Electricity data sample construction, model training, and account category determination method and device

Abstract

The application discloses a kind of electric data sample construction, model training, account category determination method and device, the electric data sample construction method includes: obtaining electric load time series;With the proportion of the preset value in electric load time series and the corresponding coefficient of variation, generate load curve jitter data;Similarity calculation is carried out based on electric load time series, to determine the first periodic characteristic data;According to the correlation coefficient between sub time series, average calculation is carried out, to obtain the second periodic characteristic data;Electric load time series in specified time period, load curve jitter data, first periodic characteristic data, second periodic characteristic data are combined, to obtain the data sample for training electric account classification model.It can effectively improve the accuracy of electric user classification by constructing the characteristic data related to electric load law characteristics for basic electric load time series data.

Description

Technical Field

[0001] This invention relates to the field of power technology, and in particular to a method and apparatus for constructing electricity data samples, training models, and determining account categories. Background Technology

[0002] Dedicated transformers refer to a mode of power supply that uses a dedicated transformer independently. With the increasing number and scale of dedicated transformer users, accurately classifying them according to different electricity consumption patterns is of great significance for electricity management, load forecasting, and electricity behavior analysis.

[0003] However, due to the diverse electricity consumption patterns of dedicated transformer users, the accuracy of electricity user classification in related technologies needs to be improved. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. To this end, the first objective of this invention is to propose a method for constructing electricity consumption data samples. This method constructs feature data related to the characteristics of electricity consumption patterns by building upon basic electricity load time series data, thereby improving the accuracy of classification model training and ultimately increasing the accuracy of electricity user classification, ensuring that users with similar electricity load patterns are grouped into the same category.

[0005] The second objective of this invention is to propose a model training method.

[0006] The third objective of this invention is to provide a method for determining the category of electricity account.

[0007] The fourth objective of this invention is to provide a method for determining the category of electricity account.

[0008] The fifth objective of this invention is to provide an apparatus for constructing electricity data samples.

[0009] The sixth objective of this invention is to provide a model training device.

[0010] The seventh objective of this invention is to provide an apparatus for determining the category of electricity account.

[0011] The eighth objective of this invention is to provide an apparatus for determining the category of electricity account.

[0012] The ninth objective of this invention is to provide a computer device.

[0013] The tenth object of the present invention is to provide a computer-readable storage medium.

[0014] To achieve the above objectives, one embodiment of the present invention proposes a method for constructing electricity consumption data samples. The method includes: acquiring an electricity load time series; wherein the electricity load time series is divided into several sub-time series; generating load curve jitter data using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; wherein the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; performing similarity calculation based on the electricity load time series to determine first periodic feature data; performing average calculation based on the correlation coefficients between the sub-time series to obtain second periodic feature data; and combining the electricity load time series within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data to obtain data samples for training an electricity account classification model.

[0015] According to one embodiment of the present invention, the second periodic feature data includes annual periodic feature data; the step of averaging the correlation coefficients between the sub-time series to obtain the second periodic feature data includes: determining the electricity consumption trend time series corresponding to the electricity load time series; dividing the electricity consumption trend time series into at least two annual electricity consumption trend time series using years as the time unit; and averaging the correlation coefficients between any two annual electricity consumption trend time series to determine the annual periodic feature data.

[0016] According to one embodiment of the present invention, determining the electricity consumption trend time series corresponding to the electricity load time series includes: averaging the electricity load time series within a specified time period to obtain an average electricity consumption time series; and performing data smoothing filtering on the average electricity consumption time series to obtain the electricity consumption trend time series corresponding to the electricity load time series.

[0017] According to one embodiment of the present invention, the second periodic feature data includes daily periodic feature data; the step of averaging the correlation coefficients between the sub-time series to obtain the second periodic feature data includes: dividing the electricity load time series into daily time units to obtain at least two daily electricity load series; and averaging the correlation coefficients between any two daily electricity load series to obtain the daily periodic feature data.

[0018] According to one embodiment of the present invention, the step of calculating similarity based on the electricity load time series to determine the first periodic feature data includes: dividing the electricity load time series according to a preset time unit to obtain at least two preset time unit load sequences; calculating the similarity between any two preset time unit load sequences using the Hamming distance algorithm; and taking the average of the similarity between the two preset time unit load sequences as the first periodic feature data.

[0019] According to one embodiment of the present invention, the data samples are labeled with electricity account categories; wherein, the electricity account category is any one of a horizontal category, a zero-value regularity category, an annual periodicity category, a daily periodicity category, and a random category; wherein, the load curve of the electricity account belonging to the horizontal category is approximately a straight line; the load curve of the electricity account belonging to the zero-value regularity category has a zero-value ratio exceeding a threshold and the non-zero values ​​exhibit a periodicity; the electricity load curve of the electricity account belonging to the annual periodicity category exhibits an annual periodicity; and the electricity load curve of the electricity account belonging to the daily periodicity category exhibits a daily periodicity.

[0020] One embodiment of the present invention proposes a model training method, the method comprising: acquiring a data sample constructed through any of the foregoing embodiments as a first electricity consumption data sample; inputting the first electricity consumption data sample into an electricity account classification model for prediction to obtain an electricity account prediction category; updating the parameters of the electricity account classification model according to the electricity account prediction category and the electricity account category labeled in the data sample, until the model stops training conditions are met.

[0021] One embodiment of the present invention proposes a method for determining the category of electricity account, the method comprising: acquiring the electricity load data of a target account; inputting the electricity load data into an electricity account classification model trained by the aforementioned model training method to determine the target electricity account category to which the target account belongs.

