure changes. This avoids false alarms caused by low-frequency slow changes while preserving the true characteristics of leakage transients, thus reducing the risk of missed alarms. The noise information caused by various reference pressure changes is dynamically estimated based on the actual operating reference pressure of each heat exchange station. This adaptively matches the pressure characteristics of different stations and stages, eliminating the need for repeated manual parameter adjustments and demonstrating strong adaptability.

[0069] According to an embodiment of the present invention, based on high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing high-frequency pressure disturbance information, multiple candidate pressure residuals characterizing the risk of leakage pressure waves are obtained, including: determining the difference between high-frequency pressure disturbance information and smoothed disturbance information as the pressure residuals of each sensor at multiple times; and determining the pressure residuals among the multiple pressure residuals whose absolute value is greater than the pressure threshold as candidate pressure residuals.

[0070] In embodiments of the present invention, the pressure residual can be used to counteract the interference of local stable fluctuations, amplify the abrupt change characteristics of leakage transient waves, transform absolute pressure anomalies into relative trend deviations, and adapt to the pressure baseline differences of different heat exchange stations. The pressure threshold can be a critical value used to distinguish between normal pressure fluctuations and abnormal abrupt changes, and can be adaptively adjusted according to the noise level of different heat exchange stations.

[0071] After obtaining high-frequency pressure disturbance information, a moving average smoothing interpolation method can be used to extract peak events. For example, a method for smoothing high-frequency pressure disturbance information to obtain smoothed disturbance information includes: performing moving average smoothing on the high-frequency pressure disturbance signal to obtain locally smoothed pressure disturbance information, denoted as... As smoothing perturbation information; the length of the moving smoothing window. Based on the sampling time interval and window half width Alternatively, to enhance robustness to outliers, moving median smoothing can be used to obtain moving median smoothed pressure perturbation information. This serves as information for smoothing perturbations.

[0072] For example, the pressure residual is obtained by subtracting the high-frequency pressure disturbance information from the smoothed disturbance information. As shown in the following formula (2):

[0073] (2);

[0074] in, Indicates the first The sensor at the first The pressure residual at any given moment; Indicates the first The sensor at the first High-frequency pressure disturbance signal at any given time; This represents smoothed perturbation information, also known as locally smoothed pressure perturbation information. It can be understood that if moving median smoothing is used, the pressure residual can also be obtained from the difference between high-frequency pressure perturbation information and moving median smoothed pressure perturbation information using the same approach.

[0075] In one feasible embodiment, to accommodate different noise levels at heat exchange stations, a local window can be used. The noise intensity is estimated by the median of the residuals and the absolute deviation of the median, and the pressure threshold is obtained as shown in the following formula (3):

[0076] (3);

[0077] in, Indicates the first The sensor at the first The pressure threshold at any given moment; This represents the threshold coefficient, for example, a value between 3 and 6; This represents the standard deviation estimate, also known as the robust standard deviation estimate α.

[0078] When satisfied In the case of, These are labeled as candidate pressure spike moments, also known as candidate moments. Let t be the total number of candidate spike moments identified on the i-th sensor. _{p,i} Where p = 1, 2, ..., P _i P _i is the total number of candidate peak moments for the i-th sensor. This represents the absolute value of the pressure residual.

[0079] Traditional methods directly determine whether the initial pressure exceeds a fixed threshold. However, the pressure baselines of different heat exchange stations vary significantly (up to 1.6 MPa at near-end stations and only 0.8 MPa at far-end stations), requiring station-by-station threshold adjustments. This is labor-intensive and prone to false alarms / missed alarms due to improper threshold settings. The embodiments of this invention employ relative residuals and adaptive thresholds, eliminating the need to focus on absolute pressure magnitude. They only assess the degree of deviation of pressure from local trends, automatically adapting to pressure differences between different heat exchange stations without parameter tuning. By smoothing and filtering out high-frequency noise before calculating the residual, the signal is effectively pre-denoised, reducing false alarms caused by random noise while preserving the abrupt changes in leakage transient waves. This overcomes the drawback of traditional differential operations amplifying sensor noise and triggering numerous false anomalies.

[0080] According to an embodiment of the present invention, the method further includes: obtaining multiple abnormal evaluation values ​​indicating the degree of deviation between the multiple pressure residuals and the reference pressure within the preset window based on multiple pressure residuals and the standard deviation estimate representing the fluctuation amplitude of residual information within the preset window; determining the abnormal evaluation value among the multiple abnormal evaluation values ​​whose absolute value is greater than the evaluation threshold as the target abnormal evaluation value, so as to identify multiple features extracted from multiple windows adjacent to the candidate time based on the target abnormal evaluation value.

[0081] In embodiments of the present invention, the preset window can be a local sliding window, centered on the current sampling time, covering a continuous sampling interval of tens to hundreds of milliseconds. The window length is less than the minimum duration of the leakage wave and greater than the duration of the sensor noise. The residual information fluctuation amplitude can be the dispersion of all pressure residuals within the local sliding window, quantified using the median absolute deviation (MAD), which can avoid the interference of extreme values ​​on the fluctuation estimation. The standard deviation estimate can be a robust standard deviation estimate, obtained by multiplying the median absolute deviation of the residuals within the preset window by the normal distribution conversion factor (1.4826).

[0082] The reference pressure can be the median of the residuals within a local window, i.e., the median value of all pressure residuals within a preset window, representing the expected residual benchmark when there are no abrupt changes within the window. The anomaly assessment value can be a robust outlier detection index based on the median and median absolute deviation (MAD). The target anomaly assessment value is used to perform a secondary screening of candidate residuals, eliminating low-significance noise interference and retaining only candidate moments worthwhile for subsequent feature identification.

[0083] For example, an anomaly assessment value is obtained by calculating the Hampel anomaly score. Robust outlier detection metrics based on the median and median absolute deviation (MAD) are used to measure the degree of anomaly of the current sampling point relative to the local background. The Hampel anomaly score is not sensitive to the outlier itself and is more suitable for signals containing spikes or abrupt changes.

[0084] Calculate the anomaly assessment value for each sampling point As shown in the following formula (4):

[0085] (4);

[0086] in, Indicates the first One sensor in The anomaly assessment value at any given time; This represents the median of the residuals within the local window; This represents the estimated standard deviation. This indicates the smallest positive number that prevents the denominator from being zero, for example, taking... .when When the detection point exceeds the evaluation threshold, it exhibits significant anomalies.

