A multi-source joint judgment method for pipeline leakage

Abstract

This invention discloses a multi-source joint judgment method for pipeline leakage, relating to the field of fluid tightness testing technology for structures. The method includes: acquiring the original acoustic channel sequence and flue gas temperature measurement point sequence, and obtaining a sound velocity classification map; extracting noise baseline segments from the acoustic sequence to determine the rise threshold, identifying the first arrival time sequence and extracting time window segments to form a first arrival waveform segment set, extracting features to generate a path fingerprint table; determining a location grid table by combining a heated surface layout diagram, calculating the propagation time to obtain an expected arrival ranking table, and comparing the first arrival time sequence to select a candidate location table; then generating an expected echo band table by combining the structural location table, comparing it with the path fingerprint table, and calculating candidate scores by combining the ranking difference table, providing the location result and uncertainty. This invention effectively solves the problem of spurious solutions formed by the superposition of multiple paths of acoustic signals in densely packed tube bundles on the boiler heated surface, providing a leak location reference for the operation and maintenance of thermal power plant boilers.

Description

Technical Field

[0001] This invention relates to the field of fluid tightness testing technology for structures, and more specifically, to a multi-source joint judgment method for pipeline leakage. Background Technology

[0002] In the operation and maintenance of boilers in thermal power plants, real-time monitoring and location of pipeline leaks in the heating surface area are crucial for ensuring stable unit operation. This area has a dense distribution of pipe bundles, surrounded by furnace walls, steel structures, and other components. The acoustic signals generated by leaks are easily affected by the surrounding environment and structure during propagation, posing complex challenges to signal reception and location analysis. Currently, leak location in this area often employs a solution combining acoustic sensor arrays with time-of-arrival (TOA) positioning technology, which is also the mainstream application of acoustic positioning in industrial settings.

[0003] The core principle of time-of-arrival (TOA) positioning technology is to simultaneously receive acoustic signals generated by a leak source using multiple acoustic sensors deployed in the monitoring area. The time difference between the arrival times of the signals at each sensor is calculated, and combined with the spatial location information of the sensors and preset sound velocity field distribution parameters, a set of positioning equations is constructed. Solving these equations determines the spatial coordinates of the leak source. The positioning accuracy of this technology depends on the stability of the signal propagation path, the accuracy of the time difference calculation, and the uniqueness of the solution to the equations. In scenarios with a single signal propagation path and minimal interference from surrounding structures, it can achieve relatively reliable positioning output.

[0004] However, in practical applications where boiler heating surfaces have dense tube bundles, the aforementioned technologies cannot fully meet the reliability requirements for positioning. The core problem stems from the multi-path propagation and superposition effect of acoustic signals. The acoustic signal generated by a leak does not propagate to the sensor through a single direct path, but simultaneously includes a direct wave, a diffracted wave formed by diffraction through the gaps in the tube bundles, and a reflected wave formed by reflection from the furnace wall and steel structure. These various waveforms superimpose each other during propagation and are ultimately received synchronously by the sensor. Due to the complexity of the spatial structure in dense tube bundle areas, leak sources in different spatial locations may produce nearly identical combinations of arrival time differences under the current sensor deployment and sound velocity field environment, thus forming a pseudo-solution with equal time difference. At the same time, changes in the unit's operating load will cause fluctuations in the flue gas temperature field, and temperature changes will directly change the sound velocity distribution, thereby affecting the signal propagation path and the calculation results of the time difference. This causes the spatial location of the pseudo-solution with equal time difference to drift with changes in operating conditions, resulting in different positioning results for the same leak source at different operating stages, and the positioning results often deviate from adjacent tube bundles or heating surface areas. Existing technologies do not fully consider the effects of multipath superposition and pseudo-solution drift. Without increasing the number of sensors, they lack effective means to separate direct waves from multipath waves, cannot accurately identify and eliminate isochronous pseudo-solutions, and are difficult to quantify the multimodality and uncertainty of positioning results. They can only rely on the solution of a single set of equations as the basis for positioning and cannot provide a reliability assessment reference for the joint judgment process.

[0005] The aforementioned problems lead to discrepancies between leak location results and actual leak locations, compromising accuracy. Locating the same leak source in different areas at different times interferes with monitoring personnel's judgment of the leak location, increasing the difficulty and time required for leak investigation. Due to the lack of quantitative output regarding the uncertainty of the location results, the joint judgment process may directly use spurious solutions as valid evidence to trigger alarm procedures, leading to unnecessary maintenance operations and disrupting the unit's normal operation. Furthermore, location deviations may delay the response to actual leaks, preventing timely and targeted maintenance measures, thus affecting the boiler's operational safety and stability, and hindering the unit's long-term safe and efficient operation.

[0006] In view of this, the present invention proposes a multi-source joint judgment method for pipeline leakage to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a multi-source joint judgment method for pipeline leakage, comprising:

[0008] The original acoustic channel sequence and the flue gas temperature measurement point sequence were collected, and the flue gas temperature measurement point sequence was graded to obtain a sound velocity classification map;

[0009] Segment truncation is performed within the original sequence of the acoustic channel to obtain the noise baseline segment. The rise threshold is determined based on the noise baseline segment. The moment when the rise threshold is first exceeded is taken as the first arrival time sequence. Time window segments are truncated around the first arrival time sequence to obtain the first arrival waveform segment set.

[0010] The first arrival waveform segment set is used as a reference to extract features and generate a path fingerprint table. The location grid table is determined based on the pre-acquired heated surface layout map. The propagation time of the location grid table is calculated according to the sound velocity classification map and the pre-acquired sensor location table. The expected arrival ranking table is obtained by sorting the expected arrival ranking table and comparing it with the first arrival time sequence to obtain a ranking difference table. The candidate location table is obtained by filtering from the ranking difference table.

[0011] Based on the candidate location table and the pre-acquired structural location table, an expected echo band table is generated. The expected echo band table is compared with the path fingerprint table, and the candidate score is calculated by combining the ranking difference table. A location distribution table is generated based on the candidate score, and the location result and uncertainty are given by the location distribution table.

[0012] Furthermore, the sound velocity classification diagram obtained by classifying the flue gas temperature measurement point sequence includes:

[0013] The temperature grading interval is set to divide the flue gas temperature measurement point sequence into multiple temperature ranges and assign sound velocity levels. The sound velocity meter is set to determine the sound velocity value of each sound velocity level based on the center temperature of the temperature range.

[0014] For each grid point in the location grid table, calculate the spatial distance between it and each flue gas temperature measuring point. Set a first proximity radius to filter the set of neighboring measuring points. Determine the weight according to the spatial distance and round down the weighted sound velocity level to obtain the sound velocity level of the grid point. Perform a graded update according to the graded update interval to obtain the sound velocity graded map.

[0015] Furthermore, methods for obtaining the noise baseline segment include:

[0016] Using a unified time scale as an index, the search duration is limited in the first part of the original acoustic channel sequence. For each channel within the search duration, the absolute value sequence of waveform sample values ​​is calculated. The mean amplitude sequence is obtained by averaging the absolute value sequence of waveform sample values ​​according to the unified time scale.

[0017] The low-energy threshold is determined by the preset low-energy quantile in the mean amplitude sequence. The continuous time periods in the mean amplitude sequence that do not exceed the low-energy threshold are selected to obtain the baseline candidate time period set. The segment before the event is determined according to the length of the first time window, and the noise baseline segment is extracted.

[0018] Furthermore, methods for determining the rising threshold include:

[0019] The noise baseline segment is de-biased and the maximum amplitude is extracted.

[0020] Set an exceedance factor, which is used to characterize the proportion of the rise threshold that exceeds the maximum amplitude;

[0021] The excess amount is determined based on the maximum amplitude and the excess coefficient. The excess amount is the amplitude obtained by amplifying the maximum amplitude by the excess coefficient.

[0022] The rise threshold is the sum of the maximum amplitude and the excess amount.

[0023] Furthermore, methods for obtaining the first waveform segment set include:

[0024] For each acoustic channel, a segment is extracted from the original sequence of the acoustic channel, centered on the first arrival time of that acoustic channel in the first arrival time sequence.

[0025] Set the second time window length, and use a symmetrical method for segment truncation before and after the first arrival time. The length of the first segment and the length of the second segment are both half of the second time window length. The continuous sampling segment obtained by truncation is recorded as the time window segment.

[0026] Segment extraction is performed repeatedly for each acoustic channel, and time window segments are recorded by associating the channel number with the first arrival time to obtain the first arrival waveform segment set.

[0027] Furthermore, if the baseline candidate time period set contains a continuous time period with a length not less than the length of the first time window, then the earliest continuous time period that meets the length condition is taken as the pre-event segment; if the baseline candidate time period set does not contain a continuous time period with a length not less than the length of the first time window, then the longest continuous time period in the baseline candidate time period set is taken as the pre-event segment, and the length of the first time window is taken as the length of the longest continuous time period.

[0028] Furthermore, the maximum amplitude of the baseline is extracted from the noise baseline segment, and a preset amplitude upper limit is determined based on the maximum amplitude of the baseline. Starting from the unified time mark at the end of the segment before the event, the search proceeds forward along the original sequence of acoustic channels to find the unified time mark position where the absolute value of the waveform sample value of any channel first exceeds the preset amplitude upper limit. This unified time mark is recorded as the event reference time mark. Each channel segment with a length equal to the length of the first time window before the event reference time mark is re-extracted as the segment before the event, and the noise baseline segment is updated.

[0029] Furthermore, methods for obtaining the path fingerprint table include:

[0030] For each time window segment within the first waveform segment set, the starting segment of the time window segment is extracted, and the length of the starting segment is limited by the length of the earlier segment.

[0031] The average energy is calculated from the absolute value of the waveform sampled values ​​of the noise baseline segment. The average energy is then normalized according to the uniform time scale length of the noise baseline segment to obtain the average energy of the noise baseline.