[0022] One embodiment of the present invention proposes a method for determining the category of an electricity account. The method includes: acquiring the electricity load time series of a target account; determining the load curve jitter data, first periodic feature data, and second periodic feature data of the target account; wherein the load curve jitter data is generated using the proportion of a preset value in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; the first periodic feature data is determined based on the similarity calculation of the electricity load time series of the target account; the second periodic feature data is determined based on the similarity calculation of several sub-time series of the target account; the several sub-time series are obtained by segmenting the electricity load time series of the target account; constructing account electricity load data based on the electricity load time series of the target account within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data; and determining the target electricity account category to which the target account belongs based on the account electricity load data.

[0023] One embodiment of the present invention provides an electricity consumption data sample construction device, the device comprising: a time series acquisition module for acquiring an electricity load time series; wherein the electricity load time series is segmented into several sub-time series; a data generation module for generating load curve jitter data using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; wherein the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; a similarity calculation module for performing similarity calculation based on the electricity load time series to determine first periodic feature data; an averaging calculation module for performing averaging calculation based on the correlation coefficients between the sub-time series to obtain second periodic feature data; and a data combination module for combining the electricity load time series within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data to obtain data samples for training an electricity account classification model.

[0024] One embodiment of the present invention provides a model training apparatus, the apparatus comprising: a data sample acquisition module, configured to acquire a data sample constructed by any of the aforementioned electricity consumption data sample construction methods, as a first electricity consumption data sample; a category prediction module, configured to input the first electricity consumption data sample into an electricity account classification model for prediction, thereby obtaining a predicted electricity account category; and a parameter update module, configured to update the parameters of the electricity account classification model according to the predicted electricity account category and the electricity account category labeled on the data sample, until the model training stop condition is met.

[0025] One embodiment of the present invention provides an electricity account category determination device, the device comprising: a load data acquisition module for acquiring electricity load data of a target account; and a category determination module for inputting the electricity load data of the account into an electricity account classification model trained by the aforementioned model training method to determine the target electricity account category to which the target account belongs.

[0026] One embodiment of the present invention provides an electricity account category determination device, the device comprising: a time series acquisition module for acquiring an electricity load time series of a target account; a data determination module for determining load curve jitter data, first periodic feature data, and second periodic feature data of the target account; wherein the load curve jitter data is generated using the proportion of a preset value in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; the first periodic feature data is determined based on similarity calculation of the electricity load time series of the target account; the second periodic feature data is determined based on similarity calculation of several sub-time series of the target account; the several sub-time series are obtained by segmenting the electricity load time series of the target account; a load data construction module for constructing account electricity load data based on the electricity load time series of the target account within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data; and a category determination module for determining the target electricity account category to which the target user belongs based on the account electricity load data.

[0027] One embodiment of the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the above embodiments.

[0028] One embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the above embodiments.

[0029] According to the various embodiments provided by the present invention, corresponding electricity load pattern feature data can be constructed according to the significant regularity of the user's electricity load curve in different electricity consumption cycles, so as to improve the accuracy of classification model training, thereby improving the accuracy of electricity user classification and ensuring that users with the same or similar electricity load patterns are classified into the same category.

[0030] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating a method for constructing a power consumption data sample according to one embodiment of this specification.

[0032] Figure 2 This is a schematic diagram of the process for calculating second periodic feature data according to one embodiment of this specification.

[0033] Figure 3 This is a flowchart illustrating the process of determining a time series of electricity consumption trends according to one embodiment of this specification.

[0034] Figure 4 This is a schematic diagram of the process for calculating second periodic feature data according to one embodiment of this specification.

[0035] Figure 5 This is a schematic diagram of a process for determining first periodic feature data according to one embodiment of this specification.

[0036] Figure 6 This is a flowchart illustrating a model training method provided according to one embodiment of this specification.

[0037] Figure 7 This is a flowchart illustrating a method for determining electricity account categories according to one embodiment of this specification.

[0038] Figure 8 This is a flowchart illustrating a method for determining electricity account categories according to one embodiment of this specification.

[0039] Figure 9 This is a structural block diagram of an apparatus for constructing an electricity consumption data sample according to one embodiment of this specification.

[0040] Figure 10 This is a structural block diagram of a model training apparatus provided according to one embodiment of this specification.

[0041] Figure 11 This is a structural block diagram of an electricity account category determination device provided according to one embodiment of this specification.

[0042] Figure 12 This is a structural block diagram of an electricity account category determination device provided according to one embodiment of this specification.

[0043] Figure 13a This is a structural block diagram of a computer device provided according to one embodiment of this specification.

[0044] Figure 13b This is a structural block diagram of a computer device provided according to one embodiment of this specification.

[0045] Figure 13c This is a structural block diagram of a computer device provided according to one embodiment of this specification. Detailed Implementation

[0046] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0047] Dedicated transformer users are typically numerous and exhibit diverse electricity consumption patterns, including commercial mixed-use users, office dedicated transformer users, agricultural dedicated transformer users, residential dedicated transformer users, non-general industrial dedicated transformer users, and large industrial dedicated transformer users. With the widespread adoption of smart meters and the development of information technology infrastructure, the electricity management and analysis of dedicated transformer users across various industries has received increasing attention from power companies, governments, and users. Accurately classifying dedicated transformer users based on their electricity load patterns is of paramount importance for electricity management under the development of information technology infrastructure.

[0048] Electricity load curves can reflect the pattern of electricity load changes over time. Electricity load curves typically present relevant electricity consumption information such as the magnitude of a user's electricity load, the time axis corresponding to that load, and the load fluctuation process. Therefore, analyzing and processing electricity load curves can identify different electricity load patterns of users under different electricity consumption modes, thereby enabling user classification.

[0049] In related technologies, classification or clustering algorithms are typically used to analyze and process the electricity load curves of electricity users. Users with similar electricity load curve shapes are grouped into the same category, while users with significantly different electricity load curve shapes are grouped into different categories. This indicates that users in the same category have similar electricity load patterns, while users in different categories have significantly different electricity load patterns.

