Maintenance processing method and device for fault power equipment and electronic equipment

Abstract

The invention provides a fault power equipment-oriented maintenance processing method and device and electronic equipment, and relates to the technical field of power. The method comprises the steps of performing multi-view feature extraction on multi-type equipment images of fault power equipment, constructing a feature matrix based on extracted image features under multiple views, and constructing a first self-representation adjacency matrix indicating a global geometric structure of the feature matrix through a preset method, and a first graph adjacency matrix indicative of a local geometry of the feature matrix. And then, on the basis of the first self-representation adjacency matrix and the first graph adjacency matrix, the fault power equipment is divided, and at least one equipment fault type is obtained. And determining a target maintenance scheme corresponding to the equipment fault type, and maintaining the corresponding fault power equipment based on the target maintenance scheme. According to the method, the fault power equipment can be accurately divided, so that a proper maintenance scheme is accurately selected, and the maintenance efficiency of the fault power equipment is improved.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a maintenance processing method, device and electronic equipment for faulty electric power equipment. Background Art

[0002] Fault classification of power equipment refers to the process of classifying and sorting faulty power equipment according to specific standards or categories, with the aim of better understanding the patterns and trends of equipment failures so as to more effectively diagnose faults. Fault classification of power equipment is a key step in power equipment management. By classifying faulty power equipment, it helps technicians to quickly take appropriate repair measures, thereby reducing the downtime of faulty power equipment and the impact on power supply.

[0003] The existing technology monitors the power equipment in real time by installing sensors, and obtains the single-view device image of the power equipment from the real-time monitoring. At the same time, a feature extractor is used to extract features from the single-view device image of the power equipment, and the power equipment is analyzed based on the extracted image features to determine whether the power equipment has a fault and what kind of fault has occurred. Appropriate repair measures are further taken to maintain the faulty power equipment.

[0004] However, the existing technology does not accurately classify faulty power equipment and cannot accurately select appropriate maintenance solutions, which leads to low maintenance efficiency of faulty power equipment. Summary of the invention

[0005] The present application provides a maintenance processing method, device and electronic equipment for faulty power equipment, which is used to solve the technical problem in the prior art that the classification of faulty power equipment is not accurate enough, and the appropriate maintenance plan cannot be accurately selected, which leads to low maintenance efficiency of the faulty power equipment.

[0006] In a first aspect, the present application provides a maintenance processing method for faulty power equipment, comprising:

[0007] Acquire multiple types of device images of faulty power equipment, perform multi-view feature extraction on the multiple types of device images, and obtain image features under multiple views;

[0008] Constructing a feature matrix based on the image features under the multiple views, and constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method; wherein the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix;

[0009] Based on the first self-representative adjacency matrix and the first graph adjacency matrix, the faulty power equipment is divided to obtain at least one equipment failure type; a target maintenance plan corresponding to the equipment failure type is determined, and based on the target maintenance plan, the faulty power equipment included in the equipment failure type corresponding to the target maintenance plan is maintained.

[0010] In a possible design, the partitioning of the faulty power equipment based on the first self-representative adjacency matrix and the first graph adjacency matrix includes:

[0011] According to a pre-specified preset dimension, a corresponding target tensor is constructed based on the first self-representation adjacency matrix and the first graph adjacency matrix; wherein the target tensor represents a multidimensional data structure reflecting the global geometric structure and the local geometric structure of the feature matrix;

[0012] The target tensor is subjected to rank reduction processing to convert the target tensor into a corresponding low-rank tensor; the low-rank tensor is subjected to clustering processing based on a preset spectral clustering method to divide the faulty power equipment.

[0013] In a possible design, the low-rank tensor includes a second self-representative adjacency matrix and a second graph adjacency matrix, and data of the second self-representative adjacency matrix and the first self-representative adjacency matrix, and data of the second graph adjacency matrix and the first graph adjacency matrix are different;

[0014] The clustering process of the low-rank tensor based on a preset spectral clustering method to divide the faulty power equipment includes:

[0015] Splitting the low-rank tensor to obtain the second self-representation adjacency matrix and the second graph adjacency matrix in the low-rank tensor;

[0016] averaging the second self-representation adjacency matrix and the second graph adjacency matrix, and determining the average value as a target similarity matrix;

[0017] Based on the target similarity matrix, the faulty power equipment is divided by the preset spectral clustering method.

[0018] In a possible design, the method of dividing the faulty power equipment by the preset spectral clustering method based on the target similarity matrix includes:

[0019] Constructing a corresponding Laplace matrix based on the target similarity matrix, and calculating an eigenvector of the Laplace matrix;

[0020] The feature vector is projected into a preset low-dimensional space, and clustering is performed on the feature vector projected into the preset low-dimensional space; and the faulty power equipment is divided according to the clustering result.

[0021] In a possible design, each view corresponds to a feature matrix, and each feature matrix includes n image features from n device images in the same view; wherein n is a positive integer greater than or equal to 1, indicating the number of acquired device images;

[0022] The step of constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method includes:

[0023] Calculate the first similarity between each image feature in each feature matrix and other n-1 image features to obtain n-1 similarity coefficients; based on the n-1 similarity coefficients, express each image feature as a linear combination of the other n-1 image features; construct a first self-expressive adjacency matrix based on the obtained linear combination; wherein each view corresponds to a first self-expressive adjacency matrix;

[0024] Calculate the distance value between any two image features in each feature matrix, and determine the second similarity between the any two image features based on the distance value; construct a first graph adjacency matrix based on the second similarity; wherein each view corresponds to a first graph adjacency matrix.

