Gas pipe network leakage early warning method and system based on edge calculation

Abstract

The invention relates to the field of gas pipe network leakage early warning, in particular to a gas pipe network leakage early warning method based on edge calculation, which is applied to an equipment terminal, and comprises the following steps: responding to a leakage monitoring request, and receiving pipe network current data acquired by an acquisition end, the pipe network current data comprising pressure current data and leakage current data; calling a preset screening model to process the pipe network current data to obtain abnormal deviation values, and obtaining effective pipe network data based on the abnormal deviation values and the corresponding data fluctuation thresholds; receiving pipe network annotation data, obtaining hidden layer output values of the pipe network annotation data according to the initial evaluation model, and determining a state output weight based on each hidden layer output value to obtain a pipe network state evaluation model; and inputting the effective pipe network data into the pipe network state evaluation model for evaluation, generating a state evaluation result, processing an abnormal state evaluation result based on a preset early warning strategy, generating leakage evaluation data, and sending the leakage evaluation data to a data center.

Description

Technical Field

[0001] The present invention relates to the field of gas pipeline network leakage warning, and in particular to a gas pipeline network leakage warning method and system based on edge computing. Background Art

[0002] As a kind of urban energy, natural gas is widely used in industry, commerce and residents' lives due to its high combustion efficiency and low pollutant emissions, becoming an indispensable energy choice in modern society. The natural gas pipeline network is a key facility for the transportation and distribution of natural gas. The scale of underground gas pipelines is constantly expanding. During their use, gas leaks often occur due to aging, external shocks or natural disasters, posing a potential threat to the environment and personal safety. In order to ensure the safe operation of the gas pipeline network, pipeline leakage must be monitored in real time.

[0003] Currently, gas pipeline leaks are mainly monitored by installing leak and pressure sensors at fixed monitoring points, collecting and uploading data in real time to the data center for judgment. During the monitoring process, there are delays in data transmission and insufficient data reliability, which may lead to false alarms or missed alarms, affecting the safe operation of the gas pipeline network. Summary of the invention

[0004] The present invention provides a gas pipeline network leakage early warning method and system based on edge computing, which can improve the accuracy and real-time performance of gas leakage monitoring.

[0005] In order to solve the above technical problems, a technical solution adopted in an embodiment of the present invention is: to provide a gas pipeline leakage warning method based on edge computing, which is applied to a device terminal, and the method includes: responding to a leakage monitoring request, receiving pipeline current data obtained by an acquisition end, and the pipeline current data includes pressure current data and leakage current data; calling a preset screening model to process the pipeline current data to obtain an abnormal deviation value, and screening the pipeline current data based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipeline data; receiving pipeline annotation data, obtaining the hidden layer output value of the pipeline annotation data according to an initial evaluation model, determining the state output weight based on each hidden layer output value, and obtaining a pipeline state evaluation model; inputting the valid pipeline data into the pipeline state evaluation model for evaluation, generating a state evaluation result, processing the abnormal state evaluation result based on a preset early warning strategy, generating leakage evaluation data, and sending the leakage evaluation data to a data center.

[0006] Optionally, the data fluctuation threshold includes a pressure fluctuation threshold, and a preset screening model is called to process the pipeline current data to obtain an abnormal deviation value, and the pipeline current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipeline data, including: calling a preset screening model to calculate the pressure current data to obtain a pressure difference reference value and a pressure difference fluctuation value; determining the pressure deviation value corresponding to each pressure current data according to the pressure difference reference value and the pressure difference fluctuation value; determining the pressure fluctuation range based on the pressure fluctuation threshold, and when the pressure deviation value is not within the pressure fluctuation range, determining the pressure current data corresponding to the pressure deviation value as abnormal pressure data; removing each abnormal pressure data in the pressure current data to obtain pressure retention current data, and determining the pressure retention current data as valid pipeline pressure data.

[0007] Optionally, the data fluctuation threshold also includes a leakage fluctuation threshold, and a preset screening model is called to process the pipeline network current data to obtain an abnormal deviation value, and the pipeline network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipeline network data, including: calculating the leakage current data according to the preset screening model to obtain a leakage reference value and a leakage fluctuation value; determining the leakage deviation value corresponding to each leakage current data according to the leakage reference value and the leakage fluctuation value; determining the leakage fluctuation range based on the leakage fluctuation threshold, and when the leakage deviation value is not within the leakage fluctuation range, determining the leakage current data corresponding to the leakage deviation value as abnormal leakage data; removing each abnormal leakage data in the leakage current data to obtain leakage retention current data, and determining the leakage retention current data as valid pipeline network leakage data.

[0008] Optionally, the pipeline current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold, and before obtaining the valid pipeline data, the method also includes: receiving historical fluctuation data, determining an initial fluctuation threshold based on the historical fluctuation data, and retrieving the threshold adjustment strategy to obtain the pipeline operation data; obtaining operation fluctuation characteristics based on the pipeline operation data, retrieving the threshold adjustment table to match each operation fluctuation characteristic, and obtaining a fluctuation adjustment value, wherein the fluctuation adjustment value includes a seasonal adjustment value, an operation adjustment value, and a performance adjustment value; obtaining a comprehensive adjustment value based on the seasonal adjustment value and / or the operation adjustment value and / or the performance adjustment value, and updating the initial fluctuation threshold based on the comprehensive adjustment value to obtain a data fluctuation threshold.

[0009] Optionally, receiving pipe network annotation data, obtaining hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determining state output weights based on each hidden layer output value, and obtaining a pipe network state evaluation model, including: receiving pipe network annotation data, screening the pipe network annotation data according to the preset screening model, and obtaining pipe network training data; identifying the pipe network training data, obtaining the pipe network state characteristics and the corresponding pipe network state labels, and determining the pipe network state characteristics and the corresponding pipe network state labels as prediction task groups; calling the preset numbering rule to number each prediction task group, obtaining the group numbering sequence, and dividing the prediction task group corresponding to the initial number into prediction test groups based on the group numbering sequence; dividing the prediction task groups corresponding to the remaining numbers into prediction training groups, and The initial evaluation model is trained to obtain the initial hidden layer output value, and the initial state output weight is determined according to each initial hidden layer output value to obtain the evaluation training model; the evaluation training model is tested based on the prediction test group to obtain the test result and the corresponding pipe network state label, and the prediction error of the test result and the corresponding pipe network state label is calculated; according to the group number sequence, the prediction task groups corresponding to different numbers are selected in turn to be divided into the prediction test group, and the prediction task groups corresponding to the remaining numbers are divided into the prediction training group to obtain multiple prediction errors; based on the multiple prediction errors, the prediction task group corresponding to the maximum prediction error is removed, and the remaining prediction task groups repeat the above training and testing steps until the final hidden layer output value is obtained, and the state output weight is determined according to each hidden layer output value to obtain the pipe network state evaluation model.