[0050] In some cases, when classifying or clustering a user's electricity load curve over a period of time, on the one hand, short-term fluctuations may have a significant impact on the final characteristics of the electricity load curve, leading to inaccurate user classification results; on the other hand, it may be impossible to take into account the different characteristics of the electricity load curve in the short and long time periods. For example, a user's electricity load curve may show significant differences in the short term but may show significant similarities in the long term, resulting in inaccurate user classification results.

[0051] To improve the accuracy of classifying electricity users with different electricity load patterns, it is necessary to provide a method for constructing electricity data samples, training models, and determining account categories. This method can construct corresponding electricity load pattern feature data based on the significant regularity of users' electricity load curves in different electricity consumption cycles, thereby improving the accuracy of classification model training and thus improving the accuracy of electricity user classification, ensuring that users with the same or similar electricity load patterns are classified into the same category.

[0052] This specification provides a method for constructing electricity data samples, referencing... Figure 1 As shown, the method may include the following steps.

[0053] S110. Obtain the time series of electricity load.

[0054] The electricity load time series is divided into several sub-time series.

[0055] The electricity load time series refers to actual electricity load data that reflects the fundamental characteristics of the actual electricity load curve of electricity users. Dividing the electricity load time series into several sub-time series involves dividing the electricity load time series according to a certain time period, resulting in several sub-time series with a time axis length equal to that time period. Specifically, the time period used to divide the electricity load time series can be any one of a daily, weekly, semi-monthly, monthly, quarterly, or annual period.

[0056] S120. Generate load curve jitter data by using the proportion of preset values ​​in the power load time series and the coefficient of variation corresponding to the power load time series.

[0057] The coefficient of variation is determined by the mean and standard deviation of the electricity load time series.

[0058] The preset value can be the electricity load value of a user that deviates from normal electricity consumption over a period of time but remains basically unchanged. The coefficient of variation can be used to measure the degree of deviation of the electricity load value in the electricity load time series. By using the proportion of the preset value in the electricity load time series and the coefficient of variation to generate load curve jitter data, it can be used as data on the load curve shape characteristics of the data sample to reflect the relatively stable curve shape characteristics of the electricity load curve corresponding to the electricity load time series.

[0059] Specifically, this section explains how to calculate the coefficient of variation. A time series of electricity load data from a user over a period of time can be obtained. The average (avg) of the electricity load values ​​in this time series can be calculated, along with the standard deviation (std) of the time series. Based on the obtained average and standard deviation, the coefficient of variation (CV) of the electricity load time series can be calculated as: CV = standard deviation (std) / average (avg). Using the preset percentage (default) and the coefficient of variation (CV), the load curve jitter data J = [default, CV] of the electricity load time series can be generated.

[0060] S130. Calculate similarity based on the electricity load time series to determine the characteristic data of the first period.

[0061] Among them, the first periodic feature can be used to indicate that users consume electricity less frequently and that the electricity load curve corresponding to the electricity load time series exhibits a periodic pattern.

[0062] In some cases, the similarity of electricity load time series can be used to measure the consistency of the period and trend of any two electricity load time series. Therefore, by calculating the similarity as a characteristic data of electricity load patterns, we can measure whether the electricity load time series has a corresponding electricity load pattern.

[0063] Specifically, the similarity between several sub-time series of the electricity load time series, with a first period length, can be calculated. The first period length can be determined by analyzing the periodicity of the user's electricity load curve. For example, if a user consumes little or no electricity in most months of the year, but consumes more in a few months, the corresponding user's electricity load curve will show fluctuations in those few months. After analyzing the periodicity of the user's electricity load curve, the similarity between each pair of the user's electricity load time series over several years can be calculated, with the year as the first period length. This similarity serves as the first period characteristic data for the user's electricity load data based on the first period characteristic.

[0064] S140. The second periodic characteristic data is obtained by averaging the correlation coefficients between the sub-time series.

[0065] The second periodicity characteristic can be used to represent the long-term electricity consumption of users, and the characteristic that the electricity load curve corresponding to the electricity load time series exhibits a periodic pattern. It should be noted that the sub-time series is obtained by dividing the electricity load time series according to the duration of the second period, and the electricity load curves corresponding to the sub-time series have similar changing trends between similar phases.

[0066] Specifically, the duration of the second period can be determined by analyzing the periodic patterns of the user's electricity load curve. For example, a user's monthly electricity load may exhibit similar patterns, such as higher electricity consumption in the first half of the month and lower consumption in the second half. Correspondingly, the user's electricity load curve shows a higher peak load in the first half of the month and a lower peak load in the second half. Based on the analysis of the periodic patterns of this user's electricity load curve, the user's electricity load time series can be divided into several sub-time series with a duration of one month, using a month as the second period. The correlation coefficients between each pair of these sub-time series are calculated, resulting in several correlation coefficients. The average of these correlation coefficients is then calculated to obtain the second-period characteristic data of the user's electricity load data based on the second-period characteristics.

[0067] S150. Combine the electricity load time series, load curve jitter data, first period feature data, and second period feature data within the specified time period to obtain data samples for training the electricity account classification model.

[0068] The specified time period can be a pre-set historical electricity consumption period for sampling the user's electricity load time series, and the electricity load time series within the specified time period can be data from the basic feature dimension of the data sample. Specifically, the load curve jitter data obtained by the aforementioned method can be data from the load curve shape feature dimension of the data sample; the first period feature data can be data from the first period feature dimension of the data sample; and the second period feature data can be data from the second period feature dimension of the data sample.

[0069] In some cases, by constructing load curve jitter data, first-cycle feature data, and second-cycle feature data, it is possible to describe the implicit user electricity load patterns in the user electricity load time series. Thus, the electricity account classification model can extract relevant feature data from the constructed data samples and perform analysis, so that the electricity account can be accurately classified according to the electricity load patterns.

