Method and system for online monitoring of gas network pressure anomalies

Abstract

This invention relates to the field of gas pressure monitoring technology, and more particularly to an online monitoring method and system for abnormal gas pipeline network pressure. The method includes the following steps: performing multi-cycle baseline fusion based on the synchronous pressure filter values ​​under historical normal operating conditions to construct a dynamic expected baseline for the monitoring node at the current moment; calculating the dynamic pressure deviation between adjacent monitoring nodes by combining the topological connection relationship of the gas pipeline network and the pressure wave propagation delay, and determining the corresponding dynamic hydraulic impedance coefficient of the pipeline segment by combining the flow characteristics of the pipeline segment; and performing multi-dimensional feature fusion judgment based on the degree of deviation of the pressure filter value from the dynamic expected baseline, the extent of dynamic pressure deviation exceeding the standard, and the variation characteristics of the dynamic hydraulic impedance coefficient of the pipeline segment, to output the anomaly detection result, effectively improving the accuracy of the gas pipeline network pressure anomaly detection result.

Description

Technical Field

[0001] This invention relates to the field of gas pressure monitoring technology, and in particular to an online monitoring method and system for abnormal gas pipeline pressure. Background Technology

[0002] Gas pipeline networks are a vital infrastructure for urban energy supply, and real-time monitoring of these networks is crucial for ensuring gas supply security and preventing leaks. Currently, gas pipeline pressure monitoring primarily relies on pressure transmitters deployed at pipeline nodes, which identify anomalies by comparing real-time pressure with fixed thresholds or simple historical statistical values.

[0003] For example, Chinese patent application CN121881271A discloses an intelligent decision-making method for gas state changes based on real-time pressure monitoring, which includes: focusing on the perception needs of residential scenarios, configuring sensing devices to collect real-time pressure data from each monitoring node; performing multi-dimensional preprocessing on the pressure data, enhancing the retention of signals in the pressure threshold range while reducing noise, and avoiding the loss of boundary signals of small high-frequency pressure fluctuations at the level of less than 0.1 kPa; triggering state change alarms and outputting decision commands according to the degree of pressure deviation, so as to improve the accuracy of gas state identification and control flexibility.

[0004] However, the aforementioned existing technologies do not effectively self-diagnose the operating status of the pressure sensor itself. Abnormal readings caused by faults such as zero-point drift, sensitivity decay, or signal stagnation of the sensor can easily be confused with pressure drops caused by actual pipeline leaks. In addition, the method fails to effectively separate and integrate the multi-scale periodic fluctuation characteristics such as daily, semi-daily, and weekly cycles contained in the pressure time series of the gas pipeline network, resulting in insufficient baseline accuracy. At the same time, when analyzing the pressure relationship between adjacent monitoring nodes, the method does not consider the finite propagation speed of pressure waves in the gas medium, and does not combine the pressure deviation between nodes with the flow characteristics of the pipeline segment to extract the dynamic hydraulic impedance information of the pipeline segment, thus failing to effectively reflect the internal frictional resistance state of the pipeline segment.

[0005] Therefore, how to effectively distinguish between real pipeline leaks, sensor failures and normal transient disturbances in the process of online monitoring of gas pipeline pressure is an urgent problem to be solved. This is because existing technologies lack sensor status self-diagnosis, multi-cycle high-precision baseline construction, topology delay compensation and multi-dimensional physical feature fusion judgment mechanism. Summary of the Invention

[0006] To address the technical problem of effectively distinguishing between real pipeline leaks, sensor malfunctions, and normal transient disturbances, this invention provides an online monitoring method and system for abnormal gas pipeline pressure.

[0007] In a first aspect, the present invention provides an online monitoring method for abnormal pressure in gas pipeline networks, employing the following technical solution:

[0008] Online monitoring methods for abnormal pressure in gas pipeline networks include the following steps: S1. Acquire raw pressure data from each monitoring node deployed in the gas pipeline network, perform sensor status self-diagnosis and filtering on the raw pressure data to obtain a pressure filter value reflecting the static pressure level; S2. Perform multi-cycle baseline fusion based on the synchronous pressure filter value under historical normal operating conditions to construct the dynamic expected baseline of the monitoring node at the current moment; combine the topological connection relationship of the gas pipeline network and the pressure wave propagation delay to calculate the dynamic pressure deviation between adjacent monitoring nodes, and determine the corresponding dynamic hydraulic impedance coefficient of the pipe segment based on the flow characteristics of the pipe segment; S3. Based on the degree of deviation of the pressure filter value from the dynamic expected baseline, the exceedance of the dynamic pressure deviation, and the variation characteristics of the dynamic hydraulic impedance coefficient of the pipe segment, perform multi-dimensional feature fusion judgment to output the anomaly detection result.

[0009] This invention provides an online monitoring and anomaly early warning method that closely reflects the actual physical characteristics of gas pipeline networks. Addressing the shortcomings of existing technologies that rely on simple historical comparisons and lack hydraulic constraints on the pipeline network, easily misinterpreting sensor hardware failures or fluctuations in user gas consumption as actual leaks, this invention performs self-diagnosis and cleaning of sensor data at the underlying level and integrates multi-period dynamic baselines based on gas consumption tidal patterns at the macroscopic level. Furthermore, it introduces a time-delay compensation mechanism for pressure wave propagation in the physical pipeline network and dynamic hydraulic impedance calculation of pipe sections reflecting the internal wall friction state. Finally, it employs cross-validation judgment based on multi-dimensional features, alleviating the long-standing difficulty in decoupling between actual physical pipe bursts, chronic micro-leaks, sensor distortion, and legitimate transient disturbances in complex urban low- and medium-pressure gas pipeline networks. This effectively improves the intrinsic safety protection level of the gas supply network and reduces the ineffective emergency response rate for maintenance teams.

[0010] According to the online monitoring method for abnormal gas pipeline pressure provided by the present invention, sensor status self-diagnosis is performed on the raw pressure data, including: calculating the signal variance of the raw pressure data within a local time window, and using the proportion of consecutive sampling points with the same sign in the first-order difference sequence to the total number of data points within the window as the sign consistency index; if the signal variance is lower than a preset noise floor threshold and the sign consistency index is higher than a preset stagnation judgment threshold, it is determined to be a signal stagnation state and subsequent calculations are blocked; if the signal variance is higher than a preset drift judgment upper limit and the sign consistency index is higher than a preset drift judgment threshold, it is determined to be a rapid drift state; if the signal variance is between the preset noise floor threshold and the preset drift judgment upper limit and the sign consistency index is lower than or equal to the preset drift judgment threshold, it is determined to be a normal state; otherwise, it is determined to be a transition state.

[0011] This invention introduces a dynamic health check and filtering correction mechanism based on signal variance and symbol consistency characteristics at the edge data access end. Compared with the approach of receiving all raw data or using static filtering coefficients, this solution can capture and intercept the mechanical jamming dead zone and rapid electronic zero drift phenomenon that often occurs in industrial pressure transmitters. When the sensor health is confirmed, the low-pass filter intensity is adaptively adjusted, thereby reducing the risk of false disaster signals being transmitted to the upper core algorithm due to the failure of the terminal instrument.

