Low-power-factor power load classification method and device and storage medium

Abstract

The invention discloses a low-power-factor power load classification method and device and a storage medium. The low-power-factor power load classification method comprises the steps of obtaining power load data corresponding to a to-be-classified power load generated under a low-power-factor condition; wherein the power load data comprises a plurality of types of power load related data; dimensionality reduction is carried out on the power load data to obtain power load data after dimensionality reduction; and classifying the to-be-classified power load according to the dimension-reduced power load data by using a pre-trained target power load classification model to obtain a target load category. The load classification efficiency and classification accuracy are improved.

Description

Technical Field

[0001] The present application relates to the technical field of power data processing, and in particular to a method and device for classifying low power factor power loads and a computer-readable storage medium. Background Art

[0002] In today's smart grid environment, typical low power factor load scenarios are numerous. To support power producers' production planning, provide more personalized power services, help power operators better understand and predict power demand, thereby optimizing the allocation and use of power resources, improving grid efficiency and stability, and helping different power users better understand their power consumption patterns and effectively reduce power consumption costs, it is necessary to classify low power factor power loads.

[0003] Currently, low power factor loads are typically classified based on power mean values, which group loads with similar power mean values into one category. However, this load classification method only considers the mean value and misses other key information, resulting in low accuracy in load classification.

[0004] In summary, how to effectively solve the problem of low accuracy in power load classification is an issue that those skilled in the art urgently need to solve. Summary of the Invention

[0005] The purpose of this application is to provide a low power factor electric load classification method, which improves the load classification efficiency and classification accuracy; another purpose of this application is to provide a low power factor electric load classification device and a computer-readable storage medium.

[0006] To solve the above technical problems, this application provides the following technical solutions:

[0007] A method for classifying low power factor electric loads, comprising:

[0008] Obtaining power load data corresponding to the power load to be classified generated under a low power factor condition; wherein the power load data includes data related to power loads of multiple categories;

[0009] Performing dimensionality reduction on the power load data to obtain the reduced-dimensional power load data;

[0010] The target power load classification model obtained by pre-training is used to classify the power load to be classified according to the power load data after dimensionality reduction to obtain the target load category.

[0011] In a specific embodiment of the present application, after obtaining power load data corresponding to the power load to be classified generated under a low power factor condition, and before performing dimensionality reduction on the power load data, the method further includes:

[0012] For each type of power load related data, draw a load data curve that changes over time;

[0013] The power load data are pre-processed according to the load data curves.

[0014] In a specific embodiment of the present application, preprocessing each power load data according to each load data curve includes:

[0015] For each load data curve, each collection time point in the load data curve is used as a target collection time point;

[0016] Determining a load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point;

[0017] Determining whether the load data change rate is greater than a first preset threshold;

[0018] If so, the target load data corresponding to the target collection time point is determined to be abnormal data;

[0019] If not, it is determined that the target load data corresponding to the target collection time point is normal data;

[0020] For each abnormal data, select each forward normal load data before the abnormal data and each backward normal load data after the abnormal data from the load data curve to which the abnormal data belongs;

[0021] Correcting the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data;

[0022] Normalization processing is performed on each load data included in the load data curve after abnormal data correction.

[0023] In a specific embodiment of the present application, correcting the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data includes:

[0024] Count the total number of forward load data and the total number of backward load data corresponding to the abnormal data;

[0025] Performing sum calculation on each forward normal load data corresponding to the abnormal data to obtain the sum of the forward load data corresponding to the abnormal data;

[0026] Perform sum calculation on each backward normal load data corresponding to the abnormal data to obtain the sum of the backward load data corresponding to the abnormal data;

[0027] Summing up the total amount of forward load data and the total amount of backward load data corresponding to the abnormal data to obtain a reference data amount correction parameter corresponding to the abnormal data;

[0028] Summing the forward load data and the backward load data corresponding to the abnormal data to obtain reference data and correction parameters corresponding to the abnormal data;

[0029] The reference data and correction parameter corresponding to the abnormal data are calculated as a quotient of the reference data correction parameter to obtain the correction data corresponding to the abnormal data, and the abnormal data is replaced by the correction data corresponding to the abnormal data.

[0030] In a specific embodiment of the present application, dimensionality reduction is performed on the power load data to obtain the reduced-dimensional power load data, including:

[0031] Determine the target Gaussian distribution variance corresponding to the power load data;

[0032] For each power load data, the target similarity between the power load data and other power load data is calculated according to the target Gaussian distribution variance;

[0033] Calculate the target joint probability distribution of the power load data according to the similarity of each target corresponding to the power load data;

[0034] Randomly determine initial power load data of the power load data in a low-dimensional space;

[0035] Determine the initial Gaussian distribution variance corresponding to the initial power load data;

[0036] For each initial power load data, calculating the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance;

[0037] Calculating an initial joint probability distribution of the initial power load data according to each initial similarity corresponding to the initial power load data;

[0038] Calculating a quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain a joint probability evaluation function of the power load data;

[0039] Determining a loss function of a joint probability distribution between the power load data and its corresponding initial power load data based on a joint probability evaluation function of the power load data and a target joint probability distribution;

[0040] Determining whether a loss function between the power load data and its corresponding initial power load data satisfies a second preset threshold;

[0041] If so, the initial power load data corresponding to the power load data is determined as the power load data after dimension reduction corresponding to the power load data;

[0042] If not, the process returns to the step of randomly determining initial power load data of the power load data in the low-dimensional space.

[0043] In a specific embodiment of the present application, a target power load classification model training process is also included. The target power load classification model training process includes:

[0044] Acquire sample power load data, and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category;

[0045] Construct a sample data set based on the sample power load data corresponding to each load category;

[0046] Divide the sample data set into a training data set and a test data set;

[0047] The pre-built initial power load classification model is trained using the training data set, and the trained initial power load classification model is tested using the test data set to obtain the target power load classification model.

[0048] In a specific embodiment of the present application, clustering is performed on the sample power load data to obtain sample power load data corresponding to each load category, including:

[0049] Determine the initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as each current centroid;

[0050] For each sample power load data, respectively calculating the first distance between the sample power load data and the current centroid corresponding to each load category cluster;

[0051] Dividing the sample power load data into load category clusters corresponding to the minimum first distance among the corresponding first distances;

[0052] According to the sample power load data in each load category cluster obtained by the current division, the updated centroid corresponding to each load category cluster is determined;

[0053] Determine whether the updated centroids correspond to the current centroids;

[0054] If yes, the sample power load data currently divided into each load category cluster is determined as the sample power load data corresponding to each load category;

[0055] If not, each updated centroid is determined as each current centroid, and the process returns to the step of calculating the first distance between each sample power load data and the current centroid corresponding to each load category cluster.

[0056] In a specific embodiment of the present application, before determining the initial centroids corresponding to the preset number of load category clusters, the method further includes:

[0057] Get multiple cluster quantity values;

[0058] For each cluster quantity value, determine the centroid corresponding to each initial load category cluster according to the cluster quantity value, and divide each sample power load data into each initial load category cluster to obtain the sample power load data corresponding to each initial load category cluster;

[0059] Calculate the sum of squares of the distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, and obtain the evaluation parameter of the cluster quantity value;

[0060] Draw an evaluation parameter curve according to the number of clusters and the evaluation parameters corresponding to the number of clusters;

[0061] For each cluster number value in the evaluation parameter curve, calculate the change rate of the evaluation parameter corresponding to the cluster number value compared to the evaluation parameter corresponding to the nearest neighbor forward cluster number value;

[0062] Select the first change rate whose absolute value is less than a preset change rate threshold from the change rate of the evaluation parameter corresponding to each cluster number value in the evaluation parameter curve;

[0063] The cluster quantity value corresponding to the selected change rate is determined as the preset quantity.

[0064] In a specific embodiment of the present application, determining the initial centroids corresponding to a preset number of load category clusters includes:

[0065] Determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density;

[0066] The power load data of each sample is divided into a high-density set and a low-density set according to the density mean;

[0067] Determining high-density sample power load data having a data density greater than a preset density threshold from the sample power load data in the high-density set;

[0068] A preset number of central sample power load data with a mutual distance greater than a preset distance threshold is determined in each high-density sample power load data, and each central sample power load data is determined as the initial centroid corresponding to the preset number of load category clusters.

[0069] In a specific embodiment of the present application, after obtaining the sample power load data corresponding to each load category, the method further includes:

[0070] For each sample power load data, calculating the intra-class average distance between the sample power load data and other sample power load data in the load category to which the sample power load data belongs;

[0071] Determining, from each sample power load data included in each load category other than the load category to which the sample power load data belongs, reference sample power load data that is closest to the sample power load data;

[0072] determining an average inter-class distance between the sample power load data and corresponding reference sample power load data;

[0073] Determining a first clustering effect evaluation parameter for clustering the sample power load data according to the intra-class average distance and the inter-class average distance corresponding to each sample power load data;

[0074] Determining the data volume of the sample power load data and the number of categories of each load category, and determining the category center point of each load category, the data center point of the sample power load data, and the intra-category data volume of the sample power load data in each load category;

[0075] Determine the inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and determine the intra-class covariance matrix based on the center point of the class and each load class;

[0076] Determine the second clustering effect evaluation parameter according to the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of categories;

[0077] Calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion corresponding to each load category;

[0078] Calculate the distance between the class centroids corresponding to any two load classes in each load class to obtain the inter-class dispersion corresponding to the two load classes;

[0079] The intra-class dispersion degrees corresponding to the two load categories are summed up and calculated to obtain the intra-class dispersion degree evaluation parameters corresponding to the two load categories;

[0080] The quotient of the intra-class dispersion evaluation parameter corresponding to the two load categories and the inter-class dispersion corresponding to the two load categories is calculated to obtain the class similarity between the two load categories;

[0081] According to the category similarity between each load category and other load categories, the maximum category similarity corresponding to each load category is determined;

[0082] Calculate the mean of the similarities of each maximum category to obtain the third clustering effect evaluation parameter;

[0083] Determine the weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter respectively, and perform weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain the load clustering effect evaluation parameter;

[0084] According to the load clustering effect evaluation parameters, determine whether the clustering results of the sample power load data meet the preset clustering conditions;

[0085] If so, each load category is used as the labeling information of the sample power load data contained in each load category, and a sample data set is constructed based on the sample power load data corresponding to each load category, including:

[0086] Construct a sample data set based on the sample power load data with labeled information corresponding to each load category;

[0087] If not, cluster the sample power load data again.

[0088] A low power factor electric load classification device, comprising:

[0089] An electric load data acquisition module is used to acquire electric load data corresponding to the electric load to be classified generated under low power factor conditions; wherein the electric load data includes multiple categories of electric load related data;

[0090] A module for obtaining power load data after dimensionality reduction is used to reduce the dimensionality of the power load data to obtain power load data after dimensionality reduction;

[0091] The target load category acquisition module is used to classify the unclassified power load according to the power load data after dimensionality reduction using the pre-trained target power load classification model to obtain the target load category.

[0092] In a specific embodiment of the present application, it also includes:

[0093] A curve drawing module is used to draw a load data curve that changes with time for each type of power load related data after obtaining power load data corresponding to the power load to be classified generated under low power factor conditions and before performing dimensionality reduction on the power load data;

[0094] The preprocessing module is used to preprocess each power load data according to each load data curve.

[0095] In a specific embodiment of the present application, the preprocessing module includes:

[0096] The target collection time point determination submodule is used to determine each collection time point in the load data curve as the target collection time point for each load data curve;

[0097] a load data change rate determination submodule, configured to determine the load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point;

[0098] A first judgment submodule is used to judge whether the load data change rate is greater than a first preset threshold;

[0099] an abnormal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is abnormal data when it is determined that the load data change rate is greater than a first preset threshold;

[0100] a normal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is normal data when it is determined that the load data change rate is not greater than a first preset threshold;

[0101] A data selection submodule is used to select, for each abnormal data, each forward normal load data before the abnormal data and each backward normal load data after the abnormal data from the load data curve to which the abnormal data belongs;

[0102] A data correction submodule, configured to correct the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data;

[0103] The normalization processing submodule is used to perform normalization processing on each load data included in the load data curve after abnormal data correction.