[0070] For example, the user's electricity load time series over the past month is sampled to serve as the basic feature data for the data samples. Sampling the user's electricity load time series over the past month with a 14-day period allows the user's electricity load time series from the 1st to the 14th of the month to be extracted as the basic feature data for the first data sample; the electricity load time series from the 2nd to the 15th of the month to be extracted as the basic feature data for the second data sample; the electricity load time series from the 3rd to the 16th of the month to be extracted as the basic feature data for the third data sample; and so on, obtaining at least 15 basic feature data samples for the user. Adding the user's load curve jitter data J, the first period feature data PF1, and the second period feature data PF2 calculated using the aforementioned method to each basic feature data sample yields at least 15 completed data samples for the user, which can then be used to train the electricity account classification model.

[0071] The electricity account classification model can be a time series classification model that categorizes users into corresponding electricity account categories based on their electricity load patterns. In some embodiments, the electricity account classification model can be an LSTM (Long Short Term Memory) time series classification model.

[0072] In the above embodiments, data with multiple feature dimensions related to the curve shape and periodicity of the user's electricity load curve are constructed as a supplement to the basic feature dimension of electricity load time series data. This is used to accurately describe the user's electricity load pattern, thereby improving the training effect of the electricity account classification model.

[0073] In some implementations, the second periodic characteristic data includes annual periodic characteristic data. (See reference...) Figure 2 As shown, the second periodic feature data is obtained by averaging the correlation coefficients between sub-time series, which may include the following steps.

[0074] S210. Determine the electricity consumption trend time series corresponding to the electricity load time series.

[0075] Among them, the annual cycle characteristic represents the periodicity of the annual electricity load curve corresponding to the user's electricity load time series. The electricity trend time series is a time series obtained by eliminating or weakening the short-term fluctuations in the electricity load time series, which can reflect the long-term trend in the electricity load time series.

[0076] In some cases, the periodicity of electricity load can be determined by observing the long-term trend of changes in the electricity load time series. Therefore, the periodic characteristics of the applied electricity load time series can be reflected by electricity trend time series. Specifically, electricity trend time series can be obtained by smoothing the electricity load time series data.

[0077] S220. Divide the electricity consumption trend time series into years to obtain at least two annual electricity consumption trend time series.

[0078] Specifically, the electricity load time series of user m years (m≥2) is obtained, and the electricity load time series is smoothed to obtain the corresponding electricity trend time series for m years. The electricity trend time series for m years is then divided into annual units, resulting in m annual electricity trend time series. It should be noted that, in this embodiment, at least two years of electricity load time series are required to analyze and determine that the user's electricity load pattern has an annual cycle characteristic. Therefore, at least two years of electricity load time series from the user are needed to calculate the annual cycle characteristic data.

[0079] S230. The annual cycle characteristic data is determined by averaging the correlation coefficients between any two annual electricity consumption trend time series.

[0080] Specifically, the correlation coefficient between any two sequences in the m-year electricity consumption trend time series is calculated to measure the similarity of the electricity consumption trend time series over several annual periods, thereby further measuring the similarity of user electricity load over several annual periods. The larger the correlation coefficient, the more consistent the electricity load of the corresponding two annual periods. Based on the calculation, m*(m-1) / 2 correlation coefficients can be obtained. The average value of the correlation coefficients is calculated as the annual periodic characteristic data of the user's electricity load.

[0081] In some embodiments, the Pearson correlation coefficient between any two annual electricity consumption trend time series can be calculated and averaged to determine the annual cycle characteristic data.

[0082] In some implementations, reference Figure 3 As shown, determining the electricity consumption trend time series corresponding to the electricity load time series may include the following steps.

[0083] S211. Calculate the average of the electricity load time series within the specified time period to obtain the average electricity consumption time series.

[0084] Specifically, based on the obtained electricity load time series of the user over m years, the average value of the electricity load recorded in the electricity load time series for each day of m years is taken to obtain the daily average electricity load data for each day. The calculated daily average electricity load data is used as a new element in the user's electricity load time series over m years to obtain the user's average electricity load time series over m years.

[0085] For example, the recording frequency of a user's daily electricity load data can be 15 minutes, meaning that the electricity load time series recorded each day includes 96 electricity load values ​​recorded at 96 recording time points throughout the day. The average value of the 96 electricity load values ​​for each day is then taken to represent the daily average electricity load data for that day.

[0086] S213. Perform data smoothing and filtering on the average electricity consumption time series to obtain the electricity consumption trend time series corresponding to the electricity load time series.

[0087] In some cases, a user's daily electricity load data often varies, resulting in discrepancies in the user's average daily electricity load. This variation leads to short-term fluctuations on a daily basis in the user's annual average electricity load time series. These short-term fluctuations are not significant for analyzing the long-term trend of the user's electricity load over an annual period. Therefore, data smoothing filtering can be applied to the electricity load average time series to eliminate or weaken these short-term fluctuations. Specifically, a filter can be used to smooth the aforementioned annual average electricity load time series, making the long-term trend characteristics of the time series more prominent. For example, the HP (Hodrick-Prescott) filter can be used to smooth the electricity load average time series. The HP filter can divide the electricity load average time series into trend and periodic components, thus obtaining the electricity load trend time series after smoothing the annual average electricity load time series over the annual average electricity load time series.

[0088] In some implementations, the second periodic feature data includes daily periodic feature data. (See reference...) Figure 4 As shown, the second periodic feature data is obtained by averaging the correlation coefficients between sub-time series, which may include the following steps.

[0089] S310. Divide the electricity load time series into days to obtain at least two days of electricity load series.

[0090] Among them, the daily periodicity characteristic represents the periodicity of the daily electricity load curve corresponding to the user's electricity load time series.

[0091] Specifically, the electricity load time series of a user for n (n≥2) days is obtained. This n-day electricity load time series is then divided into daily units, resulting in n daily electricity load sequences. It should be noted that, in this embodiment, at least two days' worth of electricity load time series are required to analyze and determine if the user's electricity load pattern exhibits a daily cycle characteristic. Therefore, at least two days' worth of electricity load time series from the user are needed to calculate the daily cycle characteristic data.