[0087] For example, a branched primary heating network in a northern city has 12 heat exchange stations, each equipped with a high-frequency pressure sensor on its water supply pipeline. A 1-second local window (covering 200 sampling points) is set for each sampling time. The window length is much longer than the duration of sensor noise (e.g., <10ms) and much shorter than the minimum duration of leakage waves (e.g., >20ms). The median absolute deviation (MAD) is calculated for the 200 pressure residuals within each window, and a robust standard deviation estimate is obtained. Simultaneously, the median of the residuals within the window is calculated as the reference pressure.

[0088] Calculate the Hampel anomaly score for each pressure residual as the anomaly assessment value. Set an assessment threshold, and mark the anomaly assessment values ​​whose absolute values ​​are greater than the assessment threshold as the target anomaly assessment values, resulting in 4 candidate time points. Extract features from the local waveform windows of these 4 time points for identification.

[0089] Traditional methods directly use the absolute value of the residual to determine anomalies. Occasional sensor impulse noise (such as a single-point 0.1MPa jump caused by electromagnetic interference) may be misjudged as leakage. The embodiments of the present invention use the statistical characteristics within the window to evaluate the anomaly. Single-point noise will not increase the overall standard deviation estimate or significantly change the window reference pressure. Therefore, it will not be judged as a high anomaly score, thus reducing the false alarm rate caused by impulse noise.

[0090] According to an embodiment of the present invention, multiple features include: an upper limit of amplitude, an energy ratio between a first sub-window and a second sub-window, waveform duration, an anomaly assessment value indicating the degree of deviation between multiple pressure residuals and a reference pressure within a preset window, and waveform gradient intensity indicating the change of pressure residual information within the preset window over time. Multiple features extracted from multiple windows adjacent to candidate times are identified to obtain a probability value for a leakage pressure wave, including: determining the probability value of a leakage pressure wave as a preset value when the upper limit of amplitude is greater than an amplitude threshold, the energy ratio is greater than an energy ratio threshold, the waveform duration is between a lower limit of duration and an upper limit of duration, the anomaly assessment value is greater than an anomaly assessment threshold, and the waveform gradient intensity is greater than an intensity threshold.

[0091] In embodiments of the present invention, the amplitude characteristics of the local waveform may include a maximum positive peak, a maximum negative peak, a peak-to-peak value, and an upper limit of amplitude. The upper limit of amplitude may be the maximum absolute value of the pressure residual within adjacent windows of the candidate time point. The energy ratio may refer to the energy ratio of a short-window to a long-window, where the first sub-window is a short-time window (e.g., 20ms wide) and the second sub-window is a long-time window (e.g., 200ms wide), and the ratio between the energy of the short-time window and the energy of the long-time window is used as the energy ratio. The waveform duration may be the length of the time period during which the pressure residual continuously exceeds the amplitude threshold.

[0092] The anomaly assessment value can be a standardized value representing the deviation of the pressure residual from the reference pressure within a preset window. The waveform gradient intensity can be the maximum rate of change of the pressure residual over time within a local waveform window. The preset value can be a deterministic output value obtained when the identification model is a regular identifier, for example, a preset value of 1 (or a normalized 1, representing a 100% probability of leakage). It should be noted that the amplitude threshold, energy ratio threshold, lower duration limit, upper duration limit, anomaly assessment threshold, and intensity threshold can be flexibly adjusted according to the pipeline network type and operating conditions; specific values ​​are not limited here.

[0093] For example, a local waveform window of length sum before and after the candidate time point is extracted for subsequent feature extraction. The local waveform energy and the energy ratio of the short and long windows are used to characterize the degree of sudden increase in transient pressure wave energy. When the energy ratio of the short and long windows increases significantly, it indicates a sudden increase in pressure energy within a short period of time, which may indicate the existence of a transient pressure wave event. The waveform duration, rise time, and decay time are determined by the threshold exceedance start and end times and the candidate time point, respectively.

[0094] For example, the ratio of positive and negative peak amplitudes and the ratio of rise / fall time can be used to describe the symmetry and morphology of waveforms. Leakage pressure waves can exhibit a pattern of sudden arrival, rapid rise, and subsequent decay. Therefore, characteristics such as the rise / fall time ratio and the ratio of positive and negative peak amplitudes can be used to distinguish leakage waves from other operational disturbance waves.

[0095] In one feasible implementation, a regular waveform recognizer is used as the recognition model. A candidate waveform can be identified as a suspected leakage pressure wave when it simultaneously meets the following conditions, as shown in formula (5):

[0096] (5);

[0097] in, Indicates the upper limit of the amplitude of the candidate waveform; The amplitude threshold representing the minimum effective waveform; Indicates the energy ratio; Indicates the energy ratio threshold; Indicates the duration of the waveform; This indicates the lower limit of the duration of the leakage waveform; This indicates the upper limit of the duration of the leakage waveform; Indicates an abnormal assessment value; Indicates the evaluation threshold; Indicates the waveform gradient intensity; This represents the intensity threshold. The waveform gradient intensity can be taken as the maximum intensity of the pressure residual signal changing over time within a local waveform window.

[0098] Traditional leakage detection relies solely on whether the absolute pressure value exceeds a fixed threshold, making it difficult to distinguish the morphological differences between leakage waves, pump frequency modulation waves, and valve action waves. This often leads to misjudging hydraulic regulation during normal operation as leakage. The embodiments of this invention use a five-dimensional feature joint judgment to cover the core characteristics of leakage waves: "sufficiently large amplitude, rapid energy increase, moderate duration, high anomaly significance, and steep leading edge." This effectively filters out various non-leakage disturbances and significantly reduces the false alarm rate.

[0099] According to an embodiment of the present invention, multiple features extracted from multiple windows adjacent to candidate times are identified to obtain the probability value of leakage pressure wave. The method further includes: determining the value of each feature step by step along the branch nodes of multiple decision trees until the leaf node is reached, and obtaining the score corresponding to the leaf node; summing the scores of each decision tree to obtain the cumulative score and mapping it to a preset interval to obtain the probability value of leakage pressure wave.

[0100] In embodiments of this invention, the decision tree can be a base learner for a recognition model (e.g., a random forest). Each decision tree can be a tree structure that makes binary judgments based on specific features. Branch nodes can be non-leaf nodes within the decision tree, corresponding to a judgment condition for a feature. Samples move to the left or right child node based on the judgment result. Leaf nodes can be the terminal nodes of the decision tree, where no further branches occur. Each leaf node is associated with a score, representing the confidence level (e.g., 0.9) or sample proportion of the sample reaching that node that belongs to the "missing wave".

