Partial discharge interference atlas rejection method and device, storage medium and electronic equipment
By extracting and constructing an abnormal feature set of local discharge defects, identifying local discharge maps that are disturbed by external signals, the problem of misidentification of interference maps in the prior art is solved, and the accuracy of GIS equipment operation status evaluation is improved.
Patent Information
- Application Number
- CN202510135752.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-07
- Publication Date
- 2025-05-16
AI Technical Summary
The existing defect identification method based on PRPD maps will misidentify the maps interfered with by external signals as localized defect maps, resulting in errors in the evaluation of the operating status of GIS devices.
By obtaining the local discharge map and defect map set to be identified of the gas insulated metal closed switchgear, sample feature data is extracted, and abnormal feature sets are extracted from the sample feature data of each localized defect, a defect recognition model is constructed, and interference map is identified.
It effectively avoids misidentified interference maps, improves the accuracy of map recognition, and ensures the accuracy of evaluation of the operating status of GIS equipment.
Smart Images
Figure CN120011787A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of power transmission technology, and in particular to a partial discharge interference spectrum rejection method and device, a storage medium and an electronic device. Background Art
[0002] As the operating life of the Gas Insulated Switcher (GIS) equipment increases year by year, different types of defects are inevitably generated inside the equipment, which seriously threatens the reliability of the equipment. In order to ensure the normal operation of the equipment and the safe production of the power system, ultra-high frequency detection means are used to enable maintenance personnel to identify the type of partial discharge in advance, help determine the degree of danger of equipment defects, respond quickly to related faults, and quickly make correct operation and maintenance measures to ensure the normal operation of the system.
[0003] The phase resolved partial discharge (PRPD) spectrum based on UHF detection means contains the statistical information of the partial discharge signal. The existing defect identification methods are mostly based on the PRPD spectrum for identification.
[0004] The current defect identification method based on PRPD spectra is to identify the type of partial discharge defect of the GIS equipment to which the PRPD spectra belong. However, some PRPD spectra are generated when the GIS equipment is interfered by external signals. Such spectra do not belong to the spectra of partial discharge defects of GIS equipment. Traditional identification methods will also identify such spectra as spectra of a certain partial discharge defect, resulting in misidentification and incorrect assessment of the operating status of the GIS equipment. Summary of the invention
[0005] In view of this, an embodiment of the present application provides a partial discharge interference spectrum rejection scheme. Using the scheme provided in the embodiment of the present application, the interference spectrum is rejected as a partial discharge defect spectrum, and the interference spectrum is identified without using a data set of the interference spectrum. Thus, it is possible to avoid misidentifying the interference spectrum as a partial discharge defect spectrum, improve the accuracy of spectrum recognition, and avoid subsequent errors in the evaluation of the operating status of the GIS equipment.
[0006] To achieve the above objectives, the present application provides the following technical solutions:
[0007] The first aspect of the present application discloses a partial discharge interference spectrum rejection method, comprising:
[0008] Obtaining a partial discharge atlas to be identified and a defect atlas set of a gas-insulated metal-enclosed switchgear, wherein the defect atlas set contains a sample atlas of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear;
[0009] Extracting sample feature data of each of the sample graphs; the sample feature data includes features of multiple dimensions;
[0010] Extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and constructing a defect recognition model of each partial discharge defect using the abnormal feature set of each partial discharge defect;
[0011] The partial discharge spectrum is processed using each of the defect recognition models, and a recognition result of the partial discharge spectrum by each of the defect recognition models is output. When each of the recognition results indicates that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum.
[0012] The second aspect of the present application discloses a partial discharge interference spectrum rejection device, comprising:
[0013] An acquisition unit, used for acquiring a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear, wherein the defect spectrum set contains a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear;
[0014] A first extraction unit, configured to extract sample feature data of each of the sample graphs; the sample feature data includes features of multiple dimensions;
[0015] A second extraction unit is used to extract an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and use the abnormal feature set of each partial discharge defect to construct a defect recognition model for each partial discharge defect;
[0016] A processing unit is used to process the partial discharge spectrum using each of the defect recognition models, output a recognition result of each of the defect recognition models on the partial discharge spectrum, and determine that the partial discharge spectrum is an interference spectrum when each of the recognition results characterizes that the partial discharge spectrum is abnormal.
[0017] A third aspect of the present application discloses a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the partial discharge interference spectrum rejection method as described above.
[0018] The fourth aspect of the present application discloses an electronic device, comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to implement the partial discharge interference pattern rejection method as described above when one or more processors execute the one or more instructions.
[0019] Compared with the prior art, this application has the following advantages:
[0020] The present application provides a method and device for rejecting partial discharge interference spectrum, a storage medium and an electronic device, including: based on sample spectrums of various partial discharge defects of gas-insulated metal-enclosed switchgear, extracting an abnormal feature set of each partial discharge defect, using each abnormal feature set to build a defect recognition model for each partial discharge defect, and then using each defect recognition model to recognize the partial discharge spectrum to be recognized, and based on the recognition results output by each defect recognition model, when each recognition result characterizes that the partial discharge spectrum is abnormal, determining that the partial discharge spectrum is an interference spectrum. In the process of identifying the interference spectrum, the present application refuses to recognize the interference spectrum as a partial discharge defect spectrum, and identifies the interference spectrum without using a data set of the interference spectrum, thereby avoiding misidentification of the interference spectrum as a partial discharge defect spectrum, improving the accuracy of spectrum recognition, and avoiding subsequent evaluation errors of the operating status of GIS equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0022] Figure 1 A flow chart of a partial discharge interference spectrum rejection method provided in an embodiment of the present application;
[0023] Figure 2 A flow chart of extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect provided in an embodiment of the present application;
[0024] Figure 3 A flow chart of a method for processing a partial discharge map using each defect recognition model and outputting a recognition result of the partial discharge map by each defect recognition model provided in an embodiment of the present application;
[0025] Figure 4 A flowchart of another partial discharge interference spectrum rejection provided in an embodiment of the present application;
[0026] Figure 5 This is an example diagram of an example of calculating the abnormal score result of the sample to be identified by constructing an isolation forest based on the to-be-identified graph and four sub-datasets in the embodiment of the present application;
[0027] Figure 6This is an example diagram of the abnormal score results of four sub-data sets and the tip corona after the simulation parameters are changed to the sample to be identified belongs to the tip corona spectrum in the embodiment of the present application;
[0028] Figure 7 A schematic diagram of the structure of a partial discharge interference spectrum rejection device provided in an embodiment of the present application;
[0029] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.
