A power distribution area daily line loss rate curve abnormal feature extraction method and system

Abstract

The application discloses a power distribution area daily line loss rate curve abnormal feature extraction method and system, comprising the following steps: according to the obtained daily line loss rate curve data and area line loss abnormal treatment record work order, an abnormal power distribution area daily line loss rate dataset is constituted; the abnormal power distribution area daily line loss rate dataset is classified according to the abnormal type, and a plurality of abnormal dataset subsets are obtained; local section data of the plurality of abnormal dataset subsets is extracted through a preset scanning window, and local section data corresponding to each abnormal dataset subset is obtained; statistical value features of the local section data are calculated and clustering calculation is performed, a plurality of first abnormal features are obtained; the plurality of first abnormal features are determined according to a self-defined rule, a plurality of second abnormal features are screened out, and abnormal features corresponding to each abnormal dataset subset in the plurality of abnormal dataset subsets are obtained, thereby improving the universality of abnormal diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power grid operation and maintenance technology, and in particular to a method and system for extracting abnormal features of daily line loss rate curves in distribution substations. Background Technology

[0002] Data analysis techniques for diagnosing abnormal line losses in distribution substations typically involve collecting and analyzing data from the distribution substation to identify the causes of abnormal line losses. Existing diagnostic techniques involve analyzing data such as voltage, current, and power in the distribution substation to determine the causes of abnormal line losses. For example, analyzing the voltage data of the distribution substation can help determine whether there is overvoltage or undervoltage, thereby identifying the cause of the abnormal line losses.

[0003] However, the main drawback of existing technologies is that they require the collection of voltage and current data from electricity meters. Current power grid systems and communication equipment do not support large-scale voltage and current data collection. They can only perform voltage and current-based anomaly diagnosis for certain distribution substations with good hardware conditions. When the hardware conditions of a distribution substation are insufficient to support the acquisition of large-scale voltage data, it is impossible to perform anomaly diagnosis for the current distribution substation. Therefore, existing anomaly diagnosis methods cannot perform large-scale anomaly diagnosis for distribution substations and are not universal. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses a method and system for extracting abnormal features of daily line loss rate curves in distribution substations, solving the problem of large-scale, universally applicable line loss anomaly diagnosis.

[0005] To achieve the above objectives, in a first aspect, the present invention discloses a method for extracting abnormal features of the daily line loss rate curve of a distribution station area, comprising:

[0006] Based on the acquired daily line loss rate curve data and the work orders for handling abnormal line loss in the distribution area, a daily line loss rate dataset for abnormal distribution areas is constructed.

[0007] The daily line loss rate dataset of the abnormal distribution substations is classified according to the abnormality type of the work order records of the abnormal line loss handling records of the substations to obtain several subsets of abnormal datasets.

[0008] Local cross-sectional data are extracted from each subset of the several types of abnormal datasets through a preset scanning window to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets.

[0009] Statistical value feature calculation and clustering calculation are performed on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset.

[0010] A set of custom rules is applied to determine the plurality of first abnormal features, and a plurality of second abnormal features are selected. The plurality of second abnormal features are then merged and summarized to obtain the abnormal features corresponding to each subset of the plurality of abnormal dataset subsets.

[0011] This invention discloses a method for extracting abnormal features from the daily line loss rate curve of a distribution substation. First, it constructs an abnormal daily line loss rate dataset using acquired daily line loss rate curve data and line loss anomaly handling records. This allows for the extraction of abnormal features based on past anomaly handling records, improving extraction accuracy. Further, after constructing the dataset, it classifies anomalies according to their types, ensuring different anomalies correspond to different abnormal features. Then, it obtains cross-sectional data from a subset of the abnormal dataset, reducing data processing volume and improving feature extraction efficiency. Statistical value feature calculations and clustering calculations are performed on the obtained cross-sectional data to search for the optimal clustering result and optimal abnormal features corresponding to different subsets of the abnormal dataset. The obtained clustering results are then filtered to further improve the accuracy of the abnormal features. This invention utilizes past line loss rate data and anomaly handling records to construct abnormal features, enabling line loss diagnosis based on these features. This avoids the need for prior art techniques that rely on voltage and current data for diagnosis, and it has no hardware requirements for the distribution substation, making it universally applicable.

