An online calibration control method for a smart meter communication module

Abstract

The present invention relates to the technical field of online calibration control of smart meter communication modules, and specifically to an online calibration control method of a smart meter communication module; the method generates a training set by acquiring historical data of the smart meter, and uses the training set to train an abnormal monitoring model of the meter. The abnormal monitoring model is trained by a neural network, and clustering, feature learning and weighting mechanisms are used to identify abnormal conditions in the meter data. By calculating and analyzing the current meter data and historical data, the abnormal state of the meter and its influencing factors are identified, and the cause of the abnormality is analyzed. In addition, the meter data and abnormal influencing data are corrected by an error compensation method to achieve accurate online calibration. The method ensures that the meter is always in a normal working state during operation by continuously optimizing and updating the abnormal monitoring model of the meter, thereby improving the accuracy and reliability of the meter and providing an effective solution for online calibration and abnormal monitoring of smart meters.

Description

Technical Field

[0001] The present invention relates to the technical field of online calibration control of a smart meter communication module, and in particular to an online calibration control method of a smart meter communication module. Background Art

[0002] With the rapid development of computer technology, the design and function of electricity meters have changed significantly. Traditional mechanical meters mainly rely on dials and pointers to display electricity consumption, while today's meters use digital technology, with stronger functions and higher accuracy.

[0003] A smart meter is a modern electricity metering device that integrates multiple functional modules. It not only has the basic functions of a traditional meter (such as electric energy measurement), but can also improve the level of intelligent power management through advanced communication technology, data storage, remote monitoring and other functions. The communication module is one of the cores for the meter to achieve efficient, reliable and intelligent operation.

[0004] In the smart meter system, the communication module plays a vital role. It not only supports the core functions such as data transmission, remote control and real-time monitoring between the meter and the external system, but also plays a key role in the online calibration of the meter. Since smart meters may be affected by various factors during long-term use, resulting in errors in their measurement parameters, regular meter calibration is an important measure to ensure the accuracy and stability of the meter.

[0005] At present, a variety of methods have been proposed for online calibration of smart meters. Among them, most methods usually only calibrate by comparing the collected data before and after when the meter is abnormal. Although these methods can improve the accuracy of the meter to a certain extent, they often ignore the type of meter abnormality and its potential influencing factors, resulting in errors in the calibration results, thereby affecting the calibration accuracy and long-term stability. Therefore, how to perform more accurate calibration based on different abnormality types and influencing factors is still a major challenge in current research.

[0006] To this end, the present invention proposes an online calibration control method for a smart meter communication module. Summary of the invention

[0007] The purpose of the present invention is to provide an online calibration control method for a smart meter communication module, which is used for online calibration of a smart meter when an abnormality occurs. The present invention mainly starts from the following points. First, the present invention proposes a meter abnormality monitoring model for real-time abnormality monitoring of smart meters. The model includes: an input layer, a clustering cluster feature learning layer, an abnormal feature weighting layer, a meter abnormality recognition layer and an output layer. Among them, by clustering analysis, feature extraction and weighting of smart meter data, and combining neural network for abnormal classification and recognition, the abnormal situation in the meter data can be detected in real time. During the training process, the model optimizes clustering features, key features and error matrices, and uses the back propagation method to continuously adjust parameters, and finally realizes efficient and accurate abnormality detection, thereby improving the reliability and safety of meter operation. Secondly, based on the abnormal monitoring results of the smart meter, the present invention proposes a method for analyzing abnormal influencing factors. The method constructs parameter abnormality vectors and normal data vectors by real-time monitoring of the abnormal situation of the meter and obtaining the meter data at the current and previous moments, and continuously monitors until the meter state is normal. According to the difference between the current abnormal vector and the normal vector, the candidate abnormal influencing factors are analyzed, the influencing factor index is established, and the data change time and abnormal impact change data are calculated. The abnormal influencing factors are comprehensively analyzed through the meter abnormal cause analysis function, and the abnormal cause analysis results are output. Finally, the abnormal influencing factors are identified through abnormal analysis and the importance and correlation analysis are performed to obtain the abnormal impact weight vector and correlation matrix; based on the slicing and change calculation of the meter data and the abnormal impact data, the extended data set is merged and compensated through error calculation, thereby improving the accuracy of the online calibration of the smart meter data.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] An online calibration control method for a smart meter communication module, comprising:

[0010] Obtain historical data of the smart meter; wherein the historical data includes: meter usage time, power data, current data, voltage data, power data, frequency data, temperature data and load data.

[0011] Further, the historical data is preprocessed to obtain standard historical feature data; wherein the preprocessing process of the historical data includes: performing data cleaning on the historical data to obtain first historical data; performing noise processing on the first historical data to obtain second historical data; performing data conversion on the second historical data to obtain third historical data; performing normalization processing on the third historical data to obtain fourth historical data; performing feature construction on the fourth historical data to obtain the standard historical feature data;

[0012] Further, the standard historical feature data is marked and integrated to obtain a training set;

[0013] Furthermore, the training set is used to train the meter anomaly monitoring model; wherein the training process of the meter anomaly monitoring model includes: using an input layer to receive data samples in the training set; wherein the input layer includes: a data receiving sublayer and a data clustering sublayer; the data receiving sublayer is used to receive data; the data clustering sublayer is used to cluster the data samples to obtain K cluster sets;

[0014] Further, the K cluster sets are input into a cluster feature learning layer to obtain K cluster feature vectors;

[0015] The cluster feature learning layer includes: obtaining data samples in the same cluster set; inputting the data samples into the trained neural network to obtain abnormal classification results; determining abnormal factors according to the abnormal classification results; obtaining key feature vectors of each data sample in the cluster set according to the abnormal factors, and constructing a key feature matrix; calculating the error between the features in the key feature matrix and the non-abnormal features to obtain a key feature error matrix; calculating the cluster feature vector according to the key feature error matrix; wherein the calculation formula of the cluster feature vector is: ;in, The cluster set The clustering feature vector of The cluster set The data sample size; The cluster set Medium data sample The key feature vector of is the weight coefficient, which is used to balance the influence of key eigenvectors and errors; The cluster set Medium data sample The key feature error vector.

[0016] Furthermore, the K cluster sets and the K cluster feature vectors are input into an abnormal feature weighted layer to obtain K weighted cluster sets; wherein the weighting function of the abnormal feature weighted layer is: ;in, The cluster set The weighted data sample ; For data samples ; For clustering set Medium data sample The weighting coefficient of .

[0017] Wherein, the formula of the weighted coefficient is: ;in, For data samples The degree of abnormality; For data samples and cluster feature vector distance; is a hyperparameter.

