Power grid fault graph intelligent identification and classification early warning method, system, equipment and medium

Abstract

The invention discloses a power grid fault graph intelligent identification and classification early warning method, system and device and a medium, and the method comprises the steps: obtaining a to-be-identified power graph, inputting the power graph into a power grid fault graph intelligent identification model, and obtaining a power fault graph identification result; based on a first clustering algorithm, carrying out clustering processing on the power fault graph recognition result to obtain a power fault graph classification result; and an early warning report is generated based on the power fault graph classification result, so that classified early warning is realized, and the influence of faults on production and operation is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of power grid fault graphic recognition technology, and in particular to a method, system, device and medium for intelligent recognition, classification and early warning of power grid fault graphics. Background Technology

[0002] With the rapid development of the power industry, the stability and reliability of the power grid have become crucial factors in ensuring social production and people's lives. However, during operation, the power grid may experience faults due to various reasons such as natural factors, equipment aging, and human error, leading to serious consequences such as power outages, equipment damage, and even grid paralysis. Therefore, timely and effective monitoring, diagnosis, and early warning of power grid faults are of great significance for ensuring the safe and stable operation of the power grid.

[0003] Traditional methods for monitoring and diagnosing power grid faults mainly rely on manual inspections and experience-based judgments. This approach is not only time-consuming and labor-intensive, but also inefficient, and it is difficult to achieve real-time monitoring and intelligent early warning of the power grid status. Therefore, there is an urgent need in the existing technology for a graphical intelligent recognition and classification early warning method for power grid faults to solve the above problems. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, device, and medium for intelligent graphic recognition and classification of power grid faults to solve the technical problems of existing power grid fault monitoring and diagnosis methods, which mainly rely on manual inspection and experience-based judgment, resulting in low efficiency and accuracy in power grid fault monitoring and diagnosis, and making it difficult to achieve real-time monitoring and intelligent early warning of power grid status.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for intelligent identification, classification, and early warning of power grid fault graphics, including:

[0008] Obtain the power grid image to be identified, input the power grid fault image into the intelligent identification model, and obtain the power grid fault image identification result;

[0009] The power fault image recognition results are clustered based on the first clustering algorithm to obtain the power fault image classification results;

[0010] Early warning reports are generated based on the power fault graphic classification results to achieve classified early warning.

[0011] As a preferred embodiment of the intelligent identification and classification early warning method for power grid fault graphics described in this invention, the method of inputting the power grid graphics into the intelligent identification model for power grid fault graphics includes:

[0012] Extract the corresponding graphic features from the power graph to be identified;

[0013] The first calculation is performed on the similarity matrix between the graphic features and the target features;

[0014] Applying the first activation function to the similarity matrix yields the attention weight matrix;

[0015] The graphic feature map is obtained based on the attention weight matrix and graphic features;

[0016] By comparing the graphic feature map and the fault graphic feature map, the power fault graphic recognition result is obtained.

[0017] The beneficial effects of this preferred technical solution are that, through deep feature extraction and attention mechanism, it significantly improves the accuracy and efficiency of power fault image recognition, while enhancing the robustness and interpretability of the system.

[0018] As a preferred embodiment of the intelligent identification, classification, and early warning method for power grid fault graphics described in this invention, the method for obtaining graphic feature maps based on attention weight matrices and graphic features includes:

[0019] Obtain the weight matrix and eigenvalue matrix; the product of the weight matrix and eigenvalue matrix yields the graphic feature matrix.

[0020] Based on the graphic feature matrix, the feature points in the graphic feature matrix are mapped onto the feature map, and the feature map is preprocessed to obtain the graphic feature map.

[0021] The beneficial effects of this preferred technical solution are that by multiplying the weights and eigenvalue matrices to generate a graphic feature matrix and preprocessing the feature map, the feature representation capability is effectively enhanced, and the recognition accuracy and robustness are improved.

[0022] As a preferred embodiment of the intelligent identification, classification, and early warning method for power grid fault graphics described in this invention, the comparison of the graphic feature map and the fault graphic feature map includes:

[0023] Based on the graphical feature map, the fault feature map is compared and learned to obtain the node's learned representation.

