Method and system for quickly retrieving and matching inspection images of power distribution network

Abstract

The invention relates to the technical field of power grid image retrieval, and discloses a power distribution network inspection image rapid retrieval matching method and system, and the method comprises the steps: obtaining a to-be-retrieved inspection image of power distribution network equipment, and extracting the equipment structure features of the inspection image through a hierarchical convolutional network; quantifying a surface texture attenuation index of the power distribution network equipment through fractal dimension based on the equipment structure characteristics; performing mapping relation coupling on the surface texture attenuation index and a space coordinate of an equipment connecting piece to generate a dynamic feature coding sequence containing an equipment structure topological relation; performing time sequence consistency matching on the dynamic feature coding sequence and a pre-constructed reference image library, and aligning an equipment aging track through a dynamic time warping algorithm to generate a similarity sorting result; according to the method, the problems that effective features cannot be extracted during retrieval matching and the retrieval precision is low are solved.

Description

Technical Field

[0001] This invention relates to the field of power grid image retrieval technology, and in particular to a method and system for rapid retrieval and matching of distribution network inspection images. Background Technology

[0002] Intelligent retrieval of distribution network inspection images relies on static feature comparison and manual interpretation strategies. At the feature extraction level, it is impossible to quantify the progressive structural degradation differences of equipment surface textures, and there is a lack of quantitative evaluation models for attenuation modes such as local aging cracks and rust diffusion. At the same time, it is impossible to correlate the spatial location correlation of equipment connectors in the entire power grid topology, resulting in the extracted features being disconnected from the actual physical connection status of the equipment.

[0003] Secondly, existing retrieval technologies treat equipment aging as a fixed-time slice of its state, ignoring the continuous evolution of the aging process over time. This leads to misalignment in the time scale of historical defect sample retrieval. Furthermore, current feature representation schemes do not embed equipment spatial topology information, making it impossible for matching results to reflect the fault propagation risk level of a specific connection point's degradation within the power grid. Consequently, when associating historical defect records, old defects in non-critical nodes of the same type of equipment are often mistakenly identified as high-priority alarms, severely reducing the accuracy of maintenance decisions. Summary of the Invention

[0004] This invention provides a method and system for rapid retrieval and matching of images from power distribution network inspections. Its main purpose is to solve the problems of inability to extract effective features and low retrieval accuracy during retrieval and matching.

[0005] To achieve the above objectives, the present invention provides a method for rapid retrieval and matching of images from power distribution network inspections, the method comprising: S1. Obtain the inspection image of the distribution network equipment to be retrieved, and extract the equipment structure features of the inspection image through a hierarchical convolutional network; S2. Based on the structural characteristics of the equipment, the surface texture attenuation index of the power distribution network equipment is quantified by fractal dimension; S3. The surface texture attenuation index is coupled with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure; S3. Perform time-series consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library, and align the device aging trajectory using a dynamic time warping algorithm to generate a similarity ranking result; S4. Based on the similarity ranking results, associate the historical defect records and location information of the power distribution network equipment.

[0006] Preferably, acquiring the inspection image of the distribution network equipment to be retrieved includes: The original inspection image dataset of the power distribution network equipment is captured based on the configured imaging equipment; The original inspection image dataset is filtered, and the filtering results are stored to obtain the inspection images to be retrieved.

[0007] Preferably, the step of extracting the device structural features of the inspection image through a hierarchical convolutional network includes: Separate the shape and contour features of the preliminary features in the inspection image; The equipment structure features of the inspection image are constructed based on the shape contour features and the preliminary features.

[0008] Preferably, based on the structural characteristics of the equipment, the surface texture attenuation index of the power distribution network equipment is quantified through fractal dimensions, wherein the calculation formula for the surface texture attenuation index is:

[0009] in: As a surface texture attenuation index, For local texture partitioning, For the set of all partitions, for The fractal dimension value, As a reference fractal dimension constant, As a texture stability weighting factor, for The gradient amplification in the fractal dimension, Find the maximum value in R.

[0010] Preferably, the formula for calculating the amplification of the fractal dimension gradient is:

[0011] in: for The gradient amplification in the fractal dimension, for The neighborhood set, This is the average fractal dimension value. For neighborhood partitioning index, Neighborhood partitioning The fractal dimension value.

[0012] Preferably, the step of mapping and coupling the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure includes: By binding the surface texture attenuation index with the spatial coordinates of the power distribution network equipment, the location attenuation relationship of the power distribution network equipment can be obtained; The location attenuation relationship is mapped to the topology of the power distribution network equipment to generate a topological adjacency list with enhanced features; Decode the sequence of the topological adjacency list to obtain the dynamic feature encoding sequence.

[0013] Preferably, the step of performing temporal consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library includes: Based on the aging period of the dynamic feature coding sequence, the benchmark image library is divided into a benchmark partition subset; Calculate the multi-sequence similarity between the dynamic feature encoding sequence and the baseline partition subset to obtain an initial similarity matrix; The dynamic time warping path of the initial similarity matrix is ​​taken as the optimal bend path; The cumulative distance of the optimal set of bend paths is normalized to obtain a preliminary matching result for temporal consistency.

