Inspection Management Method for Distribution Network

Abstract

The present invention relates to the technical field of data processing, and discloses an inspection management method for a distribution network, which is used to improve the efficiency of inspection management of the distribution network. The method includes: collecting and preprocessing multi-source data of multiple distribution network inspection nodes to obtain denoised multi-dimensional image data; performing feature extraction processing on the multi-dimensional image data to obtain device anomaly feature data of multiple distribution network inspection nodes; performing data fusion processing on the device anomaly feature data of multiple distribution network inspection nodes to obtain spatio-temporally correlated device anomaly comprehensive data; performing continuity analysis processing on the device anomaly comprehensive data through a spatio-temporal tracking algorithm to obtain device anomaly spatio-temporal change data; performing intelligent positioning processing on the device anomaly spatio-temporal change data to obtain device fault space coordinate data; performing anomaly level evaluation processing on the device fault space coordinate data based on a priority sorting algorithm to obtain device anomaly priority information.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an inspection management method for a distribution network. Background Art

[0002] In the operation management of modern distribution networks, inspection, as an important link to ensure the normal operation of power equipment, has received increasing attention. Existing technologies usually adopt regular manual inspections and basic automatic monitoring means to evaluate the status and detect faults of equipment in the distribution network. These methods collect the operation data and image information of the equipment and rely on traditional image processing technologies and simple feature extraction algorithms to achieve abnormal detection of the equipment. However, with the increase in the types and quantities of distribution network equipment, the traditional manual inspection method is inefficient and prone to missing problems, and cannot reflect the equipment status in a timely and comprehensive manner.

[0003] Although existing technologies have improved the inspection efficiency of distribution networks to a certain extent, there are still multiple deficiencies. On the one hand, traditional image processing algorithms are sensitive to image noise in complex environments, resulting in poor image data processing effects and difficulty in accurately extracting equipment features. On the other hand, existing abnormal detection means lack intelligence and automation, cannot respond to the dynamic changes of equipment faults in real time, and cannot provide effective decision-making support. This makes the positioning and evaluation of equipment faults have a certain lag, affecting the safe and stable operation of the distribution network. Therefore, there is an urgent need for a more intelligent and efficient inspection management method to solve the many problems existing in the prior art. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides an inspection management method for a distribution network, which is used to improve the inspection management efficiency of the distribution network.

[0005] The present invention provides an inspection management method for a distribution network, including:

[0006] Collect and preprocess multi-source data of multiple distribution network inspection nodes to obtain denoised multi-dimensional image data; perform feature extraction processing on the multi-dimensional image data through an image recognition algorithm to obtain equipment abnormal feature data of multiple distribution network inspection nodes; perform data fusion processing on the equipment abnormal feature data of multiple distribution network inspection nodes to obtain spatio-temporally correlated equipment abnormal comprehensive data; perform continuity analysis processing on the equipment abnormal comprehensive data through a spatio-temporal tracking algorithm to obtain equipment abnormal spatio-temporal change data; perform intelligent positioning processing on the equipment abnormal spatio-temporal change data to obtain equipment fault space coordinate data; perform abnormal level evaluation processing on the equipment fault space coordinate data based on a priority sorting algorithm to obtain equipment abnormal priority information.

[0007] In the technical solution provided by the present invention, multi-source data acquisition and denoising processing are adopted, which can effectively filter out environmental interference and equipment noise, ensuring the quality of the acquired original data. Secondly, through image recognition algorithms for feature extraction, the potential abnormal features of the equipment can be deeply explored. It can not only identify minor faults of the equipment in a timely manner, but also gradually improve the feature library through learning historical data to achieve the ability to identify new types of faults. Moreover, through data fusion processing, the equipment feature data of multiple inspection nodes are integrated to generate comprehensive equipment anomaly data with spatio-temporal correlation. This process can effectively capture the change trends of the equipment in terms of time and space, thereby improving the overall understanding of the equipment's state. In addition, combining spatio-temporal tracking algorithms for continuous analysis and processing of the comprehensive equipment anomaly data enables the monitoring of the equipment's state changes over a long period of time, providing strong data support for fault prediction. It is especially suitable for high-risk and complex power distribution environments, and can effectively reduce the risk of power outages caused by equipment failures. This efficient fault location ability greatly reduces the workload of manual inspections and improves the inspection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0009] Figure 1 It is a flowchart of an inspection management method for a distribution network in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0011] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0012] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0013] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Please refer to Figure 1 , Figure 1 which is a flowchart of an inspection and management method for a distribution network according to an embodiment of the present invention. As Figure 1 shown, it includes the following steps:

[0014] S101. Collect and preprocess multi-source data of multiple distribution network inspection nodes to obtain denoised multi-dimensional image data;

[0015] S102. Extract feature data of equipment anomalies of multiple distribution network inspection nodes by performing feature extraction on the multi-dimensional image data through an image recognition algorithm;

[0016] S103. Perform data fusion on the equipment anomaly feature data of multiple distribution network inspection nodes to obtain spatio-temporally correlated comprehensive equipment anomaly data;

[0017] S104. Perform continuity analysis on the comprehensive equipment anomaly data through a spatio-temporal tracking algorithm to obtain spatio-temporal change data of equipment anomalies;

[0018] S105. Perform intelligent positioning on the spatio-temporal change data of equipment anomalies to obtain spatial coordinate data of equipment failures;

[0019] S106. Perform anomaly level evaluation on the spatial coordinate data of equipment failures based on a priority sorting algorithm to obtain equipment anomaly priority information.