[0092] S320. The daily cycle characteristic data is obtained by averaging the correlation coefficient between any two daily electricity load sequences.

[0093] Specifically, the correlation coefficient between any two sequences in the n-day electricity load series is calculated to measure the similarity of the user's electricity load time series over several daily periods. The larger the correlation coefficient, the more consistent the electricity load is between the corresponding two daily periods. Based on the calculation, n*(n-1) / 2 correlation coefficients can be obtained. The average of these correlation coefficients is then calculated as the daily periodic characteristic data of the user's electricity load.

[0094] In some embodiments, the Pearson correlation coefficient between any two daily electricity load sequences can be calculated and averaged to determine the daily cycle characteristic data.

[0095] In some implementations, reference Figure 5 As shown, the process of calculating similarity based on the electricity load time series to determine the characteristic data of the first period may include the following steps.

[0096] S131. Divide the power load time series according to the preset time unit to obtain at least two preset time unit load series.

[0097] The preset time unit can be determined based on the periodicity of the electricity load curve corresponding to the electricity load time series. Specifically, the preset time unit can typically be any one of a half-month, monthly, quarterly, semi-annual, or annual time unit. For example, if a user consumes more electricity in April and October each year, and no electricity in other months, meaning the electricity load curve exhibits an annual periodicity, the user's electricity load time series can be segmented using the year as the preset time unit. Alternatively, if the user's electricity load curves for April and October each year show a high degree of similarity, meaning the electricity load curve exhibits a semi-annual periodicity, the user's electricity load time series can be segmented using a semi-annual periodicity.

[0098] For example, consider a user whose electricity consumption is highest in April and October each year, and lowest in other months. Furthermore, the user's electricity load curves for April and October show a high degree of similarity in periodicity and trend. We obtain the user's electricity load time series for m years (m≥1). Dividing this m-year time series into semi-annual units, we can divide each year's electricity load time series into two corresponding semi-annual electricity load series, resulting in 2m semi-annual electricity load series.

[0099] S133. Calculate the similarity between any two preset time unit load sequences using the Hamming distance algorithm.

[0100] In some cases, the first periodicity characteristic can be used to indicate that users consume electricity relatively infrequently, and the electricity load curve corresponding to the electricity load time series exhibits a periodic pattern, meaning that the electricity load series contains a large number of zero values. Therefore, it's understandable that the load series obtained by segmenting the electricity load time series according to preset time units also contains a large number of zero values. Thus, the Hamming distance algorithm can be used to calculate the Hamming distance between any two load series within preset time units, which can be used as the similarity between the corresponding load series within preset time units.

[0101] S135. The average similarity between any two preset time unit load sequences is used as the first period feature data.

[0102] For example, consider a user who consumes more electricity in April and October each year, and no electricity in other months, and whose electricity load curves in April and October each year show a high degree of similarity in periodicity and trend. Calculating the similarity between any two semi-annual electricity load sequences from 2m semi-annual electricity load sequences yields m*(2m-1) Hamming distances. The average of these Hamming distances is then used as the first-period feature data for the user's electricity load on the first-period feature.

[0103] In some implementations, the data samples are labeled with electricity account categories.

[0104] Among them, the electricity account category can be any one of the following: horizontal category, zero-value regularity category, annual periodic category, daily periodic category, or random category.

[0105] Among them, the load curve of the electricity account belonging to the horizontal type is approximately a straight line; the load curve of the electricity account belonging to the zero value regularity type has a zero value ratio exceeding the threshold and the non-zero values ​​have a periodic regularity; the electricity load curve of the electricity account belonging to the annual periodicity type has an annual periodic regularity; and the electricity load curve of the electricity account belonging to the daily periodicity type has a daily periodic regularity.

[0106] Among them, the electricity account category serves as the label for each data entry in the data sample, and can be used to verify the data classification effect. In some cases, by analyzing the actual electricity load of dedicated power transformer users, and based on the periodicity and trend characteristics of different electricity load patterns under different electricity consumption modes, electricity account categories can be further summarized to describe different electricity load patterns.

[0107] Specifically, horizontally categorized electricity accounts exhibit a horizontal curve shape, indicating that the user has not used electricity for a relatively long period, or that the electricity load remains essentially constant. For example, cold storage facilities typically require year-round low-temperature operation. Based on the electricity load time series of horizontally categorized electricity accounts, the corresponding load curve fluctuation data can be calculated. In some cases, horizontally categorized electricity accounts may experience periods of no electricity use, affecting the calculation of the coefficient of variation for the corresponding electricity load time series. To improve the accuracy of describing the electricity load patterns of horizontally categorized electricity accounts, the preset value proportion in the load curve fluctuation data can be set to a proportion of 0 values.

[0108] In the electricity load time series of accounts exhibiting a zero-value pattern, zero values ​​constitute a large proportion, and non-zero values ​​show periodicity. The proportion of zero values ​​in the load curve of these accounts exceeds a threshold; specifically, in the electricity load time series of these accounts, at least half of the electricity load values ​​are typically zero. For example, heating boiler rooms that typically only require large amounts of electricity for heating during the winter months. Based on the electricity load time series of these zero-value pattern accounts, the corresponding first-period characteristic data can be calculated.

[0109] Electricity accounts with annual cyclical characteristics exhibit long-term electricity consumption, and the electricity load curves corresponding to the electricity load time series show an annual cyclical pattern. For example, agricultural production users typically generate electricity based on the crop growth cycle. Based on the electricity load time series of annual cyclical electricity accounts, corresponding second-cycle characteristic data can be calculated, where the second-cycle characteristic data is the annual cyclical characteristic data.

