A load curve-based unsupervised classification method for power dedicated metering users

Abstract

This application discloses an unsupervised classification method for dedicated power transformer users based on load curves. The method includes: acquiring user load curves; pre-classifying users based on their load curves to determine user types, including horizontal, zero-value regular, annual-cycle, daily-cycle, and random types; and performing secondary classification on annual-cycle and daily-cycle users respectively to obtain the target classification result. This application classifies dedicated power transformer users based on the characteristics of their load curves. This classification method is more suited to the electricity consumption habits of dedicated power transformer users, is highly practical, has good classification results, and can achieve accurate classification.

Description

Technical Field

[0001] This application relates to the field of power user classification technology, specifically to an unsupervised classification method for dedicated power transformer users based on load curves. Background Technology

[0002] In classifying or segmenting dedicated power transformer users, supervised classification methods are generally used. However, in practical applications, labeled data for power users is often lacking, making it impossible to directly apply supervised classification methods to classify dedicated power transformer users. Furthermore, user segmentation tasks typically involve clustering based on users' electricity consumption patterns, characteristics, and industry features. However, the segmented users often exhibit significant differences, such as inconsistent electricity consumption patterns and varying electricity consumption characteristics. In real-world business scenarios, dedicated power transformer users within the same industry segment may have significantly different electricity consumption patterns and characteristics, failing to achieve the true purpose of classification or user segmentation. Manually labeling users is also challenging due to the varying characteristics of dedicated power transformer users, making accurate user type identification difficult. Therefore, supervised methods are difficult to implement.

[0003] Existing unsupervised classification methods for dedicated power transformer users mainly rely on clustering algorithms or user classification based on business operations. However, in practical applications, various abnormal and special dedicated power transformer users exist, directly leading to poor clustering results and a lack of interpretability. When used for downstream tasks, this affects the performance of those tasks, such as load forecasting. Poor clustering results directly impact load forecast accuracy. These techniques or methods ignore special or abnormal dedicated power transformer users, such as users with constant load, users with long-term zero load, or users with random load. These special and abnormal users directly affect the final classification results. Clustering all dedicated power transformer users leads to poor classification results and a lack of interpretability from a business perspective.

[0004] Therefore, the existing classification methods for dedicated power transformer users have problems such as low practicality, low fault tolerance, and poor classification effect. Summary of the Invention

[0005] The purpose of this application is to provide an unsupervised classification method for dedicated power transformer users based on load curves, in order to solve the problems of low practicality, low fault tolerance and poor classification effect of the existing classification methods for dedicated power transformer users.

[0006] To achieve the above objectives, the first aspect of this application provides an unsupervised classification method for dedicated power transformer users based on load curves, the method comprising:

[0007] Obtain user load curves;

[0008] Users are pre-classified based on their load curves to determine their user types, which include horizontal, zero-value regular, annual cycle, daily cycle, and random types.

[0009] Users with annual cycle patterns and users with daily cycle patterns are classified in two ways to obtain the target classification results.

[0010] In this embodiment of the application, obtaining the user load curve includes:

[0011] The system acquires users' power load data according to a preset cycle to form user load curves.

[0012] In this embodiment of the application, pre-classifying users based on user load curves to determine user types includes:

[0013] Determine whether the user type is horizontal based on the user load curve;

[0014] If the user type is not horizontal, determine whether the user type is a zero-value regularity type;

[0015] If the user type is not a zero-value regular type, determine whether the user type is an annual cycle type;

[0016] If the user type is not an annual cycle type, determine whether the user type is a daily cycle type.

[0017] If the user type is not a daily cycle type, the user type is determined to be random.

[0018] In this embodiment of the application, determining whether a user type is horizontal based on the user load curve includes at least one of the following:

[0019] Determine the mode of the user load curve. If the proportion of the mode is not lower than the first threshold, determine that the user type is horizontal; or

[0020] Determine the coefficient of variation of the user load curve. If the coefficient of variation is not higher than the second threshold, the user type is determined to be horizontal.

[0021] In this embodiment of the application, determining whether the user type is a zero-value regularity type includes:

[0022] Determine the initial two-dimensional matrix based on the user load curve;

[0023] Based on Hamming distance, the similarity matrix is ​​obtained from the initial two-dimensional matrix;

[0024] Determine the mean of the similarity matrix;

[0025] If the mean of the similarity matrix is ​​greater than the third threshold, the user type is determined to be a zero-value regularity type.

[0026] In this embodiment of the application, determining whether the user type is an annual cycle type includes:

[0027] Determine the initial two-dimensional matrix based on the user load curve;

[0028] The initial two-dimensional matrix is ​​filtered to obtain the first correlation coefficient matrix;

[0029] Determine the mean of the first correlation coefficient matrix;

[0030] Determine whether the user load curve exhibits periodicity based on the mean of the first correlation coefficient matrix;

[0031] If the user load curve exhibits a periodic pattern, the user type is determined to be annual periodic.

[0032] In this embodiment of the application, determining whether the user type is a daily cycle type includes:

[0033] Filter out the load sequence from the user load curve that is most recent first preset number of days;

[0034] The second correlation coefficient matrix is ​​determined based on the load sequence;

[0035] Determine the mean of the second correlation coefficient matrix;

[0036] The mean of the second correlation coefficient matrix is ​​used to determine whether the user load curve has a periodicity.