[0018] Further, the K weighted cluster sets are input into the meter abnormality recognition layer to obtain the abnormality recognition probability of each data sample in the K weighted cluster sets;

[0019] Furthermore, the abnormality recognition probability is outputted by the output layer to obtain an abnormality monitoring result, including: monitoring status (normal / abnormal) and abnormality type;

[0020] Further, using a loss function to calculate the error between the abnormal monitoring result and the mark;

[0021] Furthermore, the parameters of the electric meter abnormality monitoring model are updated by using a back propagation method, and when the loss value reaches convergence, the training of the electric meter abnormality monitoring model is completed.

[0022] Further, the communication module is used to collect current meter data;

[0023] Further, the current meter data is transmitted to the trained meter abnormality monitoring model to obtain a current abnormality monitoring result; wherein the current abnormality monitoring result;

[0024] Further, analyzing the current abnormal monitoring result to obtain an abnormal analysis result;

[0025] The process of analyzing the current abnormal monitoring result includes: obtaining the current meter data corresponding to the abnormality according to the current abnormal monitoring result, and constructing the current meter abnormal data vector; obtaining multiple meter data of the smart meter at the previous moment;

[0026] Further, multiple parameters at the previous moment are input into the meter abnormality monitoring model to obtain the meter abnormality monitoring result; when the monitoring result shows normal, the multiple parameters at the previous moment are used as normal meter data, and the meter normal data vector is constructed; otherwise, the previous meter data is continuously collected, and the meter abnormality monitoring model is used for continuous monitoring until the monitoring result is normal;

[0027] Further, according to the current electric meter abnormal data vector, a candidate abnormal influencing factor field is obtained;

[0028] Further, according to the candidate abnormal influencing factor field, a candidate abnormal influencing factor index is established;

[0029] Further, calculating the data change time between the normal data vector of the electric meter and the abnormal data vector of the current electric meter;

[0030] Further, calculating abnormal impact change data about the candidate abnormal impact factor field within the data change time;

[0031] Furthermore, the current meter abnormal data vector, the meter normal data vector, the candidate abnormal influencing factor index, the data change time, the abnormal impact change data, and the abnormal impact change data are input into the meter abnormal cause analysis function to obtain the abnormal analysis result. The meter abnormal cause analysis function is expressed as: ;in, The abnormal reason analysis vector of the electric meter at the current moment; is the modulus of the abnormal data vector of the electric meter at the current moment; is the data change time; is the candidate abnormal influencing factor at the current moment The modulus of the change in is the candidate abnormal influencing factor at the current moment; is the total number of candidate abnormal factors; is the influence function of the size of the meter abnormal data vector on the meter abnormal cause analysis; is the influence function of the change of the candidate abnormal influencing factors on the abnormality of the electric meter; The action function for the candidate abnormal influencing factor index.

[0032] Further, according to the abnormal analysis result, the abnormal influencing factor is obtained;

[0033] Further, abnormal importance analysis and abnormal correlation analysis are performed on the abnormal influencing factors to obtain an abnormal influence weight vector and an abnormal influence correlation matrix;

[0034] The calculation process of the abnormal impact weight vector includes: characterizing the abnormal impact factor to obtain an abnormal impact factor feature vector;

[0035] Further, setting an abnormality target of the electric meter according to the current abnormality monitoring result;

[0036] Furthermore, the influence probability of each feature in the abnormal influencing factor feature vector on the abnormal target of the electric meter is calculated; wherein the calculation formula of the influence probability is: ;in, Abnormal influencing factors The eigenvector of It is the abnormal target of the electric meter; Abnormal influencing factors Probability of impact on abnormal targets of electric meters; Abnormal influencing factors The eigenvalues ​​in the eigenvector of Abnormal target value of electric meter The joint probability of Abnormal influencing factors The eigenvalues ​​in the eigenvector of The marginal probability of is the abnormal target value of the electric meter The marginal probability of

[0037] Furthermore, the impact probabilities are sorted in descending order to obtain the abnormal impact weight vector.

[0038] Furthermore, the electric meter data and abnormal impact data within a continuous period of time are obtained;

[0039] Further, the meter data and the abnormal impact data are sliced, and a first data set is established according to the meter state; the change coefficients of the meter data and the abnormal impact data in the first data set are calculated respectively to obtain a meter data change vector and an abnormal impact data change vector;

[0040] Further, the first data set is expanded using the meter data change vector and the abnormal impact data change vector to obtain a second data set;

[0041] Further, an error calculation is performed according to the abnormal influence weight vector, the abnormal influence association matrix and the second data set to obtain an error compensation result; wherein the calculation of the error compensation result adopts an iterative error compensation calculation method, including: obtaining the abnormal influence weight vector and the abnormal influence association matrix, respectively denoted as and ;

[0042] Further, an error function is defined according to the meter data change vector and the abnormal impact data change vector in the second data set; wherein the error function is: ;in, For the The error compensation result at the current moment after iteration; is the meter change vector at the current moment; is the data change vector affected by the abnormality at the current moment; For the Error compensation coefficient after iteration; For the Correction error constant after iteration;

[0043] Further, the gradient of the error function is calculated and recorded as ;

[0044] Furthermore, the error compensation coefficient is updated according to the gradient, and the formula is: ;in, For the The error compensation coefficient at the iteration; is the learning rate;

[0045] Further, the correction error constant is updated, and the formula is: ;in, For the The correction error constant at the iteration;

[0046] Furthermore, the error compensation result is recalculated according to the updated error compensation coefficient and the corrected error constant; when the error compensation result has the best adjustment effect on the smart meter, the final error compensation result is obtained.

[0047] Furthermore, the smart meter is calibrated online according to the error compensation result.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention proposes a smart meter anomaly monitoring model for real-time monitoring of abnormal conditions of the meter. The model includes an input layer, a cluster feature learning layer, an abnormal feature weighting layer, a meter anomaly identification layer and an output layer. The model performs cluster analysis, feature extraction and weighting on the smart meter data, and combines neural networks for anomaly classification and identification, thereby achieving real-time monitoring of abnormal conditions in the meter data; in addition, the monitoring model is also closely integrated with the online calibration system of the smart meter communication module, so that timely reminders can be made, and the monitoring results are conducive to the subsequent online calibration of the smart meter, thereby improving the accuracy of the calibration.

