A remote electric energy meter fault monitoring method and device

Abstract

The application provides a remote electric energy meter fault monitoring method and device, and the steps thereof include: constructing a two-level fault monitoring network composed of an edge device and a cloud server; the edge device collects operation data of the electric energy meter based on a sliding window method, and performs initial screening of features through a principal component analysis method; when the result of the initial screening of features is an abnormal condition, the edge device sends early warning information to the cloud server and transmits the operation data; the cloud server analyzes the operation data through a machine learning model and obtains a predicted fault type.The application constructs a two-level fault monitoring network, the edge device performs initial screening, and when an abnormality occurs, the edge device sends early warning and data to the cloud server, and the cloud server analyzes the fault type through a machine learning model, which takes into account the real-time performance of the edge side and the deep analysis capability of the cloud, reduces the edge calculation amount and data transmission amount, improves the fault monitoring efficiency and accuracy, reduces the operation and maintenance cost, and adapts to the monitoring needs of the occasional faults of the electric energy meter in a complex electromagnetic environment.

Description

Technical Field

[0001] This application belongs to the field of electrical digital data processing technology, specifically relating to a method and device for remote monitoring of faults in electricity meters. Background Technology

[0002] With the rapid development of the new energy vehicle industry, fast charging technology for electric vehicles has become a key link supporting the industry's large-scale application. As a core infrastructure for energy replenishment, charging piles are seeing continuous expansion in deployment scale, increasing power levels, and significantly higher switching frequencies. This directly leads to an increasingly complex electromagnetic environment inside and around charging piles. Strong electromagnetic interference and instantaneous high-power surges pose severe challenges to the stable operation of key components within charging piles. Among these, the reliability of the electricity meter, as a core component for metering electricity transactions and ensuring fair billing, is particularly important.

[0003] In practical applications, electricity meters often face two typical types of faults: First, measurement accuracy deviation, manifested as a discrepancy between the measured electricity consumption and the actual consumption. This can not only lead to disputes between users and operators but also potentially result in economic losses. Second, communication link anomalies, i.e., interruptions or delays in data transmission between the electricity meter and the charging pile host, causing the failure of functions such as equipment status monitoring and remote maintenance. A more prominent problem is that these faults are significantly sporadic and insidious. Under the dynamic influence of strong electromagnetic environments, faults often occur suddenly and are short-lived. Traditional manual inspection methods struggle to capture the instantaneous data characteristics of faults in real time, while long-term on-site monitoring results in a significant waste of human, material, and financial resources.

[0004] In existing technologies, although some charging piles are equipped with basic fault alarm functions, they mostly rely on fixed thresholds for single parameters, such as voltage exceeding the upper limit or current exceeding the lower limit. This makes it difficult to adapt to the diversity and correlation of fault characteristics in complex electromagnetic environments, resulting in high false alarm and false alarm rates.

[0005] Meanwhile, with the development of real-time online monitoring technology for power devices, remote monitoring solutions have gradually become an important direction for equipment operation and maintenance. However, because the manufacturing and use of electricity meters must comply with relevant standards and cannot be arbitrarily modified or upgraded, there are still many technical bottlenecks in the current field of electricity meter fault monitoring. On the one hand, the conventional monitoring method is to deploy local monitoring equipment. However, local equipment is limited by computing resources and storage capacity, making it difficult to directly run complex fault analysis algorithms. Uploading all raw data to a server for processing faces problems such as high bandwidth pressure and poor real-time performance, failing to meet the needs of rapid fault response. On the other hand, electricity meter faults are diverse. Different faults, such as metering chip malfunctions, communication module interference, or power fluctuations, have significantly different corresponding operating data characteristics. Traditional single models or threshold systems are difficult to effectively distinguish the characteristic patterns of multiple types of faults, leading to a disconnect between fault warnings and type judgment, and failing to provide accurate guidance for subsequent operation and maintenance.

