A method and system for detecting meter failure

Abstract

The application discloses a meter fault detection method and system, comprising: collecting meter data and extracting feature data; constructing a parameter threshold library to identify the feature data, and generating an abnormal monitoring index; constructing an expert knowledge base to classify the abnormal monitoring index, and determining the category of the abnormal monitoring index, including: an immediate fault index type and a predictive maintenance index type; based on the expert knowledge base, the abnormal monitoring index in the predictive maintenance index type and / or the combination of the abnormal monitoring index is subjected to index reconstruction to generate a fault prediction index, including a predicted fault type and a fault prediction probability; based on the abnormal monitoring index of the immediate fault index type and / or the fault prediction index, a corresponding maintenance suggestion is provided. The application can distinguish the abnormal monitoring index, realize the prediction of the fault type and the fault occurrence time, improve the utilization efficiency of maintenance resources, and reduce the probability of meter unplanned downtime and maintenance cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of meter detection, and in particular to a meter fault detection method and system. BACKGROUND

[0002] With the increasing demand for natural gas energy, abnormal detection and fault diagnosis of meters have become an important part of intelligent management. However, although the remote meter has the function of reporting abnormalities, the reporting indicators are relatively single, and misjudgment, false reporting and other phenomena are prone to occur, resulting in waste of maintenance resources and inability to timely handle meter abnormalities. Therefore, how to accurately analyze meter abnormalities and make corresponding treatment has become a technical problem to be solved and a focus of research for those skilled in the art. SUMMARY

[0003] In order to solve the problems in the prior art, the present application adopts the following technical solutions:

[0004] In a first aspect, the present application provides a meter fault detection method, which comprises the following steps:

[0005] Collecting multi-dimensional data of the meter, and extracting a plurality of feature data reflecting the running state of the meter from the meter data through feature engineering;

[0006] Constructing a parameter threshold library, and performing abnormal identification on each of the feature data based on the parameter threshold library to generate abnormal monitoring indicators;

[0007] Constructing an expert knowledge base, and classifying the abnormal monitoring indicators based on the expert knowledge base to determine the categories of the abnormal monitoring indicators, including immediate fault indicator types and predictive maintenance indicator types;

[0008] Based on the expert knowledge base, the abnormal monitoring indicators belonging to the predictive maintenance indicator types and / or combinations of the abnormal monitoring indicators are reconstructed to generate fault prediction indicators, which are configured to identify predicted fault types and fault prediction probabilities;

[0009] Based on the abnormal monitoring indicators of the immediate fault indicator types and / or the fault prediction indicators, corresponding maintenance suggestions are provided.

[0010] In summary, the meter fault detection method provided in the application classifies abnormal monitoring indicators based on an expert knowledge base, distinguishes between immediate fault indicator types and predictive maintenance indicator types, and reconstructs abnormal monitoring indicators belonging to the predictive maintenance indicator type, predicts the fault type and fault occurrence time corresponding to the abnormal monitoring indicators belonging to the predictive maintenance indicator type, thereby providing sufficient windows for planned maintenance, avoiding treating all meter abnormalities as immediate faults, improving the utilization efficiency of maintenance resources, and reducing the probability of non-planned shutdown of the meter and maintenance costs.

[0011] Further, the classification of the abnormal monitoring indicators comprises:

[0012] Based on the expert knowledge base, historical abnormal monitoring indicator data that has been classified is obtained;

[0013] The binary classification model is trained using the historical abnormal monitoring indicator data;

[0014] The unclassified abnormal monitoring indicators are classified by the trained binary classification model to determine the category to which the abnormal monitoring indicators belong.

[0015] Further, the meter fault detection method further comprises:

[0016] Based on the principle of a large language model, an abnormal combination model and a fault prediction model are constructed;

[0017] The abnormal combination model is used to obtain the association strength between the abnormal monitoring indicators in the predictive maintenance indicator type and the fault types, and / or the association strength between the combination of the abnormal monitoring indicators in the predictive maintenance indicator type and the fault types;

[0018] Based on the association strength, the abnormal monitoring indicators in the predictive maintenance indicator type and / or the combination of the abnormal monitoring indicators are predicted by the fault prediction model to obtain the predicted fault type and fault prediction probability.

[0019] Further, the meter fault detection method further comprises:

[0020] The abnormal combination model, based on the expert classification rules, meter fault types, abnormal performance combinations, meter degradation processes, and handling records of faults that have occurred stored in the expert knowledge base, uses the prompt word mode to:

[0021] Establish and output the association strength between the abnormal monitoring indicators in the predictive maintenance indicator type and the fault types, and / or the association strength between the combination of the abnormal monitoring indicators in the predictive maintenance indicator type and the fault types.

[0022] Further, the meter fault detection method further comprises:

[0023] The fault prediction model, by means of prompt words, is based on the expert classification rules, meter fault types, abnormal performance combinations, meter deterioration processes and handling records of faults that have occurred stored in the expert knowledge base, and combines the correlation strength,

[0024] The abnormal monitoring indicators in the predictive maintenance indicator type and / or combinations of the abnormal monitoring indicators are predicted, and the predicted fault type and the fault prediction probability are output.

[0025] Further, the expert classification rules are generated based on expert experience and data analysis summary, and the expert classification rules are used to provide a basis for judgment when classifying the abnormal monitoring indicators, and assist in determining whether the abnormal monitoring indicators belong to the immediate fault indicator type or the predictive maintenance indicator type.

[0026] Further, the meter fault detection method further comprises:

[0027] Based on the expert knowledge base, maintenance experience about fault types, repair processes and component replacement cycles is obtained;

[0028] According to the abnormal monitoring indicators of the immediate fault indicator type and / or the fault types of the fault prediction indicators, the corresponding maintenance experience is searched in the expert knowledge base to generate the maintenance suggestion.

[0029] Further, the meter fault detection method further comprises:

[0030] Based on the fault-free operation period of the meter, statistical values of meter parameters are calculated, the statistical values including one or more of the following types of statistical values: mean, variance, quantile and standard deviation of the meter parameters;

[0031] At least one statistical value is used to construct an abnormality judgment threshold value for different kinds of meter parameters, and the abnormality judgment threshold value is stored in the parameter threshold value library, and the parameter threshold value library dynamically updates the abnormality judgment threshold value;

[0032] The abnormality judgment threshold value is obtained from the parameter threshold value library, the feature data is compared with the abnormality judgment threshold value to determine whether the feature data is abnormal, and according to the determination result, if there is an abnormality, the abnormal monitoring indicator is generated.

[0033] In a second aspect, the application also provides a meter fault detection system, which comprises:

[0034] The data acquisition module is configured to acquire multi-dimensional data of the meter and pre-process the acquired data.

