Partial discharge atlas database construction method and system

Abstract

The invention provides a partial discharge map database construction method and system, and the method comprises the steps: collecting multi-format partial discharge maps which are actually measured by a plurality of manufacturers, and carrying out the repeated cleaning of samples and the removal of ground noise of the collected partial discharge maps, and obtaining a first data set; preprocessing the first data set to obtain a second data set; constructing a partial discharge feature set of a partial discharge type according to the second data set; constructing a partial discharge graph database based on the set feature parameters and the partial discharge feature set; in other words, a high-quality PRPD data set, namely a partial discharge atlas database, suitable for actual engineering requirements is constructed; the problems that in the classification and recognition process of a machine learning algorithm, PRPD atlas collection is particularly difficult, and waste is caused in the data cleaning process due to the scarcity of partial discharge type verification data and the diversity of data formats are solved. And the diversity and quantity of samples in the partial discharge atlas database are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and more specifically, particularly relates to a method and system for constructing a partial discharge pattern database. Background Art

[0002] With the increasing number of years of operation of GIS (Gas Insulated Switchgear), ensuring its safe and stable operation has become the primary task of the power industry. As a key factor in the degradation of the insulation performance of GIS electrical equipment, effective monitoring and identification of partial discharge can not only provide valuable reference information for equipment maintenance, but also quickly locate insulation faults and prevent accidents.

[0003] Currently, the identification of partial discharge in GIS equipment mainly relies on two methods based on PRPD (Phase Resolved Partial Discharge) patterns: one is pattern recognition driven by expert knowledge, and the other is a deep learning method driven by data.

[0004] However, constructing a high-quality PRPD dataset suitable for actual engineering requirements faces many challenges; in the classification and recognition process of machine learning algorithms, the collection of PRPD patterns is particularly difficult, not only because of the scarcity of partial discharge type verification data, but also because of the waste in the data cleaning process caused by the diversity of data formats. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method and system for constructing a partial discharge pattern database, which is used to enhance the diversity and quantity of samples in the partial discharge pattern database.

[0006] The first aspect of the present application discloses a method for constructing a partial discharge pattern database, including:

[0007] Collect partial discharge patterns measured by multiple manufacturers in multiple formats, and perform sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first dataset;

[0008] Preprocess the first dataset to obtain a second dataset;

[0009] According to the second dataset, construct a partial discharge feature set for partial discharge types;

[0010] Based on the set feature parameters and the partial discharge feature set, construct a partial discharge pattern database.

[0011] Optionally, the step of collecting partial discharge patterns measured by multiple manufacturers in multiple formats, and performing sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first dataset includes:

[0012] Collect the partial discharge spectrograms measured from GIS devices verified through disassembly by the partial discharge monitoring systems of each manufacturer; the formats of the partial discharge spectrograms corresponding to the partial discharge monitoring systems of different manufacturers are not completely the same;

[0013] Remove duplicate samples from the partial discharge spectrograms to obtain a dataset to be processed;

[0014] Based on the Hough transform line detection method, perform background noise removal on each partial discharge spectrogram in the dataset to be processed to obtain a first dataset.

[0015] Optionally, the step of performing background noise removal on each partial discharge spectrogram in the dataset to be processed based on the Hough transform line detection method to obtain a first dataset includes:

[0016] Perform gray-scale threshold segmentation on each partial discharge spectrogram in the dataset to be processed, and remove background pixels to obtain a first pixel set containing partial discharge feature pixels and background grid line pixels;

[0017] Convert all pixel points in the first pixel set to the Hough space;

[0018] According to the representation of each pixel point in the first pixel set in the Hough space, locate the position and direction of the line background noise in the first pixel set, and remove the line background noise pixels from the first pixel set to obtain a first dataset.

[0019] Optionally, the step of preprocessing the first dataset to obtain a second dataset includes:

[0020] Unify the background color of each partial discharge spectrogram in the first dataset and detect the background pixels;

[0021] According to the background pixels of each unified partial discharge spectrogram, correct the redundant borders of the partial discharge spectrogram and achieve size normalization;

[0022] Remove the grid lines and cosine lines in the background pixel area of each partial discharge spectrogram to obtain a second dataset; the partial discharge spectrograms in the second dataset include partial discharge feature pixels and background pixels of a unified color.

[0023] Optionally, the step of removing the grid lines and cosine lines in the background pixel area of each partial discharge spectrogram to obtain a second dataset includes:

[0024] Perform a spatial-frequency transformation from the spatial domain to the frequency domain on the partial discharge spectrogram after size normalization;

[0025] Move the zero-frequency point of the spatial-frequency transformation result to the center position of the spectrogram and calculate the amplitude of the spectrogram;

[0026] Transform the amplitude of the spectrogram into the logarithmic interval and perform grid filtering;

[0027] Perform the inverse spatio - frequency transform on the spectrogram after grid filtering to obtain a second data subset;

[0028] Generate a cosine line with a period of l and an amplitude of k / 2, use the gray - scale values of the surrounding pixels of the cosine line to replace the cosine line, and remove the cosine line in the background of the second data subset to obtain a second data set; where the length of the partial discharge pattern after normalization is l and the width is k.

[0029] Optionally, constructing a partial - discharge feature set of partial - discharge types according to the second data set includes:

[0030] Based on the even - discharge feature detection method of pixel density, determine the density of partial - discharge feature pixel points in the map space within the second data set, and remove the even - discharge features with a density greater than the threshold to obtain a set of partial - discharge feature pixel positions;

[0031] Based on the density - based clustering algorithm, cluster the set of partial - discharge feature pixel positions, and generate partial - discharge feature sets for tip discharge, insulation discharge, floating discharge, and metal - particle discharge according to the partial - discharge types of the partial - discharge map.

[0032] Optionally, the density - based clustering algorithm clusters the set of partial - discharge feature pixel positions and generates partial - discharge feature sets for tip discharge, insulation discharge, floating discharge, and metal - particle discharge according to the partial - discharge types of the partial - discharge map, including:

[0033] Determine the clustering radius of the clustering algorithm and the minimum number of pixels in its neighborhood space area;

[0034] For all pixel points in the set of partial - discharge feature pixel positions, count the number of pixels in the space area with the clustering radius as the neighborhood for each pixel point;

[0035] If the number of pixels of any pixel point in the space area with the clustering radius as the neighborhood is greater than the minimum number of pixels, then mark the corresponding pixel point as a core point;

[0036] Define the core points with overlapping neighborhoods and all pixel points within the neighborhoods as the same partial - discharge type; where the partial - discharge types include: tip discharge, insulation discharge, floating discharge, or metal - particle partial discharge;

[0037] Obtain the sets of partial - discharge feature pixel positions of different partial - discharge types, and define the spatial form formed by each pixel point of the same partial - discharge type after clustering as the partial - discharge feature set of the corresponding partial - discharge type.

