Voltage sag data processing method and device, storage medium and program product

Abstract

The embodiment of the invention provides a voltage sag data processing method and device, a storage medium and a program product. The method comprises the steps that voltage is monitored, and a no-label voltage sag data set is acquired; based on an improved semi-supervised fuzzy C-means algorithm, according to the prior voltage sag data set, determining a category label corresponding to each data in the no-label voltage sag data set, and obtaining a labeled voltage sag data set; and according to the prior voltage sag data set and the labeled voltage sag data set, performing generative adversarial network training to obtain an expanded voltage sag data set. The method is used for achieving the technical effect of effectively expanding the voltage sag data.

Description

Technical Field

[0001] The present application relates to the field, and in particular to a method, device, storage medium and program product for processing voltage sag data. Background Art

[0002] Voltage sag is a phenomenon in which the root mean square value of the power frequency voltage at a certain point in the power system suddenly drops to 0.1 p.u. to 0.9 p.u. and returns to normal after a short duration of 10 ms to 1 min. In high-end manufacturing and large-scale production equipment factories, the direct and indirect losses caused by voltage sags are huge, reaching up to tens of millions of yuan. Therefore, with the rapid development of industrial automation, the market demand for high-quality electric energy further increases. By accurately processing voltage sag data, reliable data support can be provided for subsequent prediction of the working state of the power system based on voltage sag data, thereby saving labor and material costs and improving the stability and reliability of the power system.

[0003] In the prior art, deep learning technology is mostly used to process voltage sag data. However, the current methods for processing voltage sag data have high requirements for the data sample size. Due to the complexity and scarcity of voltage sag data in the distribution network, there is a shortage of samples in the voltage sag data. Therefore, there is a technical problem of scarce samples in the voltage sag data in the prior art. Summary of the Invention

[0004] Embodiments of the present application provide a method, device, storage medium and program product for processing voltage sag data, so as to achieve the technical effect of effectively expanding the voltage sag data.

[0005] In a first aspect, an embodiment of the present application provides a method for expanding voltage sag data, including:

[0006] Monitoring the voltage to obtain an unlabeled voltage sag data set;

[0007] Based on an improved semi-supervised fuzzy C-means algorithm, determining the class label corresponding to each data in the unlabeled voltage sag data set according to the prior voltage sag data set to obtain a labeled voltage sag data set; wherein, each prior voltage sag data in the prior voltage sag data set corresponds to a class label;

[0008] Performing adversarial generative network training according to the prior voltage sag data set and the labeled voltage sag data set to obtain an expanded voltage sag data set.

[0009] In a possible implementation manner, performing adversarial generative network training according to the prior voltage sag data set and the labeled voltage sag data set to obtain an expanded voltage sag data set, including:

[0010] Based on the generation objective function, the generator generates an initial generated voltage sag dataset;

[0011] Merge the data in the prior voltage sag dataset and the labeled voltage sag dataset to obtain a real voltage sag dataset;

[0012] Input the real voltage sag dataset and the initial generated voltage sag dataset into the discriminator, train the discriminator, and obtain the target generated voltage sag dataset;

[0013] According to the real voltage sag dataset and the target generated voltage sag dataset, obtain an augmented voltage sag dataset.

[0014] In a possible implementation, training the discriminator to obtain the target generated voltage sag dataset includes:

[0015] Based on the discriminator objective function, the discriminator determines the probability that each input data is real voltage sag data, and determines the probability that each discriminant data belongs to different class labels;

[0016] Judge whether the probability that each discriminant data is real voltage sag data and the probability that each discriminant data belongs to different class labels meet the first preset condition. If so, determine the target generated voltage sag dataset according to the initial generated voltage sag dataset;

[0017] If not, continue training according to the adversarial generation network until the first preset condition is met.

[0018] In a possible implementation, after obtaining the target generated voltage sag dataset, it includes:

[0019] According to the following formula, obtain the maximum mean difference between the target generated voltage sag dataset and the real voltage sag dataset;

[0020]

[0021] where MMD 2 is the maximum mean difference; P r is the data distribution of the real voltage sag dataset; P g is the data distribution of the target generated voltage sag dataset; k is a fixed kernel function; E （xr～Pr） is the expectation that each real voltage sag data satisfies the P r distribution in the Hilbert space; E （xg～Pg） is the expectation that each target generated voltage sag data satisfies the P g distribution in the Hilbert space; x r ＇ is the replicated data of x r ; x g＇ is x g duplicate data of; E （xr，xr＇）～Pr For every two identical true voltage sag data, it satisfies P in the Hilbert space r expectation of the distribution; E （xg，xg＇）～Pg For every two identical true voltage sag data, it satisfies P in the Hilbert space r expectation of the distribution;

[0022] Judge whether the maximum mean difference is greater than the first preset threshold. If so, adjust the generation objective function and the discriminant objective function, and continue to train according to the generative adversarial network until the maximum mean difference is less than or equal to the first preset threshold.

[0023] In a possible implementation, monitor the voltage to obtain an unlabeled voltage sag data set, including:

[0024] Monitor the voltage, identify the occurrence time of each voltage sag event, and determine the monitoring period of each unlabeled voltage sag data according to the occurrence time;

[0025] According to the preset time interval, obtain multiple sample points of each unlabeled voltage sag data in the monitoring period and the feature information of each sample point; wherein, the feature information of each sample point includes a voltage sag discrimination factor, a voltage sag amplitude, active power, and reactive power;

[0026] In the unlabeled voltage sag data set, each unlabeled voltage sag data is characterized by a feature value; wherein, the feature value is determined according to the feature information of multiple sample points.

[0027] In a possible implementation, the feature value is determined according to the feature information of multiple sample points, including:

[0028] The feature value includes a voltage sag duration, a target voltage sag amplitude, an active power change amount, and a reactive power change amount;

[0029] Among them, according to the voltage sag discrimination factors of multiple sample points, determine the voltage sag duration of each unlabeled voltage sag data;

[0030] According to the voltage sag discrimination factors of multiple sample points and multiple voltage sag amplitudes, determine the target voltage sag amplitude of each unlabeled voltage sag data;

[0031] According to multiple active powers and multiple reactive powers, determine the active power change amount and the reactive power change amount of each unlabeled voltage sag data respectively.