[0012] According to the online monitoring method for abnormal gas pipeline pressure provided by the present invention, the filtering process includes: for a signal stagnation state, marking the monitoring node as a sensor fault, suspending subsequent abnormal detection operations and generating a sensor maintenance prompt; for a rapid drift state, temporarily increasing the filtering coefficient; for a normal state, correcting the filtering coefficient according to a piecewise linear rule based on the interval comparison relationship between the signal variance and the historical reference variance; for a transition state, maintaining the current filtering coefficient; and using the corrected or maintained filtering coefficient to perform a digital first-order low-pass filter on the original pressure data to obtain a pressure filter value.

[0013] According to the online monitoring method for abnormal gas pipeline pressure provided by the present invention, the construction of the dynamic expected baseline corresponding to the monitoring node at the current moment includes: selecting a set of benchmark working days that meet the normal operating conditions from the historical database; extracting historical pressure filtering values ​​of the same period under multiple characteristic cycles including daily cycle, half-day cycle and weekly cycle from the set of benchmark working days, and calculating the cycle baseline corresponding to each characteristic cycle, determining the prediction residual variance of each cycle baseline in the historical data; determining the corresponding fusion weight by the proportion of the inverse of each prediction residual variance in the sum of all inverses, and weighting and superimposing each cycle baseline according to the corresponding fusion weight to synthesize the dynamic expected baseline.

[0014] This invention divides historical data into multiple specific frequency bands such as daily cycles, half-day cycles, and weekly cycles, and uses the prediction residuals for inverse weighting. Compared with simply taking the arithmetic average of the previous few days, this method restores the complex and real tidal consumption rhythms of urban residents during morning and evening cooking peaks, midday secondary peaks, and differences in gas consumption between weekdays and holidays. This allows the synthesized dynamic baseline to be attached to the legitimate gas consumption fluctuation envelope, thereby effectively reducing false alarms during special periods.

[0015] According to the online monitoring method for abnormal pressure in a gas pipeline network provided by the present invention, the dynamic pressure deviation between adjacent monitoring nodes is calculated, and the corresponding dynamic hydraulic impedance coefficient of the pipeline segment is determined in combination with the flow characteristics of the pipeline segment. The method includes: dividing the physical length of the pipeline segment between two adjacent monitoring nodes by the propagation speed of the pressure wave in the gas medium to obtain the pressure propagation delay; using the pressure propagation delay to perform time-shift correction on the pressure filter value of the downstream monitoring node to eliminate the phase difference; subtracting the pressure filter value of the downstream monitoring node after time-shift correction from the current pressure filter value of the upstream monitoring node to obtain the dynamic pressure deviation; obtaining the real-time flow meter reading of the corresponding pipeline segment, and dividing the dynamic pressure deviation at the current moment by the square of the real-time flow meter reading to obtain the dynamic hydraulic impedance coefficient of the pipeline segment.

[0016] This invention introduces Darcy's law of fluid dynamics and the acoustic propagation speed of pressure waves for spatial time delay compensation. Compared with the huge systematic phase error caused by directly subtracting asynchronous transient pressures in pipe networks with large node distances, this solution accurately aligns upstream and downstream data of the same wavefront on the time axis, restores the equivalent resistance coefficient that reflects the aging and cracking state of the pipe material, and makes hydraulic impedance a core independent evidence for judging the physical integrity of the pipe section.

[0017] The online monitoring method for abnormal gas pipeline pressure provided by this invention performs multi-dimensional feature fusion judgment, including: subtracting the historical baseline deviation from the current dynamic pressure deviation to obtain the abnormal deviation increment; taking the daily cycle as the recent history, determining the deviation threshold based on the product of the standard deviation of the dynamic pressure deviation sequence of the recent historical data and a preset confidence coefficient; running a first-scale window and a second-scale window with different window depth configurations in parallel, and accumulating the counters in the two windows respectively when the abnormal deviation increment is greater than the deviation threshold; triggering a preliminary alarm when the ratio of the counter value in any window to the window depth reaches the corresponding proportional threshold.

[0018] According to the online monitoring method for abnormal gas pipeline pressure provided by the present invention, after triggering the preliminary alarm, the method further includes: expanding the topological neighbors to obtain the set of associated pipeline segments of the current pipeline segment; calculating the association anomaly index of each associated pipeline segment within a preset time window, wherein the association anomaly index is the proportion of the number of sampling points whose abnormal deviation increment exceeds the corresponding deviation threshold to the total number of sampling points in the window; if at least one associated pipeline segment in the set of associated pipeline segments has an association anomaly index exceeding the preset association judgment threshold, then a real pipeline physical fault alarm is confirmed to be triggered.

[0019] According to the online monitoring method for abnormal gas pipeline pressure provided by the present invention, after confirming the triggering of a real pipeline physical fault alarm, the method further includes: pre-establishing a leakage hydraulic fingerprint database containing standard template sequences of leakage location, orifice size, and dynamic pressure deviation; using a dynamic time warping algorithm to calculate the similarity distance between the dynamic pressure deviation sequence of the current abnormal period and each standard template sequence in the leakage hydraulic fingerprint database; selecting a preset number of templates with the smallest distance, and outputting the leakage location interval and leakage magnitude estimate by weighting the distance inversely.

[0020] According to the online monitoring method for abnormal pressure in gas pipeline networks provided by the present invention, the output of abnormal detection results includes: when the deviation of the pressure filter value of a single monitoring node from the dynamic expected baseline exceeds the single-point alarm threshold, and the abnormal deviation increment between the monitoring node and the adjacent monitoring node does not exceed the deviation threshold, and the dynamic hydraulic impedance coefficient of the corresponding pipe section is within the normal deviation range of the design hydraulic impedance coefficient, it is determined to be a single-point sensor fault; when the abnormal deviation increment continuously exceeds the deviation threshold, meets the conditions of multi-timescale parallel accumulation decision and spatial correlation diffusion verification, and the dynamic hydraulic impedance coefficient of the pipe section exceeds the normal deviation range, it is determined to be a real physical fault in the pipeline network.

[0021] This invention establishes a fault classification decoupling matrix. Compared with the management method of uniformly sounding alarms, this invention can clearly distinguish between sensor electronic soft faults and underground pipeline physical real damage. This allows for the generation of different types of work orders, which can optimize the scheduling of human resources in emergency situations.

[0022] Secondly, the present invention provides an online monitoring system for abnormal pressure in gas pipeline networks, employing the following technical solution: An online monitoring system for abnormal gas pipeline pressure includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned online monitoring method for abnormal gas pipeline pressure.

