A power distribution anomaly identification method and device, electronic equipment and medium

Abstract

The application relates to the technical field of power distribution equipment state monitoring, and specifically provides a power distribution anomaly identification method and device, electronic equipment and a medium, which comprises the following steps: acquiring first visual data of a power distribution equipment monitoring area; performing hierarchical processing on the spatially registered first visual data through guided filtering to obtain a base layer representing overall brightness distribution and equipment structure information of the power distribution equipment monitoring area and an enhancement layer containing texture features and edge detail information; after contrast self-adaptive adjustment of the base layer, weighting fusion is performed on the base layer and the enhancement layer to generate second visual data; a time sequence change feature vector of equipment appearance is extracted based on the time sequence continuous second visual data; and matching analysis is performed on the time sequence change feature vector to determine an anomaly identification result. Through guided filtering hierarchical processing combined with contrast self-adaptive adjustment, the second visual data effectively supports subsequent time sequence change feature extraction and anomaly matching analysis, and the accuracy of power distribution equipment anomaly identification is significantly improved.

Description

Technical Field

[0001] This application relates to the field of power distribution equipment condition monitoring technology, and in particular to a power distribution anomaly identification method, device, electronic equipment and medium. Background Technology

[0002] In the operation and maintenance of power distribution systems, visual perception-based equipment anomaly identification technology is a core supporting means to replace manual inspections, improve operation and maintenance efficiency, and increase fault response speed. It has been widely used in online monitoring and status screening of various power distribution equipment such as switchgear, distribution transformers, insulators, and disconnector mechanisms. Existing power distribution visual monitoring solutions typically input the collected monitoring images directly into deep learning models to complete fault type determination and status identification. Although this reduces labor costs to some extent, it still has significant technical shortcomings in actual engineering deployments.

[0003] However, the lighting conditions in power distribution monitoring scenarios are complex and variable. Outdoor equipment is prone to local overexposure areas due to direct sunlight and specular reflection, while equipment inside cabinets and the backlight surfaces of equipment often have large dark areas. Existing image enhancement solutions mostly use globally uniform brightness adjustment or color space transformation with fixed parameters, which makes it difficult to simultaneously suppress highlights and enhance details in dark areas. This can easily lead to the loss of surface structure in overexposed areas and the submersion of texture defects in dark areas, making it impossible for subsequent recognition models to fully extract key abnormal features such as equipment edges and textures, ultimately resulting in missed or false positives. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a power distribution anomaly identification method, device, electronic equipment, and medium. By guiding filtering and layering combined with adaptive contrast adjustment, the final generated second visual data effectively supports subsequent temporal change feature extraction and anomaly matching analysis, significantly improving the accuracy, positioning precision, and scene adaptability of power distribution equipment anomaly identification.

[0005] To achieve the objectives of this application, the following technical solution is provided: In a first aspect, this application provides a method for identifying power distribution anomalies, including: Acquire first-view data of the power distribution equipment monitoring area; Spatial registration is performed on the first visual data to obtain spatially registered first visual data; After normalizing the spatially registered first visual data, guided filtering is used to perform layered processing on the spatially registered first visual data to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that contains texture features and edge detail information. The contrast of the base layer is adaptively adjusted, and the adjusted base layer is weighted and fused with the enhancement layer to generate second visual data. The temporal change feature vector of the device appearance is extracted based on the second visual data with continuous temporal sequence; the temporal change feature vector is matched and analyzed in combination with the preset abnormal feature library to determine the abnormal identification result.

[0006] A further improvement of this application is that the spatial registration of the first visual data to obtain the spatially registered first visual data includes: establishing a first planar coordinate system on the actual physical plane of the power distribution equipment monitoring area, using a preset corner point of the first planar coordinate system as the origin and two mutually perpendicular boundary directions as coordinate axis directions to determine the actual coordinate information; converting the first visual data into a single-frame image, and establishing a second planar coordinate system on the single-frame image; for each single-frame image, using the geometric center of the single-frame image as the origin and the horizontal and vertical directions of the single-frame image as coordinate axis directions to determine the image coordinate information; constructing a correction matrix based on the angular offset relationship between the actual coordinate information and the image coordinate information, and performing a geometric transformation on the single-frame image based on the correction matrix to obtain the spatially registered first visual data; wherein, the remapping of the coordinates of each pixel in the single-frame image based on the correction matrix specifically includes: ; In the formula, For the correction matrix; These are the cosine transform coefficients. ; These are the coefficients of the sinusoidal transform. ; This is the angular offset. These are the coordinates of the geometric center of a single frame image.

[0007] A further improvement in this application is that the step of performing layered processing on the first visual data after spatial registration through guided filtering to obtain a base layer characterizing the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer containing texture features and edge detail information, includes: using normalized visual data as a guide map to extract the base layer that can characterize the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area; and performing pixel-level difference operations on the normalized visual data and the base layer to separate the enhancement layer containing the texture features and edge detail information of the power distribution equipment.

