A state identification method and system of an electric energy meter calibration device

Abstract

A state identification method and system of an electric energy meter calibration device, comprising the following steps: step one, historical data preprocessing; step two, based on the data processed in step one, using an electric energy meter feature clustering model for preliminary clustering, and based on the obtained preliminary clustering center, secondary clustering to find out the clustering number and clustering center; step three, judging the state meaning of the clustering center trained in step two; step four, inputting real-time electric energy meter calibration data and step two clustering result to identify the state of the calibration device and the meter position. The present application can realize full-automatic real-time monitoring and improve monitoring efficiency on the one hand, and can identify the state of the calibration device in a large amount of complex data, avoiding the misjudgment of the performance of the electric energy meter caused by the fault of the calibration device.

Description

Technical Field

[0001] This invention relates to the field of power system automation and intelligent fault diagnosis technology, and in particular to a method and system for identifying the status of an electricity meter calibration device. Background Technology

[0002] In power systems, electricity meter calibration devices are crucial equipment for accurately testing and evaluating the performance of users' electricity metering instruments. Their reliability directly impacts the accuracy and impartiality of electricity metering. With the advancement of smart grid construction, electricity meter calibration devices are typically equipped with multiple meter positions to achieve efficient centralized testing of a large number of electricity metering instruments. However, in actual operation, the calibration device and its various meter positions may malfunction due to various factors such as aging electrical components, environmental interference, and mechanical wear. If a malfunction occurs and is not detected in a timely manner, it can not only lead to inaccurate metering data but also potentially cause safety hazards in power supply, increase the operation and maintenance costs of power companies, and raise the risk of user complaints.

[0003] Currently, fault detection of electricity meter calibration devices and meter positions largely relies on manual inspections or simple monitoring methods based on threshold judgments. Manual inspections suffer from low efficiency, strong subjectivity, and the inability to monitor in real time; threshold-based monitoring methods are difficult to adapt to complex and changing operating conditions and cannot effectively identify early faults and potential fault hazards.

[0004] In existing technologies, there have been some explorations based on data mining and artificial intelligence algorithms for monitoring and analyzing the operating status of electricity meters. For example, invention patent CN119355626A discloses a method and system for monitoring the operation of electricity meters. This method first obtains initial cluster centers using the K-Means algorithm, and then uses the Dragonfly algorithm to iteratively optimize the cluster centers, with the optimization objective being to minimize the sum of squared errors within clusters. Finally, abnormal operating data is identified by calculating the clustering probability of data points and the final cluster centers and comparing it with a preset anomaly probability threshold. However, the K-Means and Dragonfly algorithm optimization processes used have limited effectiveness in identifying non-spherical and unevenly dense cluster structures, and the distance threshold in the clustering process depends on fixed parameters, lacking the ability to adapt to dynamic changes in data distribution, thus affecting the final accuracy of the status identification of the electricity meter verification device.

[0005] Therefore, there is an urgent need for a technical solution that can predict the failure of the testing device and meter position in real time and accurately, so as to ensure the stable and reliable operation of the electricity meter testing system and improve the quality of electricity metering services. Summary of the Invention

[0006] To solve the problems in the prior art, the present application improves the existing clustering analysis method in view of the problems of the existing electric energy meter operation monitoring method, such as difficulty in accurately identifying subtle changes and abnormal states, poor timeliness of abnormal identification, and low monitoring efficiency. The present application designs an electric energy meter feature clustering model that can more accurately reflect the data distribution characteristics. The model uses sample density as the clustering division criterion to provide high-quality initial values for the subsequent secondary clustering process, and can more accurately cluster data, so as to more accurately identify the state of the testing device and effectively avoid misjudgment of the performance of the electric energy meter due to device failure. On the one hand, it can realize full-automatic real-time monitoring and improve the monitoring efficiency; on the other hand, it can identify the state of the testing device in a large amount of complex data and avoid misjudgment of the performance of the electric energy meter due to testing device failure.

[0007] The present application adopts the following technical solutions.

[0008] In a first aspect, the present application provides a state identification method of an electric energy meter testing device, comprising:

[0009] Step 1: obtaining and preprocessing historical electric energy meter testing data to obtain a historical electric energy meter testing feature vector;

[0010] Step 2: based on the historical electric energy meter testing feature vector, using an electric energy meter feature clustering model for preliminary clustering, including determining a first preliminary clustering center through a two-stage screening process combining high-density screening set and centripetal scoring, and determining a new preliminary clustering center according to multi-factor combination weight iteration; performing secondary clustering on the preliminary clustering center to obtain a final clustering center;

[0011] Step 3: based on the final clustering center and a preset state identification rule, identifying and outputting a clustering center representation state;

[0012] Step 4: obtaining and preprocessing real-time electric energy meter testing data, using a trend analysis model to establish and analyze a time series queue composed of current and historical failure rates according to the final clustering center and the clustering center representation state, calculating an electric energy meter failure history baseline, an electric energy meter failure change trend, and an electric energy meter failure history volatility, and identifying and outputting a real-time electric energy meter testing device state.

[0013] Preferably, the historical electric energy meter testing data comprises:

[0014] historical precision readings, K times of detection are performed on an electric energy meter testing device having N meter positions, and M historical precision readings are generated on each meter position in each detection;

[0015] a historical status label corresponding to a specific epitope in a specific detection, and indicating a status of the specific epitope in the specific detection as one of a normal status, an abnormal status, or a caution status.

[0016] Preferably, the process of preprocessing the historical electric energy meter calibration data comprises:

[0017] organizing the historical precision readings into K*N electric energy meter calibration data groups; wherein each of the electric energy meter calibration data groups is composed of M precision readings corresponding to a specific epitope in a specific detection;

[0018] For each of the electric energy meter calibration data groups, a historical electric energy meter calibration feature vector is constructed by sorting the M historical precision readings in the group, and extracting values from a preset position in the sorted historical precision readings sequence.

