A method and system for gas collection, monitoring and early warning

Abstract

This invention relates to the field of gas acquisition and monitoring technology, specifically a gas acquisition, monitoring, and early warning method and system. The method includes: acquiring gas transmission and distribution scenario information; collecting gas concentration and valve pressure values ​​in real time based on the scenario information; synchronously recording monitoring locations along each path; outputting early warning parameter combinations and trigger time periods based on the parameter overlap and correlation of monitoring locations; integrating all early warning parameter combinations and clustering them by spatial distance between monitoring locations to obtain early warning categories; statistically analyzing the number of early warnings per unit time for each category based on the trigger time period, generating an early warning trend curve, extracting key differences in early warning trend changes, and forming a difference feature set; quantifying the common patterns of the difference feature set, identifying strongly correlated parameters with early warning categories, and updating monitoring indicators based on the scenario-location dimension corresponding to each correlated parameter. This achieves accuracy and timeliness in gas early warning.

Description

Technical Field

[0001] This invention relates to the field of gas collection and monitoring technology, specifically a gas collection, monitoring, and early warning method and system. Background Technology

[0002] City gas is the mainstay of urban energy supply and the cornerstone of energy security. Gas leaks are a significant factor affecting the safe operation and supply of gas in cities. Traditional gas monitoring systems typically use fixed thresholds to trigger early warnings, without dynamically grouping and monitoring locations based on the transmission and distribution topology. This leads to inaccurate location tracking, and when issuing warnings based on multiple parameters, they easily overlook the parameter combinations present during a gas leak, resulting in reduced monitoring accuracy, delayed warnings, and passive responses.

[0003] For example, Chinese Patent Publication No. CN115424427A discloses a method, device, smart terminal, and storage medium for full-scene detection of gas leaks, including: determining a gas leak detection device based on scene information; performing leak detection on gas transmission and distribution equipment based on the gas leak detection device, and issuing a leak signal when a gas leak is detected in the gas transmission and distribution equipment; locating the gas transmission and distribution equipment based on the leak signal and a preset base map to obtain visualized leak data of the gas transmission and distribution equipment; issuing early warning information based on the visualized leak data, and determining corresponding response measures based on the visualized leak data, so as to handle the gas transmission and distribution equipment where a gas leak has occurred according to the response measures.

[0004] For example, Chinese Patent Publication No. CN115063950A discloses an intelligent monitoring method and system for gas leakage in open spaces of gas stations. Based on the gas concentration at key locations of various gas pipelines in the station and the gas concentration at outdoor collection points corresponding to the key locations of various gas pipelines, relevant electronic equipment is used for data processing and identification to determine the detection confusion index corresponding to the key locations of various gas pipelines, and finally determine the degree of importance attached to the inspection of the key locations of various gas pipelines.

[0005] Existing technologies determine the current gas detection method by classifying gas scenarios; and quantify the relationship between theoretical pressurization and gas leakage by using the confusion index and pressurization time of various gas pipeline fittings. However, when analyzing gas leaks, it is necessary to highlight the multi-parameter linkage effect caused by changes in a single parameter, and solve the problems of limited early warning accuracy, delayed early warning, and passive response by using the combination of other features. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a gas collection, monitoring and early warning method, including: S1, acquiring gas transmission and distribution scenario information, collecting gas concentration and gas pressure value at valve in real time according to the scenario information, synchronously recording the monitoring position on each path, and verifying the early warning conditions of each monitoring position.

[0007] S2, when an early warning is triggered at a monitoring location, outputs the combination of early warning parameters and the trigger time period based on the parameter overlap and correlation of the monitoring location.

[0008] S3 integrates all the combinations of early warning parameters and clusters them based on the spatial distance between the monitoring locations to obtain the early warning classification under the corresponding scene information.

[0009] S4. Based on the triggering time period of the warning, count the number of warnings for each warning category per unit time, generate a warning trend curve, and retrieve the key difference points of each warning in the warning trend change to form a difference feature set.

[0010] S5 quantifies the common patterns of the differential feature set, identifies the correlation parameters that are strongly correlated with the early warning classification, and updates the monitoring indicators with the scene-location dimension corresponding to each correlation parameter to determine the monitoring indicators output at the current monitoring location.

[0011] A gas collection, monitoring and early warning system includes: a scene acquisition module, used to acquire scene information of gas transmission and distribution, collect gas concentration and gas pressure value at valve in real time based on scene information, synchronously record the monitoring location on each path, and verify the early warning conditions of each monitoring location.

[0012] The parameter combination module is used to output the combination of warning parameters and the trigger time period based on the parameter overlap and correlation of the monitoring location when an early warning is triggered at the monitoring location.

[0013] The early warning classification module is used to integrate all combinations of early warning parameters and cluster them based on the spatial distance between monitoring locations to obtain the early warning classification under the corresponding scene information.