[0010] Optionally, the initial evaluation model is trained based on each prediction training group to obtain the initial hidden layer output value, and the initial state output weight is determined according to each initial hidden layer output value to obtain the evaluation training model, including: calling the basic evaluation model, configuring the number of input nodes, the number of output nodes and the number of hidden layer nodes of the basic evaluation model to obtain the initial evaluation model; determining the activation function of each hidden layer node according to the initial evaluation model, and randomly initializing the weight and bias of each hidden layer node; calculating the pipe network state characteristics of the prediction training group based on the weight, the bias and the activation function to obtain the initial hidden layer output value; obtaining the initial state output weight according to each initial hidden layer output value, and obtaining the evaluation training model based on the initial state output weight.

[0011] Optionally, calculating the pipe network state characteristics of the prediction training group based on the weight, the bias and the activation function to obtain an initial hidden layer output value includes: The network state characteristics of the prediction training group are calculated by the following formula to obtain the initial hidden layer output value: in, is the output value of the first hidden layer node, is the activation function, 1 is the weight from the input node to the first hidden layer node, is the bias of the first hidden layer node, For the The input network status characteristics, Indicates transpose.

[0012] Optionally, obtaining an initial state output weight according to each initial hidden layer output value, and obtaining an evaluation training model based on the initial state output weight, includes: The evaluation training model is obtained by the following formula: in, To evaluate the training model for the input pipeline network status characteristics The predicted output value, L is the number of hidden layer nodes, is the output value of the first hidden layer node, is the weight from the first hidden layer node to the output node, is the sum of the products of the output values ​​of all hidden layer nodes and the corresponding weights, For the The output value of each hidden layer node corresponding to the network status feature, Output weights for the initial state.

[0013] Optionally, the method also includes: receiving the warning processing result sent by the processing end, and obtaining error warning data when the warning processing result is a missed alarm or a false alarm; sending the error warning data to the device terminal, and obtaining the model adjustment parameters of the pipeline status assessment model according to the error warning data; updating the pipeline status assessment model based on the model adjustment parameters to obtain the updated pipeline status assessment model.

[0014] In order to solve the above technical problems, another technical solution adopted in the embodiment of the present invention is: to provide a gas pipeline leakage warning system based on edge computing, the system comprising: a data acquisition module, used to respond to leakage monitoring requests, receive pipeline current data acquired by the acquisition end, the pipeline current data including pressure current data and leakage current data; an abnormal screening module, used to call a preset screening model to process the pipeline current data, obtain abnormal deviation values, and screen the pipeline current data based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipeline data; a model training module, used to receive pipeline annotation data, obtain the hidden layer output value of the pipeline annotation data according to the initial evaluation model, determine the state output weight based on each hidden layer output value, and obtain a pipeline state evaluation model; a state evaluation module, used to input the valid pipeline data into the pipeline state evaluation model for evaluation, generate a state evaluation result, process the abnormal state evaluation result based on a preset early warning strategy, generate leakage evaluation data, and send the leakage evaluation data to a data center.

[0015] Different from the prior art, the present invention provides a gas pipeline leakage warning method based on edge computing, which is applied to a device terminal, and the method includes: responding to a leakage monitoring request, receiving pipeline current data obtained by an acquisition end, and the pipeline current data includes pressure current data and leakage current data; calling a preset screening model to process the pipeline current data to obtain an abnormal deviation value, and screening the pipeline current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline data; receiving pipeline annotation data, obtaining a hidden layer output value of the pipeline annotation data according to an initial evaluation model, determining a state output weight based on each hidden layer output value, and obtaining a pipeline state evaluation model; inputting the valid pipeline data into the pipeline state evaluation model for evaluation, generating a state evaluation result, processing the abnormal state evaluation result based on a preset warning strategy, generating leakage evaluation data, and sending the leakage evaluation data to a data center. The present invention allows real-time data processing near the data source, thereby significantly reducing data transmission time. By processing data at the device terminal, it is possible to determine whether the gas pipeline network is leaking more quickly and accurately. By moving the data processing task to the device terminal, most of the data does not need to be uploaded to the data center, which can save a lot of server resources, bandwidth, electricity, and physical space, etc., thereby reducing overall operating costs. By reducing the transmission of data in the network, edge computing reduces the risk of data being intercepted, stolen, or tampered with, and improves the privacy and security of the data. Even if the connection with the data center is disconnected, the device terminal can still operate independently and continue to process data, improving the reliability of determining whether the gas pipeline network is leaking, and ensuring that it can continue to work without interruption when the network fails. In summary, the present invention realizes real-time data processing, reduces the time delay of data transmission, improves the real-time, safety, and reliability of gas leak monitoring, and also reduces the cost of data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] One or more embodiments are exemplarily described by corresponding drawings, which do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and the figures in the drawings do not constitute proportional limitations unless otherwise stated.

[0017] Figure 1 It is a flow chart of a gas pipeline network leakage warning method based on edge computing provided in an embodiment of the present application; Figure 2 This is a schematic diagram of the data before and after filtering of the pipe network current data when the acquisition end is normal, provided by an embodiment of the present application; Figure 3 This is a schematic diagram of the data before and after the pipe network current data is filtered when there is a large fluctuation at the collection end provided by an embodiment of the present application; Figure 4It is a schematic diagram of a pipeline network status assessment model training and testing process provided by an embodiment of the present application; Figure 5 It is a schematic diagram of the working principle of a pipeline network status assessment model provided in an embodiment of the present application; Figure 6 It is a structural diagram of a gas pipeline leakage warning system based on edge computing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] It should be noted that, if there is no conflict, the various features in the embodiments of the present invention can be combined with each other, and all are within the protection scope of the present invention. In addition, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order from the module division in the device schematic diagram or the order in the flow chart.

[0020] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0021] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0022] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0023] See also Figure 1 , Figure 1 It is a flow chart of a gas pipeline network leakage warning method based on edge computing provided in an embodiment of the present application, which is applied to a device terminal and includes steps S1 to S4, which are as follows: S1. Respond to a leakage monitoring request and receive pipe network current data acquired by a collection end, wherein the pipe network current data includes pressure current data and leakage current data.

