Transformer substation inspection method and device, computer equipment, storage medium and computer program product

Abstract

The invention relates to a substation inspection method and device, computer equipment, a storage medium and a computer program product. Relates to the technical field of power inspection. The method comprises the following steps: performing feature extraction on an inspection image of a to-be-inspected substation to obtain a time sequence feature library of electrical equipment, and performing marking of a to-be-supplemented collection area on the inspection image to obtain a to-be-supplemented collection area marking graph; performing spatial registration and state labeling on the original point cloud data according to the feature marks of the electrical equipment included in the time sequence feature library to obtain spatial reference data; carrying out feature matching and construction on the data based on a time sequence feature library to obtain a three-dimensional semantic model; then visual dead zone detection and path planning are carried out to obtain a visual dead zone to be inspected and a collaborative inspection path; and controlling the target unmanned aerial vehicle to obtain the inspection supplementary image, and fusing the inspection supplementary image and the inspection image to obtain an inspection result of the to-be-inspected substation. The method can improve the inspection accuracy of the transformer substation.

Description

Technical Field

[0001] The present application relates to the technical field of power inspection, and particularly to a substation inspection method, device, computer device, storage medium, and computer program product. Background Art

[0002] In the field of power system operation and maintenance, the status monitoring and fault diagnosis of substation equipment are key links to ensure the safe and stable operation of the power grid, and the status monitoring and fault diagnosis of substation equipment generally rely on inspections. Therefore, improving the accuracy of substation inspections has become a problem to be solved.

[0003] Currently, fixed intelligent image terminals are generally used in substations to inspect and monitor equipment. However, due to factors such as terminal layout limitations and fixed shooting angles, there are often visual dead angles, making it difficult to comprehensively obtain equipment status information. In particular, the recognition accuracy of the opening and closing states of key equipment such as switches and disconnectors is relatively low. In addition, attempts have been made to use single unmanned aerial vehicles (UAVs) to assist inspections or use simple three-dimensional reconstruction technology for flight path planning. However, these methods have many deficiencies: First, the single UAV inspection efficiency is low and it is difficult to meet the inspection requirements of large substations. Second, traditional two-dimensional flight path planning methods ignore the complex three-dimensional spatial structure of substations and are prone to safety hazards. Third, existing inspection systems lack a coordination mechanism between fixed terminals and UAVs, making it difficult to effectively integrate multi-source image data. Therefore, existing substation inspection methods cannot effectively solve the visual limitations of fixed terminals and have the problem of low substation inspection accuracy. Summary of the Invention

[0004] Based on this, in view of the above technical problem of low substation inspection accuracy, it is necessary to provide a substation inspection method, device, computer device, computer-readable storage medium, and computer program product.

[0005] In a first aspect, the present application provides a substation inspection method, including:

[0006] Performing feature extraction on inspection images of a substation to be inspected to obtain a time-series feature library of electrical equipment in the substation to be inspected, and marking areas to be supplemented for collection on the inspection images to obtain a marked map of areas to be supplemented for collection; wherein, the time-series feature library includes feature marks of the electrical equipment; the marked map of areas to be supplemented for collection represents the original point cloud data of multiple inspection angles of the electrical equipment;

[0007] Performing spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the time-series feature library to obtain spatial reference data;

[0008] Performing feature matching and construction on the spatial reference data based on the temporal feature library to obtain the three-dimensional semantic model of the substation to be inspected;

[0009] Detecting visual dead zones and planning paths for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected;

[0010] Controlling the target unmanned aerial vehicle to inspect the visual dead zones to be inspected based on the collaborative inspection paths to obtain supplementary inspection images, and fusing the supplementary inspection images and the inspection images to obtain the inspection results of the substation to be inspected.

[0011] In one embodiment, extracting features from the inspection images of the substation to be inspected to obtain the temporal feature library of electrical equipment in the substation to be inspected, and marking the areas to be supplemented for acquisition in the inspection images to obtain a marked map of areas to be supplemented for acquisition, including: obtaining multi-scale luminance equalized inspection image layers based on the inspection images; obtaining the positioning results of the equipment in the substation to be inspected based on the multi-scale luminance equalized inspection image layers; extracting feature vectors and calculating vector quality for the positioning results to obtain equipment feature quality scores, and obtaining a supplementary acquisition requirement map based on the equipment feature quality scores; obtaining the temporal feature library of electrical equipment and the marked map of areas to be supplemented for acquisition in the substation to be inspected based on the supplementary acquisition requirement map.

[0012] In one embodiment, performing spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the temporal feature library to obtain spatial reference data, including: obtaining feature registration parameters based on the original point cloud data and the feature marks of the electrical equipment included in the temporal feature library; performing spatial transformation calculations on the point cloud data of adjacent sub-regions in the original point cloud data based on the feature registration parameters until an inspection point cloud model is obtained; performing statistical outlier calculations on the point cloud model to obtain a point cloud density distribution map, and obtaining filtered point cloud data based on the point cloud density distribution map; obtaining coordinate point clouds based on the filtered point cloud data and annotating the coordinate point clouds to obtain the spatial reference data.