[0050] 2. The present invention proposes a method for analyzing abnormal influencing factors. The method monitors the abnormal situation of the electric meter in real time, obtains the electric meter data at the current moment and the previous moment, constructs the parameter abnormality vector and the normal data vector. In this process, according to the difference between the current abnormal vector and the normal vector, the potential abnormal influencing factors are analyzed, and the influencing factor index is established. At the same time, the data change time and the abnormal impact change data are calculated to identify and quantify the key factors affecting the abnormality of the electric meter. When the electric meter detects an abnormality, the abnormality analysis can accurately identify the abnormality type and its potential influencing factors, thereby providing an effective calibration reference for the communication module.

[0051] 3. The present invention proposes an error compensation method for online calibration control of smart meter communication modules; the method identifies key factors affecting abnormalities through abnormal analysis, and performs importance and correlation analysis to obtain the weight vector and correlation matrix of abnormal impact. Based on the slice and change calculation of meter data and abnormal impact data, the data set is expanded, and effective compensation is performed through error calculation, thereby improving the online calibration accuracy of smart meter data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of an online calibration control method for a smart meter communication module provided by an embodiment of the present invention;

[0053] Figure 2 A structural diagram of an abnormality monitoring model for an electric meter provided in an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of the training loss of the electric meter anomaly monitoring model provided in an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of changes in the coefficients of change of the electric meter data and abnormal impact data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Smart meter is a modern electricity metering device that integrates multiple functional modules. It not only has the basic functions of traditional meters (such as electric energy measurement), but also can improve the intelligent level of power management through advanced communication technology, data storage, remote monitoring and other functions. Its components usually include: measurement module, communication module, display module, storage module and operation module;

[0058] In smart meters, the communication module is crucial, as it directly affects functions such as data transmission between the meter and the external system, remote control, and real-time monitoring. In addition, the communication module is also used to implement online calibration of smart meters. Since parameter errors may occur during the use of smart meters, regular calibration of the meter is an important measure to ensure its accuracy and stability.

[0059] At present, researchers have proposed a variety of online calibration methods for smart meters; most methods only calibrate by comparing the collected data before and after when the meter is abnormal. However, these methods often ignore the type of meter abnormality and its influencing factors, resulting in low calibration accuracy. To this end, the present invention proposes an online calibration control method for a smart meter communication module. The present invention will be described in detail from two embodiments below;

[0060] Embodiment 1:

[0061] In the embodiment of the present application, the method proposed in the present invention is used to calibrate the smart meter A online; wherein the smart meter A can monitor and record household electricity consumption in real time, and collect and transmit data through the communication module; according to Figure 1 The contents of the present application are described in detail. Figure 1 The method steps of the present invention are described in the description, including: S10. Obtaining the historical data of the smart meter; S20. Preprocessing the historical data to obtain standard historical feature data; S30. Marking the standard historical feature data and integrating them to obtain a training set; S40. Using the training set to train the meter abnormality monitoring model; S50. Using the communication module to collect the current meter data; S60. Transmitting the current meter data to the trained meter abnormality monitoring model to obtain the current abnormality monitoring result; S70. Analyzing the current abnormality monitoring result to obtain the abnormality analysis result; S80. Performing error compensation calculation on the smart meter to obtain the error compensation result; S90. Online calibration of the smart meter according to the error compensation result. The specific implementation process includes: collecting the historical data of the smart meter A according to the content of step S10; wherein the historical data includes: meter usage time, power data, current data, voltage data, power data, frequency data, temperature data and load data. Refer to Table 1, which gives the content and description of each historical data;

[0062]

[0063] In the embodiment of the present application, historical data of smart meter A is obtained, including: meter usage time, power data, current data, voltage data, power data, frequency data, temperature data and load data, so as to perform a more comprehensive abnormality analysis on smart meter A later.

[0064] Furthermore, the historical data of smart meter A is preprocessed, corresponding to step S20; the specific process includes: performing data cleaning on the historical data to obtain first historical data; performing noise processing on the first historical data to obtain second historical data; performing data conversion on the second historical data to obtain third historical data; performing normalization processing on the third historical data to obtain fourth historical data; performing feature construction on the fourth historical data to obtain the standard historical feature data.

[0065] For example, (1) Data cleaning (obtaining the first historical data): ensure the accuracy and completeness of the data by eliminating missing values, duplicate values, outliers, or invalid data; interpolation can be used to fill in missing data, or records with too much missing data can be deleted.

[0066] (2) Noise processing (obtaining the second historical data): In response to possible noise in the data (such as sensor error or external interference), filtering technology (such as median filtering, low-pass filtering, etc.) is used to smooth the data, reduce interference, and improve the signal-to-noise ratio of the data.

[0067] (3) Data transformation (obtaining third historical data): transform the data to adapt it to the needs of subsequent analysis. Common transformation methods include: Logarithmic transformation: scale the data logarithmically to deal with the problem of uneven data distribution. Differential transformation: used to remove the trend or seasonality of the data to make it more stable.

[0068] (4) Normalization processing (obtaining the fourth historical data): Scale the data to a uniform range (such as [0, 1]) to eliminate the dimensional differences between different features and avoid unnecessary influence of certain features on the analysis results due to their large values; commonly used methods include Z-score normalization.

[0069] (5) Feature construction (obtaining standard historical feature data): Based on the preprocessed data, feature extraction and construction are performed to form standard historical feature data for analysis. Features include the mean, standard deviation, maximum value, minimum value, volatility, etc. of the time series, or new composite features (such as power factor, load change rate, etc.) are generated by combining existing data.

[0070] In the embodiments of the present application, historical data is preprocessed by means of data cleaning, noise processing, data conversion, normalization processing, and feature construction. The above preprocessing steps can make historical data more standardized, clean, and usable, thereby providing more accurate and reliable support for subsequent analysis, modeling, and decision-making.

[0071] Further, the standard historical feature data is marked according to the content in step S30, and integrated to obtain a training set; wherein the marking of the standard historical feature data is based on the expert's abnormal analysis of the historical data, and the marking content is whether there is an abnormality and the type of abnormality; as shown in Table 2, Table 2 gives a description of some data in the historical data;

[0072]

[0073] Table 2 gives the timestamp, anomaly label and anomaly type of the historical data. In the anomaly label, "0" indicates normal; "1" indicates abnormal. The anomaly types include voltage fluctuation, current overload and temperature abnormality. The above anomaly types are only some examples, and multiple anomaly types may be included in the actual implementation process.