[0038] Optionally, constructing a partial discharge pattern database based on the set characteristic parameters and the partial discharge characteristic set includes:

[0039] Generate a zero-pixel matrix of size l×k;

[0040] In the zero-pixel matrix, process the partial discharge characteristic set according to the number N of partial discharge characteristics, the scaling ratio σ for setting partial discharge characteristics, the number M of phase shift times, and the position difference between two characteristic centers in the set characteristic parameters to obtain a partial discharge characteristic pattern;

[0041] Use the set of partial discharge characteristic patterns as the partial discharge pattern database.

[0042] The second aspect of this application discloses a partial discharge pattern database construction system, including:

[0043] A cleaning module for collecting partial discharge patterns of multiple formats measured by multiple manufacturers, and performing sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first data set;

[0044] A preprocessing module for preprocessing the first data set to obtain a second data set;

[0045] A characteristic set module for constructing a partial discharge characteristic set of a partial discharge type according to the second data set;

[0046] A construction module for constructing a partial discharge pattern database based on set characteristic parameters and the partial discharge characteristic set.

[0047] Optionally, when the characteristic set module is used to construct a partial discharge characteristic set of a partial discharge type according to the second data set, it specifically is used for:

[0048] Based on the even discharge characteristic detection method of pixel density, determine the density of partial discharge characteristic pixel points in the map space in the second data set, and eliminate the even discharge characteristics with a density greater than the threshold to obtain a set of partial discharge characteristic pixel positions;

[0049] Based on the density-based clustering algorithm, cluster the set of partial discharge characteristic pixel positions, and generate partial discharge characteristic sets of tip discharge, insulation discharge, floating discharge, and metal particle discharge according to the partial discharge type of the partial discharge pattern.

[0050] As can be seen from the above technical solution, a method for constructing a partial discharge pattern database provided by the present invention includes: collecting partial discharge patterns of multiple manufacturers in multiple formats, and performing sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first data set; preprocessing the first data set to obtain a second data set; constructing a partial discharge feature set of partial discharge types according to the second data set; constructing a partial discharge pattern database based on the set feature parameters and the partial discharge feature set; that is, realizing the construction of a high-quality PRPD data set applicable to actual engineering requirements - the partial discharge pattern database; solving the problem that it is particularly difficult to collect PRPD patterns in the classification and recognition process of machine learning algorithms, not only because of the scarcity of partial discharge type verification data, but also because of the waste in the data cleaning process caused by the diversity of data formats; enhancing the diversity and quantity of samples in the partial discharge pattern database. Brief Description of the Drawings

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 It is a flowchart of a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0053] Figures 2 - 5 It is a schematic diagram of the collected PRPD patterns involved in a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0054] Figures 6 - 9 It is a schematic diagram of the preprocessed PRPD patterns involved in a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0055] Figures 10 - 13 It is a schematic diagram of the PRPD patterns in the partial discharge feature set involved in a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0056] Figure 14 It is the generation result of the metal particle discharge pattern involved in a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0057] Figure 15 It is a schematic diagram of the partial discharge pattern database involved in a method for constructing a partial discharge pattern database provided by an embodiment of the present invention;

[0058] Figure 16 It is a schematic diagram of a partial discharge pattern database construction system provided by an embodiment of the present invention. Detailed implementation manners

[0059] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] In this application, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "including one..." does not exclude the presence of additional identical elements in the process, method, article or device including the element. In addition, the terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of this application described herein can be implemented in an order other than those illustrated or described herein.

[0061] The embodiment of this application provides a method for constructing a partial discharge pattern database, which is used to solve the many challenges faced in constructing a high-quality PRPD data set applicable to the actual engineering requirements in the prior art; in the classification and recognition process of machine learning algorithms, the collection of PRPD patterns is particularly difficult, not only because of the scarcity of partial discharge type verification data, but also because of the waste in the data cleaning process caused by the diversity of data formats.

[0062] See Figure 1 , the method for constructing the partial discharge pattern database includes:

[0063] S101. Collect partial discharge patterns measured by multiple manufacturers in multiple formats, and perform sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first data set.

[0064] This first data set can be named Data_Raw; that is, duplicate samples are removed from the collected partial discharge pattern data to obtain a data set D; the background noise influence on the partial discharge patterns in the data set D is repaired to obtain the data set Data_Raw after data cleaning.

[0065] Collect actual measured partial discharge patterns from multiple power equipment manufacturers, and these patterns may be in different formats (such as CSV, Excel, MAT files, etc.). The partial discharge pattern can be a two-dimensional PRPD pattern.

[0066] Use an automated tool or manual method to check and remove duplicate samples to ensure that each sample is unique in the dataset.

[0067] Apply signal processing techniques such as filtering and denoising to remove the background noise components in the pattern to improve the signal-to-noise ratio of the data. After this step, a first dataset that has been preliminarily cleaned and processed is obtained.

[0068] S102. Preprocess the first dataset to obtain a second dataset.

[0069] This second dataset can be named Data_Norm.

[0070] The preprocessing can be to unify the format of each partial discharge pattern in the first dataset to ensure that all partial discharge patterns are converted to a unified format (such as a MAT file but not limited to this) for subsequent processing. That is, partial discharge patterns in different formats can be converted to partial discharge patterns in the same format, rather than directly screening out a certain format. Thus, partial discharge patterns in multiple formats can be stored in the partial discharge pattern database provided in this application, increasing the diversity of the partial discharge pattern database; but of course, the format of the partial discharge pattern can also not be adjusted, which will not be elaborated here one by one.

[0071] The preprocessing can also be to normalize the data so that data from different sources and different scales can be compared and analyzed on the same scale.

[0072] The preprocessing can also include feature extraction; such as extracting key features such as the amplitude and phase of the discharge pulse to further simplify the dataset and retain key information. After this step, a second dataset that is in a unified format, normalized, and contains key features is obtained.

[0073] Of course, the preprocessing method can also be other content, which will not be elaborated here one by one and can be determined according to the actual situation, and all are within the protection scope of this application.

[0074] It should be noted that some image preprocessing in the dataset production process mainly includes image scaling, cropping, rotation, flipping, image grayscale conversion, and normalization processing. These traditional preprocessing methods are not applicable to changes in formats such as image background color and feature pixel rendering methods. The PRPD pattern preprocessing method proposed in this application mainly solves the problem of standardizing patterns in different formats.

[0075] S103. Construct a partial discharge feature set of partial discharge types according to the second dataset.

[0076] Label the samples in the second dataset, that is, assign corresponding labels to each sample according to known partial discharge types, such as tip discharge, insulation discharge, floating discharge, and metal particle discharge, etc.

[0077] Use clustering algorithms or classification algorithms, such as clustering algorithms like K-means and DBSCAN, and classification algorithms like decision trees and support vector machines, to further analyze and divide the features to form a more distinct partial discharge type feature set.