[0014] The difference extraction module is used to count the number of warnings for each warning category per unit time according to the warning trigger time period, generate a warning trend curve, and retrieve the key difference points of each warning in the warning trend change to form a difference feature set.

[0015] The indicator update module is used to quantify the common patterns of the difference feature set, identify the correlation parameters that are strongly correlated with the early warning classification, and update the monitoring indicators according to the scene-location dimension corresponding to each correlation parameter, and determine the monitoring indicators output at the current monitoring location.

[0016] The beneficial effects of this invention are as follows: First, this invention monitors locations in groups according to the gas transmission and distribution topology, converts gas concentration and pressure values ​​into symbolic values ​​for early warning condition judgment, and filters effective paths in combination with path coverage thresholds; reducing the problems of missed and false alarms during gas early warning and ensuring the reliability of early warning triggering.

[0017] Second, this invention determines parameter overlap by using time interval clustering, and uses complete overlap, partial overlap and non-overlap to describe the clustering of current warnings. It combines random forest regression to fit the pressure-concentration relationship and outputs parameter combinations ranked by feature importance, thereby improving the accuracy of parameter combination output and preventing excessive analysis complexity caused by unclear parameter correlation.

[0018] Third, this invention uses the spatial distance of the monitoring location as a basis, combined with the similarity of the warning type and the cosine similarity of the parameter features to calculate the weighted spatial distance, performs clustering and calculates the average amplitude of exceeding the threshold, so as to quickly identify gas leak anomalies and avoid the classification distortion problem caused by the separation of spatial and attribute dimensions in clustering.

[0019] Fourth, this invention determines whether the trend curve formed by the number of warnings conforms to historical trends. If it conforms, the inflection point is taken; if it does not conform, the extreme value is taken as the key difference point to form a difference feature set. The warning trend is extracted in real time. Then, the principal components of the difference feature set are extracted by principal component analysis. The chi-square test is used to verify the significant correlation between the features and the warning classification. The strongly correlated parameters are bound to the scene-location dimension to update the monitoring indicators. This achieves rapid response to warnings and accuracy in dynamic classification. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This is a flowchart illustrating a gas collection, monitoring, and early warning method.

[0022] Figure 2 This is a flowchart illustrating step S1 of a gas collection, monitoring, and early warning method.

[0023] Figure 3 This is a flowchart illustrating step S2 of a gas collection, monitoring, and early warning method.

[0024] Figure 4 This is a flowchart illustrating step S3 of a gas collection, monitoring, and early warning method.

[0025] Figure 5 This is a system framework diagram of a gas collection, monitoring, and early warning system. Detailed Implementation

[0026] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0027] See Figure 1 A gas collection, monitoring and early warning method includes: S1, acquiring gas transmission and distribution scenario information, collecting gas concentration and gas pressure value at valve in real time based on the scenario information, synchronously recording the monitoring location on each path, and verifying the early warning conditions of each monitoring location.

[0028] S2, when an early warning is triggered at a monitoring location, outputs the combination of early warning parameters and the trigger time period based on the parameter overlap and correlation of the monitoring location.

[0029] S3 integrates all the combinations of early warning parameters and clusters them based on the spatial distance between the monitoring locations to obtain the early warning classification under the corresponding scene information.

[0030] S4. Based on the triggering time period of the warning, count the number of warnings for each warning category per unit time, generate a warning trend curve, and retrieve the key difference points of each warning in the warning trend change to form a difference feature set.

[0031] S5 quantifies the common patterns of the differential feature set, identifies the correlation parameters that are strongly correlated with the early warning classification, and updates the monitoring indicators with the scene-location dimension corresponding to each correlation parameter to determine the monitoring indicators output at the current monitoring location.

[0032] In step S1, gas sensors and pressure sensors are used to collect gas concentration and gas pressure values ​​at valves in real time. The monitoring location of each type of parameter is recorded simultaneously, such as the specific location of valve A in the kitchen on the 3rd floor of a residential building and monitoring point B of pipeline No. 2 in an industrial workshop, as well as the collection timestamp and corresponding scene information, such as the specific scene of a residential kitchen or an industrial workshop. The collected data is stored in a location-scene-time association to identify gas leaks at multiple locations.

[0033] like Figure 2 As shown, the implementation of step S1 includes: S11, based on the topology of gas transmission and distribution, grouping the current monitoring location into different paths.

[0034] S12, based on the data collected for each path, convert the gas concentration and valve pressure values ​​into symbolic values, and use the symbolic values ​​to determine the warning conditions for each path.

[0035] S13, when the warning conditions meet the coverage of the current path, output the monitoring locations on the current path.

[0036] In step S11, the monitoring locations need to be grouped into different paths according to the physical layout of the gas transmission and distribution. Each path represents a continuous main pipeline segment or a branch pipeline segment, so as to manage data collection and early warning analysis by zone. Each monitoring location can represent a valve or sensor point, and these locations will serve as the main monitoring data for the gas.