[0008] A further improvement in this application is that the step of adaptively adjusting the contrast of the base layer and weightedly fusing the adjusted base layer with the enhancement layer to generate second visual data includes: statistically analyzing the pixel proportion of high-brightness regions in the grayscale histogram of the base layer; if the pixel proportion of high-brightness regions is greater than a preset highlight threshold, non-linear grayscale compression is performed on the high-brightness region pixels of the base layer using a first coefficient to obtain a compressed base layer; statistically analyzing the pixel proportion of low-brightness regions in the grayscale histogram of the base layer, and grayscale stretching is performed on the low-brightness regions in the compressed base layer using a second coefficient to obtain a restored base layer; and weightedly fusing the restored base layer with the enhancement layer to generate second visual data; wherein the first coefficient and the second coefficient are updated in real time with the grayscale histogram of each frame of the image; the first coefficient is less than 1, and the second coefficient is greater than 1.

[0009] A further improvement of this application is that the first coefficient and the second coefficient are updated in real time with the gray-level histogram of each frame of the image, including: for each frame of the image, statistically analyzing the gray-level histogram of the image's base layer to obtain a cumulative distribution function; determining the dynamic range span between the two cutoff points of the gray-level histogram based on the cumulative distribution function, and obtaining the first coefficient and the second coefficient; wherein, the first coefficient is the difference between the high-end threshold and the low-end threshold: ; ; ; ; In the formula, For the low-end threshold, For the high-end threshold, For cumulative distribution values, For the truncation ratio, As the first coefficient, As the second coefficient, It is a very small positive number.

[0010] A further improvement in this application is that the extraction of the temporal change feature vector of the device appearance based on the temporally continuous second visual data includes: extracting the temporally change feature vector of the device appearance based on the temporally continuous multi-frame second visual data using an anomaly recognition model; the anomaly recognition model is a hybrid architecture based on a cascaded 3D convolutional neural network and ConvLSTM, wherein a three-dimensional convolutional layer is used to extract short-term spatiotemporal features layer by layer, the extracted feature map sequence is input into the ConvLSTM layer, the hidden state is updated frame by frame, and the modeling of long-term temporal dependencies is completed; finally, a comprehensive spatiotemporal feature vector is obtained by global average pooling.

[0011] A further improvement of this application is that the step of performing matching analysis on the temporal change feature vector in conjunction with a preset anomaly feature library to determine the anomaly identification result includes: calculating the cosine similarity between the comprehensive spatiotemporal feature vector and the prototype feature vector of each anomaly category in the anomaly feature library one by one, and filtering and retaining candidate anomaly types with a cosine similarity greater than or equal to the corresponding matching threshold; weighting and fusing the cosine similarity of the candidate anomaly type with the probability score output by the anomaly identification model to obtain the final confidence level; comparing the maximum value of the final confidence level with the preset confidence threshold, and determining the candidate anomaly type corresponding to the maximum value of the final confidence level as the anomaly identification result if the maximum value of the final confidence level is greater than the preset confidence threshold.

[0012] Secondly, this application provides a power distribution anomaly identification device for implementing the above-mentioned power distribution anomaly identification method, the device comprising: The data acquisition module is used to acquire first-view data of the power distribution equipment monitoring area; The spatial registration module is used to perform spatial registration on the first visual data to obtain the spatially registered first visual data. The data layering module is used to normalize the spatially registered first visual data and then perform layering processing on the spatially registered first visual data through guided filtering to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that contains texture features and edge detail information. An adaptive adjustment module is used to adaptively adjust the contrast of the base layer, and then weightedly fuse the adjusted base layer with the enhancement layer to generate second visual data. The identification and analysis module is used to extract the temporal change feature vector of the device appearance based on the temporally continuous second visual data; and to perform matching analysis on the temporal change feature vector in combination with a preset abnormal feature library to determine the abnormal identification result.

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described power distribution anomaly identification method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions, which are adapted to be loaded and executed by a processor to implement the steps of the above-described power distribution anomaly identification method.