[0078] After completing the construction of the partial discharge feature set, the accuracy and reliability of the feature set can also be verified to ensure that the features between different partial discharge types have obvious distinguishability.

[0079] S104. Based on the set feature parameters and the partial discharge feature set, construct a partial discharge pattern database.

[0080] Set feature parameters according to actual requirements. These parameters may include the number N of partial discharge features, the scaling ratio σ of the features, the number M of phase shift times, the position differences dx and dy of the centers of multiple partial discharge feature templates, etc. Of course, other parameters may also be included, which will not be elaborated here one by one and can be determined according to the actual situation, all within the protection scope of this application.

[0081] Combine the partial discharge feature set with the set feature parameters to construct a pattern database containing complete partial discharge information.

[0082] Verify and optimize the database to ensure the accuracy and integrity of the data. At the same time, provide a friendly user interface and query function so that users can easily access and utilize these data.

[0083] Through the above steps, a high-quality PRPD dataset applicable to actual engineering requirements - the partial discharge pattern database is successfully constructed. This database not only solves the problem of scarce data for partial discharge type verification, but also overcomes the waste problem in the data cleaning process caused by the diversity of data formats, providing strong support for the classification and recognition of machine learning algorithms. That is, this application involves fields such as image processing, PRPD pattern data analysis, and dataset construction, mainly involving the generation of two-dimensional PRPD patterns of partial discharges in GIS. Based on the characteristics of typical GIS partial discharge PRPD patterns measured by various partial discharge monitoring manufacturers, it is committed to proposing a PRPD pattern data enhancement method. Starting from the collection of actual monitoring PRPD data from multiple manufacturers and in multiple formats, it successively constructs data collection and cleaning methods, image preprocessing methods, partial discharge feature extraction methods, and PRPD pattern feature generation methods designed based on expert knowledge guidance, so as to realize the construction of a high-quality PRPD pattern partial discharge defect dataset; thereby generating a high-quality and diverse PRPD two-dimensional pattern dataset.

[0084] In this embodiment, measured partial discharge maps in multiple formats from multiple manufacturers are collected, and the collected partial discharge maps are repeatedly cleaned and background noise is removed to obtain a first data set; the first data set is preprocessed to obtain a second data set; based on the second data set, a partial discharge feature set of the partial discharge type is constructed; based on the set feature parameters and the partial discharge feature set, a partial discharge map database is constructed; that is, a high-quality PRPD data set suitable for actual engineering needs, namely a partial discharge map database, is constructed; the problem that the collection of PRPD maps is particularly difficult in the classification and recognition process of the machine learning algorithm is solved, not only because of the scarcity of partial discharge type verification data, but also because of the waste in the data cleaning process caused by the diversity of data formats; and the diversity and quantity of samples in the partial discharge map database are enhanced.

[0085] It is worth noting that one of the challenges currently faced by GIS equipment disassembly verification is the scarcity of PRPD spectrum data for partial discharge and the lack of a public PD identification training dataset suitable for actual industrial scenarios. Existing PRPD datasets are mainly constructed based on laboratory simulation conditions, in which the GIS equipment structure and UHF PD monitoring products are in a fixed state. However, in actual applications, the PD phenomena generated by GIS equipment are diverse, and the two-dimensional spectrum formats generated by different spectrum acquisition devices are different. Therefore, in order to accurately identify the PRPD spectrum types of various PD monitoring products, the training dataset must cover the PRPD spectrum obtained by different monitoring products under different GIS equipment structures. In addition, constructing a diverse and high-quality PRPD spectrum dataset is crucial for training a PD identification network with wide applicability. The construction of this dataset includes three core links: (1) data collection; (2) data preprocessing; and (3) data deep mining. In the data collection stage, previous work has difficulty in fully covering the diversity of GIS equipment structures and different types of PD monitoring products. In the data preprocessing stage, existing methods mainly remove duplicate data samples and fill or eliminate samples with incomplete features. However, given the scarcity of PRPD measured samples and the characteristics of small sample data, the use of traditional methods for data preprocessing may over-consume precious sample resources. As for data deep mining, when the number of partial discharge samples of each category is extremely limited, the traditional data enhancement technology based on GAN (generative adversarial network) is difficult to converge effectively. Therefore, it is urgent to develop a data enhancement strategy specifically for the PRPD two-dimensional atlas dataset of GIS equipment, aiming to lay a solid foundation for the accurate identification of partial discharge types in future engineering applications.

[0086] This application proposes an innovative solution, namely, a UHF partial discharge pattern database enhancement technology based on optimized parameter adjustment, and proposes a method for constructing a partial discharge pattern database, which can generate a high-quality and diverse dataset, thus providing a solid foundation for the partial discharge type recognition based on PRPD patterns in practical industrial applications.

[0087] Optionally, in the above step S101, collect partial discharge patterns of multiple manufacturers in multiple formats, and perform sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first dataset, including:

[0088] First, through the partial discharge monitoring systems of each manufacturer, collect the measured partial discharge patterns of GIS devices verified by disassembly; the formats of the partial discharge patterns corresponding to the partial discharge monitoring systems of different manufacturers are not completely the same. Secondly, remove duplicate samples from the partial discharge patterns to obtain a dataset to be processed D. Finally, based on the Hough transform line detection method, perform background noise removal on each partial discharge pattern in the dataset to be processed D to obtain a first dataset Data_Raw.

[0089] Specifically, through the partial discharge monitoring systems provided by each power equipment manufacturer, such as the continuous partial discharge monitoring system of Glasgow DMS Company in the UK, the partial discharge monitoring system of Shanghai Mok, and the self-developed partial discharge intensive monitoring system of Xi'an High Voltage Apparatus Research Institute Co., Ltd., widely collect the measured partial discharge patterns of GIS devices, such as Figure 2 the PRPD pattern of the tip partial discharge verified by disassembly as shown, Figure 3 the PRPD pattern of the insulation partial discharge verified by disassembly as shown; Figure 4 the PRPD pattern of the metal particle partial discharge verified by disassembly of; Figure 5 shows the PRPD pattern of the floating partial discharge verified by disassembly. These GIS devices have been verified by disassembly before data collection, ensuring the authenticity and reliability of the partial discharge data. It should be noted that due to the differences in the design and technical implementation of the partial discharge monitoring systems adopted by different manufacturers, the collected partial discharge patterns are not completely unified in format, which brings certain challenges to subsequent data processing.

[0090] After collecting a large number of partial discharge patterns, it is necessary to preliminarily screen the data and remove the duplicate samples. This step is crucial because the existence of duplicate samples will not only waste storage space but also may interfere with subsequent data analysis and the training of machine learning models. Use advanced algorithms and technical means, such as hash algorithms, similarity calculations, etc., to comprehensively check the collected partial discharge patterns to ensure that each sample is unique in the dataset; of course, other duplicate removal algorithms can also be used, which will not be elaborated here one by one and can be determined according to the actual situation, all within the protection scope of this application.