[0037] Preferably, when monitoring locations are grouped onto different paths, five monitoring locations are grouped together to encompass the continuous branch pipelines and continuous main pipelines at the corresponding locations.

[0038] Step S12 converts continuous parameter values ​​(such as concentration and gas pressure) into discrete symbolic values, and then defines warning conditions based on these symbolic values. Normal, warning, and dangerous gas concentrations are represented by symbols (such as 0, 1, 2), respectively; the same method is used to set symbolic values ​​for gas pressure values. It should be noted that the current thresholds distinguishing between normal, warning, and dangerous conditions will be selected from historical data, specifically the confidence intervals for normal and dangerous gas concentrations. These confidence intervals are set at a 95% confidence level, i.e., the average ± 1.96 standard deviations. The portion exceeding the upper limit of the confidence interval for normal gas concentrations but less than the lower limit of the confidence interval for dangerous gas concentrations will be marked with warning symbols, and the symbolic values ​​will be set to 0, 1, and 2 sequentially.

[0039] The warning conditions will trigger warnings for the corresponding path based on the value of the symbol. For example, if the concentration symbols at multiple monitoring points along the path are all in the dangerous range, a warning will be triggered. The warning conditions here can be retrieved directly from the database. For example, "IF concentration symbol is high AND pressure symbol is low THEN trigger leak warning" will be expressed as a regular expression, matching the symbol values ​​along the current path.

[0040] As for considering the coverage of the current path, it is to ensure that the warning is not caused by a single anomaly, but by a common phenomenon on the path, so as to improve the reliability of subsequent warnings.

[0041] The coverage of the current path will be represented by dividing the number of valid warnings by the number of monitored locations on the current path; the number of valid warnings is represented by removing the parts with missing symbol values, unsigned value outputs, and instantaneous changes in symbol values, and then subtracting the number of duplicate warnings.

[0042] Preferably, step S13 is further implemented by setting warning conditions based on the symbol value combination on each path, and checking the coverage of the current path when the warning conditions are triggered.

[0043] If the coverage does not meet the preset coverage threshold, the path constraint sub-items of each path are collected using the coverage of the current path, and the priority of each path is set by the path constraint sub-items; the path constraint sub-items will define the data format contained on each path based on the user type and transmission medium of the path service.

[0044] For example, path constraint sub-items will be divided into path priorities according to the service objects: high priority: residential building branch pipes, hospitals, schools and other places where service personnel are concentrated; medium priority: shopping malls, hotels and other commercial places; low priority: open-air municipal main pipes and other places with sparse personnel.

[0045] The transmission medium will be assigned priority based on the pipe material and topology location. For example, PE pipes, which are prone to aging, will be assigned high priority, while steel pipes with anti-corrosion treatment will be assigned low priority. Monitoring locations in the transmission and distribution network will be assigned high priority if they are on the core path and low priority if they are at the end of a branch.

[0046] Calculate the similarity between the priority and coverage of each path and the paths that meet the coverage threshold, and output the path with the highest similarity.

[0047] It should be noted that the preset coverage threshold is calculated based on data from one hundred adjacent batches in historical data. The confidence interval of the coverage rate at the time of the warning is regarded as the current preset coverage threshold, that is, the average value ± 1.96 standard deviations is selected as its confidence interval. If the value is within the confidence interval, the corresponding path is considered to meet the preset coverage threshold; otherwise, it does not meet the threshold. The similarity between the path that does not meet the threshold and the path that does meet the coverage threshold will be calculated. The path that does not meet the preset coverage threshold will use its priority and coverage value, while the path that meets the coverage threshold will use the average of the corresponding path priority and coverage. The similarity will be calculated using the formula for the Pearson correlation coefficient. At this time, an additional path will be output. This path will be close to the obvious warning situation. It is necessary to focus on investigating whether the failure to identify it is due to temporary fluctuations. Ultimately, this is to improve the accuracy of gas warnings and reduce false alarms.

[0048] It should be noted that subsequent similarity calculations will be based on either cosine similarity or Pearson correlation coefficient.

[0049] In one embodiment of the present invention, in step S2, the parameter combination for gas warning will be further identified, and these parameters will be output in the form of parameter overlap to trigger the warning and the trigger time period, so as to further record the warning form under gas transmission and distribution.

[0050] The above parameter association method will initially identify the triggered warnings and divide them into single warning triggers where only one parameter exceeds the threshold and multiple warning triggers where multiple parameters exceed the threshold simultaneously or the same parameter exceeds the threshold consecutively.

[0051] The parameter overlap will then be explained to illustrate the complete overlap of parameters from the same source, the partial overlap of parameters from the same source but different sources, and the non-overlap of parameters from different sources but different sources; the combination of these parameters will be the subject of identification and analysis in the current step.