[0019] Preferably, the process of selecting the remaining feature vector set using the electric energy meter feature clustering model comprises:

[0020] The process of determining the first primary clustering center comprises:

[0021] calculating the sample density of each of the historical electric energy meter calibration feature vectors, and determining a high-density candidate set composed of vectors with top P% density according to the sample density;

[0022] For each of the candidate vectors in the high-density candidate set, calculate the average distance of the candidate vector from all the historical electric energy meter calibration feature vectors to obtain a centricity score; and determine the candidate vector with the smallest centricity score in the high-density candidate set as the first primary clustering center.

[0023] Preferably, removing the historical electric energy meter calibration feature vectors belonging to the first primary clustering center according to the first primary clustering center comprises:

[0024] calculating the sample set average distance between the historical electric energy meter calibration feature vectors;

[0025] calculating the distance of each of the historical electric energy meter calibration feature vectors from the first primary clustering center;

[0026] identifying all the historical electric energy meter calibration feature vectors with a distance from the first primary clustering center smaller than the sample set average distance as first primary clustering center members;

[0027] removing the first primary clustering center members from the historical electric energy meter calibration feature vectors to obtain a remaining feature vector set.

[0028] Preferably, the step of determining the new initial clustering center according to the multi-factor combination weight iteration comprises:

[0029] Step 2.2.1: Calculate the sample set average distance of the remaining feature vector set, denoted as:

[0030] ;

[0031] In the formula: is the sample set average distance, , is the th and the th vector of the remaining feature vector set, , denotes the distance between the th and the th vector;

[0032] Step 2.2.2: Traverse the remaining feature vector set, and calculate the sample density and the intra-class average distance of each vector;

[0033] Step 2.2.3: Traverse the remaining feature vector set, and calculate the class distance of each sample according to the sample density;

[0034] Step 2.2.4: Calculate the multi-factor combination weight of each vector using an adaptive weight coefficient according to the sample density, the intra-class average distance, and the class distance;

[0035] Wherein, the adaptive weight coefficient is dynamically adjusted according to the ratio of the current remaining historical electric energy meter calibration data quantity to the initial historical electric energy meter calibration data quantity;

[0036] Step 2.2.5: Determine one vector with the largest multi-factor combination weight as the new initial clustering center; for each of the historical electric energy meter calibration feature vectors, calculate the distance from the new initial clustering center; identify all remaining feature vector sets with a distance from the new initial clustering center less than the sample set average distance as new initial clustering center members; remove the new initial clustering center members from the remaining feature vector set to obtain an updated remaining feature vector set;

[0037] Step 2.2.6: Repeat steps 2.2.1 to 2.2.5 until all vectors are removed from the remaining feature vector set, and output the initial clustering center.

[0038] Preferably, the preset state recognition rule comprises:

[0039] Set the first condition, denoted as:

[0040] and ;

[0041] The second condition is set as follows:

[0042] and ;

[0043] in, Indicates the th in the final cluster center Cluster centers; They represent Component data at preset positions q, w, e, and r Indicates a preset threshold;

[0044] like If the first condition is met, the status label of the electricity meter calibration device is determined to be in normal condition.

[0045] like If the first condition is not met, but the second condition is met, then the status label of the electricity meter calibration device is determined to be in a state of alert.

[0046] like If the first and second conditions are not met, the status label of the electricity meter calibration device is determined to be in an abnormal state.

[0047] Preferably, the trend analysis model includes:

[0048] Based on the cluster center characterization status of the current batch, calculate the current comprehensive failure rate, which represents the proportion of abnormal state meter positions, and store the current comprehensive failure rate in a time series queue; the time series queue is used to store the most recent L records of electricity meter failure rates.

[0049] Based on the electricity meter failure rate records, the historical baseline of electricity meter failures, the trend of electricity meter failure changes, and the historical volatility of electricity meter failures are calculated; the historical baseline of electricity meter failures represents the arithmetic mean of the electricity meter failure rate records excluding the current comprehensive failure rate; the trend of electricity meter failure changes represents the difference between the current comprehensive failure rate and the historical baseline of electricity meter failures; the historical volatility of electricity meter failures represents the degree of dispersion of the electricity meter failure rate records.

[0050] According to a preset priority order, the real-time status of the electricity meter verification device is determined by applying composite rules based on the historical baseline of the electricity meter fault, the trend of the electricity meter fault change, and the historical volatility of the electricity meter fault.

[0051] Secondly, the present invention provides a state identification system for an electricity meter calibration device, and the state identification method for the aforementioned electricity meter calibration device includes:

[0052] A historical electric energy meter calibration data processing module is configured to acquire and pre-process historical electric energy meter calibration data to obtain a historical electric energy meter calibration feature vector.

[0053] A historical electric energy meter calibration data clustering module is configured to perform preliminary clustering using an electric energy meter feature clustering model based on the historical electric energy meter calibration feature vector, including determining a first preliminary clustering center through a two-stage screening process combining high-density screening set and centripetal scoring, and determining a new preliminary clustering center according to multi-factor combination weight iteration; and performing secondary clustering on the preliminary clustering center to obtain a final clustering center.

[0054] A historical electric energy meter calibration data state representation module is configured to identify and output a clustering center state representation based on the final clustering center and a preset state recognition rule.

[0055] A real-time electric energy meter calibration data state representation module is configured to acquire and pre-process real-time electric energy meter calibration data, use a trend analysis model to establish and analyze a time series queue composed of current and historical failure rates according to the final clustering center and the clustering center state representation, calculate an electric energy meter failure history baseline, an electric energy meter failure change trend and an electric energy meter failure history volatility, and identify and output a real-time electric energy meter calibration device state.

[0056] A third aspect of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program, when loaded into the processor, implements the state recognition method of the electric energy meter calibration device.

[0057] A fourth aspect of the present application provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the state recognition method of the electric energy meter calibration device.

[0058] Compared with the prior art, the present application has the following advantages:

[0059] Improved monitoring efficiency: traditional manual inspection is low in efficiency and cannot be monitored in real time, while the present application can realize full-automatic real-time monitoring, can obtain the running state information of the electric energy meter calibration device in time, does not need frequent on-site inspection by manual work, greatly saves the labor and time cost, can discover potential problems in the first time, and significantly improves the monitoring efficiency.

