Distributed electric meter remote monitoring system and method based on internet of things

Abstract

The application discloses a kind of based on Internet of Things distributed electric meter remote monitoring system and method, it is related to electric meter monitoring technical field, it is first in electric meter terminal acquisition real-time electric signal and carries out analog-digital conversion, then utilize lightweight machine learning model to short-time data flow is carried out real-time analysis to generate high priority exception event, and periodically statistical analysis is simultaneously used to aggregate long-time data, further, by event-driven and periodic driven dual-channel reporting mechanism, exception event is reported to cloud platform with highest priority immediately, simultaneously, aggregate record is packed and time-sharing sent.This way, combined with the acuteness of edge real-time analysis and the economy of data hierarchical reporting, effectively solve the network congestion and delay problem caused by mass data concurrency, significantly improve the real-time and accuracy of power consumption anomaly monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric meter monitoring, and more particularly, to a distributed electric meter remote monitoring system and method based on Internet of Things. BACKGROUND

[0002] With the rapid development of Internet of Things and smart grid technology, fine and intelligent remote monitoring of massive distributed electric meters has become a key to ensuring the stability of the power grid and improving energy efficiency. As the sensing terminal of the smart grid, the real-time and accuracy of the operation data of the electric meter is directly related to the dispatching optimization, fault warning and load management of the power system. Efficient and low-cost monitoring of hundreds of millions of electric meter terminals is an important cornerstone for building future energy internet.

[0003] In the prior art, traditional electric meter monitoring mainly relies on periodic manual meter reading, which not only consumes a lot of manpower and resources, but also has slow response and poor data real-time performance, making it difficult to meet the dynamic management needs of modern power grids. The existing remote monitoring system based on Internet of Things mostly regards the electric meter as a simple data acquisition terminal, and uploads the massive raw data collected to the cloud platform for centralized analysis without distinction. This "cloud centralized processing" mode has significant drawbacks: first, the continuous uploading of massive data puts a huge pressure on the communication network, resulting in high bandwidth cost and easy network congestion; second, the delay introduced by data transmission and centralized analysis makes it impossible to discover and deal with abnormal events such as electricity theft and line faults in a timely manner; finally, the cloud platform has heavy computing and storage load, and the overall operation cost and scalability of the system are severely restricted.

[0004] Therefore, there is an urgent need for an optimized distributed electric meter remote monitoring system and method based on Internet of Things. SUMMARY

[0005] In order to solve the above technical problems, the present application is proposed.

[0006] According to an aspect of the present application, a distributed electric meter remote monitoring method based on Internet of Things is provided, which comprises: S1: a micro control unit (MCU) built in an electric meter samples and digitizes an analog electric signal, and stores it in a ring buffer of the electric meter to obtain a local data stream.

[0007] S2: the MCU continuously detects anomalies in the local data stream and generates an anomaly event message upon successful anomaly detection.

[0008] S3: the MCU aggregates data in the local data stream at a first preset time period to obtain a data aggregation record.

[0009] S4: in response to detecting the abnormal event message, sending it to the cloud platform through the wireless network at the highest priority.

[0010] S5: storing the data aggregation records in the non-volatile memory, and sending the periodic aggregation data packets obtained by packing a plurality of data aggregation records according to a second preset time period to the cloud platform.

[0011] Step S2 comprises: cleaning the local data stream in the short-time sliding window according to the data sample dimension to obtain an instantaneous current short-time time series and an instantaneous voltage short-time time series; extracting a short-time current mode feature and a short-time voltage mode feature from the instantaneous current short-time time series and the instantaneous voltage short-time time series; fusing the short-time current mode feature and the short-time voltage mode feature to obtain a voltage-current short-time time series joint feature; and inputting the voltage-current short-time time series joint feature into a lightweight machine learning model for anomaly detection.

[0012] In the above-mentioned distributed electric meter remote monitoring method based on the Internet of Things, the analog electric signal is output by a metering chip built in the electric meter, the sampling frequency of the analog electric signal is 200 Hz, and the local data stream comprises an instantaneous current value, an instantaneous voltage value, and an instantaneous active power.

[0013] In the above-mentioned distributed electric meter remote monitoring method based on the Internet of Things, inputting the voltage-current short-time time series joint feature into a lightweight machine learning model for anomaly detection comprises: inputting the voltage-current short-time time series joint feature into the lightweight machine learning model to obtain an initial anomaly score vector composed of a series of initial anomaly score values; constructing a correlation parameter space matrix based on the voltage-current short-time time series joint feature and the initial anomaly score vector; performing bidirectional interaction correction on the voltage-current short-time time series joint feature and the initial anomaly score vector based on the correlation parameter space matrix to obtain a corrected voltage-current short-time time series joint feature and a corrected initial anomaly score vector; and inputting the corrected voltage-current short-time time series joint feature and the corrected initial anomaly score vector into the lightweight machine learning model for anomaly detection.

[0014] In the above-mentioned distributed electric meter remote monitoring method based on the Internet of Things, the abnormal event message comprises an abnormal event type, an abnormal event occurrence time, an instantaneous current short-time time series, and an instantaneous voltage short-time time series.

[0015] In the above-mentioned distributed electric meter remote monitoring method based on the Internet of Things, the first preset time period is 15 minutes, and the second preset time period is 4 hours.

[0016] In the above-mentioned remote monitoring method for distributed electricity meters based on Internet of Things, step S3 comprises: performing statistical analysis on the local data stream in the first preset time period to obtain average voltage, average current, average active power, total power consumption, maximum voltage, minimum voltage and peak current; and packing the average voltage, average current, average active power, total power consumption, maximum voltage, minimum voltage and peak current to obtain the data aggregation record.

[0017] In the above-mentioned remote monitoring method for distributed electricity meters based on Internet of Things, step S4 comprises: immediately connecting to the wireless network in response to detecting the abnormal event message; and sending the abnormal event message to the cloud platform through the MQTT protocol with the highest priority.