[0052] For example, complete overlap indicates that the warnings are caused by the same warning condition or the same event, and all warnings are linked to the gas concentration parameter at the same monitoring location, with the concentration change trend showing a monotonically increasing or periodic fluctuation. For example, a valve may issue three consecutive warnings showing that the concentration at the valve has increased from 15% LEL to 25% LEL, with the time interval being less than 1 minute.

[0053] The system analyzes the corresponding monitoring location by checking the warning timestamp of the corresponding monitoring location. When the parameters completely overlap, the system will directly associate the concentration change rate of the valve as the primary monitoring indicator.

[0054] The identification methods for pressure values ​​and gas concentration parameters under complete overlap are the same, but the trend of pressure value changes shows a monotonically decreasing or periodic fluctuation.

[0055] Partial overlap indicates that although the parameters bound to the warnings are all gas concentrations, they involve different monitoring locations. In this case, homogeneity means that the warnings are caused by the same warning condition or the same event. These warnings are close in time and are not bound to the same monitoring location. By analyzing the differences in gas concentration and the co-analysis of gas pressure and gas concentration, a composite monitoring index is generated. For example, warning 1 is bound to the concentration of valve A (20% LEL), and warning 2 is bound to the concentration of valve B (18% LEL), with a time interval of less than 1 minute. Since the output warnings are based on path division and labeled with warning conditions, parameters under the same warning condition will default to being on a continuous branch pipeline and main pipeline, without the need for direct spatial clustering.

[0056] Non-overlapping parameters refer to parameters involving different scenarios. For example, warning 1 is associated with the concentration (22% LEL) of valve X, and warning 2 is associated with the pressure (0.8 MPa) of valve Y. In this case, it is necessary to identify the non-linear relationship between the two parameters and generate a pressure-concentration coupling index as a monitoring basis.

[0057] like Figure 3As shown, the implementation of step S2 also includes: S21, for the monitoring location on the current path, determine whether the current situation is an abnormal situation triggered by a single warning. If it is a single warning trigger, retrieve the first trigger time and the last trigger time of the corresponding monitoring location, and use the time interval between two adjacent triggers and the complete features of the single parameter as the output data; the first trigger time represents the timestamp of the parameter first exceeding the threshold, and the last trigger time represents the timestamp of the parameter last exceeding the threshold. At this time, the time interval between two adjacent triggers is obtained as the trigger time period to describe the situation of the corresponding monitoring location; as for the warning parameter combination triggered by a single warning, it will be obtained through the complete features of the single parameter (type + threshold + actual value). For example, when the type is gas concentration, the output will be gas concentration type + threshold when gas concentration triggers warning conditions + actual value of gas concentration, thereby obtaining the corresponding warning parameter combination.

[0058] S22, if it does not belong to a single warning trigger, the time period of the warning trigger at the corresponding monitoring location is taken as the output trigger time period, and the parameter overlap of the corresponding monitoring location is determined. Parameter overlap includes complete overlap of the same source and same parameter, partial overlap of the same source and different parameter, and non-overlap of different source and different parameter. As for the time period output at this time, it is not the time of a single warning, but the time period from multiple warnings to the recovery cycle. The earliest trigger time and the latest time period among multiple warnings are selected to obtain the warning trigger time period. This time period will represent the trigger time period output in the current step.

[0059] S23, when there is complete overlap, the trend change rate of gas concentration and air pressure values ​​will be used as the output warning parameter combination. When there is complete overlap, the warning parameter combination will use the trend change rate of gas concentration and air pressure values ​​as the main monitoring indicators, and combine them with the gas concentration and air pressure values ​​under the corresponding scenario to form a warning parameter combination. This parameter combination will represent the main changes in the current gas concentration or air pressure value.

[0060] S24, in cases of partial overlap, the concentration gradient, pressure, and the rate of coordinated change of concentration are used as the output warning parameter combination. In cases of partial overlap, the same warning event or condition triggers concentration changes at multiple monitoring locations. In this case, the concentration values ​​of two consecutive monitoring locations are subtracted and divided by the distance between them. The distance, concentration, and pressure values ​​are all standardized. The rate of coordinated change of pressure and concentration is expressed as |concentration change rate / pressure change rate| × sign (concentration change rate) × sign (pressure change rate). This is achieved by multiplying the absolute value of the relevant rate of change by the relevant sine value to determine if the current data shows a consistent direction of coordinated change. The output concentration gradient focuses on whether the gas concentration measured at the current monitoring location is affected by the ventilation environment, emphasizing the value of a single indicator. The pressure change gradient is not output because the pressure change is relatively small compared to the gas concentration change; it serves as an auxiliary monitoring indicator, using the rate of coordinated change to illustrate the coordinated changes in gas concentration and pressure at multiple monitoring locations.

[0061] S25, when the relationship is non-overlapping, uses random forest regression to fit the pressure-concentration relationship, and uses the feature values ​​ranked by feature importance as the output combination of warning parameters.