[0020] It should be noted that first, multi-source data of multiple distribution network inspection nodes are collected and preprocessed to obtain multi-dimensional image data after denoising. The multi-source data includes monitoring data from different sensors, such as video monitoring, infrared imaging, and temperature sensors. These data will undergo unified formatting processing to ensure their consistency. For example, the original images collected by image acquisition devices may contain high-frequency noise. Therefore, first, the original images are denoised using the wavelet transform method. Through multi-scale decomposition, the signal and noise are separated. Specifically, the image is decomposed into different frequency bands, the low-frequency part is retained, and the high-frequency noise is suppressed to obtain multi-dimensional image data after denoising. Furthermore, an image recognition algorithm is used to perform feature extraction processing on the multi-dimensional image data to extract device anomaly feature data of multiple distribution network inspection nodes. Image recognition algorithms such as convolutional neural networks (CNNs) can automatically extract important features in images. The trained model can identify different states of the device and extract key features of the device from the denoised image, such as abnormal features like cracks and deformations. This process includes region segmentation, dividing the image into multiple parts for individual analysis, and then classifying and identifying the features of each region. For example, for an image containing a distribution box, the algorithm may detect whether the closing state of the door is normal and whether there are obvious corrosion or aging marks.

[0021] After data fusion processing of the device anomaly feature data of multiple distribution network inspection nodes, spatio-temporally correlated device anomaly comprehensive data can be obtained. This process involves integrating data obtained at different times and spaces. Usually, a weighted fusion algorithm is used, and weighted processing is performed according to the importance and anomaly degree of each node to ensure the dominant position of key data in the comprehensive result. Specifically, assume that the anomaly degree of a certain node is evaluated as 0.8, while that of another node is 0.5. The comprehensive data will be synthesized according to these weights to obtain more representative anomaly comprehensive data. For example, if the temperatures of two nodes are 80°C and 75°C respectively, the fused data may show 78°C and indicate the anomaly situations of the two nodes. Through spatio-temporal tracking algorithms, after continuous analysis processing of the device anomaly comprehensive data, device anomaly spatio-temporal change data can be obtained. This algorithm uses time series analysis and can identify abnormal change trends, such as continuously rising temperatures or frequently occurring device failures. For data analysis, the sliding window technique can be applied to update the device status regularly to more accurately reflect the real-time situation of the device. For example, if the operating temperature of a device has been rising every day in the past week and the change amplitude exceeds the set threshold, this indicates that there may be a potential fault.

[0022] After completing the intelligent positioning process of the spatio-temporal change data of equipment anomalies, the spatial coordinate data of equipment failures can be obtained. This process involves using a spatial positioning algorithm to infer the actual failure location of the equipment based on the processed anomaly data. By comparing historical failure records and combining current anomaly data, the specific location where the failure occurred can be located, such as a certain module in the distribution box. For example, by analyzing the temperature changes and load conditions before and after the failure, the system may determine the failure location of a certain module. Finally, based on the priority sorting algorithm, the anomaly level assessment process is carried out on the spatial coordinate data of equipment failures, so as to obtain the equipment anomaly priority information. In this step, the algorithm will consider the importance of the equipment, the severity of the failure, and the possible impact on the overall distribution network, and comprehensively evaluate the priority of each failure. Suppose there are three devices with failures respectively, their importance scores are 0.9, 0.7, and 0.5, and the failure impact scores are 0.8, 0.4, and 0.6. After weighted sorting, the highest-priority failure will be determined finally, so as to facilitate the operation and maintenance team to handle it quickly.

[0023] By performing the above steps and adopting multi-source data acquisition and denoising processing, environmental interference and equipment noise can be effectively filtered out, ensuring the quality of the acquired raw data. Secondly, through feature extraction using image recognition algorithms, the potential anomaly features of the equipment can be deeply explored. It can not only identify minor equipment failures in a timely manner, but also gradually improve the feature library through learning historical data, realizing the ability to identify new types of failures. Moreover, through data fusion processing, the equipment feature data of multiple inspection nodes are integrated to generate spatio-temporal associated comprehensive equipment anomaly data. This process can effectively capture the change trends of the equipment in time and space, thereby improving the overall understanding of the equipment status. In addition, by combining spatio-temporal tracking algorithms to perform continuous analysis processing on the comprehensive equipment anomaly data, the status changes of the equipment can be monitored within a relatively long time range, providing strong data support for fault prediction. It is especially suitable for high-risk and complex distribution environments, and can effectively reduce the risk of power outages caused by equipment failures. This efficient fault location ability greatly reduces the workload of manual inspections and improves the inspection efficiency.

[0024] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0025] (1) Perform data acquisition processing on the multi-source data of multiple distribution network inspection nodes to obtain raw image data, and perform format conversion processing on the raw image data to obtain image data in a unified format;

[0026] (2) Perform grayscale processing on the image data in a unified format to obtain grayscale image data, and perform normalization processing on the grayscale image data to obtain standardized image data;

[0027] (3) Denoise the standardized image data through the wavelet transform algorithm to obtain denoised image data, and perform edge smoothing processing on the denoised image data to obtain smoothed image data;

[0028] (4) Perform feature enhancement processing on the smoothed image data to obtain enhanced feature image data, and perform image reconstruction processing on the enhanced feature image data to obtain multi-dimensional image data after denoising processing.