[0025] In one possible design, the method further includes:

[0026] Calculating a first reconstruction error between the first self-representative adjacency matrix and a corresponding feature matrix, and if the first reconstruction error is within a first error range, retaining the first self-representative adjacency matrix; if the first reconstruction error is not within the first error range, reconstructing the first self-representative adjacency matrix until the first reconstruction error is within the first error range;

[0027] A second reconstruction error between the first graph adjacency matrix and the corresponding feature matrix is ​​calculated. If the second reconstruction error is within a second error range, the first graph adjacency matrix is ​​retained; if the second reconstruction error is not within the second error range, the first graph adjacency matrix is ​​reconstructed until the second reconstruction error is within the second error range.

[0028] In a possible design, before the rank reduction processing is performed on the target tensor, the method further includes:

[0029] Determine the dimension and position of the first self-representation adjacency matrix and the first graph adjacency matrix in the target tensor;

[0030] According to the dimension and the position, the target tensor is rotated based on a preset target.

[0031] In a second aspect, the present application provides a maintenance processing device for faulty power equipment, comprising:

[0032] An acquisition module, used for acquiring a device image of a faulty power device;

[0033] An extraction module, used for performing multi-view feature extraction on the device image to obtain image features under multiple views;

[0034] A construction module, configured to construct a feature matrix based on the image features under the multiple views, and to construct a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method; wherein the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix;

[0035] A processing module is used to classify the faulty power equipment based on the first self-representative adjacency matrix and the first graph adjacency matrix to obtain at least one equipment fault type; determine a target maintenance plan corresponding to the equipment fault type, and based on the target maintenance plan, maintain the faulty power equipment included in the equipment fault type corresponding to the target maintenance plan.

[0036] In a possible design, the construction module is further used to construct a corresponding target tensor based on the first self-representation adjacency matrix and the first graph adjacency matrix according to a pre-specified preset dimension; wherein the target tensor represents a multidimensional data structure reflecting the global geometric structure and the local geometric structure of the feature matrix;

[0037] The processing module is further used to perform rank reduction processing on the target tensor to convert the target tensor into a corresponding low-rank tensor;

[0038] The processing module further includes: a clustering module, configured to perform clustering processing on the low-rank tensor based on a preset spectral clustering method to divide the faulty power equipment.

[0039] In a possible design, the low-rank tensor includes a second self-representative adjacency matrix and a second graph adjacency matrix, and data of the second self-representative adjacency matrix and the first self-representative adjacency matrix, and data of the second graph adjacency matrix and the first graph adjacency matrix are different;

[0040] The clustering module further includes: a division module, a calculation module, and a determination module.

[0041] The partitioning module is used to partition the low-rank tensor to obtain the second self-representation adjacency matrix and the second graph adjacency matrix in the low-rank tensor;

[0042] The calculation module is used to average the second self-representation adjacency matrix and the second graph adjacency matrix;

[0043] The determination module is used to determine the average value as a target similarity matrix;

[0044] The partitioning module is further configured to partition the faulty power equipment based on the target similarity matrix by using the preset spectral clustering method.

[0045] In a possible design, the calculation module is further used to construct a corresponding Laplace matrix based on the target similarity matrix, and calculate the eigenvector of the Laplace matrix;

[0046] The clustering module is further used to project the feature vectors into a preset low-dimensional space and perform clustering processing on the feature vectors projected into the preset low-dimensional space;

[0047] The classification module is also used to classify the faulty power equipment according to the clustering result.

[0048] In a possible design, each view corresponds to a feature matrix, and each feature matrix includes n image features from n device images in the same view; wherein n is a positive integer greater than or equal to 1, indicating the number of acquired device images;

[0049] The calculation module is further used to calculate the first similarity between each image feature in each feature matrix and other n-1 image features to obtain n-1 similarity coefficients;

[0050] The construction module is further used to represent each image feature as a linear combination of other n-1 image features based on the n-1 similarity coefficients; and construct a first self-representation adjacency matrix based on the obtained linear combination; wherein each view corresponds to a first self-representation adjacency matrix;

[0051] The calculation module is further used to calculate the distance value between any two image features in each feature matrix;

[0052] The construction module is further used to determine the second similarity between the arbitrary two image features based on the distance value; and construct a first graph adjacency matrix based on the second similarity; wherein each view corresponds to a first graph adjacency matrix.

[0053] In a possible design, the calculation module is further used to calculate a first reconstruction error between the first self-representation adjacency matrix and a corresponding feature matrix;

[0054] The processing module is further configured to retain the first self-representation adjacency matrix if the first reconstruction error is within a first error range; and reconstruct the first self-representation adjacency matrix if the first reconstruction error is not within the first error range until the first reconstruction error is within the first error range;

[0055] The calculation module is further used to calculate a second reconstruction error between the first graph adjacency matrix and the corresponding feature matrix;

[0056] The processing module is further configured to retain the first graph adjacency matrix if the second reconstruction error is within a second error range; and reconstruct the first graph adjacency matrix if the second reconstruction error is not within the second error range until the second reconstruction error is within the second error range.

[0057] In a possible design, the determination module is further used to determine the dimension and position of the first self-representation adjacency matrix and the first graph adjacency matrix in the target tensor;

[0058] The processing module is further used to perform rotation processing on the target tensor based on the dimension and the position and based on a preset target.