[0037] If the user load curve shows periodicity, the user type is determined to be daily periodic.

[0038] In this embodiment of the application, users with annual cycle types and users with daily cycle types are classified in two ways to obtain the target classification results, including:

[0039] Abnormal user load curves were identified using density clustering algorithms;

[0040] Remove users whose load curves correspond to abnormal users to obtain the target users;

[0041] Time series clustering is performed on the user load curves of the target users to obtain the target classification results.

[0042] The second aspect of this application provides an unsupervised classification device for dedicated power transformer users based on load curves, characterized in that it includes:

[0043] The memory is configured to store instructions; and

[0044] The processor is configured to retrieve instructions from memory and, when executing instructions, to implement the aforementioned unsupervised classification method for dedicated power transformer users based on load curves.

[0045] A third aspect of this application provides a machine-readable storage medium, characterized in that the machine-readable storage medium stores instructions for causing a machine to execute the above-described unsupervised classification method for dedicated power transformer users based on load curves.

[0046] The above technical solution obtains user load curves and pre-classifies users based on these curves to determine user types. User types include horizontal, zero-value regular, annual cycle, daily cycle, and random types. Annual cycle users and daily cycle users are then further classified to obtain the target classification result. This application achieves classification based on the characteristics of the user load curves themselves, which better aligns with the electricity consumption habits of users, enabling accurate classification with good results and high practicality.

[0047] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0048] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:

[0049] Figure 1 A flowchart illustrating an unsupervised classification method for dedicated power transformer users based on load curves, provided in this application embodiment;

[0050] Figure 2 A schematic diagram illustrating a pre-classification process for dedicated power transformer users provided in this application embodiment;

[0051] Figure 3 This is a structural block diagram of an unsupervised classification device for dedicated power transformer users based on load curves, provided in an embodiment of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0053] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0054] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0055] In existing unsupervised classification tasks for dedicated power transformer users, classification is achieved solely through clustering. However, these techniques or methods neglect special or anomalous dedicated power transformer users, such as users with constant load, users with long-term zero load, or users with random load. These special and anomalous users directly affect the final classification results. Clustering all dedicated power transformer users leads to poor classification performance and lacks interpretability from an operational perspective. Therefore, the identification and classification of special dedicated power transformer users is particularly important.

[0056] The classification of dedicated power transformer users can use upstream tasks and the classification results of dedicated power transformer users as inputs for downstream tasks or as the basis for model strategies, such as load forecasting tasks. Generally, load forecasting tasks are based on independent modeling of dedicated power transformer users. However, when dealing with a large number of dedicated power transformer users, independently modeling each user leads to a significant increase in model training time, hardware cost, and inference time cost. Therefore, it is necessary to reduce the number of models while ensuring prediction accuracy. Classifying dedicated power transformer users can achieve this goal by grouping users with similar load curves into the same category, thereby constructing downstream tasks for that category and reducing the number of models. Therefore, this application provides an unsupervised classification method for dedicated power transformer users based on load curves.

[0057] Figure 1 This is a flowchart illustrating an unsupervised classification method for dedicated power transformer users based on load curves, provided as an embodiment of this application. Figure 1 As shown in the figure, this application provides an unsupervised classification method for dedicated power transformer users based on load curves. The method may include the following steps.

[0058] Step 101: Obtain the user load curve.

[0059] In this embodiment, the user load curve is the load curve of the user's electricity consumption. The load curve reflects the pattern of load change over time and serves as the basis for power system dispatching and planning. It is categorized by load type into active power load curves and reactive power load curves; and by duration into daily load curves and annual load curves. In this embodiment, unsupervised classification of dedicated power transformer users can be performed based on the user load curve. First, the user load curve is obtained, and then users are classified according to the curve. Generally, the user load curve data resembles a time series. In one example, the data collection for the user load curve can be done by acquiring power load data according to a preset period, thereby forming the user load curve. Classifying load users through the user load curve aligns with the electricity consumption habits of load users and facilitates accurate classification.

[0060] Step 102: Pre-classify users based on user load curves to determine user types. User types include horizontal, zero-value regular, annual cycle, daily cycle, and random types.

[0061] In this application embodiment, based on multiple experimental statistical analyses, users can be classified into five types according to their load curves: horizontal, zero-value regular, annual cycle, daily cycle, and random. Horizontal users are those whose load curves show essentially the same load at any given time, remaining largely unchanged. Zero-value regular users are those whose load curves show load only for a limited period within a year, with loads being very similar; for example, agricultural users only have load values ​​during certain solar terms. Annual cycle users have load curves with an annual cycle, showing very similar load trends on the same dates each year. Daily cycle users have load curves with a certain periodicity on a daily or weekly basis, such as general industrial and commercial users. Random users are those that do not fall into any of the horizontal, zero-value regular, annual cycle, or daily cycle categories.

[0062] In one example, electricity load users can be pre-classified based on their load curves, sequentially determining whether their load curves meet the criteria for horizontal, zero-value regular, annual cycle, daily cycle, or random types. Furthermore, based on the pre-classification in the first stage, all electricity load users can be divided into five categories. After multiple experimental statistical analyses, this classification method has proven to be highly practical and realistic.