[0074] Taking the timestamp "2024-01-11 13:30" as an example, the marking process is as follows: Original data: The anomaly mark of the timestamp "2024-01-1113:30" is 1, and the anomaly type is "voltage fluctuation". Anomaly marking: Since the data is abnormal, the anomaly mark is 1. Anomaly type representation: Assume that in actual operation, there are four types of anomalies, namely: voltage fluctuation, current overload, temperature anomaly, and high humidity (assumption); the corresponding marking form should be a four-dimensional binary vector, where: if "voltage fluctuation" occurs at a time point, the bit is 1, and other anomaly types are 0. For the anomaly type "voltage fluctuation" of "2024-01-11 13:30", its label should be: [1 0 0 0], indicating that the voltage fluctuation is abnormal.

[0075] Further, according to step S40, the electric meter abnormality monitoring model is trained using the training set; wherein the structure of the electric meter abnormality monitoring model is referred to Figure 2 , including: input layer, cluster feature learning layer, abnormal feature weighting layer, meter abnormality recognition layer and output layer; the specific training process of the meter abnormality monitoring model includes:

[0076] The input layer is used to receive data samples in the training set; wherein the input layer includes: a data receiving sublayer and a data clustering layer; the data receiving sublayer transmits the data samples to the data clustering sublayer to obtain K clustering sets;

[0077] The clustering process of the data clustering sublayer includes the following steps:

[0078] Step 1: Randomly select K initial cluster centers; where K is a natural number greater than 1;

[0079] Step 2: Assign each data sample to the nearest initial cluster center; the data sample assignment method is determined by calculating the Euclidean distance from the initial cluster center;

[0080] Step 3: Update the center point of the initial cluster center to obtain the updated cluster center;

[0081] Step 4: Repeat steps 2 and 3 until the center point no longer changes, and obtain K cluster sets;

[0082] Furthermore, the K cluster sets are input into the cluster feature learning layer to obtain K cluster feature vectors;

[0083] The cluster feature learning layer includes: obtaining data samples in the same cluster set;

[0084] Further, the data sample is input into a trained neural network to obtain an abnormal classification result; wherein the neural network mentioned in the embodiment of the present application refers to a convolutional neural network (CNN) model trained based on data of all abnormal types in the smart meter;

[0085] Further, determining abnormal factors according to the abnormal classification result;

[0086] Further, a key feature vector of each of the data samples in the cluster set is obtained according to the abnormal factors, and a key feature matrix is ​​constructed;

[0087] Further, the error between the features in the key feature matrix and the features without abnormality is calculated to obtain a key feature error matrix; the cluster feature vector is calculated according to the key feature error matrix;

[0088] The calculation formula of the cluster feature vector is: ;in, The cluster set The clustering feature vector of The cluster set The data sample size; The cluster set Medium data sample The key feature vector of is the weight coefficient, which is used to balance the influence of key eigenvectors and errors; The cluster set Medium data sample The key characteristic error vector of

[0089] Further, the K cluster sets and the K cluster feature vectors are input into an abnormal feature weighting layer to obtain K weighted cluster sets;

[0090] Among them, the weighting function of the abnormal feature weighting layer is: ;in, The cluster set The weighted data sample ; For data samples ; For clustering set Medium data sample The weighting coefficient of

[0091] In the implementation of this application, the abnormal feature weighting layer is used to achieve the weighting of the abnormal data features of the electric meter; this layer is weighted by clustering and the corresponding cluster feature vector; this can more accurately weight and process the abnormal data of the electric meter. The clustering process helps to group data samples according to similarity, so that the data in each cluster has a higher correlation in features, and the weighted feature vector can help the model more sensitively capture subtle changes in abnormal data.

[0092] The formula of the weighting coefficient is: ;in, For data samples The degree of abnormality; For data samples and cluster feature vector distance; is a hyperparameter.

[0093] In the formula of the weighted coefficient, the abnormality of the data sample is expressed as: ;in, For data samples The sensitivity of abnormal features in the feature learning process of trained neural networks;

[0094] The weighted coefficient formula combines the distance between the sample and the cluster feature vector and the degree of abnormality. The distance reflects the similarity between the sample and the normal pattern, while the degree of abnormality reflects the sensitivity of the sample to abnormal features during the training process. By combining these two factors, the model can comprehensively consider the normality and abnormality of the sample, which helps to improve the accuracy of the meter abnormality monitoring model.

[0095] Furthermore, the K weighted cluster sets are input into the meter anomaly recognition layer to obtain the anomaly recognition probability of each data sample in the K weighted cluster sets; wherein, the above-trained neural network can continue to be used for the meter anomaly recognition layer;

[0096] Furthermore, the abnormality recognition probability is outputted by the output layer to obtain an abnormality monitoring result; wherein the abnormality monitoring result includes: monitoring state (normal / abnormal) and abnormality type;

[0097] Furthermore, a loss function is used to calculate the error between the abnormal monitoring result and the mark; wherein, in the embodiment of the present application, the loss function is set to a combination of a binary classification loss function and a cross entropy loss function, which can be expressed as:

[0098] ;

[0099] in, Expressed as loss value; is the number of exception types; For data samples The tag value of For data samples Abnormal monitoring results; and is the weight coefficient, which is used to balance the impact of these two losses.

[0100] Furthermore, the parameters of the electric meter abnormality monitoring model are updated by using a back propagation method, and when the loss value reaches convergence, the training of the electric meter abnormality monitoring model is completed.

[0101] See also Figure 3 ,exist Figure 3 The loss changes of the meter anomaly monitoring model during 200 rounds of training are given in Figure 3 It can be seen that the model tends to converge when the training is in the 200th round. In addition, the performance data of the meter anomaly monitoring model after 100 rounds of training are given in Table 3;

[0102]

[0103] Table 3 shows the performance changes of the meter anomaly monitoring model during the training process, especially focusing on the training results of the last 100 rounds (i.e., rounds 110 to 200). Through the gradual improvement of the accuracy, recall rate, F1 score, anomaly recognition accuracy, and anomaly type recognition accuracy of the training set, it can be seen that the model gradually tends to be optimized in the final stage of training, and the performance continues to improve.

[0104] Training set accuracy: gradually increased from 80% in round 110 to 95% in round 200. The increase in accuracy means that the proportion of samples that the model can correctly predict gradually increases, and the overall performance has been significantly improved.

[0105] Recall: Recall represents the model’s ability to correctly identify anomalies. It increased from 72% at 110 rounds to 89% at 200 rounds. The increase in recall indicates that the model is better able to detect all anomaly instances and reduce missed detections.

[0106] F1 score: This value increased from 77% at 110 epochs to 93% at 200 epochs, showing a balanced improvement in the model’s precision and recall.