[0015] Compared with the prior art, this application has the following beneficial effects: The distribution anomaly identification method, device, electronic equipment, and medium provided in this application, after acquiring the first visual data of the monitoring area of ​​the distribution equipment, first perform spatial registration on the acquired first visual data, and then perform contrast adaptive adjustment based on layered processing, simultaneously optimizing from two dimensions: spatial reference and image quality, to obtain second visual data; then, extract the temporal change feature vector of the equipment appearance based on the temporally continuous second visual data; and perform matching analysis on the temporal change feature vector in combination with a preset anomaly feature library to determine the anomaly identification result. Based on eliminating the interference of image offset on temporal analysis, the processing method of guided filtering layering combined with contrast adaptive adjustment can dynamically optimize the contrast performance of highlights and shadows according to the actual illumination distribution of the image. While suppressing overexposed areas and restoring equipment structural information, it improves the visibility of details in shadows, fully preserving key features highly correlated with anomaly identification, such as equipment texture and edges. This ensures that the final generated second visual data effectively supports subsequent temporal change feature extraction and anomaly matching analysis, significantly improving the accuracy, positioning precision, and scene adaptability of distribution equipment anomaly identification, and guaranteeing the engineering practical value of the distribution visual monitoring solution. Attached Figure Description

[0016] The accompanying drawings are provided to further understand this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. Figure 1 This is a schematic diagram of an optional power distribution anomaly identification method provided in an embodiment of this application; Figure 2 This is a schematic diagram of an optional structure of the power distribution anomaly identification device provided in the embodiments of this application; Figure 3 This is a structural block diagram of an electronic device for implementing a power distribution anomaly identification method, provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature; in the description of this application, unless otherwise stated, "multiple" means two or more.

[0019] The power distribution anomaly identification method provided in this application can be executed by an electronic device, such as a terminal or a server. The terminal can be a smartphone, tablet, laptop, or other similar device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is understood that this application does not limit the specific entity executing the power distribution anomaly identification method.

[0020] The following is combined Figures 1 to 3 The embodiments shown illustrate the technical solutions of this application: Example 1 An embodiment of this application provides a method for identifying power distribution anomalies, referring to... Figure 1 This includes the following steps S101 to S105: Step S101: Obtain the first visual data of the power distribution equipment monitoring area.

[0021] In this embodiment, an image acquisition unit is used to acquire first visual data of the power distribution equipment monitoring area, and the first visual data is obtained from the image acquisition unit. The first visual data in this embodiment refers to raw image or video frame data directly acquired from the power distribution equipment monitoring area; wherein, the image acquisition unit includes several industrial cameras deployed at key locations in the power distribution room, and key locations in the power distribution room refer to key acquisition points such as incoming line cabinets, outgoing line cabinets, and transformer areas.

[0022] For example, the industrial camera is a fixed-focal-length industrial camera with a resolution of 1920×1080 and a frame rate of 60FPS.

[0023] Step S102: Spatial registration is performed on the first visual data to obtain spatially registered first visual data.

[0024] In this embodiment, the first visual data obtained in step S101 is spatially registered and feature-enhanced via a network switch, and the optimized second visual data is output. The spatial registration and feature enhancement are performed in an image preprocessing unit (preprocessing software) deployed on an edge computing device. This software sequentially executes coordinate correction, filtering, decomposition, and fusion algorithms on each input frame image to obtain a processed image sequence, forming an optimized visual data stream.

[0025] In this embodiment, a first planar coordinate system is established on the actual physical plane of the power distribution equipment monitoring area. The first planar coordinate system is used as a preset corner point as the origin and the two mutually perpendicular boundary directions are used as coordinate axis directions to determine the actual coordinate information. The first visual data is converted into single-frame image information, and a second planar coordinate system is established on the single-frame image information. The geometric center of the single-frame image is used as the origin and the horizontal and vertical directions of the single-frame image are used as coordinate axis directions to determine the image coordinate information.

[0026] Based on the angular offset relationship between the actual coordinate information and the image coordinate information, a correction matrix is ​​constructed. Then, based on the correction matrix, a geometric transformation is performed on the single-frame image information to obtain the first visual data after spatial registration. The angular offset relationship is determined and stored during the initial installation and debugging using a calibration board or markers of known physical dimensions. The geometric transformation refers to remapping the coordinates of each pixel in the image using the correction matrix, thereby eliminating the problem of image-physical space mismatch caused by camera installation position and angular deviations, ensuring that the device position in the image accurately matches the actual position.

[0027] As a feasible implementation method, the geometric transformation is achieved through the following correction matrix: ; In the formula, For the correction matrix; The coefficients are the cosine transform coefficients, which are the projection component coefficients of the rotation transformation onto the horizontal coordinate axis. , This is the angular offset. The coefficients are the sinusoidal transform coefficients, which are the projection component coefficients of the rotational transform onto the vertical coordinate axis. , These are the coordinates of the geometric center of a single frame image.

[0028] Step S103: After normalizing the first visual data after spatial registration, a guided filter is used to extract the base layer that characterizes the overall brightness distribution and structural information of the power distribution equipment monitoring area, and an enhancement layer containing texture features and edge information is obtained.

[0029] In this embodiment of the application, the spatially registered single-frame visual data (first visual data) is normalized, and a base layer is obtained by guided filtering. The base layer represents the overall brightness and structural information of the device area.