[0018] According to another aspect of the present application, a remote monitoring system for distributed electricity meters based on Internet of Things is provided for performing the above-mentioned remote monitoring method for distributed electricity meters based on Internet of Things, which comprises: a data acquisition and storage module for sampling and analog-digital conversion of analog electrical signals by a micro control unit built in the electricity meter and storing the same in a ring buffer of the electricity meter to obtain a local data stream; a data stream anomaly detection module for continuously detecting anomalies in the local data stream by the micro control unit and generating an abnormal event message after successful anomaly detection; a data aggregation module for data aggregation of the local data stream by the micro control unit in a first preset time period to obtain a data aggregation record; an abnormal event sending module for sending the abnormal event message to the cloud platform through the wireless network with the highest priority in response to detecting the abnormal event message; and a data storage and sending module for storing the data aggregation record in a non-volatile memory and sending periodic aggregation data packets obtained by packing a plurality of data aggregation records according to a second preset time period to the cloud platform.

[0019] The data stream anomaly detection module is configured to: perform data cleaning on the local data stream in a short-time sliding window according to data sample dimensions to obtain an instantaneous current short-time time series and an instantaneous voltage short-time time series; extract short-time current pattern features and short-time voltage pattern features from the instantaneous current short-time time series and the instantaneous voltage short-time time series; fuse the short-time current pattern features and the short-time voltage pattern features to obtain voltage-current short-time time series joint features; and input the voltage-current short-time time series joint features into a lightweight machine learning model for anomaly detection.

[0020] Compared with the prior art, the application provides a distributed electric meter remote monitoring system and method based on Internet of Things, which first collects real-time electric signals at an electric meter terminal and performs analog-digital conversion, then utilizes a lightweight machine learning model to perform real-time analysis on short-time data flow to generate high-priority abnormal events, and synchronously adopts periodic statistical analysis to aggregate long-time data, and further, through a double-channel reporting mechanism driven by events and periodicity, the abnormal events are reported to a cloud platform in real time with the highest priority, and the aggregated records are packaged and sent in time-sharing mode. In this way, the sensitivity of edge real-time analysis and the economy of data hierarchical reporting are combined, effectively solving the network congestion and delay problem caused by massive data concurrency, and significantly improving the real-time performance and accuracy of electric abnormality monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The drawings provided in the present application are used to provide further understanding of embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 A flowchart of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiments of the present application.

[0023] Figure 2 A data flow diagram of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiments of the present application.

[0024] Figure 3 A flowchart of sub-step S2 of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiments of the present application.

[0025] Figure 4 A flowchart of sub-step S3 of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiments of the present application.

[0026] Figure 5 A flowchart of sub-step S4 of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiments of the present application.

[0027] Figure 6 A block diagram of the distributed electric meter remote monitoring system based on Internet of Things according to the embodiments of the present application. DETAILED DESCRIPTION

[0028] Embodiments of the present disclosure will be described below in greater detail with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein, but rather, these embodiments are provided so as to more completely and thoroughly understand the present disclosure. It is understood that the drawings and embodiments of the present disclosure are for exemplary purposes only and are not intended to limit the scope of protection of the present disclosure.

[0029] To solve the problems in the background art, the present application provides a remote monitoring method for distributed electric meters based on Internet of Things. Figure 1 A flow chart of the remote monitoring method for distributed electric meters based on Internet of Things according to an embodiment of the present application. Figure 2 A data flow chart of the remote monitoring method for distributed electric meters based on Internet of Things according to an embodiment of the present application. As shown in Figure 1 and Figure 2 The remote monitoring method for distributed electric meters based on Internet of Things comprises the following steps: S1, a micro control unit built in an electric meter samples and digitizes an analog electric signal and stores it in a ring buffer of the electric meter to obtain a local data stream; S2, the micro control unit continuously detects the local data stream for abnormalities and generates an abnormal event message after successful abnormality detection; S3, the micro control unit aggregates data of the local data stream at a first preset time period to obtain a data aggregation record; S4, in response to detecting the abnormal event message, the abnormal event message is sent to a cloud platform through a wireless network at the highest priority; and S5, the data aggregation record is stored in a non-volatile memory, and a plurality of data aggregation records are packaged into a periodic aggregation data packet according to a second preset time period, and the periodic aggregation data packet is sent to the cloud platform.

[0030] In the remote monitoring method for distributed electric meters based on Internet of Things, the step S1, the micro control unit built in the electric meter samples and digitizes the analog electric signal and stores it in the ring buffer of the electric meter to obtain the local data stream. In a possible embodiment, the analog electric signal is output by a metering chip built in the electric meter, the sampling frequency for sampling the analog electric signal is 200 Hz, and the local data stream comprises instantaneous current value, instantaneous voltage value and instantaneous active power. It can be understood that, since the original analog electric signal cannot be directly processed by a digital system, the micro control unit built in the electric meter is used to sample and digitize the analog electric signal in the present application, so as to convert the continuous analog electric signal into discrete digital signal and store it efficiently, to provide reliable local data stream basis for subsequent data abnormality analysis, to meet the data processing and interaction requirements of the smart meter, to realize accurate digital conversion and orderly storage of the electric signal, to ensure data integrity and timeliness, and to provide data guarantee for the meter to realize metering, monitoring, communication and other functions.

[0031] In particular, in one possible embodiment, the implementation process of step S1 is as follows: first, the metering chip built-in the electric meter outputs an analog electric signal, the micro control unit is configured to a sampling frequency of 200Hz by the built-in timer, and the analog electric signal is sampled according to the Nyquist sampling theorem, which can effectively capture the complete waveform characteristics of the 50Hz power frequency electric signal and its high-order harmonics, and avoid data distortion caused by insufficient sampling. Secondly, the analog-to-digital converter of the micro control unit adopts a successive approximation architecture, and discretizes the continuous analog electric signal into digital quantities with a resolution of 12 bits. The sampling and holding circuit in the front end of the ADC ensures the stability of the signal during conversion, and the generated digital code reflects the instantaneous state of the instantaneous current, voltage and active power in real time. Then, the converted digital signal is transmitted to the ring buffer area in the memory through the micro control unit bus, the buffer area adopts a circular queue structure, new data is written in time sequence, and the earliest data is automatically overwritten when it is full, forming a first-in first-out storage mechanism, and the micro control unit adds a time stamp to each data point for subsequent time sequence analysis. Finally, the local data stream is organized in an array form in the buffer area, including instantaneous current value, instantaneous voltage value and instantaneous active power, the micro control unit responds to the sampling timer trigger through the interrupt mechanism, and captures and stores each sampling point in time, and the design of the read and write pointers of the buffer area ensures that data acquisition and subsequent processing are executed in parallel, thereby providing raw data support for subsequent analysis and reporting.