[0013] In one embodiment, the feature matching and construction of the spatial reference data based on the timing feature library to obtain the three-dimensional semantic model of the substation to be inspected includes: obtaining feature space position data based on the timing feature library, and processing the feature space position data and the spatial reference data to obtain feature pairs; obtaining device three-dimensional models based on the feature pairs, and obtaining a device distance matrix associated with the substation to be inspected according to the device three-dimensional models; obtaining an electrical topology diagram according to the device distance matrix, and obtaining a state discrimination relation formula based on the electrical topology diagram; obtaining a three-dimensional semantic model associated with the substation to be inspected based on the state discrimination relation formula, the device three-dimensional models, and the electrical topology diagram.

[0014] In one embodiment, the detection of visual dead zones and path planning for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected includes: obtaining visual occlusion data corresponding to each electrical device in the substation to be inspected based on the three-dimensional semantic model, and obtaining the visual dead zones to be inspected based on the visual occlusion data; obtaining an observation point set based on the visual dead zones to be inspected, and obtaining continuous route data according to the observation point set; performing obstacle avoidance calculation in the substation to be inspected based on the continuous route data to obtain a safe route map, and processing the safe route map to obtain the collaborative inspection paths of the substation to be inspected.

[0015] In one embodiment, the fusion of the supplementary inspection image and the inspection image to obtain the inspection result of the substation to be inspected includes: performing feature state matching and multi-source feature fusion on the supplementary inspection image and the inspection image to obtain a multi-dimensional feature state map; extracting the state features of the electrical devices in the multi-dimensional feature state map to obtain a device operation feature set, and obtaining device anomaly marking data based on the device operation feature set; generating a device anomaly change curve based on the device anomaly marking data, and analyzing the parameters in the device anomaly change curve based on preset electrical device operation parameters to obtain device operation deviation data; performing grading boundary division on the device operation deviation data to obtain a device state evaluation result, and obtaining a device trend prediction map based on the device state evaluation result; extracting risk factors in the device trend prediction map, and calculating the harm degree of the device trend prediction map to obtain a risk warning map; integrating the inspection results of the electrical devices based on the risk warning map to obtain the inspection result of the substation to be inspected.

[0016] In a second aspect, the present application also provides a substation inspection device, including:

[0017] An area marking module is used to extract features from the inspection images of the substation to be inspected, obtain a time-series feature library of electrical equipment in the substation to be inspected, and mark the areas to be supplemented for acquisition in the inspection images to obtain an area marking map for areas to be supplemented for acquisition; wherein, the time-series feature library includes feature marks of the electrical equipment; the area marking map for areas to be supplemented for acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment;

[0018] A status annotation module is used to perform spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the time-series feature library to obtain spatial reference data;

[0019] A model acquisition module is used to perform feature matching and construction on the spatial reference data based on the time-series feature library to obtain a three-dimensional semantic model of the substation to be inspected;

[0020] A path acquisition module is used to detect visual dead zones and plan paths for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected;

[0021] A result acquisition module is used to control a target unmanned aerial vehicle to inspect the visual dead zones to be inspected based on the collaborative inspection paths, obtain inspection supplementary images, and fuse the inspection supplementary images and the inspection images to obtain the inspection results of the substation to be inspected.

[0022] In a third aspect, the present application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0023] Extract features from the inspection images of the substation to be inspected, obtain a time-series feature library of electrical equipment in the substation to be inspected, and mark the areas to be supplemented for acquisition in the inspection images to obtain an area marking map for areas to be supplemented for acquisition; wherein, the time-series feature library includes feature marks of the electrical equipment; the area marking map for areas to be supplemented for acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment;

[0024] Perform spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the time-series feature library to obtain spatial reference data;

[0025] Perform feature matching and construction on the spatial reference data based on the time-series feature library to obtain a three-dimensional semantic model of the substation to be inspected;

[0026] Detect visual dead zones and plan paths for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected;

[0027] The control target UAV performs inspection on the visual dead zone to be inspected based on the collaborative inspection path, obtains supplementary inspection images, and fuses the supplementary inspection images and the inspection images to obtain the inspection result of the substation to be inspected.