[0032] Terminology explanation:
[0033] GIS equipment: Gas Insulated Switcher, refers to gas-insulated metal-enclosed switchgear;
[0034] PRPD: Phase Resolved Partial Discharge, refers to phase resolved partial discharge.
[0035] From the background technology, it can be seen that the existing PRPD spectrum recognition scheme mainly focuses on the identification of typical defects of GIS equipment, but ignores the identification of related spectra of GIS equipment interfered by external signals. This type of spectrum will be mistakenly identified as the spectrum of typical defects of GIS equipment, resulting in identification errors and leading to incorrect assessment of the operating status of GIS equipment.
[0036] The relevant spectrum of GIS equipment interfered by external signals can be called interference spectrum. Interference spectrum is extremely diverse, and it is extremely difficult to directly construct an interference spectrum dataset. How to identify interference spectrum in the absence of interference spectrum data has become an urgent problem to be solved.
[0037] In order to solve the above-mentioned problems, the present application provides a solution for identifying partial discharge spectra, based on sample spectra of various partial discharge defects of gas-insulated metal-enclosed switchgear, extracting abnormal feature sets of each partial discharge defect, using each abnormal feature set to build a defect recognition model for each partial discharge defect, and then using each defect recognition model to identify the partial discharge spectra to be identified, and based on the recognition results output by each defect recognition model, when each recognition result characterizes the partial discharge spectra as abnormal, determining that the partial discharge spectra are interference spectra. In the process of identifying interference spectra, the present application refuses to identify interference spectra as partial discharge defect spectra, and identifies interference spectra without using a data set of interference spectra, thereby avoiding misidentification of interference spectra as partial discharge defect spectra, improving the accuracy of spectra recognition, and avoiding subsequent errors in the evaluation of the operating status of GIS equipment.
[0038] The present invention can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or devices, etc.
[0039] Reference Figure 1 , is a flow chart of a partial discharge interference spectrum rejection method provided in an embodiment of the present application, and is specifically described as follows:
[0040] S101. Obtain a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear, wherein the defect spectrum set includes a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear.
[0041] In the solution provided in the embodiment of the present application, the defect atlas set includes a sample atlas of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear, and further, there are multiple sample atlases for each partial discharge defect. The partial discharge defects include but are not limited to defects such as tip defects, insulation defects, suspended electrodes, and metal particles.
[0042] Each sample map in the partial discharge map to be identified and the defect map set is a two-dimensional map.
[0043] S102, extracting sample feature data of each sample map; the sample feature data includes features of multiple dimensions.
[0044] Use a pre-trained high-dimensional feature extractor to perform feature extraction on each sample map and output sample feature data of each sample map.
[0045] Each sample map is input into a high-dimensional feature extractor, and the high-dimensional feature extractor converts the two-dimensional sample map into a one-dimensional deep high-dimensional feature, thereby obtaining sample feature data of the sample map. Preferably, the sample feature data contains features of 1×M dimensions, where M is a positive integer, and preferably, M is 1024.
[0046] The high-dimensional feature extractor used in this application is a deep high-dimensional feature extractor. The high-dimensional feature extractor can be trained using a supervised algorithm, and the dimensions of the feature data output by the high-dimensional feature extractor are the same.
[0047] S103, extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and using the abnormal feature set of each partial discharge defect to construct a defect recognition model for each partial discharge defect.
[0048] Reference Figure 2 , which is a flowchart of extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect provided in an embodiment of the present application, is specifically described as follows:
[0049] S201. Obtain a feature set of each dimension, where the feature set includes the features of each sample feature data in the dimension.
[0050] The dimensions of the sample feature data output by the high-dimensional feature extractor are all the same. For example, the sample feature data output by the high-dimensional feature extractor are all 1×1024-dimensional data. In other words, the sample feature data contains features of 1×1024 dimensions.
[0051] For each dimension, the features of the dimension are obtained from each sample feature data, and the obtained features are aggregated into a feature set of the dimension.
[0052] Continuing with the above description, 1024 feature sets can be obtained.
[0053] S202: Process each feature set to obtain a feature variance of each feature set.
[0054] For each feature set, variance operation is performed on each feature in the feature set to obtain the feature variance of the feature set. Furthermore, the feature variance is also the feature variance of the dimension corresponding to the feature set.
[0055] S203: Determine abnormal dimensions in each dimension based on each feature variance.
[0056] The target feature variance is determined one by one according to the order of small to large feature variances until the number of determined target feature variances reaches a preset number, and the preset number can be set according to actual needs. For example, the preset number is N, and the value of N is a positive integer.
[0057] For each target feature variance, the dimension corresponding to the target variance is determined as an abnormal dimension, and further, the feature in the abnormal dimension may be an abnormal feature.
[0058] The smaller the feature variance is, the higher the degree of aggregation of the features in the dimension corresponding to the feature variance is.
[0059] S204: Apply the dimension index of each abnormal dimension to generate an abnormal dimension index set.
[0060] The dimension index is the identification mark of the dimension, and each dimension has a corresponding dimension index.
[0061] The dimension index of each abnormal dimension is aggregated to generate an abnormal dimension index set.
[0062] S205. For each dimension index in the abnormal dimension index set, determine the feature corresponding to the dimension index in each sample feature data as the first target feature.
[0063] Using each dimension index in the abnormal dimension index set, each first target feature can be determined in each sample feature data. Furthermore, each dimension index has a corresponding first target feature in each sample feature data. For example, when the dimension index is 1, for each sample feature data, the feature with the dimension index of 1 in the dimension in the sample feature data is determined as the first target feature. Furthermore, the first target feature can be called an abnormal feature.
[0064] Thus, each first target feature can be determined in each sample feature data.
[0065] S206: Aggregate the first target features of each sample feature data into an abnormal feature sample of each sample feature data.