[0032] In a possible implementation manner, based on an improved semi-supervised fuzzy C-means algorithm, according to a prior voltage sag data set, determine the class label corresponding to each data in the unlabeled voltage sag data set, including:

[0033] Perform clustering division on the prior voltage sag data set to determine the number of class labels; among them, there are multiple prior voltage sag data in each class label, and each class label corresponds to a clustering center;

[0034] According to the eigenvalue of multiple prior voltage sag data, determine the initial eigenvalue of the clustering center corresponding to the class label of multiple prior voltage sag data;

[0035] According to the ratio of the prior voltage sag data set to the unlabeled voltage sag data set, determine the balance index;

[0036] Determine the prior membership function of each prior voltage sag data and the target clustering center;

[0037] Determine the unlabeled membership function of each unlabeled voltage sag data and the target clustering center;

[0038] According to the balance index, the membership function of each prior voltage sag data and the target clustering center, the membership function of each unlabeled voltage sag data and the target clustering center, the eigenvalue of each prior voltage sag data, and the eigenvalue of each unlabeled voltage sag data, determine the updated eigenvalue of the target clustering center;

[0039] According to the balance index, multiple prior membership functions, multiple unlabeled membership functions, and the updated eigenvalues of multiple clustering centers, determine the objective function;

[0040] Perform iterative calculation on the objective function until the objective function meets the second preset condition, and / or the number of iterative calculations meets the second preset threshold, complete the iterative calculation, and determine the class label corresponding to each unlabeled voltage sag data.

[0041] In a possible implementation manner, the method further includes:

[0042] Determine the initial membership matrix of the prior voltage sag data set; where the initial membership matrix includes the degree value of each prior voltage sag data belonging to each clustering center;

[0043] According to the eigenvalue of each prior voltage sag data and the initial eigenvalue of each clustering center, determine the Euclidean distance between each prior voltage sag data and each clustering center;

[0044] According to the eigenvalue of each unlabeled voltage sag data and the initial eigenvalue of each clustering center, determine the Euclidean distance between each unlabeled voltage sag data and each clustering center;

[0045] Determine the membership function of each prior voltage sag data with respect to the target cluster center according to the degree value of each prior voltage sag data belonging to each cluster center, the balance index, the Euclidean distance between each prior voltage sag data and the target cluster center, and the Euclidean distance between each prior voltage sag data and each cluster center;

[0046] Determine the membership function of each unlabeled voltage sag data with respect to the target cluster center according to the Euclidean distance between each unlabeled voltage sag data and the target cluster center, and the Euclidean distance between each unlabeled voltage sag data and each cluster center.

[0047] In a possible implementation, after obtaining the labeled voltage sag data set, it includes:

[0048] Obtain the difference determination coefficient of the labeled voltage sag data set and the prior voltage sag data set according to the following formula;

[0049]

[0050] where, GS is the difference determination coefficient; q is the class label, k is the total number of class labels; x i is the eigenvalue of the i-th labeled voltage sag data belonging to the class label q; C q is multiple labeled voltage sag data with the class label q, a total of J; y j is the eigenvalue of the j-th prior voltage sag data belonging to the class label q; P q is multiple voltage sag data with the class label q, a total of Z; c q is the updated eigenvalue of the cluster center of each class label;

[0051] Judge the difference between the difference determination coefficient and 0. If the difference is greater than or equal to the third preset threshold, adjust the balance index, obtain a new objective function, and perform iterative calculation on the new objective function until the difference is less than the third preset threshold.

[0052] In a second aspect, an embodiment of the present application provides a voltage sag state prediction method, including:

[0053] Monitor the voltage, identify voltage sag events, and obtain the voltage sag data of the voltage sag events;

[0054] Input the voltage sag data into the voltage sag state prediction model to obtain the predicted state of the voltage sag event; wherein, the voltage sag state prediction model is trained according to the extended voltage sag data set in the method of any one of claims 1-9, and the voltage sag state prediction model can perform state prediction on the consequences generated after the voltage sag event occurs.

[0055] In a third aspect, an embodiment of the present application provides a processing device for voltage sag data, including:

[0056] An acquisition module, configured to monitor the voltage and acquire an unlabeled voltage sag data set;

[0057] A processing module, configured to determine a class label corresponding to each data in the unlabeled voltage sag data set based on an improved semi-supervised fuzzy C-means algorithm according to a prior voltage sag data set, so as to obtain a labeled voltage sag data set; wherein, each prior voltage sag data in the prior voltage sag data set corresponds to a class label;

[0058] An expansion module, configured to perform adversarial generative network training according to the prior voltage sag data set and the labeled voltage sag data set to obtain an expanded voltage sag data set.

[0059] In a fourth aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0060] The memory stores computer execution instructions;

[0061] The processor executes the computer execution instructions stored in the memory, so that the processor executes the above first aspect, second aspect, various possible implementation manners of the first aspect, and / or various possible implementation manners of the second aspect.

[0062] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored, and when the computer execution instructions are executed by a processor, they are used to implement the above first aspect, second aspect, various possible implementation manners of the first aspect, and / or various possible implementation manners of the second aspect.

[0063] In a sixth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above first aspect, second aspect, various possible implementation manners of the first aspect, and / or various possible implementation manners of the second aspect.

[0064] The method, device, storage medium and program product for processing voltage sag data provided by the embodiments of the present application collect unlabeled voltage sag data by monitoring voltage; use the optimized semi-supervised fuzzy C-means algorithm, combined with the voltage sag data set of known categories, to assign corresponding category labels to each piece of data in the unlabeled voltage sag data set, so as to form a labeled voltage sag data set; further use the voltage sag data set of known categories and the labeled voltage sag data set to perform adversarial generative network training to generate an extended voltage sag data set. This method solves the technical problem of scarce voltage sag data samples in the prior art. On the basis of accurately labeling the voltage sag data, the voltage sag data is effectively and reasonably expanded, increasing the number of samples, thereby providing data support for reasonably and effectively predicting the voltage state of the distribution network according to the voltage sag data. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0066] Figure 1 Flow chart of the method for processing voltage sag data provided by the present application Figure 1 ;

[0067] Figure 2 Flow chart of the method for processing voltage sag data provided by the present application Figure 2 ;

[0068] Figure 3 Flow chart of the adversarial generative network training provided by the present application;

[0069] Figure 4 Flow chart of the method for processing voltage sag data provided by the present application Figure 3 ;

[0070] Figure 5 Flow chart of the voltage sag state prediction method provided by the present application Figure 4 ;

[0071] Figure 6 Structure diagram of the device for processing voltage sag data provided by the present application;

[0072] Figure 7 Hardware diagram of the device for processing voltage sag data provided by the present application.