[0110] Electricity accounts categorized by daily periodicity exhibit long-term electricity consumption, and the electricity load curves corresponding to the electricity load time series show a daily periodic pattern. For example, residents typically consume less electricity and have lower loads during the day and early morning, while consuming more electricity and having higher loads at night. Based on the electricity load time series of daily periodicity accounts, corresponding second-period characteristic data can be calculated, where the second-period characteristic data is the daily periodic characteristic data.

[0111] It should be noted that if the above four electricity consumption patterns cannot be derived from the electricity load curve corresponding to the electricity load time series, then the electricity account to which this type of electricity load time series belongs can be labeled as a random type.

[0112] In the above implementation, five labeling categories are set based on the electricity load patterns of a large number of users. At the same time, corresponding electricity load pattern feature data can be constructed for electricity accounts labeled in different categories, which is used to classify users according to their electricity load patterns to improve the accuracy of classification.

[0113] This specification provides a model training method, which can be referenced. Figure 6 As shown, the method may include the following steps.

[0114] S410. Obtain the data sample constructed by the aforementioned electricity data sample construction method as the first electricity data sample.

[0115] The first electricity consumption data sample is used to train the electricity account classification model.

[0116] Specifically, through the aforementioned method for constructing electricity consumption data samples, several data samples containing basic user characteristic data, load curve fluctuation data, first-cycle characteristic data, and second-cycle characteristic data can be obtained, which can be used as the first electricity consumption data sample for training the electricity account classification model.

[0117] In some implementations, the data samples constructed by the aforementioned electricity data sample construction method include labeled electricity account categories.

[0118] S420. Input the first electricity consumption data sample into the electricity account classification model for prediction to obtain the predicted electricity account category.

[0119] Here, "prediction" represents the classification algorithm used by the electricity account classification model to classify the first electricity data sample, so as to obtain the corresponding classification result and output it; the predicted category of electricity account is the classification result output by the electricity account classification model based on the first electricity data sample.

[0120] S430. Update the parameters of the electricity account classification model according to the predicted electricity account category and the electricity account category labeled in the data sample until the model stops training.

[0121] Specifically, the parameters of the electricity account classification model can be trained and adjusted based on the model's evaluation criteria, such as the accuracy and loss value of the loss function, according to the predicted electricity account category and the electricity account category labeled in the data samples, until the accuracy and loss value of the finally trained electricity account classification model meet the conditions.

[0122] For example, 60% of the data samples in the first electricity consumption data sample can be selected as training samples to train and adjust the parameters of the electricity account classification model; 20% of the data samples can be used as validation samples to verify the model accuracy and adjust the model hyperparameters; and the remaining 20% ​​of the data samples can be used as test samples to verify the model's generalization ability. The data from the training samples is input into the electricity account classification model, and the model parameters are trained and adjusted according to the loss function, accuracy, etc., to obtain multiple trained electricity account classification models. The obtained multiple electricity account classification models are used to classify the validation samples, and the model accuracy is recorded. The parameters corresponding to the best-performing classification model are selected as hyperparameters to optimize the electricity account classification model, thus obtaining the optimal classification model. The optimal classification model is then tested using test samples to evaluate its performance and predictive ability.

[0123] This specification provides a method for determining electricity account categories, referencing... Figure 7 As shown, the method may include the following steps.

[0124] S510, Obtain the electricity load data of the target account.

[0125] It should be noted that the data input during the classification and identification of the target account using the electricity account classification model has the same data sample structure as the data input during the training of the electricity account classification model. For a description of obtaining the target account's electricity load data in the above implementation method, please refer to the description of electricity data sample construction in this specification; details will not be repeated here.

[0126] S520. Input the account electricity load data into the electricity account classification model trained by the aforementioned model training method to determine the target electricity account category to which the target account belongs.

[0127] Specifically, the electricity load data of the target account is used as input data and fed into the trained electricity account classification model. The electricity account classification model outputs the electricity account category identified by the classification algorithm, which is taken as the target electricity account category to which the target account belongs.

[0128] This specification provides a method for determining electricity account categories, referencing... Figure 8 As shown, the method may include the following steps.

[0129] S610. Obtain the time series of electricity load for the target account.

[0130] S620. Determine the load curve jitter data, first cycle characteristic data, and second cycle characteristic data of the target account.

[0131] The load curve jitter data is generated using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series. The coefficient of variation is determined by the mean and standard deviation of the electricity load time series. The first period feature data is determined by similarity calculation based on the electricity load time series of the target account. The second period feature data is determined by similarity calculation based on several sub-time series of the target account. The several sub-time series are obtained by segmenting the electricity load time series of the target account.

[0132] S630. Construct account electricity load data based on the target account's electricity load time series, load curve jitter data, first period characteristic data, and second period characteristic data within a specified time period.

[0133] S640. Determine the target electricity account category to which the target account belongs based on the account's electricity load data.

[0134] It should be noted that for the description of constructing account electricity load data and determining the target electricity account category to which the target account belongs in the above embodiments, please refer to the description of the electricity data sample construction method and the electricity account category determination method in this specification. The details will not be repeated here.

[0135] This specification provides an apparatus for constructing electricity data samples, with reference to... Figure 9 As shown, the electricity consumption data sample construction device 700 includes: a time series acquisition module 710, a data generation module 720, a similarity calculation module 730, an average calculation module 740, and a data combination module 750.

[0136] The time series acquisition module 710 is used to acquire the electricity load time series; wherein the electricity load time series is divided into several sub-time series.

[0137] The data generation module 720 is used to generate load curve jitter data by using the proportion of preset values ​​in the power load time series and the coefficient of variation corresponding to the power load time series; wherein, the coefficient of variation is determined by the mean and standard deviation of the power load time series.

[0138] The similarity calculation module 730 is used to perform similarity calculation based on the electricity load time series to determine the feature data of the first period.

[0139] The average calculation module 740 is used to perform average calculations based on the correlation coefficients between sub-time series to obtain the second period feature data.