[0091] Next, the background noise of each partial discharge spectrum in the processing data set is removed. Background noise is a common interference factor in partial discharge spectrum, which may come from the background noise of the device itself, interference from the external environment, etc. In order to accurately extract the characteristics of the partial discharge signal, the line detection method based on Hough transform can also be used to remove the background noise. Hough transform is an algorithm widely used in the fields of image processing and computer vision, which can effectively detect the straight line structure in the image. Here, Hough transform is applied to the background noise processing of partial discharge spectrum, and a clearer and more accurate partial discharge signal is obtained by detecting and removing the linear noise components in the spectrum.

[0092] After the above steps, a first data set with duplicate sample removal and background noise removal is obtained. This data set is not only accurate, but also lays a solid foundation for subsequent preprocessing, feature extraction and database construction.

[0093] In this embodiment, the Hough transform technique is used to detect and extract background noise features from the image. The PRPD feature map of partial discharge usually shows phase correlation, that is, the number and amplitude of discharges will be different at different phases. In contrast, the background noise does not have this phase correlation. It maintains the same amplitude displayed on the feature map over the entire phase range of 0° to 360°. Therefore, in the PRPD map, the background noise appears as an approximate straight line spanning 0-360°. By detecting the straight line using the Hough transform, these background noise features can be effectively identified and removed.

[0094] Optionally, based on the Hough change line detection method, background noise is removed from each partial discharge spectrum in the data set to be processed to obtain a first data set, including:

[0095] Firstly, grayscale threshold segmentation is performed on each partial discharge map in the processed data set, and background pixels are eliminated to obtain a first pixel set including partial discharge feature pixels and background grid line pixels; secondly, all pixel points in the first pixel set are converted to Hough space; finally, according to the representation of each pixel point in the first pixel set in the Hough space, the position and direction of the linear background noise in the first pixel set are located, and the linear background noise pixels are eliminated from the first pixel set to obtain the first data set.

[0096] It should be noted that according to the gray - value differences among the background pixels, partial - discharge feature pixels, and background grid - line pixels in the PRPD pattern, these three types of pixel sets can be respectively defined as Data_back (background pixels), Data_F (partial - discharge feature pixels), and Data_grid (background grid - line pixels). Among them, the gray - value difference between Data_back and Data_F is significant, so the method of gray - threshold segmentation can be directly applied to distinguish them. However, for Data_grid, due to the usually non - obvious gray - value difference, it is difficult to directly obtain and distinguish it only through gray - value estimation.

[0097] Specifically, perform gray - threshold segmentation processing on each partial - discharge pattern in the dataset to be processed. The purpose of this step is to remove those background pixels irrelevant to the partial - discharge features from the original pattern, so as to retain the partial - discharge feature pixels and possibly existing background grid - line pixels. By setting a suitable gray - threshold, the pixels in the pattern can be divided into two categories: pixels above the threshold are regarded as foreground pixels (including partial - discharge features and background grid - lines), while pixels below the threshold are regarded as background pixels and are removed. In this way, a first pixel set containing partial - discharge feature pixels and background grid - line pixels is obtained.

[0098] Next, transform all pixel points in this first pixel set into the Hough space. The Hough space is a special space used to detect straight - line structures in an image. By transforming the points in the image into the Hough space, the straight lines in the image can be mapped to points or lines in the Hough space. This transformation process utilizes the powerful ability of the Hough transform to detect straight - line structures in the image. It should be noted that when all data points in Data_F of the PRPD pattern are transformed into the Hough space, the spatial position of the i - th pixel point in the feature space is represented as (xi, yi), and after transformation into the Hough space, it is a curve, represented as ri = xicosθ + yisinθ.

[0099] Finally, in the Hough space, locate the position and direction of the straight - line background noise according to the representation of each pixel point in the first pixel set. Since the background grid - lines usually appear as a series of regular straight lines, they will form obvious clustering points or lines in the Hough space. By identifying these clustering points or lines, the straight - line background noise in the image can be accurately located. Once the position and direction of the straight - line background noise are determined, these straight - line background - noise pixels can be removed from the first pixel set. After this step of processing, a first dataset with straight - line background noise removed is obtained. This dataset is purer and only contains partial - discharge feature pixels, providing a more accurate data basis for subsequent data analysis and the training of machine - learning models.

[0100] Specifically, a corner point solving method can be adopted to locate the position and direction of the background noise line in the partial discharge characteristics, and the background noise pixels are removed from the Data_F set. More specifically, all curve sets are input into the Hough space, and the intersection set of these curves is recorded, in the specific form of {(r1, θ1), (r2, θ2),..., (ri, θi)}. Considering that the characteristic of the background noise pixels in the PRPD pattern is that they are evenly distributed in the range of 0° to 360°, which means that in the Hough space, the lines corresponding to the background noise will be manifested as an intersection set with a quantity approximately equal to the width k of the pattern. Based on this judgment criterion, the pixels corresponding to the background noise in the PRPD pattern are defined as the pixels represented by the curves having k intersections in the Hough space. Subsequently, these pixels determined to be background noise are removed from the Data_F set.

[0101] Optionally, the above step S102 of preprocessing the first data set to obtain the second data set includes:

[0102] First, for each partial discharge pattern in the first data set, the background color is unified and the background pixels are detected; this step ensures the consistency of the background color of all patterns and lays a solid foundation for subsequent processing.

[0103] Secondly, according to the background pixels of each unified partial discharge pattern, the redundant borders of the partial discharge pattern are corrected and size normalization is achieved. Among them, the length of the normalized partial discharge pattern is l and the width is k.

[0104] That is, on the basis of the unified background color, according to the detected background pixel information of each partial discharge pattern, the redundant borders in the partial discharge pattern are corrected. This step not only removes unnecessary border information but also further realizes the normalization processing of the pattern size, making all patterns reach a unified standard in size and facilitating subsequent analysis and processing.

[0105] Finally, the grid lines and cosine lines in the background pixel area of each partial discharge pattern are removed to obtain the second data set; the partial discharge patterns in the second data set include partial discharge characteristic pixels and background pixels with a unified color. Among them, the spatial position set of the partial discharge characteristic pixels is represented as Data_F.

[0106] That is, further cleaning work is carried out on the corrected partial discharge pattern to remove interference elements such as grid lines and cosine lines in the background pixel area. After this series of preprocessing steps, the second data set is finally obtained (such as Figure 6 the result after preprocessing the tip partial discharge spectrum diagram shown, Figure 7 the result after preprocessing the metal particle partial discharge spectrum diagram shown, Figure 8 the result after preprocessing the insulation partial discharge spectrum diagram shown, and Figure 9The result after preprocessing the floating partial discharge spectrogram shown). In the second dataset, each partial discharge spectrogram only contains local discharge characteristic pixels and background pixels of a unified color, which greatly improves the purity and usability of the data.

