A method and system for predicting current overload of electric energy meter

Abstract

The present invention provides a method for predicting electric energy meter current overload, which obtains electric energy data features related to current overload and user electric energy sample data; preprocesses the data to obtain electric energy data with time series features; classifies the electric energy data with time series features according to different time granularities to obtain multi-level electric energy time series data, extracts key features in the electric energy data sequence according to set rules, and obtains candidate shapelet sequences of different time windows; based on the shapelet sequence features and the electric energy characteristics, a gated neural network is used to construct and train an electric energy meter current overload prediction model to perform electric energy meter current overload discrimination; based on the electric energy meter current overload discrimination result, the current overload distribution of each time level is integrated to form a formatted early warning message for sending, on-site verification, and data update; the present invention can achieve normal monitoring and efficient management of current overload problems.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent electricity utilization technology, and particularly relates to a method and system for predicting current overload of an electric energy meter. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] A current overload occurs when an energy meter exceeds its rated current. This can occur due to a variety of factors, including increased voltage at the power supply end, an increase in the number or power of electrical devices in the circuit, and increased resistance due to aging wiring. In recent years, with the development of the power industry, and in particular the continuous innovation of user electricity consumption data collection technology, manual collection has evolved to intelligent collection systems. These systems have significantly improved the quality and efficiency of electricity consumption collection, with core components including smart energy meters. An energy meter is an instrument used to measure electrical energy. It measures and collects electricity consumption information at various collection points, primarily data related to current and voltage. It converts analog current and voltage signals into digital data, which is then transmitted to the data collection layer via a communication channel. If a meter experiences a current overload, it can exacerbate the aging of wiring, sockets, and other equipment in the circuit, leading to data collection failures and disrupting normal user use. It can also cause configuration issues in the collection system, leading to potential risks. For example, excessive overloads can cause wiring to overheat, potentially posing a fire hazard. Therefore, how to more effectively manage and prevent current overload in electricity meters is an urgent problem that needs to be solved.

[0004] In the big data analysis module of the electricity consumption information collection system, commonly used methods for predicting meter current overload include classification models such as logistic regression, random forest, BP neural network (BPNN), and radial basis function (RBF) neural network. Traditional machine learning models require feature engineering before training the classifier, which is labor-intensive. In addition, electricity consumption data has certain periodic and aperiodic patterns, making it difficult to manually filter features to uncover hidden information. Deep learning models such as BPNN and RBF can, to a certain extent, uncover potential correlations in electricity consumption data and achieve good classification results. However, these models struggle to identify sudden changes in user electricity consumption data and long-term dependencies. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a method and system for predicting current overload of electric energy meters. The present invention conducts multi-dimensional and hierarchical analysis on abnormal data, mines the potential correlation between current change trends in electricity consumption data, thereby effectively predicting the occurrence of current overload of electric energy meters, and realizing normal monitoring and efficient management of current overload problems.

[0006] According to some embodiments, the present invention adopts the following technical solutions:

[0007] A method for predicting current overload of an electric energy meter comprises the following steps:

[0008] Obtaining current overload-related electricity consumption data characteristics and user electricity consumption sample data;

[0009] Preprocess the data to obtain electricity consumption data with time series characteristics;

[0010] For electricity consumption data with time series characteristics, we classify it according to different time granularities to obtain multi-level electricity consumption time series data. We then use set rules to extract key features from the electricity consumption data sequence and obtain candidate shapelet sequences for different time windows.

[0011] Based on shapelet sequence features and electricity consumption characteristics, a gated neural network is used to build and train an electric energy meter current overload prediction model to identify electric energy meter current overload.

[0012] Based on the current overload judgment results of the electricity meter, the current overload distribution at each time level is integrated to form a formatted early warning message for sending, conduct on-site verification, and update the data.

[0013] As an optional implementation, the relevant electricity usage data includes basic user information, metering point information, energy meter information, electricity consumption information and load information.

[0014] As an optional implementation, the preprocessing includes data cleaning, data completion, data integration and normalization.

[0015] As an optional implementation, the specific process of classifying according to different time granularities includes: dividing the sample data according to time windows of different lengths.

[0016] As an optional implementation, the specific process of obtaining candidate shapelet sequences for different time windows includes: calculating the similarity between each candidate shapelet feature extraction rule and the electricity consumption data sequence based on KL divergence, and obtaining the best candidate shapelet feature through measurement and calculation to represent the key features of the electricity consumption data within the corresponding time range.