[0032] The average energy is calculated from the absolute value of the waveform sampled values ​​of the initial segment. The average energy is then normalized according to the uniform time scale length of the initial segment to obtain the average energy of the initial segment. The early energy ratio is the ratio of the average energy of the initial segment to the average energy of the noise baseline.

[0033] The peak time scale is recorded with the first arrival time in the first arrival time sequence as the zero point, and the peak interval sequence is obtained according to the time scale difference between adjacent peaks.

[0034] For each time window segment within the initial waveform segment set, extract the frequency band energy column;

[0035] The path fingerprint table includes channel number, early energy ratio, peak interval sequence, and frequency band energy column.

[0036] Furthermore, methods for extracting frequency band energy trains include:

[0037] Set the frequency band boundary, with the lower bound being 0 Hz and the upper bound being half the sampling rate of the original sequence of the acoustic channel;

[0038] The frequency band from 0 Hz to the upper limit is divided into several frequency bands in the first frequency band according to the frequency band boundary;

[0039] Calculate the discrete frequency components for the time window segment based on the sampling point sequence corresponding to the second time window length;

[0040] For each frequency band, select discrete frequency components that fall within the boundary of that frequency band, and accumulate the squared amplitudes of the discrete frequency components to obtain the band energy of that frequency band.

[0041] Record the frequency band energy in frequency band order to obtain the frequency band energy sequence.

[0042] Compared with the prior art, the technical effects and advantages of the multi-source joint judgment method for pipeline leakage of the present invention are as follows:

[0043] This invention addresses leakage detection in densely packed tube bundle areas of boiler heating surfaces in thermal power plants. First, it collects the original acoustic channel sequence and flue gas temperature measurement point sequence. A sound velocity classification map synchronized with the operating conditions is constructed through temperature classification. Then, a noise baseline segment is extracted from the original acoustic channel sequence, and the rise threshold is determined accordingly. After identifying the first arrival time sequence, a first arrival waveform segment set is formed. Subsequently, features of the first arrival waveform segment set are extracted to generate a path fingerprint table. A location grid table is constructed by combining the heating surface layout map. Based on the sound velocity classification map and sensor location table, the propagation time is calculated, and an expected arrival ranking table is generated. A ranking difference table is obtained by comparing the first arrival time sequences, and a candidate location table is selected. Then, an expected echo band table is generated based on the candidate location table and the structural location table. This table is compared with the path fingerprint table, and a candidate score is calculated based on the ranking difference table. A location distribution table is generated based on the candidate scores, and finally, the location result and uncertainty are output synchronously. The entire solution forms a complete process around the acquisition, processing, feature extraction, and localization analysis of acoustic signals. It incorporates the operating conditions of the flue gas temperature field into the sound velocity calculation, dynamically determines the detection threshold through the noise baseline, extracts multi-dimensional features of acoustic signals to form path fingerprints, constructs the expected reverberation band by combining the structural features of the heated surface, and achieves leak localization through multi-dimensional comparison, screening, and scoring. At the same time, it quantifies the uncertainty of the localization results, forming a multi-source joint judgment system for leaks that is adapted to the complex environment of the boiler heated surface.

[0044] This invention solves the problems of acoustic signals in densely packed tube bundles on boiler heating surfaces propagating through multiple paths and superimposing to form spurious solutions with equal time differences, as well as the spatial position drift of spurious solutions caused by changes in operating conditions. It also solves the problems of existing technologies being unable to effectively separate direct waves from multipath waves, and having difficulty quantifying the multi-peak nature and uncertainty of positioning results. Furthermore, it solves the problems of positioning results easily deviating to adjacent tube bundles or heating surfaces, and the joint judgment process easily using spurious solution positioning results as valid evidence to trigger unnecessary actions.

[0045] This invention improves the matching degree between the leak location result and the actual leak location. The location result will not have a significant positional drift due to the fluctuation of flue gas temperature field caused by changes in unit load. It can distinguish the signal characteristics brought by direct waves and multipath waves, screen out the real leak candidate locations, and avoid the location result falling into invalid locations outside the heated surface area.

[0046] This invention can quantify the uncertainty of the positioning results, providing a reliability assessment reference for the joint judgment method. It avoids unnecessary maintenance work triggered by positioning deviations in the joint judgment method, and also prevents delays in handling actual leaks due to positioning deviations. This makes the pipeline leak location in the boiler heating surface area more consistent with the actual on-site working conditions, improves the accuracy and reliability of leak detection and location, ensures the stability of boiler operation, provides a leak location reference for the operation and maintenance of thermal power plant boilers, and improves the effectiveness of inspection and monitoring work in thermal power plants. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of a multi-source joint detection system for pipeline leakage according to an embodiment of the present invention;

[0048] Figure 2 This is a flowchart of a multi-source joint judgment method for pipeline leakage according to an embodiment of the present invention;

[0049] Figure 3 A flowchart illustrating the method for determining the rising threshold in an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0051] Example 1:

[0052] Please see Figure 1 As shown in the figure, this embodiment discloses a multi-source joint judgment system for pipeline leakage, including a data acquisition module, a sequence interception module, a candidate sorting module, and a result output module. Each module is connected by wired or wireless means to realize data transmission.

[0053] The data acquisition module is used to acquire the original sequence of the acoustic channel and the sequence of flue gas temperature measurement points, and to classify the flue gas temperature measurement point sequence to obtain a sound velocity classification map.

[0054] The following describes a specific implementation method for obtaining a sound velocity classification map by classifying the flue gas temperature measurement point sequence, combining the original sequence of the acquired acoustic channel and the flue gas temperature measurement point sequence:

[0055] An acoustic sensor array is defined for the heated surface area, arranged along the outer side of the heated surface tube bundle. The acoustic sensor array uses gas-based acoustic sensors, which are fixed to a fixed support on the outer side of the heated surface area via mounting bases. The sampling surface of the gas-based acoustic sensors faces the flue gas space of the heated surface area. A coupling medium is placed between the mounting base and the gas-based acoustic sensors to reduce the influence of gaps at the mounting interface on the transmission of high-frequency components. The frequency response range of the gas-based acoustic sensors covers the upper limit of the dominant frequency band of the acoustic signal from pipe leakage in the heated surface area, with a preferred range of 1 kHz to 200 kHz. The selection rule for the frequency response range is that the higher the upper limit of the dominant frequency band of the acoustic signal from pipe leakage in the heated surface area, the larger the upper limit of the frequency response range; the lower the dominant frequency band of the background mechanical noise in the heated surface area, the larger the lower limit of the frequency response range, in order to reduce the proportion of low-frequency background components entering the sampling link. During installation, a channel number is recorded for each gas-based acoustic sensor, with each channel number corresponding to its installation location.

[0056] The sampling rate is set and used by the acoustic sensor array to sample the waveforms of each channel. The preferred range for the sampling rate is 50 kHz to 500 kHz. The selection rule for the sampling rate is that the higher the upper limit of the dominant frequency band of the acoustic signal from pipe leakage in the heated surface area, the larger the sampling rate value; the lower the dominant frequency band of the background mechanical noise in the heated surface area, the larger the sampling rate value, to widen the separation space between the leakage acoustic signal and the background mechanical noise in the frequency band. An anti-aliasing cutoff frequency is set for each channel, with a preferred range of 0.35 to 0.45 times the sampling rate. The selection rule for the anti-aliasing cutoff frequency is that the higher the amplitude consistency requirement between channels of the acoustic sensor array, the smaller the anti-aliasing cutoff frequency value; the higher the upper limit of the dominant frequency band of the acoustic signal from pipe leakage in the heated surface area, the larger the anti-aliasing cutoff frequency value. Anti-aliasing filtering is performed on the waveform of each channel before sampling, and the anti-aliasing filter corresponds to the anti-aliasing cutoff frequency.

[0057] The waveforms of each channel of the acoustic sensor array are synchronously sampled according to the sampling rate. Synchronous sampling uses a hardware timing signal output from the same timing source, which uniformly constrains the sampling triggering of each channel. The quantization bit depth is set, with an optimal range of 12 to 24 bits. The selection rule for the quantization bit depth is that the larger the fluctuation range of the background mechanical noise amplitude in the heated surface area, and the longer the weak initial segment to be retained in the original acoustic channel sequence, the larger the quantization bit depth. A unified timescale is marked for each sampling point. The unified timescale uses a timescale sequence generated by the same timing source, and the unified timescale corresponds one-to-one with the sampling process of each channel of the acoustic sensor array. An upper bound for the timescale error is set, with an optimal range of 1 microsecond to 50 microseconds. The selection rule for the upper bound of the timescale error is that the higher the sampling rate, the smaller the upper bound of the timescale error; the more channels in the acoustic sensor array, the smaller the upper bound of the timescale error. The original acoustic channel sequence is formed using the channel number, unified timescale, and waveform sampling value.

[0058] The waveform sampling values ​​of the original acoustic channel sequence are used to characterize the time position of pressure pulsations in the flue gas reaching each channel. The first arrival time sequence corresponds to the sound wave arrival process in the flue gas, and the propagation time order corresponds to the propagation time order in the flue gas. The flue gas temperature measurement point sequence is used to characterize the spatial variation of flue gas temperature within the heated surface area. The sound velocity classification map obtained by classifying the flue gas temperature measurement point sequence is used to characterize the sound velocity level distribution in the flue gas. The sound velocity value range corresponding to the sound velocity classification map is consistent with 200 m / s to 800 m / s. This correspondence defines the physical meaning of the sound velocity classification map. In the propagation time calculation, the sound velocity classification map maintains the same propagation medium aperture as the first arrival time sequence of the original acoustic channel sequence, reducing the feasibility risk caused by the inconsistency of the physical basis of the expected arrival order table, and improving the usability of the positioning input data from the source.