[0104] In a specific embodiment of the present application, the data correction submodule includes:

[0105] A quantity statistics unit is used to count the total number of forward load data and the total number of backward load data corresponding to the abnormal data;

[0106] A forward load data sum obtaining unit is used to sum each forward normal load data corresponding to the abnormal data to obtain the forward load data sum corresponding to the abnormal data;

[0107] A backward load data sum obtaining unit is used to sum and calculate each backward normal load data corresponding to the abnormal data to obtain the backward load data sum corresponding to the abnormal data;

[0108] a data volume correction parameter obtaining unit, configured to sum the total amount of forward load data and the total amount of backward load data corresponding to the abnormal data to obtain a reference data volume correction parameter corresponding to the abnormal data;

[0109] a data and correction parameter obtaining unit, configured to calculate the sum of the forward load data and the backward load data corresponding to the abnormal data, and obtain reference data and correction parameters corresponding to the abnormal data;

[0110] The data replacement unit is used to calculate the quotient of the reference data and correction parameter corresponding to the abnormal data and the reference data amount correction parameter to obtain the correction data corresponding to the abnormal data, and replace the abnormal data with the correction data corresponding to the abnormal data.

[0111] In a specific embodiment of the present application, the module for obtaining power load data after dimensionality reduction includes:

[0112] A target Gaussian distribution variance determination submodule is used to determine the target Gaussian distribution variance corresponding to the power load data;

[0113] A target similarity calculation submodule is used to calculate the target similarity between each power load data and other power load data according to the target Gaussian distribution variance;

[0114] The target joint probability distribution calculation submodule is used to calculate the target joint probability distribution of power load data according to the similarity of each target corresponding to the power load data;

[0115] An initial power load data determination submodule, used for randomly determining initial power load data of power load data in a low-dimensional space;

[0116] An initial Gaussian distribution variance determination submodule is used to determine the initial Gaussian distribution variance corresponding to the initial power load data;

[0117] An initial similarity calculation submodule is used to calculate, for each initial power load data, the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance;

[0118] An initial joint probability distribution calculation submodule, configured to calculate an initial joint probability distribution of the initial power load data according to each initial similarity corresponding to the initial power load data;

[0119] A joint probability evaluation function obtaining submodule is used to calculate the quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain the joint probability evaluation function of the power load data;

[0120] A loss function determination submodule is used to determine the loss function of the joint probability distribution between the power load data and its corresponding initial power load data based on the joint probability evaluation function of the power load data and the target joint probability distribution;

[0121] A second judgment submodule is used to judge whether the loss function between the power load data and the corresponding initial power load data meets a second preset threshold;

[0122] a submodule for determining the power load data after dimensionality reduction, configured to determine the initial power load data corresponding to the power load data as the power load data after dimensionality reduction corresponding to the power load data when a loss function between the determined power load data and the initial power load data corresponding thereto satisfies a second preset threshold;

[0123] The repeated execution submodule is used to repeatedly execute the step of randomly determining the initial power load data of the power load data in the low-dimensional space when the loss function between the determined power load data and its corresponding initial power load data does not meet the second preset threshold.

[0124] In a specific embodiment of the present application, the device further includes a model training module, which includes:

[0125] The sample power load data acquisition submodule is used to obtain sample power load data and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category;

[0126] A sample data set construction submodule is used to construct a sample data set based on the sample power load data corresponding to each load category;

[0127] The training data set and test data set division submodule is used to divide the sample data set into a training data set and a test data set;

[0128] The target power load classification model acquisition submodule is used to train the pre-built initial power load classification model using the training data set, and test the trained initial power load classification model using the test data set to obtain the target power load classification model.

[0129] In a specific embodiment of the present application, the sample power load data acquisition submodule includes:

[0130] a current centroid determining unit, configured to determine initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as each current centroid;

[0131] A first distance calculation unit is used to calculate, for each sample power load data, a first distance between the sample power load data and a current centroid corresponding to each load category cluster;

[0132] a load category clustering unit, configured to divide the sample power load data into a load category cluster corresponding to the minimum first distance among the corresponding first distances;

[0133] An updated centroid determining unit, configured to determine the updated centroid corresponding to each load category cluster based on the sample power load data in each load category cluster currently divided;

[0134] A judgment unit, used to judge whether each updated centroid corresponds to each current centroid;

[0135] a sample power load data determining unit, configured to determine the sample power load data currently classified into each load category cluster as the sample power load data corresponding to each load category when it is determined that each updated centroid corresponds to each current centroid;

[0136] The repeated execution unit is used to determine each updated centroid as each current centroid when it is determined that the updated centroid does not correspond to each current centroid, and return to execute the step of calculating the first distance between the sample power load data and the current centroid corresponding to each load category cluster for each sample power load data.

[0137] In a specific embodiment of the present application, the device further includes:

[0138] A cluster quantity value acquisition module is used to obtain a plurality of cluster quantity values before determining the initial centroids corresponding to the preset number of load category clusters;

[0139] A sample power load data determination module is used to determine the centroid corresponding to each initial load category cluster according to the cluster number value for each cluster number, and to divide each sample power load data into each initial load category cluster to obtain the sample power load data corresponding to each initial load category cluster;

[0140] An evaluation parameter acquisition module is used to calculate the sum of squares of distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, and obtain an evaluation parameter for the cluster quantity value;

[0141] An evaluation parameter curve drawing module is used to draw an evaluation parameter curve according to the number of clusters and the evaluation parameters corresponding to the number of clusters;

[0142] A change rate calculation module is used to calculate the change rate of the evaluation parameter corresponding to each cluster number value in the evaluation parameter curve compared with the evaluation parameter corresponding to the nearest neighbor forward cluster number value;

[0143] A change rate selection module is used to select the first change rate whose absolute value is less than a preset change rate threshold from the change rates of the evaluation parameters corresponding to each cluster number value in the evaluation parameter curve;

[0144] The preset number determination module is used to determine the cluster number value corresponding to the selected change rate as the preset number.

[0145] In a specific embodiment of the present application, the current center of mass determination unit includes:

[0146] a density mean determination subunit, configured to determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density;

[0147] A high-density set and a low-density set division subunit is used to divide each sample power load data into a high-density set and a low-density set according to the density mean;

[0148] a high-density sample power load data determining subunit, configured to determine, from the sample power load data in the high-density set, high-density sample power load data having a data density greater than a preset density threshold;

[0149] The initial centroid determination subunit is used to determine a preset number of central sample power load data with a mutual distance greater than a preset distance threshold in each high-density sample power load data, and determine each central sample power load data as the initial centroid corresponding to a preset number of load category clusters.

[0150] In a specific embodiment of the present application, the device further includes:

[0151] An intra-class average distance calculation module is used to calculate, after obtaining the sample power load data corresponding to each load category, the intra-class average distance between each sample power load data and other sample power load data in the load category to which the sample power load data belongs;

[0152] A module for determining the nearest reference sample power load data, configured to determine the reference sample power load data closest to the sample power load data from among the sample power load data respectively included in the load categories other than the load category to which the sample power load data belongs;

[0153] an inter-class average distance determination module, configured to determine an inter-class average distance between the sample power load data and corresponding reference sample power load data;

[0154] a first clustering effect evaluation parameter determination module, configured to determine a first clustering effect evaluation parameter for clustering the sample power load data based on the intra-class average distance and the inter-class average distance corresponding to each sample power load data;

[0155] An intra-class data volume determination module is used to determine the data volume of the sample power load data and the number of categories of each load category, and to determine the category center point of each load category, the data center point of the sample power load data, and the intra-class data volume of the sample power load data in each load category;

[0156] The intra-class covariance matrix determination module is used to determine the inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and to determine the intra-class covariance matrix based on the center point of the class and each load class;

[0157] a second clustering effect evaluation parameter determination module, configured to determine a second clustering effect evaluation parameter based on the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of classes;

[0158] The intra-class dispersion degree acquisition module is used to calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion degree corresponding to each load category;

[0159] The inter-class dispersion degree obtaining module is used to calculate the distance between the class centroids corresponding to any two load classes in each load class, and obtain the inter-class dispersion degrees corresponding to the two load classes;

[0160] The intra-class dispersion evaluation parameter acquisition module is used to calculate the sum of the intra-class dispersions corresponding to the two load categories to obtain the intra-class dispersion evaluation parameters corresponding to the two load categories;

[0161] A category similarity obtaining module is used to calculate the quotient of the intra-category dispersion evaluation parameters corresponding to the two load categories and the inter-category dispersion corresponding to the two load categories to obtain the category similarity between the two load categories;

[0162] A maximum category similarity determination module is used to determine the maximum category similarity corresponding to each load category based on the category similarity between each load category and other load categories;

[0163] A third clustering effect evaluation parameter obtaining module is used to calculate the mean of the maximum category similarities to obtain the third clustering effect evaluation parameter;

[0164] a load clustering effect evaluation parameter acquisition module, used to determine the weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter, respectively, and perform weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain the load clustering effect evaluation parameter;

[0165] A judgment module is used to judge whether the clustering result of the sample power load data meets the preset clustering conditions according to the load clustering effect evaluation parameters;

[0166] a labeling information determining module, configured to, when determining that the clustering result of the sample power load data satisfies a preset clustering condition, use each load category as labeling information of the sample power load data included in each load category;

[0167] The sample data set construction submodule is specifically a module for constructing a sample data set based on the sample power load data with labeled information corresponding to each load category;

[0168] The re-clustering module is used to re-cluster the sample power load data when it is determined that the clustering result of the sample power load data does not meet the preset clustering condition.

[0169] A low power factor electric load classification device comprising:

[0170] Memory for storing computer programs;

[0171] The processor is used to implement the steps of the low power factor electric load classification method when executing the computer program.

[0172] A computer-readable storage medium stores a computer program, which implements the steps of the above low power factor electric load classification method when executed by a processor.

[0173] The low power factor electric load classification method provided in the present application obtains electric load data corresponding to the electric load to be classified generated under low power factor conditions; wherein the electric load data includes electric load-related data of multiple categories; the dimensionality of the electric load data is reduced to obtain the reduced-dimensional electric load data; and the target electric load classification model obtained by pre-training is used to classify the electric load to be classified according to the reduced-dimensional electric load data to obtain the target load category.

[0174] The above technical solution demonstrates that by comprehensively analyzing multiple categories of power load data to classify low-power-factor loads, the accuracy of load classification is improved. By reducing the dimensionality of load data, the computational effort is reduced while retaining important information, significantly improving the efficiency of load classification. By utilizing a pre-trained target load classification model to classify loads, both efficiency and accuracy are further improved.

[0175] Correspondingly, the present application also provides a low power factor power load classification device and a computer-readable storage medium corresponding to the above-mentioned low power factor power load classification method, which have the above-mentioned technical effects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0176] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0177] Figure 1 This is a flowchart of an implementation method for classifying low power factor electric loads in an embodiment of the present application;

[0178] Figure 2 Another implementation flow chart of the low power factor power load classification method in the embodiment of the present application;

[0179] Figure 3 This is a schematic diagram of the distribution of power load data after dimensionality reduction in an embodiment of the present application;

[0180] Figure 4 This is a structural block diagram of a low power factor electric load classification device in an embodiment of the present application;

[0181] Figure 5 This is a structural block diagram of a low power factor power load classification device in an embodiment of the present application;

[0182] Figure 6 A schematic diagram of the specific structure of a low power factor electric load classification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0183] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the embodiments described are only a part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present application.

[0184] See also Figure 1 , Figure 1 This is a flowchart of an implementation method for classifying low power factor power loads in an embodiment of the present application. The method may include the following steps:

[0185] S101: Obtaining power load data corresponding to the power load to be classified generated under a low power factor condition.

[0186] The power load data includes multiple categories of power load related data.

[0187] When it is necessary to classify unclassified power loads generated under low power factor conditions, power load data corresponding to the unclassified power loads generated under low power factor conditions is obtained. Power load data includes multiple categories of power load-related data, such as voltage data, current data, low power factor data, device attribute data of the power-consuming device to which the power load data belongs, environmental attribute data surrounding the power-consuming device, and user characteristic data of the user to whom the power load data belongs. Various data, such as voltage data, current data, low power factor data, device attribute data, environmental attribute data, and user characteristic data, can be obtained from the power database. Device attribute data may include data describing device characteristics, operating status, and historical information, such as rated frequency, rated capacity, and device type. Environmental attribute data may include climatic environment data (such as temperature and humidity), mechanical environment data (such as vibration and shock that the device may be subjected to during operation), electromagnetic environment data (such as electromagnetic interference from other electronic devices during operation), and biochemical environment data (such as corrosive gases or liquids). User characteristic data may include the user's electricity consumption scale, user type, and electricity usage behavior patterns.

[0188] It should be noted that the voltage data, current data, low power factor data, equipment attribute data, environmental attribute data, user characteristic data, etc. obtained in the embodiments of the present application are not private data, but non-private data related to load classification tasks that can be obtained in the power system management platform.