[0032] In the above-mentioned distributed electric meter remote monitoring method based on Internet of Things, the micro control unit continuously detects the local data stream for abnormalities and generates an abnormal event message after successful abnormality detection in step S2. In one possible embodiment, the abnormal event message includes the abnormal event type, the abnormal event occurrence time, the instantaneous current short-time time sequence and the instantaneous voltage short-time time sequence. It should be understood that, during the transmission and storage of the local data stream, abnormalities such as abnormal fluctuations of the instantaneous current value, the instantaneous voltage value and the instantaneous active power may occur, which may mean electric meter failure, line problem or abnormal power consumption, etc. In order to discover these potential problems in time and avoid problem expansion, the micro control unit continuously detects the local data stream for abnormalities, which can discover abnormal conditions in the data stream in the first time, and generates an abnormal event message containing the abnormal event type, the occurrence time, the instantaneous current short-time time sequence and the instantaneous voltage short-time time sequence, etc. after successful detection, thereby providing accurate data support for the rapid reporting and processing of subsequent abnormal events, so that relevant personnel can take measures to solve abnormal problems in time.

[0033] In particular, in one possible embodiment, Figure 3 The flow chart of the sub-step S2 of the distributed electric meter remote monitoring method based on Internet of Things according to the embodiment of the present application is shown in FIG. 2. As shown in FIG. 2, the micro control unit continuously detects the local data stream for abnormalities in step S201, and generates an abnormal event message after successful abnormality detection in step S202. In one possible embodiment, the abnormal event message includes the abnormal event type, the abnormal event occurrence time, the instantaneous current short-time time sequence and the instantaneous voltage short-time time sequence. Figure 3As shown, the step S2 comprises: S21, cleaning the local data stream in the short-time sliding window according to the data sample dimension to obtain an instantaneous current short-time time sequence and an instantaneous voltage short-time time sequence; S22, extracting a short-time current mode feature and a short-time voltage mode feature from the instantaneous current short-time time sequence and the instantaneous voltage short-time time sequence; S23, fusing the short-time current mode feature and the short-time voltage mode feature to obtain a voltage-current short-time time sequence joint feature; and S24, inputting the voltage-current short-time time sequence joint feature into a lightweight machine learning model for anomaly detection.

[0034] Specifically, the step S21 comprises: cleaning the local data stream in the short-time sliding window according to the data sample dimension to obtain an instantaneous current short-time time sequence and an instantaneous voltage short-time time sequence. Specifically, by cleaning the data stream in the short-time sliding window according to the data sample dimension, the noise points, invalid data and abnormal fluctuation values are removed, the original data stream is converted into continuous and accurate instantaneous current short-time time sequence and instantaneous voltage short-time time sequence, the purity of the local data stream is significantly improved, the invalid data caused by sampling errors, electromagnetic interference and other factors are effectively filtered out, the generated instantaneous current and voltage short-time time sequence is closer to the actual power consumption, reliable data support is provided for the subsequent feature extraction step, and the accuracy of anomaly detection is improved, the false alarm problem caused by data noise is reduced, and it is ensured that the system can accurately identify real power consumption anomalies.

[0035] Specifically, in a possible embodiment, the implementation process of the step S21 is as follows: first, the time span of the short-time sliding window is determined, for example, in seconds, and each instantaneous current value and voltage value in the window is traversed according to the data sample dimension. For each data point, first, a reasonable threshold is set based on the historical normal data range or industry standard, the current data is checked whether it is out of the threshold range, and if it is out, it is determined as an abnormal value; at the same time, it is detected whether there is data missing or format error. For abnormal values, the adjacent value interpolation method is used for correction, for example, the average value of the previous and subsequent valid data is used to replace the abnormal point; for missing data, linear interpolation is used to supplement according to the change trend of the previous and subsequent data. In the cleaning process, the time stamp and processing method of data cleaning are recorded synchronously to ensure that the data is traceable. After cleaning all data points in a single window, the valid data is arranged in time sequence, and continuous instantaneous current short-time time sequence and instantaneous voltage short-time time sequence are generated, so that the current and voltage data at each time point meet the quality requirements of subsequent feature extraction.

[0036] Specifically, the step S22 extracts short-time current mode features and short-time voltage mode features from the instantaneous current short-time time series and the instantaneous voltage short-time time series. It should be understood that the instantaneous current short-time time series and the instantaneous voltage short-time time series contain a large amount of raw data information, but if directly used for subsequent calculation, the complex data form will increase the calculation burden, and invalid information may cover the real abnormal mode. Based on this, the application focuses on key data features through feature extraction, eliminates redundant information, and avoids the complexity of data leading to low analysis efficiency. Specifically, the application extracts short-time mode features that can represent the change rule of current and voltage from the instantaneous current short-time time series and the instantaneous voltage short-time time series after data cleaning, such as waveform trend, fluctuation amplitude, peak value distribution, etc. The instantaneous current short-time time series and the instantaneous voltage short-time time series are converted into short-time current mode features and short-time voltage mode features with physical meaning, providing structured input for subsequent fusion feature processing.

[0037] In particular, in one possible embodiment, the implementation process of the step S22 is as follows: first, time domain feature extraction is performed on the instantaneous current short-time time series, the mean, variance, and root mean square value of the instantaneous current short-time time series are calculated to represent the overall level and fluctuation degree of the current, the peak value, valley value, and occurrence time are extracted to determine whether there is a current mutation, and the number of zero-crossing points and the rising / falling edge slope are counted to analyze the stability of the current waveform. At the same time, similar operations are performed on the instantaneous voltage short-time time series to obtain the mean, variance, peak value, etc. of the instantaneous voltage short-time time series. Then, frequency domain analysis is performed, the time series is converted to the frequency domain through fast Fourier transform, the amplitude and phase of each harmonic component are extracted, and the distribution of high-order harmonics such as 3rd and 5th harmonics is focused on to determine whether there is voltage and current abnormality caused by harmonic pollution. Then, the time sequence mode features of the time series are analyzed, the sliding window method is used to identify periodic fluctuation patterns, the correlation coefficient of adjacent data points is calculated, and whether there is non-periodic abnormal fluctuation is detected. In addition, transient features such as voltage transient increase / decrease amplitude, duration, and occurrence time, current sudden increase / decrease rate, etc. are extracted, which are crucial for identifying transient faults. Finally, all the time domain, frequency domain, and transient features extracted are standardized to eliminate the dimension influence and form short-time current mode features and short-time voltage mode features with unified dimensions, providing standardized input for subsequent feature fusion.