[0073] Through the above-mentioned accompanying drawings, the clear embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Specific Embodiments

[0074] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of methods and methods consistent with some aspects of the present application as detailed in the appended claims.

[0075] In the prior art, for the processing of voltage sag data, deep learning techniques are mostly used. This method relies on a large number of data samples for model training in order to achieve the purpose of accurately identifying and analyzing voltage sag events. However, due to its complex and variable characteristics and the difficulty in actual monitoring, the voltage sag data in the distribution network often shows a scarce state. This scarcity is not only reflected in the total amount of data, but also in various sag situations and conditions covered by the data. When training a prediction model based on voltage sag data and then predicting the state of voltage sag events according to the prediction model, it often leads to insufficient model training due to the lack of sufficient diverse and representative samples, thereby affecting its recognition accuracy and generalization ability for voltage sag events. Therefore, there is a technical problem of scarce samples in the voltage sag data in the prior art.

[0076] The method, device, storage medium and program product for processing voltage sag data provided by the embodiments of the present application collect unlabeled voltage sag data by monitoring voltage; use an optimized semi-supervised fuzzy C-means algorithm, combined with a voltage sag data set of known categories, to assign corresponding category labels to each piece of data in the unlabeled voltage sag data set, thereby forming a labeled voltage sag data set; further use the voltage sag data set of known categories and the labeled voltage sag data set for adversarial generative network training to generate an extended voltage sag data set. This method solves the technical problem of scarce samples in the voltage sag data in the prior art. On the basis of accurately labeling the voltage sag data, the voltage sag data is effectively and reasonably expanded, increasing the number of samples, thereby providing data support for reasonably and effectively predicting the voltage state of the distribution network based on the voltage sag data.

[0077] The technical solution of the present application and how the technical solution of the present application solves the above technical problems will be described in detail below with specific embodiments. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0078] Figure 1Flow schematic of the method for processing voltage sag data provided by this application Figure 1 , as Figure 1 shown, the method includes:

[0079] S101. Monitor the voltage to obtain an unlabeled voltage sag data set;

[0080] In this embodiment, a power quality detection device is used to detect voltage and current to obtain a plurality of voltage sag data. These voltage sag data are unlabeled data, and the label is a category label.

[0081] S102. Based on the improved semi-supervised fuzzy C-means algorithm, according to the prior voltage sag data set, determine the category label corresponding to each data in the unlabeled voltage sag data set to obtain a labeled voltage sag data set;

[0082] In this embodiment, each prior voltage sag data in the prior voltage sag data set corresponds to a category label; according to the improved semi-supervised fuzzy C-means algorithm, using the prior voltage sag data set to label the unlabeled voltage sag data, this step can make each voltage sag data have a corresponding category label, providing data support for training a prediction model using the labeled voltage sag data subsequently.

[0083] S103. According to the prior voltage sag data set and the labeled voltage sag data set, perform adversarial generative network training to obtain an augmented voltage sag data set.

[0084] In this embodiment, to solve the scarcity of voltage sag data samples, based on the adversarial generative network, with the prior voltage sag data set and the labeled voltage sag data set as data support, an augmented voltage sag data set can be generated, thereby achieving the technical efficiency of increasing the amount of voltage sag data samples.

[0085] The method for processing voltage sag data provided by the embodiment of this application collects unlabeled voltage sag data by monitoring the voltage; uses the optimized semi-supervised fuzzy C-means algorithm and combines the voltage sag data set with known categories to assign corresponding category labels to each data in the unlabeled voltage sag data set, thereby forming a labeled voltage sag data set; further uses the voltage sag data set with known categories and the labeled voltage sag data set to perform adversarial generative network training to generate an extended voltage sag data set. This method solves the technical problem of the scarcity of voltage sag data samples in the prior art, effectively and reasonably expands the voltage sag data on the basis of accurately labeling the voltage sag data, increases the number of samples, and thus provides data support for reasonably and effectively predicting the voltage state of the distribution network according to the voltage sag data.

[0086] Figure 2 Flow schematic of the method for processing voltage sag data provided by this application Figure 2 , as Figure 2 shown. Based on the embodiments, the steps of how to obtain an extended voltage sag data set are described in detail. The method includes: Figure 1 On the basis of the embodiments, the steps of how to obtain an extended voltage sag data set are described in detail. The method includes:

[0087] S201. Based on the generated objective function, the generator generates an initial generated voltage sag data set;

[0088] In this embodiment, the format of the category label is transformed into a category label vector, and category label information is added to each sample, which can further guide the training of the adversarial generation network. It can not only learn the information of multiple category samples at the same time and generate samples with category labels, but also further avoid the problems of gradient disappearance and model collapse. The sample category label is converted into a five-dimensional category label vector h through one-hot encoding, where the category shell includes tripping category and loss category, 10 represents tripping, 01 represents non-tripping, 100 represents zero loss, 010 represents low loss, and 001 represents high loss; the category label vector h = {10100, 10010, 10001, 01100, 01010, 01001}. The generator obtains the initial generated voltage sag data set according to the random noise and multiple category label vectors; where each initial generated voltage sag data corresponds to a category label vector.

[0089] S202. Merge the data in the prior voltage sag data set and the labeled voltage sag data set to obtain a real voltage sag data set;

[0090] In this embodiment, each data in the prior voltage sag data set and the labeled voltage sag data set corresponds to a category label. Similarly, the category label is transformed into a category label vector, and the merged data set is called a real voltage sag data set. The real voltage sag data set is the data basis for data augmentation of voltage sag data by the adversarial generation network.