[0060] Improve state identification accuracy: the application adopts the electric energy meter feature clustering model to carry out data clustering to innovate. The traditional clustering algorithm such as K-Means, when processing non-spherical or density uneven data set, the clustering effect is often poor. In addition, the clustering effect of traditional clustering algorithm depends on the selection of initial clustering center, and in the selection of initial clustering center, the sample distance is generally taken as the measurement standard, and in the processing of a large amount of complex data, the initial clustering effect is poor, which leads to the difficulty of achieving ideal clustering effect in the subsequent clustering process. The electric energy meter feature clustering model of the application takes the sample density as the clustering division criterion, which can more accurately reflect the data distribution characteristics, and can achieve better clustering effect when facing a large amount of complex operation data generated by the electric energy meter calibration device. At the same time, the electric energy meter feature clustering model can obtain the clustering center of each class, which can directly provide high-quality initial value for the subsequent clustering process, so that the electric energy meter feature clustering model can carry out preliminary clustering, and based on the preliminary clustering center, secondary clustering can be carried out, which can more accurately cluster the data, so as to more accurately identify the state of the calibration device in a large amount of complex data, effectively avoid the misjudgment of the performance of the electric energy meter caused by the fault of the calibration device, and ensure the accuracy and fairness of electric power measurement. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is a flow chart of the state identification method of the electric energy meter calibration device in the embodiment of the application;

[0062] Figure 2 is a clustering effect display diagram of the training set in the specific example in the embodiment of the application;

[0063] Figure 3 is a clustering effect display diagram of the test set in the specific example in the embodiment of the application;

[0064] Figure 4 is a structure principle block diagram of the state identification system of the electric energy meter calibration device in the embodiment of the application. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme of the application will be described clearly and completely below in combination with the drawings in the embodiment of the application. The described embodiments of the application are only a part of the embodiments of the application, not all embodiments. Based on the spirit of the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0066] Embodiment one:

[0067] With reference to Figure 1 , the embodiment of the application provides a state identification method of an electric energy meter calibration device, comprising the following steps:

[0068] Step one: obtain and preprocess historical electric energy meter calibration data to obtain historical electric energy meter calibration feature vectors;

[0069] Step two: based on the historical electric energy meter calibration feature vectors, use the electric energy meter feature clustering model for preliminary clustering, including determining the first preliminary clustering center through a two-stage screening process combining high-density screening set and centripetal score, and determining the new preliminary clustering center according to the multi-factor combination weight iteration; based on the secondary clustering of the preliminary clustering center, the final clustering center is obtained;

[0070] In the application scenario of the present application, due to the existence of multiple attention states and abnormal states, the accurate clustering of the traditional clustering algorithm will be affected by the selection of the initial clustering center and cannot be used alone. In addition, it also depends on the manually set distance threshold, and the electric meter detection device may be aged and worn, and the state of different devices is also different, which makes the traditional clustering algorithm not universal and adaptive, and the distance threshold needs to be adjusted manually, resulting in the decline of efficiency and accuracy. The electric energy meter feature clustering model can adaptively adjust the clustering process according to the density and distribution of data points, and dynamically determine the clustering boundary by calculating the local density between data points, so it has universality and adaptability. Therefore, the electric energy meter feature clustering model must be used to obtain the initial clustering center first, and then secondary clustering is performed to achieve the best effect.

[0071] Step three: based on the final clustering center and the preset state recognition rule, identify and output the state represented by the clustering center;

[0072] Step four: obtain and preprocess real-time electric energy meter calibration data, use the trend analysis model to establish and analyze the time series queue composed of current and historical failure rates according to the final clustering center and the state represented by the clustering center, calculate the electric energy meter failure history baseline, the electric energy meter failure change trend and the electric energy meter failure history volatility, and identify and output the real-time electric energy meter calibration device state.

[0073] Specifically:

[0074] Step one, historical data preprocessing;

[0075] The historical electric energy meter calibration data includes:

[0076] The historical precision reading is obtained by detecting K times in an electric energy meter calibration device with N table positions, and M historical precision readings are generated on each table position in each detection.

[0077] The historical state label corresponds to a specific table position in a specific detection, and indicates that the state of the specific table position at the specific detection is one of normal state, abnormal state or attention state.

[0078] Specifically, assume that a certain tested energy meter calibration device has performed a total of This test has a total of Each tabletop is detected, and each tabletop is tested in each detection step. A multi-meter instrument is used in actual electricity meter testing to meet the needs of testing efficiency. It is usually much greater than 7. Therefore, there are a total of Each historical accuracy reading reflects the difference between the reading of the tested electricity meter calibration device and the reading of a high-precision standard source. The smaller the absolute value, the higher the accuracy of the electricity meter calibration device. In addition, for each test, a historical status label of normal, caution, or abnormal is given for each meter position.

[0079] Next, the process of preprocessing the historical electricity meter calibration data includes:

[0080] The historical accuracy readings are organized into K*N electricity meter calibration data groups; wherein each electricity meter calibration data group consists of M accuracy readings corresponding to a specific meter position in a specific test;

[0081] For each of the energy meter verification data groups, the M historical accuracy readings within the group are sorted, and values ​​are extracted from preset positions in the sorted historical accuracy reading sequence to form a historical energy meter verification feature vector.

[0082] Here, a specific detection refers to one of the K detections (e.g., the k-th detection, where 1≤k≤K). A specific table position refers to one of the N table positions (e.g., the n-th table position, where 1≤n≤N).

[0083] In this embodiment, after obtaining historical data, the process is iterated through. Each detection iteration iterates through the data from each detection. The data for each table position will be used to detect each table position. The instrument accuracy data are sorted in ascending order, according to... The seven data points were indexed; this operation aims to extract features from the detection data. Engineering experience shows that directly using the raw detection data results in high computational cost, long convergence time, and unsatisfactory clustering effects. However, the data feature extraction method proposed in this invention effectively reduces computational cost and improves clustering performance. This invention achieves better results by extracting seven key data features. In practical applications, the number of data features can be flexibly adjusted according to computational requirements. Indicates less than The largest integer.