[0023] By adopting the above technical solution, the online monitoring method for abnormal gas pipeline pressure is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0024] The present invention has the following technical effects: Based on the above technical solutions, the online monitoring method and system for abnormal gas pipeline pressure provided by the present invention constructs a sensor drift self-diagnosis mechanism to intercept hardware distortions such as zero-point drift, sensitivity attenuation and signal stagnation at the filtering front end, and performs piecewise linear adaptive correction of the digital first-order low-pass filter coefficient according to the signal fluctuation intensity, so that the filtering bandwidth dynamically fits the real spectrum characteristics of gas consumption tides in the pipeline network, thereby reducing the pollution of subsequent abnormal detection links by sensor error data from the source. Based on this, the present invention determines the fusion weights according to multi-scale periodic components such as daily, semi-daily, and weekly cycles, and superimposes the baselines of each cycle into a dynamic expected baseline. This enables the reference boundary under normal operating conditions to adaptively track the complex tidal changes of urban gas consumption rhythm, effectively suppressing the systematic offset distortion of the single arithmetic mean baseline during peak-to-valley transition periods. Simultaneously, time-shift correction is performed on the downstream node pressure filter value to eliminate phase difference, and based on Darcy fluid dynamics constraints, the dynamic pressure deviation is coupled with the real-time flow of the pipe segment to form the dynamic hydraulic impedance coefficient of the pipe segment. This elevates the one-dimensional pressure sensor signal to a two-dimensional physical correlation feature characterizing the internal frictional resistance state of the pipe segment, truly reflecting the equivalent resistance mutation caused by the surge in flow at the leak point. Furthermore, a multi-dimensional feature fusion decision framework is constructed to decouple sensor single-point faults, legitimate transient disturbances, and actual physical leaks in the pipeline network into independent feature subspaces for separate handling. This can effectively reduce the invalid alarm rate and the risk of missed reporting of real micro-leaks caused by sensor hardware distortion and misjudgment of gas consumption fluctuations while meeting the requirements of online real-time monitoring response. This effectively improves the accuracy of gas pipeline network pressure anomaly detection results and the reliability of operation and maintenance decisions. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the online monitoring method for abnormal gas pipeline pressure provided in an embodiment of the present invention. Figure 2 This is a performance comparison diagram provided for an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0027] This invention discloses an online monitoring method for abnormal pressure in gas pipeline networks. For details, please refer to [link to relevant documentation]. Figure 1 As shown, Figure 1 This is a flowchart illustrating the online monitoring method for abnormal gas pipeline pressure provided in an embodiment of the present invention. The method specifically includes the following steps: S1: Obtain the pressure filter value of each monitoring node.

[0028] For example, the raw pressure data of each monitoring node deployed in the gas pipeline network is obtained, and the raw pressure data is subjected to sensor status self-diagnosis and filtering to obtain a pressure filter value that reflects the static pressure level.

[0029] In this embodiment of the invention, the gas pipeline network pressure anomaly detection system includes pressure transmitters deployed at key nodes of the pipeline network, analog-to-digital converters, an industrial control network communication module, a monitoring center server, and real-time and historical databases stored on the server. Optionally, the pressure transmitter is an intelligent pressure transmitter with an accuracy class of not less than 0.5 and an output signal of 4mA to 20mA standard current signal. Its range is determined according to the design operating pressure of the monitoring node. For example, for a medium-pressure pipeline monitoring point with an operating pressure of 0.3MPa, the pressure transmitter range can be selected from 0MPa to 0.6MPa. The pressure transmitter collects the pipeline network pressure analog signal at a set sampling frequency, converts it into a digital pressure signal by the analog-to-digital converter, and then uploads it to the real-time database of the monitoring center server in real time through the industrial control network. In this embodiment of the invention, the industrial control network is, for example, an industrial Ethernet; the sampling frequency is set to 1Hz, that is, one pressure sample is collected per second; the analog-to-digital converter uses 16-bit resolution to ensure pressure quantization accuracy.

[0030] For example, the sensor state self-diagnosis of the raw pressure data includes: calculating the signal variance of the raw pressure data within a local time window, and using the proportion of consecutive sampling points with the same sign in the first-order difference sequence to the total number of data points in the window as the symbol consistency index; if the signal variance is lower than a preset noise floor threshold and the symbol consistency index is higher than a preset stagnation threshold, it is determined to be a signal stagnation state and subsequent calculations are blocked; if the signal variance is higher than a preset drift threshold and the symbol consistency index is higher than a preset drift threshold, it is determined to be a rapid drift state; if the signal variance is between the preset noise floor threshold and the preset drift threshold and the symbol consistency index is lower than or equal to the preset drift threshold, it is determined to be a normal state; otherwise, it is determined to be a transitional state.

[0031] It is understood that the "otherwise state" refers to the situation where the signal variance is lower than the preset noise floor threshold and the symbol consistency index is lower than or equal to the preset slack threshold, or the signal variance is between the preset noise floor threshold and the preset drift threshold and the symbol consistency index is between the preset slack threshold and the preset drift threshold. In this case, it can be determined as a transition state.

[0032] For example, during filtering, for a signal stagnation state, the monitoring node is marked as a sensor fault, subsequent anomaly detection operations are paused, and a sensor maintenance prompt is generated; for a rapid drift state, the filter coefficient is temporarily increased; for a normal state, the filter coefficient is corrected according to the interval comparison relationship between the signal variance and the historical reference variance, following a piecewise linear rule; for a transition state, the current filter coefficient is maintained; the original pressure data is then subjected to a digital first-order low-pass filter using the corrected or maintained filter coefficient to obtain the pressure filter value.

[0033] Specifically, in the data preprocessing stage, the monitoring center server reads the raw pressure data from the real-time database through a polling mechanism and uses a first-order digital low-pass filter algorithm to perform noise reduction preprocessing on the raw pressure data. It should be noted that the first-order low-pass filter algorithm is a digital signal processing method based on recursive weighted averaging. It filters out high-frequency random noise and retains the low-frequency pressure evolution trend by weightedly fusing the current pressure measurement value with the filtered output value from the previous time step. The filter coefficients of the first-order low-pass filter algorithm are shown in the figure. Based on sampling frequency The calculation relationship is determined by the expected cutoff frequency, and the formula is as follows: .

[0034] It should be noted that this formula is an engineering approximation of the first-order low-pass filter coefficients under the condition that the sampling frequency is much greater than the cutoff frequency, and it belongs to the existing commonly used expressions in the field of digital signal processing. Among them, The desired cutoff frequency, physically defined as the highest allowed signal frequency component, is used. In this embodiment of the invention, for high-frequency noise generated by the instantaneous opening and closing of user-side valves in the gas pipeline pressure signal, the desired cutoff frequency is, for example, taken as 0.05Hz; combined with... =1Hz, the calculated filter coefficient is 0.314. Furthermore, the initial filter value can be the arithmetic mean of the original pressure measurements at the first N moments after the pressure transmitter is powered on, where N is typically 10.

[0035] Based on this The pressure filter value at any given time can be calculated recursively according to the following formula: , express Pressure filter value at any time, Represents the filter coefficients. express The raw pressure data obtained from constant measurements express Pressure filter value at any given time.

[0036] By using the above settings, random disturbances with frequencies higher than 0.05Hz can be effectively filtered out, and a smooth sequence reflecting the static pressure level of the pipe section can be obtained.