[0012] As a preferred example, the step of extracting local cross-sectional data for each subset of the plurality of abnormal datasets through a preset scanning window includes:

[0013] Based on the preset step size and width in the scanning window, local cross-sectional data are extracted from each subset of the abnormal dataset to obtain several columns of data;

[0014] By merging the data from the several columns, local cross-sectional data corresponding to each subset of the abnormal dataset are obtained.

[0015] In this invention, a scanning window with a set step size and width is used to extract local cross-sectional data for each subset of the abnormal dataset. Under random conditions, certain data are selected for abnormal feature extraction, which not only ensures the accuracy of subsequent abnormal feature extraction but also reduces the amount of data processing and improves the efficiency of feature extraction.

[0016] As a preferred example, in the step of performing statistical value feature calculation and clustering calculation on the local cross-sectional data, several first-class anomaly features are obtained, including:

[0017] The mean, variance, slope, and dispersion of each row of data in the local cross-sectional data are calculated according to a preset statistical value calculation method, and the mean, variance, slope, and dispersion of each row of data are statistically analyzed to obtain the local feature dataset corresponding to the local cross-sectional data.

[0018] The local feature dataset is clustered using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

[0019] This invention improves the efficiency and accuracy of abnormal feature extraction by performing statistical value calculations and clustering calculations on the local cross-sectional data and effectively aggregating the abnormal features corresponding to the dataset into a unified classification.

[0020] As a preferred example, the clustering calculation of the local feature dataset using a preset fuzzy mean clustering algorithm includes:

[0021] The local feature dataset is clustered using a pre-defined fuzzy mean clustering algorithm; wherein the objective function of the fuzzy mean clustering algorithm is:

[0022]

[0023] Let X = {x1, x2, x3, ..., x} n}, XсR S Given the number of samples as N, the dimension of the sample space as S, and the number of clusters as C; and:

[0024]

[0025]

[0026] u ij ≥0, 1≤i≤N, 1≤j≤C

[0027] Where U is the membership matrix; V is a matrix consisting of C cluster centers; u ij x represents the membership degree of the i-th sample to the j-th class; i For the i-th sample; v j Let m be the j-th cluster center; m is the fuzzy coefficient; ||x i -v j || represents the sample point x i To the cluster center v j Euclidean distance;

[0028] The clustering results of the fuzzy mean clustering algorithm are evaluated using a preset clustering validity function; wherein, the clustering validity function is:

[0029]

[0030] Where, x i For the i-th sample; v j v is the j-th cluster center; i Let i be the i-th cluster center.

[0031] This invention utilizes the objective function for clustering iteration to obtain the optimal clustering result when the objective function is minimized. Furthermore, the clustering validity function is used to evaluate the clustering result of the objective function to improve the accuracy of the clustering result, that is, to improve the precision of the abnormal feature extraction.

[0032] As a preferred example, in the process of determining the plurality of first abnormal features using custom rules to filter out a plurality of second abnormal features, including:

[0033] Obtain the number of clusters corresponding to each of the plurality of first abnormal features, and sort the number of clusters in ascending order of numerical value;

[0034] The number of transformer area IDs corresponding to each of the first abnormal features is counted, and duplicates are removed to obtain the number of transformer areas corresponding to each of the first abnormal features.

[0035] Based on the total number of transformer substations and the number of transformer substations, calculate the abnormal frequency corresponding to each of the first abnormal features;

[0036] Based on the ascending order, each of the plurality of first abnormal features is judged, and the first abnormal feature corresponding to the abnormal frequency being greater than or equal to a preset threshold is selected as the second abnormal feature, thereby obtaining a plurality of second abnormal features.

[0037] This invention improves the accuracy of abnormal feature extraction by using custom rules to determine several abnormal features contained in each abnormal type and deleting abnormal features that do not meet the determination rules from the dataset.

[0038] Secondly, the present invention discloses a system for extracting abnormal features of the daily line loss rate curve of a distribution station area. The system includes a dataset construction module, a dataset classification module, a cross-sectional data extraction module, a data clustering calculation module, and a feature extraction module.

[0039] The dataset construction module is used to construct the daily line loss rate dataset of abnormal distribution substations based on the acquired daily line loss rate curve data and the work orders for handling abnormal line loss in the substation area.

[0040] The dataset classification module is used to classify the daily line loss rate dataset of the abnormal distribution substations according to the abnormality type of the work order records of the abnormal line loss handling records of the substations, and obtain several subsets of abnormal datasets.