[0084] After the above processing, the features of historical data can be extracted, namely... indivual The matrix of each The matrix of each The matrix of each

[0085] Step two, based on the data processed in step one, i.e. The matrix of each The matrix of each

[0086] Step 2.1: using the electric energy meter feature clustering model to determine the first primary clustering center, and removing the historical electric energy meter calibration feature vectors belonging to the first primary clustering center, the specific process including:

[0087] First, the process of determining the first primary clustering center includes:

[0088] Calculate the sample density of each of the historical electric energy meter calibration feature vectors, and according to the sample density, determine a high-density candidate set composed of the top P% density vectors;

[0089] For each candidate vector in the high-density candidate set, calculate the average distance between it and all historical electric energy meter calibration feature vectors to obtain the centripetal score; the candidate vector with the smallest centripetal score in the high-density candidate set is determined as the first primary clustering center.

[0090] Then, according to the first primary clustering center, remove the historical electric energy meter calibration feature vectors belonging to the first primary clustering center, including:

[0091] Calculate the average distance of the sample set between the historical electric energy meter calibration feature vectors;

[0092] For each of the historical electric energy meter calibration feature vectors, calculate the distance between it and the first primary clustering center;

[0093] Identify all historical electric energy meter calibration feature vectors with a distance from the first primary clustering center less than the average distance of the sample set as the first primary clustering center members;

[0094] Remove the first primary clustering center members from the historical electric energy meter calibration feature vectors to obtain the remaining feature vector set.

[0095] In this embodiment, the sample set including historical electricity meter verification feature vectors is traversed, and the sample density of each sample is calculated (each sample is a unique identifier). The matrix, the sample set contains (A total of 100 samples) Then, based on the calculated sample density, all historical electricity meter verification feature vectors are sorted in descending order, and the vectors with the highest density (P%) are selected from the sorted sequence to form a high-density candidate set. In this embodiment, the value of P can be a small percentage set based on experience, such as P=5.

[0096] For each candidate vector in the candidate set, calculate the average distance between that candidate vector and all other vectors in the dataset, and define this average distance as the centripetal score of the candidate vector. The smaller the centripetal score of a candidate vector, the closer its position is to the geometric center in the overall data space. Iterate through the centripetal scores of all candidate vectors in the high-density candidate set, and define the candidate vector with the smallest centripetal score as the first preliminary cluster center. Iterate through the sample set; if the distance from a sample to the first preliminary cluster center is less than the average distance of the sample set, the sample is considered to belong to the first cluster center, identified as a member of the first preliminary cluster center, and removed from the sample set, resulting in the remaining feature vector set. By ensuring that the first cluster center accurately captures the features of the most core cluster in the dataset, a chain reaction caused by improper initial center selection is avoided, such as mistakenly merging two clusters that should be separate, providing a better starting point for subsequent clustering processes.

[0097] The formula for calculating the average distance of the sample set is as follows:

[0098]

[0099] In the formula: The average distance of the sample set, , The first The, the One sample, Indicates the first The and the first The distance between samples is calculated using the following formula:

[0100]

[0101] In the formula: it is easy to see that each sample is a 1*7 matrix; , Samples , The data located at the p-th position .

[0102] The sample density is calculated as follows:

[0103] The sample density of the i-th sample element in the data set is defined as:

[0104]

[0105] In the formula: is the sample density of the i-th sample, is the distance between the i-th and j-th samples, and the function is defined as follows:

[0106]

[0107] After step 2.1, the first cluster center and the samples belonging to the class are obtained.

[0108] Step 2.2: Assume that there are n samples left in the current sample set; iterate through the n samples, calculate the product weight of each sample, and consider the sample with the largest product weight as a new cluster center. Remove the samples belonging to the cluster center from the sample set. Repeat this step until all samples are removed from the sample set. Specifically, the step of determining a new preliminary cluster center according to the multi-factor combination weight iteration includes:

[0109] Step 2.2.1: Calculate the sample set average distance of the remaining feature vector set (n samples), which is calculated as follows:

[0110]

[0111]

[0112] In the formula: is the sample set average distance of the remaining feature vector set, , are the i-th and j-th samples of the remaining feature vector set, , , , , , ,

[0113]

[0114] In the formula: it is known that each sample is a 1*7 matrix; , are the i-th and j-th samples of the remaining feature vector set,​​​​​​​ , Located in the middle Data at each location .

[0115] Step 2.2.2: Traverse the remaining feature vector set, calculate the sample density and intra-class average distance for each sample, and so on. The formula for calculating the sample density of a sample is as follows:

[0116]

[0117] In the formula: For the first Sample density of each sample.

[0118] No. The formula for calculating the intra-class average distance of a sample is as follows:

[0119]

[0120] In the formula: For the first The average intra-class distance of each sample.

[0121] Step 2.2.3: Traverse the remaining feature vector set and calculate the class distance for each sample. Class distance of each sample The calculation formula is as follows:

[0122]

[0123] In the formula: It represents existence. It means it does not exist; Indicates searching The minimum value within the sample is found by considering the condition that the sample is within the sample. The sample density ratio of the sample The sample density is high; Indicates searching The maximum value within.

[0124] Step 2.2.4: Based on the sample density, intra-class average distance, and class distance, calculate the multi-factor combined weights for each vector using adaptive weight coefficients;

[0125] The adaptive weighting coefficient is dynamically adjusted based on the ratio of the current remaining historical energy meter verification data to the initial historical energy meter verification data.

[0126] Specifically, assuming the calculation of the first The multi-factor combination weights of the nth sample, the nth The formula for calculating the multi-factor combination weights of a sample is as follows:

[0127]

[0128] In the formula: For the first Multi-factor combination weights for each sample , and As adaptive weight coefficients, their values ​​are not fixed, but are dynamically adjusted according to the progress of clustering iteration. , and The value can be correlated with the proportion of the remaining historical electricity meter calibration data.

[0129] Specifically, in this embodiment, a decay function is designed to adjust... The value, while increasing and The values ​​include:

[0130] Let R be the ratio of the current number of remaining data points to the initial total number of data points, therefore , , .

[0131] In this way, more emphasis is placed on the early stages of iteration, when the R value is relatively large. The algorithm tends to prioritize selecting final cluster centers with clear boundaries and high separation. However, in later iterations, when the R value is low and only some mixed points remain, the algorithm places greater emphasis on... and (corresponding) , (With a large value), it tends to find the final cluster centers with higher density and more compact internal structure from the remaining points, thus better identifying clusters of different scales and shapes, thereby greatly improving the accuracy and robustness of the final clustering results.