[0101] The score of each decision tree can be the result of independent voting by each decision tree in the random forest, which consists of multiple distinct decision trees. The cumulative score can be the average (or weighted average) of the output scores of all decision trees. The preset interval can refer to a continuous interval of [0,1], used to map the cumulative score (such as the average between 0 and 1) to a numerical value that conforms to the definition of probability theory.

[0102] For example, machine learning models can be used to identify or classify multiple features to construct a feature set of candidate waveforms. As shown in the following formula (6):

[0103] (6);

[0104] in, Indicates the first A set of features for candidate waveforms from a sensor; Indicates peak-to-peak value; Represents local waveform energy; Indicates the rise time of the waveform; Indicates the duration of waveform decay; Indicates the ratio of positive to negative peak amplitude; Indicates the ratio of rise time to decay time; This represents the transpose of a vector.

[0105] Input the feature vector into the classification model: ,in, Indicates the first The probability value of a candidate waveform from a sensor belonging to a leakage pressure wave, ranging from 0 to 1; This represents a waveform classification model, such as a random forest, support vector machine, gradient boosting tree, convolutional neural network, or lightweight neural network.

[0106] When satisfied Then it can be determined that the first Several sensors detected a pressure wave suspected of causing a leak. Among them, This represents the leakage probability threshold, for example, a value between 0.5 and 0.9.

[0107] According to embodiments of the present invention, traditional identifiers rely on manually set fixed thresholds, and the rules are prone to failure once the pipeline network conditions change (such as adding a pumping station or changing the operating mode) or an unseen type of leakage occurs (such as slow seepage). Embodiments of the present invention, by automatically learning the complex boundaries between leakage waves and non-leakage waves from historical data, can identify subtle feature differences that are difficult to list using rules, significantly enhancing generalization ability.

[0108] According to an embodiment of the present invention, the above method further includes: sequentially extracting a local pressure waveform sequence of a preset length composed of multiple sampled values ​​within a target window; dividing the pressure waveform sequence into a preceding subsequence and a following subsequence using multiple candidate segmentation points; multiplying the variance of the preceding subsequence and the variance of the following subsequence corresponding to each candidate segmentation point by a penalty factor and summing them to obtain a value characterizing the data matching degree between the preceding and following subsequences of the multiple candidate segmentation points; determining the segmentation point corresponding to the lower limit value among the multiple values ​​as the sequence number of the updated first wave arrival point in the local pressure waveform sequence; updating the start time of the pipeline leakage pressure wave arriving at the target sensor based on the sequence number and the sampling time interval to obtain the update time, and determining the number of risk heat exchange stations that meet the leakage conditions based on the update time.

[0109] In embodiments of the present invention, the local pressure waveform sequence can be a pressure residual sequence truncated forward and backward for a fixed duration, centered on the initial trigger time of the short-term average ratio (STA / LTA). The length n is the total number of sampling points, for example, 40-500 sampling points (corresponding to a duration of 0.1-1s). The candidate segmentation point can be a possible first wave point in the local pressure waveform sequence, traversing the non-boundary region in the middle of the sequence (to avoid distortion in variance estimation at both ends). The candidate segmentation point number is denoted as k. The preceding subsequence can be the subsequence before the candidate segmentation point k, denoted as y1,...,y k This corresponds to the stable pressure period before the arrival of the first wave (containing only background noise / normal operation disturbances). The subsequent subsequence can be the subsequence after the candidate segmentation point k, denoted as y. k+1 ,...,y n This corresponds to the sudden pressure segment after the arrival of the first wave (including leakage transient wave characteristics).

[0110] The penalty factor can include the length of the preceding subsequence k and the length of the following subsequence nk-1, used to penalize excessively short split points. The split point corresponding to the lower limit value can be the split point with the smallest AIC value among all candidate split points. The sequence number is the position or location identifier of the discrete sampling point in the local sequence. The sampling time interval is the time difference between two adjacent sampling points, which can be determined by the reciprocal of the sampling frequency. The update time can be the arrival time of the first wave after the update and optimization.

[0111] For example, for candidate events identified by waveform recognition, the STA / LTA method can be used to obtain the initial trigger time (i.e., the start time). When the amplitude or energy of the short window relative to the long window suddenly increases to the threshold λ, the corresponding sampling time is taken as the initial trigger time.

[0112] The method for updating and optimizing the start time based on the Akaike Information Criterion (AIC) includes: selecting a local sequence of length n near the start time as the AIC optimization target. For candidate split points... The optimization method for the AIC criterion is shown in the following formula (7):

[0113] (7);

[0114] in, Indicates the first Sensors at candidate segmentation points The AIC value at a given point is a numerical value used to characterize the data matching degree between the preceding and following subsequences of multiple candidate split points; Indicates the candidate split point number; Indicates the length of the local sequence; Represents the variance operator; This represents the waveform sequence before the split point, i.e., the preceding subsequence; This represents the waveform sequence after the segmentation point, i.e., the subsequent subsequence. The method for determining the sequence number of the first wave arrival point in the local pressure waveform sequence is as follows: (8)

[0115] (8);

[0116] in, Indicates the first The sequence number of the first arrival point of each sensor in the local pressure waveform sequence; Let represent the independent variable that minimizes the objective function. The method for determining the arrival time of the first wave from each sensor is shown in the following formula (9):

[0117] (9);

[0118] in, Indicates the first The arrival time of the first wave was identified by each sensor; Indicates the start time of the AIC local window; Indicates the first The sequence number of the first arrival point of each sensor in the local pressure waveform sequence; Indicates the sampling time interval.

[0119] Taking a branched primary heating network in a northern city with 12 heat exchange stations as an example, the sampling frequency fs = 200Hz and the sampling interval Δt = 5ms. The sensor at heat exchange station 5, through waveform recognition, obtains a target window with a probability value of 0.92. The window's start time... The time interval is 10.00s, and the window length is 0.8s. Centered on the initial trigger point of STA / LTA, 0.1s is taken forward and 0.1s is taken backward to obtain a local pressure waveform sequence Y5 with a length of n=40 (containing 40 sample values), covering the abrupt change interval near the first wave.