[0140] The data combination module 750 is used to combine the electricity load time series, load curve jitter data, first period feature data, and second period feature data within a specified time period to obtain data samples for training the electricity account classification model.

[0141] Specific limitations regarding the electricity data sample construction device can be found in the limitations of the electricity data sample construction method described above, and will not be repeated here. Each module in the aforementioned electricity data sample construction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0142] This specification provides a model training device, with reference to... Figure 10 As shown, the model training device 800 includes: a data sample acquisition module 810, a category prediction module 820, and a parameter update module 830.

[0143] The data sample acquisition module 810 is used to acquire the data sample constructed by the aforementioned electricity data sample construction method as the first electricity data sample.

[0144] The category prediction module 820 is used to input the first electricity consumption data sample into the electricity account classification model for prediction, and obtain the predicted category of the electricity account.

[0145] The parameter update module 830 is used to update the parameters of the electricity account classification model according to the predicted electricity account category and the electricity account category labeled in the data sample, until the model stops training.

[0146] Specific limitations regarding the model training device can be found in the limitations of the model training method described above, and will not be repeated here. Each module in the aforementioned model training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0147] This specification provides an electricity account category determination device, referencing... Figure 11 As shown, the electricity account category determination device 900 includes: a load data acquisition module 910 and a category determination module 920.

[0148] The load data acquisition module 910 is used to acquire the account electricity load data of the target account.

[0149] The category determination module 920 is used to input the account electricity load data into the electricity account classification model trained by the aforementioned model training method to determine the target electricity account category to which the target account belongs.

[0150] Specific limitations regarding the device for determining electricity account categories can be found in the limitations of the electricity account category determination method described above, and will not be repeated here. Each module in the aforementioned electricity account category determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0151] This specification provides an electricity account category determination device, referencing... Figure 12 As shown, the electricity account category determination device 1000 includes: a time series acquisition module 1010, a data determination module 1020, a load data construction module 1030, and a category determination module 1040.

[0152] The time series acquisition module 1010 is used to acquire the electricity load time series of the target account.

[0153] The data determination module 1020 is used to determine the load curve jitter data, first-cycle feature data, and second-cycle feature data of the target account. The load curve jitter data is generated using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series. The coefficient of variation is determined by the mean and standard deviation of the electricity load time series. The first-cycle feature data is determined based on similarity calculations of the target account's electricity load time series. The second-cycle feature data is determined based on similarity calculations of several sub-time series of the target account. These sub-time series are obtained by segmenting the target account's electricity load time series.

[0154] The load data construction module 1030 is used to construct the account's electricity load data based on the target account's electricity load time series, load curve jitter data, first cycle characteristic data, and second cycle characteristic data within a specified time period.

[0155] The category determination module 1040 is used to determine the target electricity account category to which the target account belongs based on the account electricity load data.

[0156] Specific limitations regarding the device for determining electricity account categories can be found in the limitations of the electricity account category determination method described above, and will not be repeated here. Each module in the aforementioned electricity account category determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0157] This specification also provides a computer device, see embodiments thereof. Figure 13a As shown, the computer device 1100 includes a memory 1110, a processor 1120, and an electricity data sample construction method program 1130 stored in the memory 1110 and executable on the processor 1120. When the processor 1120 executes the electricity data sample construction method program 1130, it implements the aforementioned electricity data sample construction method.

[0158] This specification also provides a computer device, see embodiments thereof. Figure 13b As shown, the computer device 1200 includes a memory 1210, a processor 1220, and a model training method program 1230 stored in the memory 1210 and executable on the processor 1220. When the processor 1220 executes the model training method program 1230, it implements the aforementioned model training method.

[0159] This specification also provides a computer device, see embodiments thereof. Figure 13c As shown, the computer device 1300 includes a memory 1310, a processor 1320, and an electricity account category determination method program 1330 stored in the memory 1310 and executable on the processor 1320. When the processor 1320 executes the electricity account category determination method program 1330, it implements the aforementioned electricity account category determination method.

[0160] This specification also provides a computer-readable storage medium storing a method program for constructing electricity data samples, which, when executed by a processor, implements the aforementioned method for constructing electricity data samples.

[0161] This specification also provides a computer-readable storage medium storing a model training method program thereon, which, when executed by a processor, implements the aforementioned model training method.

[0162] This specification also provides a computer-readable storage medium storing a method program for determining an electricity account category, which, when executed by a processor, implements the aforementioned method for determining an electricity account category.

[0163] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0164] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0165] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0166] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0167] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0168] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for constructing electricity consumption data samples, characterized in that, The method includes: Obtain the electricity load time series; wherein the electricity load time series is divided into several sub-time series; Load curve jitter data is generated using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; wherein, the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; Similarity calculations are performed based on the electricity load time series to determine the first period's characteristic data; The second periodic feature data is obtained by averaging the correlation coefficients between the sub-time series. The electricity load time series within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data are combined to obtain data samples for training the electricity account classification model. The step of calculating similarity based on the electricity load time series to determine the first period feature data includes: The electricity load time sequence is divided according to a preset time unit to obtain at least two preset time unit load sequences; The similarity between any two load sequences with a preset time unit is calculated using the Hamming distance algorithm; The average similarity between any two preset time unit load sequences is used as the first periodic feature data.

2. The method according to claim 1, characterized in that, The second periodic feature data includes annual periodic feature data; the step of averaging the correlation coefficients between the sub-time series to obtain the second periodic feature data includes: Determine the electricity consumption trend time series corresponding to the electricity load time series; The electricity consumption trend time series is divided into at least two annual electricity consumption trend time series by using years as the time unit; The annual cycle characteristic data are determined by averaging the correlation coefficients between any two annual electricity consumption trend time series.

3. The method according to claim 2, characterized in that, Determining the electricity consumption trend time series corresponding to the electricity load time series includes: The average time series of electricity load over a specified period is calculated to obtain the average electricity consumption time series. The average electricity consumption time series is smoothed and filtered to obtain the electricity consumption trend time series corresponding to the electricity load time series.