[0014] Preferably, the step of aligning the device aging trajectory using a dynamic time warping algorithm to generate a similarity ranking result includes: Based on the preliminary matching results, the key degradation events of the inspection image are identified; Obtain the weight factors of the topology propagation nodes in the key degradation events; The weighting factors and the preliminary matching results are fused according to the timeline to obtain the final similarity score for spatiotemporal alignment. The final similarity scores are then ranked based on risk transmission to generate a similarity ranking result.

[0015] Preferably, the step of associating the historical defect records and location information of the power distribution network equipment based on the similarity ranking result includes: Extract the high-order sequence of the similarity ranking results to obtain the target device identifier set; Batch query the defect records of the target device identifier set to obtain a historical defect information table; By associating the historical defect information table with the spatial coordinates, a historical binding dataset of defect-location is obtained; Based on the priority of the historical binding dataset and the similarity ranking results, a final association report is generated.

[0016] A rapid image retrieval and matching system for power distribution network inspection, the system comprising: The structural feature module is used to acquire inspection images of power distribution network equipment to be retrieved, and extract the equipment structural features of the inspection images through a hierarchical convolutional network. Attenuation index generation module. Used to quantify the surface texture attenuation index of the power distribution network equipment based on the equipment's structural characteristics and through fractal dimensions. The encoding generation module is used to couple the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure. The similarity ranking module is used to perform temporal consistency matching between the dynamic feature encoding sequence and the pre-built benchmark image library, and to align the device aging trajectory through a dynamic time warping algorithm to generate similarity ranking results. The positioning module is used to associate the historical defect records and positioning information of the power distribution network equipment based on the similarity ranking results.

[0017] Beneficial effects: 1. By extracting device structural features through hierarchical convolutional networks and combining them with fractal dimensions to quantify surface texture attenuation indicators, the progressive structural degradation differences in device surface textures, such as localized aging cracks and rust diffusion, can be accurately captured, overcoming the limitation of traditional methods in quantifying such differences. Simultaneously, by coupling the surface texture attenuation indicators with the spatial coordinate mapping of device connectors, a dynamic feature encoding sequence containing topological relationships is generated. This ensures that the extracted features are closely related to the actual physical connection state of the device, solving the problem of feature decoupling from the physical connection state in existing technologies.

[0018] 2. By aligning equipment aging trajectories using a dynamic time warping algorithm, the continuous evolution of the aging process over time is fully considered, effectively avoiding time scale misalignment in historical defect sample retrieval. Furthermore, by embedding equipment spatial topology information into the feature representation scheme, the matching results accurately reflect the fault propagation risk level of specific connection point degradation in the power grid. When associating historical defect records, it avoids mistaking old defects of non-critical nodes of the same type of equipment as high-priority alarms, significantly improving the accuracy of operation and maintenance decisions and enabling rapid and accurate retrieval and matching of distribution network inspection images. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for rapid image retrieval and matching in power distribution network inspection according to an embodiment of the present invention. Figure 2 This is a functional module diagram of a power distribution network inspection image rapid retrieval and matching system provided in an embodiment of the present invention. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] This application provides a method for rapid retrieval and matching of distribution network inspection images. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for rapid retrieval and matching of distribution network inspection images can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0022] Reference Figure 1 The diagram shown is a flowchart illustrating a fast image retrieval and matching method for power distribution network inspections according to an embodiment of the present invention. In this embodiment, the fast image retrieval and matching method for power distribution network inspections includes: S1. Obtain the inspection images of the distribution network equipment to be retrieved, and extract the equipment structural features of the inspection images through a hierarchical convolutional network.

[0023] In this embodiment, acquiring the inspection image of the distribution network equipment to be retrieved includes: The original inspection image dataset of the power distribution network equipment is captured based on the configured imaging equipment; The original inspection image dataset is filtered, and the filtering results are stored to obtain the inspection images to be retrieved.

[0024] Specifically, based on the raw inspection image dataset of distribution network equipment captured by the configured imaging equipment, in the actual operation scenario of the distribution network, the imaging equipment is deployed at locations such as substations and transmission line towers to collect images of distribution network equipment such as transformers, circuit breakers, and insulators. These devices operate in outdoor open-air environments and may be affected by physical factors such as light intensity, weather conditions, and equipment installation angles. The imaging equipment needs to be able to adapt to different environments to ensure that the captured raw inspection images can clearly reflect the actual condition of the equipment.

[0025] Specifically, due to the wide distribution of power distribution network equipment, the number of raw images collected by imaging equipment is enormous, and environmental factors may cause image blurring, noise interference, or partial equipment obstruction. Therefore, it is necessary to screen the raw images in the physical storage environment. The screening process involves manual visual inspection or automatic detection through image quality assessment algorithms to remove images that do not meet the requirements, retaining clear, complete images that accurately reflect the characteristics of the equipment as the data to be retrieved.

[0026] Imaging equipment parameters include resolution, focal length, aperture, and shutter speed. For example, high-resolution cameras can capture finer structural features and surface textures of equipment. In power distribution network inspections, imaging equipment typically needs to meet certain resolution standards to clearly identify defects such as cracks and corrosion.

[0027] In detail, the focal length can be adjusted according to the distance of the device's installation location to ensure that the device can produce a clear image; the aperture and shutter speed settings need to be adapted to different lighting conditions, such as increasing the aperture and extending the shutter speed in low-light environments to obtain an image with sufficient brightness.