[0006] Furthermore, existing fault monitoring methods are inefficient in utilizing historical fault data. Most solutions only establish monitoring models for a single fault type. When faced with multiple fault types, multiple independent models need to be deployed repeatedly. This not only increases the deployment complexity of the equipment but also consumes excessive resources due to redundant calculations between models. At the same time, the extraction of fault features often relies on manual experience and lacks a data-driven scientific screening mechanism, making it difficult to capture subtle changes in characteristics before a fault occurs in the electromagnetic environment, resulting in early warnings lagging behind the actual time of fault occurrence.

[0007] These issues collectively hinder the intelligent level of charging pile electricity meter fault monitoring: it cannot achieve early warning of faults, it is difficult to quickly locate fault types, and it cannot support efficient remote operation and maintenance. Therefore, how to build a remote monitoring solution that balances real-time performance and accuracy, has low deployment costs, can effectively warn of faults, and can distinguish between multiple fault types has become a key technical problem that needs to be solved. Summary of the Invention

[0008] This application addresses the problems existing in the prior art by providing a method and apparatus for remote monitoring of electricity meter faults, thereby resolving the aforementioned issues.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] On the one hand, this application provides a method and apparatus for remote monitoring of energy meter faults, including the following steps:

[0011] Construct a two-tier fault monitoring network consisting of edge devices and cloud servers;

[0012] The edge device is connected to the electricity meter, collects the electricity meter's operating data based on the sliding window method, and performs initial feature screening using principal component analysis.

[0013] When the initial feature screening result is an anomaly, the edge device sends a warning message to the cloud server and transmits the operational data;

[0014] The cloud server analyzes the operational data using a machine learning model and obtains a predicted fault type.

[0015] Furthermore, the initial feature screening also includes the following steps:

[0016] Collect historical fault information of electricity meters and obtain the principal component matrix using principal component analysis.

[0017] Calculate fault characteristic values ​​based on the principal component matrix and the historical fault information of the electricity meter; divide the normal operating range according to the fault characteristic values;

[0018] Real-time data collection of electricity meter operation data, and calculation of real-time eigenvalues ​​based on the principal component matrix;

[0019] When the real-time feature value exceeds the normal operating range, the result of the initial feature screening is an abnormal situation.

[0020] Furthermore, the historical fault information of the electricity meter includes multiple types of faults, and each type of fault corresponds to the operating data of the electricity meter in the most recent window before the fault.

[0021] The initial feature screening also includes the following steps:

[0022] Principal component analysis was performed on the fault information according to its type, and the eigenvalue sequence was obtained.

[0023] All eigenvalue sequences are statistically analyzed, and the principal component matrix is ​​obtained by filtering based on contribution rate and similarity principles.

[0024] Furthermore, the calculation steps for the fault characteristic values ​​include:

[0025] The fault feature vector is calculated using the principal component matrix and the historical fault information of the electricity meter.

[0026] Calculate the magnitude of the fault feature vector, where the magnitude is the fault feature value.

[0027] Furthermore, the steps for defining the normal operating range include:

[0028] Based on the principal component matrix and historical data of the electricity meter, calculate the operating characteristic value; calculate the upper and lower limits of the operating characteristic value to obtain the theoretical operating range;

[0029] The upper and lower limits of the fault characteristic values ​​are statistically analyzed to obtain the fault operating range.

[0030] Based on the theoretical operating range and the fault operating range, the normal operating range is obtained through the difference set calculation method.

[0031] Furthermore, the operating data of the electricity meter includes voltage, current, power, energy, and temperature.

[0032] Furthermore, the machine learning model is a long short-term memory network;

[0033] The pre-training of the machine learning model includes the following steps:

[0034] Collect historical fault information of electricity meters and establish a time-series-based training dataset; in the training dataset, the input is the operating data of the electricity meter in the most recent window before the fault, and the target result is the fault type;

[0035] The machine learning model is obtained by training the training dataset.

[0036] Furthermore, it also includes the following steps:

[0037] The cloud server retrieves data from the database and filters online fault handling strategies based on the predicted fault type.