[0059] The acquisition range of the flue gas temperature measurement point sequence is determined. The flue gas temperature measurement point sequence comes from the unit measurement points. The unit measurement points are arranged in the upstream flue, adjacent flue, and downstream flue of the heating surface area. Temperature sampling intervals are set, with an optimal range of 0.5 seconds to 10 seconds. The selection rule for the temperature sampling interval is that the higher the frequency of unit operating load changes, the smaller the temperature sampling interval; the greater the spatial gradient of flue gas temperature in the heating surface area, the smaller the temperature sampling interval, to improve the temporal resolution of temperature field changes being recorded. Temperature values ​​are continuously collected from the unit measurement points according to the temperature sampling interval. A unified time scale is marked for each temperature sampling point. The unified time scale uses the same timing source as the original sequence of the acoustic channel. The flue gas temperature measurement point sequence is formed by the measurement point number, unified time scale, and temperature sampling value, resulting in a flue gas temperature measurement point sequence that shares a unified time scale with the original sequence of the acoustic channel. This reduces the probability of temporal mismatch between sound velocity classification and acoustic events, and alleviates the inducing conditions of sound velocity distribution drift caused by operating condition changes on the drift of the positioning results.

[0060] Time coverage processing is performed on the flue gas temperature measurement point sequence. Using the unified time scale of the original acoustic channel sequence as a benchmark, the time coverage range of the original acoustic channel sequence is determined. The flue gas temperature measurement point sequence is then truncated within the same time coverage range. For the truncated flue gas temperature measurement point sequence, if a measurement point lacks temperature sampling values ​​at certain unified time scale positions, the most recent valid temperature sampling value for that measurement point is used to fill the corresponding unified time scale position, forming a time-continuous flue gas temperature measurement point sequence. For each measurement point's flue gas temperature measurement point sequence, the direction and amplitude information of temperature changes are preserved to avoid replacing the entire change with a single fixed value. Forming a continuous sequence of flue gas temperature measurement point sequences under a unified time scale reduces the breakage of the sound velocity classification map on the time axis and alleviates the problem of inconsistent inputs to the localization equations caused by gaps in temperature sampling leading to empty segments in the sound velocity classification map.

[0061] A sound velocity classification map is obtained by classifying the flue gas temperature measurement point sequence. Temperature classification intervals are set. The preferred range for temperature classification intervals is 2 to 30 degrees Celsius. The selection rules for temperature classification intervals are as follows: the greater the fluctuation range of flue gas temperature in the heated surface area, the larger the temperature classification interval value, to control frequent classification switching; the longer the acoustic propagation path length in the heated surface area, the smaller the temperature classification interval value, to reduce the accumulation of sound velocity classification errors over propagation time. The flue gas temperature measurement point sequence is divided into multiple temperature intervals according to the temperature classification intervals. Each temperature interval is assigned a sound velocity level, represented by discrete levels. The sound velocity levels correspond one-to-one with the temperature intervals, forming a measurement point sound velocity level sequence using measurement point numbers, unified time scales, and sound velocity levels.

[0062] To ensure a definite correspondence between sound velocity levels and sound velocity values ​​in propagation time calculations, a sound velocity table is established. This table records the sound velocity value corresponding to each sound velocity level, with values ​​ranging from 200 m / s to 800 m / s. The sound velocity table is constructed based on the physical relationship between air sound velocity and temperature. The table selects the corresponding sound velocity value according to the center temperature of the temperature range, and the sound velocity value changes monotonically with the center temperature. The upper limit of the sound velocity table is determined by the rule that the higher the temperature range covered by the flue gas temperature measurement point sequence, the larger the upper limit value; conversely, the lower limit is determined by the rule that the lower the temperature range covered by the flue gas temperature measurement point sequence, the smaller the lower limit value. The sound velocity levels in the sound velocity table are consistent with the sound velocity levels in the measurement point sound velocity level sequence to avoid inconsistencies between the sound velocity levels during the classification stage and the propagation time calculation stage.

[0063] The sound velocity level sequence of measuring points is mapped to the monitoring area according to the spatial location of the heated surface region, resulting in a sound velocity classification map. The monitoring area is the coverage area of ​​the grid points corresponding to the location grid table, and the three-dimensional coordinates of the grid points and the sensor location table use the same coordinate reference. For each grid point, the spatial distance between the grid point and each measuring point is calculated, and the spatial distance is determined by the three-dimensional coordinates of the grid point and the measuring point. A first proximity radius is set, with an optimal range of 2 to 30 meters. The selection rule for the first proximity radius is that the larger the spacing between the flue gas temperature measuring points around the heated surface region, the larger the value of the first proximity radius; the larger the spatial gradient of the flue gas temperature in the heated surface region, the smaller the value of the first proximity radius. Measuring points whose spatial distance does not exceed the first proximity radius are selected to form a set of proximity measuring points. If the set of proximity measuring points is empty, the measuring point with the smallest spatial distance is selected as the set of proximity measuring points. For each measuring point in the set of proximity measuring points, a weight is determined according to the spatial distance, with the smaller the spatial distance, the larger the weight, and the weight is normalized according to the set of proximity measuring points. The sound velocity levels of neighboring measurement points are weighted and rounded to obtain the grid point sound velocity levels. The grid point sound velocity levels are represented using the same discrete level representation as the standard sound velocity levels. A sound velocity classification map is formed using the grid point number, a unified time scale, and the grid point sound velocity levels. The sound velocity classification map shows the sound velocity level distribution within the monitoring area at each unified time scale location.

[0064] The flue gas temperature measurement point sequence is transformed into a sound velocity classification map by mapping temperature range, sound velocity level, sound velocity meter and spatial distance. The sound velocity classification map has a reproducible assignment caliber under a unified time scale, which reduces the cost of expressing the direct disturbance of the sound velocity distribution by temperature field fluctuations, alleviates the basic conditions for the position drift of the isochronous pseudo solution in different operating stages, and improves the interpretability of the positioning results under changing operating conditions.

[0065] The update rhythm of the sound velocity classification map is constrained to maintain consistent time progression with the original acoustic channel sequence. A classification update interval is set, with an optimal range of 1 to 30 seconds. The selection rule for the classification update interval is that the more frequent the changes in unit operating load, the smaller the classification update interval; the smaller the temperature sampling interval of the flue gas temperature measurement point sequence, the smaller the classification update interval, to reduce the time difference between the sound velocity classification map and temperature changes. Classification updates are performed on the flue gas temperature measurement point sequence according to the classification update interval to obtain a continuous time-series sound velocity classification map. The sound velocity classification map shares a unified time scale with the original acoustic channel sequence. The sound velocity classification map covers the time coverage range corresponding to the original acoustic channel sequence. Under conditions of changing operating conditions, a sound velocity classification map synchronized with the original acoustic channel sequence is continuously output to reduce propagation time deviation caused by sound velocity distribution lag, alleviate the phenomenon of location jumps in adjacent operating periods, and thus improve the reliability of pipeline leak location in the heated surface area.

[0066] The sequence truncation module is used to truncate segments within the original sequence of the acoustic channel to obtain noise baseline segments. Based on the noise baseline segments, the rise threshold is determined, and the moment when the rise threshold is first exceeded is taken as the first arrival time sequence. Time window segments are truncated around the first arrival time sequence to obtain the first arrival waveform segment set.

[0067] Please see Figure 3 As shown, the following describes a specific implementation method for obtaining a noise baseline segment by segmenting the original acoustic channel sequence, determining the rise threshold based on the noise baseline segment, and using the moment when the rise threshold is first exceeded as the first arrival time sequence. The method then describes how to extract a time window segment around the first arrival time sequence to obtain the first arrival waveform segment set.

[0068] Segment extraction is performed within the original acoustic channel sequence to obtain the noise baseline segment. The original acoustic channel sequence is arranged according to a uniform time scale. Using the uniform time scale as an index, a search duration is first selected in the first part of the original acoustic channel sequence. The preferred range for the search duration is 30 to 300 seconds, and the selection rule is that the more frequent the changes in unit operating load, the smaller the search duration value. The absolute value sequence of waveform sample values ​​is calculated for each acoustic channel within the search duration. Then, the mean of the absolute value sequences of waveform sample values ​​of each acoustic channel at the same uniform time scale position is calculated to obtain the mean amplitude sequence. A low-energy quantile is set, with a preferred range of 10th percentile to 40th percentile. The selection rule for the low-energy quantile is that the larger the amplitude fluctuation range of the background mechanical noise in the heated surface area, the smaller the low-energy quantile value. The level corresponding to the low-energy quantile of the mean amplitude sequence is used as the low-energy threshold to filter continuous time periods where the mean amplitude sequence does not exceed the low-energy threshold, thus obtaining the baseline candidate time period set. The first time window length is set, with an optimal range of 2 to 20 seconds. The selection rule for the first time window length is that the longer the fluctuation period of the background mechanical noise in the heated surface area, the larger the value of the first time window length; the more frequent the changes in unit operating load, the smaller the value of the first time window length. If the baseline candidate time period set contains a continuous time period with a length not less than the first time window length, the earliest continuous time period that meets the length condition is taken as the pre-event segment; if the baseline candidate time period set does not contain a continuous time period with a length not less than the first time window length, the longest continuous time period in the baseline candidate time period set is taken as the pre-event segment, and the first time window length is set to the length of this longest continuous time period. The pre-event segment is extracted from each acoustic channel according to the first time window length to obtain the noise baseline segment for each acoustic channel.

[0069] After the pre-event segment is determined, a preset amplitude upper limit is set based on the noise baseline segment. The maximum absolute value of the waveform sample value is extracted from the noise baseline segment of each channel to obtain the maximum baseline amplitude. A first amplitude margin is set, preferably ranging from 10% to 100% of the maximum baseline amplitude. The selection rule for the first amplitude margin is that the higher the frequency of short-time spikes in the background mechanical noise of the heated surface area, the larger the value of the first amplitude margin. The preset amplitude upper limit is set as the first amplitude margin above the maximum baseline amplitude. Starting from the unified time marker at the end of the pre-event segment, the search proceeds forward along the original acoustic channel sequence to find the unified time marker position where the absolute value of the waveform sample value of any acoustic channel first exceeds the preset amplitude upper limit; this is recorded as the event reference time marker. Each acoustic channel segment with a length equal to the first time window before the event reference time marker is re-trimmed as the pre-event segment, resulting in an updated noise baseline segment. The updated noise baseline segment does not contain sample values ​​exceeding the preset amplitude upper limit.