[0059] In a third aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method described in the first aspect above and various possible designs.

[0060] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the method described in the first aspect and various possible designs is implemented.

[0061] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect and various possible designs of the first aspect.

[0062] The maintenance processing method, device and electronic device for faulty power equipment provided by the present application obtain multiple types of equipment images of faulty power equipment, perform multi-view feature extraction on the multi-type equipment images, and obtain image features under multiple views. Afterwards, a feature matrix is ​​constructed based on the image features under multiple views, and based on the feature matrix, a corresponding first self-representation adjacency matrix and a first graph adjacency matrix are constructed by a preset method. Among them, the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix. By extracting the multi-view image features of the faulty power equipment and adopting two different methods of measuring sample similarity, namely self-representation learning and graph learning, the adjacency matrix of the feature matrix is ​​obtained respectively, so as to fully explore the intrinsic structural information of the feature matrix. Next, based on the first self-representation adjacency matrix and the first graph adjacency matrix, the faulty power equipment is divided to obtain at least one equipment fault type. Determine the target maintenance plan corresponding to the equipment fault type, and based on the target maintenance plan, maintain the faulty power equipment included in the equipment fault type corresponding to the target maintenance plan. The image features under multiple views corresponding to multi-type equipment images can fully reflect the characteristic information of faulty power equipment, and the first self-representation adjacency matrix and the first graph adjacency matrix retain both the consistency information of the feature matrix and the heterogeneous information from a global and local perspective. Therefore, the first self-representation adjacency matrix and the first graph adjacency matrix can accurately divide the faulty power equipment, and then accurately select the appropriate maintenance plan to improve the maintenance efficiency of the faulty power equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0064] Figure 1 Schematic diagram of the maintenance process of the faulty power equipment provided in the embodiment of the present application Figure 1 ;

[0065] Figure 2 A structural diagram of a maintenance processing method for faulty power equipment provided in an embodiment of the present application;

[0066] Figure 3 Schematic diagram of the maintenance process of the faulty power equipment provided in the embodiment of the present application Figure 2 ;

[0067] Figure 4 A schematic diagram of the structure of a maintenance processing device for faulty power equipment provided in an embodiment of the present application;

[0068] Figure 5A hardware structure diagram of an electronic device provided in an embodiment of the present application.

[0069] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0070] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0071] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein, for example.

[0072] In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0073] It should be noted that the user 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, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0074] Various faults may occur in power equipment during operation, which may affect the normal operation of power equipment and even cause power outages or damage to power equipment. Therefore, identifying the faults of power equipment is the basis for ensuring the stability and reliability of the power system.

[0075] Since different types of faults may require different maintenance methods, by classifying faulty power equipment, maintenance and repair plans can be formulated more effectively. At the same time, through systematic fault classification of power equipment, resource allocation can be optimized according to fault type, frequency and impact range, reducing unnecessary downtime and maintenance costs.

[0076] The existing technology realizes real-time monitoring of power equipment by installing sensors on the power equipment, and captures the single-view device image of the power equipment through these sensors. A feature extractor is used to extract features from the single-view device image of the power equipment, and the power equipment is analyzed based on the extracted image features to determine whether a fault occurs and what kind of fault it is.

[0077] However, in the era of big data, the channels for obtaining equipment images, the ways of collecting equipment images, the means of processing equipment images, and the diversity of feature extractors make equipment images present different sources, different modes, and different features. It can be seen that equipment images in real life are diversified, and single-view equipment images cannot fully contain the description information of power equipment.

[0078] The effect of faulty power equipment segmentation is largely affected by the quality of the original equipment image. If the original equipment image is insufficient or inaccurate, it may not be possible to effectively identify or distinguish different types of faulty power equipment. In addition, power equipment may have a variety of complex failure modes, and it is impossible to accurately segment faulty power equipment with a single-view equipment image alone.

[0079] Inaccurate fault classification may lead to misdiagnosis or missed diagnosis. Misdiagnosis may lead to unnecessary maintenance operations, while missed diagnosis may lead to the failure to handle the fault in a timely manner. If the faulty power equipment is not correctly identified, it will lead to the inability to accurately select the appropriate maintenance plan, and the selected maintenance plan may not be suitable for the actual situation, which will lead to low maintenance efficiency of the faulty power equipment.

[0080] In view of the above technical problems, in order to solve the technical problem that the classification of faulty power equipment is not accurate enough due to relying only on single-view device images, the inventors have come up with the idea of ​​collecting multiple types of device images of faulty power equipment to comprehensively reflect the diverse information of the faulty power equipment. At the same time, in order to further improve the accuracy of the classification of faulty power equipment, the inventors consider performing multi-view feature extraction on the collected multiple types of device images to obtain image features of multiple types of device images under multiple views. The faulty power equipment is accurately classified according to the extracted image features, and the appropriate maintenance plan is accurately selected according to the fault type after classification to maintain the power equipment.

[0081] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0082] The embodiment of the present application provides a maintenance processing method for faulty power equipment. Figure 1 Schematic diagram of the maintenance process of the faulty power equipment provided in the embodiment of the present application Figure 1 ,like Figure 1 As shown, the maintenance processing method for faulty power equipment includes:

[0083] S101, obtaining multiple types of device images of faulty power equipment, performing multi-view feature extraction on the multiple types of device images, and obtaining image features under multiple views.