[0035] The feature engineering module is configured to extract features from the acquired data to obtain feature data reflecting the running state of the meter.

[0036] The anomaly monitoring module is configured to include a parameter threshold library, and generate anomaly monitoring indicators by identifying anomalies in the feature data based on the parameter threshold library.

[0037] The expert knowledge base module is configured to store meter fault types, meter degradation processes, anomaly performance combinations, expert classification rules, handling records of faults that have occurred, and maintenance experience.

[0038] The anomaly classification module is configured to classify the anomaly monitoring indicators based on the expert knowledge base module, determine the categories of the anomaly monitoring indicators, and the categories include immediate fault indicator types and predictive maintenance indicator types.

[0039] The fault prediction module is configured to reconstruct indicators based on the expert knowledge base, the anomaly monitoring indicators belonging to the predictive maintenance indicator types and / or combinations of the anomaly monitoring indicators, and generate fault prediction indicators configured to identify predicted fault types and fault prediction probabilities.

[0040] The result output module is configured to output corresponding maintenance recommendations based on the anomaly monitoring indicators of the immediate fault indicator types and / or the fault prediction indicators.

[0041] Further, the fault prediction module is further configured to:

[0042] train a large language model based on the expert knowledge base module using the large language model;

[0043] construct an anomaly combination model and a fault prediction model based on the trained large language model;

[0044] input the anomaly monitoring indicators of the predictive maintenance indicator types and / or combinations of the anomaly monitoring indicators into the anomaly combination model to generate the association strength between the anomaly monitoring indicators and the fault types, and / or generate the association strength between the combinations of the anomaly monitoring indicators and the fault types;

[0045] The abnormal monitoring indicators in the predictive maintenance indicator type and / or the combination of the abnormal monitoring indicators are input into the fault prediction model, fault prediction is performed on the abnormal monitoring indicators in the predictive maintenance indicator type and / or the combination of the abnormal monitoring indicators, and the predicted fault type and the fault prediction probability are obtained. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A step flowchart of a meter fault detection method provided by an embodiment of the present application is provided.

[0047] Figure 2 A step flowchart of abnormality identification of feature data in a meter fault detection method provided by an embodiment of the present application is provided.

[0048] Figure 3 A step flowchart of classification of abnormal monitoring indicators in a meter fault detection method provided by an embodiment of the present application is provided.

[0049] Figure 4 A step flowchart of indicator reconstruction of abnormal monitoring indicators in a meter fault detection method provided by an embodiment of the present application is provided.

[0050] Figure 5 A step flowchart of generation of a maintenance suggestion in a meter fault detection method provided by an embodiment of the present application is provided.

[0051] Figure 6 A structural schematic diagram of a meter fault detection system provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0052] The present application will be described in detail below with reference to the specific embodiments shown in the drawings, but these embodiments do not limit the present application, and the structural, method, or functional changes made by those of ordinary skill in the art based on these embodiments are included in the protection scope of the present application.

[0053] To solve the problems in the prior art, in a first aspect, an embodiment of the present application provides a meter fault detection method, which can be applied to a gas meter, and monitors and analyzes the temperature, pressure, flow, and gas consumption data of the gas meter during operation to detect whether the meter has an abnormal fault. Figure 1 As shown in the figure, the meter fault detection method includes the following steps:

[0054] In step S101, multi-dimensional data of the meter is collected, and multiple feature data reflecting the running state of the meter are extracted from the meter data through feature engineering.

[0055] Step S102, a parameter threshold library is constructed, and each feature data is identified based on the parameter threshold library to generate an abnormal monitoring index.

[0056] Step S103, an expert knowledge base is constructed, and the abnormal monitoring index is classified based on the expert knowledge base to determine the category of the abnormal monitoring index, including: an immediate fault index type and a predictive maintenance index type.

[0057] Step S104, based on the expert knowledge base, the abnormal monitoring index and / or the combination of the abnormal monitoring index belonging to the predictive maintenance index type are reconfigured to generate a fault prediction index, and the fault prediction index is configured to identify the predicted fault type and the fault prediction probability.

[0058] Step S105, based on the abnormal monitoring index of the immediate fault index type and / or the fault prediction index, a corresponding maintenance suggestion is provided.

[0059] Specifically, data collection is carried out for the gas meter, and the collected data covers the original data of the meter operation (such as flow, pressure, temperature, cumulative reading, battery status, alarm code, etc.) and external data related to the meter (such as local ambient temperature, user water history data, user electricity history data, etc.). The collected meter data is preprocessed, including data cleaning, denoising, format conversion, time synchronization, missing value filling, etc., to generate structured preprocessed data. Further, feature extraction is performed on the preprocessed data of the meter to obtain feature data that can reflect the running state of the meter, including gas consumption, temperature, pressure, flow, etc.

[0060] An abnormality judgment threshold of the meter parameter is set, and the abnormality judgment threshold is stored in the parameter threshold library, which is configured to dynamically update the abnormality judgment threshold based on the collected meter data. Each feature data is identified based on the abnormality judgment threshold in the parameter threshold library, the size of the feature data and the abnormality judgment threshold is judged, and according to the judgment result, if there is an abnormal phenomenon, an abnormal monitoring index is generated. In order to further illustrate the abnormal monitoring index provided by the embodiments of the present application, an example of the abnormal monitoring index is provided as follows: based on the collected meter data, the temperature standard deviation corresponding to each meter is calculated, and according to the normal distribution principle, three times of the meter temperature standard deviation is taken as the abnormality judgment threshold, and the temperature fluctuation of each meter is independently judged; if the meter temperature is within the abnormality judgment threshold, it is judged that the temperature fluctuation is normal, and if the meter temperature exceeds the abnormality judgment threshold, it is judged that the temperature fluctuation is abnormal, and an abnormal monitoring index is generated. Exemplarily, the abnormal monitoring index includes: meter id, abnormal condition description, abnormal occurrence time, etc.

[0061] ​In step S103, an expert knowledge base is constructed based on domain expert knowledge, historical data analysis, and actual maintenance experience, etc. The domain knowledge related to meter fault diagnosis and prediction is stored in the expert knowledge base, which is used for classifying abnormal monitoring indicators and predicting meter faults. Based on the expert knowledge base, the generated abnormal monitoring indicators are classified to determine whether the abnormal monitoring indicators belong to the immediate fault indicator type or the predictive maintenance indicator type. The immediate fault indicator type corresponds to abnormal monitoring indicators indicating that the meter has an immediate fault, and the target meter needs to be processed in priority. The predictive maintenance indicator type corresponds to early warning signals of faults, which indicates that the meter may fail in the future, and provides a basis for preventive maintenance of the meter.