[0189] After acquiring various power load data, the power loads can be classified according to the aforementioned various power load data. Therefore, in the process of classifying loads, the embodiment of the present application comprehensively analyzes various data such as voltage data, current data, low power factor data, device attribute data, environmental attribute data, and user characteristic data, taking into account all factors related to load classification, thereby improving the classification accuracy of low power factor power loads.

[0190] S102: Perform dimensionality reduction on the power load data to obtain dimensionality-reduced power load data.

[0191] Since the acquired power load data usually contains a large number of features, these features will bring a large burden in calculation and storage. Therefore, the power load data can be reduced in dimensionality to obtain the power load data after dimensionality reduction. Specifically, the dimensionality of the voltage data is reduced to obtain the voltage data after dimensionality reduction; the dimensionality of the current data is reduced to obtain the current data after dimensionality reduction; the dimensionality of the low power factor data is reduced to obtain the low power factor data after dimensionality reduction; the dimensionality of the device attribute data is reduced to obtain the device attribute data after dimensionality reduction; the dimensionality of the environmental attribute data is reduced to obtain the environmental attribute data after dimensionality reduction; the dimensionality of the user feature data is reduced to obtain the user feature data after dimensionality reduction. Finally, the above-mentioned various power load data after dimensionality reduction are used to realize the classification of the power load. The embodiment of the present application can remove the noise and irrelevant features contained in the power load data and improve the purity of the data by performing dimensionality reduction processing on the power load data.

[0192] S103: Using the pre-trained target power load classification model to classify the power load to be classified according to the power load data after dimensionality reduction, to obtain a target load category.

[0193] After reducing the dimensionality of the power load data to obtain the reduced-dimensional power load data, the target power load classification model obtained by pre-training is used to classify the power load to be classified according to the reduced-dimensional power load data to obtain the target load category.

[0194] The target power load classification model is pre-trained using a sample power load dataset with labeled information. For example, dimensionality-reduced voltage data, dimensionality-reduced current data, dimensionality-reduced low power factor data, dimensionality-reduced device attribute data, dimensionality-reduced environmental attribute data, and dimensionality-reduced user feature data can be input into the target power load classification model. The target power load classification model can then output the load category corresponding to the power load to be classified. The embodiments of the present application implement power load classification through the target power load classification model, thereby improving the efficiency and accuracy of power load classification.

[0195] The process of predicting the power load to be classified through the target power load classification model may include: determining the voltage feature vector corresponding to the voltage data after dimensionality reduction, the current feature vector corresponding to the current data after dimensionality reduction, the low power factor feature vector corresponding to the low power factor data after dimensionality reduction, the device attribute feature vector corresponding to the device attribute data after dimensionality reduction, the environmental attribute data feature vector corresponding to the environmental attribute data after dimensionality reduction, and the user feature vector corresponding to the user feature data after dimensionality reduction; performing feature cross processing on the voltage feature vector, the current feature vector, the low power factor feature vector, the device attribute feature vector, the environmental attribute data feature vector, and the user feature vector to obtain a feature cross vector; performing element-level cross processing on the voltage feature vector, the current feature vector, the low power factor feature vector, the device attribute feature vector, the environmental attribute data feature vector, and the user feature vector to obtain an element cross vector; using a preset transformation function to transform the feature cross vector and the element-level cross processing to obtain a category cross feature vector; inputting the category cross feature vector into the target power load classification model to obtain the load category corresponding to the power load to be classified.

[0196] The process of predicting the power load to be classified by the target power load classification model can specifically include: first, using word embedding and other methods to determine the feature vectors corresponding to various power load data after dimensionality reduction. Since the voltage data after dimensionality reduction, the current data after dimensionality reduction, the low power factor data after dimensionality reduction, the device attribute data after dimensionality reduction, the environmental attribute data after dimensionality reduction, and the user feature data after dimensionality reduction are different fields, the processing method is to embed the data of different fields into vectors of the same latitude. For example, if the voltage feature vector is , the current characteristic vector is , the low power factor characteristic vector is , the device attribute feature vector is , the environmental attribute feature vector is , the user feature vector is The cross processing method may include: performing feature-level crossover between different feature vectors, that is, performing Hadamard product on all elements between vectors, and then performing convolution transformation under certain weights to obtain the feature cross vector as ; and perform element-level crossover on all eigenvector data, that is, after performing Hadamard product on each element between the vectors, assign different weight values to the results of each product. The weight of each parameter can be determined by analyzing the sample data and using statistical methods. For example, principal component analysis (PCA) can be used to identify the most important features in the data and assign weights based on the importance of the features. The more important the parameter, the larger its weight value should be. After performing linear transformation, the element crossover vector obtained is Finally, the above feature cross vectors and element cross vectors are combined together, such as linearly combining the feature cross vector and the element cross vector, and performing transformation processing using a preset function, such as using an activation function to perform nonlinear transformation on the feature cross vector and the element cross vector to obtain a category cross feature vector. The preset function can be set according to actual conditions, and the embodiment of the present application does not limit this.

[0197] It should be noted that the above examples are only illustrative and do not limit the embodiments of the present application. Thus, by cross-processing the voltage feature vector, the current feature vector, the low power factor feature vector, the device attribute feature vector, the environment attribute feature vector, and the user feature vector, different features can be automatically or explicitly combined to generate new feature combinations. These combined features may contain complex nonlinear relationships between the original features, so that the model can capture more refined and rich information in the data, that is, it can make full use of the relationship between various data and extract more implicit features. Feature cross-processing can combine different features to generate new feature combinations. These feature combinations may contain complex nonlinear relationships between the original features, so that the model can capture more refined and rich information in the data. Feature cross-processing can simultaneously consider the low-order and high-order relationships between features, thereby taking into account both high-order and low-order processing, making data utilization more sufficient, making the classification of power loads more accurate, and meeting the needs of actual application scenarios.

[0198] Load categories can be categorized by power generation and supply usage, including power load, line loss load, power supply load, plant load, and power generation load. They can also be categorized by time of occurrence, including peak load, minimum load, and average load. They can also be categorized by interruption loss, including primary, secondary, and tertiary loads, with the higher the level, the lower the loss. They can also be categorized by electricity consumption sector, including residential load, commercial load, and industrial load. It should be noted that the above load categories are merely illustrative, and the embodiments of this application are not limited to these load categories.

[0199] The above technical solution demonstrates that by comprehensively analyzing multiple categories of power load data to classify low-power-factor loads, the accuracy of load classification is improved. By reducing the dimensionality of load data, the computational effort is reduced while retaining important information, significantly improving the efficiency of load classification. By utilizing a pre-trained target load classification model to classify loads, both efficiency and accuracy are further improved.

[0200] It should be noted that, based on the above embodiment, the present application also provides corresponding improved solutions. In subsequent embodiments, the same steps or corresponding steps as those in the above embodiment can be referenced to each other, and the corresponding beneficial effects can also be referenced to each other, and will not be described in detail in the following improved embodiments.

[0201] See also Figure 2 , Figure 2 This is another implementation flow chart of the low power factor power load classification method in an embodiment of the present application. The method may include the following steps:

[0202] S201: Acquire sample power load data, and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category.

[0203] In order to train the target power load classification model, it is necessary to accurately label the sample power load data with load category labels. For example, by classifying by occurrence time, peak load, minimum load, average load, etc. can be labeled, that is, the sample power load data needs to be accurately clustered.

[0204] Sample power load data should cover power load data under low power factors across various industries, types, and attributes. For example, the power load conditions are first determined based on dimensions such as time, region, and substation type. Then, a sampling method is used to select sample power load data from the massive power load data. For example, some of the sample power load data comes from a certain area, where low power factor power load data for 441 households over a period of one month is recorded every 15 minutes from 00:00 to 24:00 daily. Therefore, there are 96 data sampling points per day, forming a daily load matrix of sample power load data with a size of 441×96.

[0205] In a specific embodiment of the present application, clustering the sample power load data to obtain the sample power load data corresponding to each load category may include the following steps:

[0206] Step 1: Determine the initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as the current centroid;

[0207] Step 2: for each sample power load data, respectively calculate the first distance between the sample power load data and the current centroid corresponding to each load category cluster;

[0208] Step 3: Divide the sample power load data into load category clusters corresponding to the minimum first distance among the corresponding first distances;

[0209] Step 4: Determine the updated centroid corresponding to each load category cluster based on the sample power load data in each load category cluster currently divided;

[0210] Step 5: Determine whether the updated centroids are consistent with the current centroids. If so, proceed to step 6; if not, proceed to step 7.

[0211] Step 6: Determine the sample power load data currently divided into each load category cluster as the sample power load data corresponding to each load category;

[0212] Step 7: Determine each updated centroid as the current centroid and return to step 2.

[0213] For the convenience of description, the above seven steps can be combined for explanation.

[0214] The process of clustering the sample power load data may include determining the initial centroids corresponding to a preset number of load category clusters, and determining each initial centroid as each current centroid, calculating for each sample power load data the first distance between the sample power load data and the current centroid corresponding to each load category cluster, dividing the sample power load data into the load category cluster corresponding to the minimum first distance among the corresponding first distances, determining the updated centroid corresponding to each load category cluster based on the sample power load data in each load category cluster currently divided, and judging whether each updated centroid corresponds to each current centroid; if so, determining the sample power load data currently divided into each load category cluster as the sample power load data corresponding to each load category; if not, determining each updated centroid as each current centroid, and recalculating for each sample power load data the first distance between the sample power load data and the current centroid corresponding to each load category cluster for re-clustering.

[0215] The process of clustering the sample power load data can include first determining the load characteristic vector corresponding to each sample power load data, calculating the distance between each sample power load and the initial centroid based on the load characteristic vector and the centroid vector of the initial centroid corresponding to each load category cluster, and assigning each sample power load data to the load category cluster corresponding to the centroid vector with the closest distance, then recalculating the centroid of each load category cluster and its corresponding centroid vector for each load category cluster, and re-dividing the sample power load data into different load category clusters, thereby continuously dividing the sample power load data until the position of the centroid does not change, that is, the centroid vector does not change, and finally determining the sample power load data divided into different load category clusters as sample power load data under different load categories. Then, labeling each sample power load data with different categories as category labels, and finally forming a sample data set from each sample power load data labeled with the category labels. By clustering the sample power load data, the accuracy of the division of the sample power load data corresponding to each load category is greatly improved.

[0216] In a specific embodiment of the present application, before determining the initial centroids corresponding to the preset number of load category clusters, the method may further include the following steps:

[0217] Step 1: Get multiple cluster quantity values;

[0218] Step 2: For each cluster quantity value, determine the centroid corresponding to each initial load category cluster according to the cluster quantity value, and divide each sample power load data into each initial load category cluster to obtain the sample power load data corresponding to each initial load category cluster;

[0219] Step 3: Calculate the sum of the squares of the distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, and obtain the evaluation parameter of the cluster quantity value;

[0220] Step 4: Draw an evaluation parameter curve based on the number of clusters and the evaluation parameters corresponding to each cluster number;

[0221] Step 5: for each cluster quantity value in the evaluation parameter curve, calculate the change rate of the evaluation parameter corresponding to the cluster quantity value compared to the evaluation parameter corresponding to the nearest neighbor forward cluster quantity value;

[0222] Step 6: Select the first change rate whose absolute value is less than a preset change rate threshold from the change rates of the evaluation parameters corresponding to each cluster number value in the evaluation parameter curve;

[0223] Step 7: Determine the cluster quantity value corresponding to the selected change rate as the preset quantity.

[0224] For the convenience of description, the above seven steps can be combined for explanation.

[0225] Before determining the initial centroids corresponding to the preset number of load category clusters, it is necessary to determine the number of load category clusters. The process of determining the number of load category clusters may include obtaining multiple cluster quantity values for each cluster quantity value, determining the centroid corresponding to each initial load category cluster according to the cluster quantity value, dividing each sample power load data into each initial load category cluster, obtaining the sample power load data corresponding to each initial load category cluster, calculating the sum of the squares of the distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, obtaining an evaluation parameter of the cluster quantity value, and drawing an evaluation parameter curve according to each cluster quantity value and the evaluation parameter corresponding to each cluster quantity value. A change rate threshold is preset, and for each cluster quantity value in the evaluation parameter curve, a change rate of the evaluation parameter corresponding to the cluster quantity value compared to the evaluation parameter corresponding to the nearest neighbor forward cluster quantity value is calculated, and from the change rates of the evaluation parameters corresponding to each cluster quantity value in the evaluation parameter curve, the first change rate whose absolute value is less than the preset change rate threshold is selected, and the cluster quantity value corresponding to the selected change rate is determined as the preset number.