[0091] S203. Input the real voltage sag data set and the initial generated voltage sag data set into the discriminator;

[0092] S204. Based on the discriminant objective function, the discriminator determines the probability that each input data is real voltage sag data, and determines the probability that each discriminant data belongs to different category labels; determines whether the probability that each discriminant data is real voltage sag data and the probability that each discriminant data belongs to different category labels meet the first preset condition. If so, determine the target generated voltage sag data set according to the initial generated voltage sag data set; if not, continue to train according to the adversarial generation network until the first preset condition is met;

[0093] In this embodiment, Figure 3 is a flowchart for training a generative adversarial network. As Figure 3 shown, the generator generates an initial generated voltage sag dataset, which is input into the discriminator together with the real voltage sag dataset. The task of the discriminator is to distinguish whether the input data is real voltage sag data or fake data generated by the generator. At the same time, the discriminator also determines which class labels these data belong to. Through continuous adversarial training, the generator can learn the distribution characteristics of real voltage sag data, generate more realistic and diverse data, thereby effectively expanding the voltage sag data, reducing the dependence on real data to a certain extent, and lowering the cost of data acquisition.

[0094] S205. Obtain an expanded voltage sag dataset according to the real voltage sag dataset and the target generated voltage sag dataset;

[0095] In this embodiment, after merging the real voltage sag dataset and the target generated voltage sag dataset, an expanded voltage sag dataset is obtained. The expanded voltage sag dataset can increase the number of samples, making the results of data analysis more accurate and reliable. In power system analysis, it provides sufficient sample support for predicting the consequence state of voltage sag events, ensures the effectiveness of subsequent analysis results, reduces the cost of data acquisition, and provides more favorable data support for maintaining the stable operation of the power system and preventing faults in the future.

[0096] S206. Obtain the maximum mean difference between the target generated voltage sag dataset and the real voltage sag dataset; determine whether the maximum mean difference is greater than a first preset threshold. If so, adjust the generative objective function and the discriminative objective function, and continue training according to the generative adversarial network until the maximum mean difference is less than or equal to the first preset threshold.

[0097] In this embodiment, the calculation formula for the maximum mean difference is shown as follows:

[0098]

[0099] where MMD 2 is the maximum mean difference; P r is the data distribution of the real voltage sag dataset; P g is the data distribution of the target generated voltage sag dataset; k is a fixed kernel function; E （xr～Pr） is the expectation of each real voltage sag data satisfying the P r distribution in the Hilbert space; E （xg～Pg） is the expectation of each target generated voltage sag data satisfying the P g distribution in the Hilbert space; x r ＇ is x rReplicated data; x g ＇ For x g Replicated data; E （xr，xr＇）～Pr For every two identical true voltage sag data, they satisfy P in the Hilbert space r Expectation of the distribution; E （xg，xg＇）～Pg For every two identical true voltage sag data, they satisfy P in the Hilbert space r Expectation of the distribution;

[0100] Figure 4 Schematic flow of the method for processing voltage sag data provided by this application Figure 3 As Figure 4 shown, based on the Figure 1 embodiment, steps S101 to S102 are described in detail. Among them, steps S401 to S403 are the process of obtaining the unlabeled voltage sag data samples; steps S404 to S408 are the process of obtaining the labeled voltage sag data set. The method includes:

[0101] S401. Monitor the voltage, identify the occurrence time of each voltage sag event, and determine the monitoring period of each unlabeled voltage sag data according to the occurrence time;

[0102] In this embodiment, a power quality detection device is used to detect the voltage. When the monitored voltage is lower than 0.9 p.u., the power quality monitoring device starts to record the voltage sag event. Optionally, the recorded data is from five cycles before the start point Dks(0.9 p.u.) of the voltage sag event to five cycles after the end point Djs(0.9 p.u.).

[0103] S402. According to the preset time interval, obtain multiple sample points of each unlabeled voltage sag data in the monitoring period and the characteristic information of each sample point;

[0104] In this embodiment, the preset time interval can be set to 10 ms, and the instantaneous voltage and current values are converted into effective values, so as to obtain the voltage sag amplitude data with a 10-ms interval as a sample point. After calculation, the active power and reactive power data are obtained. The characteristic information of each sample point includes the voltage sag discrimination factor, voltage sag amplitude, active power, and reactive power; the voltage sag discrimination factor is used to discriminate whether the current is in the voltage sag state. When the voltage sag amplitude is less than 90%, the value is 1; when the voltage sag amplitude is greater than 90%, the value is 0; the voltage sag amplitude characterizes the reduction amplitude of the voltage compared with the standard voltage at this time. Optionally, each unlabeled voltage sag data can be represented by the following formula:

[0105]

[0106] Among them, C is each unlabeled voltage sag data; Y is the voltage sag discrimination factor; U is the amplitude of the voltage sag; P is the active power; Q is the reactive power; N is the total number of N sample points collected.

[0107] S403. In the unlabeled voltage sag dataset, each unlabeled voltage sag data is characterized according to the eigenvalue.

[0108] In this embodiment, the eigenvalue is determined according to the characteristic information of multiple sample points. According to the characteristic information of multiple sample points in each unlabeled voltage sag data, the determined eigenvalues include the voltage sag duration, the target voltage sag amplitude, the change in active power, and the change in reactive power.

[0109] In a possible implementation, according to the voltage sag discrimination factors of multiple sample points, the voltage sag duration of each unlabeled voltage sag data is determined; according to the voltage sag discrimination factors of multiple sample points and multiple voltage sag amplitudes, the target voltage sag amplitude of each unlabeled voltage sag data is determined; according to multiple active powers and multiple reactive powers, the change in active power and the change in reactive power of each unlabeled voltage sag data are respectively determined. The eigenvalue corresponding to the unlabeled voltage sag data is determined according to the characteristic information of multiple sample points. Optionally, each unlabeled voltage sag data can be expressed as: C = (T, U’, ΔP, ΔQ); where, T is the voltage sag duration; U’ is the target voltage sag amplitude; ΔP is the change in active power; ΔQ is the change in reactive power.

[0110] S404. Cluster and partition the prior voltage sag dataset to determine the number of class labels; according to the eigenvalues of multiple prior voltage sag data, determine the initial eigenvalues of the cluster centers of the class labels corresponding to the multiple prior voltage sag data; according to the ratio of the prior voltage sag dataset and the unlabeled voltage sag dataset, determine the balance index.