[0107] Anomaly recognition accuracy: This metric reflects the accuracy of the model in identifying abnormal data. The accuracy of anomaly recognition increased from 80% to 95%, indicating that the model has greatly improved its accuracy in identifying anomalies.

[0108] Anomaly type recognition accuracy: This indicator reflects the accuracy of the model in identifying different types of anomalies. The accuracy of anomaly type recognition increased from 78% to 93%, indicating that the accuracy of the model in classifying different anomaly types has also been significantly improved.

[0109] Summary: From the data in Table 3, with the increase of training rounds, the performance indicators of the model have shown a steady improvement, indicating that the training process of the model is effective and gradually converges. In the 200th round, the accuracy, recall rate and F1 score of the model have reached a high level, and the accuracy of anomaly recognition and the accuracy of anomaly type recognition are also close to the optimal state, indicating that the model has basically completed training and has a strong anomaly monitoring capability.

[0110] In the implementation of this application, the meter anomaly monitoring model is used to monitor the smart meter A for anomalies; the model effectively improves the monitoring and classification capabilities of abnormal data through clustering feature learning, abnormal feature weighting and identification. By combining key feature vectors, error calculation and weighting mechanisms, the model can accurately identify different types of anomalies, and gradually converge with the optimization of back propagation during the training process, ultimately achieving efficient anomaly monitoring effects. This gives the model significant advantages in high-precision demand scenarios such as meter anomaly detection.

[0111] Further, according to Figure 1 In step S50, the current meter data of the smart meter A is collected by using the communication module;

[0112] Further, according to the content of step S60, the current meter data is input into the trained meter abnormality monitoring model to obtain the current abnormality monitoring result of the smart meter A; wherein, according to the monitoring of the meter abnormality monitoring model, the smart meter A currently has an abnormality, and the abnormality types include: abnormal current value, abnormal voltage value and abnormal energy consumption;

[0113] Further, according to step S70, an abnormality analysis is performed on the monitoring result of the smart meter according to the meter abnormality monitoring model; wherein the abnormality analysis process of the smart meter A includes:

[0114] According to the current abnormal monitoring results, obtain the current meter data corresponding to the abnormality and construct the current meter abnormal data vector; for example, obtain the current monitoring data of smart meter A as follows: (1) current current value: 10A, normal range center: 8A; (2) current voltage value: 240V, normal range center: 220V; (3) current energy consumption value: 1000kWh, normal range center: 900kWh; according to the deviation calculation, the corresponding current meter abnormal data vector can be expressed as [2,20,100];

[0115] Further, multiple electric meter data of smart electric meter A at the last monitoring moment are obtained;

[0116] Further, multiple parameters at the previous moment are input into the meter abnormality monitoring model to obtain the meter abnormality monitoring result; when the monitoring result shows normal, the multiple parameters at the previous moment are used as normal meter data, and the meter normal data vector is constructed; otherwise, the previous meter data is continuously collected, and the meter abnormality monitoring model is used for continuous monitoring until the monitoring result is normal;

[0117] During the abnormality analysis of smart meter A at the current moment, the meter status at the previous moment is checked to confirm whether the abnormality of the meter is a temporary fluctuation or emergency, rather than a long-standing fault or problem.

[0118] Further, according to the current abnormal monitoring result, a candidate abnormal influencing factor field is obtained; wherein, according to the abnormal type of smart meter A, the following influencing causes can be obtained, such as: load change, meter failure, measurement error, environmental factors, electromagnetic interference, and power consumption mode change;

[0119] Further, according to the candidate abnormal influencing factor field, a candidate abnormal influencing factor index is established; see Table 4, which gives an example description of the index;

[0120]

[0121] Further, the data change time between the normal data vector of the electric meter and the abnormal data vector of the current electric meter is calculated;

[0122] Further, calculating abnormal impact change data of the candidate abnormal impact factor field within the data change time;

[0123] Furthermore, the current meter abnormal data vector, meter normal data vector, candidate abnormal influencing factor index, data change time, abnormal influencing change data and abnormal influencing change data are input into the meter abnormal cause analysis function to obtain the abnormal analysis result.

[0124] Among them, the abnormal cause analysis function of the electric meter is expressed as: ;in, Analyze the vector quantity of the abnormal cause of the electric meter at the current moment; is the modulus of the abnormal data vector of the electric meter at the current moment; is the data change time; is the candidate abnormal influencing factor at the current moment At the moment The modulus of the change in is the candidate abnormal influencing factor at the current moment; is the total number of candidate abnormal factors; is the influence function of the size of the meter abnormal data vector on the meter abnormal cause analysis; is the influence function of the change of the candidate abnormal influencing factors on the abnormality of the electric meter; The action function for the candidate abnormal influencing factor index.

[0125] In this application, the abnormality analysis function of the electric meter is , and They are defined as follows:

[0126] ;

[0127] in, is the number of data items obtained in the smart meter; for The moment Normal data of each electric meter; For the current moment Abnormal data of electric meters;

[0128] ;

[0129] in, The dimension of each candidate abnormal influencing factor; for Abnormal factors affecting time In the The amount of change in dimension; is the abnormal influencing factor at the current moment In the The amount of change in dimension;

[0130] ;

[0131] in, For the The weight of each candidate abnormal influencing factor; For the current moment Candidate abnormal influencing factors;

[0132] In order to analyze the cause of the abnormal impact of smart meter A in the implementation of this application, the meter abnormal cause analysis function is used for calculation; this function quantifies the relationship between the meter abnormal data and the candidate abnormal influencing factors to find out the main cause of the abnormality. Among them, the core of the meter abnormal cause analysis function is to quantify the changes in various types of meter abnormal data and candidate abnormal influencing factors, and construct a multi-dimensional analysis framework to help identify and diagnose the real cause of the smart meter abnormality.

[0133] Further, according to the abnormal analysis result, the abnormal influencing factor is obtained;

[0134] Based on the above, it can be seen that the abnormal factors affecting smart meter A include the following: load changes, temperature changes, electromagnetic interference, and changes in power consumption patterns;

[0135] In the embodiment of the present application, the abnormal cause of smart meter A is analyzed by an analysis method of abnormal influencing factors; the method obtains the meter data at the current moment and the previous moment, constructs the parameter abnormality vector and the normal data vector, and in this process, analyzes the potential abnormal influencing factors according to the difference between the current abnormal vector and the normal vector, and establishes the influencing factor index. At the same time, the data change time and the abnormal impact change data are calculated to identify and quantify the key factors affecting the meter abnormality. And the meter abnormal cause analysis function takes into account multiple dimensions such as the meter abnormal data vector, candidate abnormal influencing factors and their changes, and data change time, to ensure that the analysis of the abnormality of smart meter A is more accurate and comprehensive.