[0132] Step 2.2.5: Determine the vector with the largest multi-factor combination weight as the new preliminary cluster center; for each of the historical energy meter verification feature vectors, calculate its distance to the new preliminary cluster center; identify all remaining feature vectors whose distance to the new preliminary cluster center is less than the average distance of the sample set as new preliminary cluster center members; remove the new preliminary cluster center members from the remaining feature vector set to obtain the updated remaining feature vector set.

[0133] Specifically, traversing the sample set Given a set of samples, find the sample with the largest product weight and treat it as a new cluster center. Then iterate through the sample set again; if a sample is less than [a certain value] from the new cluster center, [then the sample is considered a new cluster center]. If the sample is considered to belong to the new cluster center, it will be removed from the sample set.

[0134] Step 2.2.6: Repeat steps 2.2.1 to 2.2.5 until all samples are removed from the sample set.

[0135] After the calculations in steps 2.2.1 to 2.2.6, the original Each sample will get There are several cluster centers, and each cluster center is easily represented as a 1x7 matrix. Furthermore, the cluster center to which each sample belongs can be determined during the calculation process.

[0136] Step 2.3: Based on the preliminary cluster centers, perform secondary clustering to obtain the final cluster centers.

[0137] After the calculation in step 2.2, we can obtain There are several cluster centers. However, the cluster centers calculated in step 2.2 are all based on samples, resulting in limited clustering effectiveness. A second clustering process is performed to re-cluster the cluster centers calculated in step 2.2. Using the initial cluster centers as the first set, the clustering is iteratively calculated again to obtain more accurate cluster centers. The specific steps of secondary clustering include:

[0138] Step 2.3.1: Initialize the fuzzy coefficient m, and set the allowable error. Based on experience Take 1.8, Pick .

[0139] Step 2.3.2: Calculate the membership matrix from the initial cluster centers. The membership matrix is ​​easy to obtain. For one OK The expressions for the column matrix and membership matrix are as follows:

[0140]

[0141] In the formula Line number Column data Indicates the first The nth sample pair Membership degree of each cluster center The calculation formula is as follows:

[0142]

[0143] In the formula: For the first Cluster centers, Indicates to open secondary roots. The calculation formula is as follows:

[0144]

[0145] In the formula: Indicates sample Located in The data above, For class center Located in The data above, .

[0146] Step 2.3.3: Calculate the cluster center matrix from the membership matrix. It is easy to obtain a cluster center matrix of one A matrix with 7 rows and 7 columns, cluster center matrix The expression is as follows:

[0147]

[0148] In the formula, the position is located at the th row data For the first Cluster centers ( ), The format is a 1-row, 7-column matrix. The calculation formula is:

[0149]

[0150] Step 2.3.4: Repeat steps 2.3.2 and 2.3.3. If, after a certain calculation, the membership matrix... If the change in each data point is less than the allowable error, the algorithm stops; otherwise, steps 2.3.2 and 2.3.3 are repeated for iterative calculation.

[0151] After the calculation in step 2.3, we can obtain Find new cluster centers and use them as the final cluster centers. The final membership matrix is ​​then used to determine the cluster centers. It is easy to obtain the cluster center to which each sample belongs, that is, the final cluster center.

[0152] Step 3: Based on the training results of Step 2, identify and output the cluster center representation state according to the final cluster center and the preset state identification rules;

[0153] The preset state identification rules include:

[0154] The first condition is set as follows:

[0155] and ;

[0156] The second condition is set as follows:

[0157] and ;

[0158] in, Indicates the th in the final cluster center Cluster centers; They represent Component data at preset positions q, w, e, and r Indicates a preset threshold;

[0159] like If the first condition is met, the status label of the electricity meter calibration device is determined to be in normal condition.

[0160] like If the first condition is not met, but the second condition is met, then the status label of the electricity meter calibration device is determined to be in a state of alert.

[0161] like If the first and second conditions are not met, the status label of the electricity meter calibration device is determined to be in an abnormal state.

[0162] Specifically, we can obtain from step two... There are *n* cluster centers, each in the format of a 1x7 matrix. Traverse each cluster center and, based on the data in each center, identify the meaning of that cluster. (The process continues with the *n*th cluster center.) Cluster centers For example, the identification process is as follows:

[0163] Based on engineering experience, whether the detection error of an electricity meter is less than 1% is an important criterion for judging whether an electricity meter is qualified. Therefore, 1% is selected as... The state represented by the data at different positions of each cluster center is determined. However, the data at positions 1*1 and 1*7 of the cluster center are easily affected by single outliers, while the data at position 1*5 is the median, which usually meets the standard of less than 1% error. Therefore, the following conditions are used as the criteria for identifying the state of the electricity meter:

[0164] First condition: and Second condition: ,and ,in Represent The data in positions 2, 3, 5, and 6.

[0165] If the first Cluster centers If the first condition is met, it is considered to represent a normal state; if the first condition is not met, but the second condition is met, it is considered to be in a state of attention; if the first and second conditions are not met, it is considered to be in an abnormal state.

[0166] Step 4: Acquire and preprocess real-time energy meter calibration data, and identify and output the status of the real-time energy meter calibration device based on the final cluster center and the cluster center characterization status.

[0167] Step 4.1: Acquire and preprocess real-time energy meter verification data to obtain real-time energy meter verification feature vector;

[0168] Measurement with a newly acquired calibration device Each table position After collecting data, iterate through... Each table has a number of positions, and each table has a number of positions. The data is sorted in ascending order, according to... The 7 data points were indexed. After the above processing, the following can be obtained: A 1*7 real-time energy meter verification feature vector, namely A sample that can represent the characteristics of the instrument panel measurement data.

[0169] Step 4.2: Determine the status of the calibration device and the meter position.