[0038] Specifically, the step S23 fuses the short-time current mode feature and the short-time voltage mode feature to obtain a voltage-current short-time time sequence joint feature. That is, the short-time current mode feature representing the current change rule is organically fused with the short-time voltage mode feature representing the voltage change rule to generate a voltage-current short-time time sequence joint feature containing the cooperative change information of the current and the voltage, which can not only reflect the respective feature attributes of the current and the voltage, but also reflect the coupling relationship such as the phase difference and the power factor change between the current and the voltage, effectively capture the cooperative abnormal mode of the current and the voltage such as the short-circuit feature of the current sudden increase accompanied by the voltage sudden drop, and avoid the missed detection problem caused by the one-sided information in the single feature analysis. In this way, the recognition ability of the system to the composite anomaly is improved, more comprehensive data support is provided for the type positioning and reason analysis of the abnormal event, the feature dimension redundancy is reduced, and the model reasoning efficiency is improved.

[0039] In particular, in a possible embodiment, the implementation process of the step S23 is as follows: first, the extracted short-time current mode feature and the short-time voltage mode feature are dimensionally aligned, and if the dimensions of the two features are different, the dimension consistency is achieved through principal component analysis dimension reduction. Then, the feature splicing method is adopted to sequentially connect the short-time current mode feature and the short-time voltage mode feature to form a multi-dimensional joint feature containing the current time domain feature, the current frequency domain feature, the voltage time domain feature, the voltage frequency domain feature, and the transient feature. Then, the electrical physical quantity correlation calculation is introduced, such as the power factor and the phase difference coupling features based on the instantaneous values of the current and the voltage, and the new dimension is embedded in the multi-dimensional joint feature to strengthen the correlation representation of the current and the voltage. Then, the multi-dimensional joint feature is standardized to eliminate the dimension influence through the minimum-maximum normalization or Z-score standardization, and the contribution weight of each feature dimension to the model is ensured to be consistent. In order to adapt to the calculation efficiency requirement of the lightweight machine learning model, the linear discriminant analysis is adopted to reduce the dimension of the joint feature, eliminate the redundant features, and retain the key feature combination with the best discriminability. Finally, the voltage-current short-time time sequence joint feature containing the independent features and the coupling features of the current and the voltage is generated to provide high-quality feature input for the subsequent input of the lightweight model for anomaly detection.

[0040] Specifically, the step S24 inputs the voltage-current short-time time series joint feature into a lightweight machine learning model for anomaly detection. It can be understood that, since the voltage-current short-time time series joint feature has performed dimension cleaning and feature fusion on the local data stream, removed redundant information and retained key patterns, and the lightweight machine learning model has high processing efficiency and low resource consumption for such parameters with small amount, the anomaly detection can be completed locally and in real time at the power meter terminal. Therefore, the application utilizes the fast analysis capability of the lightweight machine learning model on the voltage-current short-time time series joint feature to identify abnormal patterns in the short-time data stream locally and in real time at the power meter terminal, such as abnormal fluctuations of instantaneous current or instantaneous voltage, power mutation, etc., to realize real-time early warning of electricity use anomaly, while reducing data transmission amount, network load and cloud computing pressure through local processing, and balancing real-time performance and economy.

[0041] In particular, in a possible embodiment, the implementation process of the step S24 is as follows: after obtaining the voltage-current short-time time series joint feature, the feature needs to be preprocessed to meet the input requirements of the lightweight machine learning model. Considering the limited computing resources and storage capacity of the power meter terminal, the lightweight machine learning model needs to have the characteristics of low complexity and high real-time performance. In a possible embodiment, an integrated model based on multiple extremely shallow decision trees is selected as the network structure of the lightweight machine learning model.

[0042] First, considering the limited computing resources of the power meter terminal, a lightweight machine learning model composed of multiple decision trees with a depth of only 3-5 layers is selected. By limiting the tree depth and feature splitting dimension, the computing complexity is significantly reduced while ensuring detection accuracy, which adapts to the edge computing capability of the micro control unit. Before feature input, the voltage-current short-time time series joint feature needs to be preprocessed. The instantaneous current and voltage time series data with a sampling frequency of 200 Hz in the short-time sliding window are normalized to eliminate the dimension difference between current and voltage, and the statistical features of the time series data are extracted, such as mean, variance, peak value, rising slope, etc., to convert the multi-dimensional time series features into fixed-dimensional feature vectors suitable for decision tree processing.

[0043] Then, in the model inference stage, the lightweight machine learning model improves the detection robustness by parallel computing the classification results of multiple decision trees. Each decision tree splits the input feature vector layer by layer based on information gain, for example, preferentially selecting the feature with the highest correlation with abnormal events as the splitting node, such as current mutation slope or voltage drop amplitude. Since the tree depth is extremely shallow, the inference process of a single tree can be completed in a few conditional judgments. The lightweight machine learning model integrates the output results of multiple trees using a voting mechanism. When more than half of the decision trees determine that it is abnormal, the model outputs an abnormal signal, effectively reducing the misjudgment probability of a single decision tree.

[0044] For a lightweight machine learning model based on a decision tree, each branch obtains an anomaly score based on rules or proceeds to the next branch, so as to finally obtain a series of anomaly scores. Considering that the voltage-current short-time time series features may have certain complex relationships, such as the linear impedance relationship of voltage-current, the short-time time series harmonic component relationship of voltage and current, etc., it is expected that the anomaly score sequence of the decision tree can also fully consider these complex relationships beyond the rules, thereby improving the accuracy of anomaly detection.

[0045] Based on this, in another embodiment, the voltage-current short-time time series joint features are input into a lightweight machine learning model for anomaly detection, including: first, inputting the voltage-current short-time time series joint features into the lightweight machine learning model to obtain an initial anomaly score vector composed of a series of initial anomaly score values. Here, the lightweight machine learning model is a model based on a decision tree, and each branch generates an initial anomaly score according to a preset rule or empirical value. For example, when the instantaneous current value exceeds a certain threshold (such as 10A), the initial anomaly score is 0.7; when the voltage harmonic distortion rate exceeds a certain threshold (such as 5%), the initial anomaly score is 0.8. These initial anomaly score values are based on a single feature or a simple rule for preliminary judgment, which reflects the model's preliminary assessment of the degree of abnormality of the current data point. However, this preliminary assessment may ignore the internal relevance between different features, leading to false positives or false negatives.