[0084] Specifically, multiple types of equipment images can be taken on-site by using an image acquisition device carried by humans for the faulty part of the faulty power equipment; or multiple types of equipment images can be taken by using a camera carried by a substation inspection robot for the faulty part of the faulty power equipment; or multiple types of equipment images can be taken by using a camera carried by a drone for the faulty part of the faulty power equipment; or multiple types of equipment images can be obtained through other means. How to obtain multiple types of equipment images of the faulty power equipment depends on the actual situation and is not specifically limited here.

[0085] Among them, there are multiple types of equipment images of faulty power equipment such as visible light images, infrared images, hyperspectral images, etc.

[0086] Next, multi-view feature extraction is performed on multiple types of device images. The image features obtained under multiple views include but are not limited to the following features: (1) Low-level features refer to features that can be obtained through simple operations based on the device image itself, such as color, texture, shape, gradient, etc. (2) Mid-level features refer to features obtained by multi-feature fusion processing based on low-level features, mainly the diversity fusion of color and texture. (3) Deep features refer to the use of large-scale neural network models such as VGG network and residual network to mine deeper and more abstract features of device images.

[0087] S102, constructing a feature matrix based on the image features in multiple views, and constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method.

[0088] Each view corresponds to a feature matrix, and each feature matrix includes Device images in the same view image features, is a positive integer greater than or equal to 1. In order to facilitate the description of mathematical formulas, the constructed multiple feature matrices are collectively referred to as multi-view feature sets. ,in, Indicates The feature matrix corresponding to each view is Indicates the number of device images. Indicates The feature dimensions of the view, Indicates the number of views.

[0089] In order to better understand and mine the structure and relationship of feature matrices, self-representation learning and graph learning are applied to each feature matrix respectively, association analysis is performed from different angles, and the corresponding first self-representation adjacency matrix and first graph adjacency matrix are constructed.

[0090] Among them, self-representation learning represents each image feature by using a linear combination of the image features themselves, thereby forming the first self-representation adjacency matrix. Self-representation learning focuses on learning its representation by reconstructing the input data, usually by minimizing the reconstruction error. The first self-representation adjacency matrix aims to reveal the global geometric structure of the feature matrix.

[0091] Graph learning constructs the first graph adjacency matrix by measuring the distance between image features. Graph learning captures the relationship between image features by constructing a graph structure (including representations of nodes and edges). Therefore, the first graph adjacency matrix focuses more on revealing the local geometric structure of the feature matrix.

[0092] Specifically, the first similarity between each image feature in each feature matrix and the other n-1 image features is calculated to obtain n-1 similarity coefficients; based on the n-1 similarity coefficients, each image feature is represented as a linear combination of the other n-1 image features; a first self-expressive adjacency matrix is ​​constructed based on the obtained linear combination, wherein each view corresponds to a first self-expressive adjacency matrix. The distance value between any two image features in each feature matrix is ​​calculated, the second similarity between any two image features is determined based on the calculated distance value, and a first graph adjacency matrix is ​​constructed based on the second similarity, wherein each view corresponds to a first graph adjacency matrix.

[0093] Explanatory, each data in the first self-representation adjacency matrix refers to the weight that the image feature can be represented by a linear combination of other image features.

[0094] It should be noted that after the first self-representation adjacency matrix and the first graph adjacency matrix are constructed, the first self-representation adjacency matrix and the first graph adjacency matrix need to be checked for correctness. In a possible implementation, the first reconstruction error between the first self-representation adjacency matrix and the corresponding feature matrix is ​​calculated. If the first reconstruction error is within the first error range, the first self-representation adjacency matrix is ​​retained. If the first reconstruction error is not within the first error range, the first self-representation adjacency matrix is ​​reconstructed until the first reconstruction error is within the first error range.

[0095] The second reconstruction error between the first graph adjacency matrix and the corresponding feature matrix is ​​calculated. If the second reconstruction error is within the second error range, the first graph adjacency matrix is ​​retained. If the second reconstruction error is not within the second error range, the first graph adjacency matrix is ​​reconstructed until the second reconstruction error is within the second error range.

[0096] S103. Based on the first self-representative adjacency matrix and the first graph adjacency matrix, the faulty power equipment is divided to obtain at least one equipment failure type; a target maintenance plan corresponding to the equipment failure type is determined, and based on the target maintenance plan, the faulty power equipment included in the equipment failure type corresponding to the target maintenance plan is maintained.

[0097] Since the first self-representation adjacency matrix and the first graph adjacency matrix both represent the relationship between the image features of the feature matrix, ideally, the first self-representation adjacency matrix and the first graph adjacency matrix should be similar. Under this assumption, the first self-representation adjacency matrix and the first graph adjacency matrix can be stacked into a target tensor according to a pre-specified preset dimension. The target tensor is a multidimensional data structure that reflects the global and local geometric structures of the feature matrix.

[0098] It is understandable that since the target tensor is a multidimensional array structure, it usually requires a lot of computing resources to process. The method of rotating the target tensor can be used to reduce the computational complexity. Specifically, the dimension and position of the first self-representation adjacency matrix and the first graph adjacency matrix in the target tensor are determined. According to the dimension and position, the target tensor is rotated based on the preset target.