[0024] Retrieve the representation of the fault image and compare it with the representation learned by the node;

[0025] The learned representations of the nodes are judged, and the fault category is obtained based on the judgment result.

[0026] The beneficial effects of this preferred technical solution are that by comparing and learning with fault feature maps for accurate matching, combined with intelligent judgment, the fault category can be accurately identified, thereby improving diagnostic efficiency and accuracy.

[0027] As a preferred embodiment of the intelligent identification, classification, and early warning method for power grid fault graphics described in this invention, the judgment of the representation learned by the nodes includes:

[0028] The first judgment is made on the representation learned by the node. If the representation learned by the node shows a transmission line fault, then the power graph to be identified has a transmission line fault.

[0029] A second judgment is made on the representation learned by the node. If the representation learned by the node includes a device fault representation, then the power graph to be identified has a device fault.

[0030] A third judgment is made on the representation learned by the node. If the representation learned by the node shows a bus fault, then the power graph to be identified has a bus fault.

[0031] As a preferred embodiment of the intelligent identification and classification early warning method for power grid fault graphics described in this invention, the clustering process of the power fault graphics identification results based on the first clustering algorithm includes:

[0032] The value of k is determined based on cross-validation, and k points are created as centroids.

[0033] Each power fault graphic identification result is recorded as a data point, and the distance between each data point and k centroids is calculated;

[0034] Each data point is assigned to the cluster closest to the centroid;

[0035] Data points in the cluster whose distance from the centroid of the cluster exceeds the first threshold are removed to obtain the first data;

[0036] Use the first data as the dataset, and set the k value according to the characteristics of the dataset.

[0037] As a preferred embodiment of the power grid fault graphic intelligent recognition and classification early warning method of the present invention, it further includes:

[0038] Select a data point from the first dataset as the initial centroid;

[0039] When the initial centroid data is less than k, the dataset of the first data is obtained;

[0040] The distance from the existing initial centroid to each data point in the dataset is calculated based on the objective function.

[0041] The data point corresponding to the maximum distance of each data point from the existing initial centroid is used as the next initial centroid;

[0042] k initial centroids are obtained sequentially, and the dataset is clustered based on the k initial centroids to obtain several clustering result sets. Each clustering result set is used as a power fault image classification.

[0043] Early warning reports are generated based on the graphical classification results of power faults.

[0044] Secondly, the present invention provides a power grid fault graphic intelligent identification and classification early warning system, including:

[0045] The image recognition module is used to acquire the power grid image to be recognized, input the power grid fault image intelligent recognition model, and obtain the power grid fault image recognition result.

[0046] The clustering module is used to perform clustering processing on the power fault image recognition results based on the first clustering algorithm to obtain the power fault image classification results;

[0047] The early warning module is used to generate early warning reports based on the power fault graphic classification results, thereby achieving classified early warning.

[0048] Thirdly, the present invention provides an electronic device, comprising:

[0049] Memory and processor;

[0050] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for intelligent identification and classification early warning of power grid fault graphics are implemented.

[0051] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the power grid fault graphic intelligent identification and classification early warning method.

[0052] Compared with existing technologies, the beneficial effects of this invention are as follows: The power grid fault image to be identified is input into the intelligent power grid fault image recognition model to obtain the power grid fault image recognition result. Based on the k-means algorithm, the power grid fault image recognition result is clustered to obtain the power grid fault image classification result. This allows for fault identification and classification of the power grid fault images, improving the efficiency and accuracy of power grid fault monitoring and diagnosis, and enabling real-time monitoring of the power grid status. In terms of fault early warning, the system can determine whether there are abnormalities in the equipment through real-time monitoring and analysis of equipment operating data, and promptly send early warning information to relevant personnel. When the power grid fault diagnosis and early warning system predicts a high probability of a power grid fault, it will issue timely warnings through various alarm methods, enabling early detection and resolution of problems. This not only improves the response speed and processing efficiency of power grid faults but also significantly reduces the impact of faults on production and operation. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of the overall process logic of the power grid fault graphic intelligent identification and classification early warning method according to an embodiment of the present invention. Detailed Implementation

[0055] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0056] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for intelligent identification, classification, and early warning of power grid fault graphics is provided, comprising:

[0057] S100: Obtain the power grid image to be identified, input the power grid fault image into the intelligent identification model, and obtain the power grid fault image identification result;

[0058] S200: Based on the first clustering algorithm, the power fault image recognition results are clustered to obtain the power fault image classification results;

[0059] S300: Generates early warning reports based on power fault graphic classification results to achieve classified early warning.