[0046] Then, based on the voltage-current short-time time series joint features and the initial anomaly score vector, an association parameter space matrix is constructed, expressed as: ; wherein, represents the eigenvalue of the feature value at the position in the association parameter space matrix, represents the th feature value in the voltage-current short-time time series joint features, represents the th initial anomaly score value in the initial anomaly score vector, represents and corresponding first weighting parameters, represents corresponding second weighting parameters, represents corresponding third weighting parameters, wherein the weighting parameters are determined through offline training and optimization. For example, these parameters can be learned through historical normal and abnormal data using a regression analysis method, so that the matrix can maximize the internal relationship between the features and the anomaly scores.

[0047] Here, for the parameters , and , the path independence between the feature values and the anomaly scores needs to be considered, that is, satisfying: This means that for any pair of features and anomaly scores, the correlation between them is either mainly determined by the features or mainly determined by the anomaly scores, avoiding redundant or conflicting correlations.

[0048] At the same time, in order to make each decision rule have a linear matching response, that is, to maximize the matching fitting of the nonlinear correlation, the application also needs to consider the linear matching response to the correlation path, that is, satisfying: ; wherein, is a linear proportionality coefficient. The introduction of the linear proportionality coefficient allows the nonlinear correlation between the features and the anomaly scores to be maximally fitted into a linear space, thereby avoiding its abnormal topological influence on the linear rule distribution structure of the decision tree. Specifically, the linear proportionality coefficient is determined based on historical normal data fitting and expert experience calibration.

[0049] In addition, in order to apply the existing decision tree rules, the value of is limited to .

[0050] wherein, is a preset threshold. By introducing the preset threshold, it is ensured that the decision tree can be based on a stable feature energy range when making an anomaly judgment. It actually limits the value of by limiting the effective range of the feature after being adjusted by the weight , so that it conforms to the linear rule distribution structure of the decision tree. Specifically, the preset threshold is determined based on the normal fluctuation range of the feature and the sensitivity tuning of the decision tree.

[0051] In this way, the correlation parameter space matrix can be generated, which is a linear response space with feature value-anomaly value independent sub-dimensions. The independent sub-dimensions are used to ensure the relative independence between each fault indicator, so as to improve the applicability of each individual rule under the binary rule structure of the decision tree. The linear response space can make each decision rule have a linear matching, that is, to maximize the matching fitting of the nonlinear correlation, thereby avoiding its abnormal topological influence on the linear rule distribution structure of the decision tree, such as short-time time series harmonic component relationship.

[0052] Then, based on the correlation parameter space matrix, the voltage-current short-time time series joint feature and the initial anomaly score vector are bidirectionally interactively corrected to obtain a corrected voltage-current short-time time series joint feature and a corrected initial anomaly score vector, represented as: ; wherein, Represents the associated parameter space matrix, This represents the initial anomaly score vector. Represents the transpose of a vector. This indicates the short-time timing joint characteristics of the voltage-current system. Represents matrix multiplication. This represents the corrected voltage-current short-time timing joint characteristic. This represents the initial anomaly score vector for correction.

[0053] In other words, bidirectional interactive correction allows features and outlier scores to influence and correct each other. For example, if a feature value is initially considered normal, but it has a strong correlation with multiple features with high outlier scores, then after correction, the weight of that feature or its representation in the joint features may be adjusted to better reflect potential anomalies. Conversely, if an initial outlier score is high, but its corresponding feature is considered normal after correction, then that outlier score may be lowered. This bidirectional interactive correction enables the model to more comprehensively consider the synergistic effects between features and the inherent consistency of outlier scores, thereby improving the accuracy of anomaly detection.

[0054] Finally, the corrected voltage-current short-time time series joint features and the corrected initial anomaly score vector are input into the lightweight machine learning model for anomaly detection. This allows the anomaly score value of the rule-based decomposition threshold of the decision tree to achieve a highly interactive composite mapping based on the correlation distribution with the voltage-current short-time time series features. In other words, when making the final judgment, the model not only considers the threshold of a single feature but also integrates the complex correlations between features and the mutual verification of anomaly scores, thereby significantly improving the accuracy of anomaly detection. For example, in a power grid, slight voltage fluctuations may not constitute an anomaly on their own, but if they are accompanied by abnormal harmonic components of the current, and the two show a strong correlation in the correlation parameter space matrix, then after correction, the model is more likely to identify them as potential equipment failures or electricity theft, avoiding potential false negatives in traditional methods.

[0055] In the above-mentioned remote monitoring method for the distributed electric meter based on the Internet of Things, the micro control unit aggregates the local data stream to obtain the data aggregation record in the first preset time period. The first preset time period is 15 minutes. Specifically, if the local data stream is directly uploaded, the network transmission pressure will increase dramatically and the cloud storage and calculation load will be heavy. However, the data aggregation in the fixed period can greatly compress the data volume, thereby avoiding network congestion and resource waste caused by the full transmission of the original data. Therefore, the application converts the instantaneous electric signal data in the first preset time period into the data aggregation record with statistical significance through periodic data aggregation, thereby reducing the data volume to be uploaded to the cloud platform, providing periodic statistical basis for power consumption trend analysis and load management, and realizing efficient compression and value extraction of the data. Through the data aggregation manner, the cloud platform can quickly master the power consumption profile based on the aggregated data, thereby providing data support for power dispatching, abnormal trend prediction, and the like, and improving the overall operation efficiency of the system.

[0056] In particular, in one possible embodiment, Figure 4 The flow chart of the step S3 of the remote monitoring method for the distributed electric meter based on the Internet of Things according to the embodiment of the application is shown in FIG. 3. As shown in FIG. 3, the step S3 includes the following steps. Figure 4 The step S3 includes the following steps. S31: statistically analyzing the local data stream in the first preset time period to obtain the average voltage, the average current, the average active power, the total power consumption, the maximum voltage, the minimum voltage, and the peak current. S32: packaging the average voltage, the average current, the average active power, the total power consumption, the maximum voltage, the minimum voltage, and the peak current to obtain the data aggregation record.