[0136] Furthermore, according to the content of step S80, the error compensation calculation is performed on the smart meter A; the specific process includes:

[0137] Perform abnormal importance analysis and abnormal correlation analysis on the analyzed abnormal influencing factors to obtain the abnormal influence weight vector and abnormal influence correlation matrix;

[0138] Among them, the calculation process of the abnormal impact weight vector is:

[0139] Characterize the abnormal influencing factors to obtain the abnormal influencing factor feature vector;

[0140] Set meter abnormality targets based on current abnormality monitoring results;

[0141] The influence probability of each feature in the feature vector of the abnormal influencing factor on the abnormal target of the electric meter is calculated; wherein the calculation formula of the influence probability is: ;in, Abnormal influencing factors The eigenvector of It is the abnormal target of the electric meter; Abnormal influencing factors Probability of impact on abnormal targets of electric meters; Abnormal influencing factors The eigenvalues ​​in the eigenvector of and the abnormal target value of the electric meter The joint probability of Abnormal influencing factors The eigenvalues ​​in the eigenvector of The marginal probability of is the abnormal target value of the electric meter The marginal probability of

[0142] Sort the impact probability in descending order to obtain the abnormal impact weight vector. According to the above calculation results, the abnormal impact weight vector is shown in Table 5;

[0143]

[0144] In the embodiment of the present application, by calculating the influence weight of each factor, analyzing the factors affecting the abnormality of the smart meter, and calculating the influence weight of each factor, the key factors affecting the abnormality of the meter can be more accurately identified and quantified. Targeted consideration can be given to subsequent error compensation, and accurate error compensation can be achieved.

[0145] Among them, the abnormal impact correlation matrix adopts the big data analysis method, by using the co-occurrence value of the corresponding abnormal influencing factors when the same abnormality occurs in the same type of smart meters as the correlation value between the influencing factors; see Table 6;

[0146]

[0147] Further, the meter data and abnormal impact data of smart meter A within a continuous period of time are obtained;

[0148] Furthermore, the meter data and abnormal impact data are sliced ​​according to the time series; for example, sliced ​​by hour: the meter data has a high frequency of change, and each hour can be selected as a slice; and a first data set is established according to the meter state; wherein the first data set is expressed as: ;in, For the Slice number; For the Meter data of each slice; For the The abnormality of each slice affects the data; For the The meter status of each slice; is the number of slices, when the data subscript in the first data set is , it is represented by the current meter data, current abnormal impact data and current meter status of the smart meter;

[0149] Further, the variation coefficients of the electric meter data and the abnormal impact data in the first data set are calculated respectively to obtain the electric meter data variation vector and the abnormal impact data variation vector; Figure 4 ,exist Figure 4 The trend diagram of the meter data of smart meter A in the previous 10 hours and the coefficient of variation of abnormal impact data is given in the figure;

[0150] Further, the first data set is expanded using the meter data change vector and the abnormal impact data change vector to obtain a second data set;

[0151] Further, error calculation is performed according to the abnormal impact weight vector, the abnormal impact association matrix and the second data set to obtain an error compensation result;

[0152] The error compensation result is calculated by an iterative error compensation calculation method, including:

[0153] Obtain the abnormal impact weight vector and the abnormal impact association matrix, which are respectively denoted as and ;

[0154] Further, an error function is defined according to the meter data change vector and the abnormal impact data change vector in the second data set;

[0155] Wherein, the error function is: ;in, For the The error compensation result at the current moment after iteration; is the meter change vector at the current moment; is the data change vector affected by the abnormality at the current moment; For the Error compensation coefficient after iteration; For the Correction error constant after iteration;

[0156] Further, the gradient of the error function is calculated and recorded as ;

[0157] Furthermore, the error compensation coefficient is updated according to the gradient, and the formula is: ;in, For the The error compensation coefficient at the iteration; is the learning rate; wherein, in the embodiment of the present application, Set to 0.05;

[0158] Further, the correction error constant is updated, and the formula is: ;in, For the The correction error constant at the iteration;

[0159] Further, the error compensation result is recalculated according to the updated error compensation coefficient and the corrected error constant; when the error compensation result reaches the best adjustment effect on the smart meter A, the final error compensation result is obtained. Refer to Table 7, which gives the error compensation coefficient and corrected error constant of the last 10 rounds of iterative learning;

[0160]

[0161] Table 7 shows the coefficient changes of the error compensation results calculated for the last 10 rounds of smart meter A. It can be seen from Table 7 that when the number of iterations is the second to last round, the error compensation coefficient is 0.07; the correction error constant is 0.96; the calculated compensation error result is used for online adjustment, so that the error of smart meter A reaching the normal state is only 0.010; in the first to last round, the normal state error is improved to 0.013.

[0162] In the embodiment of the present application, a method for calculating the compensation error based on an iterative method is adopted. By slicing the meter data and the abnormal impact data in time series, the meter data, the abnormal impact data and the meter state of each slice are organized into a first data set, and then the change coefficients of the meter data and the abnormal impact data are calculated to obtain the meter data change vector and the abnormal impact data change vector. Based on these change vectors, the data set is further expanded and combined and the error is corrected by the error compensation calculation method. The core of this method is to continuously adjust and optimize the error compensation coefficient and the correction error constant through an iterative process until the error compensation result converges to the minimum value, thereby achieving accurate calibration of the meter data and effective compensation of abnormal impacts.

[0163] Further, according to the content of step S90, the smart meter A is calibrated online according to the calculated error compensation.

[0164] In this application, the method of the present invention is used to realize online abnormal monitoring and calibration of smart meter A; it mainly starts from the following aspects; first, a meter abnormality monitoring model for real-time abnormal monitoring of smart meters is adopted. The model includes the following main layers: input layer, clustering cluster feature learning layer, abnormal feature weighting layer, meter abnormality identification layer and output layer. The model performs cluster analysis, feature extraction and weighting on smart meter data, and classifies and identifies abnormalities in combination with neural network technology, so as to detect abnormal situations in meter data in real time. During the training process, the model optimizes clustering features, key features and error matrix, and uses the back propagation algorithm to continuously adjust the model parameters, and finally realizes efficient and accurate abnormality detection, thereby improving the reliability and safety of meter operation.