[0099] In addition, since the target tensor often contains noise, the rotated target tensor can be subjected to rank reduction processing to approximate the original target tensor, thereby achieving the goal of removing noise while retaining the main structure and features of the original target tensor. The conversion of the rotated target tensor into the corresponding low-rank tensor includes the second self-representation adjacency matrix and the second graph adjacency matrix.

[0100] It should be understood that the data of the second self-representative adjacency matrix is ​​different from that of the first self-representative adjacency matrix, and that of the second graph adjacency matrix is ​​different from that of the first graph adjacency matrix. This is because when constructing the target tensor, the first self-representative adjacency matrix and the first graph adjacency matrix are fused or combined in some form. The second self-representative adjacency matrix and the second graph adjacency matrix in the low-rank tensor capture the main features of the target tensor rather than complete information.

[0101] Furthermore, after the rotated target tensor is converted into a corresponding low-rank tensor, the low-rank tensor is clustered based on a preset spectral clustering method to divide the faulty power equipment.

[0102] Explanatory,preset spectral clustering method is a clustering method based on graph theory,,the core idea is to regard data points as nodes of a graph,construct the graph by calculating the similarity between nodes,and then use the spectral properties (eigenvalues ​​and eigenvectors) of the graph to,identify natural clusters in the data.

[0103] Specifically, the low-rank tensor is segmented to obtain a second self-representation adjacency matrix and a second graph adjacency matrix in the low-rank tensor. The second self-representation adjacency matrix and the second graph adjacency matrix are averaged, and the average is determined as a target similarity matrix. Based on the target similarity matrix, the faulty power equipment is divided by a preset spectral clustering method.

[0104] Among them, the specific process of dividing the faulty power equipment by the preset spectral clustering method based on the target similarity matrix is ​​as follows: constructing the corresponding Laplace matrix based on the target similarity matrix, calculating the eigenvector of the Laplace matrix, projecting the eigenvector to the preset low-dimensional space, clustering the eigenvector projected to the preset low-dimensional space, and dividing the faulty power equipment according to the clustering results.

[0105] Next, the classification process of faulty power equipment is explained in detail through a specific example. In a specific example, according to the multi-view feature set constructed based on image features, , self-representation learning and graph learning are applied to each feature matrix in the multi-view feature set respectively.

[0106] Among them, self-representation learning obtains the first self-representation adjacency matrix by minimizing the following formula (1-1): :

[0107] (1-1)

[0108] in, Indicates The feature matrix of each view, It is The first self-representation adjacency matrix of the views. Indicates of the views Image features of faulty power equipment, yes No. Ledi Elements of a row. is a column vector whose elements are all 1, represents the transpose of a matrix, and diag(·) represents the diagonal elements of a matrix. represents the square of the Frobenius norm of the matrix, is a balance parameter.

[0109] Formula (1-1) mainly focuses on the global geometric structure of the feature matrix hidden in multiple views. In order to better reveal the local geometric structure of the feature matrix, graph learning is also introduced. Graph learning obtains the first graph adjacency matrix by minimizing the following formula (1-2): :

[0110] (1-2)

[0111] in, Indicates The feature matrix of each view, It is The first graph adjacency matrix of the views. Indicates of the views Image features of faulty power equipment, yes No. Ledi Elements of a row. is a column vector whose elements are all 1, Represents the transpose of a matrix. represents the square of the L2-norm of the matrix, is a balance parameter.

[0112] Furthermore, the first self-representation adjacency matrix based on self-representation learning and the first graph adjacency matrix based on graph learning are stacked into a target tensor, and the target tensor is rotated to be , which allows for better study of correlations and greatly reduces computational complexity.

[0113] In addition, considering the noise The impact of To approximate , whose mathematical formula is defined as:

[0114] (1-3)

[0115] in, is a constant that controls the effect of noise, Represents a low-rank tensor The nuclear norm of the low-rank tensor The norm table is defined as:

[0116]

[0117] Low rank tensor With the first self-representing adjacency matrix And the first graph adjacency matrix is in a separated state, which may make the solution suboptimal. Therefore, the first self-representing adjacency matrix in formula (1-1) is , the first graph adjacency matrix in formula (1-2) , and the low-rank tensor in formula (1-3) Combined and optimized together. Therefore, the final objective function can be expressed as:

[0118] (1-4)

[0119] Among them, the function For each view and Merge into a third-order tensor, and then Axis rotation.

[0120] Furthermore, through the learned low-rank tensor To calculate the target similarity matrix:

[0121]

[0122] And the preset spectral clustering method is used to obtain the final division results of faulty power equipment.