[0111] In this embodiment, by clustering and partitioning the prior voltage sag dataset, multiple class labels can be obtained. There are multiple prior voltage sag data in each class label, and each class label corresponds to a cluster center; the mean value of the eigenvalues of multiple prior voltage sag data belonging to the same class label is taken to determine the initial eigenvalues of the cluster centers of the class labels corresponding to the multiple prior voltage sag data; according to the number of samples in the overall sample set obtained from the prior voltage sag dataset and the unlabeled voltage sag dataset, the value of the balance index is determined according to the ratio of the prior voltage sag dataset and the overall sample set. This value characterizes the importance of the prior voltage sag data in the subsequent determination of the objective function.

[0112] S405. Determine the prior membership function of each prior voltage sag data and the target cluster center; determine the unlabeled membership function of each unlabeled voltage sag data and the target cluster center;

[0113] In this embodiment, to determine the prior membership function of each prior voltage sag data and the target cluster center; determine the unlabeled membership function of each unlabeled voltage sag data and the target cluster center, it can be obtained according to the following steps. Step a. Determine the initial membership matrix of the prior voltage sag data set, where the initial membership matrix includes the degree values of each prior voltage sag data belonging to each cluster center. Step b. Determine the Euclidean distance between each prior voltage sag data and each cluster center according to the eigenvalue of each prior voltage sag data and the initial eigenvalue of each cluster center. Step c. Determine the Euclidean distance between each unlabeled voltage sag data and each cluster center according to the eigenvalue of each unlabeled voltage sag data and the initial eigenvalue of each cluster center. Step d. Determine the membership function of each prior voltage sag data and the target cluster center according to the degree value of each prior voltage sag data belonging to each cluster center, the balance index, the Euclidean distance between each prior voltage sag data and the target cluster center, and the Euclidean distance between each prior voltage sag data and each cluster center. The calculation formula is as shown in the following formula:

[0114]

[0115] Where, is the membership function of the j-th prior voltage sag data and the target cluster center i; α is the balance index; f ij is the degree value of the j-th prior voltage sag data belonging to the target cluster center i; d ij is the Euclidean distance between the j-th prior voltage sag data and the target cluster center i; d kj is the Euclidean distance between the j-th prior voltage sag data and each cluster center, and there are c cluster centers in total.

[0116] Step e. Determine the membership function of each unlabeled voltage sag data and the target cluster center according to the Euclidean distance between each unlabeled voltage sag data and the target cluster center, and the Euclidean distance between each unlabeled voltage sag data and each cluster center. The calculation formula is as shown in the following formula:

[0117]

[0118] Where, is the membership function of the j-th unlabeled voltage sag data and the target cluster center i; d ij is the Euclidean distance between the j-th unlabeled voltage sag data and the target cluster center i; d kjis the Euclidean distance between the j-th unlabeled voltage sag data and each cluster center, and there are c cluster centers in total.

[0119] S406. Determine the updated eigenvalue of the target cluster center according to the balance index, the membership function of each prior voltage sag data and the target cluster center, the membership function of each unlabeled voltage sag data and the target cluster center, the eigenvalue of each prior voltage sag data, and the eigenvalue of each unlabeled voltage sag data;

[0120] In this embodiment, the updated eigenvalue of the target cluster center can be determined according to the following formula:

[0121]

[0122] where c (t,...),i is the updated eigenvalue of the target cluster center being i; x (t,…),k represents the k-th sample belonging to the prior voltage sag data set or the unlabeled voltage sag data set; α is the balance index; X L is the prior voltage sag data set; X U is the unlabeled voltage sag data set; is the membership function of the j-th unlabeled voltage sag data and the target cluster center being i; is the membership function of the j-th unlabeled voltage sag data and the target cluster center being i.

[0123] S407. Determine the objective function according to the balance index, multiple prior membership functions, multiple unlabeled membership functions, and the updated eigenvalues of multiple cluster centers; perform iterative calculation on the objective function until the objective function meets the second preset condition, and / or the number of iterative calculations meets the second preset threshold, complete the iterative calculation, and determine the class label corresponding to each unlabeled voltage sag data;

[0124] In this embodiment, the objective function can be determined according to the following formula:

[0125]

[0126] where U is the membership matrix; C is the cluster center matrix; F is the initial membership matrix determined by multiple prior membership functions; c (t,u,p,q),i is the updated eigenvalue of the i-th cluster center; u (t,u,p,q),ij is the membership function of each sample and the target cluster center being i; x (t,u,p,q),j is the eigenvalue of each voltage sag data; b j includes prior voltage sag data and unlabeled voltage sag data.

[0127] 408. Obtain the difference determination coefficient between the labeled voltage sag dataset and the prior voltage sag dataset; judge the difference between the difference determination coefficient and 0. If the difference is greater than or equal to the third preset threshold, adjust the balance index to obtain a new objective function, and perform iterative calculation on the new objective function until the difference is less than the third preset threshold.

[0128] In this embodiment, the calculation formula of the difference determination coefficient is shown as follows:

[0129]

[0130] where GS is the difference determination coefficient; q is the class label, k is the total number of class labels; x i is the eigenvalue of the i-th labeled voltage sag data belonging to the class label q; C q is multiple labeled voltage sag data with the class label q, and there are J in total; y j is the eigenvalue of the j-th prior voltage sag data belonging to the class label q; P q is multiple voltage sag data with the class label q, and there are Z in total; c q is the updated eigenvalue of the clustering center of each class label.

[0131] Figure 5 is the flow chart of the voltage sag state prediction method provided by this application Figure 4 , as Figure 5 shown, this embodiment will elaborate on how to perform state prediction using the augmented voltage sag data. The method includes:

[0132] S501. Monitor the voltage, identify voltage sag events, and obtain the voltage sag data of the voltage sag events;

[0133] In this embodiment, when the monitored voltage is lower than 0.9 p.u., start recording the voltage sag event, and obtain the sag data of the voltage sag event according to the preset time interval;

[0134] S502. Input the voltage sag data into the voltage sag state prediction model to obtain the predicted state of the voltage sag event.

[0135] In this embodiment, the voltage sag state prediction model is trained according to the Figure 1 method in the augmented voltage sag dataset. The voltage sag state prediction model can realize the state prediction of the consequences generated after the voltage sag event. Optionally, through the above-obtained augmented voltage sag dataset, perform deep learning algorithm training on the initial voltage sag state prediction model to obtain the voltage sag state prediction model, so as to obtain the predicted state of the voltage sag event according to the input voltage sag data.