[0060] Specifically, the early warning report includes the fault type, location, and confidence level;

[0061] It should be noted that the intelligent identification model accurately identifies power fault patterns and uses clustering algorithms for efficient classification, generating structured early warning reports, which significantly improves the efficiency, accuracy, and real-time performance of fault diagnosis, and helps to make power grid operation and maintenance more intelligent.

[0062] In this embodiment of the application, step S100 includes the following sub-steps A1-A5;

[0063] In A1: Extract the corresponding graphic features from the power graph to be identified;

[0064] In A2: The first calculation is performed on the similarity matrix between graphic features and target features;

[0065] In A3: The first activation function is applied to the similarity matrix to obtain the attention weight matrix;

[0066] In A4: The graphic feature map is obtained based on the attention weight matrix and graphic features;

[0067] In A5: The graphic feature map and the fault graphic feature map are compared to obtain the power fault graphic recognition result.

[0068] Specifically, before obtaining the power fault image recognition results, the system includes denoising, contrast adjustment and sharpening of the power grid image. The power grid image is subject to various noise interferences during the acquisition and transmission process, such as camera sensor noise, transmission noise and environmental noise.

[0069] In one alternative embodiment, the denoising method can be spatial domain denoising, specifically including mean filtering, which replaces the value of each pixel in the image with the average value of its surrounding pixels to remove noise through smoothing; median filtering, which replaces the value of each pixel in the image with the median value of its surrounding pixels, which is particularly effective for removing salt-and-pepper noise while better preserving image edges; and Gaussian filtering, which uses a Gaussian kernel to perform a weighted average on the image, which can both smooth the image and preserve image details to some extent.

[0070] In one optional embodiment, the denoising method can be frequency domain denoising, specifically including Fourier transform to convert the image from the spatial domain to the frequency domain, removing high-frequency noise through a low-pass filter, and then converting the image back to the spatial domain through inverse Fourier transform; it also includes decomposing the image into sub-bands of different scales and directions, and achieving denoising by removing noise sub-bands;

[0071] In the embodiments of this application, all denoising methods can be used for denoising in this application. The purpose of denoising is to remove these unnecessary noises and improve the image clarity and signal-to-noise ratio.

[0072] Specifically, contrast adjustment aims to improve the visual effect of an image, making details in the image clearer and more discernible;

[0073] In this embodiment, contrast adjustment can be histogram equalization, which adjusts the histogram distribution of the image to make the pixel value distribution of the image more uniform, thereby improving the contrast of the image. This is suitable for images where both the background and foreground are dark or both are bright. Contrast adjustment can also be gamma correction, which changes the brightness and contrast of the image by adjusting the gamma value of the image. When the gamma value is less than 1, the image becomes brighter and the contrast is enhanced; when the gamma value is greater than 1, the image becomes darker and the contrast is reduced.

[0074] In one alternative embodiment, sharpening can be a high-pass filter, which, in contrast to a low-pass filter, can preserve high-frequency components (i.e., edges and details) in an image while removing low-frequency components (i.e., smooth areas). Therefore, high-pass filtering is often used for image sharpening.

[0075] In one alternative embodiment, sharpening can be gradient sharpening, which enhances edges by calculating the gradient value of each pixel in the image and adding it to the original pixel value. Gradient sharpening can highlight edge information in the image, making the image look clearer.

[0076] In one alternative embodiment, sharpening can be achieved through bilateral filtering, which smooths the image while preserving edge sharpness. Therefore, by adjusting the parameters of the bilateral filter (such as the Gaussian kernel size in the spatial and pixel domains), image sharpening can be achieved to some extent. However, bilateral filtering is primarily used for noise reduction and edge preservation, rather than direct sharpening.

[0077] In this embodiment, sharpening is mainly used to enhance the edges and details of the image, making the image appear clearer. Since power grid graphics often contain a large amount of lines and edge information, sharpening is of great significance for improving the recognition accuracy of power grid graphics.