[0029] Specifically, first perform data acquisition processing on the multi-source data of multiple distribution network inspection nodes to obtain original image data. This data can come from different types of sensors, such as surveillance cameras, infrared imaging devices, and temperature sensors, etc., covering the visual information and thermal imaging data of the equipment. During the data acquisition process, by setting reasonable sampling frequencies and resolutions, ensure that the key state information of the equipment can be captured. For example, the surveillance camera takes pictures at a rate of 30 frames per second, and the obtained image data constitutes a rich original image data set. Furthermore, perform format conversion processing on the original image data to obtain image data in a unified format. The key to this step is to convert the images from different sensors into a consistent file format, such as JPEG or PNG, for subsequent processing. During this process, use image processing tools to adjust all images according to a unified size and color depth to ensure that each image works within the same processing framework. This format unification helps with subsequent data processing and feature extraction.

[0030] The image data in the unified format will then be grayscaled to obtain grayscale image data. Grayscaling is the process of converting a color image into a black-and-white image, usually using the weighted average method, and converting each pixel value in the RGB color space into a grayscale value through a formula. Specifically, the grayscale value can be calculated by the following formula:

[0031] ;

[0032] where represents the grayscale value, , and Color values of red, green, and blue respectively. Through grayscale processing, the image information becomes simpler and is convenient for subsequent analysis. Subsequently, the grayscale image data is normalized to obtain standardized image data. The purpose of normalization is to map the grayscale value range to between [0, 1], avoiding image analysis errors caused by lighting changes. Through normalization, the standardized image data can eliminate the influence of lighting changes on subsequent processing and make feature extraction more accurate. Then, the standardized image data is denoised through the wavelet transform algorithm to obtain denoised image data. Wavelet transform is a signal processing technology that can effectively separate signals from noise. Specifically, first, the image data is wavelet decomposed to obtain different frequency components. Then, through thresholding, the high-frequency components below the set threshold are removed, and the main low-frequency information is retained. In this way, the denoised image data can retain key information while removing useless noise and improving the clarity of the image.

[0033] After denoising processing, the denoised image data is further subjected to edge smoothing processing to obtain smoothed image data. Edge smoothing usually uses filtering techniques such as Gaussian blur. By performing weighted average processing on each pixel and its neighborhood, the edge noise of the image is reduced. For example, when using a 3×3 Gaussian filter, the calculation can be performed through the following weighted coefficients:

[0034] ;

[0035] Through this method, the smoothing effect of the image is significant, which can eliminate noise in a small range and improve the accuracy of subsequent feature extraction. Furthermore, the smoothed image data is subjected to feature enhancement processing to obtain enhanced feature image data. The purpose of feature enhancement is to make the important features of the device more prominent by adjusting the contrast or applying image enhancement algorithms. This step can use the histogram equalization method to improve the visibility of weak features in the image by adjusting the contrast distribution of the image. For example, after enhancement processing, the fault features (such as cracks) of some devices may be significantly highlighted, facilitating identification. Finally, the enhanced feature image data is subjected to image reconstruction processing to obtain multi-dimensional image data after denoising processing. The process of image reconstruction reconstructs the processed data into a higher-quality image through inverse wavelet transform. This process can not only retain the information after denoising and feature enhancement but also ensure the integrity and accuracy of the image. By synthesizing each frequency component, the finally output multi-dimensional image data can clearly display the state of the power distribution network inspection nodes, laying a foundation for subsequent anomaly detection and analysis.

[0036] For example, in practical applications, after the above processing steps, the standard deviation of the image data of a certain distribution node before denoising is 20. After wavelet transform denoising, the standard deviation is reduced to 5, indicating that the noise is significantly reduced and the image quality is significantly improved. Through these data processing means, the inspection efficiency of the distribution network is effectively improved, and equipment faults can be identified and processed more timely and accurately.

[0037] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0038] (1) Perform feature preprocessing on the multi-dimensional image data to obtain the processed image data, and perform feature enhancement processing on the processed image data to obtain the enhanced image feature data;

[0039] (2) Perform region segmentation processing on the enhanced image feature data to obtain the segmented feature region data, and perform shape description processing on the segmented feature region data to obtain the shape feature data;

[0040] (3) Perform feature extraction processing on the shape feature data through a convolutional neural network to obtain a preliminary feature set, and perform feature selection processing on the preliminary feature set to obtain an optimized feature set;

[0041] (4) Perform abnormal pattern recognition processing on the optimized feature set to obtain equipment abnormal pattern data, and perform classification processing on the equipment abnormal pattern data to obtain classification feature data;

[0042] (5) Perform abnormal feature marking processing on the classification feature data to obtain the equipment abnormal feature data of multiple distribution network inspection nodes.

[0043] Specifically, noise filtering and information extraction are performed on the multi-dimensional image data to obtain the processed image data. This process includes applying a smoothing algorithm, such as a Gaussian filter, which reduces random noise in the image by weighted averaging of surrounding pixel values. In a specific implementation, if the signal-to-noise ratio of the original image data is 5 dB, after Gaussian filtering, the signal-to-noise ratio is increased to 15 dB, indicating that the noise interference is effectively reduced and the image quality is improved. After feature preprocessing, feature enhancement processing is performed with the aim of highlighting important features to make them more obvious in subsequent analysis. This process can be achieved through contrast enhancement and edge enhancement techniques. For example, through histogram equalization technology, the gray distribution of the image becomes more uniform, thereby enhancing the contrast of the image. The set contrast enhancement coefficient is 2, which means that the contrast between the brightest and darkest parts of the image is increased by two times. The enhanced image feature data can more clearly display the fault features of distribution equipment, such as cracks or deformations.