[0037] For example, before performing first-order low-pass filtering, a sensor drift self-diagnostic operation can be performed on the raw pressure data. This can be understood as automatically determining whether the pressure transmitter exhibits abnormal conditions such as zero-point drift, sensitivity decay, or signal stagnation based on the statistical characteristics of the raw pressure data within a local time window. Specifically, a length of... Local time window, For example, a time window of 120 seconds is used, with each sampling point corresponding to one time moment, resulting in 120 sampling points. Calculate the signal variance of the original pressure data within this local time window, as well as the sign consistency index of the first-order difference sequence of the original pressure data.

[0038] It should be noted that the sign consistency index refers to the proportion of consecutive sampling points with the same sign in the first-order difference sequence to the total number of data points in the window, and its value ranges from 0 to 1. When the sign consistency index approaches 1, it indicates that the original pressure data shows a unidirectional monotonic change trend; when the sign consistency index approaches 0.5, it indicates that the original pressure data shows random fluctuation characteristics.

[0039] The sensor drift self-diagnosis judgment rules are as follows: If the signal variance is lower than the preset noise floor threshold and the sign consistency index is higher than the preset jamming judgment threshold, the pressure transmitter is judged to be in a signal jamming state. At this time, the monitoring node is marked as a sensor fault, subsequent abnormal detection operations are suspended, and a sensor maintenance prompt is generated. If the signal variance is higher than the preset drift judgment upper limit and the sign consistency index is higher than the preset drift judgment threshold, the pressure transmitter is judged to be in a rapid drift state. At this time, the filter coefficient can be temporarily increased to [value missing]. , For example, a value of 0.5 is used to accelerate the tracking of actual pressure trend changes and generate sensor calibration prompts. If both the signal variance and the sign consistency index are within the normal range, the pressure transmitter is determined to be in normal condition. In this embodiment of the invention, the noise floor threshold is for example 0.0001 MPa², the jamming judgment threshold is for example 0.95, the drift judgment upper limit is for example 0.01 MPa², and the drift judgment threshold is for example 0.85.

[0040] Furthermore, for pressure transmitters determined to be in a normal state, the filter coefficients are adaptively corrected based on the signal variance to obtain a filter strength that matches the current signal fluctuation characteristics. The adaptively corrected filter coefficients... Determined according to the piecewise linearity rule: when the signal variance is less than or equal to the reference variance... When the signal variance is greater than the reference variance and less than or equal to twice the reference variance, When the signal variance is greater than twice the reference variance, The reference variance is typically taken as the median of the signal variance during historical normal operation periods. In this embodiment of the invention, the filter coefficients... The initial value is approximately 0.314, therefore the adaptively corrected filter coefficients... It varies in segments between 0.251 and 0.408.

[0041] Through the above adaptive correction, when the pipeline network is in a low gas consumption period at night and the signal fluctuation is small, the filter coefficient can be automatically reduced to retain pressure details; when the pipeline network is in a high gas consumption period and the signal fluctuation is severe, the filter intensity can be appropriately increased to suppress noise interference, thus achieving good filtering effect under different operating conditions.

[0042] S2: Calculate the dynamic hydraulic resistance coefficient of the pipe section.

[0043] For example, multi-cycle baseline fusion is performed based on the synchronous pressure filter value under historical normal operating conditions to construct the dynamic expected baseline of the monitoring node at the current moment; the dynamic pressure deviation between adjacent monitoring nodes is calculated by combining the topological connection relationship of the gas pipeline network and the pressure wave propagation delay, and the corresponding dynamic hydraulic impedance coefficient of the pipeline segment is determined by combining the flow characteristics of the pipeline segment.

[0044] It should be noted that the dynamic expected baseline refers to a reference value synthesized based on historical normal operation data at the current analysis time, used to characterize the normal fluctuation range of pressure at that time.

[0045] For example, constructing the dynamic expected baseline for the monitoring node at the current moment includes: selecting a set of benchmark working days that meet normal operating conditions from the historical database; extracting historical pressure filter values ​​for multiple characteristic cycles, including daily, half-day, and weekly cycles, and calculating the cycle baselines corresponding to each characteristic cycle; determining the prediction residual variance of each cycle baseline in the historical data; determining the corresponding fusion weight based on the proportion of the inverse of each prediction residual variance in the sum of all inverses; and weighting and superimposing each cycle baseline according to the corresponding fusion weight to synthesize the dynamic expected baseline.

[0046] In this embodiment of the invention, when filtering historical data for the same period, data from the past 30 daily cycles can be retrieved from the historical database. For each retrieved historical day, it is determined whether it is a normal operating day. The determination criteria include: the pressure fluctuation range throughout the day does not exceed ±15% of the historical average pressure of the monitoring point, and no historical pressure anomaly alarms, pipeline maintenance, or equipment maintenance events with manual markings have been triggered on that day. Only dates that meet all of the above conditions are included in the baseline working day set. Further, the daily cycle is 24 hours, or 86,400 seconds. When analyzing the pressure status at the current moment, historical sampling points that are precisely aligned with the current moment in terms of hour, minute, and second are retrieved from the baseline working day set, and their pressure filter values ​​are extracted.

[0047] In this embodiment of the invention, a dynamic expected baseline can be constructed using data from the previous three benchmark working days. An example illustrating the process of obtaining the three benchmark working days is as follows: If the current time is... At any time, one can obtain Pressure filter values ​​at the same time in the previous 1, 2, and 3 daily cycles.

[0048] Understandably, when the number of available baseline working days in the historical database is less than three, linear interpolation can be used to supplement the missing data, or regression prediction can be performed based on the historical gas load curve of the monitoring point to generate a virtual baseline value, thus ensuring the continuity of baseline calculation. Furthermore, a baseline rolling update operation can be automatically performed every morning to incorporate newly completed operating day data into the historical database and re-select baseline working days, thereby enabling the dynamic expected baseline to adaptively track the slow changes in pipeline network operating characteristics.

[0049] Furthermore, before constructing the dynamic expected baseline, a multi-period baseline fusion operation can be performed on the historical pressure filter values. It should be noted that multi-period baseline fusion refers to the separation, extraction, and weighted fusion of the periodic fluctuation features at different time scales contained in the historical pressure filter values ​​to construct a high-precision expected baseline. Specifically, characteristic periods existing in the gas pipeline network pressure time series are identified, including at least a 24-hour daily cycle, a 12-hour semi-daily cycle, and a 168-hour weekly cycle. For each characteristic period, a corresponding periodic baseline is constructed. Taking the daily cycle as an example... The baseline of the period at any given time is calculated using the following formula: ; in, , , They represent The pressure filter values ​​at the same time in the previous 1, 2, and 3 daily cycles. Correspondingly, the semi-daily cycle baseline and the weekly cycle baseline are calculated using the semi-daily cycle and the weekly cycle, respectively, in the same construction method as the daily cycle baseline.

[0050] For example, calculate the variance of the predicted residuals for each period's baseline in the most recent 30 days of historical data, denoted as follows: , , It should be noted that the prediction residual variance refers to the average of the squares of the differences between the historical actual pressure filtered value and the predicted value when using the baseline of that period as the prediction value. The smaller the prediction residual variance, the stronger the explanatory power of that baseline for the current monitoring point's pressure fluctuations.