[0028] The parameters of the original inspection image dataset include the image format, size, shooting time, and shooting location. The shooting time and location parameters are of great significance for subsequent analysis in combination with the equipment aging timeline and spatial topology, and can be used to record the status of the equipment at a specific time and location.

[0029] In this embodiment, the extraction of device structural features from the inspection image using a hierarchical convolutional network includes: Separate the shape and contour features of the preliminary features in the inspection image; The equipment structure features of the inspection image are constructed based on the shape contour features and the preliminary features.

[0030] Specifically, a hierarchical convolutional network is used to perform feature extraction operations on the inspection images. First, preliminary features are distinguished from the original image data through filtering, activation, and other operations of the convolutional layers. Then, shape and contour features are selectively separated—focusing on the edges and contour structures of the physical form of the equipment, such as identifying the arc-shaped contour of the insulator string and the rectangular contour of the switch cabinet door.

[0031] In detail, images from power distribution network inspections are greatly affected by the environment, and the network needs to adapt to these variables. For example, the kernel size parameter needs to be balanced: small kernels preserve the details of the equipment outline, while large kernels capture the overall outline. Specifically, illumination parameters affect the contour extraction effect. The network can eliminate illumination interference under different weather conditions through normalization operations, so that the contour features can be stably separated.

[0032] In detail, the installation angle of the device will distort the contour. The network adapts to contour extraction under different viewpoints through feature mapping of multiple convolutions, ensuring that the real shape contour can be separated even if the device is tilted.

[0033] In detail, the separated shape and contour features are fused with the preliminary features. For example, by feature splicing, point-by-point addition, or attention mechanism weighting, the contour and detailed information are combined to construct the complete device structural features.

[0034] In detail, the structures of power distribution network equipment are diverse, and the weight parameters for feature fusion need to be adapted accordingly. For example, for equipment with simple structures that rely on contour recognition, the weight of contour features should be increased; for equipment with complex structures that require detailed assistance, the proportion of preliminary feature fusion should be increased.

[0035] In detail, environmental noise can interfere with feature construction. The network uses dropout layers, regularization, and other parameters to suppress noisy features and ensure that the fused structural features focus on the device itself.

[0036] In detail, the background of images in different inspection scenarios varies greatly. During fusion, spatial attention parameters are used to allow the network to ignore irrelevant backgrounds and only strengthen the construction of the device's own structural features, thus adapting to the feature extraction needs in complex physical environments.

[0037] S2. Based on the structural characteristics of the equipment, the surface texture attenuation index of the power distribution network equipment is quantified by fractal dimension.

[0038] In this embodiment, based on the structural features of the device, the surface texture attenuation index of the power distribution network equipment is quantified by fractal dimension, wherein the calculation formula for the surface texture attenuation index is:

[0039] in: As a surface texture attenuation index, For local texture partitioning, For the set of all partitions, for The fractal dimension value, As a reference fractal dimension constant, As a texture stability weighting factor, for The gradient amplification in the fractal dimension, Find the maximum value in R.

[0040] Specifically, The surface texture attenuation index is used to quantify the degree of texture degradation caused by environmental factors on the surface of power distribution network equipment. A higher value indicates more severe texture deterioration. Power distribution network equipment is exposed to the outdoors for extended periods, and is affected by salt spray, dust, ultraviolet radiation, etc., causing the surface texture to gradually show wear, cracks, and dirt accumulation. It involves converting these physical degradations into calculable numerical indicators to help assess the risk of equipment aging / failure.

[0041] Local texture partitioning involves dividing inspection images of power distribution network equipment into local blocks based on physical regions. For example, dividing an insulator image into multiple sector-shaped partitions along the circumference, or dividing the surface of a switchgear into small areas using a grid, each... A texture analysis unit corresponding to a local region.

[0042] Surface texture degradation on equipment often starts locally and can be addressed by partitioning. It can accurately locate the starting point of degradation and adapt to the physical characteristics of localized degradation and gradual spread of power distribution network equipment.

[0043] The set of all partitions contains all local texture partitions on the device surface. The set is the basic range for traversing and calculating all partitions, ensuring coverage of the entire surface of the device.

[0044] The surface of power distribution network equipment is a continuous physical whole. As a complete set, it allows the algorithm to statistically analyze texture decay from a global perspective, avoiding the omission of key deterioration areas.

[0045] for The fractal dimension value describes the local texture partition. The fractal dimension is a numerical value indicating the complexity of surface texture. The higher the fractal dimension, the coarser and more irregular the texture, such as the high fractal dimension of severely corroded equipment surfaces.

[0046] The fractal dimension of the texture on healthy surfaces of power distribution network equipment is relatively stable. For example, the surface of a brand-new insulator is smooth and has a low fractal dimension. When cracks, dirt accumulation, or corrosion occur, the texture complexity increases, and the fractal dimension value changes accordingly. This change in physical form can be quantified.

[0047] The reference fractal dimension constant is a baseline value representing the fractal dimension of the surface texture of distribution network equipment under healthy conditions. It is an empirical constant based on statistics from a large number of healthy devices. For example, the reference fractal dimension of a new insulator. It can be calibrated using experimental / historical data.