[0066] For each sample feature data, each first target feature in the sample feature data is aggregated to obtain an abnormal feature sample of the sample feature data, that is, the abnormal feature sample contains each first target feature of the sample feature data.
[0067] S207 . For each partial discharge defect, aggregate abnormal feature samples of each sample feature data of the partial discharge defect into an abnormal feature set of the partial discharge defect.
[0068] The partial discharge defect here corresponds to the partial discharge defect in the defect atlas set in step S101. When the defect atlas set contains sample atlases of four types of partial discharge defects, abnormal feature sets of four types of partial discharge defects will be generated here.
[0069] For each partial discharge defect, abnormal feature samples of sample feature data of each sample spectrum corresponding to the partial discharge defect are aggregated into an abnormal feature set of the partial discharge defect.
[0070] In the method provided in the embodiment of the present application, steps S201 to S204 are the process of processing each sample feature data to obtain an abnormal dimension index set, and steps S205 to S207 are the process of using the abnormal dimension index set to extract the abnormal feature set of each partial discharge defect from each sample feature data.
[0071] The characteristic data of each sample in the present application are high-dimensional characteristic data of different partial discharge defects. Not all features in the characteristic data meet the requirements of dispersion and concentration of the same type of spectrum features. The characteristic variance of each dimension is calculated using the characteristic data of each sample. The top N dimensions with high aggregation degree can be determined from each dimension. These N dimensions are used as abnormal dimensions. These N abnormal dimensions are subsequently used for subsequent interference spectrum recognition, and the features of MN dimensions can be eliminated. This process can reduce feature redundancy and greatly improve recognition efficiency.
[0072] Furthermore, after obtaining the abnormal feature set of each partial discharge defect, a defect recognition model of the partial discharge defect is constructed using the abnormal feature set. Exemplarily, the defect recognition model of the partial discharge defect can be constructed using an isolation forest algorithm.
[0073] S104, using each defect recognition model to process the partial discharge spectrum, outputting the recognition result of each defect recognition model on the partial discharge spectrum, and when each recognition result indicates that the partial discharge spectrum is abnormal, determining that the partial discharge spectrum is an interference spectrum.
[0074] Reference Figure 3 , which is a flow chart of a method for processing a partial discharge map using each defect recognition model and outputting a recognition result of the partial discharge map by each defect recognition model provided in an embodiment of the present application, and is specifically described as follows:
[0075] S301. Apply an abnormal dimension index set to obtain abnormal feature data of a partial discharge spectrum, wherein the abnormal feature data includes features of a dimension of each dimension index of the partial discharge spectrum in the abnormal dimension index set.
[0076] The process of obtaining abnormal feature data of a local discharge spectrum is as follows: using a high-dimensional feature extractor to process the local discharge spectrum to obtain spectrum feature data of the local discharge spectrum; determining the features corresponding to each dimension index in the spectrum feature data in the abnormal dimension index set as target features, and aggregating the target features into abnormal feature data; that is, for each dimension index in the abnormal dimension index set, determining the features of the dimension corresponding to the dimension index in the spectrum feature data as target features, and then aggregating the target features into abnormal feature data.
[0077] S302, using each defect recognition model to process the abnormal feature data of the partial discharge spectrum, to obtain the recognition result of each defect recognition model on the partial discharge spectrum, wherein the recognition result includes the abnormal score of the defect recognition model on the partial discharge spectrum.
[0078] The abnormal feature data of the partial discharge spectrum to be identified is input into each defect recognition model, so that each defect recognition model processes the abnormal feature data, thereby outputting a recognition result of the partial discharge spectrum, which includes an abnormal score of the partial discharge spectrum given by the defect recognition model.
[0079] Furthermore, the recognition result also includes content indicating whether the partial discharge spectrum is abnormal.
[0080] When the recognition results output by each defect recognition model all represent that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum. At this time, the partial discharge interference spectrum is rejected. When there is a recognition result output by each defect recognition model that represents that the partial discharge spectrum is not abnormal, the recognition result corresponding to the abnormal score with the smallest numerical value is determined as the target recognition result, and the defect recognition model corresponding to the target recognition result is determined as the target model; the partial discharge spectrum is determined to be the spectrum of the partial discharge defect corresponding to the target model.
[0081] Therefore, the solution provided in the present application can identify the spectra of various partial discharge defects, and can also reject interference spectra, thereby improving the accuracy of identifying the partial discharge spectra to be identified, thereby improving the accuracy of subsequent partial discharge diagnosis results and operating status analysis for GIS equipment.
[0082] Preferably, the defect recognition model outputs the recognition result of the local discharge spectrum such as: the defect recognition model processes the input abnormal feature data and each sample feature data used to construct the defect recognition model, thereby obtaining the abnormal score of the abnormal feature data and the abnormal score of each sample feature data, and averages the abnormal scores of each sample feature data to obtain the score mean, determines the abnormal score with the largest value among the abnormal scores of each sample feature data as the target score, and determines whether the score mean, the target score, and the abnormal score of the abnormal feature data meet the preset abnormal judgment conditions. When the score mean, the target score, and the abnormal score of the abnormal feature data meet the preset abnormal judgment conditions, content characterizing the abnormality of the local discharge spectrum is generated; otherwise, content characterizing that the local discharge spectrum is not abnormal is generated.
[0083] Furthermore, the abnormality determination conditions are as follows: ,in, Anomaly score representing abnormal feature data; represents the target score; represents the mean score; The value can be set according to actual needs, for example, set to 0.3.
[0084] That is, when the abnormal score of the abnormal feature data is greater than the target score and greater than the sum of the mean score and the operation score, it is determined that the mean score, the target score and the abnormal score of the abnormal feature data meet the preset abnormality judgment condition; the operation score is calculated by multiplying the difference between the abnormal score and the target score by the preset parameter (i.e., )get.