[0062] For abnormal monitoring indicators belonging to the predictive maintenance indicator type, in step S104, based on the expert knowledge base, the correlation strength between the abnormal monitoring indicators in the predictive maintenance indicator type and the meter fault types is analyzed, and / or the correlation strength between the combination of the abnormal monitoring indicators in the predictive maintenance indicator type and the meter fault types is analyzed. The abnormal monitoring indicators in the predictive maintenance indicator type are predicted, the abnormal monitoring indicators are converted into specific fault prediction, the fault prediction indicators are generated, and the predicted fault types and fault prediction probabilities of the abnormal monitoring indicators in the predictive maintenance indicator type are obtained. For example, the abnormal monitoring indicator in the predictive maintenance indicator type is "pressure fluctuation anomaly", and through prediction analysis, the fault prediction indicator is "predicted to occur valve jamming fault in a days, with a probability of b%".

[0063] When the abnormal monitoring indicators of the immediate fault indicator type are obtained, it indicates that the meter has an immediate fault. At this time, the repair experience corresponding to the fault type is searched in the expert knowledge base, including repair process, component replacement suggestion, etc. For the fault prediction indicator, the corresponding repair experience and maintenance strategy are searched combined with the prediction fault type, prediction probability, etc. Then, the information corresponding to the immediate fault indicator and the fault prediction indicator is integrated, and specific maintenance suggestions are generated for different fault conditions or potential risks. The maintenance suggestions include real-time fault alarm, fault diagnosis report, predictive maintenance suggestion, repair work order or dispatching instruction, etc. The maintenance personnel can ensure that the meter is maintained and processed in time and accurately according to the suggestions, improve the reliability and operation efficiency of the meter, and reduce the maintenance cost and safety risk.

[0064] According to the above description, the meter fault detection method provided by the application classifies the abnormal monitoring indicators based on the expert knowledge base, distinguishes between the instant fault indicator type and the predictive maintenance indicator type, and reconstructs the abnormal monitoring indicators belonging to the predictive maintenance indicator type, predicts the fault type and fault occurrence time corresponding to the abnormal monitoring indicators belonging to the predictive maintenance indicator type, thereby providing sufficient window for planned maintenance, avoiding treating all meter abnormalities as instant faults, improving the utilization efficiency of maintenance resources, and reducing the probability of non-planned shutdown of the meter and maintenance cost.

[0065] As an optional implementation manner, in step S101, feature calculation is performed on the collected meter data, including one or more of the following type feature calculation: for the pressure data during gas use, searching for data with zero flow within a day, judging the proportion of the number of pressure values less than the minimum threshold value set by the meter to the number of data, eliminating interference data during non-gas use period and non-gas use period of more than half a day, screening pressure data of the effective gas use period, calculating the daily average value of pressure during gas use, and calculating the daily standard deviation of pressure during gas use, to reflect the overall level and fluctuation amplitude of pressure during daily gas use. At the same time, for the all-weather pressure data, the average value and standard deviation of pressure are counted by day or month without distinguishing whether gas is used, for monitoring the pressure stability of the meter in the whole period.

[0066] For the gas consumption and flow data, the average value and standard deviation are calculated by day and month, respectively, to obtain the daily average value and daily standard deviation of gas consumption, the daily average value and daily standard deviation of flow, the monthly average value and monthly standard deviation of gas consumption, and the monthly average value and monthly standard deviation of flow. The daily average value of gas consumption and flow can reflect the use scale of daily gas consumption and flow, and the standard deviation of gas consumption and flow can reflect the daily fluctuation degree; the monthly dimension statistics can capture long-term trends.

[0067] Further, based on the collected meter data, the time non-uniformity coefficient of the user can be calculated. The time non-uniformity coefficient is the ratio of the hourly gas consumption to the daily average gas consumption. Specifically, the time non-uniformity coefficient can be calculated by the following formula:

[0068] ;

[0069] Through the above formula, the calculation of the time non-uniformity coefficient can eliminate the influence of dimension, reflect the regular change of user gas data, and be beneficial to the diagnosis of user gas data.

[0070] As an optional implementation manner, as shown in Figure 2 the abnormality recognition of the feature data in step S102 further includes the following steps:

[0071] Step S201, based on the fault-free running period of the meter, calculate the statistical value of the meter parameter, the statistical value is one or more of the following types of statistical values: including the mean, variance, quantile and standard deviation of the meter parameter.

[0072] Step S202, based on at least one statistical value, respectively construct different kinds of meter parameter abnormality judgment threshold, the abnormality judgment threshold is stored in the parameter threshold library, and the parameter threshold library dynamically updates the abnormality judgment threshold.

[0073] Step S203, obtain the abnormality judgment threshold from the parameter threshold library, compare the feature data with the abnormality judgment threshold, judge whether the feature data is abnormal, and generate an abnormality monitoring index according to the judgment result.

[0074] Specifically, according to the historical maintenance record and known failure period of the meter, the normal running period of the meter is determined by reverse deduction, the data of each meter parameter (temperature, pressure, flow, gas consumption, etc.) of a specific meter, a group of meters of the same type or a group of meters in the same area in the normal running period is analyzed, and the statistical value of each meter parameter is calculated, which includes one or more of the following data: mean, variance, quantile, standard deviation, etc. Based on the statistical value of each meter parameter, the abnormality judgment threshold is set, and different types of statistical values can be used to construct the abnormality judgment threshold for different meter parameters. For example, for the temperature of the meter, the variance value can be used to construct the abnormality judgment threshold; for the pressure of the meter, the mean value can be used to construct the abnormality judgment threshold. Alternatively, in an optional embodiment, three standard deviations can also be used as the abnormality judgment threshold, if the feature data of the meter exceeds three standard deviations, it is determined that there is an abnormality; if the feature data of the meter is within three standard deviations, it is determined to be normal. Further, one or more statistical values can be used to construct the abnormality judgment threshold of the meter parameter, for example, for the flow of the meter, the variance value and the standard deviation value can be combined to construct the abnormality judgment threshold.

[0075] The abnormality judgment threshold is stored in the parameter threshold library, and the parameter threshold library can update the abnormality judgment threshold at a predetermined period or irregularly to improve the accuracy of abnormality judgment.

[0076] Further, based on the comparison between the abnormality judgment threshold and the feature data, it is judged whether the feature data is abnormal, and an abnormality monitoring index is generated according to the judgment result. In the embodiments of the present application, the feature data of the meter is subjected to abnormality identification, which includes at least temperature abnormality identification, pressure abnormality identification, flow abnormality identification and gas abnormality identification, which will be further described below.