[0062] It should be noted that when using random forest regression to fit the pressure-concentration relationship, data including pressure, pressure mean, pressure change rate, low-frequency trend of pressure, and pressure variance are constructed. This data will be used to construct a dataset related to the current pressure through multiple batches of historical data. The pressure dataset will be used as the input feature for random forest analysis, and the concentration-related features will be used as the target value. The target value for each batch of data is the gas concentration value of the corresponding batch. The gas concentration-related features can also be features such as mean and change rate, but in this case, the target value in the regression analysis must be a single variable. Here, we take the gas concentration value as an example. The resulting dataset will represent the multi-dimensional features of pressure plus the target concentration value.

[0063] During the subsequent random forest analysis, the root node is the initial splitting node of the tree. This involves selecting multiple features from the dataset as candidate features, typically 2-3 features each time. For each candidate feature, all possible values ​​are iterated, and the total mean square error (MSE) is calculated for each value as a splitting point. The feature with the smallest MSE and the splitting point are selected to split the current node, generating left and right child nodes. This process is repeated recursively until the number of samples in the current node is less than 5, or the reduction in MSE after splitting is less than 0.01. At this point, the splitting stops, and the current node is considered a leaf node, thus completing the generation of the corresponding tree structure.

[0064] As for feature importance, the mean squared error reduction of each pressure feature across all split nodes will be calculated to obtain a feature importance value. This score will also be standardized after calculation to emphasize its magnitude. If the current pressure feature exists in multiple tree structures, the average feature importance of the feature will be selected. Then, the identified features will be sorted according to this value, and the pressure features and the corresponding gas concentration values ​​will be output in descending order of feature importance value as the parameter combination for the current scenario.

[0065] Preferably, the method for determining the parameter overlap of the corresponding monitoring location also includes: verifying the warning conditions that trigger the warning, and performing cluster analysis based on the time point when the warning conditions are triggered on each path.

[0066] If the average time interval of the current clustering results is less than or equal to the preset threshold, the corresponding monitoring locations are considered to be completely overlapping when all current clustering results point to one monitoring location; and the corresponding monitoring locations are considered to be partially overlapping when all current clustering results point to multiple monitoring locations.

[0067] If the average time interval of the current clustering results is greater than the preset threshold, the corresponding monitoring location is considered to be non-overlapping.

[0068] When performing time clustering, 60 seconds will be selected as the threshold for clustering to identify whether the current warning is caused by the same warning condition or the same warning event.

[0069] In one embodiment of the present invention, step S3 emphasizes the processing of spatial clustering, that is, ignoring the path of the current warning and clustering all the warnings to identify the locations where warnings frequently occur, emphasizing the spatial location of the warnings, classifying them in local space, and setting the warning classification under the corresponding scene information.

[0070] like Figure 4 As shown, the implementation of step S3 also includes: S31, classifying the warning parameter combinations by type and configuring the warning type for the corresponding warning parameter combinations. The type classification will divide the warning parameter combinations into several warning types according to the parameters, such as leakage warning, pressure anomaly warning, and mixed anomaly warning. This warning type represents the initial warning type label. The subsequent output warning classification will combine multiple dimensions to add more specific label descriptions to the clustered warning parameter combinations.

[0071] Leakage warnings include gas concentration parameters, which can be combined with pressure / rate of change (e.g., concentration + pressure, concentration + rate of change of concentration, rate of change of concentration + rate of change of pressure). The parameters corresponding to these descriptions are labeled as leakage warnings. These data will include parameters related to the concentration in scenarios that are completely overlapping and triggered by a single warning.

[0072] The pressure anomaly category only contains pressure-related parameters, such as pressure, pressure mean, and pressure variance, but no concentration parameters; this section will include parameters containing only pressure in scenarios that are completely overlapping and triggered by a single warning.

[0073] Mixed anomaly classes include other combinations of concentration / pressure, such as the co-current rate of change of pressure and concentration, and feature values ​​ranked by feature importance; these data represent data in partially overlapping and non-overlapping scenarios.

[0074] S32, the combination of warning parameters after type classification is mapped to multiple monitoring locations, and the distance between monitoring features is weighted by the similarity of warning types and parameter features between monitoring locations to obtain the weighted spatial distance between monitoring features.

[0075] It should be noted that all data involved in the combination of warning parameters will be standardized to eliminate their dimensions, facilitating subsequent analysis.

[0076] Preferably, the weighted spatial distance is a weighted sum of the distance between monitoring locations, the similarity of warning types, and the similarity of parameter features, with weights set to 0.5, 0.3, and 0.2 respectively. The warning type similarity will be set according to a value of 1 for the same type and 0.2 for different types. The parameter feature similarity will be calculated using cosine similarity. Since the pressure anomaly class only contains pressure data, its similarity with the other two types will be calculated based on the pressure values. Simultaneously, since the mixed anomaly class contains data ranked by feature importance, when calculating similarity with this type, the feature importance values ​​of the corresponding data will be introduced, and the corresponding similarity will be obtained using weighted cosine similarity. If multiple data points are involved, the similarity will be averaged.