[0226] The process of determining the number of load category clusters can specifically include determining the number of clusters, including multiple values for the number of clusters, such as 1, 2, 3, 4, ..., and n. Specifically, a series of possible n values are first selected as candidates. These n values typically start from 1 and gradually increase to a reasonable upper limit, which can be determined based on the data size and the expected number of clusters. For each n value, the sum of the squares of the distances from each sample power load data point to the center (centroid) of the load category cluster to which it belongs is calculated, i.e., the evaluation parameter. The evaluation parameter is an important indicator of clustering effectiveness, reflecting the degree of internal compactness within each cluster after clustering. Furthermore, the evaluation parameter corresponding to each n value is plotted on an evaluation parameter curve, with the horizontal axis representing the n value and the vertical axis representing the evaluation parameter. As the n value increases, the evaluation parameter generally exhibits a decreasing trend, as more clusters better fit the data. By observing the evaluation parameter curve, an n value is sought where increasing the n value no longer significantly reduces the evaluation parameter. The inflection point corresponding to this n value marks the point where the rate of decline of the evaluation parameter begins to slow. The n value at which this inflection point occurs is considered the optimal number of clusters, i.e., the preset number. This preset number can ensure the clustering effect while avoiding the increase in computational complexity and noise sensitivity caused by excessive clustering.

[0227] In a specific embodiment of the present application, determining the initial centroids corresponding to a preset number of load category clusters may include the following steps:

[0228] Step 1: Determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density;

[0229] Step 2: Divide the sample power load data into high-density sets and low-density sets according to the density mean;

[0230] Step 3: determining high-density sample power load data having a data density greater than a preset density threshold from the sample power load data in the high-density set;

[0231] Step 4: Determine a preset number of central sample power load data with a mutual distance greater than a preset distance threshold in each high-density sample power load data, and determine each central sample power load data as the initial centroid corresponding to the preset number of load category clusters.

[0232] For the convenience of description, the above four steps can be combined for explanation.

[0233] The process of determining the initial centroids corresponding to the preset number of load category clusters may include determining the data density within a preset neighborhood corresponding to each sample power load data, determining a density mean corresponding to each data density, dividing each sample power load data into a high-density set and a low-density set based on the density mean, and determining high-density sample power load data having a data density greater than a preset density threshold from the sample power load data in the high-density set. A preset number of central sample power load data having a mutual distance greater than a preset distance threshold are determined from each high-density sample power load data, and each central sample power load data is determined as the initial centroid corresponding to the preset number of load category clusters.

[0234] The preset neighborhood, preset density threshold and preset distance threshold can be set according to actual needs. For example, the preset neighborhood corresponding to each sample power load data can be predetermined, and the load category cluster separation degree statistics can be performed by adjusting the preset radius, where , a preset radius with the best load category cluster separation is selected from the adjusted preset radii, so as to determine the preset neighborhood corresponding to each sample power load data according to the selected preset radius. Then, the data density within the preset neighborhood corresponding to each sample power load data is determined respectively. For example, the data density corresponding to sample power load data 1 is 3, the data density corresponding to sample power load data 2 is 4, the data density corresponding to sample power load data 3 is 5, the data density corresponding to sample power load data 4 is 9, the data density corresponding to sample power load data 5 is 9, and the data density corresponding to sample power load data 6 is 10. The density mean is 6.7, and the sample power load data 4, sample power load data 5, and sample power load data 6 with data density greater than 6.7 are divided into a high-density set, and the sample power load data 1, sample power load data 2, and sample power load data 3 are divided into a low-density set. If the preset density threshold is 5 and the preset number is 2, it is finally determined that the data whose mutual distance is greater than the preset threshold are sample power load data 4 and sample power load data 6. Finally, the sample power load data 4 and the sample power load data 6 are respectively determined as the initial centroids corresponding to each load category cluster.

[0235] The embodiment of the present application determines the initial centroid of the load category cluster by data density and mutual heterogeneity. By considering data density, the natural distribution characteristics of the data can be more accurately reflected. High-density areas usually correspond to the core part of the data set, while low-density areas may contain noise or outliers. Therefore, selecting the initial centroid based on data density helps to place the centroid at the core of the data, thereby improving the accuracy of the initial centroid determination and thus improving the clustering accuracy. At the same time, mutual heterogeneity measures the degree of difference between data objects. By considering mutual heterogeneity, the subtle differences between data objects can be better captured, thereby more accurately determining the initial centroid of the cluster. When the number of load category clusters and the initial centroid are accurately determined, the clustering effect can be further improved.

[0236] In a specific embodiment of the present application, after obtaining the sample power load data corresponding to each load category, the method may further include the following steps:

[0237] Step 1: For each sample power load data, calculate the intra-class average distance between the sample power load data and other sample power load data in the load category to which the sample power load data belongs;

[0238] Step 2: Determine the reference sample power load data closest to the sample power load data from each of the sample power load data included in the load category other than the load category to which the sample power load data belongs;

[0239] Step 3: Determine the average inter-class distance between the sample power load data and the corresponding reference sample power load data;

[0240] Step 4: Determine a first clustering effect evaluation parameter for clustering the sample power load data based on the intra-class average distance and the inter-class average distance corresponding to each sample power load data;

[0241] Step 5: Determine the data volume of the sample power load data and the number of categories of each load category, and determine the category center point of each load category, the data center point of the sample power load data, and the intra-category data volume of the sample power load data in each load category;

[0242] Step 6: Determine the inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and determine the intra-class covariance matrix based on the center point of the class and each load class;

[0243] Step 7: Determine the second clustering effect evaluation parameter based on the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of categories;

[0244] Step 8: Calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion corresponding to each load category;

[0245] Step 9: Calculate the distance between the centroids of any two load categories in each load category to obtain the inter-class dispersion of the two load categories.

[0246] Step 10: Sum the intra-class dispersion corresponding to the two load categories to obtain the intra-class dispersion evaluation parameters corresponding to the two load categories;

[0247] Step 11: Calculate the quotient of the intra-class dispersion evaluation parameter and the inter-class dispersion corresponding to the two load categories to obtain the category similarity between the two load categories;

[0248] Step 12: Determine the maximum category similarity corresponding to each load category based on the category similarity between each load category and other load categories;

[0249] Step 13: Calculate the mean of the similarities of each maximum category to obtain the third clustering effect evaluation parameter;

[0250] Step 14: respectively determining the weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter, and performing weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain the load clustering effect evaluation parameter;

[0251] Step 15: judging whether the clustering result of the sample power load data meets the preset clustering condition according to the load clustering effect evaluation parameter; if so, executing step 16; if not, executing step 17;

[0252] Step 16: Use each load category as the labeling information of the sample power load data contained in each load category.

[0253] And construct a sample data set based on the sample power load data with labeled information corresponding to each load category;

[0254] Step 17: Re-cluster the sample power load data.

[0255] For the convenience of description, the above seventeen steps can be combined for explanation.

[0256] After obtaining the sample power load data corresponding to each load category, for each sample power load data, the intra-class average distance between the sample power load data and other sample power load data in the load category to which the sample power load data belongs is calculated, and the reference sample power load data closest to the sample power load data is determined among the sample power load data contained in the other load categories except the load category to which the sample power load data belongs. The inter-class average distance between the sample power load data and the corresponding reference sample power load data is determined, and the first clustering effect evaluation parameter for clustering the sample power load data is determined based on the intra-class average distance and the inter-class average distance corresponding to each sample power load data.

[0257] The process of determining the first clustering effect evaluation parameter may include assuming that there is a load category cluster A, which includes two samples, sample 1 and sample 2 belong to load category cluster A, and sample 3 and sample 4 belong to load category cluster B.

[0258] The average distance between sample 1 and other samples in load category cluster A is 1, and the average distance between sample 1 and the closest sample in load category cluster B is 2.

[0259] The average distance between sample 2 and other samples in load category cluster A is 2, and the average distance between sample 2 and the closest sample in load category cluster B is 3.

[0260] The average distance between sample 3 and other samples in load category cluster B is 1, and the average distance between sample 3 and the closest sample in load category cluster A is 2.

[0261] The average distance between sample 4 and other samples in load category cluster B is 2, and the average distance between sample 4 and the closest sample in load category cluster A is 3.

[0262] The first clustering effect evaluation parameter for sample 1 is (1 - 2) / 2 = -0.5;

[0263] The first clustering effect evaluation parameter for sample 2 is (2 - 3) / 3 = -0.333;

[0264] The first clustering effect evaluation parameter for sample 3 is (1 - 2) / 2 = -0.5;

[0265] The first clustering effect evaluation parameter of sample 4 is (2 - 3) / 3 = -0.333.

[0266] Determine the amount of sample power load data and the number of categories for each load category, and determine the category center point of each load category, the center point of the sample power load data, and the amount of data within each category of the sample power load data in each load category. Determine the inter-class covariance matrix based on the category center point, the center point, and the amount of data within each category. Determine the intra-class covariance matrix based on the category center point and each load category. Determine the second clustering effect evaluation parameter based on the inter-class covariance matrix, the intra-class covariance matrix, the amount of sample power load data, and the number of categories. Calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs to obtain the intra-class dispersion corresponding to each load category. Calculate the distance between the category centroids corresponding to any two load categories in each load category to obtain the inter-class dispersion corresponding to the two load categories. Sum the intra-class dispersion corresponding to the two load categories to obtain the intra-class dispersion evaluation parameter corresponding to the two load categories. The quotient of the intra-class dispersion evaluation parameters corresponding to the two load categories and the inter-class dispersion evaluation parameters corresponding to the two load categories is calculated to obtain the class similarity between the two load categories. Based on the class similarity between each load category and the other load categories, the maximum class similarity corresponding to each load category is determined. The mean of each maximum class similarity is calculated to obtain a third clustering effect evaluation parameter. Weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter are respectively determined. Based on the weight coefficients, the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter are weighted to obtain the load clustering effect evaluation parameter. Based on the load clustering effect evaluation parameter, it is determined whether the clustering results of the sample power load data meet the preset clustering conditions. If so, each load category is used as the labeling information of the sample power load data contained in each load category, so that a sample data set can be constructed based on the sample power load data with the labeling information corresponding to each load category. If not, the sample power load data is clustered again.

[0267] The weight coefficient can be set according to actual needs. Specifically, the closeness between each sample power load data in the same load category cluster and the distance between the sample power load data in different load category clusters are determined, and the first clustering effect evaluation parameter is determined by combining the closeness and distance. The value of the first clustering effect evaluation parameter is between -1 and 1. The closer the value is to 1, the better the clustering effect is, that is, the sample power load data is close to other sample power load data in its load category cluster and distant from the sample power load data in other load category clusters; the closer the value is to -1, the worse the clustering effect is, that is, the sample power load data may be incorrectly assigned to the current load category cluster; the value close to 0 indicates that the sample power load data is located on the boundary of two load category clusters.

[0268] For each cluster, the covariance matrix (W) of the intra-class data is calculated to measure the compactness of the intra-class data, and the sum of the squares of the distances between the centroids of all classes and the center point of the sample power load data set is calculated to obtain the sum of the squares of the inter-class distances (B). Based on the sum of the squares of the inter-class distances, the inter-class covariance matrix is determined to measure the separation between clusters. Then, according to the formula Calculate the H value (the second clustering effect evaluation parameter), where n is the number of clusters, that is, the number of categories, and m is the data volume of the sample power load data. The larger the H value, the better the clustering effect.

[0269] For any two clusters C and D, calculate the sum of the average intra-cluster distances of clusters C and D. , where Si and Sj are the distances from the sample power load data in class C and class D to the class centroid, respectively. The distance between the class centroids of class C and class D (d(C, D)) is calculated, and then according to the formula Calculate the Y value (the third clustering effect evaluation parameter) and find the maximum value for all cluster pairs. The smaller the Y value, the better the clustering effect.

[0270] The embodiment of the present application evaluates the clustering effect by comprehensively analyzing the evaluation results of multiple evaluation methods. It can quantify the clustering effect from different angles and comprehensively reflect the overall quality of the clustering, including the compactness, separation, consistency and difference of the load category clusters, thereby improving the evaluation accuracy of the clustering effect.

[0271] S202: Construct a sample data set based on the sample power load data corresponding to each load category.

[0272] After clustering to obtain sample power load data corresponding to each load category, a sample data set is constructed based on the sample power load data corresponding to each load category.

[0273] S203: Divide the sample data set into a training data set and a test data set.

[0274] After constructing the sample data set, the sample data set is divided into a training data set and a test data set.

[0275] S204: Using the training data set to train the pre-built initial power load classification model, and using the test data set to test the trained initial power load classification model to obtain a target power load classification model.