[0051] When the variance of the prediction residuals for daily, semi-daily, or weekly cycles is not zero, the fusion weight of the baseline for each cycle can be determined based on the reciprocal of the variance of the prediction residuals. The fusion weight can be calculated using the following formula: ; in, Indicates the first The fusion weights corresponding to the periodic baseline, Taking 1, 2, and 3, they correspond to the daily cycle, the half-day cycle, and the weekly cycle, respectively. Indicates the first The variance of the predicted residuals for each periodic baseline.

[0052] When at least one of the variances of the prediction residuals for the daily, semi-daily, or weekly cycles is 0, the fusion weights can be determined using the following three scenarios: In one possible scenario, only one periodic baseline has a prediction residual variance of 0. In this case, the fusion weight of this periodic baseline can be set to twice the sum of the fusion weights of the other two periodic baselines. The remaining weights of the other two periodic baselines are allocated according to the proportion of the reciprocal of their prediction residual variances in the original two.

[0053] Understandably, a variance of 0 indicates that the prediction accuracy has reached its theoretical optimum, and its weight should be significantly higher than that of other periods with finite variances. For example: suppose the predicted residual variance of the daily cycle baseline... The variance of the predicted residuals of the half-day cycle baseline The variance of the predicted residuals of the weekly cycle baseline .because First, calculate the ratio of the reciprocal of the half-day cycle to the weekly cycle: , The sum of the two is 1250. The weight ratio of the half-day cycle to the two is 1000 / 1250=0.8, and that of the weekly cycle is 250 / 1250=0.2. The weight of the daily cycle baseline is set to twice the sum of the weights of the other two, that is, 2×(0.8+0.2)=2.0. After normalizing the three, the total weight is 2.0+0.8+0.2=3.0. The final merged weights are 2.0 / 3.0=0.667 for the daily cycle, 0.8 / 3.0=0.267 for the half-day cycle, and 0.2 / 3.0=0.067 for the weekly cycle.

[0054] In another possible scenario, there are two periodic baselines with zero predicted residual variance. In this case, the fusion weights of these two baselines can be set to be equal, each set to twice the weight of the remaining non-zero variance periodic baseline. After normalizing the three, the final fusion weights are obtained.

[0055] In another possible scenario, the predicted residual variance of three periodic baselines is 0. In this case, the fusion weights of the three baselines can be set to be equal, each taking 1 / 3.

[0056] The dynamic expected baseline is obtained by weighting and superimposing the baselines of each period according to the fusion weight. Through the above multi-period baseline fusion operation, the dynamic expected baseline can adaptively capture the multi-scale periodic fluctuation characteristics of pipeline pressure, such as the transition between intraday peaks and troughs, midday secondary peaks, and differences between weekdays and weekends. Compared with the single historical arithmetic mean method, it has higher baseline accuracy.

[0057] For example, after constructing the dynamic expected baseline, the method also includes: taking the daily cycle as the recent history, calculating the standard deviation of the predicted residuals of the dynamic expected baseline in the recent historical data, and using it as the baseline confidence index; when the baseline confidence index exceeds the preset baseline unreliability threshold, automatically expanding the retrieval time range of the benchmark working day to re-execute multi-cycle baseline fusion, and simultaneously reducing the weight coefficient of the corresponding monitoring node in the subsequent spatial correlation analysis.

[0058] It should be noted that the baseline confidence index is a statistic used to assess the reliability of the dynamic expected baseline. A smaller value indicates higher consistency of historical data from the same period, and a more reliable baseline. In this embodiment of the invention, if the baseline confidence index exceeds a preset baseline unreliability threshold, it is determined that the current dynamic expected baseline is contaminated by historical abnormal data or that the pipeline network's operating characteristics have changed significantly. In this case, the baseline working day retrieval range is automatically expanded to the past 60 days, and multi-cycle baseline fusion is re-executed. Simultaneously, the weight coefficient of this monitoring node in subsequent spatial correlation analysis is reduced to avoid interference from unreliable baselines on the overall detection results. For example, the baseline unreliability threshold is set to 0.02 MPa, but this can be adjusted according to actual needs.

[0059] After synthesizing the dynamic expected baseline based on the above steps, the topological connection relationship of the gas pipeline network can be read, and the pressure filter values ​​of spatially adjacent monitoring nodes can be extracted simultaneously to calculate the dynamic pressure deviation. It should be noted that spatially adjacent monitoring nodes refer to a pair of upstream and downstream monitoring nodes in the physical topology of the gas pipeline network, directly connected by a single pipe segment with a segment length not exceeding 5 km. In this embodiment of the invention, the pipeline network topological connection relationship is stored in the geographic information system database in the form of a pipeline connection matrix. The pipeline connection matrix records at least the pipe segment number, upstream node number, downstream node number, pipe segment length, pipe diameter, and pipe material. By querying the pipeline connection matrix, adjacent monitoring node pairs that are upstream and downstream of the currently analyzed monitoring node are identified. The determination of the upstream and downstream relationship is based on the medium flow direction determined by the pipeline network hydraulic calculation model. Optionally, in pipeline networks where a hydraulic calculation model has not yet been established, auxiliary determination can be made based on node elevation data in the geographic information system, combined with the pressure gradient direction presented by historical pressure monitoring data.

[0060] For example, calculating the dynamic pressure deviation between adjacent monitoring nodes and determining the corresponding dynamic hydraulic impedance coefficient of the pipe segment based on the pipe segment flow characteristics includes: dividing the physical length of the pipe segment between two adjacent monitoring nodes by the propagation speed of the pressure wave in the gas medium to obtain the pressure propagation delay; using the pressure propagation delay to perform time-shift correction on the pressure filter value of the downstream monitoring node to eliminate the phase difference; subtracting the time-shift corrected pressure filter value of the downstream monitoring node from the current pressure filter value of the upstream monitoring node to obtain the dynamic pressure deviation; obtaining the real-time flowmeter reading of the corresponding pipe segment, and dividing the dynamic pressure deviation at the current moment by the square of the real-time flowmeter reading to obtain the dynamic hydraulic impedance coefficient of the pipe segment.

[0061] Furthermore, considering the limited propagation speed of pressure waves in gas pipelines, when the pipe length between adjacent monitoring nodes is large, the pressure values ​​collected at the same moment do not strictly correspond to the same hydraulic transient process, and direct subtraction will introduce systematic spatial errors. Therefore, in this embodiment of the invention, a pressure propagation delay compensation mechanism can be introduced before calculating the pressure difference between nodes. It can be understood that the pressure propagation delay compensation mechanism refers to performing time-shift correction on the pressure filter value of the downstream monitoring node based on the propagation speed of the pressure wave in the pipe medium and the pipe length, in order to eliminate the phase difference caused by the pressure wave propagation delay. Specifically, the propagation speed of the pressure wave in the gas... For example, the speed is taken as 350 m / s, for a pipe section length of... Adjacent monitoring node pairs, pressure propagation delay According to Calculated.