[0077] S33 uses weighted spatial distance for clustering, calculates the average magnitude of parameters exceeding the threshold in the clustering results, and combines the average magnitude with the elements of the warning type and warning parameter combination to set the warning classification for the clustering results.

[0078] The output warning classification will include three dimensions of annotation, such as the warning type including the annotation of the three categories + the annotation of the gas concentration change rate after the combination of warning parameters + the average magnitude of the gas concentration change rate exceeding the threshold, to illustrate the local area of ​​spatial clustering in the corresponding region.

[0079] When performing spatial clustering, DBSCAN clustering will be used, with the neighborhood radius set according to the scene where the current monitoring location is located. For example, a neighborhood radius of 30 meters will be set for the scene of a residential building branch pipe, and a neighborhood radius of 100 meters will be set for the scene of a municipal main pipe. As for the minimum number of samples, 3 will be selected to illustrate the clustering situation of close proximity in most scenarios.

[0080] Preferably, when using weighted spatial distance for clustering, the implementation method further includes: determining the intra-cluster warning type of the clustering result, and outputting the corresponding clustering result when the intra-cluster warning types are consistent or the intra-cluster similarity meets the similarity threshold.

[0081] The similarity threshold here will be set to 0.8 to identify location clusters with similar warning features after weighted spatial clustering. At the same time, in order to prevent spatial distance from affecting the current clustering effect, data such as spatial distance and warning type similarity will be normalized to obtain warning classifications under the same risk or warning.

[0082] The output clustering results also need to meet the condition that the warning types within the cluster are consistent or the similarity is ≥0.8, and finally the clustering is completed.

[0083] The average magnitude of the above parameters exceeding the threshold will retrieve the threshold involved in step S1 and use the proportion to describe the dimensions in the current cluster that exceed the corresponding threshold. Since the threshold set in step S1 is used to distinguish between normal, warning and dangerous situations, the threshold that distinguishes between warning and dangerous situations will be selected to describe the urgency of the gas leak.

[0084] In one embodiment of the present invention, the unit time set in step S4 will be in the form of 1 hour or 1 day to count the number of warnings for gas transmission and distribution within a short period of time.

[0085] The implementation of step S4 includes: when the increase in the number of warnings conforms to the historical trend, the inflection point of the warning trend curve is regarded as the key difference point; if it does not conform to the historical trend, the maximum and minimum values ​​of the warning trend curve are selected as the key difference points, and the data corresponding to the key difference points are used to form a difference feature set.

[0086] Selecting the key difference point at this time is to ensure that the current data clearly points to the cause of the anomaly, which will facilitate subsequent monitoring of leaks during gas transmission and distribution.

[0087] The preferred calculation method, which conforms to historical trends, involves fitting the warning trend curve generated by the number of warnings to the historical trend line to generate a baseline trend line, calculating the absolute deviation between the current warning data and the historical trend line, and determining whether it conforms to historical trends if all absolute deviations are less than or equal to the average deviation threshold; otherwise, it does not conform.

[0088] When fitting the trend, the least squares method is used, and a linear regression model is used to describe the fitted baseline trend line. The calculation method is shown below.

[0089] ;in, This represents the intercept term. The value of the intercept term will be set based on the average number of warnings and the average time, such as... ;in, This represents the average number of warnings. Represents the average over time. Represents the regression coefficient. ;in, The index representing the historical trend line ranges from 1 to n. This represents the number of data points in the historical trend line. This represents the value at the i-th time point. This represents the number of warnings for the i-th time. The time value here is calculated based on a standardized timestamp or sequence number. By fitting the trend of historical data, a linear model is obtained. At the same time, the coefficient of determination is also used to verify the baseline trend line. Only when the coefficient of determination is greater than 0.6 is the current baseline trend line considered valid. The coefficient of determination is used to detect the goodness of fit of the trend. When it is less than 0.6, the historical trend is not obvious, which makes the baseline trend line unable to interpret the current situation.

[0090] The absolute deviation is calculated by subtracting the predicted value of the baseline trend line at that point in time from the actual value of the current warning data, and then taking the absolute value. The predicted value of the baseline trend line is calculated by substituting the current point in time into the linear regression equation obtained in the first step. At this point, a time mapping is performed between the current warning data and historical data to ensure that the time scales of the current data and historical data are consistent.

[0091] The average deviation threshold is calculated based on the residuals between historical data and the baseline trend line (i.e., the absolute deviation of the historical data). The threshold can be set as the average plus 1.5 times the standard deviation, which can cover most deviations in the historical data, ensuring that most points in the historical data fall within the threshold.

[0092] Then, the absolute deviation of each point in the current warning data is compared with the average deviation threshold. If the absolute deviation of all points is less than or equal to the threshold, it means that the fluctuation of the current warning data is within the range of historical normal fluctuations, and therefore it is determined to be in line with the historical trend.