[0165] Secondly, based on the abnormal monitoring results of smart meter A, an analysis method of abnormal influencing factors is adopted. This method monitors the abnormal state of the meter in real time, obtains the meter data at the current moment and the previous moment, constructs the parameter abnormality vector and the normal data vector, and continuously monitors until the meter state returns to normal. By comparing the difference between the current abnormal vector and the normal vector, the potential abnormal influencing factors are analyzed, the influencing factor index is established, and the data change time and the change of the abnormal impact data are calculated. The abnormal cause analysis function of the meter is used to comprehensively analyze the abnormal influencing factors and output the abnormal cause analysis results.

[0166] Finally, based on the abnormal analysis results, the present invention further identifies the abnormal influencing factors, and performs importance and correlation analysis to obtain the weight vector and correlation matrix of the abnormal impact. By slicing and calculating the changes of the meter data and the abnormal impact data, expanding the data set, and using error calculation for compensation, the accuracy of the online calibration of the smart meter data is finally improved.

[0167] Embodiment 2:

[0168] In Example 1, the method of the present invention is used to realize online abnormality monitoring and calibration of smart meter A; in order to illustrate that the present invention has the same practicality for other smart meters, the present embodiment of the application will explain the process of smart meter B, as follows:

[0169] Obtain historical data of smart meter B;

[0170] Furthermore, the historical data of the smart meter B is preprocessed and marked to form a training set for the smart meter B;

[0171] Furthermore, the electric meter anomaly monitoring model is trained using the training set; wherein the electric meter anomaly monitoring model includes: an input layer, a cluster feature learning layer, an anomaly feature weighting layer, an electric meter anomaly recognition layer and an output layer;

[0172] Furthermore, the trained meter anomaly monitoring model is used to analyze the current data of smart meter B collected by the communication module, thereby detecting the abnormality of smart meter B. According to the monitoring results of the meter anomaly monitoring model, it can be seen that smart meter B has voltage overload and power imbalance problems;

[0173] Furthermore, the abnormal cause analysis is performed, and the specific analysis process includes:

[0174] According to the current abnormal monitoring result, obtaining current meter data corresponding to the abnormality, and constructing a current meter abnormal data vector;

[0175] Obtain multiple electric meter data of smart electric meter B at the last moment;

[0176] Furthermore, multiple parameters at the previous moment are input into the meter abnormality monitoring model to obtain the meter abnormality monitoring result; when the monitoring result shows normal, the multiple parameters at the previous moment are used as normal meter data, and the meter normal data vector is constructed; otherwise, the previous meter data is continuously collected, and the meter abnormality monitoring model is used for continuous monitoring until the monitoring result is normal;

[0177] Further, according to the current abnormal monitoring result, a candidate abnormal influencing factor field is obtained;

[0178] Further, based on the candidate abnormal influencing factor field, a candidate abnormal influencing factor index is established;

[0179] Further, calculating the data change time between the normal data vector of the electric meter and the abnormal data vector of the current electric meter;

[0180] Further, calculating abnormal impact change data of the candidate abnormal impact factor field within the data change time;

[0181] Furthermore, the current meter abnormal data vector, the meter normal data vector, the candidate abnormal influencing factor index, the data change time, the abnormal influencing change data and the abnormal influencing change data are input into the meter abnormal cause analysis function to obtain the abnormal analysis result.

[0182] The abnormality reason analysis function of the electric meter is expressed as: ;in, The abnormal reason analysis vector of the electric meter at the current moment; is the modulus of the abnormal data vector of the electric meter at the current moment; is the data change time; is the candidate abnormal influencing factor at the current moment The modulus of the change in is the candidate abnormal influencing factor at the current moment; is the total number of candidate abnormal factors; is the influence function of the size of the meter abnormal data vector on the meter abnormal cause analysis; is the influence function of the change of the candidate abnormal influencing factors on the abnormality of the electric meter; The action function for the candidate abnormal influencing factor index.

[0183] Further, according to the abnormal analysis result, the abnormal influencing factor is obtained;

[0184] Further, abnormal importance analysis and abnormal correlation analysis are performed on the abnormal influencing factors to obtain an abnormal influence weight vector and an abnormal influence correlation matrix;

[0185] The calculation process of the abnormal impact weight vector includes:

[0186] Characterizing the abnormal influencing factors to obtain abnormal influencing factor feature vectors;

[0187] Further, setting an abnormality target of the electric meter according to the current abnormality monitoring result;

[0188] Furthermore, the influence probability of each feature in the abnormal influencing factor feature vector on the abnormal target of the electric meter is calculated; wherein the calculation formula of the influence probability is: ;in, Abnormal influencing factors The eigenvector of It is the abnormal target of the electric meter; Abnormal influencing factors Probability of impact on abnormal targets of electric meters; Abnormal influencing factors The eigenvalues ​​in the eigenvector of Abnormal target value of electric meter The joint probability of Abnormal influencing factors The eigenvalues ​​in the eigenvector of The marginal probability of is the abnormal target value of the electric meter The marginal probability of

[0189] Furthermore, the impact probabilities are sorted in descending order to obtain the abnormal impact weight vector.

[0190] Furthermore, the electric meter data and abnormal impact data within a continuous period of time are obtained;

[0191] Further, the meter data and the abnormal impact data are sliced, and a first data set is established according to the meter state; the change coefficients of the meter data and the abnormal impact data in the first data set are calculated respectively to obtain a meter data change vector and an abnormal impact data change vector;

[0192] Further, the first data set is expanded using the meter data change vector and the abnormal impact data change vector to obtain a second data set;

[0193] Further, performing error calculation according to the abnormal impact weight vector, the abnormal impact association matrix and the second data set to obtain an error compensation result;

[0194] Furthermore, the smart meter B is calibrated online according to the error compensation result.

[0195] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An online calibration control method for a smart meter communication module, characterized in that: include: Get historical data of smart meters; Preprocessing the historical data to obtain standard historical feature data; Marking and integrating the standard historical feature data to obtain a training set; Using the training set to train an electric meter abnormality monitoring model; Use the communication module to collect current meter data; Transmitting the current electric meter data to the trained electric meter abnormality monitoring model to obtain the current abnormality monitoring result; Analyze the current abnormal monitoring result to obtain an abnormal analysis result; Obtaining abnormal influencing factors according to the abnormal analysis results; Performing abnormal importance analysis and abnormal correlation analysis on the abnormal influencing factors to obtain an abnormal influence weight vector and an abnormal influence correlation matrix; Obtain meter data and abnormal impact data over a continuous period of time; Slicing the meter data and the abnormal impact data, and establishing a first data set according to the meter state; respectively calculating the change coefficients of the meter data and the abnormal impact data in the first data set to obtain a meter data change vector and an abnormal impact data change vector; Expanding the first data set by using the meter data change vector and the abnormal impact data change vector to obtain a second data set; Perform error calculation according to the abnormal impact weight vector, the abnormal impact association matrix and the second data set to obtain an error compensation result; The smart electric meter is calibrated online according to the error compensation result.

2. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The historical data includes: meter usage time, power data, current data, voltage data, power data, frequency data, temperature data and load data.

3. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The preprocessing process of the historical data includes: performing data cleaning on the historical data to obtain first historical data; performing noise processing on the first historical data to obtain second historical data; performing data conversion on the second historical data to obtain third historical data; performing normalization processing on the third historical data to obtain fourth historical data; and performing feature construction on the fourth historical data to obtain the standard historical feature data.

4. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The training process of the electric meter abnormality monitoring model includes: Using an input layer to receive data samples in the training set; wherein the input layer includes: a data receiving sublayer and a data clustering sublayer; the data receiving sublayer is used to receive data; the data clustering sublayer is used to cluster the data samples to obtain K cluster sets; Inputting the K cluster sets into the cluster feature learning layer to obtain K cluster feature vectors; The cluster feature learning layer includes: obtaining data samples in the same cluster set; inputting the data samples into the trained neural network to obtain abnormal classification results; determining abnormal factors according to the abnormal classification results; obtaining key feature vectors of each data sample in the cluster set according to the abnormal factors, and constructing a key feature matrix; calculating the error between the features in the key feature matrix and the non-abnormal features to obtain a key feature error matrix; calculating the cluster feature vector according to the key feature error matrix; wherein the calculation formula of the cluster feature vector is: ;in, The cluster set The clustering feature vector of The cluster set The data sample size; The cluster set Medium data sample The key feature vector of is the weight coefficient, which is used to balance the influence of key eigenvectors and errors; The cluster set Medium data sample The key characteristic error vector of Inputting the K cluster sets and the K cluster feature vectors into an abnormal feature weighting layer to obtain K weighted cluster sets; Inputting the K weighted cluster sets into the meter anomaly recognition layer to obtain the anomaly recognition probability of each data sample in the K weighted cluster sets; The abnormality recognition probability is output by the output layer to obtain the abnormality monitoring result; the error between the abnormality monitoring result and the mark is calculated using the loss function; the parameters of the meter abnormality monitoring model are updated using the back propagation method, and when the loss value reaches convergence, the meter abnormality monitoring model is trained.

5. The online calibration control method of a smart meter communication module according to claim 4, characterized in that: The weighting function of the abnormal feature weighting layer is: ;in, The cluster set The weighted data sample ; For data samples ; For clustering set Medium data sample The weighting coefficient of .

6. The online calibration control method of a smart meter communication module according to claim 5, characterized in that: The formula of the weighting coefficient is: ;in, For data samples The degree of abnormality; For data samples and cluster feature vector distance; is a hyperparameter.

7. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The process of analyzing the current abnormal monitoring result includes: According to the current abnormal monitoring result, obtaining current meter data corresponding to the abnormality, and constructing a current meter abnormal data vector; Acquire multiple electric meter data of the smart electric meter at the last moment; Input multiple parameters of the last moment into the meter abnormality monitoring model to obtain the meter abnormality monitoring result; when the monitoring result shows normal, use the multiple parameters of the last moment as normal meter data and construct the meter normal data vector; otherwise, continue to collect the previous meter data and use the meter abnormality monitoring model for continuous monitoring until the monitoring result is normal; Obtaining a candidate abnormality influencing factor field according to the current abnormal data vector of the electric meter; Establishing a candidate abnormal influencing factor index according to the candidate abnormal influencing factor field; Calculating the data change time between the normal data vector of the electric meter and the abnormal data vector of the current electric meter; Calculate the abnormal impact change data of the candidate abnormal impact factor field within the data change time; The current meter abnormal data vector, the meter normal data vector, the candidate abnormal influencing factor index, the data change time, the abnormal influencing change data and the abnormal influencing change data are input into the meter abnormal cause analysis function to obtain the abnormal analysis result.

8. The online calibration control method of a smart meter communication module according to claim 7, characterized in that: The abnormality cause analysis function of the electric meter is expressed as: ;in, The abnormal reason analysis vector of the electric meter at the current moment; is the modulus of the abnormal data vector of the electric meter at the current moment; is the data change time; is the candidate abnormal influencing factor at the current moment The modulus of the change in is the candidate abnormal influencing factor at the current moment; is the total number of candidate abnormal factors; is the influence function of the size of the meter abnormal data vector on the meter abnormal cause analysis; is the influence function of the change of the candidate abnormal influencing factors on the abnormality of the electric meter; The action function for the candidate abnormal influencing factor index.

9. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The calculation process of the abnormal impact weight vector includes: Characterizing the abnormal influencing factors to obtain abnormal influencing factor feature vectors; Setting an abnormality target for the electric meter according to the current abnormality monitoring result; The influence probability of each feature in the feature vector of the abnormal influencing factor on the abnormal target of the electric meter is calculated; wherein the calculation formula of the influence probability is: ;in, Abnormal influencing factors The eigenvector of It is the abnormal target of the electric meter; Abnormal influencing factors Probability of impact on abnormal targets of electric meters; Abnormal influencing factors The eigenvalues ​​in the eigenvector of Abnormal target value of electric meter The joint probability of Abnormal influencing factors The eigenvalues ​​in the eigenvector of The marginal probability of is the abnormal target value of the electric meter The marginal probability of The impact probabilities are sorted in descending order to obtain the abnormal impact weight vector.

10. The online calibration control method of a smart meter communication module according to claim 1, characterized in that: The calculation of the error compensation result adopts an iterative error compensation calculation method, including: Obtain the abnormal impact weight vector and the abnormal impact association matrix, which are respectively denoted as and ; defining an error function according to the meter data change vector and the abnormal impact data change vector in the second data set; Wherein, the error function is: ;in, For the The error compensation result at the current moment after iteration; is the meter change vector at the current moment; is the data change vector affected by the abnormality at the current moment; For the Error compensation coefficient after iteration; For the Corrected error constant after iteration; Calculate the gradient of the error function, denoted as ; The error compensation coefficient is updated according to the gradient, and the formula is: ;in, For the The error compensation coefficient at the iteration; is the learning rate; Update the correction error constant, the formula is: ;in, For the The correction error constant at the iteration; The error compensation result is recalculated according to the updated error compensation coefficient and the corrected error constant; when the error compensation result has the best adjustment effect on the smart meter, the final error compensation result is obtained.

Images