[0070] The range of values ​​for the segment before the event is limited by the mean amplitude sequence, and the upper limit of the preset amplitude is determined by the noise baseline segment, and the event reference time scale is determined accordingly. The noise baseline segment maintains a corresponding relationship with the background mechanical noise amplitude level under the change of operating conditions, reducing the probability that the noise baseline segment will be mixed with early leakage components and raise the threshold, and reducing the probability that the noise baseline segment length will be insufficient due to improper setting of the upper limit of the preset amplitude, thereby alleviating the problem that the first arrival information is masked by the background waveform under the condition of multipath superposition.

[0071] The rise threshold is determined based on the noise baseline segment. For each acoustic channel's noise baseline segment, the waveform samples of the noise baseline segment are first de-polarized. The de-polarization process uses the center level of the noise baseline segment as a reference. The center level is taken as the median level of the waveform samples within the noise baseline segment. The de-polarization process is used to reduce the impact of sensor zero-point drift on threshold determination. Subsequently, the maximum amplitude is extracted from the de-polarized noise baseline segment, and the maximum amplitude is taken as the maximum absolute value of the waveform samples within the noise baseline segment.

[0072] An exceedance factor is set to characterize the proportion by which the rise threshold exceeds the maximum amplitude. The preferred range for the exceedance factor is 0.1 to 1. The selection rule for the exceedance factor is that the higher the frequency of short-time spikes in the background mechanical noise of the heated surface area, the larger the exceedance factor value; the coarser the sampling quantization level of the original sequence of the acoustic channel, the larger the exceedance factor value. The exceedance amount is determined based on the maximum amplitude and the exceedance factor, and the exceedance amount is the amplitude exceedance obtained by amplifying the maximum amplitude by the exceedance factor. The rise threshold is set to the superposition level of the maximum amplitude and the exceedance amount to obtain the rise threshold of the acoustic channel. The rise threshold is determined using a noise baseline segment, so that the rise threshold varies with the noise floor of the heated surface area, reducing false first-arrival triggering caused by the superposition of reflected waves, diffracted waves and background mechanical noise, and reducing the input conditions for the occurrence of isochronous pseudo-solutions.

[0073] The first arrival time sequence is determined within the original acoustic channel sequence. Using the rise threshold of each acoustic channel as the criterion, the envelope rise is calculated point-by-point along a unified time scale within the original acoustic channel sequence. The envelope rise is expressed in the amplitude domain. The waveform sample values ​​of the original acoustic channel sequence are de-biased, with the center level of the noise baseline segment as a reference. The absolute value of the de-biased waveform sample value is taken at each unified time scale position as the envelope rise at that position. The unified time scale position that first exceeds the rise threshold and continuously satisfies the first continuous duration is determined as the first arrival time of that acoustic channel. The preferred range for the first continuous duration is 0.2 ms to 5 ms. The selection rule for the first continuous duration is that the higher the sampling rate of the original acoustic channel sequence, the smaller the value of the first continuous duration; the more frequent the short-time spikes of background mechanical noise in the heated surface area, the larger the value of the first continuous duration. This process is repeated for each acoustic channel of the acoustic sensor array, and the first arrival time is recorded according to the channel number to obtain the first arrival time sequence. By comparing the envelope rise and rise threshold within the same amplitude domain, the risk of non-reproducibility caused by inconsistent definitions is reduced, the probability of first arrival triggered by reflected or diffracted waves in multipath propagation is reduced, and the positioning uncertainty caused by the multi-peak nature of positioning output is alleviated.

[0074] A time window segment is extracted around the first arrival time sequence to obtain the first arrival waveform segment set. For each acoustic channel, a segment is extracted within the original sequence of the acoustic channel, centered on the first arrival time of that acoustic channel in the first arrival time sequence. A second time window length is set, with an optimal range of 5 milliseconds to 50 milliseconds. The selection rule for the second time window length is that the longer the rising edge duration of the acoustic signal of pipe leakage in the heated surface area, the larger the value of the second time window length; the shorter the arrival interval of the echo propagation in the heated surface area, the smaller the value of the second time window length, in order to reduce the proportion of echo components entering the time window segment. The segment extraction adopts a symmetrical approach before and after the first arrival time, with the length of the first segment and the length of the second segment both taken as half of the second time window length. The continuous sampling segment obtained by extraction is recorded as the time window segment. If the first arrival time is close to the boundary of the original sequence of the acoustic channel, the segment extraction is completed according to the available sampling points at the boundary, while still retaining the unified time scale corresponding to the first arrival time. This segment is repeated for each acoustic channel, and time window segments are recorded by associating channel number with the first arrival time to obtain the first arrival waveform segment set. The first arrival waveform segment set is extracted around the first arrival time sequence, so that each acoustic channel obtains the first arrival waveform shape under the same time reference, reducing the waveform tail caused by multipath superposition and thus reducing the probability of the positioning result deviating to adjacent pipe rows.

[0075] A consistent constraint expression is applied to the noise baseline segment, rise threshold, first arrival time sequence, and first arrival waveform segment set. For each acoustic channel, the unified time scale range corresponding to the noise baseline segment, the rise threshold value, the first arrival time in the first arrival time sequence, and the time window segment in the first arrival waveform segment set are associated with the same channel number. The association relationship uses the channel number as a unique index, and the unified time scale remains unchanged during association. For each time window segment within the first arrival waveform segment set, the original waveform sample value is retained, and no smoothing or resampling processing is performed on the waveform sample value to preserve the difference information in the rise pattern between direct waves and multipath waves. A consistent expression is established for the noise baseline segment, rise threshold, first arrival time sequence, and first arrival waveform segment set using the channel number and unified time scale, reducing the first arrival misjudgment caused by the inconsistency of cross-channel time bases, improving the reusability of first arrival information under multipath superposition conditions, and thus alleviating the problem of insufficient positioning reliability caused by the position drift of the isochronous pseudo-solution with changes in operating conditions.

[0076] The candidate sorting module is used to extract features from the first arrival waveform segment set with reference to the noise baseline segment, generate a path fingerprint table, determine the position grid table based on the pre-acquired heated surface layout map, calculate the propagation time of the position grid table according to the sound velocity classification map and the pre-acquired sensor position table, sort to obtain the expected arrival sorting table, compare the expected arrival sorting table with the first arrival time sequence to obtain the sorting difference table, and filter the candidate position table from the sorting difference table.

[0077] The following describes a specific implementation method for obtaining a path fingerprint table by extracting features from the first arrival waveform segment set and using the noise baseline segment as a reference. A location grid table is then determined based on a pre-acquired heated surface layout diagram. The propagation time of the location grid table is calculated according to the sound velocity classification diagram and a pre-acquired sensor location table. This is followed by sorting the locations to obtain a predicted arrival ranking table. Finally, the predicted arrival ranking table is compared with the first arrival time sequence to obtain a ranking difference table. The candidate location table is then selected from the ranking difference table.

[0078] The first waveform segment set uses the noise baseline segment as a reference to extract features and generate a path fingerprint table. For each acoustic channel, the noise baseline segment corresponds to a uniform timescale range. The first waveform segment set contains time window segments truncated around the first time series. The amplitude level of the noise baseline segment is used as a background reference. The average energy is calculated from the absolute value of the waveform sample values ​​of the noise baseline segment, and the average energy is normalized according to the uniform timescale length of the noise baseline segment to obtain the noise baseline average energy.

[0079] For each time window segment within the initial waveform segment set, the starting segment of the time window segment is extracted. The length of the starting segment is limited by the length of the early segment, with an optimal range of 0.5 ms to 5 ms. The selection rule for the length of the early segment is that the higher the sampling rate of the original acoustic channel sequence, the smaller the value of the early segment; the earlier the reverberation caused by multipath propagation in the heated surface region enters the time window segment, the smaller the value of the early segment. The average energy is calculated from the absolute value of the waveform sample values ​​of the starting segment. The average energy is normalized according to the uniform time scale length of the starting segment to obtain the average energy of the starting segment. The early energy ratio is the ratio of the average energy of the starting segment to the average energy of the noise baseline. The early energy ratio is used to characterize the degree of amplitude rise of the starting segment relative to the noise baseline segment.

[0080] For each time window segment within the initial waveform segment set, a peak interval sequence is extracted. The waveform sample values ​​of the time window segment undergo debiasing processing, with the center level of the noise baseline segment as a reference. The absolute value of the debiased waveform sample value is used as the peak detection sequence, and the root mean square amplitude of the noise baseline is calculated from the absolute value of the debiased waveform sample value of the noise baseline segment. A first peak threshold coefficient is set, with an optimal range of 2 to 6. The selection rule for the first peak threshold coefficient is that the higher the frequency of short-time spikes in the background mechanical noise of the heated surface area, the larger the value of the first peak threshold coefficient; the coarser the sampling quantization level of the original acoustic channel sequence, the larger the value of the first peak threshold coefficient. The peak threshold is determined by the root mean square amplitude of the noise baseline and the first peak threshold coefficient. A minimum peak interval is set, with an optimal range of 0.1 ms to 2 ms. The selection rule for the minimum peak interval is that the higher the sampling rate of the original acoustic channel sequence, the smaller the value of the minimum peak interval; the higher the frequency of short-time spikes in the background mechanical noise of the heated surface area, the larger the value of the minimum peak interval. The peaks are defined as local maxima above the peak threshold in the peak detection sequence, with the interval between peaks not less than the minimum peak interval. The peak timescale is recorded with the first arrival time in the first arrival time sequence as zero, and the peak interval sequence is obtained based on the timescale difference between adjacent peaks. The peak interval sequence records the time interval between adjacent peaks within a time window. This process establishes a correspondence between the peak detection threshold and the noise baseline segment, reducing the probability of short-duration spikes being mistaken for echo peaks, thereby reducing the probability of misrepresentation of path morphology under multipath superposition conditions.