[0276] After the training data set and the test data set are divided, the pre-built initial power load classification model is trained using the training data set, and the trained initial power load classification model is tested using the test data set to obtain the target power load classification model.

[0277] Assuming the model test requirement is a test accuracy of 98%, the power load classification model with a test accuracy greater than or equal to 98% is ultimately determined as the target power load classification model. If the test accuracy does not meet the requirement, the model parameters of the power load classification model are iteratively trained based on the loss function until the test accuracy meets the requirement or the number of iterations meets the requirement.

[0278] The embodiment of the present application uses clustering to classify the sample power load data, which can improve the accuracy of the classification of the sample power load data, and then train the model based on the sample power load data with higher accuracy to obtain the target power load classification model, thereby improving the prediction accuracy of the target power load classification model. In the process of training the power load classification model, based on the optimal hyperplane, the margin between the classes is maximized. For linearly inseparable problems, if the training sample is

[0279] , the optimal hyperplane can be expressed as follows:

[0280] ;

[0281] in, For the mapping function and inner product, you can choose a suitable kernel function to support different nonlinear mapping methods. X is the training sample, x i represents the i-th training sample, y i represents the label of the i-th training sample, and w is the normal vector of the hyperplane. To support different nonlinear mapping methods. The hinge loss function is introduced into the optimal hyperplane classification margin minimization problem as follows:

[0282] ;

[0283] ;

[0284] in, , represents Hinge loss, which is used to quantify the degree of misclassification of samples. C is a hyperparameter that controls the degree of error allowed by the model. The larger C is, the greater the error allowed. The smaller , the stricter the model, and vice versa. b is the bias term, and m is the total number of sample data. The dataset is split into training and test sets, and the radial basis function is selected as the kernel function to train the model.

[0285] S205: Acquire power load data corresponding to the power load to be classified generated under a low power factor condition.

[0286] The power load data includes multiple categories of power load related data.

[0287] S206: For each type of power load related data, draw a load data curve that changes with time.

[0288] After obtaining the power load data corresponding to the power load to be classified generated under the low power factor condition, a load data curve that changes with time is drawn for each type of power load related data.

[0289] Different types of power load data correspond to different load data curves, such as voltage data corresponding to voltage data curves, current data corresponding to current data curves, low power factor data corresponding to low power factor data curves, device attribute data corresponding to device attribute data curves, environmental attribute data corresponding to environmental attribute data curves, and user characteristic data corresponding to user characteristic data curves. Furthermore, based on current, voltage, and other data, load values can be calculated and load value curves can be plotted. The horizontal axis of each curve represents time, and the vertical axis represents the corresponding data value. For example, the horizontal axis of the load value curve represents time, and the vertical axis represents the load value corresponding to different time points. The horizontal axis of the voltage data curve represents time, and the vertical axis represents the voltage value corresponding to different time points.

[0290] S207: Pre-processing each power load data according to each load data curve.

[0291] After drawing the load data curve that changes with time, the power load data are preprocessed according to each load data curve, such as outlier detection, redundant data deletion, abnormal data correction, missing value supplementation, normalization and other preprocessing operations can be performed on each power load data.

[0292] The process of preprocessing each power load data according to each load data curve may include preprocessing the voltage data based on the voltage data curve to obtain preprocessed voltage data; preprocessing the current data based on the current data curve to obtain preprocessed current data; preprocessing the low power factor data based on the low power factor data curve to obtain preprocessed low power factor data; preprocessing the device attribute data based on the device attribute data curve to obtain preprocessed device attribute data; preprocessing the environment attribute data based on the environment attribute data curve to obtain preprocessed environment attribute data; preprocessing the user feature data based on the user feature data curve to obtain preprocessed user feature data. In the process of power load classification, the embodiment of the present application can correct errors and abnormal values in the data by preprocessing various power load data, making the data more accurate and reliable, thereby improving the accuracy of subsequent power load classification. At the same time, the preprocessing can also process redundant information in the data, avoiding the computing resources and time wasted by computing redundant information, so that the embodiment of the present application can also save computing resources and improve the efficiency of power load classification.

[0293] In a specific implementation of the present application, step S207 may include the following steps:

[0294] Step 1: For each load data curve, each collection time point in the load data curve is used as the target collection time point;

[0295] Step 2: Determine the load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point;

[0296] Step 3: Determine whether the load data change rate is greater than a first preset threshold, if so, execute step 4, if not, execute step 5;

[0297] Step 4: Determine whether the target load data corresponding to the target collection time point is abnormal data;

[0298] Step 5: Determine whether the target load data corresponding to the target collection time point is normal data;

[0299] Step 6: For each abnormal data, select each forward normal load data before the abnormal data and each backward normal load data after the abnormal data from the load data curve to which the abnormal data belongs;

[0300] Step 7: Correct the abnormal data according to the forward normal load data and the backward normal load data corresponding to the abnormal data;

[0301] Step 8: normalize the load data contained in the load data curve after abnormal data correction.

[0302] For the convenience of description, the above eight steps can be combined for explanation.

[0303] The preprocessing process for each power load data may include, for each load data curve, taking each collection time point in the load data curve as a target collection time point, determining the load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point, determining whether the load data change rate is greater than a first preset threshold, and if so, determining that the target load data corresponding to the target collection time point is abnormal data; if not, determining that the target load data corresponding to the target collection time point is normal data. For each abnormal data, selecting each forward normal load data preceding the abnormal data and each backward normal load data following the abnormal data from the load data curve to which the abnormal data belongs, correcting the abnormal data based on the forward normal load data and each backward normal load data corresponding to the abnormal data, and normalizing each load data included in the load data curve after the abnormal data correction. By preprocessing various power load data, errors and outliers in the data can be corrected, making the data more accurate and reliable, thereby improving the accuracy of subsequent power load classification.

[0304] The preset threshold can be set according to actual needs, for example, the preset threshold can be any value between 0.5 and 0.8. Specifically, abnormal data can be detected by the following load data change rate formula:

[0305] ;

[0306] in, Load data curve The load data change rate of the data at point i, Load data curve The power load data at point i+1 on the Load data curve The power load data at point i on .

[0307] The above formula can be used to calculate the load data change rate corresponding to any point on each load data curve. When the load data change rate at a certain point is greater than a preset threshold, the data at that point is determined to be abnormal data, thereby enabling abnormality detection of the data corresponding to each point in each load data curve. The number of abnormal data in each load data curve can also be counted separately. If the number of abnormal data in a load data curve is greater than a preset number threshold, all load data corresponding to that load data curve is deleted. That is, during the load classification process, the power load data corresponding to that load curve is not considered. Conversely, if the number of abnormal data in a load data curve is less than or equal to the preset number threshold, the abnormal data in that load data curve can be corrected.

[0308] The load data can be normalized according to the following formula:

[0309]

[0310] in, is the normalized power load data, that is, the pre-processed power load data, Power load curve The power load data at position i, Power load curve The minimum power load data in Power load curve The maximum power load data in.

[0311] It should be noted that the preset quantity threshold can be set according to actual needs, and the embodiments of the present application do not limit this.

[0312] In a specific embodiment of the present application, correcting abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data may include the following steps:

[0313] Step 1: Count the total number of forward load data and the total number of backward load data corresponding to the abnormal data;

[0314] Step 2: Sum the forward normal load data corresponding to the abnormal data to obtain the sum of the forward load data corresponding to the abnormal data;

[0315] Step 3: Sum the backward normal load data corresponding to the abnormal data to obtain the sum of the backward load data corresponding to the abnormal data;

[0316] Step 4: Sum the total number of forward load data and the total number of backward load data corresponding to the abnormal data to obtain a reference data amount correction parameter corresponding to the abnormal data;

[0317] Step 5: Sum the forward load data and the backward load data corresponding to the abnormal data to obtain the reference data and correction parameters corresponding to the abnormal data;

[0318] Step 6: Calculate the quotient of the reference data and correction parameter corresponding to the abnormal data and the reference data correction parameter to obtain the correction data corresponding to the abnormal data, and replace the abnormal data with the correction data corresponding to the abnormal data.

[0319] For the convenience of description, the above six steps can be combined for explanation.

[0320] When correcting abnormal data, the total number of forward load data and the total number of backward load data corresponding to the abnormal data are counted, and each forward normal load data corresponding to the abnormal data is summed to obtain the forward load data sum corresponding to the abnormal data. The each backward normal load data corresponding to the abnormal data is summed to obtain the backward load data sum corresponding to the abnormal data. The total number of forward load data and the total number of backward load data corresponding to the abnormal data are summed to obtain a reference data amount correction parameter corresponding to the abnormal data. The forward load data sum and the backward load data sum corresponding to the abnormal data are summed to obtain a reference data and correction parameter corresponding to the abnormal data. The reference data and correction parameter corresponding to the abnormal data are quotient-calculated with the reference data amount correction parameter to obtain corrected data corresponding to the abnormal data, and the corrected data corresponding to the abnormal data are used to replace the abnormal data.

[0321] The correction value can be calculated according to the following formula:

[0322] ;

[0323] in, Power load data curve The correction value corresponding to the abnormal data at point i, u is the normal data points in the power load data curve that are adjacent to the corrected abnormal data and before the abnormal data, that is, the points corresponding to the forward normal load data, v is the normal data points in the power load data curve that are adjacent to the corrected abnormal data and after the abnormal data, that is, the points corresponding to the backward normal load data, u1 is the total number of points corresponding to the forward normal load data, that is, the total number of forward load data, v1 is the total number of points corresponding to the backward normal load data, that is, the total number of backward load data. In this embodiment of the present application, u1 and v1 can be taken according to actual needs, such as 3, that is, according to the 3 normal data before the abnormal data point and the 3 normal data after the abnormal data point.

[0324] The above formula can be used to calculate the correction value corresponding to the abnormal data at each point in each load data curve. The correction value is then used to replace the corresponding abnormal data to obtain the corrected power load data. The abnormal data is corrected using the corresponding forward and backward normal load data, greatly improving the accuracy of the abnormal data correction.

[0325] S208: Perform dimensionality reduction on the power load data to obtain dimensionality-reduced power load data.

[0326] In a specific implementation of the present application, step S208 may include the following steps:

[0327] Step 1: Determine the target Gaussian distribution variance corresponding to the power load data;

[0328] Step 2: For each power load data, calculate the target similarity between the power load data and other power load data according to the target Gaussian distribution variance;

[0329] Step 3: Calculate the target joint probability distribution of the power load data according to the similarity of each target corresponding to the power load data;

[0330] Step 4: Randomly determine the initial power load data in the low-dimensional space;

[0331] Step 5: Determine the initial Gaussian distribution variance corresponding to the initial power load data;

[0332] Step 6: For each initial power load data, calculate the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance;

[0333] Step 7: Calculate the initial joint probability distribution of the initial power load data according to the initial similarities corresponding to the initial power load data;

[0334] Step 8: Calculate the quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain a joint probability evaluation function of the power load data;

[0335] Step 9: Determine the loss function of the joint probability distribution between the power load data and its corresponding initial power load data based on the joint probability evaluation function of the power load data and the target joint probability distribution;

[0336] Step 10: Determine whether the loss function between the power load data and its corresponding initial power load data meets a second preset threshold; if so, execute step 11; if not, return to step 4;

[0337] Step 11: Determine the initial power load data corresponding to the power load data as the power load data after dimension reduction corresponding to the power load data.

[0338] For the convenience of description, the above eleven steps can be combined for explanation.

[0339] The process of reducing the dimensionality of the power load data may include determining the target Gaussian distribution variance corresponding to the power load data, calculating the target similarity between the power load data and other power load data according to the target Gaussian distribution variance for each power load data, calculating the target joint probability distribution of the power load data according to each target similarity corresponding to the power load data, randomly determining the initial power load data in the low-dimensional space, determining the initial Gaussian distribution variance corresponding to the initial power load data, calculating the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance for each initial power load data, and calculating the initial power load probability distribution according to each initial similarity corresponding to the initial power load data. The target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data are calculated to obtain a joint probability evaluation function of the power load data. According to the joint probability evaluation function of the power load data and the target joint probability distribution, a loss function of the joint probability distribution between the power load data and the corresponding initial power load data is determined. It is judged whether the loss function between the power load data and the corresponding initial power load data meets a second preset threshold value. If so, the initial power load data corresponding to the power load data is determined as the power load data after dimensionality reduction corresponding to the power load data. If not, the initial power load data of the power load data in the low-dimensional space is randomly re-determined.