[0077] The temperature anomaly identification specifically includes: setting upper and lower thresholds of temperature according to the installation position of the meter, performing anomaly determination based on the collected meter data, and generating an abnormal monitoring index representing temperature exceeding the threshold if the temperature exceeds the set upper and lower thresholds; setting the number of data in which the temperature of each meter does not change based on the collected frequency and different time granularities, performing independent determination on each meter, and generating an abnormal monitoring index representing the temperature of the meter not changing if the number of data in which the temperature of the meter does not change is greater than the set number of data. The temperature anomaly identification specifically includes: setting upper and lower thresholds of temperature according to the installation position of the meter, performing anomaly determination based on the collected meter data, and generating an abnormal monitoring index representing temperature exceeding the threshold if the temperature exceeds the set upper and lower thresholds; setting the number of data in which the temperature of each meter does not change based on the collected frequency and different time granularities, performing independent determination on each meter, and generating an abnormal monitoring index representing the temperature of the meter not changing if the number of data in which the temperature of the meter does not change is greater than the set number of data.

[0078] The flow anomaly identification specifically includes: setting the number of data below the minimum measurement threshold of the meter based on the collected meter data, generating an abnormal monitoring index representing micro-flow if the number of data below the minimum measurement threshold of the meter is greater than the set number of data, and it should be noted that the minimum measurement threshold corresponding to different types of meters is different; setting the number of data in which the flow of each meter does not change based on the collected frequency and different time granularities, performing independent determination on each meter, and generating an abnormal monitoring index representing the flow of the meter not changing if the number of data in which the flow of the meter does not change is greater than the set number of data.

[0079] The pressure anomaly identification includes: setting upper and lower thresholds of pressure according to the installation position of the meter, performing anomaly determination based on the collected pressure data, and generating an abnormal monitoring index representing pressure exceeding the threshold if the pressure exceeds the set upper and lower thresholds; setting the number of data in which the pressure of each meter does not change based on the collected frequency and different time granularities, performing independent determination on each meter, and generating an abnormal monitoring index representing the pressure of the meter not changing if the number of data in which the pressure of the meter does not change is greater than the set number of data; based on the pressure daily dimension data during gas use, calculating the abnormal upper and lower limits within the month by the box plot method, performing anomaly determination according to the number of abnormal value data, and generating an abnormal monitoring index representing pressure fluctuation anomaly if the number of pressure fluctuation anomalies is greater than the set number.

[0080] Further, the pressure anomaly identification further includes determining whether there is pressure jump, specifically:

[0081] Based on the pressure daily dimension data during gas use, the correlation coefficient expression of the original sequence and the jump sequence is used to construct the relationship between the difference degrees of the means before and after the variation, and the specific calculation formula is as follows:

[0082] The original sequence pressure daily average data is marked as , and assuming that there is a mutation point in the original sequence, the original sequence is divided into two ends, the number of data in the front section is , and the length of the data in the rear section is , then the overall mean of the original sequence is

[0083] ;

[0084] In the formula, the overall mean of the original sequence is represented.

[0085] The mean of the original sequence before the jump is marked as

[0086] ;

[0087] In the formula, the mean of the original sequence before the jump is represented.

[0088] The mean of the original sequence after the jump is marked as

[0089] ;

[0090] In the formula, the mean of the original sequence after the jump is represented.

[0091] Another new sequence is constructed, which is the sequence composed of the points before and after the jump of the original sequence, called the jump sequence , and

[0092] ;

[0093] ;

[0094] The overall mean of the jump sequence is

[0095] ;

[0096] In the formula, the mean of the original sequence before the jump is represented, the mean of the original sequence after the jump is represented, and the overall mean of the jump sequence is represented.

[0097] The original sequence X and the jump sequence Y have correlation, and the correlation coefficient calculation formula is

[0098] ;

[0099] In the formula, the overall mean of the original sequence is represented, and the overall mean of the jump sequence is represented.

[0100] The , The variable is brought into the above formula, and the following can be obtained:

[0101] ;

[0102] Wherein, , the correlation coefficient is positive or negative, and when it is positive, the original sequence is proportional to the jump sequence, indicating that the sequence has upward jump variation; otherwise, the sequence has downward jump variation, and the amplitude of the jump is determined by controlling the absolute value of the correlation coefficient.

[0103] The gas anomaly identification includes: based on the collected meter data, if the daily gas consumption exceeds three times the standard deviation of any month gas consumption in the past year, an abnormal monitoring index representing sudden increase of gas consumption is generated; based on the collected meter data, if the monthly gas consumption is zero, an abnormal monitoring index representing no gas use in the month is generated; the cumulative reading data of the user meter is obtained, the first data of today is selected, and it is judged whether all are less than the first cumulative reading data of the day before yesterday, and an abnormal monitoring index representing abnormal decrease of reading is generated according to the judgment result.

[0104] Further, the gas anomaly identification further includes: based on the collected meter data, the daily gas consumption data is obtained, the regression equation of the daily gas consumption data is calculated, and the slope value of the daily gas consumption data is obtained, and the calculation formula is as follows:

[0105] ;

[0106] In the formula, represents the daily gas consumption data, x is the time, is the equation coefficient. The regression equation of the user gas and other indicators (water consumption data or electricity consumption data) is calculated respectively, if the two slope values are negatively correlated, it indicates that the gas consumption is abnormal, and an abnormal monitoring index representing abnormal trend of gas consumption is generated.

[0107] Further, the gas anomaly identification further includes: based on the collected meter data, the hourly uneven coefficient and other characteristic values are calculated, and based on the least square algorithm similarity measurement, the similarity of the daily hourly consumption mode (or the hourly uneven coefficient sequence) and the historical typical mode is calculated. In addition, the similarity calculation of the local daily temperature data and the daily gas consumption data is combined, and compared with the knowledge about the gas consumption characteristics of heating / non-heating users, whether the user is a heating user is judged, and the heating month and the non-heating month are divided according to the user type (heating / non-heating) to carry out separate regularity analysis or similarity calculation, so as to accurately judge whether the gas consumption rule is abnormal, and an abnormal monitoring index representing abnormal gas consumption rule is generated according to the judgment result.

[0108] The abnormality judgment threshold is dynamically constructed to identify the abnormality of the meter data, thereby improving the accuracy and adaptability of the meter fault detection. The abnormality judgment threshold can be updated according to the running state of the meter and the change of the environment, thereby avoiding the misjudgment caused by the fixed threshold due to the scene difference, and especially adapting to the running characteristics of the meter at different pipe network positions and altitudes. Moreover, the abnormality monitoring index generated by comparing the feature data with the abnormality judgment threshold can capture the abnormality of the meter from multiple dimensions, cover the immediate fault signal and the progressive abnormality trend, reduce the risk of misjudgment caused by the dependence on artificial experience, and provide a standardized data basis for subsequent fault type positioning and predictive maintenance index generation through systematic threshold management and abnormality classification, thereby effectively improving the efficiency and reliability of the intelligent operation and maintenance of the gas meter.