[0044] Subsequently, the image data after feature enhancement processing will undergo region segmentation processing to obtain the segmented feature region data. The region segmentation technology adopts the threshold segmentation method, which divides the image into foreground and background according to the signal intensity (such as grayscale value). The threshold is set to 120, and all pixels above this value are marked as the foreground to extract key information. After segmentation, if 30 foreground regions are obtained, it indicates that the key features in the device are effectively separated, providing a clear data basis for subsequent analysis. After obtaining the segmented feature region data, shape description processing is carried out to obtain shape feature data. Shape description usually includes methods such as boundary extraction, shape moments, and Fourier descriptors. Through these descriptions, the shape information of the device features can be accurately captured. The shape index indicates that the shape of this region is close to a circle, facilitating feature matching in subsequent abnormal pattern recognition.

[0045] Feature extraction processing is performed on the shape feature data through a convolutional neural network (CNN) to obtain a preliminary feature set. CNN is suitable for processing image data due to its good feature extraction ability. In this process, through the combination of multiple convolutional layers and pooling layers, the low-level features (such as edges and textures) of the image are gradually transformed into high-level features (such as the shape and structure of objects). The preliminary feature set will contain multi-dimensional information, such as the edge features and texture features of the shape. Feature selection processing is carried out on the preliminary feature set to obtain an optimized feature set. The purpose of feature selection is to screen out the features that have the greatest impact on the classification result from the preliminary feature set. Common methods include principal component analysis (PCA) and LASSO regression. Through feature selection, the optimized feature set may be reduced from the initial 100 features to 30. This process not only reduces the complexity of data processing but also improves the accuracy of subsequent analysis. Furthermore, abnormal pattern recognition processing is carried out on the optimized feature set to obtain device abnormal pattern data. Abnormal pattern recognition uses classification algorithms such as support vector machines (SVM) or random forests, which can effectively distinguish normal and abnormal patterns. After training, if 8 different abnormal patterns are identified, including overheating, short circuit, etc., it can provide a reliable basis for subsequent fault diagnosis.

[0046] Classify the device abnormal mode data to obtain classified feature data. The classification process usually adopts multi-class classification algorithms. By learning the training data, the model maps each input feature to a specific class. For example, if the classification result shows that there is a 60% probability of a short circuit risk at a certain power distribution node, this node needs to be focused on during the inspection. Finally, perform abnormal feature identification processing on the classified feature data to obtain the device abnormal feature data of multiple power distribution network inspection nodes. During the identification process, based on the classification result, label the status of each node to form an abnormal feature data set. This data set contains the status information and potential fault types of each node, providing an important basis for subsequent operation and maintenance decisions. For example, if a node is identified as having a short circuit risk, the maintenance personnel can immediately arrange for maintenance to reduce the risk of accidents.

[0047] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0048] (1) Normalize the device abnormal feature data of multiple power distribution network inspection nodes to obtain normalized feature data, and perform outlier detection on the normalized feature data to identify and remove outliers to obtain effective feature data;

[0049] (2) Perform time series annotation processing on the effective feature data to obtain feature data with timestamps, and perform spatial coordinate mapping processing on the feature data with timestamps to obtain spatio-temporal mapping feature data;

[0050] (3) Perform multi-dimensional data fusion processing on the spatio-temporal mapping feature data through a weighted fusion algorithm to obtain a fusion feature set, and perform clustering processing based on similarity analysis on the fusion feature set to obtain clustering feature data;

[0051] (4) Perform principal component analysis processing on the clustering feature data to extract key principal components to obtain principal component feature data, and perform time series correlation analysis on the principal component feature data to obtain time series correlation feature data;

[0052] (5) Perform multi-dimensional data visualization processing on the time series correlation feature data to obtain a visualization result, and perform interactive analysis on the visualization result to obtain comprehensive device abnormal data with spatio-temporal correlation.

[0053] Specifically, normalizing the device anomaly feature data for multiple inspection nodes is a crucial step in ensuring data consistency and comparability. The goal of normalization is to transform feature data with different dimensions and ranges into the same scale. This process ensures that the feature data is within the range of [0, 1], facilitating subsequent analysis and comparison. After completing the normalization process, the subsequent step is to detect outliers in the normalized feature data. Common outlier detection methods include the Z-score method and the IQR method (Interquartile Range method). Using the Z-score method, the Z-score of each data point is calculated, with the formula: Through this method, if the Z-score of the temperature data point of 65 degrees Celsius for a certain node is 4, then this data will be identified as an outlier and removed. After outlier detection, the resulting valid feature data will be more accurate and reliable, facilitating subsequent analysis.

[0054] After processing the valid feature data, time series annotation processing is carried out. This step is achieved by adding timestamps to each data point, and the timestamps provide context information on how the data changes over time. For example, if the time of the temperature data point of 65 degrees Celsius is 12:00 on September 30, 2024, then the labeled feature data will be (65, 2024-09-30 12:00) after annotation. This form of time series data will facilitate subsequent time series analysis and visualization. After the time series data annotation is completed, the subsequent step is to perform spatial coordinate mapping processing on the feature data with timestamps. This process realizes the visualization of the spatial distribution of data by associating each data point with corresponding spatial coordinates (such as longitude and latitude). For example, by combining the data point (65, 2024-09-30 12:00) with the coordinates (37.7749°N, 122.4194°W), the mapped feature data obtained is ((65,2024-09-30 12:00), (37.7749, -122.4194)). This processing combines the spatio-temporal characteristics of the data, contributing to the subsequent analysis of the device's performance at different times and locations.