[0024] Among them, the device terminal, namely the edge device, refers to the device that processes and analyzes data at the edge of the network, which usually has a certain computing power and can collect, process and store data at a place close to the place where the data is generated; the leakage monitoring request refers to a specific instruction or signal issued by the upper system or user, which is used to start or trigger the edge device to monitor the gas pipeline network and determine whether there is a leak; the acquisition end refers to the sensors arranged in various places of the gas pipeline network, which transmit the original data to the edge device for processing and analysis by continuously measuring key parameters, and may include pressure sensors, leakage sensors, etc.; the pipeline current data refers to the electrical signal data collected by the sensor from the gas pipeline network that reflects the state of gas flow, pressure change, etc.; the pressure current data refers to the current signal reflecting the change of gas pressure in the pipeline, and the pressure condition in the pipeline is monitored in real time by the pressure sensor and converted into the form of a current signal; the leakage current data refers to the current signal reflecting the change of gas flow in the pipeline, and the gas state (such as gas concentration, etc.) of the gas in the pipeline is collected in real time by the leakage sensor and converted into the form of a current signal.

[0025] Specifically, the terminal device can support the simultaneous reception of 4 signal channels and the collection of 4 4-20mA current signals (pipeline current data). Three of the signal channels are connected to the pressure sensor, which converts the collected pressure information into 4-20mA current signal transmission with an accuracy of 0.5%. The other channel is connected to the leakage sensor, which converts the collected leakage information into 4-20mA current signal transmission with an accuracy of 0.5%.

[0026] It should be noted that 4-20mA is a common current signal standard for industrial sensors. It is suitable for long-distance transmission and has strong anti-interference performance. It is widely used in the field of industrial automation. 4mA represents the lowest range value of the sensor, and 20mA represents the highest range value (full scale). The current changes between these values ​​represent different pressure values ​​or leakage values. The accuracy of 0.5% means that the maximum error between the measurement result and the true value will not exceed 0.5% of the full scale.

[0027] It can be understood that the installation location of the sensor can be set according to actual conditions, and the present invention does not impose specific restrictions on this. For example, the sensor can be installed at the intersection or branch node of the pipeline network. The sensor converts physical quantities such as pressure and gas concentration into standard current signals, so that the device terminal can process data from different sources, ensure the consistency and availability of the data, and further perform subsequent data preprocessing and calculation.

[0028] S2. Retrieve a preset screening model to process the pipeline network current data to obtain abnormal deviation values, and screen the pipeline network current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline network data.

[0029] Among them, the preset screening model represents a pre-set model for data screening of the current signal sent by the sensor, which can be specifically a Z-score model; the abnormal deviation value represents the difference between a data point (pipeline network current data) and the expected value, that is, the data point does not belong to the normal range; the data fluctuation threshold represents a numerical value used to define the degree of variation of the data under normal circumstances, and the threshold can be used to identify whether the data is within the normal range; the effective pipeline network data represents reliable data that conforms to the actual pipeline network status obtained after removing abnormal data.

[0030] It should be noted that the Z-score model is a commonly used statistical model for detecting abnormal data and standardizing data. The Z-score is a standardized indicator that describes the number of standard deviations of a data point from its mean. By calculating the Z-score of each data point, outliers that deviate significantly from the mean can be identified.

[0031] See also Figure 2 , Figure 2 It is a schematic diagram of the data before and after filtering of pipeline network current data when the acquisition end is normal, provided by an embodiment of the present application. Figure 2 - The fluctuation range of sensor data in a is relatively stable, with no obvious abnormal points. Figure 2 -b, the overall curve of the data remains consistent, only the noise or slight random fluctuations in the original data are removed, possible minor anomalies are eliminated, and the curve is smoother, that is, when the sensor data is fault-free, the screening of the data effectively removes the noise and maintains the main trend of the data.

[0032] See also Figure 3 , Figure 3 It is a schematic diagram of the data before and after filtering of the pipeline network current data when there is a large fluctuation at the collection end provided by an embodiment of the present application. Figure 3 - There are significant abnormal fluctuations in the data in a, such as the circled area, which shows a sudden offset, which is obviously beyond the normal fluctuation range of the data. This may be caused by factors such as sensor failure and sudden external interference. Figure 3 -b shows the results after filtering. The extreme anomalies in the data are better filtered out, and the data curve is more continuous and smooth. Due to the large fluctuations in the original data, some information may be mistaken for anomalies and removed, resulting in a certain difference between the filtered data and the actual status.

[0033] It should be noted that sensor start-stop, failure, and environmental factors may all generate abnormal data, which may lead to false positives or false negatives. Screening means identifying and removing abnormal data, thereby retaining valid data. Removing abnormal data can reduce noise, ensure that the data used for subsequent analysis and decision-making is accurate, avoid possible misleading, and reduce the risk of false positives or false negatives.

[0034] In some embodiments, the data fluctuation threshold includes a pressure fluctuation threshold, and a preset screening model is called to process the pipe network current data to obtain an abnormal deviation value, and the pipe network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipe network data, specifically including steps A1-A4: A1. Retrieve a preset screening model to calculate the pressure current data to obtain a pressure difference reference value and a pressure difference fluctuation value.

[0035] The pressure difference reference value represents the average value of all pressure and current data; and the pressure difference fluctuation value represents the standard deviation calculated based on all pressure and current data.

[0036] A2. Determine the pressure deviation value corresponding to each pressure current data according to the pressure difference reference value and the pressure difference fluctuation value.

[0037] The pressure deviation value indicates the difference between a certain pressure current data and a reference pressure current data.

[0038] Specifically, when the preset screening model is a Z-score model, the pressure deviation value can be calculated by the following formula: in, is the pressure deviation value, For a certain pressure current data, is the average value of all pressure and current data, that is, the pressure difference reference value, It represents the standard deviation calculated based on all pressure and current data, that is, the pressure difference fluctuation.

[0039] It should be noted that When it is equal to 0, it means that the pressure deviation value is equal to the average value of all pressure current data. If it is greater than 0, it means that the pressure deviation value is higher than the average value of all pressure current data. Less than 0, indicating that the pressure deviation value is lower than the average value of all pressure current data. The larger the absolute value of , the further the pressure current data is from the reference pressure current data (pressure difference reference amount).

[0040] A3. Determine a pressure fluctuation range based on the pressure fluctuation threshold, and when the pressure deviation value is not within the pressure fluctuation range, determine the pressure current data corresponding to the pressure deviation value as abnormal pressure data.

[0041] The pressure fluctuation range indicates the interval in which each pressure current data can fluctuate; and the abnormal pressure data indicates the abnormal data in the pressure current data.