[0048] Different device types have different health texture characteristics. It needs to be set according to the physical properties such as device type and material, as a baseline for judging whether the texture has decayed.

[0049] Texture stability weighting factors are used to measure local texture partitions. The stability of the texture is indicated by a higher value, which represents a more significant fluctuation or degradation of the texture in that area due to environmental influences. This can be set through training with historical data or by expert experience.

[0050] The environmental stability varies in different areas of the power distribution network. It can reflect the weight of the impact of such physical environment differences on texture attenuation, making the indicators more in line with actual scenarios.

[0051] for The fractal dimension gradient amplification describes the local texture partitioning. The gradient of the fractal dimension relative to adjacent partitions is used to amplify the effects of local degradation, such as a sudden increase in the fractal dimension of a certain partition. This will amplify the weight of this change.

[0052] Faults in power distribution network equipment often begin with local gradient anomalies. It can capture the physical characteristics of such local mutations and provide early warnings of potential failures.

[0053] It calculates local partitions. The absolute value of the difference between the fractal dimension and the health baseline reflects the degree of deterioration of the texture of that partition relative to the healthy state.

[0054] The greater the difference, the more significant the texture attenuation; for example, corrosion cracks cause the fractal dimension to be much higher than that of the texture. .

[0055] Using texture stability weighting factor Adjustments are made to the discrepancies. The more unstable the environment, the lower the actual degree of degradation corresponding to the same fractal dimension difference; in stable environmental regions, the differences are more likely to reflect the true degradation.

[0056] It enhances the abrupt changes in local texture. When the fractal dimension gradient of a certain texture partition is large, even if the absolute value of the difference is small, it will still be due to... Enlarging the area reveals the key region for texture attenuation, adapting to the physical laws governing faults caused by localized sudden changes in power distribution network equipment.

[0057] In this embodiment, the formula for calculating the fractal dimension gradient amplification is:

[0058] in: for The gradient amplification in the fractal dimension, for The neighborhood set, This is the average fractal dimension value. For neighborhood partitioning index, Neighborhood partitioning The fractal dimension value.

[0059] Specifically, for The neighborhood set is related to the local texture partition. The set of all adjacent partitions represents The surrounding physical environment.

[0060] For example, It is a certain piece on the surface of the insulator. The surface texture of the power distribution equipment is continuous, with local partitions, surrounding a ring of adjacent blocks. The degradation will affect / be perceived by the neighborhood. It can reflect this local-neighborhood physical relationship.

[0061] The average fractal dimension value is a local texture partition. Its neighborhood set The average fractal dimension of all partitions represents the baseline of texture complexity for that local region and its surroundings.

[0062] On the surface of healthy power distribution network equipment, the fractal dimension difference between the local area and the surrounding area is small. Stable; when a fault occurs, the local fractal dimension will deviate. This deviation can be used to identify degradation.

[0063] The neighborhood partition index is used to traverse the neighborhood set. The index of each partition facilitates the calculation of the fractal dimension of each neighboring partition.

[0064] Neighborhood partitions are multiple and spatially related. It can traverse in order, ensuring computational coverage. All surrounding adjacent areas, without omitting physically related neighborhoods.

[0065] S3. The surface texture attenuation index is coupled with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure.

[0066] In this embodiment, the step of mapping and coupling the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure includes: By binding the surface texture attenuation index with the spatial coordinates of the power distribution network equipment, the location attenuation relationship of the power distribution network equipment can be obtained; The location attenuation relationship is mapped to the topology of the power distribution network equipment to generate a topological adjacency list with enhanced features; Decode the sequence of the topological adjacency list to obtain the dynamic feature encoding sequence.

[0067] Specifically, the surface texture attenuation index of the equipment is first extracted from the inspection images / data, such as the density of cracks on the surface of the insulator and the grayscale change of the wear texture of the conductor, which reflect the characteristic parameters of equipment aging and failure. At the same time, the spatial coordinates of the distribution network equipment are collected / called, based on GPS positioning, the latitude and longitude corresponding to the tower number, and the three-dimensional coordinates of the equipment installation in the substation.

[0068] Then, through algorithms such as hash mapping and coordinate-index association matrices, the texture attenuation index and spatial coordinates are bound one-to-one, establishing the correlation between device location and texture attenuation degree.

[0069] The physical environment of the power distribution network is complex, with a wide distribution of poles and towers in the field and dense equipment in substations. Spatial coordinate parameters need to be adapted to the accuracy of different scenarios.

[0070] For example, latitude and longitude coordinates are used to ensure accurate positioning of poles in the field, while three-dimensional coordinates are used to distinguish different layers of equipment in switchgear inside substations.

[0071] Texture attenuation index is affected by the environment. For example, the texture of equipment in coastal salt spray areas corrodes quickly, and the texture of insulators in rainy areas is heavily polluted. During the bonding process, environmental correction parameters need to be incorporated to make the positional attenuation relationship more consistent with the actual physical loss. Different equipment types require different texture features and coordinate binding logic. For example, transformers are large and their coordinates are at the center of the base. The texture focuses on the corrosion on the surface of the oil tank, so the binding algorithm parameters need to be adjusted according to the equipment shape to ensure accurate association.