[0107] In this embodiment, it aims to eliminate the background grid interference in the two-dimensional image of the PRPD spectrogram obtained by monitoring the equipment of different manufacturers. Based on a priori knowledge of frequency domain transformation: image features that appear as horizontal stripes in the spatial domain will be transformed into vertical lines passing through the origin of the frequency domain in the frequency domain; correspondingly, vertical stripes in the spatial domain will be manifested as horizontal lines passing through the origin in the frequency domain. Based on this principle, a specific frequency domain filter can be designed to effectively remove the grid lines. At the same time, for the sine line interference that appears in the background, the present invention accurately fits its function expression, that is, a cosine function with a period of l and an amplitude of k / 2, to achieve accurate identification. Subsequently, using the information of the pixels around the cosine line, the gray values of the cosine lines in the background and features can be directly removed. This process ensures that the background cosine lines in the image are replaced by background pixels, while the cosine lines in the local discharge features are replaced by corresponding partial discharge characteristic pixels, thereby effectively purifying the PRPD spectrogram without destroying useful information.

[0108] Optionally, removing the grid lines and cosine lines in the background pixel regions of each partial discharge spectrogram to obtain a second dataset, including:

[0109] First, perform a spatial-frequency transformation on the size-normalized partial discharge spectrogram from the spatial domain to the frequency domain; this step transforms the partial discharge spectrogram from the spatial domain to the frequency domain. This transformation can more clearly identify and separate different frequency components in the spectrogram.

[0110] Specifically, transform the normalized data Data(i, j) from the spatial domain to the frequency domain F(u, v), where i, j represent the specific positions of the background pixels in the PRPD spectrogram space with a size of l*k, i is the abscissa, and j is the ordinate. u, v represent the positions of the effective content of the spectrogram in the frequency domain, and the spatial-frequency transformation formula is:

[0111]

[0112] Among them, M and N respectively represent the number of length and width pixels of the normalized PRPD spectrogram, then M = l, N = k, and the formula I(i, j) in the formula represents the gray value of the background of the PRPD spectrogram at the (i, j) spatial coordinate position.

[0113] The formula for determining the gray value is:

[0114]

[0115] Secondly, move the zero-frequency point of the spatio-frequency transformation result F(u, v) to the center position of the spectrogram, and calculate the amplitude of the spectrogram; this operation is considered based on the symmetry and analysis convenience of the spectrogram; move the zero-frequency point (i.e., the DC component) of the spatio-frequency transformation result to the center position of the spectrogram; calculate the amplitude of the spectrogram to quantify the intensity of each frequency component.

[0116] Specifically, since most of the information of the image is distributed in the low-frequency part of the spectrum, when no zero-frequency shift is performed, the energy of the spectrogram is mainly concentrated in the four corners, which is not conducive to observation; the centering transformation formula is:

[0117]

[0118] The formula used to calculate the amplitude |F’(u,v)| of the spectral image is:

[0119]

[0120] where R and Q are the real and imaginary parts of the frequency-domain function F’(u, v) after centering.

[0121] Next, transform the amplitude of the spectrogram to the logarithmic interval and perform grid filtering. To filter out the grid lines, convert the amplitude of the spectrogram to the logarithmic interval; the logarithmic transformation can enhance the weak signals in the spectrogram while suppressing the strong signals, making the regular interference components such as grid lines more prominent on the spectrogram, thus facilitating subsequent filtering operations. In the logarithmic interval, perform the grid filtering step to effectively remove the grid line components in the spectrogram.

[0122] Specifically, to more conveniently observe the low-frequency part in the frequency-domain image, convert the amplitude of the frequency-domain image to the logarithmic scale to obtain Fl’(u,v), and the formula used is:

[0123]

[0124] Define a frequency-domain grid removal filter D(u,v), which is the product of four sub-filters D1(u,v), D2(u,v), D3(u,v) and D4(u,v), where D1, D2, D3 and D4 represent the filters in the upper, lower, left and right directions around the frequency-domain origin respectively.

[0125] When performing grid filtering in the frequency domain, multiply the defined filter D(u,v) by the transformed frequency-domain image F’(u,v). The formula used is:

[0126]

[0127] Through this step, the grid interference in the frequency-domain image can be effectively removed.

[0128] In addition, an inverse transform of the spatial-frequency transform is performed on the spectrogram after grid filtering to obtain a second data subset. That is, after completing the grid filtering, an inverse transform of the spatial-frequency transform is performed on the processed spectrogram to convert it from the frequency domain back to the spatial domain.

[0129] Specifically, an inverse transform operation is performed on the filtered spectrogram image to obtain a PRPD map with grid lines removed, and its data is denoted as Data'. At this time, Data' only contains two major components: background data Data_back and partial discharge feature data Data_F, and no longer contains the grid line pixel set Data_grid.

[0130] Finally, a cosine line with a period of l and an amplitude of k / 2 is generated, and the gray values of the surrounding pixels of the cosine line are used to replace the cosine line to remove the cosine line in the background of the second data subset, obtaining a second data set; where the length of the partial discharge map after normalization is l and the width is k.

[0131] To remove the cosine line in the background, a cosine line template with a period of l (the same as the length of the normalized map) and an amplitude of k / 2 (related to half of the width of the normalized map) is generated. Then, the gray values of the surrounding pixels of the cosine line are used to replace the gray value of the cosine line itself, thereby achieving smooth removal of the cosine line. This step ensures that while removing the cosine line, the original features of the background pixels are maintained as much as possible. After this processing, a second data set is finally obtained, in which the grid lines and cosine lines in the background pixel area of each partial discharge map have been completely removed, and only the partial discharge feature pixels and the background pixels of a unified color are retained.

[0132] Optionally, in the above step S103, according to the second data set, a partial discharge feature set of the partial discharge type is constructed, including:

[0133] First, based on the even discharge feature detection method of pixel density, determine the density of the partial discharge feature pixel points in the second data set within the map space, and remove the even discharge features with a density greater than the threshold to obtain a set of partial discharge feature pixel positions. This set of partial discharge feature pixel positions can be named I_xz.

[0134] Specifically, an even discharge feature detection method based on pixel density is adopted. This method deeply analyzes the distribution of partial discharge (PD) feature pixels in the set Data_F of spatial positions of PD feature pixels in the atlas space. By quantifying the density of these feature pixels, even discharge features that are too sparse or dense can be identified. To ensure the quality and representativeness of the feature set, even discharge features whose density exceeds a preset threshold are removed. The output of this step is a refined set I_xz of PD feature pixel positions, which only contains feature pixels that are reasonable and representative in spatial distribution.

[0135] The even discharge feature detection method based on pixel density can be as follows: The maximum radius region is defined as ri = l / 2, the minimum region background is defined as l / 12, and i can be defined as 7 in this embodiment. Of course, other values can also be used. Then the density function f(pixel) is:

[0136]

[0137] where n_j represents the number of PD feature pixels within the circumference of the jth radius region rj.