[0028] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0029] Feature extraction is performed on the inspection images of the substation to be inspected to obtain a time-series feature library of electrical equipment in the substation to be inspected, and marking of areas to be supplemented for acquisition is performed on the inspection images to obtain a marked map of areas to be supplemented for acquisition; wherein, the time-series feature library includes feature marks of the electrical equipment; the marked map of areas to be supplemented for acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment;

[0030] According to the feature marks of the electrical equipment included in the time-series feature library, spatial registration and status annotation are performed on the original point cloud data to obtain spatial reference data;

[0031] Based on the time-series feature library, feature matching and construction are performed on the spatial reference data to obtain a three-dimensional semantic model of the substation to be inspected;

[0032] Detection and path planning of visual dead zones are performed on the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection path of the substation to be inspected;

[0033] The control target UAV performs inspection on the visual dead zone to be inspected based on the collaborative inspection path, obtains supplementary inspection images, and fuses the supplementary inspection images and the inspection images to obtain the inspection result of the substation to be inspected.

[0034] In a fifth aspect, the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0035] Feature extraction is performed on the inspection images of the substation to be inspected to obtain a time-series feature library of electrical equipment in the substation to be inspected, and marking of areas to be supplemented for acquisition is performed on the inspection images to obtain a marked map of areas to be supplemented for acquisition; wherein, the time-series feature library includes feature marks of the electrical equipment; the marked map of areas to be supplemented for acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment;

[0036] Perform spatial registration and status annotation on the original point cloud data according to the feature markers of the electrical equipment included in the timing feature library to obtain spatial reference data;

[0037] Based on the timing feature library, perform feature matching and construction on the spatial reference data to obtain the three-dimensional semantic model of the substation to be inspected;

[0038] Detect the visual dead zone and plan the path for the three-dimensional semantic model to obtain the visual dead zone to be inspected and the collaborative inspection path of the substation to be inspected;

[0039] Control the target unmanned aerial vehicle to inspect the visual dead zone to be inspected based on the collaborative inspection path to obtain inspection supplementary images, and fuse the inspection supplementary images and the inspection images to obtain the inspection results of the substation to be inspected.

[0040] The above substation inspection method, device, computer device, storage medium and computer program product have the following beneficial effects during the substation inspection process: First, extract the features of the inspection images of the substation to be inspected to obtain the timing feature library of the electrical equipment in the substation to be inspected, and mark the areas to be supplemented for image acquisition of the inspection images to obtain the marked map of the areas to be supplemented for image acquisition; among them, the timing feature library includes the feature markers of the electrical equipment; the marked map of the areas to be supplemented for image acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment; then, according to the feature markers of the electrical equipment included in the timing feature library, perform spatial registration and status annotation on the original point cloud data to obtain spatial reference data; then, based on the timing feature library, perform feature matching and construction on the spatial reference data to obtain the three-dimensional semantic model of the substation to be inspected; then, detect the visual dead zone and plan the path for the three-dimensional semantic model to obtain the visual dead zone to be inspected and the collaborative inspection path of the substation to be inspected; finally, control the target unmanned aerial vehicle to inspect the visual dead zone to be inspected based on the collaborative inspection path to obtain inspection supplementary images, and fuse the inspection supplementary images and the inspection images to obtain the inspection results of the substation to be inspected. In the above process, through systematic feature extraction and image analysis, the virtual state and changes of electrical equipment can be identified more quickly; on the basis of the known visual dead zone, targeted image supplementation is carried out to ensure no omission; by using unmanned aerial vehicles for automatic inspection, the input of human resources is reduced and work safety is improved; real-time data monitoring can be realized, and inspection data is collected and analyzed through a software system to timely reflect the equipment status. Therefore, the above process improves the accuracy of substation inspection. Description of the Drawings

[0041] To more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the related art. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0042] Figure 1 It is a schematic flow chart of the substation inspection method in an embodiment;

[0043] Figure 2 It is a schematic flow chart of the acquisition steps of the marked map of the area to be supplemented for acquisition in an embodiment;

[0044] Figure 3 It is a schematic flow chart of the substation inspection method in another embodiment;

[0045] Figure 4 It is a structural block diagram of the substation inspection device in an embodiment;

[0046] Figure 5 It is an internal structure diagram of a computer device in an embodiment. Specific embodiments

[0047] In order to make the purpose, technical solutions and advantages of the present application clearer, the following further details the present application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] In the field of power system operation and maintenance, the status monitoring and fault diagnosis of substation equipment are key links to ensure the safe and stable operation of the power grid, and equipment status inspection is an important means of monitoring and diagnosis. At present, substations generally use fixed intelligent image terminals to inspect and monitor equipment. However, due to factors such as terminal layout limitations and fixed shooting angles, there are often visual dead angles, and it is difficult to comprehensively obtain equipment status information. In particular, the recognition accuracy of the opening and closing states of key equipment such as switches and disconnectors is relatively low.

[0049] In the prior art, some studies have tried to use single - machine drones to assist in inspection or use simple three - dimensional reconstruction techniques for route planning. However, these methods have many deficiencies: First, the single - machine inspection efficiency is low and it is difficult to meet the inspection requirements of large substations; second, traditional two - dimensional route planning methods ignore the complex three - dimensional spatial structure of substations and are prone to safety hazards; third, the existing inspection systems lack a coordination mechanism between fixed terminals and drones and it is difficult to effectively integrate multi - source image data. That is, the existing substation inspection methods cannot effectively solve the visual limitations of fixed terminals and have obvious deficiencies in multi - machine coordination and safety.