[0063] Step 103: Perform secondary classification on users with annual cycle type and users with daily cycle type respectively to obtain the target classification result.

[0064] In this embodiment, after classifying all user load curves into five categories, inconsistencies in cyclical trends or fluctuations still exist among users with annual and daily cyclical loads. Consequently, user load curves within the same annual or daily cyclical category may exhibit significant differences, making load forecasting impossible using a single model. Therefore, a second-stage classification of annual and daily cyclical load users is necessary to achieve secondary classification and determine the final classification result, i.e., the target classification result.

[0065] In one example, by performing secondary classification on users with annual and daily cycles, load curves of users with different trends and cycles can be clustered, grouping load curves of the same shape into the same cluster, thus classifying load curves with the same period and fluctuations into the same category. In another example, after performing secondary classification on users with annual and daily cycles, the final classification result is obtained. This final classification result can then be used to build a load forecasting model for users of the same category, facilitating subsequent classification of users and improving classification efficiency.

[0066] In this embodiment, firstly, classification is achieved based on the characteristics of the load curves of dedicated power transformer users. Compared with existing methods, this approach better reflects the electricity consumption habits of load users and can accurately group load curves with similar electricity consumption behaviors and fluctuations / trends together. Secondly, this embodiment classifies load users in an unsupervised manner, solving the problem of insufficient labeled data in classification tasks and making it easier to apply in real-world scenarios. Furthermore, when facing load prediction problems with a large number of users, unsupervised classification eliminates the need for independent modeling of different user types or for designing models using the classification results as input. This significantly reduces the number of models, lowers training and inference costs, and improves efficiency.

[0067] The above technical solution obtains user load curves and pre-classifies users based on these curves to determine user types. User types include horizontal, zero-value regular, annual cycle, daily cycle, and random types. Annual cycle users and daily cycle users are then further classified to obtain the target classification result. This application achieves classification based on the characteristics of the user load curves themselves, which better aligns with the electricity consumption habits of users, enabling accurate classification with good results and high practicality.

[0068] In this embodiment of the application, obtaining the user load curve may include:

[0069] The system acquires users' power load data according to a preset cycle to form user load curves.

[0070] Specifically, the user load curve is the load curve of a user's electricity consumption. The load curve reflects the pattern of load change over time and serves as the basis for power system dispatching and planning. Based on load type, it is divided into active power load curves and reactive power load curves; based on time duration, it is divided into daily load curves and annual load curves. In one example, the user load curve data for a typical user resembles a time series, and data collection is performed at fixed time intervals, such as once every 15 minutes, once every hour, etc. The specific collection method depends on the actual situation. In another example, the daily data volume is represented by T, then the original user load curve can be represented as: X T ={x1,x2,...,x T}

[0071] Figure 2 This is a schematic diagram illustrating a pre-classification process for dedicated power transformer users, provided as an embodiment of this application. Figure 2 As shown in the embodiment of this application, step 102, pre-classifying users according to user load curves to determine user types, may include:

[0072] Step 201: Determine whether the user type is horizontal based on the user load curve;

[0073] Step 202: If the user type is not horizontal, determine whether the user type is a zero-value regularity type;

[0074] Step 203: If the user type is not a zero-value regularity type, determine whether the user type is an annual cycle type;

[0075] Step 204: If the user type is not an annual cycle type, determine whether the user type is a daily cycle type;

[0076] Step 205: If the user type is not a daily cycle type, determine that the user type is random.

[0077] In this embodiment of the application, the electricity load users are pre-classified in the first stage according to the user load curve, thereby dividing all user load curves into five categories: horizontal type, zero value regular type, annual cycle type, daily cycle type and random type.

[0078] In one example, after multiple experimental statistical analyses, the user type of an electricity load user can be determined according to a set order based on the user load curve. Specifically, the first step is to determine whether the user load curve meets the criteria for a horizontal type. If it does, the user type corresponding to that load curve is determined to be horizontal; otherwise, proceed to the second step. The second step is to determine whether the user load curve meets the criteria for a zero-value regular type. If it does, the user type corresponding to that load curve is determined to be zero-value regular; otherwise, proceed to the third step. The third step is to determine whether the user load curve meets the criteria for an annual cycle type. If it does, the user type corresponding to that load curve is determined to be annual cycle; otherwise, proceed to the fourth step. The fourth step is to determine whether the user load curve meets the criteria for a daily cycle type. If it does, the user type corresponding to that load curve is determined to be daily cycle; otherwise, the user type corresponding to that load curve is determined to be random. These four steps are performed on each user's load curve until all user load curves are classified into the five categories. After multiple experiments and statistical analyses, this classification method is highly practical and closely aligned with reality.

[0079] In this embodiment of the application, determining whether a user type is horizontal based on the user load curve may include at least one of the following:

[0080] Determine the mode of the user load curve. If the proportion of the mode is not lower than the first threshold, determine that the user type is horizontal; or

[0081] Determine the coefficient of variation of the user load curve. If the coefficient of variation is not higher than the second threshold, the user type is determined to be horizontal.