[0123] The maintenance processing method for faulty power equipment provided in the present application obtains multi-type equipment images of the faulty power equipment, performs multi-view feature extraction on the multi-type equipment images, and obtains image features under multiple views. Afterwards, a feature matrix is ​​constructed based on the image features under multiple views, and each image feature in each feature matrix is ​​represented as a linear combination of other image features by calculating the first similarity and similarity coefficient between each image feature and other image features, thereby constructing a first self-representation adjacency matrix. At the same time, by calculating the distance value between any two image features in each feature matrix, and determining the second similarity between any two image features based on the distance value, a first graph adjacency matrix is ​​constructed. Among them, the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix. By extracting multi-view image features of the faulty power equipment and adopting two different methods of measuring sample similarity, namely self-representation learning and graph learning, the adjacency matrix of the feature matrix is ​​obtained respectively, so as to fully explore the intrinsic structural information of the feature matrix. Next, the first self-expressive adjacency matrix and the first graph adjacency matrix are stacked into a target tensor, and the target tensor is rank-reduced to convert the target tensor into a low-rank tensor containing the second self-expressive adjacency matrix and the second graph adjacency matrix. The low-rank tensor is segmented to obtain the second self-expressive adjacency matrix and the second graph adjacency matrix, and then the second self-expressive adjacency matrix and the second graph adjacency matrix are averaged to obtain the corresponding target similarity matrix. Based on the target similarity matrix, the faulty power equipment is divided by a preset spectral clustering method to obtain at least one equipment fault type. The target maintenance plan corresponding to the equipment fault type is determined, and based on the target maintenance plan, the faulty power equipment included in the equipment fault type corresponding to the target maintenance plan is maintained. The image features under multiple views corresponding to the multi-type equipment images can fully reflect the feature information of the faulty power equipment, and the first self-expressive adjacency matrix and the first graph adjacency matrix retain both the consistency information of the feature matrix and the heterogeneous information from a global and local perspective. Therefore, the first self-representation adjacency matrix and the first graph adjacency matrix can be used to accurately divide the faulty power equipment, and then accurately select the appropriate maintenance plan to improve the maintenance efficiency of the faulty power equipment.

[0124] Next, a specific embodiment is used to Figure 1 A summary description is given of the classification process of the faulty electrical equipment involved. Figure 2 A structural diagram of a maintenance processing method for a faulty power device provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the maintenance processing method for faulty power equipment includes: an image acquisition unit, a feature extraction unit, an adjacency matrix learning unit, a tensor constraint unit, a similarity matrix learning unit, and a spectral clustering unit.

[0125] The image acquisition unit is used to acquire various types of equipment images (visible light, infrared, and hyperspectral) of the faulty power equipment, including a visible light image acquisition subunit, an infrared image acquisition subunit, and a hyperspectral image acquisition subunit.

[0126] The feature extraction unit is used to extract features of multiple views of the device image to obtain image features under multiple views, including a bottom-level feature extraction subunit, a middle-level feature extraction subunit and a deep feature extraction subunit.

[0127] The adjacency matrix learning unit is used to construct a first self-representation adjacency matrix that reveals the global geometric structure of the feature matrix through self-representation learning technology, and to construct a first graph adjacency matrix that reveals the local geometric structure of the feature matrix through graph learning technology, including a self-representation learning subunit and a graph learning subunit.

[0128] The tensor constraint unit is used to fully explore the high-order correlation and spatial structure information between multiple views, the similarity matrix learning unit is used to construct the target similarity matrix, and the spectral clustering unit is used to perform spectral clustering to obtain the final partitioning result of the faulty power equipment.

[0129] It should be noted that the image acquisition unit is connected to the feature extraction unit, the feature extraction unit is connected to the adjacency matrix learning unit, the adjacency matrix learning unit is connected to the tensor constraint unit, the tensor constraint unit is connected to the similarity matrix learning unit, and the similarity matrix learning unit is connected to the spectral clustering unit.

[0130] based on Figure 2 The structure shown, Figure 3 Schematic diagram of the maintenance process of the faulty power equipment provided in the embodiment of the present application Figure 2 ,like Figure 3 As shown, the maintenance processing method for faulty power equipment specifically includes the following steps:

[0131] S301, acquiring different types of device images of faulty power equipment through an image acquisition unit;

[0132] S302, performing multi-view feature extraction on different types of device images through a feature extraction unit to obtain image features under multiple views, and constructing a feature matrix based on the image features under multiple views;

[0133] S303, constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix through an adjacency matrix learning unit;

[0134] Among them, S303 specifically includes: step a, using self-representation learning on the feature matrix through the self-representation learning subunit to construct a first self-representation adjacency matrix; step b, using graph learning on the feature matrix through the graph learning subunit to construct a first graph adjacency matrix.

[0135] S304, stacking the first self-representation adjacency matrix and the first graph adjacency matrix into a target tensor through a tensor constraint unit;

[0136] S305, converting the target tensor into a corresponding low-rank tensor through a similarity matrix learning unit, and determining a target similarity matrix based on the low-rank tensor;

[0137] S306, dividing the faulty power equipment through a spectral clustering unit based on the target similarity matrix;

[0138] S307: Determine at least one equipment failure type according to the classification result of the faulty power equipment, and perform maintenance on the faulty power equipment based on a target maintenance plan corresponding to the equipment failure type.

[0139] By fully considering the impact of different types of equipment images and image features of different views on the division of faulty power equipment. Two different methods of measuring the similarity of feature matrices are used to obtain the adjacency matrix of multi-source feature matrices respectively, so as to mine the global and local geometric structures of the feature matrices, thereby providing diversified information for the division of faulty power equipment. The target tensor is further introduced to fully mine the high-order correlation and spatial structure information between multiple views. At the same time, the target similarity matrix is ​​calculated by the learned low-rank tensor, and the preset spectral clustering method is used to obtain the final accurate division result of the faulty power equipment. Then, according to the accurate fault division results, the appropriate maintenance plan is accurately selected to improve the maintenance efficiency of the faulty power equipment.