[0062] Using the network time protocol clock of the monitoring center server as a reference, the pressure filtering values ​​of upstream node A and downstream node B are timestamped, with an alignment accuracy better than 100 milliseconds. In the calculation... When calculating the dynamic pressure deviation at a given time, the pressure filter value of downstream node B uses the interpolated result after time delay compensation. Optionally, the calculation formula for the dynamic pressure deviation is: ; in, express Dynamic pressure deviation at any given time express The pressure filter value of upstream node A at time 1. This represents the pressure filtering value of downstream node B after time delay compensation.

[0063] Through the above spatial topology constraints and time delay compensation processing, the one-dimensional pressure sensor signal is converted into a two-dimensional correlation feature that can reflect the flow resistance state of the physical medium in the pipe section.

[0064] Furthermore, after calculating the dynamic pressure deviation, this embodiment of the invention can perform online identification of the hydraulic impedance characteristics of the pipe segment. It should be noted that online identification of the hydraulic impedance characteristics of the pipe segment refers to the operation of calculating the dynamic hydraulic impedance coefficient of the pipe segment online based on the Darcy-Weisbach formula in fluid mechanics, utilizing the real-time correspondence between the pressure difference at both ends of the pipe segment and the flow rate, to characterize the internal frictional resistance state of the pipe segment. Specifically, the real-time flow rate of the pipe segment is read from a gas flow meter deployed upstream or downstream of the pipe segment. Dynamic hydraulic resistance coefficient of the pipe section at any given time It can be calculated according to the following formula: ;in, express Dynamic pressure deviation at any given time express The flow rate of the pipe section at any given time.

[0065] Under normal operating conditions, the dynamic hydraulic impedance coefficient of a pipe section should remain stable within ±20% of its design hydraulic impedance coefficient. When a pipe section ruptures and leaks, the instantaneous surge in flow at the leak point leads to a significant increase in the pressure difference between the two ends of the pipe section. However, the flow meter reading fails to reflect the leakage increment synchronously, resulting in an abnormal rise in the dynamic hydraulic impedance coefficient of the pipe section. Continuous monitoring of the deviation of the dynamic hydraulic impedance coefficient of the pipe section is used as an auxiliary physical criterion for dynamic pressure deviation and participates in subsequent multi-dimensional anomaly fusion judgment.

[0066] S3: Output the anomaly detection results.

[0067] For example, based on the degree of deviation of the pressure filter value from the dynamic expected baseline, the extent of dynamic pressure deviation exceeding the standard, and the variation characteristics of the dynamic hydraulic impedance coefficient of the pipe section, a multi-dimensional feature fusion decision is performed to output the anomaly detection result.

[0068] For example, performing multi-dimensional feature fusion decision includes: subtracting the historical baseline deviation from the current dynamic pressure deviation to obtain the abnormal deviation increment; determining the deviation threshold based on the product of the standard deviation of the recent historical dynamic pressure deviation sequence and a preset confidence coefficient; running a first-scale window and a second-scale window with different window depth configurations in parallel, and accumulating the counters in the two windows respectively when the abnormal deviation increment is greater than the deviation threshold; triggering a preliminary alarm when the ratio of the counter value in any window to the window depth reaches the corresponding proportional threshold.

[0069] The historical baseline deviation refers to the statistical average of the dynamic pressure deviation between adjacent monitoring node pairs under historical normal operating conditions. In this embodiment of the invention, the historical baseline deviation is calculated by taking the dynamic pressure deviation sequence of all normal operating days within the past 30 days, summing the dynamic pressure deviations of all sampling points in the sequence, and then dividing the sum by the total number of sampling points. The quotient obtained is the historical baseline deviation.

[0070] For example, the preset confidence coefficient can be set to 3, corresponding to the 99.7% confidence interval under the normal distribution assumption, that is, the deviation threshold is 3 times the standard deviation of the historical normal deviation. Optionally, for pipeline areas with more drastic pressure fluctuations, it can be appropriately reduced to 2.5 to balance detection sensitivity and false alarm rate.

[0071] In this embodiment of the invention, the sliding window cumulative decision mechanism adopts a multi-time-scale sliding window parallel cumulative decision architecture. The multi-time-scale sliding window parallel cumulative decision architecture refers to running multiple alarm decision ring queues with different time scales at the same time. Through parallel accumulation and logical fusion, a decision mechanism that unifies fast response and robust confirmation is achieved.

[0072] Specifically, at least a first-scale window and a second-scale window should be configured. The first-scale window is a short window with a depth of [missing information]. The example window is set to 10, corresponding to 10 seconds, to capture sudden, rapid pressure anomalies, such as pipe bursts or large ruptures; the second-scale window is a long window with a depth of... The example uses 300, corresponding to 300 seconds, to exclude slow drift and transient disturbances, such as minor leaks or gentle valve adjustments. The first and second scale windows operate the counters independently. and The counting rule is as follows: In each sampling period, if the abnormal deviation increment > deviation threshold, then and Increment by 1 for each sample; if the abnormal deviation increment is less than or equal to the deviation threshold, the corresponding counter is cleared only when the abnormal deviation increment is less than or equal to the deviation threshold for K consecutive sampling times. K is typically set to 5. The proportional threshold for the first scale window. For example, we take 0.8 as the scale threshold for the second scale window. The example value is 0.7.

[0073] Optionally, the initial alarm triggering condition can be: or With the above settings, a rapid response can be achieved within 10 seconds for sudden and severe leaks; for slow and minor leaks, robust accumulation over a long window avoids false alarms caused by transient disturbances.

[0074] After triggering the initial alarm based on the above steps, the spatial correlation diffusion verification operation is performed. The spatial correlation diffusion verification refers to the automatic retrieval of related pipe segments that are directly connected to the target pipe segment in the topology or separated by one pipe segment when the target pipe segment meets the initial alarm conditions. The operation checks whether the related pipe segments synchronously show abnormal pressure trends in order to eliminate local transient disturbances and confirm the pipeline network-level physical fault.

[0075] For example, after triggering the initial alarm, the process also includes: expanding the topology neighbors to obtain the set of associated pipe segments for the current pipe segment; calculating the association anomaly index of each associated pipe segment within a preset time window, where the association anomaly index is the proportion of the number of sampling points whose abnormal deviation increment exceeds the corresponding deviation threshold to the total number of sampling points in the window; if at least one associated pipe segment in the set of associated pipe segments has an association anomaly index exceeding the preset association judgment threshold, then a real pipeline physical fault alarm is confirmed to be triggered.

[0076] Specifically, taking the target pipe segment that triggered the initial alarm as the center, a layer of topological neighbors is expanded outward to obtain the set of associated pipe segments. For each associated pipe segment, its nearest neighbor is extracted. The abnormal deviation increment sequence within each sampling time is analyzed, and the correlation anomaly index is calculated. It should be noted that the correlation anomaly index refers to the proportion of sampling points where the abnormal deviation increment of the associated pipe segment exceeds its own deviation threshold to the total number of sampling points in the window. If the target pipe segment meets the preliminary alarm conditions, and at least one associated pipe segment in its associated pipe segment set has a correlation anomaly index exceeding the preset correlation judgment threshold, then a real pipeline physical fault alarm is confirmed. For example, the correlation judgment threshold is set to 0.5. Through spatial correlation diffusion verification, it is possible to effectively distinguish between real pipeline network leakage and local transient disturbances, thereby significantly reducing the false alarm rate. Real pipeline network leakage pressure anomalies propagate along the topology to adjacent pipe segments, while the impact of local transient disturbances is usually limited to a single pipe segment or a single node.