[0136] The voltage sag data processing method provided by the embodiments of the present application monitors the voltage, identifies the occurrence time of each voltage sag event, and determines the monitoring period of each unlabeled voltage sag data according to the occurrence time; obtains multiple sample points of each unlabeled voltage sag data in the monitoring period and the characteristic information of each sample point according to a preset time interval; determines the characteristic value of each unlabeled voltage sag data through the characteristic information of multiple sample points; based on the improved semi-supervised fuzzy C-means algorithm, determines the class label corresponding to each data in the unlabeled voltage sag data set according to the prior voltage sag data set to obtain a labeled voltage sag data set; based on the generated objective function, the generator generates an initial generated voltage sag data set; combines the data in the prior voltage sag data set and the labeled voltage sag data set to obtain a real voltage sag data set; inputs the real voltage sag data set and the initial generated voltage sag data set into the discriminator to train the discriminator to obtain a target generated voltage sag data set; obtains an extended voltage sag data set according to the real voltage sag data set and the target generated voltage sag data set. This method overcomes the problem of scarce voltage sag data samples faced in the current technical field, significantly increases the number of voltage sag data samples, and at the same time ensures the diversity and representativeness of the samples, providing strong data support for reasonably and effectively predicting the voltage state of the distribution network based on voltage sag data, constructing a more robust and accurate prediction model, thereby improving the prediction ability of the voltage state change of the distribution network, and providing important technical support for the stable operation and fault warning of the power system.

[0137] Figure 6 is a structural schematic diagram of the voltage sag data processing device provided by the present application, as Figure 6 shown, the voltage sag data processing device 60 provided in this embodiment includes:

[0138] An acquisition module 601, configured to monitor the voltage and acquire an unlabeled voltage sag data set;

[0139] A processing module 602, configured to determine the class label corresponding to each data in the unlabeled voltage sag data set according to the prior voltage sag data set based on the improved semi-supervised fuzzy C-means algorithm to obtain a labeled voltage sag data set; wherein, each prior voltage sag data in the prior voltage sag data set corresponds to a class label;

[0140] An extension module 603, configured to perform adversarial generation network training according to the prior voltage sag data set and the labeled voltage sag data set to obtain an extended voltage sag data set.

[0141] In a possible implementation manner, the extension module 603 is further configured to:

[0142] Based on the generated objective function, the generator generates an initial generated voltage sag dataset;

[0143] Merge the data in the prior voltage sag dataset and the labeled voltage sag dataset to obtain a real voltage sag dataset;

[0144] Input the real voltage sag dataset and the initial generated voltage sag dataset into the discriminator, train the discriminator, and obtain a target generated voltage sag dataset;

[0145] Obtain an augmented voltage sag dataset according to the real voltage sag dataset and the target generated voltage sag dataset.

[0146] In a possible implementation, the augmentation module 603 is further configured to:

[0147] Based on the discriminative objective function, the discriminator determines the probability that each input data is real voltage sag data, and determines the probability that each discriminative data belongs to different class labels;

[0148] Judge whether the probability that each discriminative data is real voltage sag data and the probability that each discriminative data belongs to different class labels meet the first preset condition. If so, determine the target generated voltage sag dataset according to the initial generated voltage sag dataset;

[0149] If not, continue to train according to the generative adversarial network until the first preset condition is met.

[0150] In a possible implementation, the augmentation module 603 is further configured to:

[0151] Obtain the maximum mean difference between the target generated voltage sag dataset and the real voltage sag dataset according to the following formula;

[0152]

[0153] where MMD 2 is the maximum mean difference; P r is the data distribution of the real voltage sag dataset; P g is the data distribution of the target generated voltage sag dataset; k is a fixed kernel function; E （xr～Pr） is the expectation that each real voltage sag data satisfies the P r distribution in the Hilbert space; E （xg～Pg） is the expectation that each target generated voltage sag data satisfies the P g distribution in the Hilbert space; x r ＇ is the replicated data of x r ; x g ＇ is xg duplicated data; E （xr，xr＇）～Pr For every two identical true voltage sag data, they satisfy P r distribution expectation; E （xg，xg＇）～Pg For every two identical true voltage sag data, they satisfy P r distribution expectation;

[0154] Determine whether the maximum mean difference is greater than a first preset threshold. If so, adjust the generated objective function and the discriminative objective function, and continue training according to the generative adversarial network until the maximum mean difference is less than or equal to the first preset threshold.

[0155] In a possible implementation, the obtaining module 601 is further configured to:

[0156] Monitor the voltage, identify the occurrence time of each voltage sag event, and determine the monitoring period of each unlabeled voltage sag data according to the occurrence time;

[0157] According to a preset time interval, obtain multiple sample points of each unlabeled voltage sag data in the monitoring period and the feature information of each sample point; wherein, the feature information of each sample point includes a voltage sag discrimination factor, a voltage sag amplitude, active power, and reactive power;

[0158] In the unlabeled voltage sag dataset, each unlabeled voltage sag data is characterized by a feature value; wherein, the feature value is determined according to the feature information of multiple sample points.

[0159] In a possible implementation, the obtaining module 601 is further configured to:

[0160] The feature value includes a voltage sag duration, a target voltage sag amplitude, an active power change amount, and a reactive power change amount;

[0161] Wherein, according to the voltage sag discrimination factors of multiple sample points, determine the voltage sag duration of each unlabeled voltage sag data;

[0162] According to the voltage sag discrimination factors of multiple sample points and multiple voltage sag amplitudes, determine the target voltage sag amplitude of each unlabeled voltage sag data;

[0163] According to multiple active powers and multiple reactive powers, respectively determine the active power change amount and the reactive power change amount of each unlabeled voltage sag data.