[0093] In step S5, the differential feature set is analyzed by principal component analysis or chi-square test. The core purpose is to test whether different warning categories have statistical independence in the distribution of differential features. The strongly correlated parts of the differential features of the corresponding warning categories are regarded as the output correlation parameters. These correlation parameters are used to bind the scene-location dimension and update the independently monitored index values ​​at each monitoring location.

[0094] Therefore, the implementation of step S5 also includes: S51, determining the difference dimension of the difference feature set under different early warning classifications by performing principal component analysis on the difference feature set.

[0095] S52 uses the chi-square test to determine the significant association between different dimensions of the differential feature set based on the differential dimensions, and regards the data with significant association as the output association parameter.

[0096] The associated parameters are then bound to the corresponding scene and location as the common model for current identification. The actual values ​​of the associated parameters are used to set the monitoring indicators. For example, if the associated parameter for the warning classification related to pressure anomalies in region A is the standard deviation of pressure value, then the monitoring indicator for that location is updated to the standard deviation threshold of pressure value. By observing the values ​​of these parameters, the occurrence of pressure anomalies can be determined, thereby monitoring the accuracy of warnings appearing at the corresponding locations.

[0097] Regarding the aforementioned principal component analysis, a correlation coefficient matrix is ​​constructed from multiple features in the differential feature set, and principal components with a cumulative variance contribution rate ≥ 80% are extracted. That is, the top N principal components are selected, where N represents the number of differential dimensions currently selected. The variance contribution rate is obtained by calculating the eigenvalues ​​of these differential dimensions. Then, these eigenvalues ​​are displayed according to the corresponding warning categories. The p-value of the corresponding dimension under each warning category is determined using chi-square statistics. When the p-value is less than 0.05, it is considered that there is a significant correlation, and the corresponding data is regarded as the current correlation parameter. The correlation parameter will represent one or more features, emphasizing the type of warning that is likely to occur at the corresponding location. These types can be used as indicators for subsequent main monitoring to promptly verify the leakage situation during gas transmission and distribution and prevent false leak identification.

[0098] like Figure 5 As shown, the present invention also provides a gas acquisition, monitoring and early warning system, including: a scene acquisition module, a parameter combination module, an early warning classification module, a difference extraction module and an indicator update module; wherein, the output end of the scene acquisition module is connected to the parameter combination module, the output end of the parameter combination module is connected to the early warning classification module, the output end of the early warning classification module is connected to the difference extraction module, and the output end of the difference extraction module is connected to the indicator update module.

[0099] The scene acquisition module is used to acquire scene information of gas transmission and distribution, collect gas concentration and gas pressure value at valve in real time based on scene information, record the monitoring location on each path simultaneously, and verify the early warning conditions of each monitoring location.

[0100] The parameter combination module is used to output the combination of warning parameters and the trigger time period based on the parameter overlap and correlation of the monitoring location when an early warning is triggered at the monitoring location.

[0101] The early warning classification module is used to integrate all combinations of early warning parameters and cluster them based on the spatial distance between monitoring locations to obtain the early warning classification under the corresponding scene information.

[0102] The difference extraction module is used to count the number of warnings for each warning category per unit time according to the warning trigger time period, generate a warning trend curve, and retrieve the key difference points of each warning in the warning trend change to form a difference feature set.

[0103] The indicator update module is used to quantify the common patterns of the difference feature set, identify the correlation parameters that are strongly correlated with the early warning classification, and update the monitoring indicators according to the scene-location dimension corresponding to each correlation parameter, and determine the monitoring indicators output at the current monitoring location.

[0104] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. A gas collection monitoring and early warning method, characterized in that, Comprise: S1, acquire the scene information of gas transmission and distribution, collect gas concentration and valve gas pressure value in real time according to the scene information, synchronously record the monitoring position on each path, and check the early warning condition of each monitoring position; S2, when triggering early warning at the monitoring position, output the early warning parameter combination and triggering time period based on the parameter overlap and correlation form of the monitoring position; S3, integrate all early warning parameter combinations, cluster them according to the spatial distance between monitoring positions, and obtain the early warning classification under the corresponding scene information; S4, according to the triggering time period of early warning, count the number of early warnings of each early warning classification in unit time, generate an early warning trend curve, and call the key difference points of each early warning in the early warning trend change to form a difference feature set; S5, quantize the common mode of the difference feature set, identify the associated parameters strongly related to the early warning classification, and update the monitoring indicators in the scene-position dimension corresponding to each associated parameter to determine the monitoring indicators output by the current monitoring position; The implementation of step S2 further comprises: S21, for the monitoring position on the current path, determine whether it is an abnormal situation of single early warning triggering, when it belongs to single early warning triggering, call the first triggering time and the last triggering time of the corresponding monitoring position, and use the time interval between the adjacent two times of triggering and the complete feature of a single parameter as the output data; S22, if it does not belong to single early warning triggering, take the time period of early warning triggering at the corresponding monitoring position as the output triggering time period, and determine the parameter overlap of the corresponding monitoring position, which includes complete overlap of homologous same parameters, partial overlap of homologous different parameters and non-overlap of heterologous different parameters; S23, when it belongs to complete overlap, take the trend change rate of gas concentration and gas pressure value as the output early warning parameter combination; S24, when it belongs to partial overlap, take the concentration gradient, pressure and concentration synergistic change rate as the output early warning parameter combination; S25, when it belongs to non-overlap, use random forest regression to fit the pressure-concentration relationship, and use the feature value sorted by feature importance as the output early warning parameter combination; The implementation of determining the parameter overlap of the corresponding monitoring position further comprises: checking the early warning condition triggering early warning, clustering analysis is performed on the time points of the early warning condition triggering on each path; if the time interval average of the current clustering result is less than or equal to the preset threshold, when the current clustering result all points to one monitoring position, it is considered that the corresponding monitoring position is complete overlap; when the current clustering result all points to multiple monitoring positions, it is considered that the corresponding monitoring position is partial overlap; if the time interval average of the current clustering result is greater than the preset threshold, it is considered that the corresponding monitoring position is non-overlap.