[0050] In an exemplary embodiment, as Figure 1 shown, a substation inspection method is provided, which is applied to a terminal or a server, and includes the following steps S102 to S110. Among them:

[0051] Step S102: Extract features from the inspection images of the substation to be inspected to obtain a time-series feature library of electrical equipment in the substation to be inspected, and mark the areas to be supplemented for acquisition in the inspection images to obtain a marked map of the areas to be supplemented for acquisition; wherein, the time-series feature library includes feature marks of electrical equipment; the marked map of the areas to be supplemented for acquisition represents the original point cloud data of multiple inspection angles of electrical equipment.

[0052] Among them, feature extraction refers to identifying the key features of electrical equipment from inspection images, such as information on size, shape, color, etc.; the time-series feature library refers to a database containing the state and feature data of electrical equipment at different time points; the marked map of the areas to be supplemented for acquisition refers to a specific area in the inspection image that identifies the areas where additional data needs to be collected.

[0053] As an example, it is possible to obtain the fixed inspection images of the fixed acquisition terminals in the substation to be inspected, and perform extraction of multi-dimensional image features and marking and division of the areas to be supplemented for acquisition on the fixed inspection images to obtain a time-series feature library of electrical equipment in the substation to be inspected and a marked map of the areas to be supplemented for acquisition.

[0054] Step S104: Perform spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the time-series feature library to obtain spatial reference data.

[0055] Among them, spatial registration refers to aligning point cloud data from different sources, such as taken at different times or from different angles, to a unified coordinate system; status annotation refers to describing and identifying the status of electrical equipment in the spatial reference data.

[0056] As an example, it is possible to obtain the original point cloud data corresponding to multiple inspection angles of the marked map of the areas to be supplemented for acquisition, and perform spatial registration and status annotation on the original point cloud data based on the corresponding electrical equipment feature marks in the time-series feature library to obtain high-precision spatial reference data.

[0057] Step S106: Perform feature matching and construction on the spatial reference data based on the time-series feature library to obtain a three-dimensional semantic model of the substation to be inspected.

[0058] Among them, the three-dimensional semantic model refers to a three-dimensional model constructed based on the extracted features and annotations, including electrical equipment and their relationships.

[0059] As an example, based on the time-series feature library, spatio-temporal feature matching can be performed on the spatial reference data to construct the device constraints of electrical equipment, thereby generating a three-dimensional semantic model of the substation to be inspected.

[0060] Step S108: Detect the visual dead zones and plan the paths for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected.

[0061] Among them, visual dead zone detection refers to identifying the areas where the equipment cannot be covered during the inspection process; path planning refers to designing an optimal inspection path to efficiently cover the visual dead zones.

[0062] As an example, by detecting the visual dead zones corresponding to the fixed acquisition terminals and planning multi-aircraft paths for the three-dimensional semantic model, the visual dead zones to be inspected and the collaborative inspection paths can be obtained.

[0063] Step S110: Control the target unmanned aerial vehicle (UAV) to inspect the visual dead zones to be inspected based on the collaborative inspection paths, obtain the supplementary inspection images, and fuse the supplementary inspection images and the inspection images to obtain the inspection results of the substation to be inspected.

[0064] Among them, controlling the target UAV to conduct inspections means using the UAV to perform automated or semi-automated inspection tasks; the inspection results refer to the final analysis results formed by merging images at different times and angles.

[0065] As an example, based on the collaborative inspection paths, the target UAV can be controlled to inspect the visual dead zones to be inspected, obtain the supplementary inspection images, perform feature state matching and multi-source feature fusion on the supplementary inspection images and the fixed inspection images (inspection images), and obtain a multi-dimensional feature state map; then, perform abnormal state detection and evaluation on the multi-dimensional feature state map to obtain the electrical equipment state evaluation results, and conduct risk assessment and integration on the electrical equipment state evaluation results to generate the substation inspection results.