[0078] It should be noted that by denoising, contrast adjustment, and sharpening, the quality and clarity of the power grid image are effectively improved. The denoising method removes interference noise and improves the image signal-to-noise ratio; the contrast adjustment enhances image details, making them easier to identify; and the sharpening process highlights edge and detail information, improving the recognition accuracy and precision of the power grid image. Overall, this significantly enhances the system's ability to recognize complex power grid images and its robustness, providing high-quality input data for subsequent fault identification and classification, thereby improving the efficiency and reliability of power grid fault diagnosis.

[0079] In an optional embodiment, the first calculation can be cosine similarity, which measures the similarity by calculating the cosine of the angle between two vectors. The similarity matrix is ​​obtained by the ratio of the dot product of the two vectors to the magnitude of the two vectors. The value of cosine similarity ranges from -1 to 1, and the closer the value is to 1, the higher the similarity.

[0080] In another alternative embodiment, the first calculation can also be the Euclidean distance, which calculates the straight-line distance between two vectors in multidimensional space, the square of the difference between the two vectors, the sum of the squares, and the square root of the sum to obtain the similarity matrix.

[0081] In this embodiment of the application, the first calculation includes: let Q be a query matrix, representing the feature matrix of the affected elements, with a dimension of N*d, where N represents the number of elements in the query matrix and d represents the feature dimension of each element; K represents the indicator feature matrix of the influencing factors, with a dimension of M*d. b M represents the number of factors affecting power grid faults, and d b Characteristic dimensions representing factors influencing power grid faults;

[0082] Calculate the similarity matrix S between graphic features and target features:

[0083]

[0084] Among them, K T The similarity matrix S represents the transpose of the indicator feature matrix of the influencing factors. It has N rows and M columns, representing N affected elements and M power grid fault influencing factors. Each row represents the similarity between an affected element and all power grid fault influencing factors. ij Let i represent the similarity between the affected element i and the power grid fault influencing factor j, where 1≤i≤N and 1≤j≤M.

[0085] Factors affecting power grid failures include natural factors, equipment factors, human factors, and factors related to power grid structure and management.

[0086] Specifically, natural factors include meteorological conditions. Changes in meteorological conditions such as temperature, humidity, wind speed, and precipitation can all affect power grid equipment. For example, strong winds may cause power transmission towers to tilt, lines to loosen, or break; heavy rain and lightning may cause line short circuits, equipment damage, and other faults. Icing and snow disasters are also important factors affecting the safe operation of the power grid. In cold seasons, lines may become heavier due to snow and ice accumulation, leading to tower collapses or line breaks. Animal damage is also a concern. Birds, squirrels, and other animals nesting and moving on transmission line towers can easily cause line short circuits, insulator flashover, and other faults. Birds flying or fighting between conductors, as well as hunting birds near conductors, can all cause line grounding or short circuit accidents. Environmental changes are also a factor. Transmission lines are mostly located in outdoor environments, and dense tree growth may cause trees to grow too tall or fall onto the lines, causing line short circuits, equipment damage, and other faults. In addition, lines in chemical industrial zones are also prone to flashover discharge due to pollution.

[0087] Equipment factors include equipment aging, as power grid distribution equipment needs to operate continuously for extended periods, and the continuous transmission of high current by equipment leads to aging. Equipment aging can cause deformation of the distribution structure, restricting power distribution operation and failing to meet power supply requirements. Equipment defects, such as aging insulators or broken wires, may exist during the manufacturing, installation, or operation of the equipment, and can all lead to power grid failures. Insufficient equipment maintenance can also cause failures. For example, failure to replace aging equipment in a timely manner or failure to clean dirt from the lines in a timely manner can affect the normal operation of the equipment.

[0088] Human factors include operational errors. If maintenance personnel fail to identify the cause of a fault during troubleshooting, they may rely on subjective judgment to diagnose the problem and arbitrarily operate switches, which may worsen the issue. Furthermore, some maintenance personnel habitually attribute faults to switch failure, neglecting to check the switch's current limits. This can lead to automatic tripping during switch replacement, making it difficult to restore power in the short term. External damage is also a factor. Damage caused by mechanical construction, vehicle collisions, and other human factors can trigger power outages and electric shocks. For example, roadside power lines are vulnerable to vehicle collisions, and a crane boom working under power lines can touch the lines, causing phase-to-phase short circuits or even line breaks.