[0038] The cloud server distributes the online fault handling strategy to the edge device, and sends fault handling instructions to the electricity meter through the edge device.

[0039] On the other hand, this application also provides a remote energy meter fault monitoring device for performing the aforementioned remote energy meter fault monitoring method, which includes an edge device and a cloud server; the edge device is connected to the cloud server via a public network;

[0040] The edge device is also connected to the electricity meter via a communication module to read the electricity meter's operating data; the edge device is equipped with a first computing unit and a feature screening module.

[0041] The feature screening module is equipped with a program for feature screening;

[0042] The first computing unit executes the program of the feature screening module to perform feature screening on the operating data of the electricity meter;

[0043] The cloud server is equipped with a second computing unit and a machine learning model analysis module;

[0044] The machine learning model analysis module is equipped with machine learning models.

[0045] The second computing unit runs the machine learning model of the machine learning model analysis module to analyze the operating data of the electricity meter uploaded by the edge device and obtain the predicted fault type.

[0046] Furthermore, the cloud server is also configured with a preset module; the preset module is configured with the edge device configuration and update program;

[0047] The second calculation unit executes the program of the preset module, collects historical fault information of the electricity meter, and obtains the principal component matrix through principal component analysis; calculates fault characteristic values ​​based on the principal component matrix and the historical fault information of the electricity meter; and divides the normal operating range according to the fault characteristic values.

[0048] The preset module sends configuration information and / or update information to the edge device via the public network; the configuration information and / or update information includes at least the principal component matrix and the normal operating range.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] This invention constructs a two-level fault monitoring network consisting of edge devices and a cloud server. Edge devices collect electricity meter operating data using a sliding window and perform initial screening using principal component analysis. When an anomaly occurs, they send warnings and data to the cloud server, which then analyzes the fault type using a machine learning model. This architecture balances real-time performance at the edge with deep analysis capabilities in the cloud, reducing edge computing and data transmission loads, improving fault monitoring efficiency and accuracy, lowering maintenance costs, and adapting to the monitoring needs of occasional electricity meter faults in complex electromagnetic environments. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a method in a specific embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the device connection in a specific embodiment of the present invention.

[0054] In the diagram: 1. Electricity meter, 2. Edge device, 3. Cloud server. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0057] It should also be noted that, unless otherwise specified, the methods used in this invention are conventional methods; and the raw materials and apparatus used are, unless otherwise specified, conventional commercially available products.

[0058] On the one hand, this embodiment proposes a method for remote monitoring of energy meter faults, such as... Figure 1 As shown, the specific steps include:

[0059] A two-level fault monitoring network consisting of edge devices and cloud servers is constructed. The edge devices are deployed locally at the charging piles and have lightweight computing and data storage capabilities, while the cloud servers are deployed in remote data centers and have large-scale data storage and complex model calculation capabilities. The two communicate bidirectionally through an encrypted transmission protocol.

[0060] Edge devices connect to electricity meters and collect their operational data using a sliding window method. Principal component analysis (PCA) is then used for initial feature screening. The sliding window duration is T, meaning the edge device collects and stores data within a time period T prior to the current collection time. When new data is collected, data exceeding the T-time period is discarded. Furthermore, the collected electricity meter operational data is tagged with corresponding timestamps, thus forming a time-series dataset. Even further, the electricity meter operational data includes voltage... U Current I Active power P Cumulative electrical energy E and the temperature of the surface T emp .

[0061] Therefore, the initial feature screening also includes the following steps:

[0062] Collect historical fault information of electricity meters, which includes multiple types of faults, and each type of fault includes the operating data of the electricity meter in the most recent window before the fault; at the same time, collect normal operating data under the same working conditions to form a complete comparison dataset.

[0063] Principal component matrix is ​​obtained through principal component analysis; specifically, principal component analysis is performed on fault information according to its type, and eigenvalue sequences are obtained; for example: for the first... m Class of faults, let its corresponding data matrix be... , This represents the number of samples for this type of fault. p =5 represents the parameter dimension. First, regarding... Standardization processing is performed, that is:

[0064] ;

[0065] In the formula, For the mean phasor of each parameter, This is the standard deviation vector of each parameter.