[0170] Specifically, in this embodiment, step two can be used to obtain... For each cluster center, iterate through each sample in step 4.1 and calculate the sample... The distance to each cluster center is calculated using the same formula as in step 2.3.2. The calculation formula. After calculation... After determining the distance to each cluster center, select the cluster center corresponding to the minimum value. , considering the sample Cluster Center They have the same state. Specifically, when a feature vector to be classified is equidistant from two or more final cluster centers, the real-time energy meter detection feature vector is assigned to the final cluster center with the smallest index number among all equidistant final cluster centers, and labeled to ensure the uniqueness and certainty of the classification results. Finally, based on the state represented by each cluster center obtained in step three, the samples are confirmed. The state, i.e., the first The state of each table position.

[0171] This embodiment introduces a trend analysis model based on historical state sequences to make a dynamic final determination of the overall health status of the calibration device. The process includes:

[0172] Step 4.2.1: Calculate the current overall failure rate.

[0173] Based on the cluster center characterization status of the current batch, calculate the current overall failure rate, which represents the proportion of epitopes in abnormal states; specifically, assume that in the current batch t, ​​the number of epitopes identified as "attention state" is... The number of epitopes for "abnormal states" Divide it by the total number of table digits N to obtain the current overall failure rate. , is represented as:

[0174]

[0175] Step 4.2.2, update the state history sequence.

[0176] Current overall failure rate The data is stored in a time-series queue H of fixed length L. This queue dynamically stores the fault rate records of the most recent L energy meters, represented as:

[0177]

[0178] Step 4.2.3, calculate key trend indicators.

[0179] Based on the updated historical sequence H, calculate the historical baseline of electricity meter faults. This is the arithmetic mean of all values ​​in the historical sequence except the current value. It is expressed as:

[0180]

[0181] Calculate the trend of electricity meter faults The value represents the dispersion of the electricity meter failure rate record, that is, the difference between the current overall failure rate and the historical baseline of the electricity meter failure rate, expressed as:

[0182]

[0183] Calculate the historical fluctuations of electricity meter faults That is, the standard deviation of all values ​​in the historical sequence except the current value, expressed as:

[0184]

[0185] Step 4.2.4: Perform multidimensional intelligent decision-making.

[0186] The status of the real-time energy meter calibration device is determined by applying composite rules according to a preset priority order:

[0187] Rule A (Absolute Threshold Rule): If If the absolute fault threshold is greater than the preset threshold, the percentage of time the meter position is in a state of alert or abnormality when the calibration device is functioning normally is usually less than 20%. For example, if the absolute fault threshold is set to 20%, it is judged as "calibration device failure".

[0188] Rule B (Rapid Deterioration Rule): If If the value exceeds the preset warning threshold (e.g., 10%), and If the slope exceeds the preset threshold, it is determined to be a "faulty calibration device".

[0189] Rule C (Statistical Significance Rule): If If k is a multiple (e.g., k=3), then it is determined as "the calibration device is faulty".

[0190] Rule D (Default Rule): If none of the above fault rules are met, the problem is identified as a "tablet problem".

[0191] Finally, the system outputs a specific status diagnosis for each individual station, as well as a final overall status judgment for the entire calibration device based on the aforementioned trend analysis model.

[0192] Specifically, the implementation process of this patent is illustrated with examples as follows:

[0193] Table 1. Five preliminary cluster centers.

[0194]

[0195] After the secondary clustering in step 2.3, five precise cluster centers were finally obtained, as shown in Table 2.

[0196] Table 2 shows the five final cluster centers.

[0197]

[0198] The calibration device has 16 gauge positions. Each test can test 100 gauges per gauge position. A total of 10 tests were performed, resulting in a total of 16,000 accuracy data.

[0199] Step 1: The data processing method shown in Step 1 yields 160 samples, each of which is a 1*7 matrix.

[0200] Step 2: After the calculations in Step 2.1 and Step 2.2, five cluster centers are initially obtained, as shown in Table 1.

[0201] Furthermore, secondary clustering reveals which cluster center each sample belongs to. The clustering results are as follows: Figure 2 As shown, a magnified view of the clustering results is as follows: Figure 3 As shown in the figure, one shape represents a sample belonging to a certain cluster center. It can be seen from the figure that there are five shapes in total, that is, five cluster centers and five classes. (The sample is 7-dimensional data, and only 3Ds 1, 4 and 7 are selected for plotting).

[0202] Depend on Figure 2 and Figure 3 As can be seen, the algorithm proposed in this invention can effectively identify abnormal states and attentional states that are similar to normal states. Traditional clustering algorithms, on the other hand, either struggle to identify attentional states or cluster a large number of normal state samples and abnormal state samples into the same class, making it difficult to achieve good clustering results.

[0203] Step 3: Determine the state meaning of the obtained cluster centers. Cluster center 1 corresponds to samples where more than 90% are normal samples, therefore cluster center 1 is judged to be normal. Cluster centers 2 and 3 correspond to samples where more than 90% are attention samples, therefore cluster centers 2 and 3 are judged to represent attention. Cluster centers 4 and 5 correspond to samples where more than 90% are abnormal samples, therefore cluster centers 4 and 5 are judged to represent abnormal.

[0204] After processing in step four, the 16*100 newly obtained data from the testing device are processed to obtain 16 new samples, i.e., 16 1*7 matrices. Each sample is traversed, and its distance to all cluster centers is calculated. If a sample is closest to a certain cluster center, it is considered to belong to that cluster center. The clustering results of the test samples are shown in Table 3. Of the 16 samples, 15 belong to cluster center 1, and 1 belongs to cluster center 2. Combining this with the judgment results from step three, we can conclude that 15 of the 16 epitopes are in a normal state, and 1 epitope is in a state of attention. The proportion of attention or abnormal epitopes is less than 20%, indicating that the test system is normal.

[0205] Table 3 Clustering results of test samples

[0206]

[0207] The beneficial effects of this invention are as follows:

[0208] Improved monitoring efficiency: Traditional manual inspections are inefficient and cannot monitor in real time, while this invention can achieve fully automatic real-time monitoring, which can obtain the operating status information of the electricity meter calibration device in a timely manner. It eliminates the need for frequent on-site manual inspections, greatly saving manpower and time costs, and can detect potential problems at the first time, thus significantly improving monitoring efficiency.