[0109] As an optional implementation manner, as shown in FIG. 10, in step S103, after the abnormality monitoring index is generated, the classification of the abnormality monitoring index further includes the following steps: Figure 3

[0110] Step S301: Based on the expert knowledge base, historical abnormality monitoring index data that has been classified is obtained.

[0111] Step S302: The binary classification model is trained using the historical abnormality monitoring index data.

[0112] Step S303: The unclassified abnormality monitoring index is classified by using the trained binary classification model, and the category to which the abnormality monitoring index belongs is determined.

[0113] Specifically, the expert knowledge base stores the domain knowledge related to the meter fault diagnosis, which includes the historical abnormality monitoring index data that has been classified by the domain experts or according to the historical maintenance records. These data are accumulated for a long time, cover the abnormality monitoring index under different meter models, running environments and fault scenes, and each abnormality monitoring index is clearly labeled as “immediate fault indicator type” or “predictive maintenance indicator type”. The historical abnormality monitoring index data that has been classified is obtained from the expert knowledge base, thereby forming a targeted training data set and providing reliable labeled data support for subsequent model training.

[0114] ​The classified historical abnormal monitoring indicator data is input into a binary classification model as a training data set. The binary classification model is trained with the instant failure indicator type and the predictive maintenance indicator type as class labels to learn the association rules between different abnormal monitoring indicators in the historical abnormal monitoring indicator data and the class labels, so as to improve the classification accuracy. Optionally, the binary classification model can be a support vector machine (SVM) classification model. The unclassified abnormal monitoring indicators are classified based on the trained binary classification model. The unclassified abnormal monitoring indicators and related gauge types, maintenance records and other information are input into the binary classification model. The binary classification model outputs the probability that the abnormal monitoring indicators belong to the instant failure indicator type or the predictive maintenance indicator type based on the classification rules learned in the training stage. Alternatively, the binary classification model can be configured to directly output the classification results.

[0115] The binary classification model is trained based on the historical abnormal monitoring indicator data accumulated in the expert knowledge base. The binary classification model is used to classify the abnormal monitoring indicators, which realizes the automation of the classification process, avoids the classification bias caused by relying solely on human experience, reduces the misjudgment and omission in complex abnormal scenarios, and improves the accuracy and efficiency of the classification of abnormal monitoring indicators. Moreover, by classifying the abnormal monitoring indicators into instant failure indicators and predictive maintenance indicators, the utilization rate of maintenance resources is improved by avoiding treating all abnormalities as instant failures.

[0116] As an optional implementation manner, as shown in FIG. 10, in step S104, the abnormal monitoring indicators belonging to the predictive maintenance indicator type and / or the combination of the abnormal monitoring indicators are subjected to indicator reconstruction, which further includes the following steps: Figure 4

[0117] In step S401, based on the principle of a large language model, an abnormal combination model and a failure prediction model are respectively constructed.

[0118] In step S402, the abnormal combination model is used to obtain the association strength between the abnormal monitoring indicators in the predictive maintenance indicator type and the failure types, and / or the association strength between the combination of the abnormal monitoring indicators in the predictive maintenance indicator type and the failure types.

[0119] In step S403, based on the association strength, the abnormal monitoring indicators in the predictive maintenance indicator type and / or the combination of the abnormal monitoring indicators are predicted by the failure prediction model to obtain the predicted failure types and failure prediction probabilities.

[0120] ​Specifically, by using the learning ability of the large language model for meter failure related knowledge, an abnormal combination model and a fault prediction model are respectively constructed. The abnormal combination model focuses on analyzing the correlation strength between abnormal monitoring indicators (or combinations of abnormal monitoring indicators) and fault types. The fault prediction model focuses on further predicting the specific fault type and the probability of occurrence in combination with the correlation strength.

[0121] Through the abnormal combination model, the correlation strength between abnormal monitoring indicators and fault types is quantified. Single abnormal monitoring indicators (such as "pressure fluctuation anomaly") or combinations of abnormal monitoring indicators (such as "pressure fluctuation anomaly + flow invariance") belonging to the predictive maintenance indicator type are input into the abnormal combination model. The abnormal combination model analyzes the frequency of occurrence of abnormal monitoring indicators or combinations of abnormal monitoring indicators in historical fault cases, thereby obtaining the correlation strength between abnormal monitoring indicators in the predictive maintenance indicator type and fault types, and / or obtaining the correlation strength between combinations of abnormal monitoring indicators in the predictive maintenance indicator type and fault types. For example, there are 5 cases of gas stealing in the historical fault case library, 4 of which are discovered to be "gas stealing" through gas usage trend and gas usage regularity. Therefore, the two indicators of "gas usage trend" and "gas usage regularity" are associated with the fault prediction indicator "gas stealing", and the correlation strength is calculated.

[0122] Further, the correlation strength can be calculated in the following way: all cases with the target fault type are filtered from the historical fault case library, and the total number of cases is counted respectively; for each case of the target fault type, the abnormal records within a certain time window before the case occurs are checked; the number of related cases in which the target abnormal monitoring indicator appears in the time window is counted, and the correlation strength between the target abnormal monitoring indicator and the target fault type is calculated by the following formula:

[0123] ;

[0124] After obtaining the correlation strength, the abnormal monitoring indicators and / or combinations of abnormal monitoring indicators in the predictive maintenance indicator type and the obtained correlation strength are input into the fault prediction model. The fault prediction model predicts the fault type for the abnormal monitoring indicators and / or combinations of abnormal monitoring indicators in the predictive maintenance indicator type, and the fault prediction model can also predict the time window and probability of fault occurrence in combination with the degradation process knowledge in the expert knowledge base. For example, according to the duration of similar anomalies and the interval between anomalies and faults, it is concluded that "the probability of sensor failure occurring in the next 10-15 days is 65%", thereby generating a fault prediction indicator.

[0125] Optionally, the fault prediction indicator can also include structured information such as predicted fault type, predicted fault probability, abnormal monitoring indicator name, correlation strength, and possible root cause analysis.

[0126] Through the double-model mechanism of the anomaly combination model and the fault prediction model, the anomaly combination model can quantify the correlation degree between the multi-dimensional anomaly monitoring indicators and the fault types, solve the problem of high false positive rate of a single indicator, the fault prediction model combines the correlation degree and the expert knowledge, realizes the conversion of the anomaly monitoring indicators into the prediction of specific faults and the fault occurrence time, provides sufficient window for planned maintenance, reduces the probability of non-planned shutdown of the instrument and the maintenance cost.

[0127] As an optional implementation manner, in the step S402, the anomaly combination model further comprises: the anomaly combination model establishes and outputs the correlation degree between the anomaly monitoring indicators in the predictive maintenance indicator type and the fault types and / or the correlation degree between the combination of the anomaly monitoring indicators in the predictive maintenance indicator type and the fault types based on the expert classification rules, the fault types of the instrument, the anomaly performance combination, the instrument degradation process and the processing records of the occurred faults stored in the expert knowledge base in the form of prompt words.