[0066] Next, calculate the covariance matrix of the standardized matrix. The formula is:

[0067] ;

[0068] Solving by eigenvalue decomposition eigenvalues and corresponding feature vectors This forms a sequence of characteristic values ​​for this type of fault. .

[0069] All eigenvalue sequences are statistically analyzed, and principal component matrices are obtained by filtering based on contribution rate and similarity principles. Specifically, the cumulative contribution rate of each eigenvalue sequence is calculated. The formula is:

[0070] ;

[0071] In the formula, select Corresponding front k Each feature value is calculated; simultaneously, the similarity (e.g., cosine similarity) of feature value sequences for different fault types is calculated. Sequences with similarity higher than a threshold (e.g., 0.8) are merged, and the sequence with the highest overall contribution and best discriminative power is finally selected. k Eigenvectors form the principal component matrix .

[0072] Fault feature values ​​are calculated based on the principal component matrix and historical fault information of the electricity meter; during the calculation of fault feature values, fault feature vectors are calculated using the principal component matrix and historical fault information of the electricity meter.

[0073] Specifically, standardized data for each type of fault. Calculate its principal component scores. ,Right now:

[0074] ;

[0075] In the formula, For the eigenvector of the th i Each component.

[0076] Therefore, the magnitude of the fault feature vector is denoted as the fault feature value.

[0077] The normal operating range is defined based on fault characteristic values. The definition process includes:

[0078] The operational characteristic values ​​are calculated based on the principal component matrix and historical data from the electricity meter. Specifically, for historical normal data... Principal component scores were calculated after standardization. Its mold length This involves setting the running feature values. Next, the upper limit value of the running feature values ​​is calculated. and lower limit value In order to obtain the theoretical operating range .

[0079] Therefore, the upper limit of the statistical fault characteristic value is... and lower limit value To obtain the fault operating range .

[0080] Based on the theoretical operating range and the fault operating range, the normal operating range is obtained through the difference set calculation method. Furthermore, if the fault operating range is completely contained within the theoretical operating range, then the normal operating range is... If there is partial overlap, the interval after removing the overlapping part from the theoretical operating range is taken to ensure that the boundary between normal and fault characteristic values ​​is clearly distinguishable.

[0081] Real-time acquisition of electricity meter operating data, and calculation of real-time eigenvalues ​​based on principal component matrices; real-time window data. First, based on historical averages μ and standard deviation σ Standardized to Then calculate the principal component scores. Its mold length This refers to the real-time feature value.

[0082] When the real-time feature value exceeds the normal operating range, the result of the initial feature screening is an abnormal situation.

[0083] When the initial feature screening result is abnormal, the edge device sends an early warning message to the cloud server and transmits operational data.

[0084] The cloud server analyzes operational data using a machine learning model to predict fault types. The machine learning model is a Long Short-Term Memory (LSTM) network. The network structure used in this embodiment includes an input layer (dimension: [dimension not specified]). p × T s , T s The output consists of: the number of sampling points within the window, two LSTM hidden layers (with 64 and 32 hidden units respectively, using the ReLU activation function), a dropout layer (to prevent overfitting, with a dropout rate of 0.2), and a fully connected output layer (the output dimension is the number of fault types, using the softmax activation function).

[0085] The pre-training of a machine learning model includes the following steps:

[0086] Historical fault information of electricity meters was collected, and a time-series-based training dataset was established; for each fault sample, the period before the fault occurred was extracted. T The duration of the window data is used as the input sequence, and the corresponding fault type is used as the label (represented by one-hot encoding). The data is divided into training set, validation set and test set in a ratio of 7:2:1.

[0087] In the training dataset, the input is the operating data of the energy meter in the most recent window before the fault, and the target result is the fault type.