[0041] The cross-sectional data extraction module is used to extract local cross-sectional data from each subset of the several types of abnormal datasets through a preset scanning window, so as to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets.

[0042] The data clustering calculation module is used to perform statistical value feature calculation and clustering calculation on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset.

[0043] The feature extraction module is used to determine the plurality of first abnormal features using custom rules, filter out a plurality of second abnormal features, merge and summarize the plurality of second abnormal features, and obtain the abnormal features corresponding to each subset of the plurality of abnormal dataset subsets.

[0044] This invention discloses a system for extracting abnormal features from the daily line loss rate curve of a distribution substation. First, it constructs an abnormal daily line loss rate dataset using acquired daily line loss rate curve data and line loss anomaly handling records. This allows for the extraction of abnormal features based on past anomaly handling records, improving extraction accuracy. Furthermore, after constructing the dataset, it first classifies the anomalies according to their types, ensuring that different anomalies correspond to different abnormal features. Next, it obtains cross-sectional data from a subset of the abnormal dataset, reducing data processing volume and improving feature extraction efficiency. Statistical value feature calculations and clustering calculations are performed on the obtained cross-sectional data to search for the optimal clustering result and optimal abnormal features corresponding to different subsets of the abnormal dataset. The obtained clustering results are then filtered to further improve the accuracy of the abnormal features. This invention utilizes past line loss rate data and anomaly handling records to construct abnormal features, enabling line loss diagnosis based on these features. This avoids the need for prior art techniques that rely on voltage and current data for diagnosis, and it has no hardware requirements for the distribution substation, making it universally applicable.

[0045] As a preferred example, the cross-sectional data extraction module includes a scanning unit and a merging unit;

[0046] The scanning unit is used to extract local cross-sectional data from each subset of the abnormal dataset according to the preset step size and width in the scanning window, and obtain several columns of data.

[0047] The merging unit is used to merge the data from the plurality of columns to obtain local cross-sectional data corresponding to each subset of the abnormal dataset.

[0048] In this invention, a scanning window with a set step size and width is used to extract local cross-sectional data for each subset of the abnormal dataset. Under random conditions, certain data are selected for abnormal feature extraction, which not only ensures the accuracy of subsequent abnormal feature extraction but also reduces the amount of data processing and improves the efficiency of feature extraction.

[0049] As a preferred example, the data clustering calculation module includes a statistical unit and a clustering unit;

[0050] The statistical unit is used to calculate the mean, variance, slope and dispersion of each row of data in the local cross-sectional data according to the preset statistical value calculation method, and to calculate the mean, variance, slope and dispersion of each row of data respectively, so as to obtain the local feature dataset corresponding to the local cross-sectional data.

[0051] The clustering unit is used to perform clustering calculations on the local feature dataset using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

[0052] This invention improves the efficiency and accuracy of abnormal feature extraction by performing statistical value calculations and clustering calculations on the local cross-sectional data and effectively aggregating the abnormal features corresponding to the dataset into a unified classification.

[0053] As a preferred example, the feature extraction module includes an ascending order unit, a calculation unit, and a feature filtering unit;

[0054] The ascending unit is used to obtain the number of clusters corresponding to each of the plurality of first abnormal features, and sort the number of clusters in ascending order of numerical value;

[0055] The calculation unit is used to count the transformer area IDs corresponding to each of the first abnormal features, and to count the duplicates to obtain the number of transformer areas corresponding to each of the first abnormal features; and to calculate the abnormal frequency corresponding to each of the first abnormal features based on the total number of transformer areas and the number of transformer areas.

[0056] The feature filtering unit is used to determine each of the plurality of first abnormal features according to the ascending order, and to filter out the first abnormal features corresponding to the abnormal frequency being greater than or equal to a preset threshold as the second abnormal features, thereby obtaining a plurality of second abnormal features.

[0057] This invention improves the accuracy of abnormal feature extraction by using custom rules to determine several abnormal features contained in each abnormal type and deleting abnormal features that do not meet the determination rules from the dataset.

[0058] Thirdly, the present invention discloses an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the method for extracting abnormal features of the daily line loss rate curve of a distribution station area as described in the first aspect. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a method for extracting abnormal features from the daily line loss rate curve of a distribution station, as provided in an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram of a system for extracting abnormal features of the daily line loss rate curve of a distribution station, provided in an embodiment of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example

[0063] This invention discloses a method for extracting abnormal features from the daily line loss rate curve of a distribution substation. The specific implementation process of the extraction method can be found in [reference needed]. Figure 1 It mainly includes steps 101 to 105, which mainly include:

[0064] Step 101: Based on the obtained daily line loss rate curve data and the work orders for handling abnormal line loss in the distribution area, construct the daily line loss rate dataset for the abnormal distribution area.