[0030] In this embodiment, the spatially registered first visual data is first normalized, linearly mapping the pixel values ​​of each channel of the original 8-bit RGB image from the integer range of 0 to 255 to... The floating-point range is defined. For outdoor scenes with significant overexposure, the highest and lowest grayscale extremes based on preset values ​​are first truncated according to the power distribution area. Invalid extreme value pixels caused by camera vignetting or direct strong light sources are then removed. Normalization is then performed to prevent extreme values ​​from interfering with the stability of subsequent layered processing. This way, images from different camera positions and at different times have a uniform numerical range after normalization, ensuring the robustness of subsequent guided filtering and dynamic range adjustment parameters.

[0031] Subsequently, guided filtering is used to extract the base layer representing the overall brightness and macroscopic structural information of the equipment. The normalized image itself is used as the guide image. Leveraging the edge-preserving properties of guided filtering, while smoothing large-area illumination changes, the edges of the main equipment structures, such as switch cabinet doors, insulator skirts, and disconnector mechanisms, are preserved, preventing blurring of equipment outlines due to smoothing. In this embodiment, the filter window radius is set based on the average scale of the equipment in the monitoring field of view, typically ranging from 15×15 to 21×21 pixels. The regularization parameter is dynamically adjusted according to the scene's lighting complexity: a smaller value is used in indoor power distribution rooms with uniform illumination to retain more structural details, while a larger value is used in outdoor scenes with high light ratios, such as backlighting and sidelighting, to fully smooth the global illumination gradient.

[0032] The final output base layer only carries the overall brightness distribution and macro structure of the power distribution equipment monitoring area, including information such as the lighting gradient, large area of ​​light and dark distribution, and the outline of the main body of the equipment, while removing high-frequency information such as instrument scales, equipment surface textures, and defect details.

[0033] After extracting the base layer, the base layer is subtracted pixel-by-pixel from the normalized visual data to separate the enhancement layer, which contains texture and edge information. This enhancement layer carries all high-frequency detail information, including texture features directly related to equipment anomalies such as dirt texture on the insulator surface, cracks in the equipment casing, oxidation and discoloration of the terminals, discharge erosion marks, and contact gaps of the disconnectors; as well as key details for status identification such as equipment identification text, instrument scales, and on / off indicators, while retaining the gradient information of all equipment edges.

[0034] By separating the enhancement layer from the base layer, the overall brightness and local details can be optimized independently, avoiding the problem of weakening details when adjusting brightness and disrupting the overall brightness balance when enhancing details.

[0035] Step S104: Perform contrast adaptive adjustment on the base layer, and then perform weighted fusion of the adjusted base layer and the enhancement layer to generate second visual data.

[0036] In this embodiment, the base layer undergoes adaptive dynamic range adjustment. This adjustment includes reducing the overall contrast of the highlight region using a first coefficient, and then restoring the contrast of the compressed base layer using a second coefficient. Both the first and second coefficients are dynamically calculated from the overall histogram distribution of the base layer.

[0037] When reducing the overall contrast of the highlight areas using the first coefficient, the pixel proportion of the high-brightness range in the grayscale histogram of the base layer is statistically analyzed. When the proportion exceeds a preset threshold, a significant overexposed area is identified, and a first coefficient less than 1 is calculated. Non-linear grayscale compression is then applied to the pixels in the high-brightness range to suppress overexposed areas caused by direct sunlight and reflected light, restoring the surface structure of the equipment in overexposed areas. When restoring the contrast of the compressed base layer using the second coefficient, the pixel proportion of the low-brightness range in the base layer is statistically analyzed. Based on this, a second coefficient greater than 1 is calculated. Grayscale stretching is then applied to the compressed mid-to-low brightness range to enhance the brightness and contrast of dark areas such as shadows inside the cabinet and the backlight surface of the equipment, highlighting the hidden details of the equipment in the dark areas.

[0038] It should be noted that the first and second coefficients are updated in real time according to the histogram distribution of each frame of the image. This allows for adaptive adjustment to changes in lighting conditions during the morning, noon, and evening, differences in weather conditions, and varying indoor and outdoor lighting conditions in the power distribution equipment monitoring area. This improves visibility in dark areas without causing secondary overexposure. The specific calculation process for the first and second coefficients is as follows: Let the normalized base layer image be Its grayscale value range is Analyze the gray-level histogram of the image to obtain the cumulative distribution function: ; In the formula, Gray levels The probability, This is the grayscale threshold; First coefficient Used to reduce overall contrast, its value is equal to the dynamic range span between the two cutoff points of the grayscale histogram. Sets the cutoff ratio. (Typically taken as 0.01~0.05), calculate the low-end threshold. and high-end threshold : ; ; First coefficient , The smaller the value, the lower the contrast of the base layer (original image), and the less compression is needed; conversely, the larger the value, the more compression is needed.