[0055] Multidimensional data fusion processing is performed on the spatio-temporal mapping feature data through a weighted fusion algorithm to generate a fusion feature set. The weighted fusion algorithm assigns different weights to each feature data to make full use of the advantages of various types of data. Assuming the weight of the temperature feature is 0.6 and the weight of the humidity feature is 0.4, the fusion feature can be calculated by the following formula:

[0056] ;

[0057] where, is the temperature feature, is the humidity feature, and is the corresponding weight. If the temperature at a certain moment is 65 degrees Celsius and the humidity is 70%, the fused feature is:

[0058] ;

[0059] The fused feature set will provide a more comprehensive perspective for subsequent clustering analysis. Furthermore, clustering processing based on similarity analysis is performed on the fused feature set. The clustering algorithm (such as the K-means algorithm) is used to group similar features. The set K value is 3. If the clustering result shows that the feature set is divided into three categories, representing different device states, it can provide more targeted analysis for device fault diagnosis.

[0060] Principal component analysis (PCA) is performed on the clustering feature data to extract key principal components to obtain principal component feature data. PCA transforms the original feature data into a new feature space through linear transformation, reducing the dimension while retaining the main features of the data. The set number of principal components is 2. If the original data is 4-dimensional, after PCA transformation, the principal component data will be 2-dimensional. This principal component feature data provides key information to help understand the overall operating state of the device.

[0061] Finally, time-series correlation analysis is performed on the principal component feature data to obtain time-series correlation feature data. This step identifies potential trends and periodic patterns by analyzing the relationships between principal component features at different time points. If the analysis results show that a certain feature appears frequently within a specific time period, this will help maintenance personnel identify potential risks of device anomalies. After completing all processing, multi-dimensional data visualization is performed to display the analysis results graphically to obtain visualization results. This visualization effect helps maintenance personnel intuitively understand the operating state and abnormal conditions of the device, facilitating subsequent decision-making. At the same time, interactive analysis is carried out, enabling users to freely explore and analyze data as needed, thereby obtaining comprehensive data on device anomalies with spatio-temporal correlation. Through the above methods, the distribution network inspection management method can effectively improve the detection and analysis capabilities of device anomalies and ensure the stable and safe operation of the distribution network.

[0062] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0063] (1) Perform time-series recombination processing on the comprehensive device anomaly data to obtain the recombined time-series data, and perform data smoothing processing on the recombined time-series data to obtain the smoothed time-series data;

[0064] (2) Perform trend analysis processing on the smoothed time-series data to identify long-term trends and seasonal fluctuations to obtain trend feature data, and perform timeliness evaluation on the trend feature data to determine the timeliness evaluation result;

[0065] (3) Dynamically track and process the timeliness evaluation results through a spatio-temporal tracking algorithm to obtain dynamically changing feature data, and identify potential abnormal patterns by recognizing abnormal patterns in the dynamically changing feature data;

[0066] (4) Conduct spatial clustering analysis on the potential abnormal patterns to obtain clustering results, and perform adjacent node correlation analysis on the clustering results to identify the correlation between nodes and obtain correlation feature data;

[0067] (5) Perform multi-dimensional visualization display on the correlation feature data, combining the time axis and spatial distribution to obtain device abnormal spatio-temporal change data.

[0068] Specifically, by performing time series recombination processing on the comprehensive device abnormal data, the abnormal states of the device at each time point can be rearranged in chronological order, thus forming a set of continuous time series data. This process helps to capture the performance of the device at different time periods and provides a more systematic perspective. After obtaining the original data, data smoothing technology is used to remove short-term random fluctuations to obtain smoothed time series data, which further lays the foundation for trend analysis. The smoothed time series data can identify long-term trends and seasonal fluctuations after being processed. By analyzing these data, the operating state of the device over a long period of time can be judged. For example, it is found that the temperature of a certain device generally shows an upward trend within a year, and the temperature fluctuation intensifies in a specific season, such as summer. This kind of analysis can not only provide the historical background of the device state but also provide a reference for future maintenance.

[0069] After obtaining the trend feature data, the process of timeliness evaluation is to determine whether these data are reliable. This step usually includes evaluating the integrity and consistency of the data to confirm that the data used in the analysis truly reflects the real state of the device. If the data is continuous and there are no obvious missing values, it can be judged as valid data, which is crucial for subsequent dynamic tracking processing. Through the spatio-temporal tracking algorithm, the timeliness evaluation results are dynamically tracked and processed with the aim of obtaining the dynamic feature data of the device state change. This algorithm can combine time series and spatial information to monitor the change of the device operating state in real time. If it is found that the temperature of a certain device continues to rise within a certain period of time and shows an abnormality compared with the historical data, potential device problems can be quickly identified and measures can be taken in a timely manner.