[0065] In this embodiment, the main steps are: to obtain the daily line loss rate curve data of the low-voltage distribution transformer area in the metering automation system, and to obtain the work order event records of abnormal line loss handling in the marketing management system to form a dataset of daily line loss rate of abnormal distribution transformer areas.

[0066] Furthermore, in this embodiment of the invention, the metering automation system is configured with n low-voltage distribution substations and m days of daily line loss rate data x. i(i∈[1,m]), there are work order event records for handling abnormal line loss in the marketing management system. An employee handles a certain type of abnormal event y (such as electricity theft, metering failure, etc.) in a certain transformer area (ID). The original dataset is constructed and merged from these two types of datasets. The data structure is as follows.

[0067] ID x1 x2 ... xm y 1 A 2 B ... C n D

[0068] Step 102: Classify the daily line loss rate dataset of the abnormal distribution substation according to the abnormality type of the work order record of the abnormal line loss handling record of the substation, and obtain several subsets of abnormal datasets.

[0069] In this embodiment, the main step is to classify the anomalies according to the anomaly types recorded in the work order records of abnormal line loss handling in the transformer area, and construct a subset of the dataset.

[0070] Furthermore, in this embodiment, assuming there are F types of anomalies, F parallel computations are performed, and serial computation is performed for each type. If the anomaly type y of the work order is classified and a subset of the dataset is constructed, for example, there are ni stations for y=A, nj stations for y=B, and nk stations for y=C. The dataset subset table of the stations corresponding to the anomaly data is as follows:

[0071] ID x1 x2 ... xm y 1 A 5 A ... A ni A

[0072] ID <![CDATA[x1]]> <![CDATA[x2]]> ... <![CDATA[x m ]]> y 2 B 7 B ... B <![CDATA[n j ]]> B

[0073]

[0074]

[0075] As can be seen from the table above, the dataset is classified according to the type of abnormal event, resulting in a subset of abnormal datasets with an equal number of abnormal data types.

[0076] Step 103: Extract local cross-sectional data for each subset of the several types of abnormal datasets through a preset scanning window to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets.

[0077] In this embodiment, the step includes: extracting local cross-sectional data from each subset of the abnormal dataset according to the preset step size and width in the scanning window to obtain several columns of data; merging the several columns of data to obtain the local cross-sectional data corresponding to each subset of the abnormal dataset.

[0078] Furthermore, in this embodiment, a scanning window with a step size of 1 and a width of d is set to extract local cross-sectional data from the aforementioned data subset. Taking dataset A as an example, a scanning window with a step size of 2 and a width of 5 is set to extract local cross-sectional data, and the datasets are then merged. See below.

[0079] Extract 5 columns of data starting from x1:

[0080] ID x1 x2 x3 x4 x5 y 1 A

[0081] Extract 5 columns of data starting from x3:

[0082] ID x3 x4 x5 x6 x7 y 1 A

[0083] Continue until all data sets A have been scanned and merged. See below.

[0084]

[0085]

[0086] In this embodiment, this step uses a scanning window with a set step size and width to extract local cross-sectional data for each subset of the abnormal dataset. Under random conditions, certain data are selected for abnormal feature extraction, which not only ensures the accuracy of subsequent abnormal feature extraction but also reduces the amount of data processing and improves the efficiency of feature extraction.

[0087] Step 104: Perform statistical value feature calculation and clustering calculation on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset.

[0088] In this embodiment, the step mainly includes: calculating the mean, variance, slope, and dispersion of each row of data in the local cross-sectional data according to a preset statistical value calculation method, and statistically calculating the mean, variance, slope, and dispersion of each row of data to obtain the local feature dataset corresponding to the local cross-sectional data; and performing clustering calculation on the local feature dataset using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

[0089] Furthermore, the objective function in the fuzzy mean clustering algorithm is:

[0090]

[0091] Let X = {x1, x2, x3, ..., x} n}, XсR S Given the number of samples as N, the dimension of the sample space as S, and the number of clusters as C; and:

[0092]

[0093]

[0094] u ij ≥0, 1≤i≤N, 1≤j≤C

[0095] Where U is the membership matrix; V is a matrix consisting of C cluster centers; u ij x represents the membership degree of the i-th sample to the j-th class; i For the i-th sample; v j Let m be the j-th cluster center; m is the fuzzy coefficient; ||x i -v j || represents the sample point x i To the cluster center v j Euclidean distance;

[0096] The clustering results of the fuzzy mean clustering algorithm are evaluated using a preset clustering validity function; wherein, the clustering validity function is:

[0097]

[0098] Where, x i For the i-th sample; v j v is the j-th cluster center; i Let i be the i-th cluster center.