[0039] Second coefficient The formula used to restore contrast in the compressed base layer is as follows: ; In the formula, For a very small positive number (such as This is to prevent division by zero. The restored grayscale value... Calculated using the following formula: ; Therefore, the original grayscale range can be... Linear stretch back At the same time, it suppresses over-enhancement in extremely low or high grayscale regions.

[0040] In this embodiment, the restored base layer and enhancement layer are weighted and fused to generate the final optimized second visual data; the fusion weight is not a globally fixed value, but is dynamically allocated according to the local variance of each pixel neighborhood to achieve a balance between structure preservation and detail enhancement.

[0041] Specifically, the local variance of the enhancement layer is calculated using a neighborhood window of 3×3 or 5×5 pixels. Higher local variance indicates richer texture and edge information in the region, corresponding to key feature areas of the device. In this case, a higher fusion weight is assigned to the enhancement layer to amplify local texture contrast and enhance anomaly-related details. Conversely, lower local variance indicates a flat region. In this case, a higher weight is assigned to the base layer to maintain overall brightness smoothness and avoid amplifying background noise in flat areas. Therefore, while preserving the overall structure and brightness naturalness of the device, the contrast of key details required for anomaly identification in power distribution equipment can be specifically enhanced, avoiding noise amplification and color distortion caused by global enhancement.

[0042] In this way, the pixel-physical coordinate correspondence brought about by spatial registration is fully preserved, and common problems such as uneven lighting, overexposure and loss of details in dark areas, and low contrast of defect features in power distribution monitoring scenarios are specifically solved. The equipment structure is clear and key details are highlighted, which can be directly input into the subsequent deep convolutional neural network anomaly recognition model, effectively improving the recognition accuracy and positioning precision of faults such as equipment appearance defects, abnormal status, and foreign object intrusion.

[0043] Step S105: Extract the temporal change feature vector of the device appearance based on the temporally continuous second visual data; perform matching analysis on the temporal change feature vector in combination with the preset abnormal feature library to determine the abnormal identification result.

[0044] In this embodiment, an anomaly recognition model based on a deep convolutional neural network is used. Based on multiple frames of second visual data in a continuous time sequence, the temporal change features of the device appearance are extracted and matched with a preset anomaly feature library to output device status signals and anomaly feature information.

[0045] The anomaly detection model is a hybrid architecture based on a cascaded 3D convolutional neural network and ConvLSTM, and its specific structure is as follows: The input layer receives temporally continuous second visual data after spatial registration and feature enhancement, with the number of frames taken as... Each frame of the image is uniformly scaled to a resolution of 224×224 pixels, and the input tensor size is [missing information]. .

[0046] The 3D convolutional layer consists of two 3D convolutional layers. The kernel size of the first convolutional layer is [size missing]. Corresponding time dimension Height dimension The width dimension has a sliding stride of 1, and the number of output channels is 64. After convolution, a batch normalization layer and a ReLU activation function are applied to extract short-term spatiotemporal features, such as frame-by-frame flickering of the electric arc, pulsating changes in the corona glow, inter-frame displacement of small foreign objects, and short-term diffusion of local discoloration at the joint, etc., which are fine-grained dynamic features. The second convolutional kernel is also... The output channel count is 128, and the sliding step size is 2 for downsampling, followed by batch normalization and ReLU activation function.

[0047] The ConvLSTM layer receives the feature map sequence (size 8×112×112×128) output from the second convolutional layer, sets the number of hidden state channels to 256, and the internal convolutional kernel size to 3×3. It is used to capture mid- to long-term temporal dependencies and outputs a high-dimensional feature map after temporal fusion. ConvLSTM uses gating mechanisms—input gate, forget gate, and output gate—to update the state of the 8-frame feature sequence frame by frame, memorizing the global state information of previous frames, thus enabling the modeling of anomalous evolution patterns over longer time spans.

[0048] The sampling global average pooling layer performs global average pooling on the high-dimensional feature map output by the ConvLSTM layer in the spatial dimension, reducing the spatial dimension of the high-dimensional feature map to [missing value]. This yields a one-dimensional feature vector of length 256. Finally, a two-stage fully connected layer is used to complete the feature transformation and classification output: the first-stage fully connected layer maps the 256-dimensional feature vector to 128-dimensional features, further completing the nonlinear fusion and dimensionality reduction of high-dimensional spatiotemporal features; the second-stage output... dimension( (The number of anomaly categories is preset), and finally, the probability score for each anomaly category is generated by the Sigmoid activation function.

[0049] Here, the anomaly recognition model is trained on a massive amount of labeled images and video data of normal and abnormal states of power distribution equipment (such as joint heating and deformation, insulator flashover, foreign objects in cabinets, etc.).