[0072] In detail, the topology of the distribution network equipment is first analyzed to form a basic topological adjacency list. Then, the location attenuation relationship is used as an enhancing feature and embedded into the topological adjacency list.

[0073] Distribution network topology has a hierarchy, including macroscopic transmission network topology and microscopic internal equipment topology. When mapping, it is necessary to distinguish the topology hierarchy parameters.

[0074] In detail, for example at the macro level, we can focus on the attenuation propagation of conductor connections between towers. For instance, aging equipment on one tower may affect adjacent towers.

[0075] At the micro level, focus on the equipment inside the switchgear, such as the impact of circuit breaker contact texture attenuation on adjacent busbars, and use different adjacency weight parameters for different levels.

[0076] Spatial distance in the physical environment affects the strength of topological associations. For example, if two towers are far apart, the equipment attenuation association is weak; if they are close, the association is strong. When mapping, a distance attenuation coefficient parameter is introduced to correct the association weights in the topological adjacency list, so that the enhanced features are more in line with the actual physical connection influence.

[0077] The cascading nature of equipment failures, such as the possibility that an aging and broken insulator may break adjacent conductors, can be addressed by using fault propagation probability parameters in the topological adjacency list to transform the location attenuation relationship into a potential fault association, thereby enhancing the feature's support for identifying cascading fault risks during inspections.

[0078] S3. Perform time-series consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library, and align the device aging trajectory using a dynamic time warping algorithm to generate a similarity ranking result.

[0079] In this embodiment, the step of performing temporal consistency matching between the dynamic feature encoding sequence and the pre-built benchmark image library includes: Based on the aging period of the dynamic feature coding sequence, the benchmark image library is divided into a benchmark partition subset; Calculate the multi-sequence similarity between the dynamic feature encoding sequence and the baseline partition subset to obtain an initial similarity matrix; The dynamic time warping path of the initial similarity matrix is ​​taken as the optimal bend path; The cumulative distance of the optimal set of bend paths is normalized to obtain a preliminary matching result for temporal consistency.

[0080] Specifically, the aging period labels are first extracted from the dynamic feature encoding sequence, such as the timestamps in the sequence and the equipment aging stages corresponding to the decay trends, like mild aging (1-3 years) and severe aging (5-8 years). Then, the baseline image library is traversed, storing historical inspection images and corresponding equipment states. The images are grouped according to the aging period, and baseline images with the same / similar aging stages are grouped into a subset.

[0081] In detail, the aging of power distribution network equipment is affected by the environment. Coastal equipment corrodes quickly, while mountainous areas experience accelerated aging due to large temperature differences. The division of aging periods needs to incorporate environmental correction parameters.

[0082] For example, in coastal areas, mild equipment aging may take 1-2 years, while in mountainous areas it may take 2-4 years. By statistically analyzing the relationship between the environment and aging rate based on historical data, the threshold for dividing the time period can be adjusted.

[0083] In detail, different equipment types have different aging cycles, and equipment type parameters are required when dividing the baseline subset. Transformers age slowly, while insulators age quickly, ensuring that the aging stages of equipment within the subset more closely match the actual physical loss patterns.

[0084] The size of the baseline image library affects the segmentation efficiency. By using block retrieval parameters, the images are first coarsely segmented by geographical location and equipment type, and then subdivided by aging period, which can meet the processing needs of massive inspection images of power distribution networks.

[0085] Dimensional similarity algorithms, such as DTW pre-computation and a fusion of cosine similarity and dynamic time warping, are used to compare feature sequences one by one. The similarity between each encoded sequence and all reference sequences in the subset is organized into a matrix, with rows representing the sequence to be retrieved, columns representing the reference sequences, and values ​​representing the similarity scores, recording the initial matching degree.

[0086] Specifically, the characteristics of power distribution network equipment are affected by environmental noise, and similarity calculation needs to incorporate noise-robust parameters. For example, smoothing filters can be applied to the feature sequences, Gaussian filter parameters can be adapted to different noise intensities, or dynamic time-warped local constraint parameters can be used to ignore small fluctuations in features caused by environmental noise, thus ensuring accurate similarity calculation.

[0087] Different devices have different key feature dimensions. For insulators, the key feature is crack length, while for transformers, it's oil color pattern. When calculating multi-sequence similarity, feature weight parameters are used. Key feature dimensions, such as crack length (0.6%), are given high weight, while secondary features, such as background sky, are given low weight. This makes the similarity more closely match the physical needs of fault identification in power distribution network equipment.

[0088] The number of samples in the baseline subset affects the matrix computation cost. Using batch computation parameters, such as GPU-accelerated matrix operations, can adapt to the fast similarity calculation of large-scale image databases in power distribution networks, thus ensuring retrieval efficiency.

[0089] For each sequence to be retrieved and the reference sequence in the initial similarity matrix, the Dynamic Time Warping (DTW) algorithm is used to find the optimal alignment path of the feature points of the two sequences.

[0090] The aging of power distribution network equipment is non-uniform, such as the abrupt aging of insulators after a lightning strike. The bending window parameters of DTW need to be adapted accordingly. By setting the window size, feature points can be aligned within a certain range, which avoids excessive bending and mismatch with irrelevant events, while capturing sudden environmental impacts, such as abrupt aging caused by lightning strikes and blizzards, thus conforming to the temporal characteristics of actual physical events.