[0138] This density function is used to calculate the sparsity of PD feature pixels in Data_F. If it is too sparse, it is considered an occasional partial discharge, and the pixel is directly removed to obtain I_xz.

[0139] More specifically, in a single preprocessed PRPD atlas, the set of spatial positions of PD feature pixels is marked as Data_F. To detect the even discharge characteristics of these feature pixels, a multi-scale method based on pixel density is adopted. This method is achieved by analyzing the distribution density of PD feature pixels in Data_F in the atlas space.

[0140] The specific steps are as follows: Taking each PD feature pixel in Data_F as the center, a series of radii {r1, r2, r3,..., ri} arranged in equal differences from small to large are set, and i circumferences are generated around each feature pixel. Then, the number of pixel points n_i within each circumference interval is calculated, and based on this, the density function f(pixel) is constructed, where pixel represents a specific pixel point. This density function f(pixel) is used to describe the density characteristics of pixels.

[0141] Using the density function f(pixel), the sparsity of PD feature pixels in Data_F can be evaluated. If the pixels around a certain feature pixel are too sparse, it is regarded as an occasional PD feature and directly removed from Data_F to obtain the pixel set I_xz after removing occasional features.

[0142] Secondly, based on the density-based clustering algorithm, cluster the set I_xz of the partial discharge feature pixel positions, and generate partial discharge feature sets for tip discharge, insulation discharge, floating discharge, and metal particle discharge according to the partial discharge types in the partial discharge pattern.

[0143] Specifically, use the density-based clustering algorithm to deeply process this set of positions. The density clustering algorithm is known for its sensitivity to irregularly shaped clusters and its flexibility in not requiring the number of clusters to be specified in advance. During the operation of the algorithm, similar partial discharge feature pixels (i.e., those that are close in position and may belong to the same partial discharge type) will be automatically clustered into clusters.

[0144] According to the known partial discharge types in the partial discharge pattern, conduct a detailed classification of these clustering results. This step matches the clustering clusters with specific partial discharge types such as tip discharge, insulation discharge, floating discharge, and metal particle discharge. Through such a matching process, a dedicated partial discharge feature set can be generated for each partial discharge type. These feature sets not only contain the precise position information of the partial discharge features but also imply the unique electrical and physical characteristics associated with specific partial discharge types, providing strong support for subsequent partial discharge identification and diagnosis.

[0145] That is, use the density-based clustering algorithm DBSCAN to process I_xz to extract the spatial morphological features of the discharge of the GIS equipment. After clustering, the set of partial discharge pixel features belonging to the same class will form a specific spatial morphology, and these morphologies are regarded as the extracted partial discharge feature templates.

[0146] According to the PRPD pattern characteristics of tip discharge, insulation discharge, floating discharge, and metal particle discharge, four typical partial discharge feature sets F are generated in sequence using the above method, that is, F = {F_jd (tip discharge feature set as shown in Figure 10 ), F_jy (insulation discharge feature set as shown in Figure 12 ), F_xf (floating discharge feature set as shown in Figure 13 ), F_js (metal particle discharge feature set as shown in Figure 11 )}.

[0147] Optionally, based on the density-based clustering algorithm, cluster the set I_xz of the partial discharge feature pixel positions, and generate partial discharge feature sets for tip discharge, insulation discharge, floating discharge, and metal particle discharge, including:

[0148] First, determine the clustering radius of the clustering algorithm and the minimum number of pixels in its neighborhood space area. The clustering radius R can be 5 or other values, and the minimum number of pixels N in the neighborhood space of the clustering radius R minIt can be 4, but it can also be other values. The two parameters jointly determine the density sensitivity and accuracy of the clustering process. That is, input the local discharge feature pixel position information I_xz(i, j) to be clustered, input the clustering radius R = 5, and the minimum number of pixels N within the area of the clustering radius min = 4;

[0149] Secondly, for all pixel points I_xz(i, j) within the set I_xz of local discharge feature pixel positions, count the number of pixels in the spatial area with the clustering radius R as the neighborhood for each pixel point. Specifically, take each pixel point as the center and its clustering radius as the radius to delimit a neighborhood spatial area, and then count the total number of pixels within this spatial area.

[0150] Then, if the number of pixels of any pixel point in the spatial area with the clustering radius as the neighborhood is greater than the minimum number of pixels N min , then mark the corresponding pixel point as a core point. That is, make a judgment based on the statistical results; if the number of pixels in the neighborhood spatial area of a certain pixel point exceeds the preset minimum number of pixels, then regard this pixel point as a core point. These core points play a crucial role in the clustering process because they are not only part of the clustering themselves but also can attract and contain other pixel points within their neighborhoods. The set of all core points is defined as P.

[0151] Again, define the core points with overlapping neighborhoods and all pixel points within their neighborhoods as the same type of local discharge; that is, after determining the core points, further mark those core points with overlapping neighborhoods and all pixel points within their neighborhoods as the same type; among them, the core points and all core points within their domains can also be marked as the same type. Among them, the types of local discharge include: tip discharge, insulation discharge, floating discharge or metal particle partial discharge.

[0152] Finally, obtain the sets of local discharge feature pixel positions of different local discharge types, and define the spatial form formed by each pixel point of the same local discharge type after clustering as the local discharge feature set of the corresponding local discharge type.

[0153] That is, input the PRPD maps of the defect types of tip discharge, insulation discharge, floating discharge, and metal particle partial discharge in sequence, and for each type of map, obtain the pixel set I_xz after removing the accidental discharge characteristics through the above processing flow. Then, use the DBSCAN clustering method mentioned above to perform clustering analysis on these I_xz. After clustering, the spatial form composed of the same type of partial discharge pixel points is defined as a feature template in the corresponding partial discharge defect feature library. In this way, a unique feature template is constructed for each type of partial discharge defect.

[0154] Specifically, according to different partial discharge types, the corresponding sets of partial discharge characteristic pixel positions are obtained. And the spatial form formed by each pixel point belonging to the same partial discharge type after clustering is defined as the partial discharge characteristic set of this partial discharge type. In this way, through the clustering analysis of the set of partial discharge characteristic pixel positions, a partial discharge characteristic set with clear partial discharge type characteristics can be generated, providing strong support for subsequent analysis and recognition.

[0155] In this embodiment, a pixel density function is constructed to quantify the isolation degree of partial discharge characteristic pixels, and accordingly, accidental discharge characteristics are eliminated. Subsequently, the DBSCAN algorithm is used to perform unsupervised clustering on the spectrogram after eliminating accidental characteristics, so as to extract a representative set of partial discharge spectrogram characteristics.