[0017] As an optional implementation, the electricity meter current overload prediction model includes an update gate and a reset gate. The update gate is used to control the extent to which the electricity consumption information state at the previous moment is transferred to the state at the current moment; the reset gate is used to control how much information of the previous state is written into the current candidate set.

[0018] As an optional implementation, the training process of the electric energy meter current overload prediction model includes:

[0019] The shapelet features of the electricity consumption data subsequences in different time windows and the initial basic information features of electricity consumption are uniformly encoded and used as the input of the GRU network respectively, and the model parameters are initialized;

[0020] Use the back-propagation algorithm to train the model and iteratively update the model parameters;

[0021] Substitute the updated model parameters into the objective function to determine whether the error between the objective function value and the true value is less than the error threshold. If so, proceed to the next step; otherwise, return to iterative update.

[0022] The objective function parameters are fixed, and the updated electric energy meter current overload prediction model is output.

[0023] A current overload prediction system for an electric energy meter, comprising:

[0024] A module for obtaining current overload-related electricity consumption data characteristics and user electricity consumption sample data;

[0025] A module for preprocessing data to obtain electricity consumption data with time series characteristics;

[0026] A module for classifying electricity consumption data with time series characteristics according to different time granularities to obtain multi-level electricity consumption time series data. It then uses set rules to extract key features from the electricity consumption data sequence and obtain candidate shapelet sequences for different time windows.

[0027] A module for identifying electric energy meter current overload using an electric energy meter current overload prediction model built and trained using a gated neural network based on shapelet sequence features and electricity consumption characteristics;

[0028] A module used to integrate the current overload distribution at each time level based on the current overload judgment results of the electricity meter, generate a formatted early warning message for transmission, conduct on-site verification, and update data.

[0029] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0030] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps in the above method are completed.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. Given the characteristics of electricity consumption data, such as large data volume, strong timeliness, and sparse distribution, there is a problem of difficulty in mining potential correlation relationships in electricity consumption data. The present invention adopts an electric energy meter current overload prediction method based on rule feature optimization GRU to predict electric energy meter current overload. On the one hand, based on existing expert experience and KL divergence, it realizes the extraction of explainable current overload anomaly features, and combines basic electricity consumption collection features to realize the characterization of unexplainable correlation features of current overload, thereby realizing multi-dimensional and multi-level mining of correlation features of current overload; on the other hand, it uses GRU to construct a current overload prediction model, deeply mining the potential time feature dependency in electricity consumption data, and further improving the prediction accuracy of current overload;

[0033] 2. In view of the time-varying patterns of electricity consumption data, the uneven distribution of collected electricity consumption data, and the identifiable time granularity, this invention adopts a time-level processing method and models according to different time windows. It can intuitively mine current overload patterns from multiple levels and granularities, and use shapelets to discover the key features of current change trends. On the one hand, this makes the model more interpretable and facilitates mutual verification of expert experience; on the other hand, it can focus on abnormal data subsequences, quickly extract key features, improve the efficiency of model training and fitting, and thus save storage space and computing resources.

[0034] 3. Due to the timeliness of electricity consumption data and the time-dependent relationship of data distribution, the present invention uses GRU to model the time series characteristics of electricity consumption data. GRU is a very effective variant of recurrent neural network. It has a simpler structure than LSTM network and faster training speed. At the same time, it can solve the long-term dependency problem in recurrent neural network. First, the correlation of candidate shapelet subsequences is calculated based on KL divergence, and the obtained optimal shapelet is used as part of the input of GRU. This can more effectively mine the key features in electricity consumption data, remove noise in the data, and improve the accuracy of model prediction. Secondly, shapelet modeling in different time windows is used to mine current overload anomalies, which can respond to current overload anomalies in a hierarchical manner, provide more extensive decision support, and further improve the accuracy of current overload prediction and the effectiveness of diagnosis of electricity meters.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 This is the overall flow chart of the current overload prediction method for electric energy meters;

[0038] Figure 2 Flowchart for shapelet discovery of electricity consumption data subsequences;

[0039] Figure 3 Illustration of data collection for low-voltage users in a certain community;

[0040] Figure 4 This is the GRU network structure diagram;