[0091] Different device topologies have varying complexity, such as the coordinated aging of multiple components within a switch cabinet. DTW's path constraint parameters differ. Complex topologies use global constraints, while simple topologies use local constraints, ensuring that the optimal path conforms to both physical connection logic and accurately aligns with aging characteristics.

[0092] In this embodiment, the step of aligning the device aging trajectory using a dynamic time warping algorithm to generate a similarity ranking result includes: Based on the preliminary matching results, the key degradation events of the inspection image are identified; Obtain the weight factors of the topology propagation nodes in the key degradation events; The weighting factors and the preliminary matching results are fused according to the timeline to obtain the final similarity score for spatiotemporal alignment. The final similarity scores are then ranked based on risk transmission to generate a similarity ranking result.

[0093] Specifically, from the preliminary matching results, significant feature changes are filtered out from the historical image-current image matching pairs that have been aligned in time sequence.

[0094] By using preset degradation judgment rules, such as considering crack length >3mm as a degradation event, the events corresponding to these characteristic changes are marked as key degradation events, and their locations are marked in the inspection image, such as using a box selection + label to mark the degradation area of ​​the insulator crack.

[0095] Specifically, the degradation threshold of power distribution network equipment is affected by the environment, and environmental correction parameters need to be introduced when calibrating key degradation events. The degradation judgment threshold should be adjusted according to the location of the equipment to make the calibration more closely reflect the actual physical loss risk.

[0096] Different equipment types have different key degradation characteristics, and the judgment rules are switched using equipment type parameters. For example, for transformers, sudden changes in oil chromatography data are detected; for surge arresters, surface discharge burn textures are detected to ensure accurate identification of key degradation events.

[0097] The similarity threshold of the initial matching results affects event filtering. Setting confidence parameters can filter out false degradation events caused by image noise and mismatches, thereby improving the reliability of calibration.

[0098] Specifically, the distribution network topology is analyzed to locate the topology transmission nodes where key degradation events occur. Using topology influence algorithms, such as graph theory-based node importance calculation, and considering the node's degree in the topology and whether its physical location is a hub node, a weight factor for each transmission node is calculated.

[0099] Specifically, due to differences in distribution network topology levels—between the macroscopic transmission network and the microscopic switchgear—the calculation of weighting factors needs to differentiate between topology level parameters. At the macroscopic level, hub towers have high weights; at the microscopic level, busbar nodes within switchgear have high weights. The calculation logic is adjusted by adjusting the level coefficients.

[0100] Specifically, spatial distance in the physical environment affects the probability of fault propagation, so a distance attenuation parameter is introduced. When calculating the weighting factor, the weights are made to better reflect the actual physical propagation laws.

[0101] Equipment redundancy design affects weights, and redundancy is used to correct parameters. For redundant nodes, their weight factors are reduced because even if the primary node deteriorates, the redundant node can act as a fallback, resulting in a low risk of fault propagation.

[0102] The weighting factors of key degradation events, spatial topological influence, and temporal similarity of the initial matching results are fused together, such as by weighted summation: Final score = Temporal similarity × 0.7 + Weighting factor × 0.3. Simultaneously, the scores are aligned along the timeline so that the fused score reflects both temporal consistency and spatial topological influence, achieving "spatiotemporal alignment."

[0103] This refers to the time-series propagation of faults in the distribution network. For example, icing in winter can lead to insulator deterioration, which may subsequently cause conductor galloping. A time decay parameter is introduced during the integration process. The more recent the historical event, the higher the time decay coefficient, highlighting the reference value of similar recent events; the larger the time difference, the lower the coefficient, adapting to the aging pattern of equipment over time.

[0104] The influence of weighting factors varies under different environments, so the fusion ratio is adjusted using an environment-weighting coefficient. In typhoon areas, the weighting factor ratio is increased to 0.4; in windless areas, it is reduced to 0.2, making the fusion score more closely reflect the regional physical risk.

[0105] The data dimensions of the initial matching results affect the fusion complexity; therefore, the parameters are adapted using dimensions. When matching multiple devices, the weight factors for each device are fused independently, and then aggregated according to topological relationships to ensure accurate scoring in complex scenarios.

[0106] Specifically, based on the final similarity score and combined with a risk transmission model, the risk diffusion path of critical degradation events in the distribution network topology is simulated. Historical images with "high risk and good similarity match" are ranked first, based on a comprehensive sorting of risk diffusion range and final score, generating the final search results.

[0107] The risk tolerance of the distribution network varies depending on the scenario, and scenario risk parameters are incorporated into the ranking. Within substations, the risk transmission weight is increased, prioritizing high-risk matches; in field scenarios, the weight is reduced to adapt to the maintenance priorities of different areas. Historical failure consequences data are used to optimize the parameters of the risk transmission model. Through machine learning, actual consequences are used as labels to iteratively adjust the weight calculation of risk transmission, making the ranking more closely reflect the development patterns of real physical failures. To meet the display requirements of search results, a truncation parameter is set. Only the top N results by score are output, which reduces redundant information while ensuring coverage of high-risk matches, thus adapting to the efficiency requirements of power distribution network on-site operation and maintenance.