[0050] The anomaly identification model extracts the temporal change feature vector of the device appearance based on multiple frames of continuous second visual data. Then, it performs matching analysis in combination with a preset anomaly feature library to output device status signals and anomaly feature information.

[0051] The anomaly feature library in this embodiment is a structured database that stores pattern descriptions of various typical anomalies and their corresponding matching features. Each record includes at least the following fields: anomaly type ID, anomaly name, feature vector prototype (128-dimensional floating-point array), matching threshold, anomaly level (minor / serious / critical), and suggested handling strategy index.

[0052] As a feasible implementation method, typical anomalies and matching features are shown in Table 1 below: Table 1

[0053] When performing matching analysis based on an anomaly feature library, the cosine similarity between the temporal change feature vector (comprehensive spatiotemporal feature vector) and the prototype feature vector of each anomaly category in the anomaly feature library is calculated one by one, and candidate anomaly types with a cosine similarity greater than or equal to the corresponding matching threshold are selected and retained. In this embodiment, for multiple temporally continuous analysis cycles (e.g., once per second), a sliding window statistical approach is used: if more than a preset number of frames in K consecutive frames are determined to be of the same anomaly type, then the anomaly is confirmed to have occurred, avoiding instantaneous noise interference. Thus, the anomaly type, anomaly region location, confidence level, and the handling strategy ID matched from the adaptive strategy library based on the anomaly level are output.

[0054] For example, the preset number of frames is 0.6K frames, where K=5.

[0055] As a feasible implementation method, 16 consecutive frames of second visual data are input into the anomaly recognition model in chronological order, and forward propagation calculation is performed. First, two 3D convolutional layers extract short-term spatiotemporal features layer by layer, extracting local dynamic features of the device appearance changing over time (such as arc flashing and insulator crack propagation). The generated feature map sequence is input into a ConvLSTM layer, and the hidden state is updated frame by frame to complete the modeling of long-term temporal dependencies. Finally, a 256-dimensional comprehensive spatiotemporal feature vector is obtained through global average pooling. Based on this feature vector and the prototype feature vector of each anomaly category in the preset anomaly feature library (obtained by the mean of similar samples in the training set), the cosine similarity is calculated. : ; In the formula, Given the feature vectors of the input image sequence, For the first The prototype feature vector of the anomaly class.

[0056] In this embodiment, the probability score output by the anomaly detection model is weighted and fused with the cosine similarity to obtain the final confidence score. For example, ,in, Let it be a probability fraction; if If the value exceeds the preset confidence threshold, the system determines that such an anomaly exists and outputs the corresponding device status signal and anomaly characteristic information.

[0057] The power distribution anomaly identification method provided in this embodiment, after acquiring the first visual data of the power distribution equipment monitoring area, first performs spatial registration on the acquired first visual data, and then performs contrast adaptive adjustment based on layered processing, simultaneously optimizing from two dimensions: spatial reference and image quality, to obtain second visual data; then, based on the temporally continuous second visual data, it extracts the temporal change feature vector of the equipment appearance; and combines it with a preset anomaly feature library to perform matching analysis on the temporal change feature vector to determine the anomaly identification result. Based on eliminating the interference of image offset on temporal analysis, the method uses guided filtering layering combined with contrast adaptive adjustment to dynamically optimize the contrast performance of highlights and shadows according to the actual illumination distribution of the image. While suppressing overexposed areas and restoring equipment structural information, it improves the visibility of details in shadows, fully preserving key features highly correlated with anomaly identification, such as equipment texture and edges. This allows the final generated visual data to effectively support subsequent temporal change feature extraction and anomaly matching analysis, significantly improving the accuracy, positioning precision, and scene adaptability of power distribution equipment anomaly identification, and ensuring the engineering practical value of the power distribution visual monitoring solution.

[0058] Example 2 Based on the above embodiments, this embodiment also provides a power distribution anomaly identification device for implementing the above power distribution anomaly identification method, referring to... Figure 2 The apparatus in this embodiment includes: Data acquisition module 201 is used to acquire first visual data of the power distribution equipment monitoring area; Spatial registration module 202 is used to spatially register the first visual data to obtain spatially registered first visual data; The data layering module 203 is used to normalize the spatially registered first visual data and then perform layering processing on the spatially registered first visual data through guided filtering to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that includes texture features and edge detail information. The adaptive adjustment module 204 is used to perform contrast adaptive adjustment on the base layer, and then perform weighted fusion of the adjusted base layer and the enhancement layer to generate second visual data. The identification and analysis module 205 is used to extract the temporal change feature vector of the device appearance based on the second visual data that is continuous in time; and to perform matching analysis on the temporal change feature vector in combination with the preset abnormal feature library to determine the abnormal identification result.