[0089] Power grid structure and management factors include weak local power grid structures, fewer local power supply points, lower voltage levels, and the fact that local power grids and the main grid are mostly connected in single-line series, resulting in poor electrical stability and weak connectivity. Once the system experiences disturbances, local power grid disconnection is highly likely, causing a chain reaction and evolving into a system accident. Other factors include poor management, with weaknesses in the planning, construction, design, maintenance, and operation of the power grid. For example, the lack of a long-term management mechanism and the failure to conduct timely equipment inspection and maintenance can all lead to power grid failures.

[0090] It should be noted that by efficiently generating a similarity matrix through matrix-based calculation, the relationship between the affected elements and the fault factors is quantified, providing a solid foundation for subsequent accurate identification and classification, and significantly improving the system's processing efficiency and diagnostic accuracy.

[0091] In an optional embodiment, the first activation function can be the Sigmoid function, which is applied to each element in the similarity matrix to map it to the (0,1) interval, thereby obtaining the normalized weight of each element.

[0092] In another alternative embodiment, the first activation function may also be the ReLU function, which applies the ReLU function to each element in the similarity matrix, setting negative values ​​to 0 and retaining positive values;

[0093] In this embodiment, the first activation function includes the Softmax function;

[0094] Specifically, the attention weight matrix W is obtained by applying the Softmax function to the similarity matrix S, and is expressed as:

[0095]

[0096] Where, x c ∈S represents an element in the similarity matrix S, c represents the element index in the similarity matrix S, 1≤c≤M×N;

[0097] It should be noted that the Softmax function normalizes the similarity matrix into a weight matrix, highlighting important features and improving recognition accuracy and stability.

[0098] In this embodiment of the application, after completing steps A1-A5, step S100 also includes steps A6-A7;

[0099] In A6: Obtain the weight matrix and eigenvalue matrix; the product of the weight matrix and eigenvalue matrix yields the graphic feature matrix.

[0100] In A7: Based on the graphic feature matrix, the feature points in the graphic feature matrix are mapped onto the feature map, and the feature map is preprocessed to obtain the graphic feature map.

[0101] Specifically, obtain the weight matrix W and the eigenvalue matrix V, where the eigenvalue matrix V represents the numerical matrix corresponding to the factors affecting power grid faults;

[0102] The graphic feature matrix is ​​calculated by multiplying the weight matrix W and the eigenvalue matrix V.

[0103] Based on the graphic feature matrix, the feature points in the graphic feature matrix are mapped onto the feature map. Usually, the positions of the feature points are marked as bright spots or high-value regions on the feature map. The feature map is then enhanced by Gaussian filtering. The feature map is then preprocessed to obtain the graphic feature map.

[0104] The first preprocessing step is visualization processing;

[0105] In one alternative embodiment, the visualization process can be a heatmap, which represents the feature intensity in the feature map with the shades of color. Typically, red is used to represent high-value areas and blue to represent low-value areas, which is suitable for scenarios that require rapid identification of feature intensity distribution.

[0106] In an alternative embodiment, the visualization process can be overlaid, displaying the feature map over the original image, typically using a semi-transparent method to highlight the high-value areas of the feature map on the original image.

[0107] In this embodiment of the application, visualization includes color encoding and grayscale encoding;

[0108] Specifically, grayscale encoding maps each feature value in the feature map to a grayscale value in the grayscale image. It assigns the grayscale value of each feature point to the corresponding pixel to generate a grayscale image. The higher the grayscale value, the brighter the image; the lower the grayscale value, the darker the image.

[0109] Color encoding involves mapping each feature value in the feature map to a color value in the color space, assigning the color value of each feature point to the corresponding pixel, and generating a color image.

[0110] It should be noted that Gaussian filtering is used to enhance the feature map, and color or grayscale encoding is used for visualization to highlight key features and improve recognition accuracy and interpretability.