[0120] The candidate split point k is set to a value range of 20~39 (avoiding the boundaries of the two ends of the sequence), and Y5 is divided into preceding subsequences y1,...,y k and subsequent subsequence y k+1 ,...,y 40 For each k, calculate the variance of the preceding subsequence multiplied by the penalty factor k, and the variance of the following subsequence multiplied by the penalty factor (40 - k - 1), summing them to obtain the AIC value AIC5(k) (i.e., the numerical value representing the matching degree). Due to the drastic change in the first wave of the burst tube, the variance of the following subsequence is much higher than that of the preceding one, and the AIC value shows a significant minimum at k=77. Find the lower limit (minimum value) of the AIC value. =77, meaning the updated first wave has the sequence number 77 in the local sequence. The update time t5 is calculated to be 10.00s + (77 - 1) × 5ms = 10.38s, which corrects the 5ms system deviation of the initial trigger point / start time.

[0121] Pressure waves generated by leakage are often accompanied by multiple reflections, with the main peak usually arriving later than the first wave and having a higher amplitude. Traditional peak detection methods take the point with the largest absolute pressure value as the first wave, which can lead to positioning errors. The embodiments of this invention are based on statistical characteristic mutation detection. Regardless of the amplitude of the first wave, it can be identified as long as the statistical characteristics of its preceding and following subsequences change significantly, which conforms to the physical laws of pressure wave propagation. By comparing the statistical characteristics of the subsequences before and after the segmentation point, only points with sufficiently significant differences in variance before and after the segmentation point are selected, effectively eliminating the interference of local noise and improving the reliability of first wave acquisition.

[0122] According to an embodiment of the present invention, based on the start time, leakage triggering information of multiple target sensors, and the center time obtained by clustering the leakage triggering times of each target sensor, the number of risk heat exchange stations that meet the leakage conditions and the leakage risk level corresponding to the number are determined, including: when the absolute value of the difference between the start time and the center time is less than or equal to a preset duration, the number of risk heat exchange stations that meet the leakage conditions is determined; when the number of risk heat exchange stations is greater than a number threshold, the leakage risk level is determined as the pipeline risk level.

[0123] In embodiments of the present invention, the preset duration can be half the width of a time consistency window, or a fault tolerance threshold set according to the pipeline network scale, pressure wave propagation speed, and time synchronization accuracy, for example, a value of 0.05~0.2s. The number of risk heat exchange stations that meet the leakage conditions can be the number of triggering stations, determined by the total number of target sensors that trigger leakage events within the center time ± preset duration window. The number threshold can be the minimum number of triggering stations, a judgment threshold set according to the pipeline network topology redundancy, for example, a value of not less than 3.

[0124] For example, pressure spikes at a single heat exchange station can be caused by local equipment operation or sensor noise. Therefore, this embodiment of the invention employs a multi-heat exchange station event consistency judgment mechanism to reduce the false alarm rate. After waveform recognition, the first... The event trigger flag for each sensor is shown in the following formula (10):

[0125] (10);

[0126] in, Indicates the first Does each sensor trigger a suspected leakage event? Indicates the probability of identifying the missing waveform; Indicates the threshold for the probability of leakage; Indicates an abnormal assessment value; This represents the evaluation threshold. The number of triggering stations that meet the consistency condition is determined by the following formula (11):

[0127] (11);

[0128] in, Indicates The number of sensors that trigger suspected leakage events within the central time window; Indicates the first Does each sensor trigger a suspected leakage event? This indicates an indicator function that takes the value 1 if the condition is true and 0 otherwise. Indicates the first The arrival time of the first wave from each sensor; Indicates the central moment of the candidate event; This indicates the half-width of the time window that triggers consistency. (The last part, "when the consistency is met," appears to be an error and doesn't translate directly.) In such cases, a suspected pipeline-level leakage event is determined to have occurred. This indicates the minimum number of triggering stations, preferably no less than 3.

[0129] In one feasible embodiment, in order to improve the subsequent positioning quality, the sensor pool can be selected based on the arrival time of the first wave and the waveform recognition confidence level, as shown in the following formula (12):

[0130] (12);

[0131] Where S represents the set of sensors used for reverse positioning; Indicates before selection A sorting operator for each sensor; This represents the probability of missing data identification. Preferably, when sorting, sensors with earlier arrival of the first wave, higher waveform identification probability, and higher anomaly score are selected first.

[0132] Traditional multi-station positioning directly uses a fixed time difference threshold, which lacks consideration for time fluctuations caused by sensor time synchronization errors, initial wave pickup deviations, and pressure wave propagation path differences. This can easily lead to misjudging different sensor trigger times of the same leakage event as independent events. The embodiments of the present invention obtain a dynamic center time through clustering, and with a preset fault-tolerant window, adapt to the time deviation characteristics of different pipe networks, significantly improving robustness.

[0133] According to an embodiment of the present invention, the method further includes: based on the pipeline network topology information, comparing the deviation between the reference time of the leakage pressure wave reaching each target sensor and the propagation time of the pressure wave reaching each target sensor at the candidate leakage node in the pipeline network; selecting nodes with a deviation less than a deviation threshold from multiple candidate leakage nodes as risk leakage nodes, wherein the risk leakage node has a leakage range; and using the combined information obtained by combining the number of target sensors, the start time of the pressure wave propagation from the risk leakage node to each target sensor, and the leakage range as a diagnostic result and outputting it.

[0134] In embodiments of the present invention, the pipeline network topology information can be a weighted undirected graph, including a set of nodes (heat sources, heat exchange stations, branches, valve nodes) and a set of pipe segments (the start and end points, length, diameter, and pressure wave propagation speed of each pipe segment). The pressure wave propagation time of a single pipe segment can be considered as a weight. The reference time can be the theoretical propagation time, characterizing the time it takes for a pressure wave to propagate from a candidate leak node to a target sensor node along the shortest path of the pipeline network, calculated by a shortest path algorithm combining the length of each pipe segment and the propagation time required for the pressure wave to pass through the pipe segment. The propagation time can also be the actual propagation time obtained based on measured data, which is the difference between the estimated start time of each target sensor and the estimated leak occurrence time corresponding to the candidate leak node.