[0066] In the above substation inspection method, first, feature extraction is performed on the inspection images of the substation to be inspected, obtaining a time-series feature library of electrical equipment in the substation to be inspected, and marking the areas to be supplemented for image acquisition on the inspection images, obtaining a marked map of areas to be supplemented for image acquisition; among them, the time-series feature library includes feature marks of electrical equipment; the marked map of areas to be supplemented for image acquisition represents the original point cloud data of multiple inspection angles of electrical equipment; then, according to the feature marks of electrical equipment included in the time-series feature library, spatial registration and status annotation are performed on the original point cloud data, obtaining spatial reference data; next, based on the time-series feature library, feature matching and construction are performed on the spatial reference data, obtaining a three-dimensional semantic model of the substation to be inspected; then, detection of visual dead zones and path planning are performed on the three-dimensional semantic model, obtaining the visual dead zones to be inspected and the collaborative inspection path of the substation to be inspected; finally, controlling the target unmanned aerial vehicle to perform inspections on the visual dead zones to be inspected based on the collaborative inspection path, obtaining supplementary inspection images, and fusing the supplementary inspection images and the inspection images, obtaining the inspection results of the substation to be inspected. In the above process, through systematic feature extraction and image analysis, the virtual states and changes of electrical equipment can be recognized more quickly; based on the known visual dead zones, targeted image supplementation is carried out to ensure no omission; by using unmanned aerial vehicles for automatic inspections, the input of human resources is reduced and work safety is improved; real-time data monitoring can be achieved, and inspection data is collected and analyzed through a software system to promptly reflect the equipment status. Therefore, the above process improves the accuracy of substation inspections.

[0067] In one embodiment, as Figure 2 shown, performing feature extraction on the inspection images of the substation to be inspected, obtaining a time-series feature library of electrical equipment in the substation to be inspected, and marking the areas to be supplemented for image acquisition on the inspection images, obtaining a marked map of areas to be supplemented for image acquisition, includes:

[0068] Step S202, based on the inspection images, obtaining multi-scale luminance equalized inspection image layers; Step S204, based on the multi-scale luminance equalized inspection image layers, obtaining the positioning results of the equipment in the substation to be inspected; Step S206, performing feature vector extraction and vector quality calculation on the positioning results, obtaining the equipment feature quality scores, and obtaining a supplementary acquisition requirement map based on the equipment feature quality scores; Step S208, based on the supplementary acquisition requirement map, obtaining a time-series feature library of electrical equipment in the substation to be inspected and a marked map of areas to be supplemented for image acquisition.

[0069] Among them, the multi-scale brightness equalized inspection image layer refers to processing the inspection image and performing brightness equalization using multi-scale technology so that the device features can be clearly displayed in the image at different scales; the positioning result refers to identifying the specific position of the device in space based on the processed inspection image; the feature vector extraction refers to converting the positioning result into a mathematical vector and extracting the key features of the device, such as shape, size, and color; the vector quality calculation refers to evaluating the effectiveness of the feature vector and calculating the quality score of the electrical device features; the device feature quality score refers to the quality index reflecting the feature vector extracted from the positioning result; the supplementary acquisition requirement map refers to an image or area formed based on the device feature quality score, indicating the device or position that requires additional data acquisition.

[0070] As an example, the brightness histogram calculation and brightness equalization can be performed on the fixed inspection image to obtain the brightness equalized inspection image, and the multi-scale decomposition can be performed on the brightness equalized inspection image to obtain the multi-scale brightness equalized inspection image layer; then the gradient calculation and texture extraction are performed on each scale brightness equalized inspection image layer to generate the multi-scale image feature data, and the initial region segmentation and the positioning of the electrical device features are performed on the multi-scale image feature data to obtain the device positioning result; then the extraction of the device feature vector and the vector quality calculation are performed on the device positioning result to obtain the device feature quality score, and the threshold evaluation and the marking of the supplementary acquisition area are performed on the device feature quality score to generate the supplementary acquisition requirement map; finally, the temporal correlation is performed on the supplementary acquisition requirement map to obtain the state sequence data, the feature organization is performed on the state sequence data to construct the temporal feature library of the electrical devices in the substation to be inspected, and the spatial mapping and the division of the marked area are performed on the supplementary acquisition requirement map to generate the marked map of the area to be supplemented for acquisition.

[0071] In this embodiment, by equalizing the brightness and reducing the influence brought by the illumination change, the readability of the image can be improved, thus helping the subsequent analysis and device identification; the positioning result can provide the basic data for the subsequent feature extraction and state analysis, ensuring the accurate positioning of the device in space; according to the feature quality score, it can be judged which devices need more detailed data acquisition, ensuring the comprehensiveness and effectiveness of the inspection.

[0072] In an exemplary embodiment, according to the feature markings of the electrical devices included in the temporal feature library, the spatial registration and state annotation are performed on the original point cloud data to obtain the spatial reference data, including:

[0073] Based on the feature markers of electrical equipment included in the original point cloud data and the time-series feature library, obtain the feature registration parameters; based on the feature registration parameters, perform spatial transformation calculations on the point cloud data of adjacent sub-regions in the original point cloud data until an inspection point cloud model is obtained; perform statistical outlier calculations on the point cloud model to obtain a point cloud density distribution map, and obtain filtered point cloud data based on the point cloud density distribution map; obtain coordinate point cloud based on the filtered point cloud data, and annotate the coordinate point cloud to obtain spatial reference data.