[0340] The process of reducing the dimensionality of power load data can specifically include reducing the dimensionality of voltage data to obtain reduced-dimensional voltage data; reducing the dimensionality of current data to obtain reduced-dimensional current data; reducing the dimensionality of low power factor data to obtain reduced-dimensional low power factor data; reducing the dimensionality of device attribute data to obtain reduced-dimensional device attribute data; reducing the dimensionality of environmental attribute data to obtain reduced-dimensional environmental attribute data; reducing the dimensionality of user feature data to obtain reduced-dimensional user feature data.

[0341] See also Figure 3 , Figure 3 The process of reducing the dimensionality of the power load data may include: first randomly determining the power load data Initial power load data in low-dimensional space , where m represents the dimension of the power load data and the initial power load data, Indicates the first initial power load data, represents the first power load data, and then the similarity between each power load data and other power load data is calculated according to the following formula:

[0342] ;

[0343] in, is the i-th power load data With the jth power load data The similarity between Power load data The corresponding Gaussian distribution variance, is the kth power load data, and exp() represents the similarity calculation function.

[0344] In this way, the similarity between each power load data and other power load data can be calculated in the above-mentioned manner.

[0345] Furthermore, the joint probability distribution of power load data is calculated according to the following formula:

[0346] ;

[0347] in, is the i-th power load data With the jth power load data The joint probability distribution between is the jth power load data With the i-th power load data The similarity between them is m, and m is the number of power load data.

[0348] In this way, the joint probability distribution between the power load data can be determined in the above manner.

[0349] The initial joint probability distribution of the initial power load data can also be calculated according to the following formula:

[0350] ;

[0351] in, is the initial joint probability distribution between initial load data i and initial load data j, w i is the i-th initial load data, w j is the jth initial load data, w k is the kth initial load data.

[0352] In this way, the initial joint probability distribution between the initial load data can be calculated in the above-mentioned manner.

[0353] The loss function of the joint probability distribution can also be calculated according to the following formula:

[0354] ;

[0355] Among them, KL divergence measures the difference between two probability distributions, h is the joint probability distribution between the power load data, The KL regression (KL) is the initial joint probability distribution between the initial load data. It measures the difference in the similarity distribution between data points in the high-dimensional space and the low-dimensional space. The goal is to minimize this difference, thereby preserving the structural characteristics of the data points in the high-dimensional space as much as possible in the low-dimensional space. By optimizing this process, high-dimensional data can be effectively mapped to the low-dimensional space while preserving the local and global structural characteristics of the data as much as possible. Finally, gradient descent is used to iteratively optimize and minimize the loss function to obtain the final low-dimensional space dataset.

[0356] The loss function is calculated as follows:

[0357] ;

[0358] Among them, loss is the loss function.

[0359] Based on the loss function, the position information of the initial power load data in the low-dimensional space is iteratively optimized until the loss function converges to a preset threshold. The initial power load data at the time when the loss function converges to the preset threshold is used as the power load data after dimensionality reduction. Since the sparsity and computational complexity of the data increase significantly with the increase of data dimension, causing the effectiveness of many algorithms to drop sharply, the embodiments of the present application can reduce the computational complexity and avoid the interference of noise in high-dimensional data by performing dimensionality reduction processing on the original power load data, thereby improving the accuracy of load classification.

[0360] S209: Using the pre-trained target power load classification model to classify the power load to be classified according to the power load data after dimensionality reduction, to obtain a target load category.

[0361] Corresponding to the above method embodiment, the present application also provides a low power factor electric load classification device. The low power factor electric load classification device described below and the low power factor electric load classification method described above can refer to each other.

[0362] See also Figure 4 , Figure 4 This is a structural block diagram of a low power factor power load classification device in an embodiment of the present application. The device may include:

[0363] The power load data acquisition module 41 is used to acquire power load data corresponding to the power load to be classified generated under low power factor conditions; wherein the power load data includes power load related data of multiple categories;

[0364] A module 42 for obtaining power load data after dimensionality reduction is used to reduce the dimensionality of the power load data to obtain power load data after dimensionality reduction;

[0365] The target load category obtaining module 43 is used to classify the unclassified power load according to the power load data after dimensionality reduction using the pre-trained target power load classification model to obtain the target load category.

[0366] The above technical solution demonstrates that by comprehensively analyzing multiple categories of power load data to classify low-power-factor loads, the accuracy of load classification is improved. By reducing the dimensionality of load data, the computational effort is reduced while retaining important information, significantly improving the efficiency of load classification. By utilizing a pre-trained target load classification model to classify loads, both efficiency and accuracy are further improved.

[0367] In a specific embodiment of the present application, the device may further include:

[0368] A curve drawing module is used to draw a load data curve that changes with time for each type of power load related data after obtaining power load data corresponding to the power load to be classified generated under low power factor conditions and before performing dimensionality reduction on the power load data;

[0369] The preprocessing module is used to preprocess each power load data according to each load data curve.

[0370] In a specific embodiment of the present application, the pre-processing module may include:

[0371] The target collection time point determination submodule is used to determine each collection time point in the load data curve as the target collection time point for each load data curve;

[0372] a load data change rate determination submodule, configured to determine the load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point;

[0373] A first judgment submodule is used to judge whether the load data change rate is greater than a first preset threshold;

[0374] an abnormal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is abnormal data when it is determined that the load data change rate is greater than a first preset threshold;

[0375] a normal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is normal data when it is determined that the load data change rate is not greater than a first preset threshold;

[0376] A data selection submodule is used to select, for each abnormal data, each forward normal load data before the abnormal data and each backward normal load data after the abnormal data from the load data curve to which the abnormal data belongs;

[0377] A data correction submodule, configured to correct the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data;

[0378] The normalization processing submodule is used to perform normalization processing on each load data included in the load data curve after abnormal data correction.

[0379] In a specific embodiment of the present application, the data correction submodule may include:

[0380] A quantity statistics unit is used to count the total number of forward load data and the total number of backward load data corresponding to the abnormal data;

[0381] A forward load data sum obtaining unit is used to sum each forward normal load data corresponding to the abnormal data to obtain the forward load data sum corresponding to the abnormal data;

[0382] A backward load data sum obtaining unit is used to sum and calculate each backward normal load data corresponding to the abnormal data to obtain the backward load data sum corresponding to the abnormal data;

[0383] a data volume correction parameter obtaining unit, configured to sum the total amount of forward load data and the total amount of backward load data corresponding to the abnormal data to obtain a reference data volume correction parameter corresponding to the abnormal data;

[0384] a data and correction parameter obtaining unit, configured to calculate the sum of the forward load data and the backward load data corresponding to the abnormal data, and obtain reference data and correction parameters corresponding to the abnormal data;

[0385] The data replacement unit is used to calculate the quotient of the reference data and correction parameter corresponding to the abnormal data and the reference data amount correction parameter to obtain the correction data corresponding to the abnormal data, and replace the abnormal data with the correction data corresponding to the abnormal data.

[0386] In a specific embodiment of the present application, the module for obtaining power load data after dimensionality reduction may include:

[0387] A target Gaussian distribution variance determination submodule is used to determine the target Gaussian distribution variance corresponding to the power load data;

[0388] A target similarity calculation submodule is used to calculate the target similarity between each power load data and other power load data according to the target Gaussian distribution variance;

[0389] The target joint probability distribution calculation submodule is used to calculate the target joint probability distribution of power load data according to the similarity of each target corresponding to the power load data;

[0390] An initial power load data determination submodule, used for randomly determining initial power load data of power load data in a low-dimensional space;

[0391] An initial Gaussian distribution variance determination submodule is used to determine the initial Gaussian distribution variance corresponding to the initial power load data;

[0392] An initial similarity calculation submodule is used to calculate, for each initial power load data, the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance;

[0393] An initial joint probability distribution calculation submodule, configured to calculate an initial joint probability distribution of the initial power load data according to each initial similarity corresponding to the initial power load data;

[0394] A joint probability evaluation function obtaining submodule is used to calculate the quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain the joint probability evaluation function of the power load data;

[0395] A loss function determination submodule is used to determine the loss function of the joint probability distribution between the power load data and its corresponding initial power load data based on the joint probability evaluation function of the power load data and the target joint probability distribution;

[0396] A second judgment submodule is used to judge whether the loss function between the power load data and the corresponding initial power load data meets a second preset threshold;

[0397] a submodule for determining the power load data after dimensionality reduction, configured to determine the initial power load data corresponding to the power load data as the power load data after dimensionality reduction corresponding to the power load data when a loss function between the determined power load data and the initial power load data corresponding thereto satisfies a second preset threshold;

[0398] The repeated execution submodule is used to repeatedly execute the step of randomly determining the initial power load data of the power load data in the low-dimensional space when the loss function between the determined power load data and its corresponding initial power load data does not meet the second preset threshold.

[0399] In a specific embodiment of the present application, the device may further include a model training module, which may include:

[0400] The sample power load data acquisition submodule is used to obtain sample power load data and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category;

[0401] A sample data set construction submodule is used to construct a sample data set based on the sample power load data corresponding to each load category;

[0402] The training data set and test data set division submodule is used to divide the sample data set into a training data set and a test data set;

[0403] The target power load classification model acquisition submodule is used to train the pre-built initial power load classification model using the training data set, and test the trained initial power load classification model using the test data set to obtain the target power load classification model.

[0404] In a specific embodiment of the present application, the sample power load data acquisition submodule includes:

[0405] a current centroid determining unit, configured to determine initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as each current centroid;

[0406] A first distance calculation unit is used to calculate, for each sample power load data, a first distance between the sample power load data and a current centroid corresponding to each load category cluster;

[0407] a load category clustering unit, configured to divide the sample power load data into a load category cluster corresponding to the minimum first distance among the corresponding first distances;

[0408] An updated centroid determining unit, configured to determine the updated centroid corresponding to each load category cluster based on the sample power load data in each load category cluster currently divided;

[0409] A judgment unit, used to judge whether each updated centroid corresponds to each current centroid;

[0410] a sample power load data determining unit, configured to determine the sample power load data currently classified into each load category cluster as the sample power load data corresponding to each load category when it is determined that each updated centroid corresponds to each current centroid;

[0411] The repeated execution unit is used to determine each updated centroid as each current centroid when it is determined that the updated centroid does not correspond to each current centroid, and return to execute the step of calculating the first distance between the sample power load data and the current centroid corresponding to each load category cluster for each sample power load data.

[0412] In a specific embodiment of the present application, the device may further include:

[0413] A cluster quantity value acquisition module is used to obtain a plurality of cluster quantity values before determining the initial centroids corresponding to the preset number of load category clusters;

[0414] A sample power load data determination module is used to determine the centroid corresponding to each initial load category cluster according to the cluster number value for each cluster number, and to divide each sample power load data into each initial load category cluster to obtain the sample power load data corresponding to each initial load category cluster;

[0415] An evaluation parameter acquisition module is used to calculate the sum of squares of distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, and obtain an evaluation parameter for the cluster quantity value;

[0416] An evaluation parameter curve drawing module is used to draw an evaluation parameter curve according to the number of clusters and the evaluation parameters corresponding to the number of clusters;

[0417] A change rate calculation module is used to calculate the change rate of the evaluation parameter corresponding to each cluster number value in the evaluation parameter curve compared with the evaluation parameter corresponding to the nearest neighbor forward cluster number value;

[0418] A change rate selection module is used to select the first change rate whose absolute value is less than a preset change rate threshold from the change rates of the evaluation parameters corresponding to each cluster number value in the evaluation parameter curve;

[0419] The preset number determination module is used to determine the cluster number value corresponding to the selected change rate as the preset number.

[0420] In a specific embodiment of the present application, the current center of mass determination unit includes:

[0421] a density mean determination subunit, configured to determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density;

[0422] A high-density set and a low-density set division subunit is used to divide each sample power load data into a high-density set and a low-density set according to the density mean;

[0423] a high-density sample power load data determining subunit, configured to determine, from the sample power load data in the high-density set, high-density sample power load data having a data density greater than a preset density threshold;

[0424] The initial centroid determination subunit is used to determine a preset number of central sample power load data with a mutual distance greater than a preset distance threshold in each high-density sample power load data, and determine each central sample power load data as the initial centroid corresponding to a preset number of load category clusters.