[0041] Figure 5 This is the overall architecture diagram of the electric energy meter current overload prediction model. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0044] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0045] The present invention utilizes shapelet, which is a time series subsequence with identification, which can effectively screen local key features and can be applied as a feature extraction method in time series classification and prediction tasks. At the same time, the gated neural network (Gated Recurrent Unit, GRU) can learn the time series features of electricity data and solve the long-term dependency problem in the RNN network. In addition, it has a simpler structure and faster training speed than the LSTM network. Therefore, the present invention first forms a specific feature extraction rule based on expert experience, extracts a subset of electricity data from the electricity time series data according to a specific time window as a candidate shapelet feature, and then uses KL divergence to calculate the correlation between the collected data sequence and the candidate shapelet feature to obtain the optimal candidate shapelet feature. Subsequently, the optimal shapelet feature data and the basic time series feature data of electricity collection are used as the input of the GRU to construct a loss function based on the output prediction data and the actual sample label and iteratively feedback and adjust the parameters, thereby completing the construction of the electric energy meter current overload prediction model, further improving the reliability and accuracy of the power data acquisition terminal.

[0046] The detailed scheme is described with reference to typical embodiments.

[0047] The implementation platform is based on the electricity consumption information collection system of a power company of State Grid. The data collection content includes basic user information, metering data of each electricity consumption collection terminal device, and corresponding input and output information.

[0048] refer to Figure 1 , is a specific implementation process of the electric energy meter current overload prediction method and system based on rule feature optimization GRU of the present invention, the steps are as follows:

[0049] A. Use the electricity consumption information collection system to screen the current overload-related electricity consumption data characteristics and the current overload sample data collected by various terminals, and summarize the data;

[0050] The electricity consumption information collection system collected more than 200,000 user electricity consumption samples from October 2019 to September 2021 in a certain area, mainly including basic user information, metering indicator data, and related electricity consumption details. These samples were summarized and some of the initial feature data were selected as examples:

[0051]

[0052]

[0053] B. Perform data preprocessing on the aggregated data, mainly including data cleaning, data completion, data integration, and normalization, to obtain electricity consumption data with time series characteristics. Specifically, by sorting out various current overload classification indicators, analyzing historical data and current overload cases, and studying the correlation between different abnormal indicators and current overload cases, through statistical results such as similarity and information entropy, the application dimensions of current overload indicators are screened and sorted out. In addition, combined with the volatility, trend, and duration of the data occurrence time, the correlation and difference between the multi-dimensional data indicators corresponding to current overload and abnormalities such as power overcapacity and voltage anomalies are explored;

[0054] C. For electricity consumption data with time series characteristics (T∈x I×M ), T∈x I×M Represents I electricity consumption data instances of length M, where each electricity consumption data instance t={[x1,F1],[x2,F2],[x3,F3],...,[x p ,F p ]}, x and F represent the electricity consumption sequence and the attribute characteristics of the electricity consumption data in a certain time range, respectively, and p represents the time node of the electricity consumption data collection. Based on the expert judgment experience, rules are formulated and used as the basis for extracting key features from the electricity consumption data sequence. Shapelet sequence features of different time windows can be obtained. Then, based on the KL divergence, further calculations are performed to screen the optimal candidate shapelet.

[0055] C1. Electricity consumption sample data T∈x obtained from the electricity consumption information collection system I×M , divide the sample data into T according to different time windows m ∈x I×M 、T w ∈x I×M 、T d ∈x I×M and T s ∈x I×M , respectively represent the monthly collection sample data set, weekly collection sample data set, daily collection sample data set and minute collection sample data set;

[0056] C2. Based on electricity consumption data in different time windows, rules are formulated using expert judgment experience. For example, for dedicated transformer users, the current values ​​at at least 5 of the 24 collection points in a day are greater than 1.2 times the rated maximum current on the meter nameplate. The abnormal intervals are divided by parameters such as the number of overcurrent points per day, the number of consecutive abnormal days, and the cumulative number of abnormal days in a month. For low-voltage household meters, an early warning is issued if the current at one collection point in a day is greater than 1.2 times the rated maximum current on the meter nameplate. Combined with the line loss in the substation, the degree of overload impact on public transformer customers is checked. Priority is given to analyzing overloaded users with at least 2 abnormal points in a day and a substation line loss rate greater than 5%. Priority is given to overloaded customers with high losses in the substation during the overload period and qualified substations during the non-overload period. Based on this type of expert judgment experience, a set of candidate shapelets that can represent subsequence features in the electricity consumption data sequence can be obtained;