[0085] In the method provided in the embodiment of the present application, a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear are obtained, wherein the defect spectrum set contains a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear; sample feature data of each sample spectrum are extracted; the sample feature data include features of multiple dimensions; an abnormal feature set of each partial discharge defect is extracted from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes multiple abnormal feature samples, and a defect recognition model of each partial discharge defect is constructed using the abnormal feature set of each partial discharge defect; each defect recognition model is used to process the partial discharge spectrum, and a recognition result of each defect recognition model for the partial discharge spectrum is output, and when each recognition result characterizes that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum. The present application does not need to construct a data set of interference spectra. By using defect spectra sets of various partial discharge defects, the abnormal feature set of each partial discharge defect is extracted, and then each abnormal feature set is used to construct a defect recognition model for each partial discharge defect. Each defect recognition model is used to process the partial discharge spectra to be identified, and when each defect recognition model outputs an identification result characterizing the abnormality of the partial discharge spectrum, the partial discharge spectrum is determined as an interference spectrum, thereby completing the rejection of the interference spectrum, avoiding the identification of the interference spectrum as a spectrum of partial discharge defects, avoiding spectrum misidentification, and improving the subsequent analysis results of the operating status of the GIS equipment.
[0086] As can be seen from the background technology, due to the diversity of interference spectra of GIS equipment, it is difficult to construct a data set with a large amount of data to use a model for identification, so the data set of interference spectra is scarce. Therefore, in the solution provided in this application, a PRPD interference spectrum rejection scheme that meets the recognition accuracy and efficiency under unsupervised conditions is proposed. While identifying the interference spectrum, this scheme can take into account the spectrum recognition of typical partial discharge defects of GIS equipment.
[0087] Reference Figure 4 , is another flowchart of partial discharge interference spectrum rejection provided by an embodiment of the present application, and is specifically described as follows:
[0088] S1: Train a high-dimensional feature extractor.
[0089] Preferably, the high-dimensional feature extractor is constructed based on the Resnet network. Preferably, when the GIS equipment has four typical partial discharge defects, a defect recognition network for the four types of partial discharge defects is constructed based on the Resnet network, and then the PRPD atlas dataset Dataset_old containing the four types of partial discharge defects is input into the constructed defect recognition network to train the partial discharge defect recognition network. The structure and corresponding parameters of the trained partial discharge defect recognition network are defined as F, the classifier structure and parameters are defined as C, and the trained F is output, that is, the output trained F is the high-dimensional feature extractor.
[0090] Exemplarily, the process of training a defect recognition network based on a Resnet network is as follows from 2.1) to 2.3):
[0091] 2.1) Input the atlas dataset Dataset_old, which includes multiple PRPD atlas samples Samples and label data Labels corresponding to each PRPD atlas sample, where the label data Labels represents the partial discharge defect type of the PRPD atlas sample. Samples and Labels in the atlas dataset Dataset_old exist in pairs. Based on the total number Ntotal of PRPD atlas samples in the atlas dataset Dataset_old, it is divided into the training set Data_train and the test set Data_test in a ratio of 8:2. The loss function adopts the cross entropy loss function, and the specific expression is as follows:
[0092] ;
[0093] where y i represents the true value of the i-th PRPD spectrum sample, z i is the feature of the i-th PRPD spectrum sample extracted by the high-dimensional feature extractor. i ) is the softmax function.
[0094] 2.2) Based on the above loss function definition, the input data of the defect recognition network is normalized, the three-dimensional data of the defect recognition network in Data_train is transformed and reshaped to obtain 4D-double data, and the label data Labels and PRPD map samples Samples are transformed into rows and columns to form the correct input format data that meets the input format of the defect recognition network.
[0095] 2.3) In order to improve the training efficiency and speed, the training epoch can be defined as 50, Batch_size is defined as 32, the number of PRPD atlas and label pairs in the training set is Ntotal*0.8, and Ntotal*0.8 are randomly divided into Ntotal*0.8 / 32 groups of PRPD atlas sample-label data pairs according to the size of Batch_size. The divided samples are input into the feature extractor F and classifier C for training and testing according to the set epoch and Batch_size, and the changes in the loss function, training set and test set are calculated. The accuracy rate acc_train related to the training set and the accuracy rate acc_test related to the test set can be specifically expressed as:
[0096] ;
[0097] ;
[0098] In the formula, is the number of all training samples, is the number of samples whose classification results are consistent with the labels during training. is the number of all test samples, It is the number of samples whose classification results are consistent with the labels during the test process.
[0099] Preferably, when the number of training epochs reaches a preset number, the training of the defect recognition network is completed, and a high-dimensional feature extractor can be output at that time.
[0100] For example, the description of each parameter in the defect recognition network structure after training is shown in Table 1:
[0101] Table 1
[0102]
[0103] Table 1 above illustrates the specific network structure of the 18-layer Resnet feature extraction network, where Output size indicates the size of the data after passing through the layer, Layer name is the name of each layer, conv-related layers in the name are all convolutional layers, and pooling is the pooling layer. Stride indicates the step size of the operation, 4-d fc indicates that the network can divide the final result into 4 categories, max pool is the maximum pooling, average pool is the average pooling, and softmax indicates the softmax function used for classification calculation.
[0104] Furthermore, since the PRPD interference map and the four types of typical defect maps have similar generation principles, consistent map backgrounds, and unified color rendering methods for partial discharge pixels, the feature extractors of the interference map and the four types of typical partial discharge maps can be shared. To train the feature extractor, this application is based on the existing four types of typical partial discharge defect map datasets, and trains a map recognition model for four types of typical partial discharge defects based on the Resnet basic framework, retaining the structure and parameters of the feature extractor in the model, thereby obtaining a high-dimensional feature extractor. Here, the supervised high-dimensional feature extractor is directly used for interference map feature extraction, which can ensure the accuracy of subsequent interference map recognition.
[0105] S2: Feature extraction is performed on the PPRD spectrum sample S to be identified and the spectrum dataset Dataset_old containing four types of typical partial discharge defects.
[0106] The PPRD spectrum sample S to be identified in this step is equivalent to Figure 1The partial discharge spectrum to be identified mentioned in S1; the spectrum dataset Dataset_old in this step can be considered as the spectrum dataset Dataset_old mentioned in S1. The spectrum dataset Dataset_old here is equivalent to Figure 1 The defect atlas mentioned in .
[0107] The PRPD spectrum sample S to be identified and the data in the spectrum data set Dataset_old are input into the high-dimensional feature extractor F; further, the PRPD spectrum sample S to be identified and the data in the spectrum data set Dataset_old input into the high-dimensional feature extractor can be regarded as PRPD spectra.