[0156] Optionally, based on the set characteristic parameters and the partial discharge characteristic set, a partial discharge spectrogram database is constructed, including:

[0157] First, a all-zero pixel matrix with size l×k is generated; this all-zero pixel matrix serves as the basis for subsequent processing and storing partial discharge characteristic spectrograms.

[0158] Secondly, in the all-zero pixel matrix, according to the number N of partial discharge characteristics, the scaling ratio σ of the set partial discharge characteristics, the number M of phase shift times, and the position differences dx and dy between the two characteristic centers in the set characteristic parameters, the partial discharge characteristic set is processed to obtain a partial discharge characteristic spectrogram.

[0159] That is to say, multiple key elements in the set characteristic parameters are used to perform fine processing on the partial discharge characteristic set. Specifically, N partial discharge characteristics are extracted from the partial discharge characteristic set, these N partial discharge characteristics are scaled according to the scaling ratio σ, M phase shift operations are performed, and the position differences between two different characteristic centers are used for processing. By comprehensively considering these elements, a series of partial discharge characteristic spectrograms with different characteristics and forms can be generated.

[0160] That is, during the processing, operations such as scaling, rotation, and translation of the partial discharge characteristics may be required to simulate various partial discharge phenomena that may occur in practice. For example, the scaling ratio σ can adjust the size of the partial discharge characteristics, the number M of phase shift times can simulate the phase change of the partial discharge characteristics, and the position differences between the two characteristic centers can reflect the relative position relationship between different partial discharge sources.

[0161] Finally, the set of partial discharge characteristic spectrograms is used as the partial discharge spectrogram database.

[0162] Specifically, all the generated partial discharge characteristic spectrograms are combined to form a complete partial discharge spectrogram database. This database contains various partial discharge spectrograms with different characteristics and forms, providing rich data support for subsequent applications such as pattern recognition and fault diagnosis.

[0163] By constructing such a partial discharge pattern database, the partial discharge phenomenon can be understood more comprehensively and deeply, improving the safety and reliability of power equipment.

[0164] It should be noted that each PRPD pattern in the partial discharge pattern database is processed using the above method, and will not be elaborated here one by one.

[0165] In this embodiment, the generation process of the pattern depends on the parameter configuration preset by experts. To enhance the diversity of the dataset, the shape of the template features in the feature library can be scaled. At the same time, to address the possible phase asynchronization problem during the recognition process, it is adjusted by setting the phase offset parameter of the pattern.

[0166] In the above embodiments, the specific parameter settings of the algorithm are as follows: For the DBSCAN algorithm, R = 12 and Nmin = 20 are set. Taking the generation of the PRPD pattern of metal particle partial discharge as an example, a pixel matrix with a size of 200×360 is created and the partial discharge feature set F_js is input. The number of partial discharge features is specified as N = 1. In addition, the scaling ratio of the features is set as σ = 1, and the number of phase shift times is M = 10. Since only one partial discharge feature is selected, the position differences of the feature centers are set as dx = 0 and dy = 0. Based on these artificially given parameters, a high-quality two-dimensional PRPD feature pattern can be generated (as Figure 14 shown).

[0167] Figures 2 - 5 Shows the PRPD patterns of tip, suspension, insulation, and metal particle partial discharges monitored by different UHF devices and verified through disassembly. Figures 6 - 9 Shows the results after preprocessing the PRPD patterns of different manufacturers. Figures 10 - 13 Shows the construction of a partial discharge feature set F for a typical partial discharge type. Figure 14 Shows the generation results of the metal particle discharge pattern. Figure 15 Shows the effect after data augmentation of the PRPD pattern dataset (partial discharge pattern database).

[0168] In this embodiment, the main challenges encountered by the current PRPD pattern in partial discharge type recognition stem from the lack of diversity and scarcity of the dataset, which severely limits the performance of the recognition network based on deep learning, especially when facing the small sample problem. To solve this problem, this patent proposes an innovative solution that combines traditional morphological parameter settings to construct the PRPD pattern dataset and significantly enhances the sample size and diversity of the dataset through data generation strategies.

[0169] Specifically, actual PRPD samples were first accumulated through live inspection activities, and the formats of these samples were normalized to ensure data consistency. Subsequently, a set of PRPD spectrum feature extraction methods was designed, which involved building a feature template library covering various types of partial discharge models, providing a key basic feature morphological basis for the subsequent generation of PRPD spectrum data. On this basis, a parameterized PRPD spectrum enhancement technology guided by experts was further developed, which not only improved the diversity of the data, but also effectively increased the overall size of the data set. By implementing the method proposed in this patent, a rich data set for PRPD spectrum partial discharge defects was successfully constructed, which fully meets the engineering application needs of PRPD defect identification under actual operating conditions.

[0170] Another embodiment of the present application discloses a system for constructing a partial discharge spectrum database.

[0171] See also Figure 16 , the local discharge spectrum database construction system includes:

[0172] The cleaning module 101 is used to collect partial discharge spectra in multiple formats measured by multiple manufacturers, and perform repeated sample cleaning and background noise removal on the collected partial discharge spectra to obtain a first data set.

[0173] The preprocessing module 102 is used to preprocess the first data set to obtain a second data set.

[0174] The feature set module 103 is used to construct a partial discharge feature set of the partial discharge type according to the second data set.

[0175] The construction module 104 is used to construct a partial discharge spectrum database based on the set characteristic parameters and the partial discharge feature set.

[0176] Optionally, when the feature set module 103 is used to construct a partial discharge feature set of a partial discharge type according to the second data set, it is specifically used to:

[0177] The occasional discharge feature detection method based on pixel density determines the density of the partial discharge feature pixel points in the second data set in the spectrum space, eliminates the occasional discharge features with a density greater than a threshold, and obtains a partial discharge feature pixel position set.

[0178] The density-based clustering algorithm is used to cluster the PD feature pixel position sets, and according to the PD types of the PD map, the PD feature sets of tip discharge, insulating discharge, suspended discharge and metal particle discharge are generated.

[0179] The specific working process and principle of each of the above modules are detailed in the partial discharge spectrum database construction method provided in the above embodiment, which will not be described here one by one. It depends on the actual situation and is within the protection scope of this application.

[0180] In this embodiment, the cleaning module 101 is configured to collect partial discharge patterns of multiple manufacturers in multiple formats, and perform sample duplicate cleaning and background noise removal on the collected partial discharge patterns to obtain a first data set; the preprocessing module 102 is configured to preprocess the first data set to obtain a second data set; the feature set module 103 is configured to construct a partial discharge feature set of partial discharge types according to the second data set; the construction module 104 is configured to construct a partial discharge pattern database based on the set feature parameters and the partial discharge feature set; that is, to realize the construction of a high-quality PRPD data set applicable to actual engineering requirements - the partial discharge pattern database; to solve the problem that in the classification and recognition process of machine learning algorithms, the collection of PRPD patterns is particularly difficult, not only because of the scarcity of partial discharge type verification data, but also because of the waste in the data cleaning process caused by the diversity of data formats; to enhance the diversity and quantity of samples in the partial discharge pattern database.