[0082] In this embodiment, a horizontal user load curve refers to a user type where the load at any given time point is essentially the same and remains largely unchanged. The first threshold is the minimum proportion of the mode in the data of a horizontal user load curve when the user type is horizontal. The second threshold is the maximum coefficient of variation of a horizontal user load curve when the user type is horizontal.

[0083] In a specific embodiment of this application, the first user load curve can be represented as X. T ={x1,x2,...,x T}, where T is the amount of data in the user load curve.

[0084] Based on user load curve X TThe first phase involves user type segmentation. First, determine if a user is a horizontally typed user. Horizontally typed users have a relatively constant load at any given time. There are two methods for determining horizontally typed users.

[0085] The specific judgment steps for Method 1 are as follows:

[0086] Calculate the user load curve X T The proportion of the median, φ, if φ≥90%, indicates a level user. Here, 90% is the first threshold. The value of the first threshold can be determined according to the actual situation. Based on experimental statistics, the first threshold is set to 90% here.

[0087] The specific judgment steps for Method 2 are as follows:

[0088] Calculate the user's load curve X T The coefficient of variation cv satisfies formula (1):

[0089]

[0090] Where cv is the coefficient of variation, T is the amount of data in the user's load curve, and x t The load data is the user load curve, and μ is the sequence X. T The mean of sequence X, σ is the mean of sequence X. T The standard deviation.

[0091] If cv≤0.01, then the user load curve corresponds to a horizontal user. Here, 0.01 is the second threshold. The value of the first threshold can be determined according to the actual situation. It can be understood that the closer the second threshold is to 0, that is, the closer the coefficient of variation is to 0, the more accurate the judgment result will be.

[0092] It is understandable that when a user's load curve meets either the criteria in Method 1 or Method 2, the user type corresponding to that load curve can be determined to be horizontal. The above method for determining whether a user's load curve is horizontal is simple and clear, which helps improve the accuracy of the determination results. By using this method, horizontal users can be identified, and dedicated power transformer users that meet the criteria for horizontal users can be grouped into one category. This provides category input or a basis for prediction model construction strategies for downstream load forecasting tasks, and helps reduce the number of models in load forecasting tasks involving a large number of dedicated power transformer users.

[0093] In this embodiment of the application, determining whether the user type is a zero-value regularity type may include:

[0094] Determine the initial two-dimensional matrix based on the user load curve;

[0095] Based on Hamming distance, the similarity matrix is ​​obtained from the initial two-dimensional matrix;

[0096] Determine the mean of the similarity matrix;

[0097] If the mean of the similarity matrix is ​​greater than the third threshold, the user type is determined to be a zero-value regularity type.

[0098] In this embodiment, the zero-value regularity type refers to a portion of the electricity load users whose load curves only show load for a certain period of the year, and whose loads are very close together. For example, agricultural users only have load values ​​during certain solar terms. When it is determined that a user's load curve does not meet the conditions for a horizontal type, it is determined whether the user's load curve is of the zero-value regularity type.

[0099] Specifically, the first user load curve can be represented as X. T ={x1, x2, ..., x T}, where T is the amount of data in the user's load curve. The specific process for determining whether a user type is a zero-value regularity type is as follows:

[0100] User load curve X T Convert the annual cycle data format. Transform the user electricity load sequence X... T Convert to X on an annual cycle y，d，T The three-dimensional matrix is ​​obtained by calculating the average daily load curve (the average load value calculated by year and day) to obtain the initial two-dimensional matrix X. y，d , where y is the year in which the curve is located, and d is the date in the middle of the year.

[0101] If the load value for the same date in the middle of the year is 0, then delete the date (and the row it belongs to), resulting in X. y，m , where m≤d, and m is the number of dates remaining after deletion.

[0102] Calculation of X based on Hamming distance y，m The similarity of the matrices along the y-axis is used to obtain the similarity matrix, and the similarity satisfies formula (2):

[0103]

[0104] Where u and v are X y，m The sequence in year y.

[0105] Obtain the lower triangular matrix of the similarity matrix and calculate its mean. If the matrix If the similarity threshold is greater than the third threshold s0, then it is a zero-value regularity type.

[0106] The above method enables the identification of zero-value regularity users, and classifies dedicated power transformer users that meet the criteria for zero-value regularity users into one category. This provides category input or a basis for prediction model construction strategies for downstream load forecasting tasks, which helps to reduce the number of models in load forecasting tasks for a large number of dedicated power transformer users.

[0107] In this embodiment of the application, determining whether the user type is an annual cycle type may include:

[0108] Determine the initial two-dimensional matrix based on the user load curve;

[0109] The initial two-dimensional matrix is ​​filtered to obtain the first correlation coefficient matrix;

[0110] Determine the mean of the first correlation coefficient matrix;

[0111] Determine whether the user load curve exhibits periodicity based on the mean of the first correlation coefficient matrix;

[0112] If the user load curve exhibits a periodic pattern, the user type is determined to be annual periodic.

[0113] In this embodiment, an annual cycle type refers to an electricity load user whose load curve has an annual cycle, and whose load trends are very similar on the same date each year. When it is determined that a user's load curve does not meet the condition of a zero-value regularity type, it is then determined whether the user's load curve is an annual cycle type.