[0164] In a possible implementation, the processing module 602 is further configured to:

[0165] Cluster and partition the prior voltage sag dataset to determine the number of class labels; among them, there are multiple prior voltage sag data in each class label, and each class label corresponds to a cluster center;

[0166] Determine the initial characteristic values of the cluster centers of the class labels corresponding to the multiple prior voltage sag data according to the characteristic values of the multiple prior voltage sag data;

[0167] Determine the balance index according to the ratio of the prior voltage sag dataset to the unlabeled voltage sag dataset;

[0168] Determine the prior membership function of each prior voltage sag data and the target cluster center;

[0169] Determine the unlabeled membership function of each unlabeled voltage sag data and the target cluster center;

[0170] Determine the updated characteristic values of the target cluster center according to the balance index, the membership function of each prior voltage sag data and the target cluster center, the membership function of each unlabeled voltage sag data and the target cluster center, the characteristic values of each prior voltage sag data, and the characteristic values of each unlabeled voltage sag data;

[0171] Determine the objective function according to the balance index, multiple prior membership functions, multiple unlabeled membership functions, and the updated characteristic values of multiple cluster centers;

[0172] Perform iterative calculation on the objective function until the objective function meets the second preset condition, and / or the number of iterative calculations meets the second preset threshold, complete the iterative calculation, and determine the class label corresponding to each unlabeled voltage sag data.

[0173] In a possible implementation, the processing module 602 is further configured to:

[0174] Determine the initial membership matrix of the prior voltage sag dataset; where the initial membership matrix includes the degree values of each prior voltage sag data belonging to each cluster center;

[0175] Determine the Euclidean distance between each prior voltage sag data and each cluster center according to the characteristic values of each prior voltage sag data and the initial characteristic values of each cluster center;

[0176] Determine the Euclidean distance between each unlabeled voltage sag data and each cluster center according to the characteristic values of each unlabeled voltage sag data and the initial characteristic values of each cluster center;

[0177] Determine the membership function of each prior voltage sag data with respect to the target cluster center according to the degree value of each prior voltage sag data belonging to each cluster center, the balance index, the Euclidean distance between each prior voltage sag data and the target cluster center, and the Euclidean distance between each prior voltage sag data and each cluster center;

[0178] Determine the membership function of each unlabeled voltage sag data with respect to the target cluster center according to the Euclidean distance between each unlabeled voltage sag data and the target cluster center, and the Euclidean distance between each unlabeled voltage sag data and each cluster center.

[0179] In a possible implementation, the processing module 602 is further configured to:

[0180] Obtain the difference determination coefficient between the labeled voltage sag data set and the prior voltage sag data set according to the following formula;

[0181]

[0182] where GS is the difference determination coefficient; q is the class label, k is the total number of class labels; x i is the feature value of the i-th labeled voltage sag data belonging to the class label q; C q is a plurality of labeled voltage sag data with the class label q, and there are J in total; y j is the feature value of the j-th prior voltage sag data belonging to the class label q; P q is a plurality of voltage sag data with the class label q, and there are Z in total; c q is the updated feature value of the cluster center of each class label;

[0183] Judge the difference between the difference determination coefficient and 0. If the difference is greater than or equal to the third preset threshold, adjust the balance index to obtain a new objective function, and perform iterative calculation on the new objective function until the difference is less than the third preset threshold.

[0184] The voltage sag data processing device provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0185] Figure 7 is a hardware schematic diagram of the voltage sag data processing device provided in this application. As Figure 7 shown, the electronic device 70 provided in this embodiment includes: at least one processor 701 and a memory 702. Optionally, the device 70 further includes a communication component 703. Among them, the processor 701, the memory 702, and the communication component 703 are connected through a bus 704.

[0186] In a specific implementation process, at least one processor 701 executes computer-executable instructions stored in a memory 702, so that at least one processor 701 executes the above-mentioned method.

[0187] For the specific implementation process of the processor 701, reference may be made to the above method embodiments. Their implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0188] In the above embodiments, it should be understood that the processor may be a central processing unit (Central Processing Unit, CPU for short), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP for short), application specific integrated circuits (Application Specific Integrated Circuit, ASIC for short), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.

[0189] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0190] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0191] This application also provides a computer program product, including a computer program, which implements the above-mentioned method when executed by a processor.

[0192] This application also provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the processor executes the computer-executable instructions, the above-mentioned method is implemented.

[0193] The above-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disk. The readable storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0194] An exemplary readable storage medium is coupled to the processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in a device.

[0195] The division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, methods, or units, and can be in electrical, mechanical, or other forms.

[0196] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0197] In addition, the functional units in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0198] If a function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, and other various media that can store program codes.

[0199] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When this program is executed, it executes the steps including the above method embodiments; and the aforementioned storage medium includes: ROMs, RAMs, magnetic disks, or optical discs, and other various media that can store program codes.

[0200] Finally, it should be noted that: after considering the specification and practicing the invention disclosed herein, those skilled in the art will readily think of other implementation manners of the present invention. The present invention is intended to cover any variations, uses, or adaptive changes of the present invention. These variations, uses, or adaptive changes follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed in the present invention. It is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for processing voltage sag data, characterized in that: include: Monitor the voltage and obtain a label-free voltage sag dataset; Based on the improved semi-supervised fuzzy C-means algorithm, according to the prior voltage sag data set, the category label corresponding to each data in the unlabeled voltage sag data set is determined to obtain a labeled voltage sag data set; wherein each prior voltage sag data in the prior voltage sag data set corresponds to a category label; According to the prior voltage sag data set and the labeled voltage sag data set, adversarial generative network training is performed to obtain an expanded voltage sag data set.

2. The method according to claim 1, characterized in that: The step of performing adversarial generative network training according to the prior voltage sag dataset and the labeled voltage sag dataset to obtain an expanded voltage sag dataset includes: Based on the generation objective function, the generator generates an initial generation voltage sag data set; Merging the data in the priori voltage sag dataset and the labeled voltage sag dataset to obtain a real voltage sag dataset; Inputting the real voltage sag data set and the initial generated voltage sag data set into a discriminator, training the discriminator, and obtaining a target generated voltage sag data set; An expanded voltage sag dataset is obtained according to the real voltage sag dataset and the target generated voltage sag dataset.

3. The method according to claim 2, characterized in that The step of training the discriminator to obtain a target generated voltage sag data set includes: Based on the discriminant objective function, the discriminator determines the probability that each input data is real voltage sag data, and determines the probability that each discriminated data belongs to a different category label; Determine whether the probability that each discrimination data is real voltage sag data and the probability that each discrimination data belongs to a different category label meet a first preset condition, and if so, determine the target generated voltage sag data set according to the initially generated voltage sag data set; If not, continue training according to the generative adversarial network until the first preset condition is met.