[0077] Furthermore, after confirming the triggering of a real pipeline physical fault alarm, the leak location and magnitude estimation operation based on hydraulic fingerprint matching is performed. Here, hydraulic fingerprint refers to a feature vector template established through pipeline hydraulic simulation or historical leak case statistics, which characterizes the time-domain evolution of dynamic pressure deviation under different combinations of leak location and leak aperture.

[0078] For example, after confirming the triggering of a real pipeline physical fault alarm, the process also includes: pre-establishing a leakage hydraulic fingerprint database containing standard template sequences of leakage location, orifice size, and dynamic pressure deviation; using a dynamic time warping algorithm to calculate the similarity distance between the dynamic pressure deviation sequence of the current abnormal period and each standard template sequence in the leakage hydraulic fingerprint database; selecting a preset number of templates with the smallest distance, such as 3; and outputting the leakage location range and leakage magnitude estimates by weighting the distance inversely.

[0079] In this embodiment of the invention, a hydraulic fingerprint database for leaks can be pre-constructed. Each hydraulic fingerprint entry in the database includes at least the leak location coordinates, leak orifice diameter level, dynamic pressure deviation standard template sequence, and pressure drop rate characteristic value. The dynamic pressure deviation sequence of the current abnormal period is matched with each template in the hydraulic fingerprint database based on similarity. The similarity measurement employs a dynamic time warping algorithm, which uses dynamic programming to find the optimal bending path between two sequences, thereby eliminating local scaling differences on the time axis. The dynamic time warping distance between the current dynamic pressure deviation sequence and each hydraulic fingerprint template is calculated. The three templates with the smallest distances are selected as candidate matching results, and the most likely leak location range and estimated leak magnitude are output using a weighted average of the inverse distance. The estimated leak magnitude is presented in the form of equivalent orifice diameter or estimated leak flow rate. For example, it is divided into three levels: micro-leakage (<5mm equivalent orifice diameter), medium-leakage (5mm to 20mm equivalent orifice diameter), and large-leakage (>20mm equivalent orifice diameter). The leak location and magnitude estimation results are written into the alarm message and pushed to the dispatch center monitoring screen to provide accurate decision-making basis for on-site repair personnel.

[0080] Furthermore, a multi-dimensional feature fusion mechanism is used to distinguish between single-point sensor faults and actual pipeline leaks. It should be noted that the multi-dimensional feature fusion mechanism refers to simultaneously examining the single-point pressure deviation characteristics of monitoring nodes, the pressure gradient correlation characteristics between adjacent nodes, and the hydraulic impedance deviation characteristics of pipe sections, in order to differentiate between local sensor anomalies and actual physical leaks.

[0081] For example, the output of anomaly detection results includes: when the deviation of the pressure filter value of a single monitoring node from the dynamic expected baseline exceeds the single-point alarm threshold, and the abnormal deviation increment between the monitoring node and the adjacent monitoring node does not exceed the deviation threshold, and the dynamic hydraulic impedance coefficient of the corresponding pipe section is within the normal deviation range of the design hydraulic impedance coefficient, it is determined to be a single-point sensor fault; when the abnormal deviation increment continuously exceeds the deviation threshold, meets the conditions of multi-timescale parallel accumulation decision and spatial correlation diffusion verification, and the dynamic hydraulic impedance coefficient of the pipe section exceeds the normal deviation range, it is determined to be a real pipeline physical fault.

[0082] A single-point sensor failure refers to a distortion in pressure readings caused by a hardware malfunction or abnormal signal transmission in a single pressure transmitter. Its typical characteristics include an abnormal deviation of the pressure value at the faulty node from the baseline, but no substantial change in the pressure gradient between adjacent monitoring nodes and the dynamic hydraulic impedance coefficient of the pipe section remaining within the normal range. In this embodiment of the invention, the single-point alarm threshold is set to ±10% of the dynamic expected baseline.

[0083] When the deviation of the pressure filter value of a monitoring node from its dynamic expected baseline exceeds the single-point alarm threshold, but the abnormal deviation increment between this node and all its adjacent monitoring nodes does not exceed the deviation threshold, and the dynamic hydraulic impedance coefficient of the pipe section is within ±20% of the design hydraulic impedance coefficient, it can be determined as a single-point sensor fault. Only a sensor maintenance prompt will be generated without triggering a pipe network leak alarm. Conversely, when the abnormal deviation increment continuously exceeds the standard, meets the multi-timescale sliding window parallel accumulation judgment condition, passes spatial correlation diffusion verification, and the dynamic hydraulic impedance coefficient of the pipe section exceeds the normal range, it can be determined as a real pipe network physical fault. An alarm message containing the pipe section number, abnormal node number, abnormal deviation value, leak location estimation, leak level, and timestamp will be generated and pushed to the dispatch center monitoring screen in JSON data format via Hypertext Transfer Protocol.

[0084] Optionally, the alarm message may also include an alarm level field, where the alarm level is incremented based on the abnormal deviation. Relative to the deviation threshold The number of times the standard is exceeded is used to divide the data. For example, when... ≤1.5 This is a general alarm, when 1.5 < ≤2.5 This is an important warning, when >2.5 This is an emergency alarm.

[0085] See also Figure 2 As shown, Figure 2 This is a performance comparison diagram provided for an embodiment of the present invention.

[0086] As can be seen from the data in the figure, the leakage detection rate of the single threshold method is 62%, and the false alarm rate is as high as 28%; the leakage detection rate of the simple filtering method is increased to 75%, and the false alarm rate is reduced to 18%; while the leakage detection rate of the method of the present invention reaches the highest of 96%, and the false alarm rate is significantly reduced to only 4%.

[0087] To improve leak detection rates, this invention extracts daily, semi-daily, and weekly cycle features from historical data and synthesizes a high-precision dynamic expected baseline through weighted superposition. Simultaneously, it utilizes pressure propagation delay to perform time-shift correction on downstream monitoring nodes to eliminate phase differences, thereby obtaining more accurate dynamic pressure deviations. Furthermore, it calculates the dynamic hydraulic impedance coefficient of the pipe section in real time. When a rupture or leak causes a surge in pressure difference across the pipe section, the hydraulic impedance coefficient exhibits an abnormal increase. This physical characteristic serves as an auxiliary criterion in multi-dimensional decision-making, significantly enhancing the ability to identify genuine physical faults.

[0088] To reduce false alarm rates, this invention effectively determines whether a sensor is experiencing signal stagnation or rapid drift by calculating signal variance and sign consistency index during the data preprocessing stage. When a single-point sensor fault is identified, only a maintenance prompt is generated without triggering a pipeline leak alarm, thus directly eliminating false alarms caused by sensor malfunctions. Furthermore, by using parallel execution of short-cycle and long-cycle dual-scale windows for cumulative decision-making, it can quickly respond to sudden severe leaks while avoiding false alarms caused by local transient disturbances such as instantaneous opening and closing of user-side valves through robust accumulation over long windows. Finally, after triggering an initial alarm, spatial correlation diffusion verification is performed, expanding the topological neighbors to obtain a set of associated pipe segments and calculating the correlation anomaly index, which can effectively distinguish between real pipeline leaks and local transient disturbances, thereby significantly reducing the false alarm rate.