[0181] The features described in the respective embodiments of this specification can be replaced or combined with each other. For the same or similar parts between the respective embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for a system or a system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiment. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative efforts.

[0182] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0183] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those 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. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a partial discharge pattern database, characterized in that, include: Collecting multi-format partial discharge spectra measured by multiple manufacturers, and repeatedly cleaning samples and removing background noise from the collected partial discharge spectra to obtain a first data set; Preprocessing the first data set to obtain a second data set; constructing a partial discharge feature set of partial discharge type according to the second data set; Based on the set characteristic parameters and the partial discharge feature set, a partial discharge spectrum database is constructed.

2. The method for constructing a partial discharge pattern database according to claim 1, characterized in that The method collects the partial discharge spectra measured by multiple manufacturers in multiple formats, and repeatedly cleans the samples and removes the background noise of the collected partial discharge spectra to obtain a first data set, including: Through the partial discharge monitoring systems of various manufacturers, the partial discharge spectra measured by the GIS equipment that has been dismantled and verified are collected; the formats of the partial discharge spectra corresponding to the partial discharge monitoring systems of different manufacturers are not exactly the same; Removing duplicate samples from the partial discharge spectrum to obtain a data set to be processed; Based on the Hough change line detection method, background noise is removed from each partial discharge spectrum in the data set to be processed to obtain a first data set.

3. The method for constructing a partial discharge pattern database according to claim 2, characterized in that The Hough change line detection method is used to remove background noise from each partial discharge spectrum in the data set to be processed to obtain a first data set, including: Performing grayscale threshold segmentation on each partial discharge map in the data set to be processed, removing background pixels, and obtaining a first pixel set including partial discharge feature pixels and background grid line pixels; Convert all pixels in the first pixel set to Hough space; According to the representation of each pixel point in the first pixel set in the Hough space, the position and direction of the linear background noise in the first pixel set are located, and the linear background noise pixels are removed from the first pixel set to obtain a first data set.

4. The method for constructing a partial discharge pattern database according to claim 1, characterized in that, The preprocessing of the first data set to obtain a second data set includes: For each partial discharge spectrum in the first data set, unify the background color and detect the background pixels; According to the unified background pixels of each of the partial discharge maps, the redundant borders of the partial discharge maps are corrected and the size is normalized; The grid lines and cosine lines in the background pixel area of ​​each partial discharge spectrum are removed to obtain a second data set; the partial discharge spectrum in the second data set includes partial discharge characteristic pixels and background pixels of uniform color.

5. The method for constructing a partial discharge pattern database according to claim 4, wherein The step of removing the grid lines and cosine lines in the background pixel area of ​​each partial discharge spectrum to obtain a second data set includes: Performing space-frequency transformation on the size-normalized partial discharge spectrum from the spatial domain to the frequency domain; Move the zero frequency point of the space-frequency transform result to the center of the spectrum graph, and calculate the amplitude of the spectrum graph; Transforming the amplitude of the spectrum graph into a logarithmic interval and performing grid filtering; Performing an inverse space-frequency transform on the frequency spectrum after grid filtering to obtain a second data subset; Generate a cosine line with a period of l and an amplitude of k / 2, use the grayscale value of the pixels around the cosine line to replace the cosine line, remove the cosine line in the background in the second data subset, and obtain a second data set; wherein the length of the normalized partial discharge spectrum is l and the width is k.

6. The method for constructing a partial discharge pattern database according to claim 1, wherein The constructing a partial discharge feature set of a partial discharge type according to the second data set includes: Based on the pixel density, an occasional discharge feature detection method is used to determine the density of partial discharge feature pixel points in the second data set within the atlas space, and the occasional discharge features with a density greater than the threshold are removed to obtain a set of partial discharge feature pixel positions; Based on the density-based clustering algorithm, the set of partial discharge feature pixel positions is clustered, and according to the partial discharge types of the partial discharge atlas, partial discharge feature sets of tip discharge, insulation discharge, suspension discharge, and metal particle discharge are generated.

7. The method for constructing a partial discharge pattern database according to claim 6, wherein The density-based clustering algorithm clusters the set of partial discharge feature pixel positions and generates partial discharge feature sets of tip discharge, insulation discharge, suspension discharge, and metal particle discharge according to the partial discharge types of the partial discharge atlas, including: Determine the clustering radius of the clustering algorithm and the minimum number of pixels in its neighborhood space area; For all pixel points in the set of partial discharge feature pixel positions, count the number of pixels in the space area with the clustering radius as the neighborhood for each pixel point; If the number of pixels of any pixel point in the space area with the clustering radius as the neighborhood is greater than the minimum number of pixels, the corresponding pixel point is marked as a core point; Define the core points with overlapping neighborhoods and all pixel points within the neighborhoods as the same partial discharge type; where the partial discharge types include: tip discharge, insulation discharge, suspension discharge, or metal particle partial discharge; Obtain the set of partial discharge feature pixel positions of different partial discharge types, and define the spatial form formed by each pixel point of the same partial discharge type after clustering as the partial discharge feature set of the corresponding partial discharge type.

8. The method for constructing a partial discharge pattern database according to claim 1, wherein Based on the set feature parameters and the partial discharge feature set, a partial discharge atlas database is constructed, including: Generate a full-zero pixel matrix with a size of l×k; In the full-zero pixel matrix, according to the number N of partial discharge features, the scaling ratio σ for setting partial discharge features, the number M of phase shift times, and the position difference between two feature centers in the set feature parameters, the partial discharge feature set is processed to obtain a partial discharge feature atlas; Use the set of partial discharge feature atlases as the partial discharge atlas database.

9. A partial discharge pattern database construction system, characterized in that Including: A cleaning module for collecting partial discharge atlases of multiple manufacturers in multiple formats, and performing sample duplicate cleaning and background noise removal on the collected partial discharge atlases to obtain a first data set; A preprocessing module for preprocessing the first data set to obtain a second data set; A feature set module for constructing a partial discharge feature set of a partial discharge type according to the second data set; A construction module for constructing a partial discharge atlas database based on set feature parameters and the partial discharge feature set.

10. The partial discharge pattern database construction system according to claim 9, characterized in that When the feature set module is used to construct a partial discharge feature set of a partial discharge type according to the second data set, it specifically: Based on the pixel density, an occasional discharge feature detection method is used to determine the density of partial discharge feature pixel points in the second data set within the atlas space, and the occasional discharge features with a density greater than the threshold are removed to obtain a set of partial discharge feature pixel positions; Based on the density-based clustering algorithm, the set of partial discharge feature pixel positions is clustered, and according to the partial discharge types of the partial discharge atlas, partial discharge feature sets of tip discharge, insulation discharge, suspension discharge, and metal particle discharge are generated.

Images