[0074] Among them, the original point cloud data refers to a three-dimensional coordinate set containing the positions of objects in space obtained through laser scanning or other three-dimensional imaging technologies; the feature registration parameters refer to the parameters obtained by comparing the time-series feature library and the original point cloud data, and are used for spatial registration of the point cloud; the spatial transformation calculation refers to performing mathematical transformations, such as rotation and translation, on the point cloud data of adjacent sub-regions according to the feature registration parameters; the point cloud density distribution map refers to a chart drawn based on the point density of each region in the point cloud, reflecting the acquisition situation of each part and the distribution characteristics of the point cloud; the filtered point cloud data refers to the point cloud data after filtering processing, usually by removing outliers or noise to improve data quality; the coordinate point cloud refers to converting the filtered point cloud data into a set of points with spatial coordinates; the spatial reference data refers to the annotated coordinate point cloud data, which is three-dimensional spatial data with attached feature and status information.

[0075] As an example, the inspection space feature data can be obtained by extracting geometric features from the original point cloud data, and the boundary fitting calculation is performed on the space feature data to generate equipment boundary data; then, based on the corresponding electrical equipment feature markers in the time-series feature library, the similarity calculation of the electrical equipment contour features is performed on the equipment boundary data to obtain a set of corresponding feature points, and the rigid transformation matrix is solved for the set of corresponding feature points to generate feature registration parameters; then, based on the feature registration parameters, spatial transformation calculations are performed on the point cloud data of adjacent sub-regions in the original point cloud data to obtain a regional stitching result, and the minimum error calculation of the overlapping region is performed on the regional stitching result to generate a complete inspection point cloud model; then, statistical outlier calculations are performed on the complete point cloud model to obtain a point cloud density distribution map, and voxel filtering is performed on the point cloud density distribution map to generate reduced point cloud data; finally, world coordinate system mapping calculations are performed on the reduced point cloud data to obtain absolute coordinate point cloud, and the electrical equipment type and position are annotated on the absolute coordinate point cloud to generate high-precision spatial reference data.

[0076] In this embodiment, the above process ensures a complete chain from the original data to the high-quality spatial reference data; through precise feature registration and calculation, the utilization value of the point cloud data can be maximized; through the calculation of outliers and data density analysis, it helps to clean and optimize the data, ensuring that the obtained coordinate point cloud and spatial reference data can accurately reflect the true state of the device and provide a reference for subsequent steps.

[0077] Furthermore, in one embodiment, based on the time-series feature library, feature matching and construction are performed on the spatial reference data to obtain a three-dimensional semantic model of the substation to be inspected, including:

[0078] Obtain feature space position data based on the time-series feature library, and process the feature space position data and the spatial reference data to obtain feature pairs; obtain the three-dimensional model of the device based on the feature pairs, and obtain the distance matrix between devices associated with the substation to be inspected according to the three-dimensional model of the device; obtain the electrical topology diagram based on the distance matrix between devices, and obtain the state discrimination relationship formula based on the electrical topology diagram; obtain the three-dimensional semantic model associated with the substation to be inspected based on the state discrimination relationship formula, the three-dimensional model of the device, and the electrical topology diagram.

[0079] Among them, the feature space position data refers to the feature data describing the device in space, including position, shape, and structure information, etc.; the feature pair refers to the matching of the feature space position data and the spatial reference data to form corresponding feature combinations; the distance matrix between devices refers to the matrix containing the relative distance information between electrical devices; the state discrimination relationship formula refers to the mathematical model or formula established according to the state of the device and the electrical topology relationship to help deduce the operating conditions of the device.

[0080] As an example, the feature space position data can be obtained by performing three-dimensional space coordinate mapping on the time-series feature library, and distance measurement calculation is performed on the feature space position data and the high-precision spatial reference data to generate effective feature pairs; the boundary contours of electrical devices are extracted and three-dimensional voxel reconstruction is performed on the effective feature pairs to generate a three-dimensional model of the device, and the relative distance and azimuth angle are calculated for the three-dimensional model of the device to obtain the distance matrix between devices; then hierarchical clustering of the distribution of the distance matrix between devices is performed to generate a spatial structure tree, and electrical terminal identification and loop tracing calculation are performed on the device nodes in the spatial structure tree to generate an electrical topology diagram; then the operating parameters of the device nodes in the electrical topology diagram are measured and time series sorting is performed to generate a state change sequence, and threshold comparison and pattern recognition are performed on the state change sequence to obtain the state discrimination relationship formula; finally, importance calculation is performed on the state discrimination relationship formula to generate state evaluation system data, and multi-modal feature combination and semantic annotation calculation are performed on the three-dimensional model of the device, the electrical topology diagram, and the state evaluation system data to generate a three-dimensional semantic model.