[0108] S4. Based on the similarity ranking results, associate the historical defect records and location information of the power distribution network equipment.

[0109] In this embodiment, associating the historical defect records and location information of the power distribution network equipment based on the similarity ranking result includes: Extract the high-order sequence of the similarity ranking results to obtain the target device identifier set; Batch query the defect records of the target device identifier set to obtain a historical defect information table; By associating the historical defect information table with the spatial coordinates, a historical binding dataset of defect-location is obtained; Based on the priority of the historical binding dataset and the similarity ranking results, a final association report is generated.

[0110] Specifically, from the similarity ranking results, the high-ranking portions are extracted; these are the historical images / records that best match the current equipment status and pose the highest risk. Then, from these high-ranking results, the corresponding unique equipment identifiers are extracted, such as pole / tower number + equipment type code, or equipment tag number within the switchgear, and these are compiled into a target equipment identifier set.

[0111] The confidence level requirement for similarity ranking is set, and confidence level filtering parameters are configured. Low-confidence matches are filtered out to prevent invalid device identifiers from being mixed in, thereby improving the efficiency of subsequent association and meeting the needs of the distribution network for accurate fault diagnosis.

[0112] Specifically, the system calls upon the power distribution network defect record database, uses the target equipment identifier set as the query condition, and performs batch retrieval. The historical defects of each piece of equipment found are then organized into a structured table—the Historical Defect Information Table.

[0113] In detail, the distribution of the defect database—whether centralized or distributed—affects the query method, and database adaptation parameters are used. The centralized database uses a unified SQL query; the distributed database uses sharded queries + aggregation, adapting to the physical architecture of defect data storage across multiple regions and sites in the power distribution network.

[0114] To determine the required time span for historical defects, whether to query the past year or the past five years, set a time window parameter. Based on the current level of equipment degradation, query for 5 years for severe degradation and 1 year for mild degradation, limiting the query time range, reducing redundant data, and conforming to the timeline pattern of equipment aging.

[0115] The system records multiple equipment statuses and identifies multiple defects on the same equipment using deduplication / merging parameters. For recurring defects on the same equipment, such as multiple instances of the same type of crack, users can choose to deduplicat or merge them, making the historical defect information table clearer and better suited to the viewing needs of maintenance personnel.

[0116] In detail, the equipment identifier is extracted from the historical defect information table, linked to the distribution network spatial coordinate database, and the defect information is bound to the spatial location. For example, for an insulator with equipment ID A123, the defect is a crack, and the spatial coordinates are (X,Y,Z). A historical dataset of defect-location binding is generated, giving the defect information a physical location attribute.

[0117] The accuracy level of spatial coordinates is determined by accuracy matching parameters during association. For high-precision requirements, millimeter-level coordinates are forcibly associated; for macroscopic analysis, meter-level coordinates are used to adapt to the physical positioning needs of different scenarios.

[0118] When equipment location changes, a timestamp-related parameter is introduced. Based on the defect occurrence time, the spatial coordinates of the corresponding time point are matched to ensure the authenticity of the defect-location binding.

[0119] Multi-source spatial data is fused using data fusion parameters. When coordinates differ between different data sources, a registration algorithm, such as least squares fusion, is used to obtain the optimal coordinate values, ensuring accurate location binding.

[0120] Specifically, the system integrates historical binding datasets and similarity ranking results, and generates a document according to a report template, including basic device information, a list of historical defects, spatial location annotations, and current matching degree analysis. High-priority devices with high similarity, high risk, and numerous historical defects are highlighted to form a final associated report, assisting maintenance personnel in decision-making.

[0121] The need for visualizing risk priorities is addressed by using colors and icons to differentiate risk levels and setting visualization mapping parameters. High-risk equipment is highlighted in red and marked with a fault icon; low-risk equipment is highlighted in green, allowing the report to intuitively reflect the risk distribution of physical equipment in the distribution network. To address the need for comparing historical and current data, trend analysis parameters are introduced. The report automatically calculates and displays trend curves, such as showing an increase in the number of defects on a certain tower from 2 to 5 over the past three years, helping maintenance personnel predict the physical progression of equipment degradation.

[0122] like Figure 2 The diagram shown is a functional module diagram of a power distribution network inspection image rapid retrieval and matching system provided in an embodiment of the present invention.

[0123] The fast image retrieval and matching system 100 for power distribution network inspection described in this invention can be installed in an electronic device. Depending on the functions implemented, the fast image retrieval and matching system 100 may include a structural feature module 101, an attenuation index generation module 102, an encoding generation module 103, a similarity ranking module 104, and a positioning module 105. The module described in this invention can also be called a unit, referring to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0124] In this embodiment, the functions of each module / unit are as follows: Structural feature module 101: used to acquire inspection images of distribution network equipment to be retrieved, and extract the equipment structural features of the inspection images through a hierarchical convolutional network; Attenuation index generation module 102: used to quantify the surface texture attenuation index of the power distribution network equipment based on the equipment structural features and through fractal dimensions; Encoding generation module 103: used to couple the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure; Similarity ranking module 104: used to perform temporal consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library, and to align the device aging trajectory through a dynamic time warping algorithm to generate similarity ranking results; Location module 105: used to associate the historical defect records and location information of the power distribution network equipment with the similarity sorting results.