[0081] For each time window segment within the initial waveform segment set, extract the frequency band energy column. Set frequency band boundaries, with the lower bound at 0 Hz and the upper bound at half the sampling rate of the original acoustic channel sequence. Preset the number of frequency bands to the first frequency band number, with an optimal range of 3 to 12. The selection rule for the first frequency band number is: the wider the frequency band distribution of background mechanical noise in the heated surface area, the larger the value of the first frequency band number; the more channels in the acoustic sensor array, the smaller the value of the first frequency band number. Divide the frequency band from 0 Hz to the upper bound into several first frequency bands according to the frequency band boundaries. Calculate the discrete frequency components for each time window segment based on the sampling point sequence corresponding to the second time window length. For each frequency band, select the discrete frequency components falling within the boundary of that frequency band, and accumulate the squared amplitudes of the discrete frequency components to obtain the frequency band energy of that frequency band. Record the frequency band energy in frequency band order to obtain the frequency band energy column. This processing confines the energy distribution of time window segments within the frequency band boundary, reducing the inconsistency in frequency band aperture caused by different sampling rates and different channels, thereby reducing the probability of incomparable frequency band energy characteristics under multipath superposition conditions.

[0082] Channel number, early energy ratio, peak interval sequence, and frequency band energy column are written into the path fingerprint table. The path fingerprint table maintains the same set of channel numbers as the first arrival time sequence. Feature extraction of the first arrival waveform segment set is constrained by the average energy of the noise baseline, reducing the dependence of the early energy ratio on the length of the noise baseline segment, improving the comparability of the early energy ratio under different channels and different operating periods, thereby reducing false triggering inputs caused by multipath superposition and reducing the probability of spurious solutions with equal time difference.

[0083] A location grid table is determined based on a pre-acquired heating surface layout diagram. The heating surface layout diagram describes the spatial extent of the tube bundles in the heating surface region and the spatial boundaries of the furnace wall and steel structure. The boundary of the heating surface region given in the heating surface layout diagram is used as the grid coverage area. A first grid spacing is set, which limits the interval length between adjacent grid points in the location grid table in the planar direction of the heating surface region. The first grid spacing corresponds to the spatial resolution of the location grid table, and its unit is meters. The first grid spacing constrains the spatial discrete scale of the candidate location table. The preferred range of the first grid spacing is 0.2 meters to 1.5 meters. The selection rule for the first grid spacing is that the smaller the spacing between the heating surface tube bundles, the smaller the value of the first grid spacing, to reduce confusion caused by candidate location tables falling between adjacent tube rows; the fewer the number of sensors in the acoustic sensor array, the larger the value of the first grid spacing, to avoid the location grid table being too dense, making it difficult to distinguish multiple location grid points in time order. Regular grid points are generated within the boundary of the heating surface region according to the first grid spacing. Each grid point records three-dimensional coordinates, and the three-dimensional coordinates use the same coordinate reference as the sensor location table. The grid point numbers and 3D coordinates are recorded as a location grid table. The spatial constraints of the heated surface layout are transformed into a location grid table, which avoids pushing the location of the leak source outside the heated surface area during the localization process, reduces the probability of false solutions caused by multipath reflection falling into the furnace wall or steel structure neighborhood, and thus improves the spatial reliability of the localization output.

[0084] The propagation time is calculated using the sound velocity classification map and the pre-acquired sensor location table against the location grid table. The sensor location table provides the three-dimensional coordinates of the sensor corresponding to each channel. The sound velocity classification map shows the sound velocity level distribution within the monitoring area at a unified time scale. Using the unified time scale position corresponding to the first arrival time sequence as the current time scale, the sound velocity level distribution is retrieved from the sound velocity classification map according to the current time scale. A sound velocity value is assigned to each sound velocity level, and the sound velocity value is taken from the sound velocity table. The preferred range for the sound velocity table is 200 m / s to 800 m / s. The selection rule for the sound velocity table is that the higher the temperature range covered by the flue gas temperature measurement point sequence, the larger the upper limit value of the sound velocity table; the lower the temperature range covered by the flue gas temperature measurement point sequence, the smaller the lower limit value of the sound velocity table.

[0085] For each grid point in the location grid table, determine the propagation path from that grid point to each sensor in the sensor location table. The propagation path is a straight line connecting the three-dimensional coordinates of the grid point and the three-dimensional coordinates of the sensor. This straight line traverses several sound velocity level regions within the spatial range corresponding to the sound velocity classification map. A first-step length is set, preferably between 0.1 meters and 1 meter. The selection rule for the first-step length is that the smaller the first grid spacing in the location grid table, the smaller the first-step length; and the denser the boundaries of the sound velocity level regions in the sound velocity classification map, the smaller the first-step length. A sequence of path sampling points is generated along the straight line according to the first-step length. For each path sampling point sequence, the sound velocity level is determined in the sound velocity classification map. Continuous segments are divided based on whether adjacent path sampling point sequences have the same sound velocity level, resulting in a sound velocity level segment sequence. For each sound velocity level segment sequence, the segment length is determined based on the number of path sampling point sequences within the segment and the first-step length. A correspondence is established between the segment length and the sound velocity value corresponding to the sound velocity level of that segment.

[0086] For each sound velocity level segment sequence, the propagation time of that segment is determined by its length and sound velocity value. The propagation times of all segments are then summed to obtain the propagation time from that grid point to the sensor. A set of propagation times is generated for each grid point. The grid point number, channel number, and propagation time are recorded as a propagation time table. This method incorporates the sound velocity classification map caused by the flue gas temperature measurement point sequence into the propagation time calculation. The propagation time maintains a correspondence with the sound velocity level distribution under a unified time scale, reducing the drift of location results caused by temperature field fluctuations and alleviating the problem of the same leak source being located in different areas at different times.

[0087] The expected arrival order table is obtained by sorting according to the propagation time table. For each grid point in the location grid table, the propagation time of each channel corresponding to that grid point is retrieved from the propagation time table. The propagation times of each channel are sorted in ascending order of value, and the sorting result forms the channel arrival order. The grid point number and the channel arrival order are recorded as the expected arrival order table. A sorting interval is set to handle cases where the propagation time difference is close. The preferred range of the sorting interval is 0.05 ms to 2 ms. The selection rule for the sorting interval is that the lower the sampling rate of the original acoustic channel sequence, the larger the sorting interval value; the higher the density of the tube bundle in the heated surface area and the smaller the difference in propagation path length, the larger the sorting interval value should be, so as to reduce frequent reordering caused by extremely small differences. If the difference in propagation time between two adjacent channels does not exceed the sorting interval, the two channels are marked as the same arrival level in the expected arrival order table, and no order is mandatory. Using the expected arrival sorting table to express the arrival order of the location grid table under the condition of sound velocity classification, the multi-peak nature of pseudo-solutions caused by random reordering when the propagation time is approximately equal is reduced, thereby reducing the probability of multiple candidate locations formed by equidistant pseudo-solutions.

[0088] The expected arrival ranking table is compared with the first arrival time sequence to obtain a ranking difference table, and the candidate location table is obtained by filtering from the ranking difference table. The first arrival time sequence records the first arrival time according to the channel number. The first arrival time sequence is sorted in ascending order to obtain the observed arrival order, and the channel set of the observed arrival order is consistent with the sensor location table. The expected arrival ranking table gives the channel arrival order for each grid point. The channel arrival order contains several arrival levels, and the arrival level is used to characterize the set of channels whose propagation time difference does not exceed the ranking interval.

[0089] For each grid point in the expected arrival ranking list, the channel arrival order of that grid point is taken. The channel arrival order is compared layer by layer with the observed arrival order. The comparison begins from the first arrival level of the channel arrival order. For the current arrival level, the number of channels contained in that arrival level is taken as the level channel number. Starting from the current position of the observed arrival order, several channels from consecutive levels are taken to form an observation level set, the number of channels in the observation level set being the same as the level channel number. The channel set of this arrival level is compared with the observation level set; if they are the same, they are marked as consistent; if they are different, they are marked as inconsistent. After completing the comparison, the current position of the observed arrival order is moved forward by several level channels to enter the comparison of the next arrival level, until all arrival levels of the grid point have been compared. The ranking difference number is obtained by counting the inconsistent levels.

[0090] The grid point numbers and sorting differences are recorded as a sorting difference table. A difference threshold is set, with an optimal range of 1 to 5. The selection rule for the difference threshold is that the more channels the acoustic sensor array has, the larger the difference threshold value; the stronger the multipath reflection conditions in the heated surface area, the smaller the difference threshold value. Grid points with sorting differences not exceeding the difference threshold are selected based on the sorting difference table to form a candidate location table. The number of channels at the arrival level constrains the value range of the observation level set. The counting rule for the sorting difference number is deterministic, reducing the fluctuation of the sorting difference number caused by different alignment methods. The first arrival time sequence is used to apply difference constraints to the expected arrival sorting table, suppressing the probability of isochronous pseudo-solutions entering the candidate location table, thereby reducing the probability of the location result deviating from adjacent pipe rows or heated surface areas and improving the reliability of leak location.

[0091] The results output module generates an expected echo band table based on the candidate location table and the pre-acquired structural location table. It compares the expected echo band table with the path fingerprint table and calculates the candidate scores in conjunction with the sorting difference table. Based on the candidate scores, it generates a location distribution table, which provides the location results and uncertainties.

[0092] The following describes the specific implementation method of generating an expected echo band table based on a candidate location table and a pre-acquired structural location table. This table is then compared with a path fingerprint table, and candidate scores are calculated using a ranking difference table. A location distribution table is generated based on these candidate scores, and the location distribution table provides the specific implementation details regarding the location results and uncertainties.