[0081] In this embodiment, starting from obtaining data in the time series feature library, a 3D model of the electrical equipment is established using feature space position data and spatial reference data, and the relationships and states between the equipment are further analyzed. This series of structured steps ensures that the modeling, performance analysis, and safety assessment of the electrical equipment are based on accurate data and logical reasoning, which helps improve the efficiency, accuracy, and timeliness of substation inspection tours.

[0082] In one embodiment, the visual dead zones of the 3D semantic model are detected and path planning is performed to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected, including:

[0083] Based on the 3D semantic model, the visual occlusion data corresponding to each electrical equipment in the substation to be inspected is obtained, and based on the visual occlusion data, the visual dead zones to be inspected are obtained; based on the visual dead zones to be inspected, an observation point set is obtained, and continuous route data is obtained according to the observation point set; obstacle avoidance calculation is performed in the substation to be inspected based on the continuous route data to obtain a safe route map, and the safe route map is processed to obtain the collaborative inspection path of the substation to be inspected.

[0084] Among them, the visual occlusion data refers to the information describing the obstruction of the line of sight caused by the presence of other objects during the observation process; the visual dead zones to be inspected refer to the areas or equipment that cannot be observed due to visual occlusion during the inspection process; the observation point set refers to the set of available observation points selected according to the distribution of the visual dead zones, and each point can effectively cover a specific area; the continuous route data refers to a series of continuous path data generated from the observation point set, indicating the trajectory that the equipment needs to pass through during the inspection process; the collaborative inspection path refers to the final inspection route plan considering the visual dead zones, the observation point set, and the safe route.

[0085] As an example, the visual occlusion data of the equipment can be obtained by calculating the perspective visibility of each electrical equipment in the 3D semantic model, and the visual dead zones to be inspected can be generated by performing regional union and intersection operations on the visual occlusion data of the equipment; then, the power supply range of each electrical equipment in the 3D semantic model is identified and the load level is calculated to obtain the equipment weight data, and the inspection order table is generated by performing sorting operations on the equipment weight data with multiple preset inspection factors; then, the visual dead zones to be inspected are rasterized and ray traced to obtain the visible area data, and the optimal inspection perspective is calculated for the visible area data to generate an effective observation point set; then, the inspection path length is estimated for the effective observation point set and the balanced distribution of multi-robot inspections is performed to generate a multi-robot collaborative control sequence, and path interpolation and curve fitting are performed on the collaborative control sequence to obtain continuous route data; finally, obstacle avoidance calculation is performed on the continuous route data to generate a safe route map, and the flight distance and turning angle are calculated for the safe route map and the secondary adjustment of the preset inspection path parameters is performed to generate the collaborative inspection path.

[0086] In this embodiment, by analyzing the visual occlusion data, the visual dead zones that need to be focused on during the inspection are identified; then, through the selection of observation points and the calculation of continuous routes, the integrity and safety of the inspection path are ensured; finally, the generated safety route map and collaborative inspection path provide clear guidance for the actual inspection, which can improve the efficiency and safety of the inspection.

[0087] Further, in an exemplary embodiment, the inspection supplementary images and inspection images are fused to obtain the inspection results of the substation to be inspected, including:

[0088] Perform feature state matching and multi-source feature fusion on the inspection supplementary images and inspection images to obtain a multi-dimensional feature state map; extract the state features of electrical equipment from the multi-dimensional feature state map to obtain a set of equipment operation features, and obtain equipment anomaly marking data based on the set of equipment operation features; generate an equipment anomaly change curve based on the equipment anomaly marking data, and analyze the parameters in the equipment anomaly change curve based on the preset electrical equipment operation parameters to obtain equipment operation deviation data; perform classification boundary division on the equipment operation deviation data to obtain an equipment state evaluation result, and obtain an equipment trend prediction map based on the equipment state evaluation result; extract risk factors from the equipment trend prediction map, and calculate the hazard degree of the equipment trend prediction map to obtain a risk warning map; based on the risk warning map, integrate the inspection results of electrical equipment to obtain the inspection results of the substation to be inspected.

[0089] Among them, multi-source feature fusion refers to integrating data features from different sources to form a unified feature representation; the multi-dimensional feature state map refers to a graph generated based on the fused multi-source features, which shows the multi-dimensional state features of electrical equipment; the set of equipment operation features refers to the set of equipment state features extracted from the multi-dimensional feature state map, which is used to reflect the current operation state of the equipment; classification boundary division refers to classifying the equipment operation deviation data and dividing it into multiple levels according to different deviation degrees; risk factors refer to the key factors extracted from the equipment trend prediction map that may affect the operation stability or safety of the equipment; the inspection result refers to the comprehensive result finally formed that reflects the equipment state and risk assessment of the substation to be inspected.