[0128] Specifically, the expert knowledge base comprises an expert rule base, a fault mode base and a historical fault case base. The expert rule base stores the expert classification rules summarized based on the expert experience and data analysis, and the expert classification rules provide the basic criteria for the anomaly combination model to judge the correlation between the anomaly monitoring indicators and the fault types; the fault mode base stores the fault types that different instrument models may occur, the typical instrument anomaly performance combination and the instrument degradation process, and the instrument degradation process describes the change law of each indicator in the process from the normal state to the occurrence of the fault; the historical fault case base stores the detailed records of the past faults, including the anomaly monitoring indicator performance before the occurrence of the fault, the fault type, the occurrence time, the maintenance record and the root cause, and provides the actual situation of the development of the anomaly into the fault in the historical cases.

[0129] The anomaly combination model is trained based on the data in the expert rule base, the fault case base and the historical fault case base, so that the anomaly combination model learns and understands the correlation logic between the anomaly monitoring indicators and the fault types, and masters the indicator evolution law in the instrument degradation process, for example, “the daily average value of the pressure is continuously lower than the threshold value by 10% and the standard deviation is increased” may indicate that there is a slight leakage in the pipeline, and “the flow remains unchanged accompanied by the abnormal decrease of the reading” may indicate the fault of the metering module of the instrument.

[0130] Further, the abnormal combination model inputs a prompt word, and the abnormal combination model performs deep thinking according to the content of the prompt word, and finally generates a result according to the content of the prompt word. In an embodiment, the prompt word includes a target fault type, a target abnormal monitoring indicator, an analysis time window, a correlation strength calculation rule, and a task requirement, and the like; the abnormal combination model performs case retrieval on the target fault type and the target abnormal monitoring indicator in the expert knowledge base according to the task requirement, filters the retrieval result based on the analysis time window, and finally calculates and outputs the correlation strength between the target fault type and the target abnormal monitoring indicator through the correlation strength calculation rule.

[0131] When facing a single abnormal monitoring indicator in the predictive maintenance indicator type, the abnormal combination model analyzes which fault types the abnormality is commonly associated with in historical cases and the closeness of the association based on the content of the expert knowledge base. For the combination of abnormal monitoring indicators, the abnormal combination model also refers to the association of a specific fault type when multiple abnormal monitoring indicators appear simultaneously in the expert knowledge base, and further determines the correlation strength between the combined abnormal monitoring indicators and the fault type, which can reflect the closeness of the association between the abnormal monitoring indicators (or the combination of abnormal monitoring indicators) and the fault type. Finally, the abnormal combination model outputs the analysis result in the form of correlation strength, which provides key support for subsequent fault prediction.

[0132] As an optional implementation, in step S403, the fault prediction model further includes: predicting the abnormal monitoring indicator and / or the combination of abnormal monitoring indicators in the predictive maintenance indicator type based on the expert classification rules, the gauge fault type, the abnormal performance combination, the gauge degradation process, and the handling record of the occurred fault stored in the expert knowledge base, in combination with the correlation strength, through the prompt word, and outputting the predicted fault type and the fault prediction probability.

[0133] Specifically, the fault prediction model receives the abnormal monitoring indicator and / or the combination of abnormal monitoring indicators in the predictive maintenance indicator type, and guides the fault prediction model to accurately call the content related to the abnormal monitoring indicator in the expert knowledge base through the prompt word, including the expert classification rules, the gauge fault type, the abnormal performance combination, the gauge degradation process, and the handling record of the occurred fault, and the like.

[0134] The fault prediction model performs comprehensive analysis in combination with the content in the expert knowledge base and the correlation strength obtained in step S402. For example, when the correlation strength between a certain abnormal monitoring indicator and a certain fault type is high, the fault prediction model mainly refers to the information such as the performance of the fault type in historical cases and the degradation process to judge the possibility of the current abnormality developing into the fault, and calculates the corresponding probability. Through the above analysis process, the fault prediction model finally determines the predicted fault type and the corresponding fault prediction probability and outputs them, which provides a clear basis for the predictive maintenance of the gauge.

[0135] Further, in an embodiment, a prompt word is input to the fault prediction model, and the fault prediction model performs deep thinking according to the content of the prompt word, and finally outputs a result generated according to the content of the prompt word. For example, the prompt word includes the target abnormal monitoring indicator, the situation description, the correlation strength, and the task requirement, etc. The fault prediction model performs case retrieval in the expert knowledge base based on the content of the prompt word according to the task requirement, and performs deep learning on the retrieval result, and outputs the predicted fault type, the fault prediction probability, the possible cause analysis, and the action suggestion, etc.

[0136] As an optional implementation, the expert classification rule is generated based on expert experience and data analysis summary, and the expert classification rule is used to provide a basis for judgment when classifying the abnormal monitoring indicator, and assist in judging whether the abnormal monitoring indicator belongs to the immediate fault indicator type or the predictive maintenance indicator type.

[0137] Specifically, the expert classification rule is formed by combining two aspects, one is the practical experience accumulated by experts who have long been engaged in fault diagnosis of gas meters, and the other is the rule summarized after data analysis on a large amount of meter operation data and historical fault cases. After repeated verification and refinement of the actual situation, the expert classification rule can accurately reflect the potential relationship between the abnormal monitoring indicator and the fault type.

[0138] In the process of classifying the abnormal monitoring indicator, the characteristics, the scene of occurrence, and the possible impact of the abnormal monitoring indicator can be analyzed according to the expert classification rule. For example, once some abnormal monitoring indicators appear, it often means that the meter has occurred an immediate fault, which needs to be processed in priority, so the abnormal monitoring indicator can be judged to belong to the immediate fault indicator type. While other abnormal monitoring indicators may only be an early signal of the fault of the meter, and the fault will not occur immediately in the short term, so the expert classification rule assists in judging that the abnormal monitoring indicator belongs to the predictive maintenance indicator type.

[0139] As an optional implementation, as shown in Figure 5 In step S105, the maintenance suggestion is generated based on the immediate fault indicator type abnormal monitoring indicator and / or the fault prediction indicator, and further includes the following steps:

[0140] Step S501, based on the expert knowledge base, obtaining maintenance experience about fault types, repair processes, and component replacement periods.

[0141] Step S502, according to the fault type of the immediate fault indicator type abnormal monitoring indicator and / or the fault prediction indicator, searching for the corresponding maintenance experience in the expert knowledge base to generate the maintenance suggestion.