[0111] In this embodiment of the application, after completing steps A6-A7, step S100 also includes the following steps A8-A13;

[0112] In A8: The fault feature map is compared and learned based on the graph feature map to obtain the representation learned by the node;

[0113] In A9: The representation of the fault image is retrieved and compared with the representation learned by the nodes;

[0114] In A10: the representations learned by the nodes are judged, and the fault category is obtained based on the judgment result;

[0115] In A11: The first judgment is made on the representation learned by the node. If the representation learned by the node shows a transmission line fault, then the power graph to be identified has a transmission line fault.

[0116] In A12: A second judgment is made on the representation learned by the node. If the representation learned by the node shows a device fault, then the power graph to be identified has a device fault.

[0117] In A13: A third judgment is made on the representation learned by the node. If the representation learned by the node shows a bus fault representation, then the power graph to be identified has a bus fault.

[0118] Specifically, based on the graphical feature map, the fault graphic feature map is compared and learned to obtain the representation learned by the node; the representation of the fault graphic is retrieved and compared and analyzed with the representation learned by the node.

[0119] Determine whether the representation learned by the node contains a transmission line fault representation. If so, the power graph to be identified contains a transmission line fault. Transmission line faults include open circuit faults, single-phase ground faults, two-phase ground faults, phase-to-phase faults, and overload faults.

[0120] Determine whether the representation learned by the node contains a device fault representation. If so, the power graph to be identified has a device fault, which includes faults in transformers, circuit breakers, and distribution cabinets.

[0121] Determine whether the representation learned by the node indicates a bus fault. If so, the power graph to be identified has a bus fault.

[0122] It should be noted that by using comparative learning to accurately identify fault characteristics, it is possible to quickly determine the specific fault types of transmission lines, equipment, busbars, etc., significantly improving the accuracy and efficiency of fault diagnosis and providing strong support for power grid operation and maintenance.

[0123] In this embodiment of the application, step S200 includes the following sub-steps B1-B5;

[0124] In B1: The value of k is determined based on cross-validation, and k points are created as centroids;

[0125] In B2: Each power fault graphic identification result is recorded as a data point, and the distance between each data point and k centroids is calculated;

[0126] In B3: Each data point is assigned to the cluster closest to the centroid;

[0127] In B4: Data points whose distance from the centroid of the cluster exceeds the first threshold are removed to obtain the first data.

[0128] In B5: The first data is used as the dataset, and the k value is set according to the characteristics of the dataset;

[0129] In B6: Select the data points in the first data set as the initial centroid;

[0130] In B7: When the initial centroid data is less than k, the dataset of the first data is obtained;

[0131] In B8: Calculate the distance of each data point in the dataset from the existing initial centroid based on the objective function;

[0132] In B9: The data point corresponding to the maximum distance of each data point from the existing initial centroid is taken as the next initial centroid;

[0133] In B10: k initial centroids are obtained sequentially, and the dataset is clustered based on the k initial centroids to obtain several clustering result sets. Each clustering result set is used as a power fault graph classification.

[0134] In B11: Generate early warning reports based on the graphical classification results of power faults.

[0135] In an optional embodiment, the first clustering algorithm can be hierarchical clustering, which organizes data points by constructing a nested hierarchical clustering structure, calculates the similarity between data points using a distance metric, selects an appropriate cluster merging strategy, constructs a hierarchical clustering tree, and determines the final number of clusters by truncating the tree.

[0136] In another alternative embodiment, the first clustering algorithm can also be a density-based clustering algorithm, which can discover clusters of arbitrary shapes and identify noise points. It selects appropriate neighborhood radius and minimum number of points parameters, traverses data points, and for each unvisited point, checks the number of points within the neighborhood radius. If the minimum number of points condition is met, it is marked as a core point and the cluster is expanded. Points that do not belong to any cluster are marked as noise points.

[0137] In this embodiment of the application, the first clustering algorithm includes the k-means algorithm;

[0138] Specifically, the k-means algorithm is used to cluster the power fault image recognition results, cross-validation is used to predict the preprocessing effect corresponding to different k values, the k value is determined, and k points are created as centroids.

[0139] Each power fault image recognition result is recorded as a data point. The distance between each data point and k centroids is calculated. Each data point is assigned to the cluster closest to the centroid. Data points in the cluster whose distance from the centroid of the corresponding cluster exceeds a preset threshold are removed to obtain the preprocessed data.

[0140] The preprocessed data is treated as a dataset. The k value is set according to the size and shape of the dataset. A data point in the preprocessed data is randomly selected as the initial centroid.