[0135] Deviation can be a time difference residual (e.g., root mean square time difference residual, mean absolute time difference residual, composite time difference residual), which is a statistically quantified value of the difference between the reference duration and the propagation duration. The deviation threshold is a critical value for screening risky leakage nodes. Candidate leakage nodes can be all nodes in the pipeline node set, i.e., all potential points that could become leaks. Risky leakage nodes can be candidate leakage nodes in the candidate node set whose deviation is less than the deviation threshold. The leakage range can be a suspected leakage range, representing the spatial area composed of all risky leakage nodes and their connected pipe segments, rather than a single node.

[0136] The combined information can be a structured diagnostic result, which is a data packet containing information such as the number of combined target sensors, the start time of each target sensor, the range of leakage, the order of candidate nodes, and the location reliability. The diagnostic result can be a conclusion on the leakage event, including whether leakage has occurred, the risk level, the location information, and the investigation guidelines.

[0137] For example, the pipeline network topology information can be abstracted into a weighted undirected graph. , This represents a set of nodes, including heat source nodes, heat exchange station nodes, pipeline branch nodes, valve nodes, and pipe segment connection nodes. This represents a set of pipe segments, with each edge corresponding to a segment of the heating pipeline. For any pipe segment, its start point, end point, length, pipe diameter, and pressure wave propagation velocity are used as basic attributes in subsequent propagation time calculations. The pressure wave propagation time on a pipe segment can be expressed as... , This can represent the pressure wave passing through the pipe section. Required transmission time; Indicates the length of the pipe section; This represents the propagation velocity of the pressure wave in the pipe section. The propagation velocity can be based on empirical values, simulation values, or historical event calibration values, and can be corrected according to pipe diameter, wall thickness, pipe material, water temperature, and operating pressure. The i-th sensor and its installation node... A one-to-one mapping is established, and the node can be used as the sensor location for subsequent backpropagation.

[0138] For example, after identifying a suspected pipeline network leak, reverse localization can be performed based on the pipeline network topology and the arrival times of the first wave from multiple sensors. This includes calculating the propagation time from candidate leak nodes to sensors, and for any candidate leak node... Candidate missing nodes To the sensor nodes The duration of transmission is , This indicates that the pressure wave originates from the candidate leakage node. Spread to the sensor nodes Duration of transmission; Indicates from candidate leak nodes To the sensor nodes A pipeline route; Representing a path A section of the pipeline; Indicates pipe section The duration of transmission; This represents the minimum propagation time among all reachable paths. The corresponding pipeline distance from the candidate leak node to the sensor node can be obtained by summing the lengths of each pipe segment on the shortest path. The aforementioned shortest propagation time and shortest pipeline distance can be calculated using classical graph theory algorithms or other graph shortest path algorithms.

[0139] If candidate missing nodes If the node is a real leakage node, then the first... The first arrival time of each sensor satisfy , Indicates the time when the leakage occurred; Indicates candidate missing nodes To the sensor nodes Duration of transmission; This represents the time residual caused by initial wave pickup error, wave velocity error, and noise. Due to... Unknown, but can be applied to each candidate missing node. Estimate the time when the corresponding leakage occurs.

[0140] For a sensor set S, candidate missing nodes The estimated time of leakage is as follows (13):

[0141] (13);

[0142] in, Indicates if candidate missing nodes The estimated time of leakage occurrence when the leakage point is the location of the leakage point; Indicates the number of sensors involved in the positioning process; Indicates the sensor number; Indicates the first The arrival time of the first wave from each sensor; Indicates candidate missing nodes To the One sensor The duration of its spread.

[0143] In one feasible embodiment, the median can be used to estimate candidate missing nodes. The corresponding leakage time is shown in the following formula (14):

[0144] (14);

[0145] in, This represents an estimate of the leakage occurrence time based on the median; Represents the median operator; Indicates by the first The leakage time was obtained by reverse engineering from the sensors.

[0146] In one feasible embodiment, by calculating the arrival time consistency residual, the interference of single-point anomalies on the overall judgment can be avoided, and candidate missing nodes can be identified. The root mean square time residual is calculated using the following formula (15):

[0147] (15);

[0148] in, Indicates candidate missing nodes The root mean square time of arrival residual. The mean absolute time difference residual and the measured time difference residual between any two sensors can be used as auxiliary residual terms to improve robustness to picking up anomalous first waves. The combined time difference residual can be expressed as: , Indicates candidate missing nodes The overall positioning residual; This represents the root mean square time residual; This represents the mean absolute time difference residual; This represents the residual time difference between the actual arrival times of any two sensors; Let represent the residual weighting coefficients, and satisfy . It is understandable that the smaller the residual of a candidate missing node, the stronger its explanatory power for the arrival times of the first wave from multiple sensors, and the more likely that node is a missing location.

[0149] For example, selecting nodes with deviations less than a deviation threshold from multiple candidate leakage nodes as risk leakage nodes includes: assuming the minimum comprehensive residual among all candidate leakage nodes is... , This represents the minimum composite residual among all candidate missing nodes; Indicates candidate missing nodes The overall positioning residual; Let represent the set of nodes. Then the set of candidate missing nodes can be defined as follows: , Indicates candidate missing nodes; This indicates the allowed residual expansion threshold. The range of candidate missing nodes can also be adjusted using a relative residual expansion coefficient.

[0150] In another feasible embodiment, in order to reduce the impact of the initial pickup error of individual sensors on the positioning result, the embodiments of the present invention adopt a multi-sensor combination voting mechanism. Sensor combinations of different sizes are selected from the sensor pool S, as shown in the following formula (16):

[0151] (16);

[0152] in, Indicates by A sensor combination consisting of individual sensors; Indicates the first indivual Sensor combination; This indicates the pool of sensors involved in the positioning process; Indicates the number of sensors in the combination; Indicates the size of the sensor combination, such as 3, 4, 5 or more.

[0153] According to embodiments of the present invention, by combining and enumerating multiple sensors, the limitation of traditional multi-station positioning relying on the input of all sensors is avoided: if some sensors experience initial pickup errors, equipment failures, or communication interruptions, traditional methods can easily lead to positioning failure. However, the embodiments of the present invention, through enumeration and combination, are equivalent to cross-validating the sensor inputs. Even if some sensors are abnormal, the positioning results of the remaining valid combinations can still be retained, which greatly improves the robustness of the positioning system in complex operating environments.