[0108] The high-dimensional feature mentioner transforms the input PRPD spectrum from a two-dimensional spectrum into a one-dimensional deep high-dimensional feature, where the deep high-dimensional feature after F is 1×1024 dimensions, and outputs the deep high-dimensional feature data of the PRPD spectrum; further, the deep high-dimensional feature data of the PPRD spectrum sample S to be identified can be represented as Sh, and the deep high-dimensional feature data set of the four types of partial discharge defects is represented as the Dataset_new data set; wherein, the deep high-dimensional feature data set of the spectrum data set Dataset_old contains the deep high-dimensional feature data of each PRPD spectrum sample in the spectrum data set Dataset_old, and the PRPD spectrum sample is equivalent to Figure 1 The sample atlas mentioned in the above article has a deep high-dimensional feature data equivalent to Figure 1 The sample feature data of the sample map mentioned in .
[0109] Using the trained high-dimensional feature extractor, deep high-dimensional feature extraction is performed on the atlas dataset containing four types of typical partial discharge defects and the PPRD atlas samples to be identified. The trained high-dimensional feature extractor can make the atlases of the same defects in the atlas dataset containing four types of typical partial discharge defects more concentrated in the deep feature space, while the distances between atlases of different types are far apart. Therefore, in the deep high-dimensional feature space, it is beneficial to distinguish the distance between the subsequent interference atlas and the atlases of each defect type, which can further improve the recognition effect.
[0110] S3: Abnormal dimension selection.
[0111] Each deep high-dimensional feature data in the Dataset_new dataset is used as each sample, and each sample has a deep high-level feature of 1×1024 dimensions (equivalent to the features mentioned above); for each dimension, the deep high-level features of all samples in that dimension are calculated to obtain the feature variance of each dimension; after obtaining the feature variance of each dimension, the feature variance of each dimension is selected in order from small to large, and the dimension corresponding to the selected feature variance is determined as the abnormal dimension, and the abnormal dimension index set is generated based on the dimension index of each abnormal dimension, such as: {I1, I2, I3, ..., I N}, where I N Represents the dimension index of the Nth abnormal dimension, where N is a positive integer.
[0112] After the PRPD spectrum passes through the high-dimensional feature extractor, deep high-dimensional feature data is obtained. Each deep high-dimensional feature data has 1024 features. Not all of the 1024 features meet the requirements of scattered and concentrated features of the same type of spectrum. Therefore, the top N features with the highest aggregation degree are found from the 1024 deep high-dimensional features as the feature dimensions for subsequent abnormal target identification. The remaining 1024-N features can be eliminated as unsuitable. This process can reduce feature redundancy in the sample and greatly improve recognition efficiency.
[0113] S4: Build an efficient defect recognition model for interference pattern rejection and pattern recognition of four typical partial discharge defects.
[0114] Furthermore, in order to specifically illustrate the process of building a defect recognition model, the defect recognition model here is illustrated using isolation forest as a specific example, and the process of building a defect recognition model using other recognition algorithms is not repeated here.
[0115] Input the PRPD spectrum sample Sh to be identified, the deep high-dimensional feature dataset of four types of partial discharge defects represented as Dataset_new dataset, the dimension index set of N abnormal dimensions {I1, I2, I3, ..., I N}. Select the 1024-dimensional features of Sh and {I1, I2, I3, ..., I N The N features corresponding to} are abnormal feature data Se (equivalent to the abnormal feature data of the partial discharge spectrum mentioned above), that is, the abnormal feature data Se includes the abnormal feature data located at I in Sh MThe features of the corresponding dimension, the abnormal feature data Se contains N features; in a similar way, the abnormal feature sample Data_e of each deep high-dimensional feature number in Datasets_new is selected, and the abnormal feature sample Data_e contains N features. The abnormal feature sample Data_e contains the features of the dimensions corresponding to each dimension index in the dimension index set in the deep high-dimensional feature number.
[0116] Furthermore, there are four types of typical partial discharge defects, such as metal particles, tip corona, insulation, and suspended electrodes. Therefore, different partial discharge defects have corresponding abnormal feature sets, and each abnormal feature set is represented by Data_e1, Data_e2, Data_e3, and Data_e4 respectively. The four abnormal feature sets and the abnormal feature data Se of the PRPD spectrum sample to be identified are used to form four groups of data, which are represented as: {Data_e1, Se}, {Data_e2, Se}, {Data_e3, Se}, and {Data_e4, Se} respectively. These four groups of data are used to construct four isolated forests iForest1, iForest2, iForest3, and iForest4 respectively.
[0117] After the interference map passes through the high-dimensional feature extractor, it has abnormal characteristics among the selected N abnormal features compared with the existing four types of typical defect map features, that is, the interference map is far away from the relatively clustered four types of typical defect maps. Therefore, this patent adopts the isolation forest detection method to determine the sparsity of the sample, and constructs four isolation forests for the sample to be identified and all sub-datasets in the four types of typical defect data sets.
[0118] The process of constructing an isolation forest is as follows: 3.1) to 3.2)
[0119] 3.1) Constructing an isolation tree: The pseudo code of the isolation tree construction method is as follows. Q sample points are randomly selected from the sample set as the input samples of the isolation forest, denoted as I'. The maximum length of each isolation tree is defined as l, and the selected N abnormal feature sets are defined as B.
[0120] Pseudo code of the isolation tree construction method:
[0121] iTree (I', h)
[0122] Input: I'-input data, h-iTree height, B-selected abnormal features, l-maximum tree height limit
[0123] Output: iTree
[0124] if |I'| ≤ 1 or h>l then
[0125] Return exNode{Size ← |I'|}
[0126] else
[0127] Randomly select a feature b ∈ B from B
[0128] Select a value a between the maximum and minimum values in the bth feature of I'
[0129] I' l ← filter(I', b < a)
[0130] I' r ← filter(I', b ≥ a)
[0131] Return inNode{Left ← iTree(I' l , h + 1, l),
[0132] Right ← iTree(I'r, h + 1, l),
[0133] Split band ← b,
[0134] Split Value ← a}
[0135] End if
[0136] The above is the pseudo code of the isolation tree construction method.