[0042] Specifically, the pressure fluctuation threshold defines the fluctuation degree of the upper and lower limits. The pressure fluctuation range can be determined according to the fluctuation degree of the upper and lower limits. For example, if the pressure fluctuation threshold is ±3, the pressure fluctuation range is [-3, 3]. When the pressure deviation value corresponding to a certain pressure current data is 4, and 4 is not within the pressure fluctuation range [-3, 3], the pressure current data is determined as abnormal data.

[0043] A4. Remove abnormal pressure data from the pressure current data to obtain pressure retention current data, and determine the pressure retention current data as effective pipe network pressure data.

[0044] Among them, the pressure retention current data represents the pressure current data remaining after removing the abnormal data, that is, the pressure current data within the normal range; the effective pipeline network pressure data represents the pressure current data that is reliable and consistent with the actual pipeline network status after removing the abnormal data.

[0045] It should be noted that by removing abnormal pressure data and retaining valid pipeline pressure data, the interference of noise and invalid data is reduced, the accuracy of the data and the reliability of subsequent analysis are improved, and unnecessary system alarms caused by untrue data are avoided.

[0046] In some embodiments, the data fluctuation threshold also includes a leakage fluctuation threshold, and a preset screening model is called to process the pipe network current data to obtain an abnormal deviation value, and the pipe network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipe network data, specifically including steps B1-B4: B1. Calculate the leakage current data according to the preset screening model to obtain a leakage reference value and a leakage fluctuation value.

[0047] The leakage reference value represents the average value of all leakage current data; and the leakage fluctuation value represents the standard deviation calculated based on all leakage current data.

[0048] B2. Determine the leakage deviation value corresponding to each leakage current data according to the leakage reference value and the leakage fluctuation value.

[0049] The leakage deviation value represents the difference between a leakage current data and a reference leakage current data. difference.

[0050] Specifically, when the preset screening model is a Z-score model, the leakage deviation value can be calculated by the following formula: in, is the leakage deviation value, For a leakage current data, is the average value of all leakage current data, i.e., the leakage reference value, It represents the standard deviation calculated based on all leakage current data, that is, the leakage fluctuation.

[0051] It should be noted that Equal to 0 means that the leakage deviation value is equal to the average value of all leakage current data. Greater than 0, indicating that the leakage deviation value is higher than the average value of all leakage current data. Less than 0, indicating that the leakage deviation value is lower than the average value of all leakage current data. The larger the absolute value of , the further the leakage current data is from the reference leakage current data (reference leakage amount).

[0052] B3. Determine a leakage fluctuation range based on the leakage fluctuation threshold, and when the leakage deviation value is not within the leakage fluctuation range, determine the leakage current data corresponding to the leakage deviation value as abnormal leakage data.

[0053] The leakage fluctuation range indicates the interval in which each leakage current data can fluctuate; and the abnormal leakage data indicates abnormal data in the leakage current data.

[0054] Specifically, the leakage fluctuation threshold defines the fluctuation degree of the upper and lower limits. The leakage fluctuation range can be determined according to the fluctuation degree of the upper and lower limits. For example, if the leakage fluctuation threshold is ±2, the leakage fluctuation range is [-2, 2]. When the leakage deviation value corresponding to a leakage current data is -3, and -3 is not within the leakage fluctuation range [-2, 2], the leakage current data is determined as abnormal data.

[0055] B4. Remove each abnormal leakage data in the leakage current data to obtain leakage retention current data, and determine the leakage retention current data as effective pipe network leakage data.

[0056] Among them, the leakage retained current data represents the leakage current data remaining after removing abnormal values, that is, the leakage current data within the normal range; the effective pipeline leakage data represents the leakage current data that is reliable and consistent with the actual pipeline status after removing abnormal data.

[0057] It should be noted that by removing abnormal leakage current data and retaining valid pipe network leakage data, the interference of noise and invalid data is reduced, the accuracy of the data and the reliability of subsequent analysis are improved, and the possibility of false alarms in the system is reduced.

[0058] In some embodiments, the pipeline network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold value, and before obtaining valid pipeline network data, the method specifically further includes C1-C3: C1. Receive historical fluctuation data, determine an initial fluctuation threshold according to the historical fluctuation data, and call a threshold adjustment strategy to obtain pipe network operation data.

[0059] Among them, the historical fluctuation data represents the pipeline current data collected in the past period of time; the initial fluctuation threshold represents the initially set data fluctuation range; the threshold adjustment strategy represents the pre-set rules for dynamically adjusting the initial fluctuation threshold according to the actual operation of the gas pipeline network; the pipeline operation data represents the operating season, whether there is construction during operation, and the health index of the pipeline network, etc.

[0060] It should be noted that by receiving historical fluctuation data, the fluctuation of pipeline current in the past period of time can be obtained, which provides a basis for subsequent analysis and threshold setting. Historical data can help determine the typical fluctuation range of the system under normal operating conditions, thereby setting the initial fluctuation threshold, such as setting the initial fluctuation threshold to ±2 based on historical fluctuation data.

[0061] C2. Acquire operation fluctuation characteristics according to the pipeline network operation data, retrieve the threshold adjustment table to match each operation fluctuation characteristic, and acquire fluctuation adjustment values, wherein the fluctuation adjustment values ​​include season adjustment values, operation adjustment values, and performance adjustment values.

[0062] Among them, the operating fluctuation data represents representative fluctuation information extracted from real-time operating data, reflecting the fluctuation law under current operating conditions; the threshold adjustment table represents a comparison table containing different fluctuation information and corresponding adjustment values; the fluctuation adjustment value represents the numerical value used to correct the initial fluctuation threshold, that is, the adjustment range obtained to make the fluctuation threshold adapt to the current operating environment and conditions; the seasonal adjustment value represents the adjustment range of the fluctuation threshold according to seasonal changes; the operation adjustment value represents the adjustment range of the fluctuation threshold according to the influence of different operation modes (such as maintenance, overhaul, etc.) of the pipeline network; the performance adjustment value represents the adjustment range of the fluctuation threshold according to the performance changes (such as aging, failure, etc.) of the pipeline network equipment.

[0063] It should be noted that changes in the operating environment such as seasonal changes, changes in the working environment or changes in equipment performance may cause fluctuations in pipeline network data. Therefore, the fluctuation threshold needs to be dynamically adjusted according to the current operating conditions to avoid unnecessary false alarms or omissions.

[0064] C3. Obtain a comprehensive adjustment value according to the season adjustment value and / or the operation adjustment value and / or the performance adjustment value, and update the initial fluctuation threshold based on the comprehensive adjustment value to obtain a data fluctuation threshold.