[0093] A predicted echo band table is generated based on the candidate location table and the structural location table. The structural location table is pre-acquired data, recording the spatial boundary points of the boiler wall and steel structure. These spatial boundary points use a three-dimensional coordinate reference consistent with the sensor location table. For each candidate location in the candidate location table, its three-dimensional coordinates are obtained. For each channel, the three-dimensional coordinates of the corresponding sensor in the sensor location table are obtained. A direct direction is determined by connecting the candidate location to the sensor. Nearby boundary points are retrieved from the structural location table along this direct direction. These nearby boundary points are the spatial boundary points closest to the connecting line and are used to approximately characterize the transit points where the channel may generate reflected or diffracted waves.

[0094] The detour distance is set, with an optimal range of 0.2 meters to 2 meters. The selection rule for the detour distance is that the smaller the width of the detourable channel outside the tube bundle in the heating surface layout diagram, the smaller the detour distance; the closer the boiler wall and steel structure are to the heating surface area, the smaller the detour distance. The distance from the adjacent boundary point to the line connecting the adjacent boundary point to the direct direction is calculated. When the distance from the adjacent boundary point to the line connecting the adjacent boundary point to the direct direction does not exceed the detour distance, the adjacent boundary point is used as the echo reverberation transfer point of that channel. When the distance from the adjacent boundary point to the line connecting the adjacent boundary point to the direct direction exceeds the detour distance, the channel does not generate the expected echo delay interval.

[0095] For channels that meet the detour distance constraints, the nearest boundary point is used as the transfer point, forming two spatial distances from the candidate location to the nearest boundary point and then to the sensor. These two spatial distances are segmented and converted according to the sound velocity level at the corresponding time-scale position on the sound velocity classification map to obtain the center reverberation delay of the channel. The reverberation interval length is set, with an optimal range of 0.5 milliseconds to 10 milliseconds. The selection rule for the reverberation interval length is that the more sound velocity levels on the sound velocity classification map, the smaller the reverberation interval length; the greater the fluctuation amplitude of the flue gas temperature measurement point sequence, the larger the reverberation interval length. Taking the center reverberation delay as the center, half the reverberation interval length forward is taken as the lower limit of the expected reverberation delay interval, and half the reverberation interval length backward is taken as the upper limit of the expected reverberation delay interval. If the lower limit of the expected reverberation delay interval is less than 0 milliseconds, then the lower limit of the expected reverberation delay interval is set to 0 milliseconds.

[0096] For each candidate location and each channel that meets the detour distance constraint, a corresponding expected echo delay interval is formed. The candidate location number, channel number, lower limit of the expected echo delay interval, and upper limit of the expected echo delay interval are recorded to obtain the expected echo band table.

[0097] The expected echo band table describes the time range within which echo components may appear in each channel at candidate locations under conditions where reflected or diffracted waves exist between the boiler wall and steel structure. The expected echo band table uses candidate location number and channel number as indexes. Each index corresponds to one expected echo delay interval record. This record includes the lower and upper limits of the expected echo delay interval. The time base for both the lower and upper limits is consistent with the first arrival time sequence, and the zero point is consistent with the first arrival time in the first arrival time sequence. The time unit is seconds. Channels that do not generate expected echo delay intervals in the expected echo band table are represented by empty records. Empty records indicate that the channel does not meet the detour distance constraint and is not included in the echo time constraint for that candidate location.

[0098] The expected echo band table is obtained by combining the structural location table and the sound velocity classification map. The structural location table provides the spatial boundary points between the boiler furnace wall and the steel structure. These spatial boundary points determine the echo reverberation reversal point, which is constrained by the detour distance. The sound velocity classification map provides the sound velocity level distribution along the echo propagation path. The sound velocity levels correspond to the first sound velocity table and are used to convert the two spatial distances from the candidate location to the echo reversal point and then to the sensor into the center echo delay. The center echo delay and the echo interval length together determine the lower and upper limits of the expected echo delay interval. The expected echo delay interval covers the time range before and after the center echo delay, and the lower limit of the expected echo delay interval is not less than 0 seconds.

[0099] By incorporating the boiler furnace wall and steel structure into the reverberation delay interval expression, the reverberation reversal point is constrained by the detour distance. The expected reverberation delay interval is determined by the central reverberation delay and the length of the reverberation interval. This allows the reflected and diffracted waves in multipath propagation to have a constrained range in time, reducing the probability of the isochronous pseudo-solution forming multiple approximate solutions in space, thereby mitigating the situation where the positioning result deviates from the adjacent tube bank or the area of ​​the heated surface.

[0100] The expected echo band table is compared with the path fingerprint table to form a constrained expression of the multipath superposition pattern. The path fingerprint table already includes channel number, early energy ratio, peak interval sequence, and frequency band energy column. For each candidate position in the candidate position table, the expected echo delay interval corresponding to that candidate position is retrieved from the expected echo band table according to the channel number. For the same channel, the peak interval sequence is retrieved from the path fingerprint table. The peak interval sequence is converted into the relative time-scale position of the peak occurrence, with the first arrival time in the first arrival time sequence as the zero point. A lower limit for peak count is set, with the preferred range of 2 to 6. The selection rule for the lower limit of peak count is that the higher the frequency of the background mechanical noise spikes in the heated surface area, the larger the value of the lower limit of peak count; the higher the sampling rate, the larger the value of the lower limit of peak count, which is used to require a more complete echo pattern. If the number of peaks falling into the expected echo delay interval at the relative time-scale position is not less than the lower limit of peak count, then the channel is considered to have an in-echo band response at that candidate position. For the same channel, the frequency band energy column is compared with the corresponding frequency band energy level of the noise baseline segment. A frequency band rise threshold is set, with an optimal range of 1.2 to 3. The selection rule for the frequency band rise threshold is that the higher the proportion of background mechanical noise in the heated surface area in the high-frequency band, the larger the frequency band rise threshold value; the higher the main frequency band of the leakage acoustic signal, the smaller the frequency band rise threshold value. If a frequency band energy rise occurs near the relative time scale position corresponding to the expected echo delay interval, and the rise exceeds the frequency band rise threshold, then the channel is considered to meet the echo band energy characteristics at that candidate position. For each candidate position in the candidate position table, the above comparison process is repeated in each channel to form the echo band matching situation for that candidate position. The peak interval and frequency band energy of the path fingerprint table are time-constrained using the expected echo band table to reduce waveform misjudgment caused by the superposition of direct waves and multipath waves, incorporate the time structure of multipath propagation into the positioning constraint, and reduce the chance of the isochronous pseudo-solution being valid in both the time and spatial dimensions.

[0101] Candidate scores are calculated using the sorting difference table. The sorting difference table records candidate position numbers and sorting difference counts. For each candidate position in the candidate position table, the sorting difference count is read from the sorting difference table; this count characterizes the consistency between the first arrival time sequence and the expected arrival sorting table. A difference deduction is set, with an optimal range of 0.5 to 5. The selection rule for the difference deduction is that the more channels the acoustic sensor array has, the smaller the difference deduction value; the stronger the reflection conditions of the heated surface area structure, the larger the difference deduction value, used to improve the suppression of false solutions. Based on the echo band matching, the number of channels that satisfy both the response within the echo band and the echo band energy characteristics is counted. A matching bonus is set, with an optimal range of 0.2 to 2. The selection rule for the matching bonus is that the more channels in the first frequency band, the smaller the matching bonus value; the longer the echo interval, the smaller the matching bonus value, to avoid an overly wide echo interval leading to overly easy matching. The candidate score is calculated by multiplying the matching bonus by the number of channels to obtain the bonus value, and then multiplying the difference deduction by the sorting difference to obtain the deduction value. The candidate score is the bonus value minus the deduction value. For each candidate location, a candidate location number and candidate score are recorded to form a candidate score table. First-arrival sorting consistency and echoband matching are simultaneously included in the candidate score to reduce the sensitivity of a single equation solution based solely on arrival time difference to multipath propagation, suppress the occupation of the candidate set by isochronous solutions, and reduce the occurrence of positioning results being included in different adjacent pipe arrays at different time periods.

[0102] A location distribution table is generated based on the candidate scores. The candidate scores are sorted from highest to lowest. For each candidate location, the three-dimensional coordinates in the candidate location table are read, and the candidate location number, three-dimensional coordinates, and candidate score are recorded as the location distribution table. A distribution cutoff threshold is set, with an optimal range of 0.5 to 0.9. The selection rule for the distribution cutoff threshold is that the smaller the spacing between tube bundles in the heated surface area, the larger the distribution cutoff threshold value, to reduce the diffusion of the location distribution table between adjacent tube bundles; the larger the fluctuation range of the flue gas temperature measurement point sequence, the smaller the distribution cutoff threshold value, to preserve the location uncertainty range caused by changes in the sound velocity classification map. Using the highest candidate score in the location distribution table as a reference, candidate locations with a ratio of candidate score to the highest candidate score not lower than the distribution cutoff threshold are selected as the high-score candidate set, and the spatial coverage of the high-score candidate set is marked in the location distribution table. Using a location distribution table to express the spatial distribution of candidate scores avoids the location output from only retaining single-point results and ignoring multimodality. It presents the spatial uncertainty range under multi-path superposition conditions in data form, thereby reducing the probability that the joint judgment process will take a false solution single point as a definite location and trigger unnecessary actions.

[0103] The positioning results and uncertainties are provided by the positioning distribution table. The table records candidate location numbers, 3D coordinates, and candidate scores. The positioning result is the 3D coordinates corresponding to the candidate location with the highest score in the positioning distribution table, and these coordinates are located within the boundary of the heated surface area. An uncertainty radius is set, with an optimal range of 0.5 meters to 6 meters. The selection rule for the uncertainty radius is that the smaller the first grid spacing of the location grid table, the smaller the uncertainty radius; and the larger the length of the response interval in the expected response zone table, the larger the uncertainty radius.

[0104] Centered on the three-dimensional coordinates of the positioning result, a local candidate set is obtained by searching within the high-scoring candidate set for candidate locations whose spatial distance from the positioning result does not exceed the uncertainty radius. If the local candidate set contains multiple candidate locations, the uncertainty is determined by the largest spatial distance from the positioning result within the local candidate set. If the local candidate set contains only the candidate location corresponding to the positioning result, the uncertainty is half of the first grid distance.