[0140] Figure 4 A schematic diagram of the structure of a maintenance processing device for faulty power equipment provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, the maintenance processing device 400 for faulty power equipment includes: an acquisition module 401, an extraction module 402, a construction module 403, and a processing module 404;

[0141] Wherein, the acquisition module 401 is used to acquire the device image of the faulty power equipment;

[0142] An extraction module 402 is used to perform multi-view feature extraction on the device image to obtain image features under multiple views;

[0143] A construction module 403 is used to construct a feature matrix based on the image features under multiple views, and to construct a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method; wherein the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the matrix features;

[0144] The processing module 404 is used to classify the faulty power equipment based on the first self-representative adjacency matrix and the first graph adjacency matrix to obtain at least one equipment fault type; determine the target maintenance plan corresponding to the equipment fault type, and based on the target maintenance plan, maintain the faulty power equipment included in the equipment fault type corresponding to the target maintenance plan.

[0145] In a possible design, the construction module 403 is further used to construct a corresponding target tensor based on the first self-representation adjacency matrix and the first graph adjacency matrix according to a pre-specified preset dimension; wherein the target tensor represents a multidimensional data structure reflecting the global geometric structure and the local geometric structure of the feature matrix;

[0146] The processing module 404 is further used to perform rank reduction processing on the target tensor to convert the target tensor into a corresponding low-rank tensor;

[0147] The processing module 404 further includes: a clustering module 405, which is used to perform clustering processing on the low-rank tensor based on a preset spectral clustering method to divide the faulty power equipment.

[0148] In a possible design, the low-rank tensor includes a second self-representative adjacency matrix and a second graph adjacency matrix, and data of the second self-representative adjacency matrix and the first self-representative adjacency matrix, and data of the second graph adjacency matrix and the first graph adjacency matrix are different;

[0149] The clustering module 405 further includes: a partitioning module 406, a calculation module 407, and a determination module 408.

[0150] A partitioning module 406 is used to partition the low-rank tensor to obtain a second self-representation adjacency matrix and a second graph adjacency matrix in the low-rank tensor;

[0151] A calculation module 407, configured to average the second self-representation adjacency matrix and the second graph adjacency matrix;

[0152] A determination module 408, configured to determine the average value as a target similarity matrix;

[0153] The division module 406 is further configured to divide the faulty power equipment by a preset spectral clustering method based on the target similarity matrix.

[0154] In a possible design, the calculation module 407 is further used to construct a corresponding Laplace matrix based on the target similarity matrix and calculate the eigenvector of the Laplace matrix;

[0155] The clustering module 405 is further used to project the feature vectors into a preset low-dimensional space and perform clustering processing on the feature vectors projected into the preset low-dimensional space;

[0156] The classification module 406 is further used to classify the faulty power equipment according to the clustering result.

[0157] In a possible design, each view corresponds to a feature matrix, and each feature matrix includes n image features from n device images in the same view; wherein n is a positive integer greater than or equal to 1, indicating the number of acquired device images;

[0158] The calculation module 407 is further used to calculate the first similarity between each image feature in each feature matrix and other n-1 image features to obtain n-1 similarity coefficients;

[0159] The construction module 403 is further used to represent each image feature as a linear combination of other n-1 image features based on the n-1 similarity coefficients; and construct a first self-representation adjacency matrix based on the obtained linear combination; wherein each view corresponds to a first self-representation adjacency matrix;

[0160] The calculation module 407 is further used to calculate the distance value between any two image features in each feature matrix;

[0161] The construction module 403 is further used to determine a second similarity between any two image features based on the distance value; and to construct a first graph adjacency matrix based on the second similarity; wherein each view corresponds to a first graph adjacency matrix.

[0162] In a possible design, the calculation module 407 is further used to calculate a first reconstruction error between the first self-representation adjacency matrix and the corresponding feature matrix;

[0163] The processing module 404 is further configured to retain the first self-representation adjacency matrix if the first reconstruction error is within the first error range; and reconstruct the first self-representation adjacency matrix if the first reconstruction error is not within the first error range until the first reconstruction error is within the first error range.

[0164] The calculation module 407 is further used to calculate a second reconstruction error between the first graph adjacency matrix and the corresponding feature matrix;

[0165] The processing module 404 is further configured to retain the first graph adjacency matrix if the second reconstruction error is within the second error range; and reconstruct the first graph adjacency matrix if the second reconstruction error is not within the second error range until the second reconstruction error is within the second error range.

[0166] In a possible design, the determination module 408 is further used to determine the dimension and position of the first self-representation adjacency matrix and the first graph adjacency matrix in the target tensor;

[0167] The processing module 404 is further used to perform rotation processing on the target tensor based on the preset target according to the dimension and the position.

[0168] The maintenance processing device for faulty power equipment provided in the embodiment of the present application can be used to execute the maintenance processing method for faulty power equipment in any of the above embodiments. Its implementation principle and technical effects are similar and will not be repeated here.

[0169] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0170] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown, the electronic device may include: a transceiver 51 , a processor 52 , and a memory 53 .

[0171] The processor 52 executes the computer execution instructions stored in the memory, so that the processor 52 executes the scheme in the above embodiment. The processor 52 can be a general-purpose processor, including a central processing unit CPU, a network processor (NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0172] The memory 53 is connected to the processor 52 via a system bus and completes communication between them. The memory 53 is used to store computer program instructions.

[0173] The transceiver 51 may be used to communicate with other devices.

[0174] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.

[0175] The electronic device provided in the embodiments of the present application can be used to execute the method provided in any of the above embodiments. The implementation principles and technical effects are similar and will not be repeated here.