[0425] In a specific embodiment of the present application, the device may further include:

[0426] An intra-class average distance calculation module is used to calculate, after obtaining the sample power load data corresponding to each load category, the intra-class average distance between each sample power load data and other sample power load data in the load category to which the sample power load data belongs;

[0427] A module for determining the nearest reference sample power load data, configured to determine the reference sample power load data closest to the sample power load data from among the sample power load data respectively included in the load categories other than the load category to which the sample power load data belongs;

[0428] an inter-class average distance determination module, configured to determine an inter-class average distance between the sample power load data and corresponding reference sample power load data;

[0429] a first clustering effect evaluation parameter determination module, configured to determine a first clustering effect evaluation parameter for clustering the sample power load data based on the intra-class average distance and the inter-class average distance corresponding to each sample power load data;

[0430] An intra-class data volume determination module is used to determine the data volume of the sample power load data and the number of categories of each load category, and to determine the category center point of each load category, the data center point of the sample power load data, and the intra-class data volume of the sample power load data in each load category;

[0431] The intra-class covariance matrix determination module is used to determine the inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and to determine the intra-class covariance matrix based on the center point of the class and each load class;

[0432] a second clustering effect evaluation parameter determination module, configured to determine a second clustering effect evaluation parameter based on the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of classes;

[0433] The intra-class dispersion degree acquisition module is used to calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion degree corresponding to each load category;

[0434] The inter-class dispersion degree obtaining module is used to calculate the distance between the class centroids corresponding to any two load classes in each load class, and obtain the inter-class dispersion degrees corresponding to the two load classes;

[0435] The intra-class dispersion evaluation parameter acquisition module is used to calculate the sum of the intra-class dispersions corresponding to the two load categories to obtain the intra-class dispersion evaluation parameters corresponding to the two load categories;

[0436] A category similarity obtaining module is used to calculate the quotient of the intra-category dispersion evaluation parameters corresponding to the two load categories and the inter-category dispersion corresponding to the two load categories to obtain the category similarity between the two load categories;

[0437] A maximum category similarity determination module is used to determine the maximum category similarity corresponding to each load category based on the category similarity between each load category and other load categories;

[0438] A third clustering effect evaluation parameter obtaining module is used to calculate the mean of the maximum category similarities to obtain the third clustering effect evaluation parameter;

[0439] a load clustering effect evaluation parameter acquisition module, used to determine the weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter, respectively, and perform weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain the load clustering effect evaluation parameter;

[0440] A judgment module is used to judge whether the clustering result of the sample power load data meets the preset clustering conditions according to the load clustering effect evaluation parameters;

[0441] a labeling information determining module, configured to, when determining that the clustering result of the sample power load data satisfies a preset clustering condition, use each load category as labeling information of the sample power load data included in each load category;

[0442] The sample data set construction submodule is specifically a module for constructing a sample data set based on the sample power load data with labeled information corresponding to each load category;

[0443] The re-clustering module is used to re-cluster the sample power load data when it is determined that the clustering result of the sample power load data does not meet the preset clustering condition.

[0444] Corresponding to the above method embodiment, see Figure 5 , Figure 5 This is a schematic diagram of a low power factor electrical load classification device provided by this application, which may include:

[0445] Memory 332, for storing computer programs;

[0446] The processor 322 is configured to implement the steps of the low power factor electric load classification method of the above method embodiment when executing the computer program.

[0447] For details, please refer to Figure 6 , Figure 6 This is a schematic diagram of the specific structure of a low-power-factor electric load classification device provided in this embodiment. This low-power-factor electric load classification device may vary significantly due to different configurations or performance. It may include a processor (central processing unit, CPU) 322 (e.g., one or more processors) and a memory 332. The memory 332 stores one or more computer programs 342 or data 344. The memory 332 may be either transient or persistent storage. The program stored in the memory 332 may include one or more modules (not shown), each of which may include a series of instruction operations in the data processing device. Furthermore, the processor 322 may be configured to communicate with the memory 332 to execute the series of instruction operations in the memory 332 on the low-power-factor electric load classification device 301.

[0448] The low power factor electrical load classification device 301 may further include one or more power supplies 326 , one or more wired or wireless network interfaces 350 , one or more input and output interfaces 358 , and / or one or more operating systems 341 .

[0449] The steps in the low power factor electric load classification method described above can be implemented by the structure of a low power factor electric load classification device.

[0450] Corresponding to the above method embodiment, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps can be implemented:

[0451] Obtain power load data corresponding to the power load to be classified generated under low power factor conditions; wherein the power load data includes power load-related data of multiple categories; perform dimensionality reduction on the power load data to obtain dimensionality-reduced power load data; use a pre-trained target power load classification model to classify the power load to be classified according to the dimensionality-reduced power load data to obtain a target load category.

[0452] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0453] For an introduction to the computer-readable storage medium provided in this application, please refer to the above method embodiment, and this application will not go into details here.

[0454] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference will be made to the description of the identical or similar parts between the various embodiments. The devices and computer-readable storage media disclosed in the embodiments correspond to the methods disclosed in the embodiments, so the description is relatively simple. For relevant details, refer to the description of the methods.

[0455] Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the technical solution and core ideas of this application. It should be noted that, for those skilled in the art, without departing from the principles of this application, various improvements and modifications may be made to this application, and such improvements and modifications also fall within the scope of protection of this application.

Claims

1. A method for classifying low power factor electric loads, characterized in that: include: Acquiring power load data corresponding to the power load to be classified generated under a low power factor condition; wherein the power load data includes data related to power loads of multiple categories; Performing dimensionality reduction on the power load data to obtain dimensionality-reduced power load data; The target power load classification model obtained by pre-training is used to classify the power load to be classified according to the power load data after dimensionality reduction to obtain a target load category.

2. The low power factor power load classification method according to claim 1, characterized in that: After obtaining power load data corresponding to the power load to be classified generated under a low power factor condition, and before performing dimensionality reduction on the power load data, the method further includes: For each type of power load related data, draw a load data curve that changes over time; The power load data are pre-processed according to the load data curves.

3. The low power factor power load classification method according to claim 2, characterized in that: Pre-process each power load data according to each load data curve, including: For each load data curve, taking each collection time point in the load data curve as a target collection time point; determining a load data change rate corresponding to the target collection time point based on load data corresponding to a next collection time point adjacent to the target collection time point and target load data corresponding to the target collection time point; Determining whether the load data change rate is greater than a first preset threshold; If so, determining that the target load data corresponding to the target collection time point is abnormal data; If not, determining that the target load data corresponding to the target collection time point is normal data; For each abnormal data, selecting each forward normal load data before the abnormal data and each backward normal load data after the abnormal data from the load data curve to which the abnormal data belongs; Correcting the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data; Normalization processing is performed on each load data included in the load data curve after abnormal data correction.

4. The low power factor power load classification method according to claim 3, characterized in that: Correcting the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data includes: Counting the total number of forward load data and the total number of backward load data corresponding to the abnormal data; performing a sum calculation on each forward normal load data corresponding to the abnormal data to obtain a sum of the forward load data corresponding to the abnormal data; performing sum calculation on each piece of backward normal load data corresponding to the abnormal data to obtain the sum of the backward load data corresponding to the abnormal data; Summing up the total amount of the forward load data and the total amount of the backward load data corresponding to the abnormal data to obtain a reference data amount correction parameter corresponding to the abnormal data; performing a sum calculation on the forward load data sum and the backward load data sum corresponding to the abnormal data to obtain reference data and correction parameters corresponding to the abnormal data; A quotient is calculated between the reference data and correction parameter corresponding to the abnormal data and the reference data amount correction parameter to obtain the correction data corresponding to the abnormal data, and the abnormal data is replaced by the correction data corresponding to the abnormal data.

5. The low power factor electric load classification method according to claim 1, characterized in that: Performing dimensionality reduction on the power load data to obtain dimensionality-reduced power load data includes: Determining a target Gaussian distribution variance corresponding to the power load data; For each power load data, calculating the target similarity between the power load data and other power load data according to the target Gaussian distribution variance; Calculating the target joint probability distribution of the power load data according to the target similarities corresponding to the power load data; Randomly determining initial power load data of the power load data in a low-dimensional space; Determining an initial Gaussian distribution variance corresponding to the initial power load data; For each initial power load data, calculating the initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance; Calculating an initial joint probability distribution of the initial power load data according to each initial similarity corresponding to the initial power load data; Calculating a quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain a joint probability evaluation function of the power load data; determining a loss function of a joint probability distribution between the power load data and the corresponding initial power load data according to the joint probability evaluation function of the power load data and the target joint probability distribution; Determining whether the loss function between the power load data and the corresponding initial power load data satisfies a second preset threshold; If yes, determining the initial power load data corresponding to the power load data as the power load data after dimension reduction corresponding to the power load data; If not, the method returns to the step of randomly determining the initial power load data of the power load data in the low-dimensional space.

6. The low power factor electric load classification method according to claim 1, characterized in that: The method further includes a training process of the target power load classification model, wherein the training process of the target power load classification model includes: Acquire sample power load data, and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category; Construct a sample data set based on the sample power load data corresponding to each load category; Dividing the sample data set into a training data set and a test data set; The pre-built initial power load classification model is trained using the training data set, and the trained initial power load classification model is tested using the test data set to obtain the target power load classification model.

7. The low power factor electric load classification method according to claim 6, characterized in that: Clustering is performed on the sample power load data to obtain sample power load data corresponding to each load category, including: Determine the initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as each current centroid; For each sample power load data, respectively calculating a first distance between the sample power load data and a current centroid corresponding to each load category cluster; Dividing the sample power load data into load category clusters corresponding to the minimum first distance among the corresponding first distances; According to the sample power load data in each load category cluster obtained by the current division, the updated centroid corresponding to each load category cluster is determined; Determine whether the updated centroids correspond to the current centroids; If yes, the sample power load data currently divided into each load category cluster is determined as the sample power load data corresponding to each load category; If not, each updated centroid is determined as each current centroid, and the process returns to execute the step of respectively calculating the first distance between each sample power load data and the current centroid corresponding to each load category cluster.

8. The low power factor electric load classification method according to claim 7, characterized in that: Before determining the initial centroids corresponding to the preset number of load category clusters, it also includes: Get multiple cluster quantity values; For each cluster quantity value, determine the centroid corresponding to each initial load category cluster according to the cluster quantity value, and divide each sample power load data into each initial load category cluster to obtain the sample power load data corresponding to each initial load category cluster; Calculating the sum of squares of distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, to obtain an evaluation parameter of the cluster quantity value; Draw an evaluation parameter curve according to the number of clusters and the evaluation parameters corresponding to the number of clusters; For each cluster quantity value in the evaluation parameter curve, calculating a change rate of the evaluation parameter corresponding to the cluster quantity value compared to the evaluation parameter corresponding to the nearest neighbor forward cluster quantity value; Selecting the first change rate whose absolute value is less than a preset change rate threshold from the change rates of the evaluation parameters corresponding to each cluster quantity value in the evaluation parameter curve; The cluster quantity value corresponding to the selected change rate is determined as the preset quantity.

9. The low power factor electric load classification method according to claim 7, characterized in that: Determine the initial centroids corresponding to a preset number of load category clusters, including: Determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density; Dividing each sample power load data into a high-density set and a low-density set according to the density mean; Determining high-density sample power load data having a data density greater than a preset density threshold among the sample power load data in the high-density set; The preset number of central sample power load data with a mutual distance greater than a preset distance threshold is determined in each high-density sample power load data, and each central sample power load data is determined as the initial centroid corresponding to the preset number of load category clusters.

10. The low power factor electric load classification method according to claim 6, characterized in that: After obtaining the sample power load data corresponding to each load category, it also includes: For each sample power load data, calculating the intra-class average distance between the sample power load data and other sample power load data in the load category to which the sample power load data belongs; Determining, from each sample power load data item included in each load category other than the load category to which the sample power load data item belongs, reference sample power load data item closest to the sample power load data item; determining an average inter-class distance between the sample power load data and corresponding reference sample power load data; Determining a first clustering effect evaluation parameter for clustering the sample power load data according to the intra-class average distance and the inter-class average distance corresponding to each sample power load data; Determining the data volume of the sample power load data and the number of categories of each load category, and determining the category center point of each load category, the data center point of the sample power load data, and the intra-category data volume of the sample power load data in each load category; Determining an inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and determining an intra-class covariance matrix based on the center point of the class and each load class; determining a second clustering effect evaluation parameter according to the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of classes; Calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion corresponding to each load category; Calculate the distance between the centroids of any two load categories in each load category to obtain the inter-class dispersion of the two load categories. Summing up the intra-class dispersion degrees corresponding to the two load categories to obtain intra-class dispersion degree evaluation parameters corresponding to the two load categories; Calculating a quotient of the intra-class dispersion evaluation parameter corresponding to the two load categories and the inter-class dispersion corresponding to the two load categories to obtain a class similarity between the two load categories; According to the category similarity between each load category and other load categories, the maximum category similarity corresponding to each load category is determined; Calculate the mean of the similarities of each maximum category to obtain the third clustering effect evaluation parameter; respectively determining weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter, and performing weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain a load clustering effect evaluation parameter; Determining whether a clustering result of the sample power load data meets a preset clustering condition according to the load clustering effect evaluation parameter; If so, each load category is used as labeling information of the sample power load data included in each load category, and a sample data set is constructed according to the sample power load data corresponding to each load category, including: constructing the sample data set according to the sample power load data with labeling information corresponding to each load category; If not, clustering is performed again on the sample power load data.