[0105] When a candidate location exists within the high-scoring candidate set and the spatial distance between that candidate location and the positioning result exceeds the uncertainty radius, the uncertainty is defined as the spatial distance from the positioning result to the highest-scoring candidate location within the high-scoring candidate set. This uncertainty covers the maximum spatial separation range of the high-scoring candidate set. The positioning result and uncertainty are output as the positioning distribution table. Under conditions of multipath propagation and graded changes in sound velocity, the positioning result and uncertainty are output simultaneously. The uncertainty covers both the local expansion of the positioning result's neighborhood and the spatial separation of the high-scoring candidate set from the positioning result, reducing the sensitivity of single-point positioning results to isochronous spurious solutions. This provides spatial constraints for handling pipeline leaks in heated areas, reducing delays caused by an insufficient investigation scope.

[0106] Example 2:

[0107] This embodiment provides a numerical example of a multi-source joint judgment system for pipeline leaks, demonstrating the determination of the maximum amplitude of a noise baseline segment in a certain channel, the determination of the rise threshold, the determination of the first arrival time, the method of obtaining the early energy ratio, the method of forming the propagation time sorting of a certain grid point, the method of counting the sorting difference number, the method of constructing the lower and upper limits of the expected echo delay interval, the method of calculating the candidate score, and the method of outputting the location result and uncertainty. In this embodiment, the acoustic sensor array includes channel 1, channel 2, and channel 3. The sensor position table gives the three-dimensional coordinates as follows: channel 1 corresponds to sensor coordinates 0m, 0m, 0m; channel 2 corresponds to sensor coordinates 4m, 0m, 0m; and channel 3 corresponds to sensor coordinates 0m, 3m, 0m. The position grid table includes grid point 1 and grid point 2. The three-dimensional coordinates of grid point 1 are 1m, 1m, 0m, and the three-dimensional coordinates of grid point 2 are 3m, 2m, 0m. The structural position table includes the spatial boundary points of the boiler furnace wall and the steel structure, where the adjacent boundary points related to channel 1 are 2m, 1m, 0m.

[0108] The original acoustic channel sequence uses a sampling rate of 200 kHz to form a unified time scale. A first time window of 5 seconds is selected for channel 1 as the noise baseline segment. De-polarization processing is performed within this segment, with the center level set to 0 mV. In the de-polarized noise baseline segment, the maximum absolute value of the waveform sample is 0.6 mV, and the maximum amplitude is also 0.6 mV. An exceedance factor of 0.5 is set, and the exceedance is amplified to 0.3 mV based on the maximum amplitude. The rise threshold is the sum of the maximum amplitude and the exceedance, resulting in a rise threshold of 0.9 mV. In this example, the rise threshold is determined by the noise baseline segment, and the threshold size is consistent with the waveform sample amplitude size.

[0109] The first arrival time sequence is determined based on the rise threshold. The first continuous duration is set to 0.5 milliseconds. The original acoustic channel sequence of channel 1 undergoes de-biasing processing, with the center level of the noise baseline segment as the de-biasing reference. The absolute value of the de-biased waveform sample value is taken as the envelope rise. Starting at a unified timescale of 100 seconds, the envelope rise of channel 1 at timescale 100.010000 is 0.85 mV, which does not exceed the rise threshold of 0.9 mV; at timescale 100.010005, the envelope rise is 0.95 mV, exceeding the rise threshold of 0.9 mV. Continuing the check along the unified timescale, the envelope rise of channel 1 within a 0.5 millisecond range starting from time 100.010005 exceeds 0.9 mV, satisfying the first continuous duration of 0.5 milliseconds. The first arrival time of channel 1 is set to 100.010005 seconds. Using the same rules, the first arrival time for channel 3 is set to 100.012585 seconds, and the first arrival time for channel 2 is set to 100.014305 seconds. The first arrival time sequence is recorded according to channel number: channel 1 corresponds to 100.010005 seconds, channel 2 corresponds to 100.014305 seconds, and channel 3 corresponds to 100.012585 seconds. The observed arrival order, from smallest to largest first arrival time, is channel 1, channel 3, and channel 2.

[0110] The initial waveform segment set is truncated according to the second time window length. The second time window length is 20 milliseconds. For channel 1, a time window segment is truncated from the original acoustic channel sequence centered at the initial arrival time of 100.010005 seconds, resulting in the time window segment of channel 1. The early segment length is 1 millisecond. The average energy of the noise baseline segment of channel 1 is calculated. The average energy of the noise baseline is obtained by averaging the squares of the absolute values ​​of the debiased waveform samples; in the example, the average energy of the noise baseline is 0.12 mV squared. The average energy of the starting segment of the time window segment of channel 1 is calculated; in the example, the average energy of the starting segment is 1.20 mV squared. The early energy ratio is the ratio of the average energy of the starting segment to the average energy of the noise baseline; in the example, the early energy ratio is 10. This early energy ratio does not change with the length of the noise baseline segment, facilitating comparison between different channels and different time periods.

[0111] The expected arrival order table and the order difference table are formed according to the propagation time table. The sound velocity classification map takes the sound velocity level distribution at the current time scale as the sound velocity level 2 covering the monitoring area. The sound velocity table gives the sound velocity value of 340 m / s corresponding to sound velocity level 2. The propagation path is a straight line connecting grid points and sensors. In the example, the sound velocity level remains unchanged along the path, and the propagation time is determined by the spatial distance and the sound velocity value. The distance from grid point 1 to the sensor corresponding to channel 1 is 1.414 meters, and the propagation time is 4.16 milliseconds; the distance from grid point 1 to the sensor corresponding to channel 3 is 2.236 meters, and the propagation time is 6.58 milliseconds; the distance from grid point 1 to the sensor corresponding to channel 2 is 3.162 meters, and the propagation time is 9.30 milliseconds. The channel arrival order of grid point 1 is channel 1, channel 3, channel 2. The distance from grid point 2 to the sensor corresponding to channel 2 is 2.236 meters, with a propagation time of 6.58 milliseconds; the distance from grid point 2 to the sensor corresponding to channel 3 is 3.162 meters, with a propagation time of 9.30 milliseconds; and the distance from grid point 2 to the sensor corresponding to channel 1 is 3.606 meters, with a propagation time of 10.61 milliseconds. The channel arrival order of grid point 2 is channel 2, channel 3, and channel 1. The grid point number, channel number, and propagation time are recorded as a propagation time table, and the channel arrival order is recorded as an expected arrival sequence table.

[0112] The sorting difference table is formed by comparing arrival levels layer by layer. In the example, each arrival level contains one channel. For grid point 1, the first arrival level channel set is Channel 1, and the observation level set is Channel 1, which is formed by taking one consecutive channel from the current position of the observed arrival order. If the two are the same, it is considered consistent. The second arrival level channel set is Channel 3, and the observation level set is Channel 3. If the two are the same, it is considered consistent. The third arrival level channel set is Channel 2, and the observation level set is Channel 2. If the two are the same, it is considered consistent. The sorting difference number for grid point 1 is 0. For grid point 2, the first arrival level channel set is Channel 2, and the observation level set is Channel 1. If the two are different, it is considered inconsistent. The second arrival level channel set is Channel 3, and the observation level set is Channel 3. If the two are the same, it is considered consistent. The third arrival level channel set is Channel 1, and the observation level set is Channel 2. If the two are different, it is considered inconsistent. The sorting difference number for grid point 2 is 2. The grid point number and the sorting difference number are recorded in the sorting difference table. The difference threshold is set to 2. Grid points whose sorting difference does not exceed the difference threshold are entered into the candidate position table. In the example, the candidate position table contains grid point 1 corresponding to candidate position number 1 and grid point 2 corresponding to candidate position number 2.

[0113] The expected echo delay interval is constructed using the structural location table, defining its lower and upper limits. A detour distance of 1 meter is used. For candidate location 1 and channel 1, the 3D coordinates of the candidate location are 1 meter, 1 meter, and 0 meters, while the corresponding sensor coordinates for channel 1 are 0 meters, 0 meters, and 0 meters. A line connecting these points in the direct direction is used to determine the desired echo delay. The adjacent boundary points are defined as 2 meters, 1 meter, and 0 meters. The distance from the adjacent boundary point to the line connecting the direct direction is calculated; in the example, it is 0.71 meters, not exceeding the 1-meter detour distance. The adjacent boundary point serves as the echo reverberation transfer point. The distance from the candidate location to the adjacent boundary point is 1 meter, and the distance from the adjacent boundary point to the corresponding sensor in channel 1 is 2.236 meters. These two spatial distances are converted to a sound speed of 340 meters per second (equivalent to sound speed level 2) to obtain a center echo delay of 9.52 milliseconds. The echo interval length is 2 milliseconds. The lower limit of the expected echo delay interval is 8.52 milliseconds, calculated by moving 1 millisecond forward from the center echo delay. The upper limit of the expected echo delay interval is 10.52 milliseconds, calculated by moving 1 millisecond backward from the center echo delay. Using the same criteria, the expected echo delay interval of candidate position number 1 in channels 2 and 3, and the expected echo delay interval of candidate position number 2 in channels 1, 2, and 3 can be obtained. The candidate position number, channel number, lower limit of the expected echo delay interval, and upper limit of the expected echo delay interval are recorded as the expected echo band table.