2. The gas collection monitoring and early warning method according to claim 1, characterized in that, The implementation of step S1 comprises: S11, based on the topological structure of gas transmission and distribution, group the current monitoring position into different paths; S12, based on the collection situation of each path, convert the gas concentration and valve gas pressure value into symbolic values, and determine the early warning condition on each path through the symbolic values; S13, when the coverage rate of the early warning condition meets the current path, output the monitoring position on the current path.

3. The gas collection monitoring and early warning method of claim 2, wherein, The implementation of step S13 further comprises: Setting an early warning condition based on a combination of symbol values on each path, when the early warning condition is triggered, checking the coverage of the current path; In the case where the coverage does not meet the preset coverage threshold, using the coverage of the current path to collect path constraint sub-items of each path, and setting the priority of each path based on the path constraint sub-items; The priority and coverage of each path are calculated with the path that meets the coverage threshold to calculate the similarity, and the path with the largest similarity is output.

4. The gas collection monitoring and early warning method of claim 1, wherein, The implementation of step S3 further includes: S31, type classification is performed on the early warning parameter combination, and the early warning type corresponding to the early warning parameter combination is configured; S32, the early warning parameter combination after type classification is mapped to a plurality of monitoring positions, the distance between the monitoring features is weighted using the early warning type similarity and the parameter feature similarity between the monitoring positions, and a weighted spatial distance between the monitoring features is obtained; S33, clustering is performed using the weighted spatial distance, the average amplitude of the parameters exceeding the threshold in the clustering result is calculated, and the average amplitude is combined with the elements of the early warning type and the early warning parameter combination to set an early warning classification for the clustering result.

5. The gas collection monitoring and early warning method of claim 4, wherein, When clustering is performed using the weighted spatial distance, the implementation further includes determining the early warning type within the cluster of the clustering result, and when the early warning type within the cluster is consistent or the similarity within the cluster meets a similarity threshold, the corresponding clustering result is output.

6. The method according to claim 1, wherein, The implementation of step S4 includes: When the growth of the number of early warnings meets the historical trend, the inflection point of the early warning trend curve is regarded as a key difference point, and if it does not meet the historical trend, the maximum value and the minimum value of the early warning trend curve are selected as the key difference point, and the data corresponding to the key difference point is composed into a difference feature set.

7. The method according to claim 1, wherein, The implementation of step S5 further includes: S51, by performing principal component analysis on the difference feature set, the difference dimension of the difference feature set under different early warning classifications is determined; S52, based on the difference dimension of the difference feature set, the chi-square test is used to judge the significant association between different difference dimensions, and the data with significant association is regarded as an output association parameter.

8. A gas collection monitoring and early warning system, characterized in that, It includes: A scene acquisition module is configured to acquire scene information of gas transmission and distribution, to collect gas concentration and valve pressure values in real time according to the scene information, to synchronously record monitoring positions on each path, and to check early warning conditions of each monitoring position; A parameter combination module is configured to output early warning parameter combinations and trigger time periods based on parameter overlap and association forms of monitoring positions when early warnings are triggered at the monitoring positions; An early warning classification module is configured to integrate all early warning parameter combinations, to cluster the early warning parameter combinations based on spatial distances between the monitoring positions, and to obtain early warning classifications under corresponding scene information; A difference extraction module is configured to count the number of early warnings of each early warning classification in a unit time according to trigger time periods of the early warnings, to generate an early warning trend curve, and to call key difference points of each early warning in early warning trend changes to compose a difference feature set; An index updating module is configured to quantify common modes of the difference feature set, to identify association parameters strongly related to the early warning classifications, and to update monitoring indexes in scene-position dimensions corresponding to the association parameters to determine monitoring indexes output by current monitoring positions.

Images