[0065] Among them, the comprehensive adjustment value represents the final adjustment range obtained after considering various adjustment factors; the data fluctuation threshold represents the final threshold for data screening determined after adjustment.

[0066] Specifically, the system weights the seasonal adjustment value, the operation adjustment value, and the performance adjustment value to obtain a final comprehensive adjustment value. Each adjustment value can be linearly superimposed, or weighted according to the importance of different influencing factors. The specific setting can be based on actual conditions, and the present invention does not impose specific restrictions on this. For example, if the initial fluctuation threshold is ±2 units, the seasonal adjustment value is +0.5, the operation adjustment value is -0.3, and the comprehensive adjustment value is +0.2, the updated fluctuation threshold will be ±2.2.

[0067] It should be noted that a single adjustment value may not be able to fully reflect the current operating conditions. The comprehensive adjustment value, through the joint effect of multiple factors, can ensure that the fluctuation threshold is more in line with the needs of the actual operating environment.

[0068] S3. Receive pipe network annotation data, obtain hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determine state output weights based on each hidden layer output value, and obtain a pipe network state evaluation model.

[0069] Among them, the pipeline network annotation data represents a data set composed of historical data related to the gas pipeline network or real-time data collected by sensors. These data have been manually labeled as different states (such as normal, leakage, etc.) and can be used for model training and testing. Each data in the pipeline network annotation data is marked with a clear state label to guide the learning of the model; the initial evaluation model represents a pre-built untrained model, which can be specifically an extreme learning machine (ELM); the hidden layer output value represents the output value of the hidden layer of the model; the state output weight represents the relationship weight between the hidden layer and the output layer; the pipeline network state evaluation model represents a machine learning model that can perform state evaluation on the gas pipeline network, which is finally formed through the training process, and can be used to determine whether the gas pipeline network is leaking.

[0070] It should be noted that ELM is a fast single-hidden layer feedforward neural network that uses randomly initialized hidden layer node weights and biases for training. The training process only needs to be done once, that is, after the hidden layer is determined, the weight of the output layer needs to be solved by minimizing the output error. Compared with traditional neural networks, the computational complexity is greatly reduced. This feature gives ELM a significant speed advantage in processing data. In addition, although the model is simple and does not require complex iterative training, ELM can still exhibit good generalization capabilities.

[0071] See also Figure 4 , Figure 4 It is a schematic diagram of a pipeline network status assessment model training and testing process provided in an embodiment of the present application.

[0072] In some embodiments, step S3 (receiving pipe network annotation data, obtaining hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determining state output weights based on each hidden layer output value, and obtaining a pipe network state evaluation model) specifically includes steps S31 to S37: S31, receiving pipe network annotation data, screening the pipe network annotation data according to the preset screening model, and obtaining pipe network training data.

[0073] Among them, the preset screening model refers to a pre-set model for screening data, which can be specifically a Z-score model; the pipeline network training data refers to the official website annotation data for training the model after removing noise and abnormal data.

[0074] S32: Identify the pipeline network training data, obtain the pipeline network status characteristics and the corresponding pipeline network status labels, and determine the pipeline network status characteristics and the corresponding pipeline network status labels as a prediction task group.

[0075] Among them, the pipeline network status characteristics represent the key attribute characteristics used to describe the operation status of the pipeline network obtained after feature extraction of the pipeline network training data, such as pressure characteristics, gas concentration characteristics, temperature characteristics, etc.; the pipeline network status labels represent the classification labels pre-defined for the feature data, including normal, leakage, etc.; the prediction task group represents a set of data sets containing features and corresponding labels, which are used to train or verify the model.

[0076] It should be noted that obtaining the characteristics of the pipeline network status helps the model focus on the key factors that affect whether the pipeline network is leaking, laying the foundation for subsequent model training.

[0077] S33, calling a preset numbering rule to number each prediction task group, obtaining a group numbering sequence, and dividing the prediction task group corresponding to the initial number into prediction test groups based on the group numbering sequence.

[0078] Among them, the preset numbering rule represents a predefined rule for assigning a unique number to each data group; the group numbering sequence represents the task group number arrangement sequence determined according to the preset numbering rule; the initial number represents the first number in the task group number arrangement sequence; the prediction test group represents a data group divided from the prediction task group for performing model performance evaluation on the trained model.

[0079] It should be noted that assigning a unique number to each prediction task group helps to clearly identify and track the task group in subsequent processing and improve the efficiency of model training.

[0080] S34, dividing the prediction task groups corresponding to the remaining numbers into prediction training groups, training the initial evaluation model based on each prediction training group, obtaining the initial hidden layer output value, determining the initial state output weight according to each initial hidden layer output value, and obtaining the evaluation training model.

[0081] Among them, the remaining numbers are the numbers except the first number; the prediction training group represents the data group used to train the model; the initial hidden layer output value represents the output value of each hidden layer in the first round of training; the initial state output weight represents the output layer weight obtained in the first round of training; the evaluation training model represents the model obtained after the first round of training of the initial evaluation model.

[0082] In some embodiments, step S34 (training the initial evaluation model based on each prediction training group, obtaining the initial hidden layer output value, determining the initial state output weight according to each initial hidden layer output value, and obtaining the evaluation training model) specifically includes steps S341 to S344: S341, calling a basic evaluation model, configuring the number of input nodes, the number of output nodes, and the number of hidden layer nodes of the basic evaluation model, and obtaining an initial evaluation model.

[0083] Among them, the basic evaluation model represents the preliminary model before configuration, which is a simple machine learning model, specifically ELM; the number of input nodes represents the number of nodes in the model input layer, which is usually the same as the number of input features; the number of output nodes represents the number of nodes in the model output layer, which is usually the same as the number of predicted categories or states, that is, how many state labels can be output; the number of hidden layer nodes represents the number of nodes in the hidden layer.

[0084] It should be noted that the number of nodes in the hidden layer needs to be adjusted according to the complexity of the task. Too few nodes may not capture the characteristics of the data, while too many nodes may lead to overfitting.

[0085] Specifically, see Figure 5 , Figure 5This is a schematic diagram of the working principle of a pipeline network status assessment model provided in an embodiment of the present application. The model has 5 input nodes, 1 output node and L hidden layer nodes.

[0086] S342. Determine the activation function of each hidden layer node according to the initial evaluation model, and randomly initialize the weight and bias of each hidden layer node.