[0070] Furthermore, the identified potential abnormal patterns will enter the spatial clustering analysis stage. Clustering analysis can help find the similarities between different devices, thereby revealing potential correlations. In this process, the distance and similarity between adjacent nodes are used to judge the degree of influence between devices. For example, if a device has an abnormal temperature rise and the temperature of its adjacent device also rises accordingly, there may be a certain correlation between the two. By analyzing these correlation feature data, it is possible to determine which devices may affect each other during operation and adjust the inspection strategy in a timely manner. Finally, the correlation feature data is subjected to multi-dimensional visual display, combined with the time axis and spatial distribution information, to form an intuitive representation of the spatio-temporal change data of device anomalies. This display method enables inspection personnel to quickly identify device anomalies and improves the efficiency of decision-making. Through visualization means such as heat maps and time series diagrams, the abnormal states of different devices at specific time periods are clearly displayed, assisting decision-makers in judging the devices that need to be processed first to ensure the safe operation of the distribution network.

[0071] For example, assume that during a one-month monitoring of a group of devices, the smoothed temperature sequence shows an obvious upward trend. After analysis, it is found that the temperature of a certain device exceeds the preset safety threshold within a specific period. Through the spatio-temporal tracking algorithm, it is found that the temperature of its adjacent devices also rises accordingly before the device shows anomalies. This indicates that there may be a certain correlation between the devices. Based on the clustering analysis results, the inspection team decides to prioritize the inspection and maintenance of this group of devices, thus effectively preventing possible failures.

[0072] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0073] (1) Perform coordinate system conversion processing on the spatio-temporal change data of device anomalies to obtain the converted spatio-temporal data, and perform feature calibration processing on the converted spatio-temporal data to obtain calibrated feature data;

[0074] (2) Perform position inference processing on the calibrated feature data through a machine learning-based spatial positioning algorithm to obtain inferred spatial coordinate data, and perform reliability evaluation on the inferred spatial coordinate data to screen and obtain valid coordinates;

[0075] (3) Perform spatial clustering processing on the valid coordinates to identify the fault hot spot areas, obtain hot spot coordinate data, and perform density analysis on the hot spot coordinate data to determine the dense areas where faults occur;

[0076] (4) Based on historical fault data, perform fuzzy logic processing on the dense areas to obtain fuzzy decision feature data, and perform precise positioning processing on the fuzzy decision feature data to obtain the spatial coordinate data of device faults.

[0077] Specifically, it is crucial to perform coordinate system conversion processing on the spatio-temporal change data during equipment anomalies. This process aims to standardize the data in different coordinate systems to obtain the converted spatio-temporal data. This conversion usually requires referring to the geographical location and spatial layout of the equipment so that data from different sources can be analyzed in the same coordinate system. After completing the coordinate system conversion, feature calibration processing is carried out, which is a process of accurately identifying the characteristic attributes of the equipment to obtain calibrated feature data. The calibrated feature data integrates information such as the operating parameters of the equipment and historical fault records, making it comparable and analyzable, laying a foundation for subsequent data processing. Based on the calibrated feature data, a spatial positioning algorithm based on machine learning is applied for position inference processing. This algorithm uses the known calibrated feature data and infers the actual spatial coordinate data of the equipment by analyzing the relationships between the data. During the inference process, multi-dimensional information such as the working state of the equipment, the surrounding environment, and historical fault data is usually combined to improve the accuracy of the inference. After completing the position inference, it is necessary to conduct a reliability assessment. By setting thresholds and criteria, valid coordinates are screened out to ensure the accuracy and reliability of subsequent analysis. Valid coordinates refer to those data points that have been verified multiple times and have high accuracy, and these data points will provide a basis for subsequent clustering analysis.

[0078] Spatial clustering processing is performed on the valid coordinates to identify fault hotspots. Clustering analysis is a process of calculating the distances between data points and finding data points with high similarity in the dataset. During this process, an algorithm is used to divide the valid coordinates into different groups, and each group represents a potential fault hotspot area. Through clustering analysis, hotspot coordinate data can be obtained, which reflects the areas where faults frequently occur during a certain period. On this basis, density analysis further helps to determine the dense areas where faults occur. By calculating the ratio of the number of faults to the number of equipment in each hotspot area, it can be judged which areas have a higher fault risk. When performing fuzzy logic processing on the dense areas, combined with historical fault data, fuzzy decision-making feature data can be obtained. Fuzzy logic processing allows reasoning and decision-making in the presence of uncertainty and ambiguity, and can take into account various possibilities and influencing factors of equipment faults. On this basis, precise positioning processing is carried out to obtain the spatial coordinate data of equipment faults. These spatial coordinate data will be provided to the inspection team so that high-risk areas can be given priority attention in subsequent inspection and maintenance work.

[0079] For example, within a specific time period, by analyzing the abnormal data of distribution network equipment, the spatio-temporal change data of 1,000 abnormal events were obtained. After coordinate system transformation and calibration of these data, machine learning algorithms were used to infer the spatial coordinates corresponding to these abnormal events, and 800 valid coordinates were obtained. Through cluster analysis, several hot spots were identified. For example, within a certain area, fault events occurred frequently with a density reaching 20%. Combining historical fault data, after performing fuzzy logic processing on the dense area, it was determined that there was a 70% probability that a fault would occur again in this area within a future period of time. At this time, the inspection team could prioritize the inspection of this area to reduce the economic losses caused by potential equipment failures.