[0089] This invention also discloses an online monitoring system for abnormal gas pipeline pressure, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the online monitoring method for abnormal gas pipeline pressure provided by this invention.

[0090] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0091] In this invention, the aforementioned memory can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0092] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method of on-line monitoring of abnormal pressure in a gas network, characterized in that, include: S1, acquire the raw pressure data of each monitoring node deployed in the gas pipeline network, perform sensor status self-diagnosis and filtering on the raw pressure data, and obtain the pressure filter value reflecting the static pressure level; S2, based on the synchronous pressure filter values ​​under historical normal operating conditions, performs multi-cycle baseline fusion to construct the dynamic expected baseline for the monitoring node at the current moment, including: selecting a set of benchmark working days that meet the normal operating conditions from the historical database; extracting historical synchronous pressure filter values ​​under multiple characteristic cycles, including daily cycle, half-day cycle, and weekly cycle, from the set of benchmark working days; calculating the cycle baseline corresponding to each characteristic cycle; determining the prediction residual variance of each cycle baseline in the historical data; determining the corresponding fusion weight based on the proportion of the reciprocal of each prediction residual variance in the sum of all reciprocals; and weighting and superimposing each cycle baseline according to the corresponding fusion weight to synthesize the dynamic expected baseline. By combining the topological connections of the gas pipeline network with the pressure wave propagation delay, the dynamic pressure deviation between adjacent monitoring nodes is calculated, and the corresponding dynamic hydraulic impedance coefficient of the pipeline segment is determined by combining the flow characteristics of the pipeline segment. This includes: dividing the physical length of the pipeline segment between two adjacent monitoring nodes by the propagation speed of the pressure wave in the gas medium to obtain the pressure propagation delay; using the pressure propagation delay to perform time-shift correction on the pressure filter value of the downstream monitoring node to eliminate the phase difference; subtracting the pressure filter value of the downstream monitoring node after time-shift correction from the current pressure filter value of the upstream monitoring node to obtain the dynamic pressure deviation; obtaining the real-time flow meter reading of the corresponding pipeline segment, and dividing the dynamic pressure deviation at the current moment by the square of the real-time flow meter reading to obtain the dynamic hydraulic impedance coefficient of the pipeline segment. S3, based on the deviation of the pressure filter value from the dynamic expected baseline, the exceedance of the dynamic pressure deviation, and the variation characteristics of the dynamic hydraulic impedance coefficient of the pipe section, performs multi-dimensional feature fusion judgment, including: subtracting the historical benchmark deviation from the current dynamic pressure deviation to obtain the abnormal deviation increment; taking the daily cycle as the recent history, determining the deviation threshold based on the product of the standard deviation of the dynamic pressure deviation sequence of the recent historical data and the preset confidence coefficient; running a first-scale window and a second-scale window with different window depth configurations in parallel, and accumulating the counters in the two windows respectively when the abnormal deviation increment is greater than the deviation threshold; triggering a preliminary alarm when the ratio of the counter value in any window to the window depth reaches the corresponding proportional threshold; and outputting the abnormal detection result.

2. The online monitoring method for abnormal gas pipeline pressure according to claim 1, characterized in that, Perform sensor condition self-diagnosis on raw pressure data, including: Within a local time window, the signal variance of the original pressure data is calculated. The proportion of consecutive sampling points with the same sign in the first-order difference sequence to the total number of data points within the window is used as the symbol consistency index. If the signal variance is lower than a preset noise floor threshold and the symbol consistency index is higher than a preset stagnation threshold, it is determined to be a signal stagnation state and subsequent calculations are blocked. If the signal variance is higher than a preset drift threshold and the symbol consistency index is higher than a preset drift threshold, it is determined to be a fast drift state. If the signal variance is between the preset noise floor threshold and the preset drift threshold and the symbol consistency index is lower than or equal to the preset drift threshold, it is determined to be a normal state. Otherwise, it is determined to be a transition state.

3. The online monitoring method for abnormal gas pipeline pressure according to claim 2, characterized in that, Filtering processes include: For signal stagnation, mark the monitoring node as a sensor fault, suspend subsequent anomaly detection operations, and generate a sensor maintenance prompt; for rapid drift, temporarily increase the filter coefficient; for normal state, adjust the filter coefficient according to the interval comparison relationship between the signal variance and the historical reference variance, and follow the piecewise linear rule; for transition state, maintain the current filter coefficient; use the adjusted or maintained filter coefficient to perform a digital first-order low-pass filter on the original pressure data to obtain the pressure filter value.

4. The online monitoring method for abnormal gas pipeline pressure according to claim 1, characterized in that, After the initial alarm is triggered, the following is also included: Expand the topological neighbors to obtain the set of associated pipe segments for the current pipe segment; Calculate the correlation anomaly index of each associated pipe segment within the preset time window. The correlation anomaly index is the proportion of the number of sampling points whose abnormal deviation increment exceeds the corresponding deviation threshold to the total number of sampling points in the window. If at least one associated pipe segment in the associated pipe segment set has an associated anomaly index exceeding the preset associated judgment threshold, then a real pipeline physical fault alarm is triggered.

5. The online monitoring method for abnormal gas pipeline pressure according to claim 4, characterized in that, After confirming that a real physical fault alarm in the pipeline network has been triggered, the following steps are also included: A leak hydraulic fingerprint database containing standard template sequences of leak locations, orifice sizes, and dynamic pressure deviations is pre-established; The similarity distance between the dynamic pressure deviation sequence during the current abnormal period and the standard template sequences in the leakage hydraulic fingerprint database is calculated using the dynamic time warping algorithm. Select the preset number of templates with the smallest distance, and output the estimated values ​​of the leakage location range and leakage magnitude by weighting the values ​​according to the reciprocal of the distance.

6. The online monitoring method for abnormal gas pipeline pressure according to claim 4, characterized in that, The output anomaly detection results include: When the deviation of the pressure filter value of a single monitoring node from the dynamic expected baseline exceeds the single-point alarm threshold, and the abnormal deviation increment between the monitoring node and the adjacent monitoring node does not exceed the deviation threshold, and the dynamic hydraulic impedance coefficient of the corresponding pipe section is within the normal deviation range of the design hydraulic impedance coefficient, it is determined to be a single-point sensor failure. When the abnormal deviation increment continuously exceeds the deviation threshold, meets the conditions for parallel cumulative judgment at multiple time scales and spatial correlation diffusion verification, and the dynamic hydraulic impedance coefficient of the pipe section exceeds the normal deviation range, it is determined to be a real physical fault in the pipe network.

7. An online monitoring system for abnormal pressure in a gas pipeline network, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the online monitoring method for abnormal gas pipeline pressure according to any one of claims 1-6.

Images