[0099] Specifically, in this embodiment, statistical value features are calculated on the merged local cross-sectional data. The calculation method is to calculate the mean M, variance D, slope K, and dispersion V of each row of data to construct a local feature dataset.

[0100] The formula for calculating the mean M is:

[0101]

[0102] The formula for calculating variance D is:

[0103]

[0104] The formula for calculating the slope K is:

[0105]

[0106] The formula for calculating the dispersion rate V is:

[0107]

[0108] After calculating the statistical features of each row of data, a new dataset is formed as follows.

[0109]

[0110]

[0111] Taking the subset A of the dataset as an example, clustering calculation is performed on the local feature dataset of A, and n i The row data is clustered into C classes using the fuzzy C-means clustering algorithm.

[0112] The fuzzy C-means clustering algorithm has three key parameters: the number of clusters, the cluster centers, and the number of iterations. FCM obtains cluster centers by minimizing an objective function, which is essentially the sum of the Euclidean distances (sum of squared errors) from each data point to each cluster. The clustering process is essentially the process of minimizing the objective function. Through repeated iterative calculations, the error value of the objective function is gradually reduced. When the objective function converges, the final clustering result is obtained.

[0113] Let X include x1, x2, x3, and so on up to x n ,and Given N samples, S sample space dimension, and C number of clusters, the objective function is:

[0114]

[0115] Make:

[0116]

[0117]

[0118] u ij ≥0, 1≤i≤N, 1≤j≤C

[0119] In the formula: U is the membership matrix; V is a matrix consisting of C cluster centers; u ij x represents the membership degree of the i-th sample to the j-th class; i For the i-th sample; v j Let m be the j-th cluster center; m is the fuzzy coefficient; ||x i -v j || represents the sample point x i To the cluster center v j Euclidean distance.

[0120] Fuzzy C-means clustering requires pre-specifying the number of clusters C, but in practice, the number of clusters is usually uncertain. Therefore, the algorithm obtains different fuzzy partitions for different values ​​of C. Clustering effectiveness metrics can evaluate the clustering effect and thus determine the optimal number of clusters. Furthermore, when the data type is unknown, effectiveness metrics can also be used to search for the optimal number of clusters. Based on a comprehensive analysis of effectiveness metrics, the Xie-Beni clustering effectiveness function is selected to determine the optimal number of clusters. The XB metric can find a balance between intra-cluster compactness and inter-cluster separation, and its formula is:

[0121]

[0122] In the formula: x i For the i-th sample; v j v is the j-th cluster center; i Let i be the i-th cluster center.

[0123] Intra-class compactness = numerator of formula / number of samples N, the smaller the better; inter-class separation = denominator of formula * number of samples N, the larger the better; the smaller the XB index, the better the clustering effect.

[0124] For example, the clusters are divided into 3 categories, labeled as a, b, and c respectively. The clustering results are shown in the table below.

[0125] ID M D K V y C 1 A a 1 A a 1 A b 1 A c 5 A a 5 A b 5 A c ... A b ni A c

[0126] In this embodiment, this step performs statistical value calculation and clustering calculation on the local cross-sectional data, effectively aggregating the abnormal features corresponding to the dataset into a unified classification, thereby improving the efficiency and accuracy of abnormal feature extraction.

[0127] Step 105: Perform custom rule judgment on the several first abnormal features, filter out several second abnormal features, merge and summarize the several second abnormal features, and obtain the abnormal features corresponding to each subset of the several abnormal datasets.

[0128] In this embodiment, the main steps are as follows: obtaining the number of clusters corresponding to each of the plurality of first abnormal features, and sorting the number of clusters in ascending order; counting the area IDs corresponding to each of the first abnormal features, and deduplicating the counts to obtain the number of areas corresponding to each of the first abnormal features; calculating the abnormal frequency corresponding to each of the first abnormal features based on the total number of areas and the number of areas; judging each of the plurality of first abnormal features according to the ascending order, and selecting the first abnormal features corresponding to the abnormal frequency being greater than or equal to a preset threshold as the second abnormal features, thereby obtaining a plurality of second abnormal features.