[0080] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0081] (1) Perform feature extraction processing on the spatial coordinate data of equipment failures to obtain fault feature data;

[0082] (2) Perform risk scoring processing on the fault feature data through a risk assessment model, calculate the risk scores of each equipment failure to obtain risk scoring data, and perform sorting processing on the risk scoring data to obtain sorted risk data;

[0083] (3) Perform weight assignment processing on the sorted risk data, combine equipment importance and fault impact degree to obtain weighted risk data, and perform abnormal level classification processing on the weighted risk data to obtain preliminary abnormal level information;

[0084] (4) Perform decision tree analysis processing on the preliminary abnormal level information to obtain a decision tree model, and re-evaluate the abnormal level through the decision tree model to obtain equipment abnormal priority information.

[0085] Specifically, the extraction and processing of fault feature data are carried out. Feature extraction is the process of identifying key attributes that affect fault judgment from the original data. In this stage, the spatial coordinate data of equipment faults are analyzed to identify multiple features, such as the time of fault occurrence, frequency, equipment type, environmental factors, and the correlation with other equipment. These features together constitute the fault feature data, laying a foundation for subsequent risk assessment. Furthermore, a risk assessment model is used to perform risk scoring on the fault feature data. The risk assessment model analyzes the extracted feature data, combines known fault histories and influencing factors, and calculates the risk score for each equipment fault. This process involves the calculation of multiple parameters, including the importance of the equipment, the probability of fault occurrence, and the impact of the fault on the system. Through the weighted calculation of these parameters, the risk scoring data for each fault are finally obtained. The scoring range is usually between 0 and 100, where 0 indicates no risk and 100 indicates extremely high risk. After completing the risk scoring, the risk data are sorted from high to low to obtain the sorted risk data. This sorting process helps to quickly identify the equipment that requires the most attention, ensuring the rational allocation of resources. The next key step is to perform weight assignment on the sorted risk data. In this process, the risk scores of each equipment are weighted in combination with the importance of the equipment and the degree of fault impact. The importance of the equipment is usually based on its role in the distribution network, such as a key transformer or distribution line, whose fault may cause power outages or equipment damage over a larger area. At the same time, the degree of fault impact takes into account the impact of the fault on surrounding equipment and the system. After weight assignment, the weighted risk data are obtained, which not only reflect the risk scores but also show the importance of the equipment in the system and the possible consequences of the fault.

[0086] After obtaining the weighted risk data, the abnormal level division process is carried out. By setting thresholds, the weighted risk data are divided into multiple abnormal levels, such as "Normal", "Attention", "Warning", and "Danger". This division helps to quickly identify the abnormal state of the equipment, enabling timely maintenance and inspection. Furthermore, based on the preliminary abnormal level information, decision tree analysis is carried out. Decision tree is a popular machine learning method used to extract decision rules from data. In this process, the preliminary abnormal level is used as input to construct a decision tree model. By analyzing historical data, it is determined which features are the most important for the judgment of different abnormal levels. Finally, the decision tree model re-evaluates the abnormal levels to obtain the equipment abnormal priority information. This priority information will provide strong guidance for the inspection team to ensure that high-risk equipment can be repaired first.

[0087] For example, assume that after feature extraction of the spatial coordinate data of a certain equipment failure, it is obtained that the equipment failure frequency is 5 times / year, the failure type is insulation failure, and the failure occurs in a high-temperature environment. Based on these features, the risk assessment model calculates a risk score of 80 for this equipment, and after sorting, it ranks third among all equipment. According to the importance of the equipment, assume that the importance score of this equipment is 0.9 and the failure impact score is 0.8, then the weighted risk data is:

[0088] ;

[0089] Next, the weighted risk data is divided into abnormal levels, and the set threshold is 120. Finally, the abnormal level of this equipment is classified as "dangerous". Subsequently, through decision tree analysis, it is found that there is a strong correlation between the insulation failure of this equipment and temperature, humidity, and service life. Finally, the equipment abnormal priority information will show that this equipment has the highest priority among all equipment, and the inspection team will give priority to checking and maintaining it.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the scope of the claims of the present invention.

Claims

1. A distribution network inspection management method, characterized in that: include: Collect and preprocess multi-source data of multiple distribution network inspection nodes to obtain multi-dimensional image data after denoising; Performing feature extraction processing on the multi-dimensional image data by using an image recognition algorithm to obtain equipment abnormality feature data of multiple distribution network inspection nodes; Perform data fusion processing on the equipment abnormality feature data of multiple distribution network inspection nodes to obtain the comprehensive equipment abnormality data with temporal and spatial correlation; Continuously analyzing and processing the equipment abnormality comprehensive data through a spatiotemporal tracking algorithm to obtain equipment abnormality spatiotemporal variation data; The step of performing continuous analysis and processing on the equipment abnormality comprehensive data by using a spatiotemporal tracking algorithm to obtain equipment abnormality spatiotemporal change data includes: performing time series reorganization processing on the equipment abnormality comprehensive data to obtain reorganized time series data, and performing data smoothing processing on the reorganized time series data to obtain smoothed time series data; performing trend analysis processing on the smoothed time series data to identify long-term trends and seasonal fluctuations to obtain trend feature data, and performing timeliness evaluation on the trend feature data to determine the timeliness evaluation result; performing dynamic tracking processing on the timeliness evaluation result by using a spatiotemporal tracking algorithm to obtain dynamic change feature data, and performing abnormal pattern recognition on the dynamic change feature data to identify potential abnormal patterns; performing spatial clustering analysis on the potential abnormal patterns to obtain clustering results, and performing adjacent node association analysis on the clustering results to identify the association between nodes to obtain association feature data; performing multi-dimensional visualization display on the association feature data, combining the time axis and spatial distribution to obtain equipment abnormality spatiotemporal change data; Intelligently positioning the abnormal spatiotemporal change data of the equipment to obtain spatial coordinate data of equipment failure; Based on the priority sorting algorithm, the equipment fault spatial coordinate data is evaluated for abnormality level to obtain equipment abnormality priority information.