11. A low power factor electric load classification device, characterized in that: include: An electric load data acquisition module is used to acquire electric load data corresponding to the electric load to be classified generated under low power factor conditions; wherein the electric load data includes multiple categories of electric load related data; A module for obtaining power load data after dimensionality reduction, configured to reduce the dimensionality of the power load data to obtain power load data after dimensionality reduction; The target load category acquisition module is used to classify the to-be-classified electric load according to the dimensionality-reduced electric load data using the pre-trained target electric load classification model to obtain the target load category.

12. The low power factor electric load classification device according to claim 11, characterized in that: Also includes: a curve drawing module for drawing a load data curve that changes over time for each type of power load related data after obtaining power load data corresponding to the power load to be classified generated under low power factor conditions and before performing dimensionality reduction on the power load data; The preprocessing module is used to preprocess each power load data according to each load data curve.

13. The low power factor electric load classification device according to claim 12, characterized in that: The preprocessing module includes: a target collection time point determination submodule, configured to, for each load data curve, use each collection time point in the load data curve as a target collection time point; a load data change rate determination submodule, configured to determine the load data change rate corresponding to the target collection time point based on the load data corresponding to the next collection time point adjacent to the target collection time point and the target load data corresponding to the target collection time point; A first judgment submodule, configured to judge whether the load data change rate is greater than a first preset threshold; an abnormal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is abnormal data when it is determined that the load data change rate is greater than the first preset threshold; a normal data determination submodule, configured to determine that the target load data corresponding to the target collection time point is normal data when it is determined that the load data change rate is not greater than the first preset threshold; A data selection submodule, configured to select, for each abnormal data, each forward normal load data preceding the abnormal data and each backward normal load data following the abnormal data from the load data curve to which the abnormal data belongs; a data correction submodule, configured to correct the abnormal data according to each forward normal load data and each backward normal load data corresponding to the abnormal data; The normalization processing submodule is used to perform normalization processing on each load data included in the load data curve after abnormal data correction.

14. The low power factor electric load classification device according to claim 13, characterized in that: The data correction submodule includes: A quantity counting unit, configured to count the total quantity of forward load data and the total quantity of backward load data corresponding to the abnormal data; a forward load data sum obtaining unit, configured to sum each forward normal load data corresponding to the abnormal data to obtain the forward load data sum corresponding to the abnormal data; a backward load data sum obtaining unit, configured to sum each backward normal load data corresponding to the abnormal data to obtain the backward load data sum corresponding to the abnormal data; a data amount correction parameter obtaining unit, configured to sum the total amount of the forward load data and the total amount of the backward load data corresponding to the abnormal data to obtain a reference data amount correction parameter corresponding to the abnormal data; a data and correction parameter obtaining unit, configured to calculate the sum of the forward load data and the backward load data corresponding to the abnormal data, and obtain reference data and correction parameters corresponding to the abnormal data; The data replacement unit is configured to calculate a quotient of the reference data and correction parameter corresponding to the abnormal data and the reference data amount correction parameter to obtain the correction data corresponding to the abnormal data, and replace the abnormal data with the correction data corresponding to the abnormal data.

15. The low power factor electric load classification device according to claim 11, characterized in that: The module for obtaining power load data after dimension reduction includes: a target Gaussian distribution variance determination submodule, configured to determine a target Gaussian distribution variance corresponding to the power load data; A target similarity calculation submodule is used to calculate, for each piece of power load data, the target similarity between the power load data and other power load data according to the target Gaussian distribution variance; A target joint probability distribution calculation submodule is used to calculate the target joint probability distribution of the power load data according to the similarities of each target corresponding to the power load data; an initial power load data determination submodule, configured to randomly determine the initial power load data of the power load data in a low-dimensional space; an initial Gaussian distribution variance determination submodule, configured to determine the initial Gaussian distribution variance corresponding to the initial power load data; an initial similarity calculation submodule, configured to calculate, for each initial power load data, an initial similarity between the initial power load data and other initial power load data according to the initial Gaussian distribution variance; an initial joint probability distribution calculation submodule, configured to calculate an initial joint probability distribution of the initial power load data according to each initial similarity corresponding to the initial power load data; a joint probability evaluation function obtaining submodule, configured to calculate a quotient of the target joint probability distribution of the power load data and the initial joint probability distribution of the corresponding initial power load data to obtain a joint probability evaluation function of the power load data; a loss function determination submodule, configured to determine a loss function of a joint probability distribution between the power load data and the corresponding initial power load data according to the joint probability evaluation function of the power load data and the target joint probability distribution; A second judgment submodule is configured to judge whether the loss function between the power load data and the corresponding initial power load data satisfies a second preset threshold; a submodule for determining power load data after dimensionality reduction, configured to, when it is determined that a loss function between the power load data and the initial power load data corresponding thereto satisfies a second preset threshold, determine the initial power load data corresponding to the power load data as the power load data after dimensionality reduction corresponding to the power load data; The repeatedly executed submodule is used to return to the step of randomly determining the initial power load data of the power load data in the low-dimensional space when it is determined that the loss function between the power load data and its corresponding initial power load data does not meet the second preset threshold.

16. The low power factor electric load classification device according to claim 11, characterized in that: It also includes a model training module, which includes: The sample power load data acquisition submodule is used to obtain sample power load data and perform clustering processing on the sample power load data to obtain sample power load data corresponding to each load category; A sample data set construction submodule is used to construct a sample data set based on the sample power load data corresponding to each load category; A training data set and a test data set division submodule, configured to divide the sample data set into a training data set and a test data set; The target power load classification model acquisition submodule is used to train the pre-built initial power load classification model using the training data set, and test the trained initial power load classification model using the test data set to obtain the target power load classification model.

17. The low power factor electric load classification device according to claim 16, characterized in that: The sample power load data acquisition submodule includes: a current centroid determining unit, configured to determine initial centroids corresponding to a preset number of load category clusters, and determine each initial centroid as each current centroid; A first distance calculation unit is configured to calculate, for each sample power load data, a first distance between the sample power load data and a current centroid corresponding to each load category cluster; a load category clustering unit, configured to divide the sample power load data into a load category cluster corresponding to a minimum first distance among the corresponding first distances; An updated centroid determining unit, configured to determine the updated centroid corresponding to each load category cluster based on the sample power load data in each load category cluster currently divided; A judgment unit, used to judge whether each updated centroid corresponds to each current centroid; a sample power load data determining unit, configured to determine the sample power load data currently classified into each load category cluster as the sample power load data corresponding to each load category when it is determined that each updated centroid corresponds to each current centroid; The repeated execution unit is used to determine each updated centroid as each current centroid when it is determined that the updated centroid does not correspond to each current centroid, and return to execute the step of calculating the first distance between the sample power load data and the current centroid corresponding to each load category cluster for each sample power load data.

18. The low power factor electric load classification device according to claim 17, characterized in that: Also includes: A cluster quantity value acquisition module is used to obtain a plurality of cluster quantity values before determining the initial centroids corresponding to the preset number of load category clusters; a sample power load data determination module, configured to determine, for each cluster quantity value, the centroid corresponding to each initial load category cluster according to the cluster quantity value, and to divide each sample power load data into each initial load category cluster to obtain sample power load data corresponding to each initial load category cluster; An evaluation parameter acquisition module is used to calculate the sum of squares of distances between each sample power load data and the centroid of the initial load category cluster to which it belongs, and obtain an evaluation parameter of the cluster quantity value; An evaluation parameter curve drawing module is used to draw an evaluation parameter curve according to the number of clusters and the evaluation parameters corresponding to the number of clusters; a change rate calculation module, configured to calculate, for each cluster quantity value in the evaluation parameter curve, a change rate of the evaluation parameter corresponding to the cluster quantity value compared to the evaluation parameter corresponding to the nearest neighbor forward cluster quantity value; a change rate selection module, configured to select, from the change rates of the evaluation parameters corresponding to each cluster quantity value in the evaluation parameter curve, the first change rate whose absolute value is less than a preset change rate threshold; The preset number determining module is configured to determine the cluster number value corresponding to the selected change rate as the preset number.

19. The low power factor electric load classification device according to claim 17, characterized in that: The current center of mass determination unit includes: a density mean determination subunit, configured to determine the data density within a preset neighborhood corresponding to each sample power load data, and determine the density mean corresponding to each data density; a high-density set and low-density set division subunit, configured to divide each sample power load data into a high-density set and a low-density set according to the density mean; a high-density sample power load data determining subunit, configured to determine, from the sample power load data in the high-density set, high-density sample power load data whose data density is greater than a preset density threshold; The initial centroid determination subunit is used to determine the preset number of central sample power load data with a mutual distance greater than a preset distance threshold in each high-density sample power load data, and determine each central sample power load data as the initial centroid corresponding to the preset number of load category clusters.

20. The low power factor electric load classification device according to claim 16, characterized in that: Also includes: An intra-class average distance calculation module is used to calculate, after obtaining the sample power load data corresponding to each load category, the intra-class average distance between each sample power load data and other sample power load data in the load category to which the sample power load data belongs; A nearest reference sample power load data determination module is configured to determine, from among each sample power load data respectively included in other load categories except the load category to which the sample power load data belongs, reference sample power load data that is nearest to the sample power load data; an inter-class average distance determination module, configured to determine an inter-class average distance between the sample power load data and corresponding reference sample power load data; a first clustering effect evaluation parameter determination module, configured to determine a first clustering effect evaluation parameter for clustering the sample power load data according to the intra-class average distance and the inter-class average distance corresponding to each sample power load data; an intra-class data volume determination module, configured to determine the data volume of the sample power load data and the number of categories of each load category, and to determine the category center point of each load category, the data center point of the sample power load data, and the intra-class data volume of the sample power load data in each load category; an intra-class covariance matrix determination module, configured to determine an inter-class covariance matrix based on the center point of each class, the center point of the data, and the amount of data within each class, and to determine an intra-class covariance matrix based on the center point of the class and each load class; a second clustering effect evaluation parameter determination module, configured to determine a second clustering effect evaluation parameter according to the inter-class covariance matrix, the intra-class covariance matrix, the data volume of the sample power load data, and the number of classes; The intra-class dispersion degree acquisition module is used to calculate the average distance between each sample power load data in each load category and the centroid of the category to which it belongs, and obtain the intra-class dispersion degree corresponding to each load category; An inter-class dispersion degree obtaining module is used to calculate the distance between the class centroids corresponding to any two load classes in each load class to obtain the inter-class dispersion degrees corresponding to the two load classes; An intra-class dispersion evaluation parameter obtaining module is used to sum the intra-class dispersions corresponding to the two load categories to obtain the intra-class dispersion evaluation parameters corresponding to the two load categories; a category similarity obtaining module, configured to calculate a quotient of the intra-category dispersion evaluation parameters corresponding to the two load categories and the inter-category dispersion corresponding to the two load categories, to obtain the category similarity between the two load categories; A maximum category similarity determination module is used to determine the maximum category similarity corresponding to each load category based on the category similarity between each load category and other load categories; A third clustering effect evaluation parameter obtaining module is used to calculate the mean of the maximum category similarities to obtain the third clustering effect evaluation parameter; a load clustering effect evaluation parameter obtaining module, configured to respectively determine weight coefficients corresponding to the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter, and perform weighted calculation on the first clustering effect evaluation parameter, the second clustering effect evaluation parameter, and the third clustering effect evaluation parameter according to the weight coefficients to obtain the load clustering effect evaluation parameter; A judgment module, configured to judge whether a clustering result of the sample power load data satisfies a preset clustering condition according to the load clustering effect evaluation parameter; a labeling information determining module, configured to, when determining that the clustering result of the sample power load data satisfies the preset clustering condition, use each load category as labeling information of the sample power load data included in each load category; The sample data set construction submodule is specifically a module for constructing the sample data set according to the sample power load data with labeled information corresponding to each load category; The re-clustering module is configured to re-cluster the sample power load data when it is determined that the clustering result of the sample power load data does not meet the preset clustering condition.

21. A low power factor electric load classification device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the low power factor electric load classification method according to any one of claims 1 to 10 when executing the computer program.

22. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the low power factor electric load classification method according to any one of claims 1 to 10 are implemented.

Images