[0154] For each combination It can independently calculate the set of candidate leaky nodes. Candidate missing nodes The voting value can be defined as , Indicates candidate missing nodes The combined voting value; This indicates the minimum sensor combination size, preferably not less than 3; Indicates the maximum sensor combination size; Indicates size The number of sensor combinations; It represents the portfolio size and also serves as the portfolio weight; This indicates the indicator function, if the candidate missing node Belongs to combination The set of candidate missing nodes is set to 1 if it is obtained, and 0 otherwise. Indicates the first indivual The set of candidate leaky nodes obtained by combining sensors.

[0155] Candidate missing nodes The average residual is shown in the following formula (17):

[0156] (17);

[0157] in, Indicates candidate missing nodes Average positioning residual in multi-sensor combinations; Indicates the first indivual Candidate missing nodes under sensor combination The overall positioning residual; The indicator function is represented. Risk leakage nodes can be determined by prioritizing voting value and then suboptimal residual, as shown in the following formula (18):

[0158] (18);

[0159] in, This indicates the node at risk of leakage, i.e., the node most likely to leak. Indicates candidate missing nodes The vote count; Indicates candidate missing nodes The average residual; Represents the normalization function; Indicates the residual penalty weight; This represents the node that maximizes the objective function.

[0160] The method for determining the leakage range of a risky leakage node includes: the maximum voting value among all candidate leakage nodes is denoted as... The pipeline distance D(v, from the candidate leakage node to the risky leakage node) The leakage range of a risk leakage node can be determined by the shortest path length. The calculation method is shown in the following formula (19):

[0161] (19);

[0162] in, Indicates candidate missing nodes; Indicates the node's voting value; This represents the consensus ratio threshold, with a preferred value of 0.6-0.9. Indicates the maximum vote value; Indicates the average residual at each node; This indicates the maximum permissible average residual; This represents the distance in the pipeline from the candidate leaking node to the most likely leaking node; This indicates the maximum pipeline extension distance of the suspected leak area. This method can output both the most likely leak point and the suspected leak area simultaneously, without forcibly compressing the results into a single node, which better meets the needs of on-site inspection and emergency repair of heating pipelines.

[0163] The final diagnostic results may include: leakage event detection results, number of triggered heat exchange stations, arrival time of the first wave of each sensor, waveform identification confidence of each sensor, most likely leakage node, suspected leakage range, candidate node ranking, average time difference residual, location confidence, and pipeline network diagnostic diagram.

[0164] In one feasible implementation, when calculating location confidence, the node voting values ​​can first be normalized to... Normalize the average residual to And calculate the average missed identification probability of the positioning sensors. Overall confidence level Can be defined as , The range of values ​​can be normalized to 0 to 1; This represents the normalized vote score; Indicates the average probability of identifying the leakage waveform; This represents the normalized residual score; Represents the weighting coefficients, satisfying .if If the value is high, output the most likely leak point; if... If the scores of multiple candidate nodes are insufficient or close, output the leakage range. .

[0165] The location error can be evaluated by predicting the pipeline distance between the leaking node and the actual leaking node, and further statistical analysis can be performed on the average error, the location accuracy under the distance threshold, and the hit rate of the suspected range. In practical applications, the diagnostic results can be displayed on the heating dispatching platform or the geographic information system platform. The system output can include: whether a leak event has occurred, the number of triggered heat exchange stations, the arrival time of the first wave at each heat exchange station, the probability of identifying the leak waveform at each sensor, the most likely leaking node, the suspected leak range, the candidate node ranking table, the comprehensive confidence score, and a pipeline diagnosis visualization. Among these, the suspected leak range can be represented by a semi-transparent area, highlighted candidate nodes, or bolded along the pipe segment; the most likely leaking point can be displayed with a prominent icon; and the triggered heat exchange station can be displayed using a sensor icon.

[0166] Based on the above-mentioned method for determining leakage risk information in heating pipe networks, this invention also provides a device for determining leakage risk information in heating pipe networks. The following will be combined with... Figure 3 The device is described in detail.

[0167] Figure 3 A structural block diagram of a device for determining leakage risk information in heating pipe networks according to an embodiment of the present invention is shown.

[0168] like Figure 3 As shown, the heating network leakage risk information determination device 300 in this embodiment includes a noise reduction module 310, a residual determination module 320, an identification module 330, a time determination module 340, and a level determination module 350.

[0169] The denoising module 310 is used to denoise the initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations, so as to obtain high-frequency pressure disturbance information characterizing the leakage risk of pipelines of each heat exchange station. In one embodiment, the denoising module 310 can be used to perform the operation S210 described above, which will not be repeated here.

[0170] The residual determination module 320 is used to obtain multiple candidate pressure residuals characterizing the risk of leakage pressure waves based on high-frequency pressure disturbance information and smoothed disturbance information obtained by smoothing high-frequency pressure disturbance information. Each candidate pressure residual has a corresponding candidate time. In one embodiment, the residual determination module 320 can be used to perform the operation S220 described above, which will not be repeated here.

[0171] The identification module 330 is used to identify multiple features extracted from multiple windows adjacent to candidate time points to obtain the probability value of the leakage pressure wave. In one embodiment, the identification module 330 can be used to perform the operation S230 described above, which will not be repeated here.

[0172] The timing determination module 340 is used to determine the start time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than a probability threshold, and the second pressure amplitude of the second sub-window in the target window, wherein the width of the second sub-window is greater than the width of the first sub-window. In one embodiment, the timing determination module 340 can be used to perform the operation S240 described above, which will not be repeated here.

[0173] The risk level determination module 350 is used to determine the number of risky heat exchange stations that meet the leakage conditions and the corresponding leakage risk level when there are multiple target sensors, based on the start time, leakage trigger information of multiple target sensors, and the center time obtained by clustering the leakage trigger times of each target sensor. In one embodiment, the risk level determination module 350 can be used to perform the operation S250 described above, which will not be repeated here.

[0174] According to an embodiment of the present invention, based on the denoising module 310, residual determination module 320, identification module 330, time determination module 340, and level determination module 350 in the heating network leakage risk information determination device 300, high-frequency pressure disturbance information is extracted through denoising, effectively filtering out the masking of leakage transient pressure waves by low-frequency pressure fluctuations and normal hydraulic regulation. Thus, based on residual analysis and multi-window feature identification between high-frequency disturbance information and smooth disturbance information, the problem of traditional single pressure thresholds being unable to distinguish between leakage, pump frequency regulation, and valve action disturbances is avoided. The starting time is determined by comparing the amplitude of double-width sub-windows within the target window, and combined with the starting time of multiple sensors, leakage trigger information, and cluster center time, the reliability of single-measuring point anomaly judgment is improved. The time difference relationship of multiple measuring points is fully utilized to quantify the number of risky heat exchange stations and the corresponding leakage risk level, providing a basis for the risk range and confidence level of the location results.