[0141] When the number of initial centroids is less than k, let Y l Let μ be the preprocessed data set, l = 1, 2…L, where L represents the number of preprocessed data sets. z Let z be the initial centroid, and z = 1, 2, ..., k;

[0142] Based on the objective function Calculate the distance D(Y) of each data point in the dataset from the existing initial centroid;

[0143] Take the data point corresponding to the maximum value in D(Y) as the next initial centroid; obtain k initial centroids in sequence, and cluster the dataset based on the k initial centroids to obtain several clustering result sets. Take each clustering result set as a power fault graphic classification.

[0144] It should be noted that an improved k-means algorithm is used to cluster power fault images. The k value is dynamically adjusted through cross-validation to ensure optimal clustering results. At the same time, outlier data points are removed to further improve the accuracy and robustness of clustering. The final clustering results can accurately classify power fault images, providing strong support for rapid diagnosis and early warning of power grid faults, and significantly improving the efficiency and reliability of power grid operation and maintenance.

[0145] The above is an illustrative scheme of the intelligent identification and classification early warning method for power grid fault graphics in this embodiment. It should be noted that the technical solution of this intelligent identification and classification early warning system for power grid fault graphics belongs to the same concept as the technical solution of the aforementioned intelligent identification and classification early warning method for power grid fault graphics. Details not described in detail in the technical solution of the intelligent identification and classification early warning system for power grid fault graphics in this embodiment can be found in the description of the technical solution of the aforementioned intelligent identification and classification early warning method for power grid fault graphics.

[0146] The power grid fault graphic intelligent recognition and classification early warning system in this embodiment includes:

[0147] The image recognition module is used to acquire the power grid image to be recognized, input the power grid fault image intelligent recognition model, and obtain the power grid fault image recognition result.

[0148] The clustering module is used to perform clustering processing on the power fault image recognition results based on the first clustering algorithm to obtain the power fault image classification results;

[0149] The early warning module is used to generate early warning reports based on the power fault graphic classification results, thereby achieving classified early warning.

[0150] This embodiment also provides an electronic device suitable for intelligent graphical recognition and classification early warning of power grid faults, including:

[0151] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the method for intelligent graphical identification and classification early warning of power grid faults as proposed in the above embodiments.

[0152] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the method for intelligent identification and classification early warning of power grid fault graphics as proposed in the above embodiments.

[0153] The storage medium proposed in this embodiment and the method for realizing intelligent graphic recognition and classification early warning of power grid faults proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0154] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0155] Example 2, referring to Tables 1-2, differs from the first example and provides a verification test of the method for intelligent identification and classification early warning of power grid fault graphics, to verify and explain the technical effects used in this method.

[0156] Using traditional methods as a control experiment, traditional methods mainly rely on manually designed features to describe power graphs, including line density, number of nodes, and topology index. In terms of similarity calculation, Euclidean distance is used to measure the difference between features. Support Vector Machine (SVM) is used as a classifier to directly generate early warning reports.

[0157] This application's method employs a deep learning model to extract deep features from power grid images, combined with an attention mechanism to further enhance the expressive power of the features. It calculates the cosine similarity matrix between the image features and the target features (fault template). Based on Softmax attention weights and fully connected layers, it uses a K-means clustering algorithm to optimize the classification results, generating a structured early warning report. The report includes detailed information such as fault type, fault location, and confidence level.

[0158] The experimental results of the traditional method and the method of this application are compared as shown in Table 1;

[0159] Table 1 Comparison of experimental results

[0160] index This application method Traditional methods Improvement rate Recognition accuracy (%) 98.2 89.5 +9.7％ False alarm rate (%) 1.5 7.8 -80.8％ Missed report rate (%) 0.3 2.7 -88.9％ Average recognition time (seconds / image) 0.85 2.1 -59.5％ Cluster purity 0.92 - -

[0161] Cluster purity measures the consistency between clustering results and true labels. Traditional methods do not have a clustering module and therefore cannot calculate it.