[0137] 3.2) Constructing an isolation forest: An isolation forest is composed of multiple isolated trees. The pseudo code for constructing an isolation forest is as follows. Where Iall' represents the entire input sample, n_t represents the number of isolated trees contained in an isolation forest, and n_s represents the number of samples contained in each isolated tree.
[0138] The pseudo code for constructing an isolation forest is as follows:
[0139] Algorithm 1: iForest (Iall', n_t, n_s)
[0140] Input: Iall'-input data, n_t-number of iTrees, n_s -number of samples in iTree
[0141] Output: n_t iTrees
[0142] 1: Initialize Forest
[0143] 2: for i = 1 to n_t do
[0144] 3: sub-I' ← sample (Iall', n_s)
[0145] 4: Forest ← Forest U iTree (sub-I', 0)
[0146] 5: end for
[0147] 6: Return to Forest
[0148] The above is the pseudo code for the construction method of the isolation forest.
[0149] S5: Interference spectrum rejection and formulation of identification rules for four types of typical partial discharge spectra.
[0150] Four groups of isolation forests are input, and the anomaly score of each sample in the isolation forest is calculated. When Se is judged as an abnormal sample in all four isolation forests, the PRPD spectrum to be identified is an interference spectrum. Otherwise, the PRPD spectrum to be identified is the partial discharge defect of this type in which the anomaly score of Se is the smallest.
[0151] Exemplarily, the calculation process of the abnormality score of the sample to be identified by the isolation forest is described as follows:
[0152] Each data input into the isolation forest is regarded as a sample to be identified. For example, the PRPD spectrum to be identified is a sample to be identified, and the spectrum in the spectrum data set containing four typical partial discharge defects is also a sample to be identified. When the input is feature data, the input abnormal feature data Se is a sample, and each abnormal feature sample in the input Data_e is also a sample.
[0153] The path length l(d j ), calculate the anomaly score Sco of the i-th sample point in the isolation forest i The expression is as follows:
[0154] ;
[0155] Where q is the total number of samples required to construct the isolation forest model. For iForest1, iForest2, iForest3, and iForest4, the value of q is the total number of sample points of the q1, q2, q3, and q4 models, respectively. j )) is the length of all paths l(d j ), c(q) is the average value of the tree height.
[0156] Among them, the expression of c(q) is as follows:
[0157] ;
[0158] ;
[0159] Where: ξ is the Euler constant, which is 0.5772156649.
[0160] In the above way, the anomaly score of each sample input into the isolation forest can be calculated. The anomaly score here is equivalent to the anomaly score above.
[0161] Furthermore, taking the iForest1 constructed by {Data_e1, Se} as an example for identification, if the sample Se to be identified is an abnormal sample, it satisfies the following expression:
[0162] ;
[0163] in is the maximum value of the abnormality scores of all samples in Data_e1, is the average of all sample anomaly scores in Data_e1, and λ is a user-defined value, which can be set to 0.3 based on experience.
[0164] Abnormal score in Se When the above expression is satisfied, iForest1 outputs the result representing the abnormality of the PRPD spectrum to be identified. When each iForest outputs the result representing the abnormality of the PRPD spectrum to be identified, the PRPD spectrum to be identified is determined to be an interference spectrum. Otherwise, the smallest score among the abnormality scores of the PRPD spectrum to be identified output by each iForest is determined as the target score, and the PRPD spectrum to be identified is determined as the spectrum of the partial discharge defect of the iForest corresponding to the target score.
[0165] In the absence of PRPD interference spectrum data sets, this application only realizes PRPD spectrum rejection based on four types of typical partial discharge defect data sets. Existing PRPD interference spectrum recognition methods all need to collect PRPD interference spectra and use the collected PRPD interference spectra to train the corresponding recognition model. However, in practical applications, the diversity of interference spectra is extremely strong, and the collected interference spectrum data sets are difficult to cover all interference types. In addition, due to the diversity of interference, the workload of collecting spectra is very large, and it is difficult to collect comprehensive interference spectra, resulting in a small amount of interference spectrum data. Therefore, the interference spectrum recognition model trained based on the interference data set has a poor recognition rate in the test phase. This application can improve the accuracy and efficiency of unsupervised recognition. The PRPD spectrum samples to be identified are input into the feature extractor trained based on four types of typical partial discharge data sets to obtain the deep high-dimensional features of the data, and the classifier is replaced by the proposed efficient isolation forest, which can significantly improve the accuracy of the recognition results and the recognition efficiency.
[0166] Furthermore, the present application also provides example contents for simulation, the specific contents are as follows:
[0167] Simulation conditions:
[0168] The given PRPD sample to be identified is a PRPD interference map. Each subclass of the four typical defect map data sets has 400 samples, that is, each partial discharge defect has 400 samples. The isolation forest constructed by each subclass data set and the sample to be identified has 401 samples. The parameters for constructing the four isolation forests are as follows: after all maps have been trained with the feature extractor, they have 1024 feature values, the number of abnormal features is defined as 16, and 16 abnormal features are selected from the 1024 features. The final abnormal score result is determined by the average of 10 isolation forests, and each isolation forest contains 4 isolated trees.
[0169] Simulation results analysis:
[0170] Reference Figure 5 , is an example diagram of calculating the abnormal score results of the PRPD sample to be identified based on the isolation forest constructed based on the PRPD sample to be identified and the four sub-datasets in an embodiment of the present application, where the 401st sample in the figure is the PRPD sample to be identified. Figure 5 The blue lines in (a), (b), (c), and (d) respectively represent the abnormal scores of the PRPD sample to be identified and the samples in the four sub-datasets of tip corona, metal particles, insulation, and suspended electrode. The red lines represent the abnormal scores of the PRPD sample to be identified. It is obvious that the PRPD sample to be identified meets the abnormal characteristics compared with the samples in the four sub-datasets, and is therefore identified as an interference spectrum.