2. The inspection management method for distribution network according to claim 1, characterized in that: The step of collecting and preprocessing multi-source data of multiple distribution network inspection nodes to obtain multi-dimensional image data after denoising includes: Performing data acquisition processing on multi-source data of multiple distribution network inspection nodes to obtain original image data, and performing format conversion processing on the original image data to obtain image data in a unified format; Performing grayscale processing on the image data in the unified format to obtain grayscale image data, and performing normalization processing on the grayscale image data to obtain standardized image data; Performing denoising processing on the standardized image data by using a wavelet transform algorithm to obtain denoised image data, and performing edge smoothing processing on the denoised image data to obtain smoothed image data; The smoothed image data is subjected to feature enhancement processing to obtain enhanced feature image data, and the enhanced feature image data is subjected to image reconstruction processing to obtain multi-dimensional image data after denoising processing.

3. The inspection management method for distribution network according to claim 1, characterized in that: The step of performing feature extraction processing on the multi-dimensional image data by using an image recognition algorithm to obtain device abnormality feature data of multiple distribution network inspection nodes includes: Performing feature preprocessing on the multidimensional image data to obtain processed image data, and performing feature enhancement processing on the processed image data to obtain enhanced image feature data; Performing region segmentation processing on the enhanced image feature data to obtain segmented feature region data, and performing shape description processing on the segmented feature region data to obtain shape feature data; Performing feature extraction processing on the shape feature data through a convolutional neural network to obtain a preliminary feature set, and performing feature selection processing on the preliminary feature set to obtain an optimized feature set; Performing abnormal pattern recognition processing on the optimized feature set to obtain device abnormal pattern data, and performing classification processing on the device abnormal pattern data to obtain classification feature data; The classified feature data is processed for abnormal feature identification to obtain equipment abnormal feature data of multiple distribution network inspection nodes.

4. The inspection management method for distribution network according to claim 1, characterized in that: The step of performing data fusion processing on the equipment abnormality feature data of multiple distribution network inspection nodes to obtain the time-space associated equipment abnormality comprehensive data includes: Normalizing the abnormal feature data of the equipment of multiple distribution network inspection nodes to obtain normalized feature data, and performing outlier detection on the normalized feature data to identify and remove outliers to obtain valid feature data; Performing time series labeling processing on the effective feature data to obtain feature data with timestamps, and performing space coordinate mapping processing on the feature data with timestamps to obtain spatiotemporal mapping feature data; Performing multi-dimensional data fusion processing on the spatiotemporal mapping feature data by a weighted fusion algorithm to obtain a fused feature set, and performing clustering processing on the fused feature set based on similarity analysis to obtain cluster feature data; Performing principal component analysis on the clustering feature data to extract key principal components to obtain principal component feature data, and performing time series correlation analysis on the principal component feature data to obtain time series correlation feature data; Multi-dimensional data visualization processing is performed on the time series correlation feature data to obtain visualization results, and interactive analysis is performed on the visualization results to obtain comprehensive equipment anomaly data with time and space correlation.

5. The inspection management method for distribution network according to claim 1, characterized in that: The step of performing intelligent positioning processing on the abnormal spatiotemporal change data of the equipment to obtain spatial coordinate data of equipment failure includes: Performing coordinate system conversion processing on the abnormal spatiotemporal change data of the equipment to obtain converted spatiotemporal data, and performing feature calibration processing on the converted spatiotemporal data to obtain calibrated feature data; Performing position inference processing on the calibration feature data through a spatial positioning algorithm based on machine learning to obtain inferred spatial coordinate data, and performing reliability assessment on the inferred spatial coordinate data to screen and obtain valid coordinates; Performing spatial clustering processing on the effective coordinates to identify fault hotspot areas, obtaining hotspot coordinate data, and performing density analysis on the hotspot coordinate data to determine dense areas where faults occur; Based on the historical fault data, fuzzy logic processing is performed on the dense area to obtain fuzzy decision feature data, and precise positioning processing is performed on the fuzzy decision feature data to obtain equipment fault spatial coordinate data.

6. The inspection management method for distribution network according to claim 5, characterized in that: The step of performing abnormality level evaluation processing on the spatial coordinate data of the equipment fault based on the priority sorting algorithm to obtain equipment abnormality priority information includes: Performing feature extraction processing on the spatial coordinate data of the equipment fault to obtain fault feature data; Performing risk scoring processing on the fault feature data through a risk assessment model, calculating the risk score of each equipment fault to obtain risk scoring data, and sorting the risk scoring data to obtain sorted risk data; Performing weight distribution processing on the ranked risk data, combining the importance of equipment and the impact of failure to obtain weighted risk data, and performing abnormality level classification processing on the weighted risk data to obtain preliminary abnormality level information; The preliminary abnormality level information is subjected to decision tree analysis to obtain a decision tree model, and the abnormality level is re-evaluated through the decision tree model to obtain equipment abnormality priority information.

Images