[0087] Among them, the activation function represents the mathematical function applied in the hidden layer node, which determines the output of the node; the weight of the hidden layer node represents the weight from the input node to the first hidden layer node; the bias of the hidden layer node represents the bias from the input node to the first hidden layer node.

[0088] It should be noted that the activation function introduces nonlinear characteristics, enabling the model to learn complex functional relationships and improve the performance of the model. Random initialization can promote different nodes to learn different features.

[0089] S343. Calculate the pipe network status characteristics of the prediction training group based on the weight, the bias and the activation function to obtain an initial hidden layer output value.

[0090] The initial hidden layer output value represents the output value of each hidden layer node obtained by performing the first round of training on the model.

[0091] In some embodiments, calculating the pipe network state characteristics of the prediction training group based on the weight, the bias, and the activation function to obtain an initial hidden layer output value includes: The network state characteristics of the prediction training group are calculated by the following formula to obtain the initial hidden layer output value: in, is the output value of the first hidden layer node, is the activation function, 1 is the weight from the input node to the first hidden layer node, is the bias of the first hidden layer node, For the The input network status characteristics, Indicates transpose.

[0092] Specifically, when predicting the network status characteristics, the input network status characteristics are first transformed linearly through weights, and then a bias term is added, and finally a nonlinear transformation is performed through an activation function to obtain the output of the first hidden layer, which is used as the input of the output layer.

[0093] It should be noted that the weights from the input nodes to the hidden layer nodes can be adjusted by linear transformation to adjust the importance of the input network state features in each dimension. Transposition can ensure that the input features and weight vectors can be multiplied mathematically. It is the dot product of the weight from the input node to the first hidden layer node and the input network state feature, which represents the linear combination of the input network state feature after weighting.

[0094] S344. Obtain initial state output weights according to each initial hidden layer output value, and obtain an evaluation training model based on the initial state output weights.

[0095] Among them, the initial state output weight represents the weight calculated according to each initial hidden layer output value, which is used to map the hidden layer output to the final output; the evaluation training model is a model formed after preliminary training and weight optimization (the first round of training and testing), which aims to improve the prediction ability of the pipeline network status and is the basis for subsequent optimization.

[0096] In some embodiments, obtaining an initial state output weight according to each initial hidden layer output value, and obtaining an evaluation training model based on the initial state output weight includes: The evaluation training model is obtained by the following formula: in, To evaluate the training model for the input pipeline network status characteristics The predicted output value, L is the number of hidden layer nodes, is the output value of the first hidden layer node, is the weight from the first hidden layer node to the output node, is the sum of the products of the output values ​​of all hidden layer nodes and the corresponding weights, For the The output value of each hidden layer node corresponding to the network status feature, Output weights for the initial state.

[0097] It should be noted that It is a weight matrix used to associate the hidden layer output with the final model output. It is used to weight the output of different hidden layer nodes, that is, to multiply the output value of each hidden layer node with the corresponding output weight and sum them to obtain the final model prediction result.

[0098] S35. Testing the evaluation training model based on the prediction test group, obtaining the test results and the corresponding pipe network status labels, and calculating the prediction errors of the test results and the corresponding pipe network status labels.

[0099] Among them, the test result represents the prediction result, that is, whether a leakage occurs, and the corresponding pipeline status label is the pre-marked actual label, which is the pipeline status label; the prediction error represents the difference between the prediction result and the actual label.

[0100] It should be noted that using the prediction test group to test the trained model and obtain the test results can reveal the performance of the model on unseen data and can be used to evaluate the model performance.

[0101] S36. According to the group number sequence, prediction task groups corresponding to different numbers are selected one by one and divided into prediction test groups, and prediction task groups corresponding to the remaining numbers are divided into prediction training groups to obtain multiple prediction errors.

[0102] Specifically, prediction task groups corresponding to different numbers are selected one by one and divided into prediction test groups, and prediction task groups corresponding to the remaining numbers are divided into prediction training groups, that is, in the first round, the prediction task group numbered 1 is selected as the prediction test group, and the remaining prediction task groups of all numbers are prediction training groups. The model is trained based on the prediction training group, and then the trained model is tested based on the prediction test group to obtain the tested model and prediction error. In the second round, the prediction task group numbered 2 is selected as the prediction test group, and the remaining prediction task groups of all numbers (including numbered 1) are prediction training groups. The model after the first round of testing is trained based on the prediction training group, and then the model after the second round of training is tested based on the prediction test group to obtain the tested model and prediction error. Repeat the above steps until all the prediction task groups with different numbers have been used as prediction test groups to verify the model, and multiple prediction errors are obtained.

[0103] S37. Based on multiple prediction errors, the prediction task group corresponding to the maximum prediction error is removed, and the remaining prediction task groups repeat the above training and testing steps until the final hidden layer output value is obtained, and the state output weight is determined according to each hidden layer output value to obtain the pipeline network state assessment model.

[0104] Among them, the maximum prediction error represents the maximum value of all prediction errors; the final hidden layer output value represents the output value of the output layer obtained after all training of the model; and the state output weight represents the output layer weight obtained after all training.

[0105] For example, there are 100 prediction task groups in total, and 100 prediction errors are obtained. When the prediction task group corresponding to No. 50 is used as the prediction test group, the prediction error obtained is the maximum prediction error. Then the prediction task group corresponding to No. 50 is removed, and the remaining 99 prediction task groups are retrained and tested, and then the prediction task group with the largest prediction error is removed until the final pipeline network status assessment model is obtained.

[0106] It should be noted that by gradually removing the task groups with large prediction errors and continuously optimizing the training set and model, the model error can be effectively reduced, the accuracy of its prediction of the pipeline network status can be improved, and ultimately a more reliable evaluation model can be obtained.

[0107] S4. Input the valid pipeline network data into the pipeline network status assessment model for assessment, generate status assessment results, process the abnormal status assessment results based on a preset early warning strategy, generate leakage assessment data, and send the leakage assessment data to the data center.

[0108] Among them, the status assessment result represents the prediction result output by the pipeline network status assessment model, which can be normal, leakage, etc.; the preset early warning strategy represents the pre-set rules for early warning of gas leakage; the abnormal status assessment result means that the prediction result is leakage; the leakage assessment data represents the relevant detailed data when a natural gas leakage is predicted, including the specific location of the gas pipeline network, the severity of the leakage, the possible affected area and other information; the data center represents the information processing core responsible for providing reliable data storage and processing capabilities, which can perform real-time analysis and storage on the leakage assessment data sent by the equipment terminal, and send early warning information to the maintenance terminal, so that the maintenance personnel can repair the gas pipeline network predicted to leak, etc.