[0171] See also Figure 6 , Figure 6 This is an example diagram of the abnormal score results of four sub-data sets and the PRPD sample to be identified after the PRPD sample to be identified in the simulation parameters is changed to belong to the tip corona spectrum in the embodiment of the present application, and the other simulation parameters remain unchanged. Figure 6 The blue lines in (a), (b), (c), and (d) represent the abnormal scores of the PRPD sample to be identified and the samples in the four sub-datasets of tip corona, metal particles, insulation, and suspended electrodes, respectively. The red lines represent the abnormal scores of the samples to be identified. It is obvious that in the tip sub-dataset, the scores of the PRPD samples to be identified are integrated into the sub-dataset, while compared with the other three sub-datasets, the identified samples have abnormal characteristics. Therefore, the identified samples are judged to be tip corona discharge.
[0172] In short, the present invention relates to the field of PRPD spectrum recognition of GIS partial discharge, and specifically to an efficient unsupervised PRPD interference spectrum rejection method. In view of the difficulty in constructing a diverse dataset of PRPD interference spectra, this patent is based on the traditional four-category typical defect PRPD spectrum dataset Dataset_old, where the defect types include tip corona, suspended electrode, metal particles, and insulation, and trains four types of typical defect Resnet recognition networks; retains the feature extractor structure and parameters F of the trained Resnet typical partial discharge defect recognition network, and inputs the traditional PRPD dataset into the trained feature extractor F to obtain a high-dimensional dataset Dataset_new; inputs the original PRPD spectrum S to be identified into F to obtain a high-dimensional feature Sh; based on Sh and the high-dimensional data features in Dataset_new, replaces the classifier in the traditional recognition network with an improved efficient isolation forest distance measurement method to realize PPRD typical insulation spectrum recognition and interference spectrum rejection. The present invention constructs an efficient PRPD interference spectrum rejection method, which can realize interference rejection and typical spectrum recognition without interference datasets, and provide technical support for partial discharge diagnosis and operation status monitoring of GIS equipment.
[0173] and Figure 1 In contrast to the method shown in the figure, the embodiment of the present application provides a partial discharge interference spectrum rejection device, which is used to support Figure 1 Specific implementation of the method shown.
[0174] Reference Figure 7 , is a schematic diagram of the structure of a partial discharge interference spectrum rejection device provided in an embodiment of the present application, and is specifically described as follows:
[0175] An acquisition unit 501 is used to acquire a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear, wherein the defect spectrum set includes a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear;
[0176] A first extraction unit 502 is used to extract sample feature data of each sample map; the sample feature data includes features of multiple dimensions;
[0177] A second extraction unit 503 is used to extract an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and construct a defect recognition model of each partial discharge defect using the abnormal feature set of each partial discharge defect;
[0178] The processing unit 504 is used to process the partial discharge spectrum using each of the defect recognition models, output recognition results of each of the defect recognition models on the partial discharge spectrum, and determine that the partial discharge spectrum is an interference spectrum when each of the recognition results indicates that the partial discharge spectrum is abnormal.
[0179] In the device provided in the embodiment of the present application, a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear are obtained, wherein the defect spectrum set contains a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear; sample feature data of each sample spectrum are extracted; the sample feature data include features of multiple dimensions; an abnormal feature set of each partial discharge defect is extracted from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes multiple abnormal feature samples, and a defect recognition model of each partial discharge defect is constructed using the abnormal feature set of each partial discharge defect; each defect recognition model is used to process the partial discharge spectrum, and a recognition result of each defect recognition model on the partial discharge spectrum is output, and when each recognition result characterizes that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum. The present application does not need to construct a data set of interference spectra. By using defect spectra sets of various partial discharge defects, the abnormal feature set of each partial discharge defect is extracted, and then each abnormal feature set is used to construct a defect recognition model for each partial discharge defect. Each defect recognition model is used to process the partial discharge spectra to be identified, and when each defect recognition model outputs an identification result characterizing the abnormality of the partial discharge spectrum, the partial discharge spectrum is determined as an interference spectrum, thereby completing the rejection of the interference spectrum, avoiding the identification of the interference spectrum as a spectrum of partial discharge defects, avoiding spectrum misidentification, and improving the subsequent analysis results of the operating status of the GIS equipment.
[0180] In another embodiment provided by the present application, the first extraction unit of the device performs a process of extracting sample feature data of each sample atlas, including:
[0181] Use a pre-trained high-dimensional feature extractor to perform feature extraction processing on each of the sample maps, and output sample feature data of each of the sample maps.
[0182] In another embodiment provided by the present application, the second extraction unit of the device performs a process of extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, including:
[0183] Processing each of the sample feature data to obtain an abnormal dimension index set, wherein the abnormal dimension index set includes a dimension index determined as an abnormal dimension in each of the dimensions;
[0184] For each of the partial discharge defects, the abnormal dimension index set is used to extract an abnormal feature set from each sample feature data of the partial discharge defect.
[0185] In another embodiment provided by the present application, the second extraction unit of the device performs a process of processing each of the sample feature data to obtain an abnormal dimension index set, including:
[0186] Acquire a feature set of each dimension, wherein the feature set includes features of each sample feature data in the dimension;
[0187] Processing each of the feature sets to obtain a feature variance of each of the feature sets;
[0188] Based on each of the feature variances, determining an abnormal dimension in each of the dimensions;
[0189] The dimension index of each of the abnormal dimensions is applied to generate an abnormal dimension index set.
[0190] In another embodiment provided by the present application, the second extraction unit of the device performs a process of extracting an abnormal feature set from each sample feature data of the partial discharge defect using the abnormal dimension index set for each partial discharge defect, including:
[0191] For each dimension index in the abnormal dimension index set, determining a feature corresponding to the dimension index in each of the sample feature data as a first target feature;
[0192] Aggregating each first target feature of each sample feature data into an abnormal feature sample of each sample feature data;
[0193] For each partial discharge defect, abnormal feature samples of each sample feature data of the partial discharge defect are aggregated into an abnormal feature set of the partial discharge defect.
[0194] In another embodiment provided by the present application, the processing unit of the device executes a process of processing the partial discharge map using each of the defect recognition models and outputting a recognition result of the partial discharge map by each of the defect recognition models, including:
[0195] Applying the abnormal dimension index set, obtaining abnormal feature data of the partial discharge map, wherein the abnormal feature data includes features of the dimension of each dimension index of the partial discharge map in the abnormal dimension index set;
[0196] Each defect recognition model is used to process the abnormal feature data of the partial discharge map to obtain a recognition result of each defect recognition model on the partial discharge map, wherein the recognition result includes an abnormality score of the partial discharge map by the defect recognition model.