[0129] Specifically, in this embodiment, a custom rule is used to determine the C-class clustering results in the clustering result table to filter out abnormal local data features. The defined determination rule is as follows: First, in the clustering result table, define the number of clustering results as 'a' as 'na', the number of clustering results as 'b' as 'nb', and the number of clustering results as 'c' as 'nc'. Sort 'na', 'nb', and 'nc' in ascending order of their values, for example, upRank(na, nb, nc) = (nc, nb, na). Count the station IDs with clustering results of a, b, and c respectively, then remove duplicates, and then count them. 'ma' represents the number of station IDs with clustering result 'a'.

[0130] ma=COUNT(UN IQUE(ID(C)=a))

[0131] mb=COUNT(UN IQUE(ID(C)=b))

[0132] mc=COUNT(UN IQUE(ID(C)=c))

[0133] Calculate frequency Where m a n is the number of transformer substations with clustering result 'a'. i This represents the total number of districts.

[0134] Calculate frequency Where m b n is the number of transformer substations with a clustering result of b. i This represents the total number of districts.

[0135] Calculate frequency Where m c n is the number of transformer substations with a clustering result of c. i This represents the total number of districts.

[0136] If upRank(na, nb, nc) = (nc, nb, na), then Step 1 evaluates P in the order of c, b, a. α can be set to 80%-90%. If P(c) > α, then proceed directly to Step 2; if P(c) < α, then evaluate the next result b.

[0137] If P>α cannot be satisfied after all checks are performed, then the value of α can be decreased.

[0138] Step 2 outputs the cluster centers of the clustering results that conform to the decision rules. For example, if P(c) < α and P(b) > α, then the cluster centers of clustering result b are output as the abnormal local features of anomaly type A. See below.

[0139]

[0140] The above calculations are performed on the F-type anomalies from step 1 to obtain the feature cluster centers of each type as local features of each anomaly type. The merged output results are as follows.

[0141]

[0142] Therefore, the local anomaly characteristics of the power distribution area can be obtained through the above statistical value calculation and clustering calculation.

[0143] Furthermore, after obtaining the aforementioned abnormal feature classification, the local features of each abnormal type can be used to diagnose anomalies in the daily line loss rate data of other distribution substations. For example, if the statistical characteristics of the daily line loss rate of other substations over the past 5 days are similar to those of type A, then it is diagnosed that type A anomaly has occurred in that substation over the past 5 days. (See below.)

[0144]

[0145]

[0146]

[0147]

[0148] In this embodiment, this step improves the accuracy of abnormal feature extraction by using custom rules to determine several abnormal features included in each abnormal type and deleting abnormal features that do not meet the determination rules from the dataset.

[0149] In addition to the methods described above, this invention also provides a system for extracting abnormal features of the daily line loss rate curve of a distribution station area. For the specific structural composition of the extraction system, please refer to [reference needed]. Figure 2The system includes a dataset construction module 201, a dataset classification module 202, a cross-sectional data extraction module 203, a data clustering calculation module 204, and a feature extraction module 205.

[0150] The dataset construction module 201 is used to construct the daily line loss rate dataset of abnormal distribution substations based on the acquired daily line loss rate curve data and the work orders for handling abnormal line loss in the substation area.

[0151] The dataset classification module 202 is used to classify the daily line loss rate dataset of the abnormal distribution area according to the abnormality type of the work order record of the abnormal line loss handling record of the transformer area, and obtain several abnormal dataset subsets.

[0152] The cross-sectional data extraction module 203 is used to extract local cross-sectional data from each subset of the several types of abnormal datasets through a preset scanning window, so as to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets.

[0153] The data clustering calculation module 204 is used to perform statistical value feature calculation and clustering calculation on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset.

[0154] The feature extraction module 205 is used to perform custom rule judgment on the plurality of first abnormal features, filter out a plurality of second abnormal features, merge and summarize the plurality of second abnormal features, and obtain the abnormal features corresponding to each subset of the plurality of abnormal dataset subsets.

[0155] In this embodiment, the cross-sectional data extraction module 203 includes a scanning unit and a merging unit.