[0059] In this embodiment, the identification and analysis module 205 is also connected to a non-visual sensor data interface for receiving temperature, current, and partial discharge monitoring data from the power distribution equipment. The identification and analysis module 205 is configured to perform cross-modal feature fusion of the optimized second visual data and the non-visual sensor data to generate a comprehensive anomaly judgment result and generate control commands. Specifically, the cross-modal feature fusion is performed at the decision level, where the visual analysis results (such as appearance anomaly category and location) and sensor data analysis results (such as whether the temperature exceeds the threshold or whether the partial discharge exceeds the standard) are independently judged, and then integrated through preset fusion rules (such as weighted voting and DS evidence theory) to reduce the false alarm rate of single-source information and improve the reliability of the overall judgment.

[0060] The power distribution anomaly identification device in this embodiment also includes an adaptive strategy library, used to store handling strategies for different equipment types, anomaly levels, and environmental conditions. When generating control commands, the identification and analysis module 205 calls the adaptive strategy library to match the optimal handling strategy. Specifically, the adaptive strategy library is an updatable rule database. For example, for an anomaly of "slightly increased cable joint temperature," the strategy might be "issue an early warning to the mobile terminal of maintenance personnel"; for a severe anomaly of "obvious electric arc appearing in the switchgear" and a sudden increase in current, the strategy might be "immediately trigger an audible and visual alarm and automatically request the disconnection of the upstream power supply."

[0061] Corresponding to the power distribution anomaly identification method provided in the embodiments of this application, the embodiments of this application also provide an electronic device for performing the power distribution anomaly identification method, referring to... Figure 3 The electronic device includes: a processor 301; and a memory 302 for storing a program for a power distribution anomaly identification method. After the device is powered on and the program for the power distribution anomaly identification method is run by the processor, the following steps are performed: Acquire first-view data of the power distribution equipment monitoring area; Spatial registration is performed on the first visual data to obtain spatially registered first visual data; After normalizing the spatially registered first visual data, guided filtering is used to perform layered processing on the spatially registered first visual data to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that contains texture features and edge detail information. The contrast of the base layer is adaptively adjusted, and the adjusted base layer is weighted and fused with the enhancement layer to generate second visual data. The temporal change feature vector of the device appearance is extracted based on the second visual data with continuous temporal sequence; the temporal change feature vector is matched and analyzed in combination with the preset abnormal feature library to determine the abnormal identification result.

[0062] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method in any of the embodiments of this application. Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the above embodiments is stored, and the computer (or CPU (Central Processing Unit) or MPU (Microprocessor Unit) of the system or apparatus may read and execute the program code stored in the storage medium.

[0063] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined above in the system of this application.

[0064] It should be noted that the computer-readable storage medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF (Radio Frequency), etc., or any suitable combination thereof.

[0065] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0066] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0067] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0068] In the several embodiments provided in this application, it should be understood that the disclosed systems, modules, and methods can be implemented in other ways. For example, the module embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules or units, and may be electrical, mechanical, or other forms.

[0069] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. This application is not limited to the exact structures described above and illustrated in the accompanying drawings, and it should not be considered that the specific implementation of this application is limited to these descriptions. For those skilled in the art, various changes and modifications made without departing from the concept of this application should be considered to fall within the protection scope of this application.

Claims

1. A method for identifying power distribution anomalies, characterized in that, include: Acquire first-view data of the power distribution equipment monitoring area; Spatial registration is performed on the first visual data to obtain spatially registered first visual data; After normalizing the spatially registered first visual data, guided filtering is used to perform layered processing on the spatially registered first visual data to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that contains texture features and edge detail information. The contrast of the base layer is adaptively adjusted, and the adjusted base layer is weighted and fused with the enhancement layer to generate second visual data. Extracting temporal variation feature vectors of device appearance based on temporally continuous second visual data; The anomaly identification result is determined by matching and analyzing the time-series change feature vectors with a pre-set anomaly feature library.

2. The power distribution anomaly identification method according to claim 1, characterized in that, The step of spatially registering the first visual data to obtain spatially registered first visual data includes: A first plane coordinate system is established on the actual physical plane of the power distribution equipment monitoring area. A preset corner point of the first plane coordinate system is used as the origin, and the two mutually perpendicular boundary directions are used as the coordinate axis directions to determine the actual coordinate information. The first visual data is converted into a single-frame image, and a second planar coordinate system is established on the single-frame image; For each single frame image, the image coordinate information is determined with the geometric center of the single frame image as the origin and the horizontal and vertical directions of the single frame image as the coordinate axes. Based on the angular offset relationship between the actual coordinate information and the image coordinate information, a correction matrix is ​​constructed, and a geometric transformation is performed on a single-frame image based on the correction matrix to obtain the spatially registered first visual data; wherein, the coordinates of each pixel in the single-frame image are remapped based on the correction matrix, specifically: ； In the formula, For the correction matrix; These are the cosine transform coefficients. ; These are the coefficients of the sinusoidal transform. ; This is the angular offset. These are the coordinates of the geometric center of a single frame image.