[0057] Specifically, the step S31 statistically analyzes the local data stream in the first preset time period to obtain the average voltage, the average current, the average active power, the total power consumption, the maximum voltage, the minimum voltage, and the peak current. Specifically, the statistical analysis of the local data stream in the 15-minute period obtains the parameters with statistical significance such as the average voltage and the average current, thereby forming the data aggregation record, providing the simplified and valuable data for the periodic packaging and sending to the cloud platform, and enabling the cloud platform to master the power consumption profile in the time period based on the aggregated data, thereby meeting the demand for the macroscopic monitoring and analysis of the long-time data. After the statistical analysis, the data transmission volume is significantly reduced, the network transmission efficiency is improved, and the processing pressure of the cloud is reduced. Meanwhile, the cloud platform can obtain the representative power consumption statistical information, thereby facilitating the power consumption trend analysis, load management, and the like, and realizing the efficiency of the data reporting under the premise of ensuring the effectiveness of the data.

[0058] In particular, in one possible embodiment, the step S31 is implemented as follows: first, the micro control unit determines the time range for which statistical analysis is required according to the first preset time period, extracts the local data stream in this time period from the ring buffer of the electric meter, which contains data such as instantaneous current value, instantaneous voltage value and instantaneous active power. Then, the extracted instantaneous current value is processed, all the instantaneous current values in this time period are added and then divided by the number of data points to obtain the average current, and in the same way, the instantaneous voltage value is summed and averaged to obtain the average voltage. The instantaneous active power is summed and averaged to obtain the average active power. Then, the total power consumption is calculated by integrating or accumulating the active power in this time period to obtain the total power consumption in 15 minutes. In the statistical process, the instantaneous voltage value is traversed simultaneously to find the maximum value and the minimum value, which are the maximum voltage and the minimum voltage respectively, and the instantaneous current value is traversed to find the maximum value, which is the peak current.

[0059] Specifically, the step S32 packs the average voltage, average current, average active power, total power consumption, maximum voltage, minimum voltage and peak current to obtain the data aggregation record. That is, by packing various statistical parameters, a standardized data aggregation record is formed, which converts scattered parameters into a unified information unit for processing, facilitating storage in the non-volatile memory and maintaining data integrity and consistency when transmitting to the cloud platform through the wireless network, ensuring that the cloud platform can accurately receive and analyze the power consumption statistical information in this time period. The packed data aggregation record becomes an independent information unit, the data structure is more regular, which facilitates storage management and transmission scheduling, reduces the fragmentation problem of data transmission, improves the reliability of data in the storage and transmission process, and improves the data processing efficiency, realizes the standardization and systematization of data reporting.

[0060] In particular, in one possible embodiment, the implementation process of the step S32 is as follows: first, determine the format standard of the data aggregation record, for example, in JSON format or custom binary format, to ensure the standardization and parsability of the data. Then, add a timestamp to each data aggregation record, which corresponds to the start or end time of the 15-minute period, so that the cloud platform can identify the time attribute of the data. Then, arrange all the statistical parameters according to the preset format structure, for example, construct an object containing parameter fields in JSON format, to ensure that each parameter has a corresponding field name and accurate value. To ensure the accuracy of data transmission, a check field can be added during packaging, such as generating a check code by calculating the hash value of the parameter. After completing the parameter arrangement and check code addition, all contents are packaged into a complete data aggregation record data packet according to the format requirements. Finally, the generated data aggregation record is stored in the non-volatile memory of the electric meter, waiting for subsequent packaging into a periodic aggregation data packet according to the second preset time period of 4 hours, and sending to the cloud platform together with the aggregation records of other time periods.

[0061] In the above-mentioned Internet of Things-based distributed electric meter remote monitoring method, the step S4 sends the abnormal event message to the cloud platform through the wireless network with the highest priority in response to detecting the abnormal event message. It should be understood that, due to the strong timeliness and immediate handling characteristics of abnormal events, if they are sent with the same priority as periodic data aggregation records, the event response may be delayed due to network congestion or transmission queue backlog, thereby causing safety accidents or economic losses. Therefore, the abnormal event message needs to be sent with the highest priority to ensure that it has absolute priority in network transmission and avoid being squeezed out of bandwidth resources by regular data transmission, thereby meeting the core needs of immediate discovery and immediate reporting of abnormal situations.

[0062] In particular, in one possible embodiment, Figure 5 The flowchart of the sub-step S4 of the Internet of Things-based distributed electric meter remote monitoring method according to the embodiment of the present application is shown in FIG. 4. As shown in FIG. 4, the step S4 includes: S41, in response to detecting an abnormal event message, immediately connecting to a wireless network; and S42, sending the abnormal event message to the cloud platform through the MQTT protocol with the highest priority. Figure 5

[0063] Specifically, the step S41, in response to detecting an abnormal event message, immediately connects to a wireless network. That is, upon detecting an abnormal event message, a wireless network connection is triggered immediately to provide an immediate transmission physical channel for the abnormal event message, ensuring that the abnormal information can quickly enter the transmission link after being generated, thereby avoiding message retention due to network connection delay, and thus realizing the timeliness guarantee of the whole process from detection to reporting of the abnormal event, and ensuring that the cloud platform can timely obtain abnormal data and start the subsequent processing mechanism. ​

[0064] In particular, in one possible embodiment, the step S41 is implemented as follows: when the micro control unit of the electric meter terminal detects the generation of an abnormal event message, first wake up the built-in wireless network module and initialize its configuration, including setting the communication frequency band, network protocol parameters such as APN access point, authentication information. Then, the micro control unit sends a connection instruction to the wireless network module, triggering the module to scan the available network signal, and selecting the base station or access point with the optimal signal strength for association. During the association process, the authentication and encryption negotiation is completed according to the preset network protocol, and the connection of the wireless link control layer and the network layer is established. After the connection is established, the micro control unit real-time monitors the state feedback of the network module, if the confirmation signal of successful connection is received, the transmission channel is immediately allocated for the abnormal event message; if the connection fails, the reconnection mechanism is started, the first reconnection interval is set to 200 ms, if the reconnection fails for 3 times in succession, the error log is recorded, and the connection is tried again after 5 minutes. During the whole connection process, the micro control unit will preferentially guarantee the power supply of the wireless network module, ensure the stability of the connection establishment, at the same time shield the transmission request of other non-emergency data, avoid occupying the network resources of the abnormal event message, and after the abnormal event message is successfully sent to the cloud platform, it is decided whether to maintain the connection or enter the low-power standby state according to the system setting.