4. The method according to claim 3, characterized in that After obtaining the target generated voltage sag data set, the method further comprises: According to the following formula, the maximum mean difference between the target generated voltage sag data set and the real voltage sag data set is obtained; Among them, MMD 2 is the maximum mean difference; P r is the data distribution of the real voltage sag data set; P g The data distribution of the voltage sag data set is generated for the target; k is a fixed kernel function; E （xr～Pr） For each real voltage sag data, satisfy P in Hilbert space r Expectation of distribution; E （xg～Pg） Generate voltage sag data for each target to satisfy P in Hilbert space g The expectation of the distribution; x r ＇ For x r Copy data of x g ＇ For x g The copy data of E （xr，xr＇）～Pr For every two identical real voltage sag data, P r Expectation of distribution; E （xg，xg＇）～Pg For every two identical real voltage sag data, P r Expectation of distribution; Determine whether the maximum mean difference is greater than a first preset threshold value. If so, adjust the generation objective function and the discrimination objective function, and continue training according to the adversarial generative network until the maximum mean difference is less than or equal to the first preset threshold value.

5. The method according to claim 1, characterized in that: The voltage is monitored to obtain an unlabeled voltage sag dataset, including: Monitor the voltage, identify the occurrence time of each voltage sag event, and determine the monitoring period of each unlabeled voltage sag data according to the occurrence time; According to a preset time interval, a plurality of sample points of each tag-free voltage sag data in the monitoring period and characteristic information of each sample point are obtained; wherein the characteristic information of each sample point includes a voltage sag discrimination factor, a voltage sag amplitude, active power and reactive power; In the unlabeled voltage sag data set, each unlabeled voltage sag data is characterized according to a characteristic value; wherein the characteristic value is determined according to characteristic information of multiple sample points.

6. The method according to claim 5, characterized in that The characteristic value is determined according to characteristic information of a plurality of sample points, including: The characteristic values ​​include voltage sag duration, target voltage sag amplitude, active power change and reactive power change; Wherein, the voltage sag duration of each unlabeled voltage sag data is determined according to the voltage sag discrimination factors of the multiple sample points; Determining a target voltage sag amplitude for each unlabeled voltage sag data according to the voltage sag discrimination factors and the multiple voltage sag amplitudes of the multiple sample points; According to the multiple active powers and the multiple reactive powers, the active power change amount and the reactive power change amount of each tag-free voltage sag data are determined respectively.

7. The method according to claim 1, characterized in that The improved semi-supervised fuzzy C-means algorithm is based on the prior voltage sag data set to determine the category label corresponding to each data in the unlabeled voltage sag data set, including: Clustering the a priori voltage sag data set to determine the number of category labels; wherein each category label contains a plurality of a priori voltage sag data, and each category label corresponds to a cluster center; Determining, according to the characteristic values ​​of the plurality of prior voltage sag data, initial characteristic values ​​of cluster centers of category labels corresponding to the plurality of prior voltage sag data; determining a balance index according to a ratio of the prior voltage sag dataset to the unlabeled voltage sag dataset; Determine a priori membership function of each priori voltage sag data and target cluster center; Determine the unlabeled membership function of each unlabeled voltage sag data and the target cluster center; Determine an updated eigenvalue of the target cluster center according to the balance index, the membership function of each prior voltage sag data and the target cluster center, the membership function of each unlabeled voltage sag data and the target cluster center, the eigenvalue of each prior voltage sag data and the eigenvalue of each unlabeled voltage sag data; Determining an objective function according to the balance index, multiple prior membership functions, multiple unlabeled membership functions, and updated characteristic values ​​of multiple cluster centers; The objective function is iteratively calculated until the objective function meets a second preset condition and / or the number of iterative calculations meets a second preset threshold, the iterative calculations are completed, and the category label corresponding to each unlabeled voltage sag data is determined.

8. The method according to claim 7, characterized in that The method further comprises: Determine an initial membership matrix of a priori voltage sag data set; wherein the initial membership matrix includes a degree value of each priori voltage sag data belonging to each cluster center; Determine the Euclidean distance between each prior voltage sag data and each cluster center according to the eigenvalue of each prior voltage sag data and the initial eigenvalue of each cluster center; According to the eigenvalue of each unlabeled voltage sag data and the initial eigenvalue of each cluster center, the Euclidean distance between each unlabeled voltage sag data and each cluster center is determined; Determine a membership function of each a priori voltage sag data and the target cluster center according to the degree value of each a priori voltage sag data belonging to each cluster center, the balance index, the Euclidean distance between each a priori voltage sag data and the target cluster center, and the Euclidean distance between each a priori voltage sag data and each cluster center; The membership function of each unlabeled voltage sag data and the target cluster center is determined according to the Euclidean distance between each unlabeled voltage sag data and the target cluster center, and the Euclidean distance between each unlabeled voltage sag data and each cluster center.

9. The method according to claim 8, characterized in that After obtaining the labeled voltage sag data set, the method includes: According to the following formula, the difference determination coefficient between the labeled voltage sag dataset and the prior voltage sag dataset is obtained; Where GS is the difference determination coefficient; q is the category label, k is the total number of category labels; x i is the characteristic value of the i-th labeled voltage sag data belonging to the category label q; C q is a number of labeled voltage sag data with category label q, a total of J; y j is the characteristic value of the j-th prior voltage sag data belonging to the category label q; q There are multiple voltage sag data with category label q, and there are Z data in total; c q The updated feature values ​​for the cluster centers of each category label; The difference between the difference determination coefficient and 0 is determined. If the difference is greater than or equal to a third preset threshold, the balance index is adjusted to obtain a new objective function, and the new objective function is iteratively calculated until the difference is less than the third preset threshold.

10. A method for predicting voltage sag state, characterized in that: include: Monitor the voltage, identify a voltage sag event, and obtain voltage sag data of the voltage sag event; The voltage sag data is input into a voltage sag state prediction model to obtain a predicted state of the voltage sag event; wherein the voltage sag state prediction model is The voltage sag state prediction model is obtained by training the expanded voltage sag data set in the method according to any one of claims 1 to 9, and can perform state prediction on the consequences of the voltage sag event.

Images