[0109] In some other embodiments, the equipment terminal may directly generate warning information based on the leakage assessment data and send it to the maintenance terminal.

[0110] In some embodiments, the method further comprises: The warning processing result sent by the receiving processing end is obtained, and when the warning processing result is a missed alarm or a false alarm, error warning data is obtained.

[0111] Among them, the processing end represents the terminal equipment used by maintenance personnel, that is, the maintenance terminal, such as the mobile phone used by maintenance personnel in the event of a natural gas leak; the early warning processing result represents the feedback data sent by the maintenance personnel after maintenance based on the model prediction results, including but not limited to processed, false alarms, missed alarms, etc.; the error warning data represents the relevant data corresponding to the false alarms and missed alarms, including but not limited to the prediction results, gas pipeline network location, maintenance results, etc.

[0112] The error warning data is sent to the device terminal, and the model adjustment parameters of the pipeline network status assessment model are obtained according to the error warning data.

[0113] Among them, the device terminal is an edge terminal, that is, a device that processes the collected pipeline current data to obtain a prediction result; the model adjustment parameter represents a parameter for adjusting the pipeline status assessment model, such as the number of hidden layer nodes.

[0114] The pipeline network status assessment model is updated based on the model adjustment parameters to obtain an updated pipeline network status assessment model.

[0115] Specifically, the overall architecture of the pipeline network status assessment model is reconstructed based on the model adjustment parameters to make the model's prediction results more accurate, such as reducing the number of hidden layer nodes.

[0116] In this embodiment, a gas pipeline leakage warning method based on edge computing is provided, which is applied to a device terminal, and the method includes: responding to a leakage monitoring request, receiving pipeline current data obtained by an acquisition end, and the pipeline current data includes pressure current data and leakage current data; calling a preset screening model to process the pipeline current data to obtain an abnormal deviation value, and screening the pipeline current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline data; receiving pipeline annotation data, obtaining a hidden layer output value of the pipeline annotation data according to an initial evaluation model, determining a state output weight based on each hidden layer output value, and obtaining a pipeline state evaluation model; inputting the valid pipeline data into the pipeline state evaluation model for evaluation, generating a state evaluation result, processing the abnormal state evaluation result based on a preset warning strategy, generating leakage evaluation data, and sending the leakage evaluation data to a data center. It allows real-time data processing near the data source, thereby significantly reducing data transmission time. By processing data at the device terminal, it can be determined more quickly and accurately whether the gas pipeline network is leaking; by moving the data processing task to the device terminal, most of the data does not need to be uploaded to the data center, which can save a lot of server resources, bandwidth, electricity, and physical space, etc., thereby reducing overall operating costs; by reducing the transmission of data in the network, edge computing reduces the risk of data being intercepted, stolen, or tampered with, and improves the privacy and security of the data; even if the connection with the data center is disconnected, the device terminal can still operate independently and continue to process data, improving the reliability of judging whether the gas pipeline network is leaking, and ensuring that it can continue to work without interruption in the event of a network failure. In summary, the present invention realizes real-time data processing, reduces the time delay of data transmission, improves the real-time, safety and reliability of gas leak monitoring, and also reduces the cost of data transmission.

[0117] like Figure 6 FIG. 1 is a schematic diagram of a gas pipeline network leakage warning system based on edge computing provided by an embodiment of the present invention. The gas pipeline network leakage warning system based on edge computing includes: A data acquisition module, used to respond to a leakage monitoring request and receive the pipe network current data acquired by the acquisition end, wherein the pipe network current data includes pressure current data and leakage current data; An abnormal screening module is used to call a preset screening model to process the pipeline network current data to obtain an abnormal deviation value, and screen the pipeline network current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline network data; A model training module is used to receive pipe network annotation data, obtain hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determine state output weights based on each hidden layer output value, and obtain a pipe network state evaluation model; The status assessment module is used to input the valid pipeline network data into the pipeline network status assessment model for assessment, generate status assessment results, process abnormal status assessment results based on preset early warning strategies, generate leakage assessment data, and send the leakage assessment data to the data center.

[0118] It should be noted that the above-mentioned gas pipeline leakage warning system based on edge computing can execute the gas pipeline leakage warning method based on edge computing provided in the embodiment of the present invention. For technical details and beneficial effects that are not fully described in the embodiment of the gas pipeline leakage warning system based on edge computing, please refer to the gas pipeline leakage warning method based on edge computing provided in the embodiment of the present invention.

[0119] Through the description of the above embodiments, it can be clearly understood by those skilled in the art that each embodiment can be implemented by means of software plus a general hardware platform, or by hardware. It can be understood by those skilled in the art that all or part of the processes in the above embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may also be combined, the steps may be implemented in any order, and there are many other changes in different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A gas pipeline network leakage early warning method based on edge computing, applied to a device terminal, characterized in that: The method comprises: In response to a leakage monitoring request, receiving pipe network current data acquired by a collection end, the pipe network current data including pressure current data and leakage current data; Retrieving a preset screening model to process the pipeline network current data to obtain abnormal deviation values, screening the pipeline network current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline network data; Receiving pipe network annotation data, obtaining hidden layer output values ​​of the pipe network annotation data according to an initial evaluation model, determining state output weights based on each hidden layer output value, and obtaining a pipe network state evaluation model; The valid pipeline network data is input into the pipeline network status assessment model for assessment to generate status assessment results, the abnormal status assessment results are processed based on a preset early warning strategy to generate leakage assessment data, and the leakage assessment data is sent to the data center.

2. The gas pipeline network leakage early warning method based on edge computing according to claim 1 is characterized in that: The data fluctuation threshold includes a pressure fluctuation threshold, calling a preset screening model to process the pipeline network current data to obtain an abnormal deviation value, screening the pipeline network current data based on each abnormal deviation value and the corresponding data fluctuation threshold, and obtaining valid pipeline network data, including: Retrieving a preset screening model to calculate the pressure current data to obtain a pressure difference reference value and a pressure difference fluctuation value; Determine the pressure deviation value corresponding to each pressure current data according to the pressure difference reference value and the pressure difference fluctuation value; determining a pressure fluctuation range based on the pressure fluctuation threshold, and determining the pressure current data corresponding to the pressure deviation value as abnormal pressure data when the pressure deviation value is not within the pressure fluctuation range; The abnormal pressure data in the pressure current data are removed to obtain pressure retention current data, and the pressure retention current data are determined as effective pipe network pressure data.