[0156] The scanning unit is used to extract local cross-sectional data from each subset of the abnormal dataset according to the preset step size and width in the scanning window, and obtain several columns of data.

[0157] The merging unit is used to merge the data from the plurality of columns to obtain local cross-sectional data corresponding to each subset of the abnormal dataset.

[0158] In this embodiment, the data clustering calculation module 204 includes a statistical unit and a clustering unit.

[0159] The statistical unit is used to calculate the mean, variance, slope and dispersion of each row of data in the local cross-sectional data according to a preset statistical value calculation method, and to statistically calculate the mean, variance, slope and dispersion of each row of data respectively, so as to obtain the local feature dataset corresponding to the local cross-sectional data.

[0160] The clustering unit is used to perform clustering calculations on the local feature dataset using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

[0161] In this embodiment, the feature extraction module 205 includes an ascending order unit, a calculation unit, and a feature filtering unit.

[0162] The ascending order unit is used to obtain the number of clusters corresponding to each of the plurality of first abnormal features, and sort the number of clusters in ascending order of numerical value.

[0163] The calculation unit is used to count the transformer area IDs corresponding to each of the first abnormal features, and to count the duplicates to obtain the number of transformer areas corresponding to each of the first abnormal features; based on the total number of transformer areas and the number of transformer areas, the abnormal frequency corresponding to each of the first abnormal features is calculated.

[0164] The feature filtering unit is used to determine each of the plurality of first abnormal features according to the ascending order, and to filter out the first abnormal features corresponding to the abnormal frequency being greater than or equal to a preset threshold as the second abnormal features, thereby obtaining a plurality of second abnormal features.

[0165] In addition to the methods and systems described above, embodiments of the present invention also provide an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the method for extracting abnormal features of the daily line loss rate curve of a distribution station area as described in the embodiments of the present invention.

[0166] This invention discloses a method and system for extracting abnormal features from the daily line loss rate curve of a distribution substation. Compared with existing substation line loss anomaly diagnosis technologies, this method only uses existing data from the power grid system to calculate abnormal features, has low data requirements, and can be widely and universally promoted and applied. At the same time, compared with curve feature extraction technologies in other industries, which mostly involve curve noise reduction and feature extraction based on set rules, this invention can automatically calculate abnormal features from the daily line loss rate data curve through scanning slicing, local features, feature clustering, and statistical judgment.

[0167] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for extracting abnormal features of the daily line loss rate curve of a distribution station area, characterized in that, include: Based on the acquired daily line loss rate curve data and the work orders for handling abnormal line loss in the distribution area, a daily line loss rate dataset for abnormal distribution areas is constructed. The daily line loss rate dataset of the abnormal distribution substations is classified according to the abnormality type of the work order records of the abnormal line loss handling records of the substations to obtain several subsets of abnormal datasets. Local cross-sectional data are extracted from each subset of the several types of abnormal datasets through a preset scanning window to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets. Statistical value feature calculation and clustering calculation are performed on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset. A custom rule is used to determine the plurality of first abnormal features, and a plurality of second abnormal features are selected. The plurality of second abnormal features are then merged and summarized to obtain the abnormal features corresponding to each subset of the plurality of abnormal datasets. The step of extracting local cross-sectional data from each subset of the plurality of abnormal datasets through a preset scanning window includes: Based on the preset step size and width in the scanning window, local cross-sectional data are extracted from each subset of the abnormal dataset to obtain several columns of data; Merge the data from the several columns to obtain the local cross-sectional data corresponding to each subset of the abnormal dataset; The statistical value feature calculation and clustering calculation of the local cross-sectional data are performed to obtain several first abnormal features corresponding to each subset of the abnormal dataset, including: The mean, variance, slope, and dispersion of each row of data in the local cross-sectional data are calculated according to a preset statistical value calculation method, and the mean, variance, slope, and dispersion of each row of data are statistically analyzed to obtain the local feature dataset corresponding to the local cross-sectional data. The local feature dataset is clustered using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

2. The method for extracting abnormal features of the daily line loss rate curve of a distribution station area as described in claim 1, characterized in that, The step of performing clustering calculations on the local feature dataset using a preset fuzzy mean clustering algorithm includes: The local feature dataset is clustered using a pre-defined fuzzy mean clustering algorithm; wherein the objective function of the fuzzy mean clustering algorithm is: Where X = {x1, x2, x3, ..., xn}, X ∈ RS, the given number of samples is N, the dimension of the sample space is S, and the number of clusters is C; Meanwhile: in, Membership matrix; It is a matrix consisting of C cluster centers; For the first The nth sample pair Membership degree of a class; For the first One sample; For the first Cluster centers; It is the fuzzy coefficient; Represents sample points To the cluster center Euclidean distance; The clustering results of the fuzzy mean clustering algorithm are evaluated using a preset clustering validity function; wherein, the clustering validity function is: in, For the first One sample; For the first Cluster centers; For the first Cluster centers.