[0088] A machine learning model is obtained by training a training dataset; the training process uses the cross-entropy loss function, calculated as follows:

[0089] ;

[0090] In the formula, c Number of fault types For real labels, To predict probabilities.

[0091] The optimizer uses Adam (with an initial learning rate of 0.001 that decays with each training epoch). Training stops when the accuracy on the validation set does not improve for 10 consecutive epochs, and the optimal model parameters are saved.

[0092] Furthermore, after predicting the fault type, the cloud server also retrieves data from the database and filters online fault handling strategies based on the predicted fault type.

[0093] The cloud server distributes online fault handling strategies to edge devices, which then send fault handling instructions to the electricity meter. These instructions are in a structured format (e.g., JSON) and include the operation type, parameter values, and execution time limit. After execution, the edge devices return the results to the cloud, forming a closed-loop control system.

[0094] The above fault monitoring method differs from common edge-cloud collaborative processing technologies. It reduces the resource requirements for data storage and processing of local monitoring devices from multiple perspectives. It only requires configuring the parameters needed for principal component analysis in the local monitoring device. With the sliding window design, it improves processing speed and, combined with similarity calculation in the preprocessing stage, coordinates different types of faults to obtain the feature quantity that best reflects the fault, thereby filtering out abnormal situations without reducing the accuracy of judgment. Meanwhile, artificial intelligence processing based on time-series datasets in the cloud can enhance the accuracy of subsequent judgments and analyze potential fault problems based on time-series data. On the other hand, it can solve the problem of large warning volume of edge devices and reduce false alarm rate through high-precision analysis.

[0095] On the other hand, such as Figure 2 As shown, this application also provides a remote electricity meter fault monitoring device for performing the above method, including an edge device 2 and a cloud server 3; the edge device is connected to the cloud server via a public network, such as 4G / 5G, Ethernet, etc.

[0096] Since common energy meter 1 has an RS485 interface, the edge device also connects to energy meter 1 via a communication module, such as an RS485 communication module, to read the energy meter's operating data. Edge device 2, as a locally deployed device, aims to reduce costs; its hardware configuration only needs to meet practical requirements. Therefore, a parallel computing architecture can be selected, allowing a single edge device to connect to multiple energy meters (see attached diagram). Figure 2 (Taking a single electricity meter as an example), it is equipped with a first computing unit, such as a microprocessor, and a feature screening module. The feature screening module is stored in Flash memory as firmware. The feature screening module is configured with a program for feature screening, which includes data normalization functions, matrix multiplication functions, modulus calculation functions, and range judgment logic.

[0097] The first computing unit executes the program of the feature screening module to perform feature screening on the operating data of the electricity meter. During the operation, the memory occupied is ≤128KB, which meets the lightweight requirements of the edge side.

[0098] Cloud server 3 is configured with a second computing unit, such as a GPU cluster and a machine learning model analysis module; the machine learning model analysis module is specifically deployed in a container. This module contains the previously trained LSTM model and supports model version management and online updates.

[0099] The second computing unit runs the machine learning model analysis module to analyze the operating data of the electricity meter uploaded by the edge device and obtain the predicted fault type.

[0100] Preferably, cloud server 3 is also configured with a preset module, which is deployed on the distributed database node. The preset module is configured with edge device configuration and update program, supporting batch device management and parameter synchronization.

[0101] The second calculation unit executes the program of the preset module, periodically collects newly added historical fault information of electricity meters, and re-executes the principal component analysis process to update the principal component matrix. P The system simultaneously updates fault characteristic values ​​and normal operating ranges. The preset module distributes configuration and update information to edge devices via the public network, specifically using incremental updates that transmit only the changed portions, thus reducing the amount of data updates. The configuration and update information includes at least the principal component matrix. P Normal operating range threshold and standardized parameters μ、σ The edge device automatically overwrites the old parameters after receiving them, ensuring the timeliness of the monitoring standards.