[0175] According to an embodiment of the present invention, the above-mentioned device further includes: a comparison module, a selection module, and a combination module. The comparison module is used to compare the deviation between the reference time of the leakage pressure wave reaching each target sensor and the propagation time of the pressure wave reaching each target sensor from the candidate leakage nodes in the pipeline network, based on the pipeline network topology information. The selection module is used to select nodes with deviations less than a deviation threshold from multiple candidate leakage nodes as risk leakage nodes, wherein the risk leakage nodes have a leakage range. The combination module is used to combine the number of target sensors, the start time of the pressure wave propagation from the risk leakage node to each target sensor, and the leakage range to obtain combined information, and output it as a diagnostic result.

[0176] According to an embodiment of the present invention, the noise reduction module 310 includes an information combination submodule and a sub-submodule. The information combination submodule is used to combine reference pressure change information characterizing the normal operating pressure of the pipeline network, pressure fluctuation information caused by leakage or pressure relief, pressure disturbance information caused by equipment regulation, and random disturbance information caused by sensor noise from multiple pressure disturbance information to obtain initial pressure information; the sub-submodule is used to subtract noise information caused by multiple reference pressure changes from the initial pressure information to obtain high-frequency pressure disturbance information.

[0177] According to an embodiment of the present invention, the residual determination module 320 includes: a first determination submodule and a second determination submodule. The first determination submodule is used to determine the difference between high-frequency pressure disturbance information and smoothed disturbance information as the pressure residual of each sensor at multiple times; the second determination submodule is used to determine the pressure residuals among the multiple pressure residuals whose absolute value is greater than a pressure threshold as candidate pressure residuals.

[0178] According to an embodiment of the present invention, the above-described apparatus further includes: an evaluation value calculation module and an evaluation value determination module. The evaluation value calculation module is used to obtain multiple abnormal evaluation values ​​indicating the degree of deviation between the multiple pressure residuals and the reference pressure within the preset window, based on multiple pressure residuals and a standard deviation estimate characterizing the fluctuation amplitude of residual information within a preset window. The evaluation value determination module is used to determine the abnormal evaluation value whose absolute value is greater than an evaluation threshold among the multiple abnormal evaluation values ​​as the target abnormal evaluation value, so as to identify multiple features extracted from multiple windows adjacent to candidate times based on the target abnormal evaluation value.

[0179] According to an embodiment of the present invention, multiple features include: an upper limit of amplitude, an energy ratio between a first sub-window and a second sub-window, waveform duration, an anomaly assessment value indicating the degree of deviation between multiple pressure residuals and a reference pressure within a preset window, and waveform gradient intensity indicating the change of pressure residual information within the preset window over time; the identification module 330 includes: a preset value determination sub-module, used to determine the probability value of the leakage pressure wave as a preset value when the upper limit of amplitude is greater than an amplitude threshold, the energy ratio is greater than an energy ratio threshold, the waveform duration is between a lower limit of duration and an upper limit of duration, the anomaly assessment value is greater than an anomaly assessment threshold, and the waveform gradient intensity is greater than an intensity threshold.

[0180] According to an embodiment of the present invention, the identification module 330 further includes a branch acquisition submodule and a mapping submodule. The branch acquisition submodule is used to make progressive judgments along the branch nodes of multiple decision trees according to the values ​​of each feature until a leaf node is reached, and to obtain the score corresponding to the leaf node; the mapping submodule is used to sum the scores of each decision tree to obtain a cumulative score and map it to a preset interval to obtain the probability value of the leakage pressure wave.

[0181] According to an embodiment of the present invention, the above-mentioned device further includes: a truncation module, a summation module, a sequence number determination module, and a time determination module. The truncation module is used to sequentially truncate a local pressure waveform sequence of a preset length, consisting of multiple sampled values, within a target window; the summation module is used to divide the pressure waveform sequence into a preceding subsequence and a following subsequence using multiple candidate segmentation points, multiply the variance of the preceding subsequence and the variance of the following subsequence corresponding to each candidate segmentation point by a penalty factor and sum them to obtain a value characterizing the data matching degree between the preceding and following subsequences of the multiple candidate segmentation points; the sequence number determination module is used to determine the segmentation point corresponding to the lower limit value among the multiple values ​​as the sequence number of the updated first wave arrival point in the local pressure waveform sequence; the time determination module is used to update the start time of the pipeline leakage pressure wave arriving at the target sensor based on the sequence number and the sampling time interval to obtain the updated time, and to determine the number of risk heat exchange stations that meet the leakage conditions based on the updated time.

[0182] According to an embodiment of the present invention, the risk level determination module 350 includes a number determination submodule and a risk level determination submodule. The number determination submodule is used to determine the number of risk heat exchange stations that meet the leakage conditions when the absolute value of the difference between the start time and the center time is less than or equal to a preset time. The risk level determination submodule is used to determine the leakage risk level as the pipeline risk level when the number of risk heat exchange stations is greater than a number threshold.

[0183] According to embodiments of the present invention, any plurality of modules among the denoising module 310, residual determination module 320, identification module 330, timing determination module 340, and level determination module 350 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the denoising module 310, residual determination module 320, identification module 330, timing determination module 340, and level determination module 350 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the noise reduction module 310, residual determination module 320, recognition module 330, time determination module 340, and level determination module 350 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0184] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0185] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for determining leakage risk information in a heating pipeline network, characterized in that, The method includes: The initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors of multiple heat exchange stations is denoised to obtain high-frequency pressure disturbance information that characterizes the pipeline leakage risk of each heat exchange station. Based on the high-frequency pressure disturbance information and the smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information, multiple candidate pressure residuals characterizing the risk of leakage pressure waves are obtained, and each candidate pressure residual has a corresponding candidate time. Multiple features extracted from multiple windows adjacent to the candidate time are identified to obtain the probability value of the leakage pressure wave. Based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than the probability threshold, and the second pressure amplitude of the second sub-window in the target window, the starting time of the pipeline leakage pressure wave arriving at the target sensor is obtained, and the width of the second sub-window is greater than the width of the first sub-window. When there are multiple target sensors, based on the start time, the leakage trigger information of multiple target sensors, and the center time obtained by clustering the leakage trigger times of each target sensor, the number of risk heat exchange stations that meet the leakage conditions and the leakage risk level corresponding to the number are determined.