[0065] Specifically, the step S42 sends the abnormal event message to the cloud platform through the MQTT protocol with the highest priority. It should be understood that the MQTT protocol is a lightweight communication protocol, which is built on the TCP / IP protocol and transmits messages through the MQTT server, which can improve the scalability of the system and the adaptability of the dynamic network topology, and is suitable for running on resource-limited devices. Based on this, the application realizes the queuing ability of the abnormal event message in network transmission through the efficient transmission characteristics and the highest priority mechanism of the MQTT protocol, so that it reaches the cloud platform in the shortest time, and the cloud platform can immediately trigger the early warning response or fault handling process, thereby improving the real-time performance of the power consumption abnormality monitoring and ensuring the safe operation of the power grid.

[0066] In particular, in one possible embodiment, the step S42 is implemented as follows: first, integrate the MQTT client library in the micro control unit of the meter terminal, complete the protocol initialization configuration, including setting the client ID, subscribing to the topic such as “ / meter / exception”, and publishing the topic such as “ / meter / exception / report”, and configuring the clean session as true to ensure the independence of each connection. When detecting the abnormal event message, first call the connect method of the MQTT client to establish the TCP / IP connection with the cloud platform MQTT server, and carry the will message during the connection process. If the meter suddenly goes offline, an abnormal disconnection notification can be automatically published. After the connection is successful, the abnormal event message is encapsulated in the JSON format, and is sent through the publish method, wherein the QoS parameter is set to 2 to ensure that the message is sent only once and is not repeated, and the retain flag is set to false. When sending, the message receipt mechanism is enabled. If the PUBCOMP confirmation packet of the server is not received, the retransmission logic is triggered, and the number of retransmissions can be set to 3 times, with an interval of 500 ms each time, to ensure that the message can be reliably delivered. After the transmission is completed, it is determined whether to disconnect according to the system strategy. If there are still abnormal events to be sent, the connection is maintained. Otherwise, the connection is disconnected after a delay of 10 seconds to save power consumption. During the whole process, the error callback function of the MQTT client is listened to, the network exception is captured in real time, and the corresponding recovery mechanism is triggered to ensure the stability of the transmission link.

[0067] In the above-mentioned distributed meter remote monitoring method based on the Internet of Things, the step S5 stores the data aggregation record in the non-volatile memory, and packs a plurality of data aggregation records into a periodic aggregation data packet according to a second preset time period, and then sends the periodic aggregation data packet to the cloud platform. The second preset time period is 4 hours. It can be understood that the non-volatile storage ensures the persistent storage of the data aggregation record, and can avoid the loss of historical statistical data due to unexpected power failure and the like. Specifically, the data aggregation record is packed and sent to the cloud platform based on a 4-hour period, so as to realize the batch transmission of long-time data, so as to reduce the network communication frequency while ensuring that the electricity consumption statistical data is reported to the cloud platform at a reasonable time interval, and to provide periodic data support for electricity consumption analysis and load management.

[0068] Specifically, in one possible embodiment, step S5 is implemented as follows: After the microcontroller completes data aggregation every 15 minutes, it writes the generated aggregated records such as average voltage and average current into the meter's built-in non-volatile memory in binary or JSON format. The records are arranged in timestamp order, and a length identifier is added before each record for subsequent parsing. Simultaneously, the system starts a 4-hour periodic timer. When the timer reaches 4 hours, it triggers a packet transmission process, reading all aggregated records from the non-volatile memory within the last 4 hours (approximately 16 records), one record every 15 minutes, and merging them into a single data packet in chronological order. If the data volume is large, a lightweight compression algorithm is used to reduce the packet size. Subsequently, if there is no network connection, the connection interface of the wireless network module is called to establish a communication link with the cloud platform. The data packet is published to a specified topic via the MQTT protocol, and a QoS1 level is set to ensure that the message is delivered at least once. After transmission is complete, the transmitted record range is marked in the non-volatile memory. This range will be overwritten when new data is stored next time, freeing up storage space. If a network anomaly occurs during transmission, such as a connection loss, the data packet will be temporarily stored in a temporary buffer, and a retransmission mechanism will be initiated. The initial retransmission interval is 500ms. If three consecutive retransmissions fail, an error log will be recorded, and the data packet will be retransmitted as soon as the network recovers, ensuring that all aggregated records are reliably reported to the cloud platform.

[0069] In summary, the IoT-based distributed meter remote monitoring method based on the embodiments of this application is explained. First, real-time electrical signals are collected at the meter terminal and converted from analog to digital. Then, a lightweight machine learning model is used to analyze short-term data streams in real time to generate high-priority abnormal events. Simultaneously, periodic statistical analysis is used to aggregate long-term data. Furthermore, through a dual-channel reporting mechanism driven by both events and periods, abnormal events are reported to the cloud platform with the highest priority in real time, while the aggregated records are packaged and sent in time-sharing fashion. This combines the sensitivity of real-time edge analysis with the economic efficiency of hierarchical data reporting, effectively solving the network congestion and latency problems caused by massive data concurrency, and significantly improving the real-time performance and accuracy of electricity anomaly monitoring.

[0070] Figure 6 This is a block diagram of a distributed electricity meter remote monitoring system based on the Internet of Things according to an embodiment of this application. Figure 6As shown, the Internet of Things based distributed electric meter remote monitoring system 100 according to the embodiment of the present application comprises: a data acquisition and storage module 110, configured to sample and analog-digital convert analog electric signals by a micro control unit built in an electric meter, and store the analog-digital converted signals in a ring buffer of the electric meter to obtain a local data stream; a data stream anomaly detection module 120, configured to continuously detect anomalies of the local data stream by the micro control unit and generate an anomaly event message after successful anomaly detection; a data aggregation module 130, configured to aggregate data of the local data stream by the micro control unit in a first preset time period to obtain a data aggregation record; an anomaly event sending module 140, configured to send the anomaly event message to a cloud platform through a wireless network in a highest priority in response to detecting the anomaly event message; and a data storage sending module 150, configured to store the data aggregation record in a non-volatile memory, and send a plurality of data aggregation records packed as a periodic aggregation data packet to the cloud platform according to a second preset time period.