[0197] In another embodiment provided by the present application, the device further includes:
[0198] A determination unit is used to determine the defect recognition model corresponding to the abnormality score with the smallest value as the target model when there is an identification result indicating that the partial discharge spectrum is not abnormal; and determine that the partial discharge spectrum is a spectrum of partial discharge defects corresponding to the target model.
[0199] Although the present invention depicts operations in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.
[0200] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.
[0201] An embodiment of the present invention further provides a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the above-mentioned partial discharge interference spectrum rejection method.
[0202] The embodiment of the present invention further provides an electronic device, the structural diagram of which is shown in FIG. Figure 8 As shown, it specifically includes a memory 601 and one or more instructions 602, wherein the one or more instructions 602 are stored in the memory 601 and are configured to be executed by one or more processors 603 to perform the following operations:
[0203] Obtaining a partial discharge atlas to be identified and a defect atlas set of a gas-insulated metal-enclosed switchgear, wherein the defect atlas set contains a sample atlas of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear;
[0204] Extracting sample feature data of each of the sample graphs; the sample feature data includes features of multiple dimensions;
[0205] Extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and constructing a defect recognition model of each partial discharge defect using the abnormal feature set of each partial discharge defect;
[0206] The partial discharge spectrum is processed using each of the defect recognition models, and a recognition result of the partial discharge spectrum by each of the defect recognition models is output. When each of the recognition results indicates that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum.
[0207] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions.
[0208] The specific implementation processes and derivative methods of the above-mentioned embodiments are all within the protection scope of the present invention.
[0209] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0210] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0211] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for rejecting partial discharge interference patterns, characterized in that: include: Obtaining a partial discharge atlas to be identified and a defect atlas set of a gas-insulated metal-enclosed switchgear, wherein the defect atlas set contains a sample atlas of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear; Extracting sample feature data of each of the sample graphs; the sample feature data includes features of multiple dimensions; Extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and constructing a defect recognition model of each partial discharge defect using the abnormal feature set of each partial discharge defect; The partial discharge spectrum is processed using each of the defect recognition models, and a recognition result of the partial discharge spectrum by each of the defect recognition models is output. When each of the recognition results indicates that the partial discharge spectrum is abnormal, the partial discharge spectrum is determined to be an interference spectrum.
2. The method according to claim 1, characterized in that The extracting of sample feature data of each sample atlas includes: Use a pre-trained high-dimensional feature extractor to perform feature extraction processing on each of the sample maps, and output sample feature data of each of the sample maps.
3. The method according to claim 1, characterized in that The step of extracting an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect comprises: Processing each of the sample feature data to obtain an abnormal dimension index set, wherein the abnormal dimension index set includes a dimension index determined as an abnormal dimension in each of the dimensions; For each of the partial discharge defects, the abnormal dimension index set is used to extract an abnormal feature set from each sample feature data of the partial discharge defect.
4. The method according to claim 3, characterized in that The processing of each of the sample feature data to obtain an abnormal dimension index set includes: Acquire a feature set of each dimension, wherein the feature set includes features of each sample feature data in the dimension; Processing each of the feature sets to obtain a feature variance of each of the feature sets; Based on each of the feature variances, determining an abnormal dimension in each of the dimensions; The dimension index of each of the abnormal dimensions is applied to generate an abnormal dimension index set.
5. The method according to any one of claims 3 to 4, characterized in that: For each of the partial discharge defects, using the abnormal dimension index set, extracting an abnormal feature set from each sample feature data of the partial discharge defect, including: For each dimension index in the abnormal dimension index set, determining a feature corresponding to the dimension index in each of the sample feature data as a first target feature; Aggregating each first target feature of each sample feature data into an abnormal feature sample of each sample feature data; For each partial discharge defect, abnormal feature samples of each sample feature data of the partial discharge defect are aggregated into an abnormal feature set of the partial discharge defect.
6. The method according to claim 3, characterized in that The using each of the defect recognition models to process the partial discharge map and outputting the recognition result of each of the defect recognition models on the partial discharge map includes: Applying the abnormal dimension index set, obtaining abnormal feature data of the partial discharge map, wherein the abnormal feature data includes features of the dimension of each dimension index of the partial discharge map in the abnormal dimension index set; Each defect recognition model is used to process the abnormal feature data of the partial discharge map to obtain a recognition result of each defect recognition model on the partial discharge map, wherein the recognition result includes an abnormality score of the defect recognition model on the partial discharge map.
7. The method according to claim 6, characterized in that Also includes: When there is an identification result indicating that the partial discharge spectrum is not abnormal, the defect identification model corresponding to the abnormality score with the smallest value is determined as the target model; The partial discharge map is determined to be a map of partial discharge defects corresponding to the target model.
8. A partial discharge interference spectrum rejection device, characterized in that: include: An acquisition unit, used for acquiring a partial discharge spectrum to be identified and a defect spectrum set of a gas-insulated metal-enclosed switchgear, wherein the defect spectrum set contains a sample spectrum of at least one partial discharge defect of the gas-insulated metal-enclosed switchgear; A first extraction unit, configured to extract sample feature data of each of the sample graphs; the sample feature data includes features of multiple dimensions; A second extraction unit is used to extract an abnormal feature set of each partial discharge defect from each sample feature data of each partial discharge defect, wherein the abnormal feature set includes a plurality of abnormal feature samples, and use the abnormal feature set of each partial discharge defect to construct a defect recognition model for each partial discharge defect; A processing unit is used to process the partial discharge spectrum using each of the defect recognition models, output a recognition result of each of the defect recognition models on the partial discharge spectrum, and determine that the partial discharge spectrum is an interference spectrum when each of the recognition results characterizes that the partial discharge spectrum is abnormal.
9. A storage medium, characterized in that: The storage medium includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the partial discharge interference spectrum rejection method according to any one of claims 1 to 7.
10. An electronic device, characterized in that: It comprises a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to be executed by one or more processors to implement the partial discharge interference spectrum rejection method as described in any one of claims 1 to 7.