[0114] Specifically, if the user load curve is a non-zero value regular pattern, then the user load curve X is determined to be... T Is the user a yearly-cycle user? A yearly-cycle user's load curve has an annual cycle, and the load trends on the same dates in each year are very similar. The specific determination method is as follows:

[0115] User load curve X T Convert the annual cycle data format. Transform the user electricity load sequence X... T Convert to X on an annual cycle y，d，T The three-dimensional matrix is ​​obtained by calculating the average daily load curve (the average load value calculated by year and day) to obtain the initial two-dimensional matrix X. d，y , where y is the year in which the curve is located, and d is the date in the middle of the year.

[0116] For the initial two-dimensional matrix X d，y HP filtering is performed column-by-column. HP filtering can decompose the sequence into trend components and periodic components. In this embodiment, only the trend component g is needed. d，y HP filtering calculations satisfy formula (3):

[0117] X d，y =f(X)d，y ;λ)=g d，y +c d，y (3)

[0118] Among them, g d，y As a trend component, c d，y λ is the periodic component, and λ is the smoothing parameter. The specific value of the smoothing parameter can be adjusted according to actual needs. After multiple experiments, for the user load curve with T=96, the value of the smoothing parameter can preferably be set to 7×T.

[0119] Based on the smoothed trend part g d，y Calculate the Pearson correlation coefficient and obtain R. d×d The first correlation coefficient matrix R y×y The Pearson correlation coefficient is calculated according to formula (4):

[0120]

[0121] Where R is the correlation coefficient, and u and v are respectively g d，y A vector of a certain column (the daily average load sequence for a certain year); Let be the mean values ​​of vectors u and v, respectively; and d be the mean value of g. d，y d day.

[0122] Determine the first correlation coefficient matrix R. y×y The mean, based on the first correlation coefficient matrix R y×y The mean determination is whether the user's load curve has a periodicity. If it does, then the user's load curve X... T The corresponding users are those with an annual cycle.

[0123] In one example, is there a method for determining periodicity as follows:

[0124] With the first correlation coefficient matrix R y×y As input, the first correlation coefficient matrix R y×y Convert to a lower triangular matrix, remove the row with the smallest row mean (a) and the column with the smallest column mean (a can be customized, but a << m) to obtain the lower triangular matrix R. (m-a)×(m-a) Then calculate the lower triangular matrix R. (m-a)×(m-a) The mean of the non-empty portion is considered to have periodicity if it is greater than 0.6; otherwise, it does not.

[0125] The above method enables the identification of annual cycle users, classifying dedicated power transformer users that meet the criteria for zero-value regularity annual cycle users into one category. This provides category input or a basis for prediction model construction strategies for downstream load forecasting tasks, which helps reduce the number of models in load forecasting tasks for a large number of dedicated power transformer users.

[0126] In this embodiment of the application, determining whether the user type is a daily cycle type may include:

[0127] Filter out the load sequence from the user load curve that is most recent first preset number of days;

[0128] The second correlation coefficient matrix is ​​determined based on the load sequence;

[0129] Determine the mean of the second correlation coefficient matrix;

[0130] The mean of the second correlation coefficient matrix is ​​used to determine whether the user load curve has a periodicity.

[0131] If the user load curve shows periodicity, the user type is determined to be daily periodic.

[0132] In this embodiment, a daily-cycle type refers to an electricity load user whose load curve has a certain periodicity on a daily or weekly basis, such as a typical industrial user or commercial electricity user. When it is determined that a user's load curve does not meet the conditions for an annual-cycle type, it is then determined whether the user's load curve is a daily-cycle type.

[0133] Specifically, if the user load curve corresponds to a non-annual cycle user, then the user load curve X is determined to be... T Whether a user is a daily-cycle user depends on whether their load curve exhibits a certain periodicity on a daily or weekly basis. Examples include typical industrial and commercial electricity users. The specific methods for determining whether a user is a daily-cycle user are as follows:

[0134] Select user load curve X T The load sequence of the most recent first preset number of days, for example, selecting the load sequence of the most recent 90 days. Convert the load curve to X. d×T It is a two-dimensional matrix, with each row representing k load values ​​for a day.

[0135] According to the load matrix X d×T Calculate the Pearson correlation coefficient matrix daily to obtain the correlation matrix between days and the second correlation coefficient matrix R. d×d .

[0136] According to the second correlation coefficient matrix R d×d Determine whether the user's load curve has a periodicity. If it does, the user is a daily periodic user; otherwise, the user is a random user.

[0137] In one example, is there a method for determining periodicity as follows:

[0138] With the first correlation coefficient matrix R y×y As input, the first correlation coefficient matrix R y×y Convert to a lower triangular matrix, remove the row with the smallest row mean (a) and the column with the smallest column mean (a can be customized, but a << m) to obtain the lower triangular matrix R. (m-a)×(m-a) Then calculate the lower triangular matrix R. (m-a)×(m-a) The mean of the non-empty portion is considered to have periodicity if it is greater than 0.6; otherwise, it does not.