[0102] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for remotely monitoring faults in an energy meter, characterized in that, Includes the following steps: Construct a two-tier fault monitoring network consisting of edge devices and cloud servers; The edge device is connected to the electricity meter, collects the meter's operating data based on the sliding window method, and performs initial feature screening using principal component analysis. The initial feature screening also includes the following steps: Collect historical fault information of electricity meters and obtain the principal component matrix using principal component analysis. Calculate fault characteristic values ​​based on the principal component matrix and the historical fault information of the electricity meter; The normal operating range is determined based on the fault characteristic values; Real-time data collection of electricity meter operation data, and calculation of real-time eigenvalues ​​based on the principal component matrix; When the real-time feature value exceeds the normal operating range, the result of the initial feature screening is an abnormal situation; The historical fault information of the electricity meter includes multiple types of faults, and each type of fault corresponds to the operating data of the electricity meter in the most recent window before the fault. The initial feature screening also includes the following steps: Principal component analysis was performed on the fault information according to its type, and the eigenvalue sequence was obtained. The principal component matrix is ​​obtained by statistically analyzing all the feature value sequences and filtering them based on the contribution rate and similarity principles. The calculation steps for the fault feature value include: calculating a fault feature vector using the principal component matrix and the historical fault information of the energy meter; and calculating the magnitude of the fault feature vector, where the magnitude is the fault feature value. The steps for defining the normal operating range include: Based on the principal component matrix and historical data of the electricity meter, calculate the operating characteristic value; calculate the upper and lower limits of the operating characteristic value to obtain the theoretical operating range; The upper and lower limits of the fault characteristic values ​​are statistically analyzed to obtain the fault operating range. Based on the theoretical operating range and the fault operating range, the normal operating range is obtained through the difference set calculation method. When the initial feature screening result is an anomaly, the edge device sends a warning message to the cloud server and transmits the operational data; The cloud server analyzes the operational data using a machine learning model and obtains a predicted fault type.

2. The remote energy meter fault monitoring method according to claim 1, characterized in that, The operating data of the electricity meter includes voltage, current, power, energy, and temperature.

3. The remote energy meter fault monitoring method according to claim 1, characterized in that, The machine learning model is a long short-term memory network; The pre-training of the machine learning model includes the following steps: Collect historical fault information of electricity meters and establish a time-series-based training dataset; in the training dataset, the input is the operating data of the electricity meter in the most recent window before the fault, and the target result is the fault type; The machine learning model is obtained by training the training dataset.

4. The remote energy meter fault monitoring method according to claim 1, characterized in that, It also includes the following steps: The cloud server retrieves data from the database and filters online fault handling strategies based on the predicted fault type. The cloud server distributes the online fault handling strategy to the edge device, and sends fault handling instructions to the electricity meter through the edge device.

5. A remote energy meter fault monitoring device, characterized in that, Used to perform the remote energy meter fault monitoring method according to any one of claims 1-4; The device includes an edge device and a cloud server; the edge device is connected to the cloud server via a public network. The edge device is also connected to the electricity meter via a communication module to read the electricity meter's operating data; the edge device is equipped with a first computing unit and a feature screening module. The feature screening module is equipped with a program for feature screening; The first computing unit executes the program of the feature screening module to perform feature screening on the operating data of the electricity meter; The cloud server is equipped with a second computing unit and a machine learning model analysis module; The machine learning model analysis module is equipped with machine learning models. The second computing unit runs the machine learning model of the machine learning model analysis module to analyze the operating data of the electricity meter uploaded by the edge device and obtain the predicted fault type.

6. The remote energy meter fault monitoring device according to claim 5, characterized in that, The cloud server is also configured with a preset module; the preset module is configured with the edge device configuration and update program; The second calculation unit executes the program of the preset module, collects historical fault information of the electricity meter, and obtains the principal component matrix through principal component analysis. Calculate fault characteristic values ​​based on the principal component matrix and the historical fault information of the electricity meter; The normal operating range is determined based on the fault characteristic values; The preset module sends configuration information and / or update information to the edge device via the public network; The configuration information and / or update information include at least the principal component matrix and the normal operating range.

Images