[0142] Specifically, the expert knowledge base further includes a maintenance experience base, which stores detailed experiences of maintenance personnel accumulated in long-term practice about fault types, maintenance processes, and component replacement periods. These experiences exist in structured and unstructured forms, including standard maintenance processes for different fault types, root cause analysis of common faults, component aging rules, and recommended replacement periods, etc.

[0143] According to the fault type corresponding to the current instant fault indicator or fault prediction indicator to be processed, the expert knowledge base is accurately searched with the fault type as an index to obtain the corresponding maintenance experience. The searched maintenance experience is sorted and structured to be converted into specific maintenance suggestions. The maintenance suggestions include instant fault alarms, fault diagnosis reports, fault prediction reports, predictive maintenance suggestions, maintenance work orders or dispatch instructions, fault handling steps, safety precautions, etc., to ensure that the maintenance suggestions are operable and can guide the operation and maintenance personnel to accurately and efficiently handle faults or preventive maintenance.

[0144] Further, if the maintenance case corresponding to the fault type is not found in the expert knowledge base, a similar case can also be searched in the expert knowledge base according to the abnormal situation of the meter and the meter model, and the maintenance suggestion is generated based on the similar case to provide the operation and maintenance personnel with valuable guidance.

[0145] According to the above description, the meter fault detection method provided by the present application trains a binary classification model based on an expert knowledge base, classifies abnormal monitoring indicators through the trained binary classification model, distinguishes between instant fault indicator types and predictive maintenance indicator types, avoids treating all meter abnormalities as instant faults, and improves the utilization efficiency of maintenance resources.

[0146] Further, the meter fault detection method provided by the present application constructs an abnormal combination model and a fault prediction model through a large language model, reconstructs the abnormal monitoring indicators belonging to the predictive maintenance indicator type through the abnormal combination model and the fault prediction model, obtains the correlation strength between the abnormal monitoring indicators and the fault types and the combination of the abnormal monitoring indicators and the fault types, and predicts the fault type and the fault occurrence time corresponding to the abnormal monitoring indicators belonging to the predictive maintenance indicator type through the obtained correlation strength, thereby providing a sufficient window for planned maintenance and reducing the probability of non-planned shutdown of the meter and the maintenance cost.

[0147] In a second aspect, the embodiments of the present application further provide a meter fault detection system. The meter fault detection system is applied to a gas meter. The meter fault detection system includes a data acquisition module, a feature engineering module, an abnormal monitoring module, an expert knowledge base module, an abnormal classification module, a fault prediction module, and a result output module.

[0148] As shown in Figure 6 The data acquisition module is configured to acquire multi-dimensional data of the meter and pre-process the acquired data. The feature engineering module is configured to extract features from the acquired data to obtain feature data reflecting the running state of the meter. The anomaly monitoring module is configured to include a parameter threshold library, and generate anomaly monitoring indicators by identifying anomalies in the feature data based on the parameter threshold library. The expert knowledge base module is configured to store meter fault types, meter degradation processes, anomaly manifestation combinations, expert classification rules, handling records of faults that have occurred, and maintenance experience.

[0149] The anomaly classification module is configured to classify the anomaly monitoring indicators based on the expert knowledge base module, determine the categories of the anomaly monitoring indicators, and the categories include: immediate fault indicator types and predictive maintenance indicator types. The fault prediction module is configured to reconstruct indicators based on the expert knowledge base, anomaly monitoring indicators and / or combinations of anomaly monitoring indicators in the predictive maintenance indicator types, to generate fault prediction indicators, which are configured to identify predicted fault types and fault prediction probabilities. The result output module is configured to output corresponding maintenance recommendations based on the anomaly monitoring indicators of the immediate fault indicator types and / or the fault prediction indicators.

[0150] According to the above description, the meter fault detection system provided by the present application classifies anomaly monitoring indicators based on an expert knowledge base, distinguishes between immediate fault indicator types and predictive maintenance indicator types, improves the classification accuracy of anomaly monitoring indicators, avoids treating all meter anomalies as immediate faults, improves the utilization efficiency of maintenance resources, reconstructs indicators for anomaly monitoring indicators in the predictive maintenance indicator types, predicts the fault types and fault occurrence times corresponding to the anomaly monitoring indicators in the predictive maintenance indicator types, thereby providing sufficient windows for planned maintenance and reducing the probability of unplanned downtime of the meter and maintenance costs.

[0151] As an optional implementation manner, the fault prediction module is further configured to: based on the principle of a large language model, construct an anomaly combination model and a fault prediction model respectively.

[0152] The anomaly combination model is used to obtain the association strength between anomaly monitoring indicators in the predictive maintenance indicator types and fault types, and / or the association strength between combinations of anomaly monitoring indicators in the predictive maintenance indicator types and fault types.

[0153] Based on the association strength, the fault prediction model is used to predict anomaly monitoring indicators and / or combinations of anomaly monitoring indicators in the predictive maintenance indicator types, to obtain predicted fault types and fault prediction probabilities.

[0154] The abnormal combination model and the fault prediction model are constructed based on the principle of a large language model. The abnormal monitoring indicators of the predictive maintenance indicator type are reconstructed by the abnormal combination model and the fault prediction model. The correlation strength between the abnormal monitoring indicators and the fault types and the combination of the abnormal monitoring indicators and the fault types is obtained. The fault type and the fault occurrence time corresponding to the abnormal monitoring indicators of the predictive maintenance indicator type are predicted by the obtained correlation strength. Thus, a sufficient window is provided for planned maintenance, and the probability of non-planned shutdown of the gauge and the maintenance cost are reduced.

[0155] As an optional implementation manner, the abnormal classification module is further configured to: acquire historical abnormal monitoring indicator data that has been classified based on the expert knowledge base module; train the binary classification model using the historical abnormal monitoring indicator data; and classify the unclassified abnormal monitoring indicators by using the trained binary classification model to determine the category to which the abnormal monitoring indicators belong.

[0156] The binary classification model is trained by relying on the historical abnormal monitoring indicator data accumulated in the expert knowledge base. The abnormal monitoring indicators are classified by using the binary classification model. The automation of the classification process is realized. The classification deviation caused by simply relying on manual experience is avoided. The misjudgment and the missed judgment in the complex abnormal scenario are reduced. The accuracy and the efficiency of the classification of the abnormal monitoring indicators are improved. Moreover, by dividing the abnormal monitoring indicators into two categories, namely, the immediate fault indicators and the predictive maintenance indicators, all abnormalities are avoided from being treated as immediate faults. The utilization rate of the maintenance resources is improved.

[0157] It can be understood that the word "exemplary" used in this paper means "as an example, example or illustration". Any embodiment described as "exemplary" is not necessarily superior to or superior to other embodiments and / or does not exclude the features combined with other embodiments. It should be understood that certain features of the present application described in the context of separate embodiments for the sake of clarity can also be provided in combination. Conversely, various features of the present application described in the context of a single embodiment for the sake of clarity can also be provided separately or in any suitable combination or as an embodiment of the present application described anywhere else.