3. The gas pipeline network leakage early warning method based on edge computing according to claim 2 is characterized in that: The data fluctuation threshold also includes a leakage fluctuation threshold. The preset screening model is called to process the pipeline network current data to obtain an abnormal deviation value. The pipeline network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold to obtain valid pipeline network data, including: Calculating the leakage current data according to the preset screening model to obtain a leakage baseline amount and a leakage fluctuation amount; Determine the leakage deviation value corresponding to each leakage current data according to the leakage reference value and the leakage fluctuation value; determining a leakage fluctuation range based on the leakage fluctuation threshold, and determining leakage current data corresponding to the leakage deviation value as abnormal leakage data when the leakage deviation value is not within the leakage fluctuation range; The abnormal leakage data in the leakage current data are removed to obtain leakage retention current data, and the leakage retention current data are determined as valid pipe network leakage data.

4. The gas pipeline network leakage early warning method based on edge computing according to claim 1 is characterized in that: The pipeline network current data is screened based on each abnormal deviation value and the corresponding data fluctuation threshold value, and before obtaining valid pipeline network data, the method further includes: Receiving historical fluctuation data, determining an initial fluctuation threshold according to the historical fluctuation data, and retrieving a threshold adjustment strategy to obtain pipe network operation data; Acquire operation fluctuation characteristics according to the pipeline network operation data, retrieve the threshold adjustment table to match each operation fluctuation characteristic, and obtain fluctuation adjustment values, wherein the fluctuation adjustment values ​​include seasonal adjustment values, operation adjustment values, and performance adjustment values; A comprehensive adjustment value is obtained according to the season adjustment value and / or the operation adjustment value and / or the performance adjustment value, and the initial fluctuation threshold is updated based on the comprehensive adjustment value to obtain a data fluctuation threshold.

5. The gas pipeline network leakage early warning method based on edge computing according to claim 1 is characterized in that: Receiving pipe network annotation data, obtaining hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determining state output weights based on each hidden layer output value, and obtaining a pipe network state evaluation model, including: Receiving pipe network annotation data, screening the pipe network annotation data according to the preset screening model, and obtaining pipe network training data; Identify the pipeline network training data, obtain the pipeline network state characteristics and corresponding pipeline network state labels, and determine the pipeline network state characteristics and corresponding pipeline network state labels as a prediction task group; Calling a preset numbering rule to number each prediction task group, obtaining a group numbering sequence, and dividing the prediction task group corresponding to the initial number into prediction test groups based on the group numbering sequence; The prediction task groups corresponding to the remaining numbers are divided into prediction training groups, and the initial evaluation model is trained based on each prediction training group to obtain the initial hidden layer output value, and the initial state output weight is determined according to each initial hidden layer output value to obtain the evaluation training model; Testing the evaluation training model based on the prediction test group, obtaining the test results and the corresponding pipe network status labels, and calculating the prediction errors of the test results and the corresponding pipe network status labels; According to the group number sequence, prediction task groups corresponding to different numbers are selected one by one and divided into prediction test groups, and prediction task groups corresponding to the remaining numbers are divided into prediction training groups, to obtain multiple prediction errors; Based on multiple prediction errors, the prediction task group corresponding to the maximum prediction error is removed, and the remaining prediction task groups repeat the above training and testing steps until the final hidden layer output value is obtained. The state output weight is determined according to each hidden layer output value to obtain the pipeline network state assessment model.

6. The gas pipeline network leakage early warning method based on edge computing according to claim 5 is characterized in that: The initial evaluation model is trained based on each prediction training group to obtain the initial hidden layer output value, and the initial state output weight is determined according to each initial hidden layer output value to obtain the evaluation training model, including: Retrieving a basic evaluation model, configuring the number of input nodes, the number of output nodes, and the number of hidden layer nodes of the basic evaluation model, and obtaining an initial evaluation model; Determine the activation function of each hidden layer node according to the initial evaluation model, and randomly initialize the weight and bias of each hidden layer node; Calculating the pipe network state characteristics of the prediction training group based on the weight, the bias and the activation function to obtain an initial hidden layer output value; The initial state output weights are obtained according to the initial hidden layer output values, and the evaluation training model is obtained based on the initial state output weights.

7. The gas pipeline network leakage early warning method based on edge computing according to claim 6 is characterized in that: The pipe network state characteristics of the prediction training group are calculated based on the weight, the bias and the activation function to obtain an initial hidden layer output value, including: The network state characteristics of the prediction training group are calculated by the following formula to obtain the initial hidden layer output value: in, is the output value of the first hidden layer node, is the activation function, 1 is the weight from the input node to the first hidden layer node, is the bias of the first hidden layer node, For the The input network status characteristics, Indicates transpose.

8. The gas pipeline network leakage early warning method based on edge computing according to claim 7 is characterized in that: Obtaining initial state output weights according to each initial hidden layer output value, and obtaining an evaluation training model based on the initial state output weights, including: The evaluation training model is obtained by the following formula: in, To evaluate the training model for the input pipeline network status characteristics The predicted output value, L is the number of hidden layer nodes, is the output value of the first hidden layer node, is the weight from the first hidden layer node to the output node, is the sum of the products of the output values ​​of all hidden layer nodes and the corresponding weights, For the The output value of each hidden layer node corresponding to the network status feature, Output weights for the initial state.

9. The gas pipeline network leakage early warning method based on edge computing according to claim 1 is characterized in that: The method further comprises: Receive the warning processing result sent by the processing end, and when the warning processing result is a missed alarm or a false alarm, obtain error warning data; Sending the error warning data to the device terminal, and obtaining the model adjustment parameters of the pipeline network status assessment model according to the error warning data; The pipeline network status assessment model is updated based on the model adjustment parameters to obtain an updated pipeline network status assessment model.

10. A gas pipeline network leakage warning system based on edge computing, characterized in that: The system comprises: A data acquisition module, used to respond to a leakage monitoring request and receive the pipe network current data acquired by the acquisition end, wherein the pipe network current data includes pressure current data and leakage current data; An abnormal screening module is used to call a preset screening model to process the pipeline network current data to obtain an abnormal deviation value, and screen the pipeline network current data based on each abnormal deviation value and a corresponding data fluctuation threshold to obtain valid pipeline network data; A model training module is used to receive pipe network annotation data, obtain hidden layer output values ​​of the pipe network annotation data according to the initial evaluation model, determine state output weights based on each hidden layer output value, and obtain a pipe network state evaluation model; The status assessment module is used to input the valid pipeline network data into the pipeline network status assessment model for assessment, generate status assessment results, process abnormal status assessment results based on preset early warning strategies, generate leakage assessment data, and send the leakage assessment data to the data center.

Images