[0162] The recognition time includes the entire process of feature extraction, classification, and clustering;

[0163] The fault identification performance of the method in this application compared with traditional methods is shown in Table 2;

[0164] Table 2 Typical Fault Identification Results

[0165]

[0166] As shown in Tables 1 and 2, the method of this application significantly improves the recognition accuracy of complex fault images through deep feature extraction and attention mechanism. The clustering module further optimizes the classification results, with a clustering purity of 0.92. The average recognition time is shortened to 0.85 seconds / image, meeting the real-time requirements. The generated early warning report includes fault type, location, and confidence level, assisting maintenance personnel in making quick decisions. Under noise interference (such as blurred images or incorrect annotations), the method of this application still maintains high stability, and the false alarm rate is reduced by 80.8%.

[0167] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent identification, classification, and early warning of power grid faults using graphical representations, characterized in that: include: Obtain the power grid image to be identified, input the power grid fault image into the intelligent identification model, and obtain the power grid fault image identification result; The power fault image recognition results are clustered based on the first clustering algorithm to obtain the power fault image classification results; Early warning reports are generated based on the power fault graphic classification results to achieve classified early warning.

2. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 1, characterized in that, Inputting the power grid graphics into the power grid fault graphic intelligent recognition model includes: Extract the corresponding graphic features from the power graph to be identified; The first calculation is performed on the similarity matrix between the graphic features and the target features; Applying the first activation function to the similarity matrix yields the attention weight matrix; The graphic feature map is obtained based on the attention weight matrix and graphic features; By comparing the graphic feature map and the fault graphic feature map, the power fault graphic recognition result is obtained.

3. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 2, characterized in that, The graphical feature map obtained based on the attention weight matrix and graphical features includes: Obtain the weight matrix and eigenvalue matrix; the product of the weight matrix and eigenvalue matrix yields the graphic feature matrix. Based on the graphic feature matrix, the feature points in the graphic feature matrix are mapped onto the feature map, and the feature map is preprocessed to obtain the graphic feature map.

4. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 2 or 3, characterized in that, The comparison between the graphic feature map and the fault graphic feature map includes: Based on the graphical feature map, the fault feature map is compared and learned to obtain the node's learned representation. Retrieve the representation of the fault image and compare it with the representation learned by the node; The learned representations of the nodes are judged, and the fault category is obtained based on the judgment result.

5. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 4, characterized in that, Judging the representations learned by the nodes includes: The first judgment is made on the representation learned by the node. If the representation learned by the node shows a transmission line fault, then the power graph to be identified has a transmission line fault. A second judgment is made on the representation learned by the node. If the representation learned by the node includes a device fault representation, then the power graph to be identified has a device fault. A third judgment is made on the representation learned by the node. If the representation learned by the node shows a bus fault, then the power graph to be identified has a bus fault.

6. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 5, characterized in that, Clustering the power fault image recognition results based on the first clustering algorithm includes: The value of k is determined based on cross-validation, and k points are created as centroids. Each power fault graphic identification result is recorded as a data point, and the distance between each data point and k centroids is calculated; Each data point is assigned to the cluster closest to the centroid; Data points in the cluster whose distance from the centroid of the cluster exceeds the first threshold are removed to obtain the first data; Use the first data as the dataset, and set the k value according to the characteristics of the dataset.

7. The method for intelligent identification, classification, and early warning of power grid fault graphics as described in claim 6, characterized in that, Also includes: Select a data point from the first dataset as the initial centroid; When the initial centroid data is less than k, the dataset of the first data is obtained; The distance from the existing initial centroid to each data point in the dataset is calculated based on the objective function. The data point corresponding to the maximum distance of each data point from the existing initial centroid is used as the next initial centroid; k initial centroids are obtained sequentially, and the dataset is clustered based on the k initial centroids to obtain several clustering result sets. Each clustering result set is used as a power fault image classification. Early warning reports are generated based on the graphical classification results of power faults.

8. A power grid fault graphic intelligent identification and classification early warning system, using the method described in any one of claims 1-7, characterized in that, include: The image recognition module is used to acquire the power grid image to be recognized, input the power grid fault image intelligent recognition model, and obtain the power grid fault image recognition result. The clustering module is used to perform clustering processing on the power fault image recognition results based on the first clustering algorithm to obtain the power fault image classification results; The early warning module is used to generate early warning reports based on the power fault graphic classification results, thereby achieving classified early warning.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images