[0071] As described above, the Internet of Things based distributed electric meter remote monitoring system 100 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a smart Internet of Things based device asset operation and maintenance management algorithm, etc. In a possible implementation manner, the Internet of Things based distributed electric meter remote monitoring system 100 according to the embodiment of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the Internet of Things based distributed electric meter remote monitoring system 100 can be a software module in an operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the Internet of Things based distributed electric meter remote monitoring system 100 can also be one of many hardware modules of the wireless terminal.

[0072] Alternatively, in another example, the Internet of Things based distributed electric meter remote monitoring system 100 and the wireless terminal can also be separate devices, and the Internet of Things based distributed electric meter remote monitoring system 100 can be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in an agreed data format.

[0073] Here, those skilled in the art can understand that the specific operations of each step in the above Internet of Things based distributed electric meter remote monitoring system have been described in detail above with reference to the description of the Internet of Things based distributed electric meter remote monitoring method of Figures 1 to 5 the present application, and therefore, the repeated description thereof will be omitted.

Claims

1. A method for remote monitoring of distributed electricity meters based on the Internet of Things, characterized in that, The method comprises the following steps: S1: a micro control unit built in the electric meter samples and digitizes the analog electric signal and stores it in a ring buffer of the electric meter to obtain a local data stream; S2: the micro control unit continuously detects the local data stream for abnormality and generates an abnormal event message after successful abnormality detection; S3: the micro control unit aggregates data of the local data stream in a first preset time period to obtain a data aggregation record; S4: in response to detection of the abnormal event message, the abnormal event message is sent to a cloud platform through a wireless network at the highest priority; S5: the data aggregation record is stored in a non-volatile memory, and a plurality of data aggregation records are packaged into a periodic aggregation data packet according to a second preset time period, and then the periodic aggregation data packet is sent to the cloud platform; Step S2 comprises: cleaning the local data stream in a short-time sliding window according to the data sample dimension to obtain an instantaneous current short-time time series and an instantaneous voltage short-time time series; extracting short-time current mode features and short-time voltage mode features from the instantaneous current short-time time series and the instantaneous voltage short-time time series; fusing the short-time current mode features and the short-time voltage mode features to obtain voltage-current short-time time sequence joint features; inputting the voltage-current short-time time sequence joint features into a lightweight machine learning model for abnormality detection; wherein inputting the voltage-current short-time time sequence joint features into the lightweight machine learning model for abnormality detection comprises inputting the voltage-current short-time time sequence joint features into the lightweight machine learning model to obtain an initial abnormality score vector composed of a series of initial abnormality score values; Based on the voltage-current short-time time sequence combined feature and the initial abnormal score vector, a correlation parameter space matrix is constructed, the correlation parameter space matrix is a linear response space with eigenvalue-abnormal value independent sub-dimension, used for guaranteeing the relative independence between each fault indicator and maximizing matching fitting of the nonlinear correlation, expressed as: ; wherein, represents the eigenvalue of the feature value in the correlation parameter space matrix , represents the eigenvalue of the th feature value in the voltage-current short-time time sequence combined feature, represents the initial abnormal score value in the initial abnormal score vector, represents the initial abnormal score value in the initial abnormal score vector, represents and corresponding first weighting parameters, represents corresponding second weighting parameters, represents corresponding third weighting parameters. For the parameters and , the path independence between the eigenvalues and the outlier scores is satisfied: ; in response to linear matching of the correlation path, the following is satisfied: ； wherein is a linear proportionality factor; based on the correlation parameter space matrix, the voltage-current short-time time sequence joint features and the initial abnormality score vector are interactively corrected in both directions to obtain corrected voltage-current short-time time sequence joint features and corrected initial abnormality score vector; inputting the corrected voltage-current short-time time sequence joint features and the corrected initial abnormality score vector into the lightweight machine learning model for abnormality detection. 2.The IoT-based distributed electricity meter remote monitoring method according to claim 1, characterized in that, The analog electric signal is output by a metering chip built in the electric meter, the sampling frequency for sampling the analog electric signal is 200 Hz, and the local data stream comprises an instantaneous current value, an instantaneous voltage value and an instantaneous active power. 3.The IoT-based distributed electricity meter remote monitoring method according to claim 1, wherein, The abnormal event message comprises an abnormal event type, an abnormal event occurrence time, an instantaneous current short-time time series and an instantaneous voltage short-time time series. 4.The IoT-based distributed electricity meter remote monitoring method according to claim 1, wherein, The first preset time period is 15 minutes, and the second preset time period is 4 hours. 5.The IoT-based distributed electricity meter remote monitoring method according to claim 4, characterized in that, Step S3 comprises: statistical analysis of the local data stream in the first preset time period to obtain average voltage, average current, average active power, total power consumption, maximum voltage, minimum voltage and peak current; packaging the average voltage, the average current, the average active power, the total power consumption, the maximum voltage, the minimum voltage and the peak current to obtain the data aggregation record. 6.The IoT-based distributed electricity meter remote monitoring method according to claim 1, wherein, Step S4 comprises: in response to detection of the abnormal event message, immediately establishing a wireless network connection; The abnormal event message is sent to a cloud platform through an MQTT protocol with the highest priority.

7. An Internet of Things based distributed electricity meter remote monitoring system for performing the Internet of Things based distributed electricity meter remote monitoring method of any one of claims 1-6, characterized in that, The application relates to an abnormal event detection method and device for a smart meter. A data acquisition and storage module is used for a micro control unit built in an electric meter to sample and analog-digital convert an analog electric signal and store the analog-digital converted signal in a ring buffer of the electric meter to obtain a local data stream; A data stream abnormality detection module is used for the micro control unit to continuously detect the local data stream for abnormality and generate an abnormal event message after successful abnormality detection; A data aggregation module is used for the micro control unit to aggregate the local data stream for a first preset time period to obtain a data aggregation record; An abnormal event sending module is used for sending the abnormal event message to a cloud platform through a wireless network with the highest priority in response to detection of the abnormal event message; A data storage and sending module is used for storing the data aggregation record in a non-volatile memory and sending a plurality of data aggregation records to the cloud platform after the plurality of data aggregation records are packaged into a periodic aggregation data packet according to a second preset time period.

Images