[0057] C3. Figure 3 The current data trends shown are from a low-voltage electricity meter in a residential complex. According to expert experience, for a low-voltage meter, an overload warning is issued if the current at one collection point during the day exceeds 1.2 times the meter's nameplate rated maximum current. This indicates that the current peaked between 9:00 AM and 11:30 AM, indicating a current overload. The actual current peak reached 26A, exceeding the meter's rated current of 20A by more than 1.2 times, resulting in negative readings on the meter. This indicates a current overload, necessitating an overload warning. Therefore, the electronic sequence for this time window can be used to represent the key features of current overload within that time window. Similarly, for a candidate shapelet set, the similarity between each candidate shapelet and each electricity data sequence is calculated based on KL divergence. This effectively extracts key features from the electricity data sequence, allowing for more accurate learning of current trends over time. KL divergence is the relative entropy, which can measure the distance between two random distributions. If the two random distributions are the same, their relative entropy is zero; if the two random distributions are very different, their relative entropy will also increase. Therefore, the relative entropy D(T||S) from the time series T(t) to its subsequence S(t) can be expressed as:

[0058]

[0059] t represents the time series distribution. By measuring and calculating the relative entropy of each time series T(t) to its subsequence S(t), the best candidate shapelet S'(t) is evaluated to represent the key features of the electricity consumption data within the corresponding time range.

[0060] D. Use the shapelet sequence feature S'(t) obtained in step C and the electricity consumption feature set F obtained in step A as inputs to the GRU network, build an electric energy meter current overload prediction model, and output the current overload prediction result;

[0061] D1. Based on the electricity consumption data subsequence shapelet (S m ,S w ,S d ,S s ), and then combined with the corresponding power consumption feature set F(f1,f2,f3,…,f n ), after unified encoding, it is used as the input of the GRU network. At the same time, the model parameters are initialized, such as setting the number of iterative training and dropout ratio. There are two gates in the GRU network: update gate z t and reset gate r t The specific structure is as shown in the attached Figure 4 As shown. The function of the update gate is to control the degree to which the power consumption information state of the previous moment is transferred to the current state. The larger the value of the update gate, the more state information of the previous moment is transferred. The reset gate can control how much information of the previous state is written into the current candidate set. The smaller the reset gate, the less information of the previous state is written. The forward propagation calculation of the GRU network is as follows:

[0062] r t =σ(W r ·[h t-1 ,S t ,F t ],

[0063] z t =σ(W z ·[h t-1 ,S t ,F t ],

[0064]

[0065]

[0066] y t =σ(W o ·h t ).

[0067] Among them, [] indicates that two vectors are connected, and * indicates matrix product. Then, according to different time windows, the corresponding shapelet is input into the corresponding GRU network in parallel. At this time, there are multiple parallel GRU network modules in the model, which use the electricity consumption data collected at different time levels (i.e. shapelet (S m ,S w ,S d ,S s ) and the corresponding electricity usage feature set F) for training.

[0068] D2. Use the back propagation algorithm to train the model and iteratively update the model parameters. From the formula of the forward propagation process, we can see that the parameters to be trained include W r ,W z , W o Etc. Among them,

[0069] W r =W rh +W rs +W rf ,

[0070] W z =W zh +W zs +W zf ,

[0071]

[0072] The output of the model is The objective function uses the square loss function is the true value.

[0073] D3. Substitute the updated model parameters into the objective function L and determine whether the error between the objective function and the true value is less than the set error threshold δ. If so, proceed to step D4, otherwise return to step D2;

[0074] D4. Fix the objective function parameters and the corresponding environment configuration, and output the updated electric energy meter current overload prediction model.

[0075] E. Based on the meter's current overload prediction results, the current overload distribution at each time level is integrated to form a formatted message. This is followed by an on-site verification, primarily to verify the effectiveness of the meter's current overload prediction model and any issues identified. The verification results are fed back to step A, updating the sample data to improve the accuracy of the meter's current overload diagnosis and prediction. If the probability of a current overload anomaly exceeds the threshold δ, the system automatically generates a work order and transmits it to the closed-loop data collection and operation management system. Anomalies below the threshold can be manually identified and a work order generated. Whitelist management can be implemented for special users.

[0076] Based on over 200,000 user electricity usage samples from a certain region collected from the electricity consumption information collection system from October 2019 to September 2021, a certain number of positive and negative samples were randomly selected for current overload identification. The test results were statistically analyzed to evaluate the performance of the model. The specific test sample data and evaluation indicators are shown in the following table:

[0077]

[0078] As can be seen from the table, the electricity meter current overload prediction model based on rule-feature optimized GRU performs well in terms of precision, recall, and F1 score. It can more efficiently predict the current overload of the electricity meter and achieve the goal of accurately managing current overload.