3. The method for extracting abnormal features of the daily line loss rate curve of a distribution station area as described in claim 1, characterized in that, The step of using custom rules to determine the plurality of first abnormal features and filtering out a plurality of second abnormal features includes: Obtain the number of clusters corresponding to each of the plurality of first abnormal features, and sort the number of clusters in ascending order of numerical value; The number of transformer area IDs corresponding to each of the first abnormal features is counted, and duplicates are removed to obtain the number of transformer areas corresponding to each of the first abnormal features. Based on the total number of transformer substations and the number of transformer substations, calculate the abnormal frequency corresponding to each of the first abnormal features; Based on the ascending order, each of the plurality of first abnormal features is judged, and the first abnormal feature corresponding to the abnormal frequency being greater than or equal to a preset threshold is selected as the second abnormal feature, thereby obtaining a plurality of second abnormal features.

4. A system for extracting abnormal features of daily line loss rate curves in a distribution station area, characterized in that, The system includes a dataset construction module, a dataset classification module, a cross-sectional data extraction module, a data clustering calculation module, and a feature extraction module; The dataset construction module is used to construct the daily line loss rate dataset of abnormal distribution substations based on the acquired daily line loss rate curve data and the work orders for handling abnormal line loss in the substation area. The dataset classification module is used to classify the daily line loss rate dataset of the abnormal distribution substations according to the abnormality type of the work order records of the abnormal line loss handling records of the substations, and obtain several subsets of abnormal datasets. The cross-sectional data extraction module is used to extract local cross-sectional data from each subset of the several types of abnormal datasets through a preset scanning window, so as to obtain the local cross-sectional data corresponding to each subset of the abnormal datasets. The data clustering calculation module is used to perform statistical value feature calculation and clustering calculation on the local cross-sectional data to obtain several first abnormal features corresponding to each subset of the abnormal dataset. The feature extraction module is used to perform custom rule judgment on the plurality of first abnormal features, filter out a plurality of second abnormal features, merge and summarize the plurality of second abnormal features, and obtain the abnormal features corresponding to each subset of the plurality of abnormal datasets. The cross-sectional data extraction module includes a scanning unit and a merging unit; The scanning unit is used to extract local cross-sectional data from each subset of the abnormal dataset according to the preset step size and width in the scanning window, and obtain several columns of data. The merging unit is used to merge the several columns of data to obtain local cross-sectional data corresponding to each subset of the abnormal dataset. The data clustering calculation module includes a statistical unit and a clustering unit; The statistical unit is used to calculate the mean, variance, slope and dispersion of each row of data in the local cross-sectional data according to the preset statistical value calculation method, and to calculate the mean, variance, slope and dispersion of each row of data respectively, so as to obtain the local feature dataset corresponding to the local cross-sectional data. The clustering unit is used to perform clustering calculations on the local feature dataset using a preset fuzzy mean clustering algorithm to obtain several first abnormal features corresponding to the local feature dataset.

5. The system for extracting abnormal features of daily line loss rate curves in a distribution station area as described in claim 4, characterized in that, The feature extraction module includes an ascending order unit, a calculation unit, and a feature filtering unit; The ascending unit is used to obtain the number of clusters corresponding to each of the plurality of first abnormal features, and sort the number of clusters in ascending order of numerical value; The calculation unit is used to count the transformer area IDs corresponding to each of the first abnormal features, and to count the duplicates to obtain the number of transformer areas corresponding to each of the first abnormal features; and to calculate the abnormal frequency corresponding to each of the first abnormal features based on the total number of transformer areas and the number of transformer areas. The feature filtering unit is used to determine each of the plurality of first abnormal features according to the ascending order, and to filter out the first abnormal features corresponding to the abnormal frequency being greater than or equal to a preset threshold as the second abnormal features, thereby obtaining a plurality of second abnormal features.

6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the method for extracting abnormal features of the daily line loss rate curve of a distribution station area as described in any one of claims 1-3.

Images