[0176] An embodiment of the present application further provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the method provided in any of the above embodiments.

[0177] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, the method provided in any of the above embodiments can be implemented.

[0178] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0179] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0180] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0181] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.

[0182] It should be understood that the above processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0183] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0184] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0185] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0186] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a main control device.

[0187] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A maintenance method for faulty power equipment, characterized in that: include: Acquire multiple types of device images of faulty power equipment, perform multi-view feature extraction on the multiple types of device images, and obtain image features under multiple views; Constructing a feature matrix based on the image features under the multiple views, and constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method; wherein the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix; Based on the first self-representative adjacency matrix and the first graph adjacency matrix, the faulty power equipment is divided to obtain at least one equipment failure type; a target maintenance plan corresponding to the equipment failure type is determined, and based on the target maintenance plan, the faulty power equipment included in the equipment failure type corresponding to the target maintenance plan is maintained.

2. The method according to claim 1, characterized in that The dividing the faulty power equipment based on the first self-representation adjacency matrix and the first graph adjacency matrix includes: According to a pre-specified preset dimension, a corresponding target tensor is constructed based on the first self-representation adjacency matrix and the first graph adjacency matrix; wherein the target tensor represents a multidimensional data structure reflecting the global geometric structure and the local geometric structure of the feature matrix; The target tensor is subjected to rank reduction processing to convert the target tensor into a corresponding low-rank tensor; the low-rank tensor is subjected to clustering processing based on a preset spectral clustering method to divide the faulty power equipment.

3. The method according to claim 2, characterized in that The low-rank tensor includes a second self-representative adjacency matrix and a second graph adjacency matrix, and the data of the second self-representative adjacency matrix and the first self-representative adjacency matrix, the second graph adjacency matrix and the first graph adjacency matrix are different; The clustering process of the low-rank tensor based on a preset spectral clustering method to divide the faulty power equipment includes: Splitting the low-rank tensor to obtain the second self-representation adjacency matrix and the second graph adjacency matrix in the low-rank tensor; averaging the second self-representation adjacency matrix and the second graph adjacency matrix, and determining the average value as a target similarity matrix; Based on the target similarity matrix, the faulty power equipment is divided by the preset spectral clustering method.

4. The method according to claim 3, characterized in that The method of dividing the faulty power equipment based on the target similarity matrix by using the preset spectral clustering method includes: Constructing a corresponding Laplace matrix based on the target similarity matrix, and calculating an eigenvector of the Laplace matrix; The feature vector is projected into a preset low-dimensional space, and the feature vector projected into the preset low-dimensional space is clustered; and the faulty power equipment is divided according to the clustering result.

5. The method according to claim 1, characterized in that Each view corresponds to a feature matrix, and each feature matrix includes n image features from n device images in the same view; where n is a positive integer greater than or equal to 1, indicating the number of acquired device images; The step of constructing a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method includes: Calculate the first similarity between each image feature in each feature matrix and other n-1 image features to obtain n-1 similarity coefficients; based on the n-1 similarity coefficients, express each image feature as a linear combination of the other n-1 image features; construct a first self-expressive adjacency matrix based on the obtained linear combination; wherein each view corresponds to a first self-expressive adjacency matrix; Calculate the distance value between any two image features in each feature matrix, and determine the second similarity between the any two image features based on the distance value; construct a first graph adjacency matrix based on the second similarity; wherein each view corresponds to a first graph adjacency matrix.

6. The method according to claim 5, characterized in that The method further comprises: Calculating a first reconstruction error between the first self-representative adjacency matrix and a corresponding feature matrix, and if the first reconstruction error is within a first error range, retaining the first self-representative adjacency matrix; if the first reconstruction error is not within the first error range, reconstructing the first self-representative adjacency matrix until the first reconstruction error is within the first error range; A second reconstruction error between the first graph adjacency matrix and the corresponding feature matrix is ​​calculated. If the second reconstruction error is within a second error range, the first graph adjacency matrix is ​​retained; if the second reconstruction error is not within the second error range, the first graph adjacency matrix is ​​reconstructed until the second reconstruction error is within the second error range.

7. The method according to any one of claims 2 to 4, characterized in that Before the rank reduction processing is performed on the target tensor, the method further includes: Determine the dimension and position of the first self-representation adjacency matrix and the first graph adjacency matrix in the target tensor; According to the dimension and the position, the target tensor is rotated based on a preset target.

8. A maintenance processing device for faulty power equipment, characterized in that: include: An acquisition module, used for acquiring a device image of a faulty power device; An extraction module, used for performing multi-view feature extraction on the device image to obtain image features under multiple views; A construction module, configured to construct a feature matrix based on the image features under the multiple views, and to construct a corresponding first self-representation adjacency matrix and a first graph adjacency matrix based on the feature matrix by a preset method; wherein the first self-representation adjacency matrix is ​​used to indicate the global geometric structure of the feature matrix, and the first graph adjacency matrix is ​​used to indicate the local geometric structure of the feature matrix; A processing module is used to classify the faulty power equipment based on the first self-representative adjacency matrix and the first graph adjacency matrix to obtain at least one equipment fault type; determine a target maintenance plan corresponding to the equipment fault type, and based on the target maintenance plan, maintain the faulty power equipment included in the equipment fault type corresponding to the target maintenance plan.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the maintenance processing method for faulty power equipment according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the maintenance processing method for faulty power equipment according to any one of claims 1 to 7.

Images