[0079] In addition, a certain number of samples were randomly selected to evaluate the effect of KL divergence on prediction improvement, and the performance of KL divergence was evaluated using Euclidean distance as the evaluation baseline. The performance was evaluated using the F1 score, and the results are shown in the following table:

[0080] Model F1 score Use Euclidean distance 93.46％ Using KL divergence 98.33％

[0081] As shown in the table, shapelet feature extraction based on KL divergence performs well, demonstrating its superior ability to mine features in current overload scenarios. Compared to Euclidean distance, it can more effectively assess the correlation between electricity consumption sequences, thereby mining key features of meter current overloads. This allows for better modeling of changing trends in electricity consumption data and improves the ability to identify and predict meter current overloads.

[0082] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0084] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0086] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting current overload of an electric energy meter, characterized in that: The following steps are involved: Obtaining current overload-related electricity consumption data characteristics and user electricity consumption sample data; Preprocess the data to obtain electricity consumption data with time series characteristics; For electricity consumption data with time series characteristics, we classify it according to different time granularities to obtain multi-level electricity consumption time series data. We then use set rules to extract key features from the electricity consumption data sequence and obtain candidate shapelet sequences for different time windows. Based on shapelet sequence features and electricity consumption characteristics, a gated neural network is used to build and train an electric energy meter current overload prediction model to identify electric energy meter current overload. Based on the current overload identification results of the electric energy meter, the current overload distribution at each time level is integrated to generate a formatted early warning message for sending, on-site verification, and data update. The specific process of classification according to different time granularities includes: dividing sample data according to time windows of different lengths; The specific process of obtaining candidate shapelet sequences for different time windows includes: calculating the similarity between each candidate shapelet feature extraction rule and the electricity consumption data sequence based on KL divergence, and obtaining the best candidate shapelet features through measurement and calculation to represent the key features of the electricity consumption data in the corresponding time range.

2. The method for predicting current overload of an electric energy meter according to claim 1, wherein: Relevant electricity consumption data includes basic user information, metering point information, energy meter information, electricity consumption information and load information.

3. The method for predicting current overload of an electric energy meter according to claim 1, wherein: The preprocessing includes data cleaning, data completion, data integration and normalization.

4. The method for predicting current overload of an electric energy meter according to claim 1, wherein: The electric energy meter current overload prediction model includes an update gate and a reset gate. The update gate is used to control the extent to which the power consumption information state at the previous moment is transferred to the current moment state; the reset gate is used to control how much information of the previous state is written into the current candidate set.

5. The method for predicting current overload of an electric energy meter according to claim 1, wherein: The training process of the electric energy meter current overload prediction model includes: The shapelet features of the electricity consumption data subsequences in different time windows and the initial basic information features of electricity consumption are uniformly encoded and used as the input of the GRU network respectively, and the model parameters are initialized; Use the back-propagation algorithm to train the model and iteratively update the model parameters; Substitute the updated model parameters into the objective function to determine whether the error between the objective function value and the true value is less than the error threshold. If so, proceed to the next step; otherwise, return to iterative update. The objective function parameters are fixed, and the updated electric energy meter current overload prediction model is output.

6. A current overload prediction system for an electric energy meter, characterized in that: include: A module for obtaining current overload-related electricity consumption data characteristics and user electricity consumption sample data; A module for preprocessing data to obtain electricity consumption data with time series characteristics; A module for classifying electricity consumption data with time series characteristics according to different time granularities to obtain multi-level electricity consumption time series data. It then uses set rules to extract key features from the electricity consumption data sequence and obtain candidate shapelet sequences for different time windows. A module for identifying electric energy meter current overload using an electric energy meter current overload prediction model built and trained using a gated neural network based on shapelet sequence features and electricity consumption characteristics; A module used to integrate the current overload distribution of each time layer based on the current overload judgment results of the electric energy meter, generate a formatted early warning message for transmission, conduct on-site verification, and update data; The specific process of classification according to different time granularities includes: dividing sample data according to time windows of different lengths; The specific process of obtaining candidate shapelet sequences for different time windows includes: calculating the similarity between each candidate shapelet feature extraction rule and the electricity consumption data sequence based on KL divergence, and obtaining the best candidate shapelet features through measurement and calculation to represent the key features of the electricity consumption data in the corresponding time range.

7. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the steps of the method according to any one of claims 1 to 5 are completed when the computer instructions are executed by the processor.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 7.

Images