[0209] Improving the accuracy of state identification: This invention innovates by employing a data clustering model based on electricity meter features. Traditional clustering algorithms, such as K-Means, often perform poorly when dealing with non-spherical or unevenly dense datasets. Furthermore, the clustering effect of traditional algorithms depends on the selection of initial cluster centers, which is generally based on sample distance. When dealing with large amounts of complex data, the initial clustering effect is poor, making it difficult to achieve ideal clustering results in subsequent processes. The electricity meter feature clustering model of this invention uses sample density as the clustering criterion, which more accurately reflects the data distribution characteristics and achieves better clustering results when dealing with large amounts of complex operational data generated by electricity meter verification devices. Meanwhile, the electricity meter feature clustering model can obtain the cluster center of each class, which can directly provide high-quality initial values ​​for the subsequent clustering process. This enables the electricity meter feature clustering model to perform preliminary clustering and secondary clustering based on the preliminary cluster centers, which can more accurately cluster the data. This allows for more accurate identification of the status of the verification device in a large amount of complex data, effectively avoiding misjudgments of electricity meter performance caused by verification device failure, and ensuring the accuracy and fairness of electricity metering.

[0210] Example 2:

[0211] This invention provides a status identification system for an electricity meter calibration device. The device is used to implement the steps of the method in Embodiment 1 above. The device specifically includes:

[0212] The historical electricity meter calibration data processing module is used to acquire and preprocess historical electricity meter calibration data to obtain historical electricity meter calibration feature vectors.

[0213] The historical electricity meter verification data clustering module is used to perform preliminary clustering based on the historical electricity meter verification feature vector using an electricity meter feature clustering model. This includes a two-stage screening process combining a high-density screening set and a centripetal score to determine the first preliminary cluster center, and iteratively determining new preliminary cluster centers based on multi-factor combined weights. The final cluster centers are obtained by performing secondary clustering on the preliminary cluster centers.

[0214] The historical electricity meter calibration data status characterization module is used to identify and output the cluster center characterization status based on the final cluster center and the preset status identification rules.

[0215] The real-time energy meter calibration data status characterization module is used to acquire and preprocess real-time energy meter calibration data. Based on the final cluster center and the status characterization of the cluster center, it uses a trend analysis model to establish and analyze a time series queue composed of current and historical failure rates, calculates the historical baseline of energy meter failures, the trend of energy meter failure changes, and the historical volatility of energy meter failures, and identifies and outputs the status of the real-time energy meter calibration device.

[0216] Optionally, the status of the real-time energy meter verification device can be determined using a trend analysis model based on a fixed threshold. This includes:

[0217] First, the total number of gauges identified as being in a "caution" or "abnormal" state is counted. Then, this number is compared with a preset equipment failure threshold. For example, based on practical engineering experience, this threshold can be set to 20% of the total number of gauges N. If the number of gauges in a "caution" or "abnormal" state is less than 20% of the total number of gauges, the gauge is considered to have a problem; if the number is greater than or equal to 20%, the calibration device is considered to have malfunctioned.

[0218] However, methods based on fixed thresholds cannot identify dynamic deterioration trends in equipment status, which may lead to delays in fault warnings. Therefore, it is preferable to use a trend analysis model based on Example 1 for the final equipment status determination.

[0219] By adopting this trend analysis model, the real-time energy meter verification data status characterization module of the present invention performs better because it does not rely on a fixed threshold. It can make intelligent decisions by combining the historical performance of the equipment and recent change trends, thereby identifying potential equipment failure risks earlier and more accurately, and realizing a technological improvement from static diagnosis to dynamic early warning.

[0220] like Figure 4As shown, firstly, historical measurement data of the electricity meter calibration device is acquired, and feature extraction is performed to generate a historical feature vector that represents the statistical distribution characteristics of the measurement event. Based on all extracted historical feature vectors, a two-stage clustering process is performed to determine the final cluster centers. In the first stage, an electricity meter feature clustering model is used to determine all subsequent preliminary cluster centers. In the second stage, these preliminary cluster centers are used as initial values ​​for secondary clustering to obtain the final cluster centers and the corresponding clustering results, including the number of clusters. According to a set of preset judgment rules and thresholds, a cluster center characterization state is assigned to them. Finally, after acquiring real-time electricity meter calibration data, feature extraction is performed to identify cluster centers, and based on the cluster center characterization state, a final judgment is made on the health status of the entire calibration device. The state identification system for an electricity meter calibration device provided in this embodiment of the invention is based on the same technical concept as the state identification method for an electricity meter calibration device provided in Embodiment 1, and can produce the beneficial effects described in Embodiment 1. For details not described in this embodiment, please refer to Embodiment 1.

[0221] Example 3:

[0222] An electronic device provided by an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the steps of any of the methods described in Embodiment 1.

[0223] Example 4:

[0224] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments.

[0225] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0226] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0227] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0228] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0229] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A state identification method of an electric energy meter calibration device, characterized by, The method comprises the following steps: Step 1: Obtain and preprocess historical electric energy meter calibration data to obtain historical electric energy meter calibration feature vectors; Step 2: Based on the historical electric energy meter calibration feature vectors, use an electric energy meter feature clustering model for preliminary clustering, including a two-stage screening process combining high-density screening set and centripetal score to determine a first preliminary clustering center, and determining a new preliminary clustering center according to multi-factor combination weight iteration; Carry out secondary clustering on the preliminary clustering center to obtain a final clustering center; Step 3: Based on the final clustering center and a preset state recognition rule, identify and output the state represented by the clustering center; Step 4: Obtain and preprocess real-time electric energy meter calibration data, use a trend analysis model to establish and analyze a time series queue composed of current and historical failure rates according to the final clustering center and the state represented by the clustering center, calculate the electric energy meter failure history baseline, the electric energy meter failure change trend and the electric energy meter failure history volatility, and identify and output the real-time electric energy meter calibration device state.

2. The electric energy meter calibration device state recognition method according to claim 1, wherein: The historical electric energy meter calibration data comprises: Historical precision readings, K times of detection are performed on an electric energy meter calibration device with N table positions, and M historical precision readings are generated on each table position in each detection; A historical state label corresponding to a specific table position in a specific detection, and indicating that the specific table position is in one of the normal state, the abnormal state or the attention state during the specific detection.