4. The method according to claim 1, characterized in that, The second periodic feature data includes daily periodic feature data; the step of averaging the correlation coefficients between the sub-time series to obtain the second periodic feature data includes: The electricity load time series is divided into segments with days as the time unit to obtain at least two-day electricity load sequences; The daily cycle characteristic data are obtained by averaging the correlation coefficients between any two of the said daily electricity load sequences.

5. The method according to any one of claims 1 to 4, characterized in that, The data samples are labeled with electricity account categories; wherein, the electricity account category is any one of the following: horizontal category, zero-value regularity category, annual periodic category, daily periodic category, and random category; Among them, the load curve of the electricity account belonging to the horizontal type is approximately a straight line; The load curves of electricity accounts belonging to the aforementioned zero-value pattern category show a zero-value ratio exceeding a threshold and non-zero values ​​exhibiting a periodic pattern. The electricity load curves of electricity accounts belonging to the aforementioned annual periodic category exhibit an annual periodic pattern. The electricity load curves of electricity accounts belonging to the aforementioned daily periodicity category exhibit a daily periodicity pattern.

6. A model training method, characterized in that, The method includes: Obtain a data sample constructed by any one of claims 1 to 5 as a first electricity consumption data sample; The first electricity consumption data sample is input into the electricity account classification model for prediction to obtain the predicted electricity account category. The parameters of the electricity account classification model are updated based on the predicted electricity account category and the electricity account category labeled in the data sample until the model stops training.

7. A method for determining the category of electricity account, characterized in that, The method includes: Obtain the electricity load data of the target account; The electricity load data of the account is input into the electricity account classification model trained by the method in claim 6 to determine the target electricity account category to which the target account belongs.

8. A method for determining the category of electricity account, characterized in that, The method includes: Obtain the time series of electricity load for the target account; The load curve jitter data, first period feature data, and second period feature data of the target account are determined. The load curve jitter data is generated using the proportion of a preset value in the electricity load time series and the coefficient of variation corresponding to the electricity load time series. The coefficient of variation is determined by the mean and standard deviation of the electricity load time series. The first period feature data is determined based on similarity calculations of the electricity load time series of the target account. The second period feature data is determined based on similarity calculations of several sub-time series of the target account. The several sub-time series are obtained by segmenting the electricity load time series of the target account. The account's electricity load data is constructed based on the target account's electricity load time series within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data. The target electricity account category to which the target account belongs is determined based on the account's electricity load data; The step of determining the first periodic feature data based on the similarity calculation of the electricity load time series includes: The electricity load time series of the target account is segmented according to a preset time unit to obtain at least two preset time unit load sequences; the similarity between any two preset time unit load sequences is calculated using the Hamming distance algorithm; and the average value of the similarity between any two preset time unit load sequences is used as the first periodic feature data.

9. An apparatus for constructing electricity data samples, characterized in that, The device includes: A time series acquisition module is used to acquire an electricity load time series; wherein the electricity load time series is divided into several sub-time series; The data generation module is used to generate load curve jitter data using the proportion of preset values ​​in the electricity load time series and the coefficient of variation corresponding to the electricity load time series; wherein, the coefficient of variation is determined by the mean and standard deviation of the electricity load time series; The similarity calculation module is used to perform similarity calculation based on the electricity load time series to determine the first period feature data; The average calculation module is used to perform average calculation based on the correlation coefficients between the sub-time series to obtain the second periodic feature data. The data combination module is used to combine the electricity load time series, the load curve jitter data, the first periodic feature data, and the second periodic feature data within a specified time period to obtain data samples for training the electricity account classification model. The similarity calculation module is specifically used to: divide the electricity load time series according to a preset time unit to obtain at least two preset time unit load sequences; calculate the similarity between any two preset time unit load sequences using the Hamming distance algorithm; and use the average of the similarities between the two preset time unit load sequences as the first periodic feature data.

10. A model training device, characterized in that, The device includes: A data sample acquisition module is used to acquire a data sample constructed by any one of the methods of claims 1 to 6, as a first electricity consumption data sample; The category prediction module is used to input the first electricity consumption data sample into the electricity account classification model for prediction, and obtain the predicted category of the electricity account. The parameter update module is used to update the parameters of the electricity account classification model according to the predicted electricity account category and the electricity account category labeled in the data sample, until the model stops training.

11. A device for determining the category of electricity account, characterized in that, The device includes: The load data acquisition module is used to acquire the electricity load data of the target account. The category determination module is used to input the account electricity load data into the electricity account classification model trained by the method in claim 7, and determine the target electricity account category to which the target account belongs.

12. A device for determining the category of electricity account, characterized in that, The device includes: The time series acquisition module is used to acquire the electricity load time series of the target account; The data determination module is used to determine the load curve jitter data, first period feature data, and second period feature data of the target account. The load curve jitter data is generated using the proportion of a preset value in the electricity load time series and the coefficient of variation corresponding to the electricity load time series. The coefficient of variation is determined by the mean and standard deviation of the electricity load time series. The first period feature data is determined based on a similarity calculation of the electricity load time series of the target account. The second period feature data is determined based on a similarity calculation of several sub-time series of the target account. The several sub-time series are obtained by segmenting the electricity load time series of the target account. The load data construction module is used to construct the account's electricity load data based on the target account's electricity load time series within a specified time period, the load curve jitter data, the first periodic feature data, and the second periodic feature data. The category determination module is used to determine the target electricity account category to which the target account belongs based on the account electricity load data. The data determination module is specifically used to: segment the electricity load time series of the target account according to a preset time unit to obtain at least two preset time unit load sequences; calculate the similarity between any two preset time unit load sequences using the Hamming distance algorithm; and use the average of the similarity between the two preset time unit load sequences as the first periodic feature data.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images