[0158] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this paper is only a description of the association between the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, "at least one" means one or more, and "multiple" means two or more. "First", "second", etc. are not limited in quantity and execution order, and "first", "second", etc. are not necessarily different.

[0159] The above-described embodiments are merely intended to illustrate the present application, but not to limit the scope of the present application. Those skilled in the art can understand that changes, modifications, substitutions, combinations, simplifications, and the like, which do not depart from the spirit and scope of the present application and the appended claims, are all equivalent replacements and still fall within the scope of the present application.

Claims

1. A method of detecting a failure of a meter, the method comprising: The method is applied to fault detection of a gas meter, and the meter fault detection method comprises the following steps: Collecting multi-dimensional data of the meter, extracting a plurality of characteristic data reflecting the running state of the meter from the meter data through feature engineering, wherein the characteristic data comprises gas consumption, temperature, pressure and flow rate; Building a parameter threshold library, identifying the abnormality of each characteristic data based on the parameter threshold library, and generating an abnormal monitoring index, wherein the abnormal monitoring index comprises meter id, abnormal condition description and abnormal occurrence time; Building an expert knowledge base, obtaining historical abnormal monitoring index data that has been classified based on the expert knowledge base, training a binary classification model using the historical abnormal monitoring index data, classifying unclassified abnormal monitoring indexes through the trained binary classification model, determining the category of the abnormal monitoring index, and the category comprises instant fault indicator type and predictive maintenance indicator type; Based on the expert knowledge base, the abnormal monitoring index and / or the combination of the abnormal monitoring index belonging to the predictive maintenance indicator type are reconstructed to generate a fault prediction index, and the fault prediction index is configured to identify the predicted fault type and the fault prediction probability; Based on the abnormal monitoring index of the instant fault indicator type and / or the fault prediction index, a corresponding maintenance suggestion is provided; The meter fault detection method further comprises: based on the principle of large language model, an abnormal combination model and a fault prediction model are respectively constructed; Through the abnormal combination model, the association strength between the abnormal monitoring index in the predictive maintenance indicator type and the fault type is obtained, and / or the association strength between the combination of the abnormal monitoring index in the predictive maintenance indicator type and the fault type is obtained; Based on the association strength, the abnormal monitoring index in the predictive maintenance indicator type and / or the combination of the abnormal monitoring index are predicted through the fault prediction model to obtain the predicted fault type and the fault prediction probability.

2. The method of claim 1, wherein, The meter fault detection method further comprises: The abnormal combination model, based on the expert classification rules, meter fault types, abnormal performance combinations, meter degradation processes and handling records of faults that have occurred stored in the expert knowledge base, establishes and outputs the association strength between the abnormal monitoring index in the predictive maintenance indicator type and the fault type, and / or the association strength between the combination of the abnormal monitoring index in the predictive maintenance indicator type and the fault type.

3. The method of claim 1, wherein, The meter fault detection method further comprises: The fault prediction model, based on the expert classification rules, meter fault types, abnormal performance combinations, meter degradation processes and handling records of faults that have occurred stored in the expert knowledge base, predicts the abnormal monitoring index in the predictive maintenance indicator type and / or the combination of the abnormal monitoring index, and outputs the predicted fault type and the fault prediction probability.

4. The method of claim 3, wherein, The expert classification rule is generated based on expert experience and data analysis summary, and is used to provide a basis for judgment when classifying the abnormal monitoring indicators, and assist in determining whether the abnormal monitoring indicators belong to the immediate failure indicator type or the predictive maintenance indicator type.

5. The method of claim 1, wherein, The meter fault detection method further comprises: Based on the expert knowledge base, maintenance experience about fault types, maintenance processes and component replacement cycles is obtained; According to the abnormal monitoring indicators of the immediate failure indicator type and / or the fault types of the failure prediction indicators, corresponding maintenance experience is searched in the expert knowledge base to generate the maintenance suggestions.

6. The method of claim 1, wherein, The meter fault detection method further comprises: Based on the fault-free operation period of the meter, statistical values of meter parameters are calculated, including one or more of the following types of statistical values: mean, variance, quantile and standard deviation of the meter parameters; Based on at least one statistical value, abnormal judgment thresholds of different kinds of meter parameters are constructed respectively, and the abnormal judgment thresholds are stored in the parameter threshold library, which dynamically updates the abnormal judgment thresholds; The abnormal judgment thresholds are obtained from the parameter threshold library, and by comparing the feature data with the abnormal judgment thresholds, it is determined whether the feature data is abnormal, and according to the determination result, if there is an abnormality, the abnormal monitoring indicators are generated.

7. A system for detecting failure of a gauge, the system comprising: The meter fault detection system is applied to the fault detection of a gas meter, and the meter fault detection system comprises: A data acquisition module configured to acquire multi-dimensional data of the meter and pre-process the acquired data; A feature engineering module configured to extract features from the acquired data to obtain feature data reflecting the running state of the meter, including gas consumption, temperature, pressure and flow; An abnormal monitoring module configured to include a parameter threshold library, and based on the parameter threshold library, to generate abnormal monitoring indicators by identifying abnormalities in the feature data, including meter id, abnormal condition description and abnormal occurrence time; An expert knowledge base module configured to store meter fault types, meter degradation processes, abnormal performance combinations, expert classification rules, handling records of faults that have occurred and maintenance experience; An abnormal classification module configured to obtain historical abnormal monitoring indicator data that has been classified based on the expert knowledge base module, train a binary classification model using the historical abnormal monitoring indicator data, classify unclassified abnormal monitoring indicators using the trained binary classification model, and determine the class of the abnormal monitoring indicators, including immediate failure indicator type and predictive maintenance indicator type; A failure prediction module configured to reconstruct indicators based on the expert knowledge base, the abnormal monitoring indicators belonging to the predictive maintenance indicator type and / or combinations of the abnormal monitoring indicators, to generate failure prediction indicators configured to identify predicted fault types and fault prediction probabilities; The result output module is configured to output a corresponding maintenance suggestion based on the abnormal monitoring indicator and / or the fault prediction indicator of the instant fault indicator type; The fault prediction module is further configured to: construct an abnormal combination model and a fault prediction model based on the principle of a large language model respectively; obtain the association strength between the abnormal monitoring indicator in the predictive maintenance indicator type and the fault type and / or the association strength between the combination of the abnormal monitoring indicator in the predictive maintenance indicator type and the fault type through the abnormal combination model; predict the abnormal monitoring indicator and / or the combination of the abnormal monitoring indicator in the predictive maintenance indicator type through the fault prediction model based on the association strength, and obtain a predicted fault type and a fault prediction probability.

Images