3. The electric energy meter calibration device state recognition method according to claim 2, wherein: The process of preprocessing the historical electric energy meter calibration data comprises: Organizing the historical precision readings into K*N electric energy meter calibration data groups; wherein each electric energy meter calibration data group is composed of M historical precision readings corresponding to a specific table position in a specific detection; For each electric energy meter calibration data group, a historical electric energy meter calibration feature vector is constructed by sorting the M historical precision readings in the group and extracting values from a preset position in the sorted historical precision reading sequence.

4. The electric energy meter calibration device state recognition method according to claim 3, wherein: The process of determining the first preliminary clustering center comprises: Calculating the sample density of each vector in the historical electric energy meter calibration feature vector, and determining a high-density candidate set composed of vectors with a density ranking of P% according to the sample density; For each candidate vector in the high-density candidate set, calculate its average distance with all historical electric energy meter calibration feature vectors to obtain a centripetal score; the candidate vector with the smallest centripetal score in the high-density candidate set is determined as the first preliminary clustering center.

5. The electric energy meter calibration device state recognition method according to claim 4, wherein: According to the first preliminary clustering center, remove the historical electric energy meter calibration feature vectors belonging to the first preliminary clustering center, comprising: calculating a sample set average distance between the historical electric energy meter calibration feature vectors; calculating the distance between each of the historical electric energy meter calibration feature vectors and the first primary clustering center; identifying all historical electric energy meter calibration feature vectors with a distance to the first primary clustering center less than the sample set average distance as first primary clustering center members; removing the first primary clustering center members from the historical electric energy meter calibration feature vectors to obtain a remaining feature vector set.

6. The state identification method of the electric energy meter calibration device according to claim 5, characterized in that: the step of determining a new primary clustering center according to multi-factor combined weight iteration comprises: Step 2.2.1: calculating a sample set average distance of the remaining feature vector set, denoted as: wherein: is the average distance of the sample set, , are the first and the first vectors of the remaining set of feature vectors, respectively, , denote the distance between the first and the first vectors. Step 2.2.2: traversing the remaining feature vector set to calculate the sample density and the intra-class average distance of each vector; Step 2.2.3: traversing the remaining feature vector set to calculate the class distance of each sample according to the sample density; Step 2.2.4: calculating the multi-factor combined weight of each vector using an adaptive weight coefficient according to the sample density, the intra-class average distance and the class distance; wherein the adaptive weight coefficient is dynamically adjusted according to the ratio of the current remaining historical electric energy meter calibration data quantity to the initial historical electric energy meter calibration data quantity; Step 2.2.5: determining a vector with the maximum multi-factor combined weight as the new primary clustering center; calculating the distance between each of the historical electric energy meter calibration feature vectors and the new primary clustering center; identifying all remaining feature vector sets with a distance to the new primary clustering center less than the sample set average distance as new primary clustering center members; removing the new primary clustering center members from the remaining feature vector set to obtain an updated remaining feature vector set; Step 2.2.6: repeating Steps 2.2.1 to 2.2.5 until all vectors are removed from the remaining feature vector set, and outputting the primary clustering center.

7. The state identification method of the electric energy meter calibration device according to claim 6, characterized in that: the preset state identification rule comprises: setting a first condition, denoted as: and ; setting a second condition, denoted as: and ; wherein, represents the qth cluster center in the final cluster centers; respectively represents component data at preset qth, wth, eth, and rth positions, represents a preset threshold value; If If the first condition is met, the state label of the electric energy meter testing device is determined as normal state. If If the first condition is not met but the second condition is met, the state label of the electric energy meter testing device is determined as an attention state. If If the first condition and the second condition are not met, the state label of the electric energy meter testing device is determined as an abnormal state.

8. The state identification method of the electric energy meter calibration device according to claim 7, characterized in that: the trend analysis model comprises: calculating a current comprehensive failure rate representing the proportion of abnormal state positions according to the state represented by the clustering center of the current batch, and storing the current comprehensive failure rate in a time sequence queue; the time sequence queue is used to store the records of the electric energy meter failure rates in the last L times; Based on the electric energy meter failure rate record, an electric energy meter failure history baseline, an electric energy meter failure change trend and an electric energy meter failure history volatility are calculated; the electric energy meter failure history baseline represents an arithmetic mean of the electric energy meter failure rate records except the current comprehensive failure rate; the electric energy meter failure change trend represents a difference between the current comprehensive failure rate and the electric energy meter failure history baseline; and the electric energy meter failure history volatility represents a dispersion degree of the electric energy meter failure rate records; According to the electric energy meter failure history baseline, the electric energy meter failure change trend and the electric energy meter failure history volatility, a composite rule is applied to determine a real-time electric energy meter calibration device state in a preset priority order.

9. A state identification system of an electric energy meter calibration device, which implements the state identification method of any one of claims 1-8, characterized in that, Comprise: A historical electric energy meter calibration data processing module is configured to acquire and pre-process historical electric energy meter calibration data to obtain a historical electric energy meter calibration feature vector; A historical electric energy meter calibration data clustering module is configured to perform preliminary clustering using an electric energy meter feature clustering model based on the historical electric energy meter calibration feature vector, including determining a first preliminary clustering center through a two-stage screening process combining a high-density screening set and a centripetal score, and determining a new preliminary clustering center according to a multi-factor combination weight iteration; Based on the preliminary clustering center, a secondary clustering is performed to obtain a final clustering center; A historical electric energy meter calibration data state representation module is configured to identify and output a clustering center state representation based on the final clustering center and a preset state recognition rule; A real-time electric energy meter calibration data state representation module is configured to acquire and pre-process real-time electric energy meter calibration data, establish and analyze a time series queue composed of current and historical failure rates using a trend analysis model according to the final clustering center and the clustering center state representation, calculate an electric energy meter failure history baseline, an electric energy meter failure change trend and an electric energy meter failure history volatility, and identify and output a real-time electric energy meter calibration device state.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is loaded into the processor to implement the state recognition method of the electric energy meter calibration device according to any one of claims 1-8.

11. A computer-readable storage medium storing a computer program, wherein the computer program comprises the following steps of: The computer program is executed by the processor to implement the state recognition method of the electric energy meter calibration device according to any one of claims 1-8.

Images