[0139] The above method enables the identification of daily-cycle users, classifying dedicated power transformer users that meet the criteria for daily-cycle users into one category. This provides category input or a basis for prediction model construction strategies for downstream load forecasting tasks, which helps reduce the number of models in load forecasting tasks involving a large number of dedicated power transformer users.

[0140] In this embodiment of the application, secondary classification is performed on users with annual cycle types and users with daily cycle types to obtain the target classification result, which may include:

[0141] Abnormal user load curves were identified using density clustering algorithms;

[0142] Remove users whose load curves correspond to abnormal users to obtain the target users;

[0143] Time series clustering is performed on the user load curves of the target users to obtain the target classification results.

[0144] In this embodiment, after completing the first stage of pre-classification of electricity load users based on their load curves, the electricity load users are divided into five categories. However, for annual and daily cycle users, there is still a problem of inconsistent cycle trends or cycle fluctuations. The load curves of users of the same type may differ significantly, making it impossible to predict the load using a unified model. Therefore, a second stage of secondary classification is performed on these two types of users to obtain the final classification result, i.e., the target classification result.

[0145] In one example, the initial load dataset can be obtained by first filtering the load sequence of the second most recent preset number of days in the user's power load curve; the initial load dataset can be normalized to obtain a sample dataset; the sample dataset can be zero-mean processed to obtain a zero-mean dataset; the zero-mean dataset can be smoothed to obtain a smoothed dataset; then the density clustering algorithm DBSCAN can be used to identify anomalies in the smoothed dataset, and finally the time series clustering algorithm K-Shape can be used to cluster user load curves with different trends and periods.

[0146] Specifically, the process of secondary classification of annual cycle users and daily cycle users based on user load curves is as follows:

[0147] Select the load sequence X for each user T A new load dataset X was recently constructed from the load data of the second preset number of days. m×l For example, a new load dataset X can be constructed by selecting 7 days of load data. m×l If l is the length of the sequence (7×T), and m is the number of users in the annual or daily cycle, also known as the number of samples.

[0148] For X m×l Data preprocessing begins with data normalization, unifying the loading values ​​of all samples to a scale of 0-1. Therefore, X... m×l Normalizing the samples yields X′. m×l The normalized calculation satisfies formula (5):

[0149]

[0150] Where t = l is the length of the user load data sequence.

[0151] To eliminate the differences caused by different load levels in different samples, the load sequences of the samples can be zero-meaned to obtain a zero-mean dataset. The zero-meaning process satisfies formula (6):

[0152]

[0153] The zero-mean dataset was smoothed using HP filtering, and the trend component g was selected. m×l The smoothed data satisfies formula (7):

[0154] X″ m×l =f(X″) m×l ;λ)=g m×l +c m×l (7)

[0155] Where λ is the smoothing coefficient, preferably a multiple of 8. In this embodiment, λ = 8.

[0156] Using the smoothed dataset g m×l DBSCAN clustering is performed. Through density clustering based on DBSCAN, some abnormal sequences can be obtained. These sequences are independent types of load users and will be classified into independent categories.

[0157] In the DBSCAN clustering process, DTW is used as the distance metric for the sample sequences. The proportion of independent or outlier samples is adjusted by the neighborhood density control parameter. After DBSCAN clustering, a cluster label is obtained for each sample. Samples with a label of -1 are grouped into an independent category C0, and samples with other labels are grouped into C1. Secondary clustering is performed on the samples in category C1.

[0158] From dataset g m×l After removing samples from category C0, we get The dataset is then subjected to a second stage of K-Shape clustering. K-Shape clustering clusters users based on the similarity of their load curve shapes, grouping load curves with the same shape into the same cluster. This means that load curves with the same period and fluctuations are grouped into the same category. After this grouping, a load prediction model for users in the same category can be built.

[0159] In K-Shape clustering, DTW is still used as the distance metric. The number of clusters N_Class can be determined by searching based on the silhouette coefficient or manually specified. However, when there are many samples, the number of clusters should not be too small. It is recommended to be between 5 and 12. This patent sets the number of clusters to 6.

[0160] After performing the above processing on the user load curves of annual cycle users and daily cycle users respectively, the clustering classification results are obtained. If the annual cycle users are represented by the value 3 to indicate the category in the first stage, the final classification result can be expressed as: in Let C0 be an independent category; similarly, daily-cycle users are represented by the value 4, and the final classification result can be expressed as: in It belongs to the independent category C0.

[0161] In a specific embodiment of this application, the electricity load users are classified using the above method. The numbers 1, 2, 3, 4, and 5 represent the labels of horizontal, zero-value regular, annual cycle, daily cycle, and random users after the first stage of pre-classification, respectively. The final classification results of the electricity load users are shown in Table 1.

[0162] Table 1

[0163]

[0164] In the first-stage pre-classification process, users are initially divided into multiple types using the aforementioned method. Each type represents a large category of users with load curves, which is equivalent to identifying and filtering special or abnormal load curve users. This process removes outlier samples from the second-stage clustering algorithm, improving the classification effect of the second stage. In this embodiment, HP filtering is used to extract the trend term of the load curve, achieving data smoothing. A periodicity judgment method based on the Pearson correlation coefficient is also employed.