[0125] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0126] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0127] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0128] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0129] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for fast retrieval and matching of images from power distribution network inspections, characterized in that, The method includes: S1. Obtain the inspection image of the distribution network equipment to be retrieved, and extract the equipment structure features of the inspection image through a hierarchical convolutional network; S2. Based on the structural characteristics of the equipment, the surface texture attenuation index of the power distribution network equipment is quantified by fractal dimension; S3. The surface texture attenuation index is coupled with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure; S4. Perform time-series consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library, and align the device aging trajectory using a dynamic time warping algorithm to generate a similarity ranking result; S5. Based on the similarity ranking results, associate the historical defect records and location information of the power distribution network equipment.

2. The fast retrieval and matching method for distribution network inspection images as described in claim 1, characterized in that, The acquisition of the inspection images of the distribution network equipment to be retrieved includes: The original inspection image dataset of the power distribution network equipment is captured based on the configured imaging equipment; The original inspection image dataset is filtered, and the filtering results are stored to obtain the inspection images to be retrieved.

3. The fast retrieval and matching method for distribution network inspection images as described in claim 1, characterized in that, The extraction of equipment structural features from the inspection image using a hierarchical convolutional network includes: Separate the shape and contour features of the preliminary features in the inspection image; The equipment structure features of the inspection image are constructed based on the shape contour features and the preliminary features.

4. The fast retrieval and matching method for distribution network inspection images as described in claim 1, characterized in that, Based on the structural characteristics of the equipment, the surface texture attenuation index of the power distribution network equipment is quantified through fractal dimensions, wherein the calculation formula for the surface texture attenuation index is: ， in: As a surface texture attenuation index, For local texture partitioning, For the set of all partitions, for The fractal dimension value, As a reference fractal dimension constant, As a texture stability weighting factor, for The gradient amplification in the fractal dimension, Find the maximum value in R.

5. The fast retrieval and matching method for distribution network inspection images as described in claim 4, characterized in that, The formula for calculating the gradient amplification in the fractal dimension is: ， in: for The gradient amplification in the fractal dimension, for The neighborhood set, This is the average fractal dimension value. For neighborhood partitioning index, Neighborhood partitioning The fractal dimension value.

6. The method for rapid retrieval and matching of distribution network inspection images as described in claim 1, characterized in that, The step of coupling the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure includes: By binding the surface texture attenuation index with the spatial coordinates of the power distribution network equipment, the location attenuation relationship of the power distribution network equipment can be obtained; The location attenuation relationship is mapped to the topology of the power distribution network equipment to generate a topological adjacency list with enhanced features; Decode the sequence of the topological adjacency list to obtain the dynamic feature encoding sequence.

7. The method for rapid retrieval and matching of distribution network inspection images as described in claim 1, characterized in that, The step of performing temporal consistency matching between the dynamic feature encoding sequence and the pre-constructed benchmark image library includes: Based on the aging period of the dynamic feature coding sequence, the benchmark image library is divided into a benchmark partition subset; Calculate the multi-sequence similarity between the dynamic feature encoding sequence and the baseline partition subset to obtain an initial similarity matrix; The dynamic time warping path of the initial similarity matrix is ​​taken as the optimal bend path; The cumulative distance of the optimal set of bend paths is normalized to obtain a preliminary matching result for temporal consistency.

8. The fast retrieval and matching method for distribution network inspection images as described in claim 7, characterized in that, The process of aligning device aging trajectories using a dynamic time warping algorithm to generate similarity ranking results includes: Based on the preliminary matching results, the key degradation events of the inspection image are identified; Obtain the weight factors of the topology propagation nodes in the key degradation events; The weighting factors and the preliminary matching results are fused according to the timeline to obtain the final similarity score for spatiotemporal alignment. The final similarity scores are then ranked based on risk transmission to generate a similarity ranking result.

9. The method for rapid retrieval and matching of distribution network inspection images as described in claim 1, characterized in that, The step of associating the historical defect records and location information of the power distribution network equipment based on the similarity ranking results includes: Extract the high-order sequence of the similarity ranking results to obtain the target device identifier set; Batch query the defect records of the target device identifier set to obtain a historical defect information table; By associating the historical defect information table with the spatial coordinates, a historical binding dataset of defect-location is obtained; Based on the priority of the historical binding dataset and the similarity ranking results, a final association report is generated.

10. A rapid image retrieval and matching system for power distribution network inspection, characterized in that, The system includes: The structural feature module is used to acquire inspection images of power distribution network equipment to be retrieved, and extract the equipment structural features of the inspection images through a hierarchical convolutional network. Attenuation index generation module. Used to quantify the surface texture attenuation index of the power distribution network equipment based on the equipment's structural characteristics and through fractal dimensions. The encoding generation module is used to couple the surface texture attenuation index with the spatial coordinates of the device connector to generate a dynamic feature encoding sequence containing the topological relationship of the device structure. The similarity ranking module is used to perform temporal consistency matching between the dynamic feature encoding sequence and the pre-built benchmark image library, and to align the device aging trajectory through a dynamic time warping algorithm to generate similarity ranking results. The positioning module is used to associate the historical defect records and positioning information of the power distribution network equipment based on the similarity ranking results.

Images