3. The power distribution anomaly identification method according to claim 2, characterized in that, The process of performing layered processing on the spatially registered first visual data through guided filtering yields a base layer characterizing the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer containing texture features and edge detail information, including: Using normalized visual data as a guide map, a basic layer that can characterize the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area is extracted. Pixel-level difference operations are performed between the normalized visual data and the base layer to separate an enhancement layer containing the texture features and edge detail information of the power distribution equipment.

4. The power distribution anomaly identification method according to claim 3, characterized in that, The step of adaptively adjusting the contrast of the base layer and weightedly fusing the adjusted base layer with the enhancement layer to generate second visual data includes: The percentage of pixels in the high-brightness region in the gray-level histogram of the base layer is statistically analyzed. If the percentage of pixels in the high-brightness region is greater than the preset highlight threshold, the high-brightness region pixels of the base layer are non-linearly compressed using the first coefficient to obtain the compressed base layer. The pixel percentage of low-brightness regions in the grayscale histogram of the base layer is statistically analyzed, and the grayscale of the low-brightness regions in the compressed base layer is stretched using the second coefficient to obtain the restored base layer. The restored base layer and the enhancement layer are weighted and fused to generate second visual data. The first coefficient and the second coefficient are updated in real time with the grayscale histogram of each frame of the image; the first coefficient is less than 1 and the second coefficient is greater than 1.

5. The power distribution anomaly identification method according to claim 4, characterized in that, The first coefficient and the second coefficient are updated in real time with the grayscale histogram of each frame of the image, including: For each frame of the image, the gray-level histogram of the base layer of the image is statistically analyzed to obtain the cumulative distribution function; Based on the cumulative distribution function, the dynamic range span between the two cutoff points of the gray-level histogram is determined, resulting in the first coefficient and the second coefficient; wherein, the first coefficient is the difference between the high-end threshold and the low-end threshold: ； ； ； ； In the formula, For the low-end threshold, For the high-end threshold, For cumulative distribution values, For the truncation ratio, As the first coefficient, As the second coefficient, It is a very small positive number.

6. The power distribution anomaly identification method according to claim 1, characterized in that, The temporal change feature vector of the device appearance extracted based on the second visual data of temporal continuity includes: An anomaly detection model is used to extract temporal variation feature vectors of device appearance based on multiple frames of temporally continuous second visual data; The anomaly recognition model is a hybrid architecture based on a cascaded 3D convolutional neural network and a ConvLSTM. It uses three-dimensional convolutional layers to extract short-term spatiotemporal features layer by layer, inputs the extracted feature map sequence into the ConvLSTM layer, updates the hidden state frame by frame, and completes the modeling of long-term temporal dependencies. Finally, it obtains a comprehensive spatiotemporal feature vector through global average pooling.

7. The power distribution anomaly identification method according to claim 6, characterized in that, The step of matching and analyzing the time-series change feature vectors using a pre-defined anomaly feature library to determine the anomaly identification result includes: Calculate the cosine similarity between the comprehensive spatiotemporal feature vector and the prototype feature vector of each anomaly category in the anomaly feature library, and filter and retain candidate anomaly types with a cosine similarity greater than or equal to the corresponding matching threshold. The cosine similarity of candidate anomaly types is weighted and fused with the probability scores output by the anomaly recognition model to obtain the final confidence level. The final maximum confidence level is compared with the preset confidence threshold. If the final maximum confidence level is greater than the preset confidence threshold, the candidate anomaly type corresponding to the final maximum confidence level is determined as the anomaly identification result.

8. A power distribution anomaly identification device, characterized in that, include: The data acquisition module is used to acquire first-view data of the power distribution equipment monitoring area; The spatial registration module is used to perform spatial registration on the first visual data to obtain the spatially registered first visual data. The data layering module is used to normalize the spatially registered first visual data and then perform layering processing on the spatially registered first visual data through guided filtering to obtain a base layer that characterizes the overall brightness distribution and equipment structure information of the power distribution equipment monitoring area, and an enhancement layer that contains texture features and edge detail information. An adaptive adjustment module is used to adaptively adjust the contrast of the base layer, and then weightedly fuse the adjusted base layer with the enhancement layer to generate second visual data. The identification and analysis module is used to extract temporal variation feature vectors of the device appearance based on temporally continuous second visual data; The anomaly identification result is determined by matching and analyzing the time-series change feature vectors with a pre-set anomaly feature library.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the power distribution anomaly identification method according to any one of claims 1-7.

10. A computer-readable storage medium storing a plurality of instructions thereon, characterized in that, The instructions are applicable to be loaded and executed by a processor to implement the power distribution anomaly identification method according to any one of claims 1 to 7.

Images