[0165] In the second stage, users with periodic patterns were further segmented. First, abnormal user load curves were identified using DBSCAN clustering. Then, the K-Shape time series clustering algorithm was used to cluster user load curves with different trends and periods. Both DBSCAN and K-Shape clustering algorithms employed the DTW sequence distance metric to more accurately measure the similarity between user load curves, resulting in better clustering performance.

[0166] The above technical solution obtains user load curves and pre-classifies users based on these curves to determine user types. User types include horizontal, zero-value regular, annual cycle, daily cycle, and random types. Annual cycle users and daily cycle users are then further classified to obtain the target classification result. This application achieves classification based on the characteristics of the user load curves themselves, which better aligns with the electricity consumption habits of users, enabling accurate classification with good results and high practicality.

[0167] Figure 3 This is a structural block diagram of an unsupervised classification device for dedicated power transformer users based on load curves, provided as an embodiment of this application. Figure 3 As shown in the figure, this application provides an unsupervised classification device for dedicated power transformer users based on load curves, which may include:

[0168] Memory 310 is configured to store instructions; and

[0169] The processor 320 is configured to retrieve instructions from the memory 310 and, when executing the instructions, to implement the aforementioned method for controlling the boom.

[0170] Specifically, in this embodiment of the application, the processor 320 can be configured to:

[0171] Obtain user load curves;

[0172] Users are pre-classified based on their load curves to determine their user types, which include horizontal, zero-value regular, annual cycle, daily cycle, and random types.

[0173] Users with annual cycle patterns and users with daily cycle patterns are classified in two ways to obtain the target classification results.

[0174] Furthermore, the processor 320 can also be configured as follows:

[0175] Obtaining user load curves includes:

[0176] The system acquires users' power load data according to a preset cycle to form user load curves.

[0177] Furthermore, the processor 320 can also be configured as follows:

[0178] Users are pre-classified based on their load curves to determine user types, including:

[0179] Determine whether the user type is horizontal based on the user load curve;

[0180] If the user type is not horizontal, determine whether the user type is a zero-value regularity type;

[0181] If the user type is not a zero-value regular type, determine whether the user type is an annual cycle type;

[0182] If the user type is not an annual cycle type, determine whether the user type is a daily cycle type.

[0183] If the user type is not a daily cycle type, the user type is determined to be random.

[0184] Furthermore, the processor 320 can also be configured as follows:

[0185] Determining whether a user type is horizontal based on the user load curve includes at least one of the following:

[0186] Determine the mode of the user load curve. If the proportion of the mode is not lower than the first threshold, determine that the user type is horizontal; or

[0187] Determine the coefficient of variation of the user load curve. If the coefficient of variation is not higher than the second threshold, the user type is determined to be horizontal.

[0188] Furthermore, the processor 320 can also be configured as follows:

[0189] Regular patterns for determining whether a user type is a zero value include:

[0190] Determine the initial two-dimensional matrix based on the user load curve;

[0191] Based on Hamming distance, the similarity matrix is ​​obtained from the initial two-dimensional matrix;

[0192] Determine the mean of the similarity matrix;

[0193] If the mean of the similarity matrix is ​​greater than the third threshold, the user type is determined to be a zero-value regularity type.

[0194] Furthermore, the processor 320 can also be configured as follows:

[0195] Determining whether a user type is an annual cycle type includes:

[0196] Determine the initial two-dimensional matrix based on the user load curve;

[0197] The initial two-dimensional matrix is ​​filtered to obtain the first correlation coefficient matrix;

[0198] Determine the mean of the first correlation coefficient matrix;

[0199] Determine whether the user load curve exhibits periodicity based on the mean of the first correlation coefficient matrix;

[0200] If the user load curve exhibits a periodic pattern, the user type is determined to be annual periodic.

[0201] Furthermore, the processor 320 can also be configured as follows:

[0202] Determining whether a user type is a daily cycle type includes:

[0203] Filter out the load sequence from the user load curve that is most recent first preset number of days;

[0204] The second correlation coefficient matrix is ​​determined based on the load sequence;

[0205] Determine the mean of the second correlation coefficient matrix;

[0206] The mean of the second correlation coefficient matrix is ​​used to determine whether the user load curve has a periodicity.

[0207] If the user load curve shows periodicity, the user type is determined to be daily periodic.

[0208] Furthermore, the processor 320 can also be configured as follows:

[0209] Users with annual or daily time cycles are further classified to obtain the target classification results, including:

[0210] Abnormal user load curves were identified using density clustering algorithms;

[0211] Remove users whose load curves correspond to abnormal users to obtain the target users;

[0212] Time series clustering is performed on the user load curves of the target users to obtain the target classification results.

[0213] The above technical solution obtains user load curves and pre-classifies users based on these curves to determine user types. User types include horizontal, zero-value regular, annual cycle, daily cycle, and random types. Annual cycle users and daily cycle users are then further classified to obtain the target classification result. This application achieves classification based on the characteristics of the user load curves themselves, which better aligns with the electricity consumption habits of users, enabling accurate classification with good results and high practicality.

[0214] This application also provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described unsupervised classification method for dedicated power transformer users based on load curves.