2. The method according to claim 1, characterized in that, The method further includes: Based on pipeline topology information, the deviation between the reference time of leakage pressure waves reaching each target sensor and the propagation time of pressure waves reaching each target sensor at candidate leakage nodes in the pipeline network is compared. Nodes with deviations less than a deviation threshold are selected from multiple candidate leakage nodes as risk leakage nodes, and risk leakage nodes have leakage ranges. The combined information obtained by combining the number of target sensors, the start time of the pressure wave propagation from the risk leakage node to each target sensor, and the leakage range is used as the diagnostic result and output.

3. The method according to claim 1, characterized in that, The initial pressure information obtained by combining multiple pressure disturbance information acquired by sensors from multiple heat exchange stations is denoised to obtain high-frequency pressure disturbance information characterizing the pipeline leakage risk of each heat exchange station, including: The initial pressure information is obtained by combining the reference pressure change information representing the normal operating pressure of the pipeline network, the pressure fluctuation information caused by leakage or pressure relief, the pressure disturbance information caused by equipment adjustment, and the random disturbance information caused by sensor noise from the multiple pressure disturbance information. The high-frequency pressure disturbance information is obtained by subtracting the noise information caused by various reference pressure changes from the initial pressure information.

4. The method according to claim 1, characterized in that, Based on the high-frequency pressure disturbance information and the smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information, multiple candidate pressure residuals characterizing the risk of leakage pressure waves are obtained, including: The difference between the high-frequency pressure disturbance information and the smoothed disturbance information is determined as the pressure residual of each sensor at multiple times. Among multiple pressure residuals, the pressure residuals whose absolute value is greater than the pressure threshold are identified as candidate pressure residuals.

5. The method according to claim 4, characterized in that, The method further includes: Based on multiple pressure residuals and the standard deviation estimate representing the fluctuation range of residual information within a preset window, multiple anomaly assessment values ​​indicating the degree of deviation between multiple pressure residuals and the reference pressure within the preset window are obtained. The abnormal evaluation value whose absolute value is greater than the evaluation threshold among the multiple abnormal evaluation values ​​is determined as the target abnormal evaluation value, so as to identify multiple features extracted from multiple windows adjacent to the candidate time based on the target abnormal evaluation value.

6. The method according to claim 5, characterized in that, The multiple features include: upper limit of amplitude, energy ratio between the first sub-window and the second sub-window, waveform duration, anomaly assessment value indicating the degree of deviation between multiple pressure residuals and the reference pressure in the preset window, and waveform gradient intensity of the pressure residual information in the preset window as time changes. Multiple features extracted from multiple windows adjacent to the candidate time are identified to obtain the probability value of the leakage pressure wave, including: When the upper limit of amplitude is greater than the amplitude threshold, the energy ratio is greater than the energy ratio threshold, the duration of the waveform is between the lower limit and the upper limit of duration, the anomaly assessment value is greater than the anomaly assessment threshold, and the waveform gradient intensity is greater than the intensity threshold, the probability value of the leakage pressure wave is determined to be a preset value.

7. The method according to claim 6, characterized in that, The method further includes identifying multiple features extracted from multiple windows adjacent to the candidate time points to obtain the probability value of the leakage pressure wave, and also includes: The decision is made step by step along the branch nodes of multiple decision trees based on the value of each feature until the leaf node is reached, and the score corresponding to the leaf node is obtained. The scores of each decision tree are summed to obtain the cumulative score, which is then mapped to a preset interval to obtain the probability value of the leakage pressure wave.

8. The method according to claim 1, characterized in that, The method further includes: Within the target window, a local pressure waveform sequence consisting of multiple sampled values ​​of a preset length is sequentially extracted; The pressure waveform sequence is divided into a preceding subsequence and a following subsequence using multiple candidate segmentation points. The variances of the preceding and following subsequences corresponding to each candidate segmentation point are multiplied by a penalty factor and summed to obtain a numerical value characterizing the data matching degree between the preceding and following subsequences of the multiple candidate segmentation points. The dividing point corresponding to the lower limit value among multiple values ​​is determined as the sequence number of the updated first wave arrival point in the local pressure waveform sequence; Based on the sequence number and sampling time interval, the start time of the pipeline leakage pressure wave arriving at the target sensor is updated to obtain the update time, so as to determine the number of risk heat exchange stations that meet the leakage conditions based on the update time.

9. The method according to claim 1, characterized in that, Based on the aforementioned start time, leakage trigger information from multiple target sensors, and the center time obtained by clustering the leakage trigger times of each target sensor, the number of risky heat exchange stations meeting the leakage conditions and the corresponding leakage risk level are determined, including: If the absolute value of the difference between the start time and the center time is less than or equal to a preset duration, the number of risk heat exchange stations that meet the leakage conditions is determined. If the number of risky heat exchange stations exceeds a certain threshold, the leakage risk level is determined as the pipeline risk level.

10. A device for determining leakage risk information in a heating pipeline network, characterized in that, The device includes: The noise reduction module is used to denoise the initial pressure information obtained by combining multiple pressure disturbance information acquired by the sensors of multiple heat exchange stations, so as to obtain high-frequency pressure disturbance information that characterizes the pipeline leakage risk of each heat exchange station. The residual determination module is used to obtain multiple candidate pressure residuals that characterize the risk of leakage pressure waves based on the high-frequency pressure disturbance information and the smoothed disturbance information obtained by smoothing the high-frequency pressure disturbance information. Each candidate pressure residual has a corresponding candidate time. The identification module is used to identify multiple features extracted from multiple windows adjacent to the candidate time to obtain the probability value of the leakage pressure wave. The timing determination module is used to determine the starting time of the pipeline leakage pressure wave reaching the target sensor based on the first pressure amplitude of the first sub-window in the target window with a probability value greater than a probability threshold and the second pressure amplitude of the second sub-window in the target window, wherein the width of the second sub-window is greater than the width of the first sub-window. The risk level determination module is used to determine the number of risk heat exchange stations that meet the leakage conditions and the leakage risk level corresponding to the number of stations when there are multiple target sensors, based on the start time, the leakage trigger information of multiple target sensors, and the center time obtained by clustering the leakage trigger times of each target sensor.

Images