[0114] Candidate scores are formed by combining the expected echo band table and the path fingerprint table. A peak threshold is required to extract the peak interval sequence from the time window segment of channel 1. The root mean square amplitude of the noise baseline is calculated for the noise baseline segment of channel 1; in the example, it is 0.25 mV. The first peak threshold coefficient is 4, and the peak threshold is 1.0 mV. The minimum peak interval is 0.3 ms. Checking the time window segment of channel 1, within the range of the lower limit of the expected echo delay interval (8.52 ms) to the upper limit of the expected echo delay interval (10.52 ms), if the peak detection sequence shows three local maxima exceeding the peak threshold, and the interval between each local maxima is not less than 0.3 ms, channel 1 satisfies the in-band echo response. The same rule is applied to channels 2 and 3. In the example, channels 1, 2, and 3 at candidate position number 1 all satisfy the in-band echo response, and channels 1, 2, and 3 at candidate position number 2 all satisfy the in-band echo response. The multipath coincidence number is the number of channels that satisfy the response within the echo band. In the example, the multipath coincidence number for candidate location number 1 is 3, and the multipath coincidence number for candidate location number 2 is 3. The difference deduction is 0.5, and the matching bonus is 1.2. The candidate score for candidate location number 1 is 3.6, corresponding to the bonus value of 3 multipath coincidence number and 0 deduction value for the sorting difference number. The candidate score for candidate location number 2 is 2.6, corresponding to the bonus value of 3 multipath coincidence number and 1.0 deduction value for the sorting difference number. The candidate location number, 3D coordinates, and candidate score are recorded in a location distribution table.

[0115] The positioning results and uncertainties are given by the positioning distribution table. The positioning results are taken as the three-dimensional coordinates of 1 meter, 1 meter, and 0 meters corresponding to candidate position number 1 with the highest candidate score. The uncertainty radius is taken as 1 meter. Centered on the positioning results, candidate positions with a spatial distance of no more than 1 meter are searched in the high-scoring candidate set. In the example, the local candidate set only contains candidate position number 1. The uncertainty is initially taken as half of the first grid distance. In the example, the first grid distance is 0.5 meters, and the uncertainty is 0.25 meters. The high-scoring candidate set is filtered according to the ratio of the candidate score to the highest candidate score being no less than 0.7. The ratio of the candidate score for candidate position number 2 is 0.72. The high-scoring candidate set includes candidate position number 1 and candidate position number 2. The spatial distance between candidate position number 2 and the positioning results is 2.24 meters, which exceeds the uncertainty radius of 1 meter. Therefore, the uncertainty is taken as the largest spatial distance from the positioning results in the high-scoring candidate set, which is 2.24 meters. The positioning distribution table outputs the positioning results as 1 meter, 1 meter, and 0 meters, with an uncertainty of 2.24 meters. When there are multipath propagation and graded changes in sound velocity, this output reflects both the expansion of the main peak neighborhood and the separation of secondary peaks far from the main peak in the uncertainty, reducing the sensitivity of single-point localization results to equidistant pseudo-solutions.

[0116] Example 3:

[0117] Please see Figure 2 As shown, this embodiment provides a multi-source joint judgment method for pipeline leakage, including:

[0118] The original acoustic channel sequence and the flue gas temperature measurement point sequence were collected, and the flue gas temperature measurement point sequence was graded to obtain a sound velocity classification map;

[0119] Segment truncation is performed within the original sequence of the acoustic channel to obtain the noise baseline segment. The rise threshold is determined based on the noise baseline segment. The moment when the rise threshold is first exceeded is taken as the first arrival time sequence. Time window segments are truncated around the first arrival time sequence to obtain the first arrival waveform segment set.

[0120] The first arrival waveform segment set is used as a reference to extract features and generate a path fingerprint table. The location grid table is determined based on the pre-acquired heated surface layout map. The propagation time of the location grid table is calculated according to the sound velocity classification map and the pre-acquired sensor location table. The expected arrival ranking table is obtained by sorting the expected arrival ranking table and comparing it with the first arrival time sequence to obtain a ranking difference table. The candidate location table is obtained by filtering from the ranking difference table.

[0121] Based on the candidate location table and the pre-acquired structural location table, an expected echo band table is generated. The expected echo band table is compared with the path fingerprint table, and the candidate score is calculated by combining the ranking difference table. A location distribution table is generated based on the candidate score, and the location result and uncertainty are given by the location distribution table.

[0122] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0123] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-source joint judgment method for pipeline leakage, characterized in that, include: The original acoustic channel sequence and the flue gas temperature measurement point sequence were collected, and the flue gas temperature measurement point sequence was graded to obtain a sound velocity classification map; Segmentation is performed within the original sequence of the acoustic channel to obtain the noise baseline segment, and the rise threshold is determined based on the noise baseline segment. The moment when the rise threshold is first exceeded is taken as the first arrival time sequence. For each acoustic channel, segmentation is performed within the original sequence of the acoustic channel with the first arrival time of that acoustic channel in the first arrival time sequence as the center. Set the second time window length, and use a symmetrical method for segment truncation before and after the first arrival time. The length of the first segment and the length of the second segment are both half of the second time window length. The continuous sampling segment obtained by truncation is recorded as the time window segment. Each acoustic channel is repeatedly segmented, and time window segments are recorded by associating channel number with the first arrival time to obtain the first arrival waveform segment set; The first arrival waveform segment set is used as a reference to extract features and generate a path fingerprint table. The location grid table is determined based on the pre-acquired heated surface layout map. The propagation time of the location grid table is calculated according to the sound velocity classification map and the pre-acquired sensor location table. The expected arrival ranking table is obtained by sorting the expected arrival ranking table and comparing it with the first arrival time sequence to obtain a ranking difference table. The candidate location table is obtained by filtering from the ranking difference table. A predicted echoband table is generated based on the candidate location table and a pre-acquired structural location table. The predicted echoband table is compared with the path fingerprint table, and candidate scores are calculated using a ranking difference table. A location distribution table is generated based on the candidate scores, providing the location results and uncertainties. The method for generating the path fingerprint table includes: For each time window segment within the first waveform segment set, the starting segment of the time window segment is extracted, and the length of the starting segment is limited by the length of the earlier segment. The average energy is calculated from the absolute value of the waveform sampled values ​​of the noise baseline segment. The average energy is then normalized according to the uniform time scale length of the noise baseline segment to obtain the average energy of the noise baseline. The average energy is calculated from the absolute value of the waveform sampled values ​​of the initial segment. The average energy is then normalized according to the uniform time scale length of the initial segment to obtain the average energy of the initial segment. The early energy ratio is the ratio of the average energy of the initial segment to the average energy of the noise baseline. The peak time scale is recorded with the first arrival time in the first arrival time sequence as the zero point, and the peak interval sequence is obtained according to the time scale difference between adjacent peaks. For each time window segment within the initial waveform segment set, extract the frequency band energy column; The path fingerprint table includes channel number, early energy ratio, peak interval sequence, and frequency band energy column.

2. The multi-source joint judgment method for pipeline leakage according to claim 1, characterized in that, The sound velocity classification diagram obtained by classifying the flue gas temperature measurement point sequence includes: The temperature grading interval is set to divide the flue gas temperature measurement point sequence into multiple temperature ranges and assign sound velocity levels. The sound velocity meter is set to determine the sound velocity value of each sound velocity level based on the center temperature of the temperature range. For each grid point in the location grid table, calculate the spatial distance between it and each flue gas temperature measuring point. Set a first proximity radius to filter the set of neighboring measuring points. Determine the weight according to the spatial distance and round down the weighted sound velocity level to obtain the sound velocity level of the grid point. Perform a graded update according to the graded update interval to obtain the sound velocity graded map.

3. The method according to claim 1, wherein, Methods for obtaining noise baseline segments include: Using a unified time scale as an index, the search duration is limited in the first part of the original acoustic channel sequence. For each channel within the search duration, the absolute value sequence of waveform sample values ​​is calculated. The mean amplitude sequence is obtained by averaging the absolute value sequence of waveform sample values ​​according to the unified time scale. The low-energy threshold is determined by the preset low-energy quantile in the mean amplitude sequence. The continuous time periods in the mean amplitude sequence that do not exceed the low-energy threshold are selected to obtain the baseline candidate time period set. The segment before the event is determined according to the length of the first time window, and the noise baseline segment is extracted.

4. The multi-source joint judgment method for pipeline leakage according to claim 1, characterized in that, Methods for determining the rising threshold include: The noise baseline segment is de-biased and the maximum amplitude is extracted. Set an exceedance factor, which is used to characterize the proportion of the rise threshold that exceeds the maximum amplitude; The excess amount is determined based on the maximum amplitude and the excess coefficient. The excess amount is the amplitude obtained by amplifying the maximum amplitude by the excess coefficient. The rise threshold is the sum of the maximum amplitude and the excess amount.

5. The multi-source joint judgment method for pipeline leakage according to claim 3, characterized in that, If the baseline candidate time period set contains a continuous time period with a length not less than the length of the first time window, then the earliest continuous time period that meets the length condition is taken as the pre-event segment; if the baseline candidate time period set does not contain a continuous time period with a length not less than the length of the first time window, then the longest continuous time period in the baseline candidate time period set is taken as the pre-event segment, and the length of the first time window is taken as the length of the longest continuous time period.

6. The multi-source joint judgment method for pipeline leakage according to claim 3, characterized in that, The maximum amplitude of the baseline is extracted from the noise baseline segment. Based on the maximum amplitude of the baseline, a preset amplitude upper limit is determined. Starting from the unified time mark at the end of the segment before the event, the search proceeds forward along the original sequence of acoustic channels to find the unified time mark position where the absolute value of the waveform sample value of any channel first exceeds the preset amplitude upper limit. This unified time mark is recorded as the event reference time mark. Each channel segment with a length equal to the length of the first time window before the event reference time mark is re-extracted as the segment before the event, and the noise baseline segment is updated.

7. The multi-source joint judgment method for pipeline leakage according to claim 1, characterized in that, Methods for extracting frequency band energy trains include: Set the frequency band boundary, with the lower bound being 0 Hz and the upper bound being half the sampling rate of the original sequence of the acoustic channel; The frequency band from 0 Hz to the upper limit is divided into several frequency bands in the first frequency band according to the frequency band boundary; Calculate the discrete frequency components for the time window segment based on the sampling point sequence corresponding to the second time window length; For each frequency band, select discrete frequency components that fall within the boundary of that frequency band, and accumulate the squared amplitudes of the discrete frequency components to obtain the band energy of that frequency band. Record the frequency band energy in frequency band order to obtain the frequency band energy sequence.

Images