[0215] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0216] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0217] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0218] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0219] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0220] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0221] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0222] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0223] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An unsupervised classification method for dedicated power transformer users based on load curves, characterized in that, include: Obtain user load curves; Users are pre-classified based on the user load curve to determine user types, including horizontal, zero-value regular, annual cycle, daily cycle, and random types. The users with the annual cycle type and the users with the daily cycle type are classified in two ways to obtain the target classification result. The step of pre-classifying users based on the user load curve to determine user type includes: Determine whether the user type is the horizontal type based on the user load curve; If the user type is not the horizontal type, determine whether the user type is the zero-value regular type; If the user type is not the zero-value regularity type, determine whether the user type is the annual cycle type; If the user type is not the annual cycle type, determine whether the user type is the daily cycle type; If the user type is not the daily cycle type, the user type is determined to be the random type. The step of determining whether the user type is a zero-value regularity includes: Determine the initial two-dimensional matrix based on the user load curve; Based on the Hamming distance, a similarity matrix is ​​obtained from the initial two-dimensional matrix; Determine the mean of the similarity matrix; If the mean of the similarity matrix is ​​greater than the third threshold, the user type is determined to be the zero-value regularity type. The step of determining the initial two-dimensional matrix based on the user load curve includes: The user load curve is represented as a user electricity load sequence. Where T is the amount of data in the user load curve; The user electricity load sequence Converting the data into a three-dimensional matrix with an annual cycle, and calculating the average daily load curve, yields an initial two-dimensional matrix. ,in y The curve is located at the first y Year, d The date in the middle of the year; The step of obtaining the similarity matrix based on the Hamming distance and the initial two-dimensional matrix includes: The initial two-dimensional matrix Delete dates with a load value of 0 that are the same as the dates in the middle age group to obtain the result. Matrix, where, y The curve is located at the first y Year, m This represents the number of dates remaining after deletion. Based on Hamming distance, calculate Matrix in y The similarity of the axes is used to obtain the similarity matrix.

2. The unsupervised classification method for dedicated power transformer users according to claim 1, characterized in that, The method for obtaining user load curves includes: The power load data of the user is acquired according to a preset period to form the user load curve.

3. The unsupervised classification method for dedicated power transformer users according to claim 1, characterized in that, The step of determining whether the user type is the horizontal type based on the user load curve includes at least one of the following: Determine the mode of the user load curve, and if the proportion of the mode is not lower than a first threshold, determine that the user type is the horizontal type; or Determine the coefficient of variation of the user load curve, and if the coefficient of variation is not higher than a second threshold, determine that the user type is the horizontal type.

4. The unsupervised classification method for dedicated power transformer users according to claim 1, characterized in that, The step of determining whether the user type is the annual cycle type includes: Determine the initial two-dimensional matrix based on the user load curve; The initial two-dimensional matrix is ​​filtered to obtain the first correlation coefficient matrix; Determine the mean of the first correlation coefficient matrix; The user load curve is determined to have periodicity based on the mean of the first correlation coefficient matrix. If the user load curve exhibits a periodic pattern, the user type is determined to be the annual periodic type. The step of determining the initial two-dimensional matrix based on the user load curve includes: The user load curve is represented as a user electricity load sequence. Where T is the amount of data in the user load curve; The user electricity load sequence Converting the data into a three-dimensional matrix with an annual cycle, and calculating the average daily load curve, yields an initial two-dimensional matrix. ,in y The curve is located at the first y Year, d The date in the middle of the year; The step of filtering the initial two-dimensional matrix to obtain the first correlation coefficient matrix includes: For the initial two-dimensional matrix Perform HP filtering calculations to obtain the trend components; Based on the trend components, the Pearson correlation coefficient is calculated to obtain the first correlation coefficient matrix.

5. The unsupervised classification method for dedicated power transformer users according to claim 1, characterized in that, The step of determining whether the user type is the daily cycle type includes: Filter out the load sequence of the most recent first preset number of days from the user load curve; Determine the second correlation coefficient matrix based on the load sequence; Determine the mean of the second correlation coefficient matrix; The user load curve is determined to have periodicity based on the mean of the second correlation coefficient matrix. If the user load curve exhibits periodicity, the user type is determined to be the daily periodic type. The step of determining the second correlation coefficient matrix based on the load sequence includes: Determine the load sequence of the user load curve for the first preset number of days, and convert the load sequence into a two-dimensional load matrix; Based on the two-dimensional load matrix, the Pearson correlation coefficient matrix is ​​calculated to obtain the second correlation coefficient matrix.

6. The unsupervised classification method for dedicated power transformer users according to claim 1, characterized in that, The step of performing secondary classification on the annual cycle type users and the daily cycle type users respectively to obtain the target classification result includes: Abnormal user load curves were identified using density clustering algorithms; Users corresponding to the abnormal user load curves are removed to obtain the target users; Time series clustering is performed on the user load curves of the target users to obtain the target classification results.

7. An unsupervised classification device for dedicated power transformer users based on load curves, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the unsupervised classification method for dedicated power transformer users based on load curves according to any one of claims 1 to 6.

8. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform the unsupervised classification method for dedicated power transformer users based on load curves according to any one of claims 1 to 6.

Images