[0090] As an example, by extracting the edge and texture features of the supplementary inspection images and calculating the local descriptors for the electrical equipment during patrol inspection, device feature description data can be generated. Based on the historical patrol inspection feature data corresponding to the fixed patrol inspection images, nearest neighbor search and ratio test are performed on the device feature description data to obtain a set of feature point pairs. Then, consistency verification of the patrol inspection space is carried out on the set of feature point pairs to generate valid feature matching pairs, and time window division and sequence alignment are performed on the valid feature matching pairs to obtain state change data. Also, determination of the fixed patrol inspection state and extraction of the time sequence are carried out on the time sequence feature library to obtain the historical state sequence and the historical feature trend. Furthermore, pattern recognition is performed on the state change data and the historical state sequence to generate a change feature map, and based on the historical feature trend, multi-source feature fusion is carried out on the change feature map to obtain feature fusion data. Then, credibility calculation and feature selection are performed on the feature fusion data to generate a weighted feature map, and based on the historical patrol inspection feature data, patrol inspection space mapping and patrol inspection parameter association are carried out on the weighted feature map to generate a state feature space. Finally, time sequence correlation calculation and dimension transformation are carried out on the state feature space to obtain a key state vector, and feature time sequence organization and spatial registration are carried out on the key state vector to generate a multi-dimensional feature state map.

[0091] Furthermore, by extracting the state features of the electrical equipment from the multi-dimensional feature state map, a set of device operation features can be obtained, and threshold statistics are carried out on the set of device operation features to generate abnormal marker data. Electrical change quantity calculation and time sequence curve calculation are carried out on the abnormal marker data to generate an abnormal change curve, and based on the preset electrical equipment operation parameters, parameter comparison and deviation statistics are carried out on the abnormal change curve to obtain operation deviation data. Then, grading boundary division is carried out on the operation deviation data to generate a device state evaluation result, and time series reconstruction and failure mode identification are carried out on the device state evaluation result to generate a device trend prediction map. Finally, risk factor extraction and hazard degree calculation are carried out on the trend prediction map to generate a risk warning map, and based on the risk warning map, integration of the patrol inspection results of the electrical equipment is carried out to obtain the substation patrol inspection results.

[0092] In this embodiment, by analyzing and extracting various features, the patrol inspection results are finally generated, ensuring comprehensive monitoring, abnormal detection, and risk assessment of electrical equipment. By combining multi-dimensional data to generate trend prediction and risk warning, not only the accuracy of device patrol inspection is improved, but also potential fault risks can be effectively predicted and addressed, optimizing the management and operation and maintenance strategies of the substation.

[0093] This application provides a substation patrol inspection method. To better understand the process of the above substation patrol inspection method, in combination with Figure 3 as shown below, the specific process of a substation patrol inspection method of this application is elaborated in detail, including the following steps:

[0094] Step S302: Obtain the fixed inspection images of the fixed acquisition terminals in the substation to be inspected, extract multi-dimensional image features from the fixed inspection images, and mark and divide the areas to be supplemented for acquisition, so as to obtain the time series feature library of the electrical equipment in the substation to be inspected and the marked map of the areas to be supplemented for acquisition.

[0095] Among them, relevant data can be obtained and processed based on artificial intelligence technology. Among them, Artificial Intelligence (AI) 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 knowledge to obtain the best results. Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0096] Furthermore, the fixed acquisition terminal is a fixed intelligent image terminal installed in a substation, such as a fixed high-definition monitoring camera: installed at a fixed position in the substation, used for continuously acquiring images of the equipment appearance, mainly monitoring the status of key equipment such as switches and disconnectors; a fixed infrared thermal imager: installed around key equipment, used for monitoring the temperature distribution of the equipment and detecting abnormal heating of the equipment; a fixed ultra-high frequency partial discharge monitoring device: arranged around high-voltage equipment, used for detecting corona discharge phenomena and monitoring the insulation condition, and the relevant equipment can also be increased or decreased according to the needs of substation monitoring; the fixed inspection images include but are not limited to visible light images acquired by fixed high-definition cameras, such as recording the appearance state of the equipment, reflecting the opening and closing positions of equipment such as switches and disconnectors, recording the mechanical damage on the surface of the equipment, temperature distribution images acquired by infrared thermal imagers, such as recording the equipment temperature distribution map, reflecting the abnormal heating condition of the equipment, showing the heating situation of the contact point, and discharge distribution images acquired by ultra-high frequency partial discharge monitoring devices, such as recording the discharge situation of the equipment, reflecting the insulation state, and showing the discharge intensity distribution; electrical equipment refers to the equipment that needs to be inspected in the substation, such as switches and disconnectors; calculating the brightness histogram and performing brightness equalization on the fixed inspection images to obtain brightness-equalized inspection images, and performing multi-scale decomposition on the brightness-equalized inspection images to obtain multi-scale brightness-equalized inspection image layers; calculating the gradient and extracting the texture for each scale brightness-equalized inspection image layer to generate multi-scale image feature data, and performing initial region segmentation and positioning of electrical equipment features on the multi-scale image feature data to obtain equipment positioning results; extracting the equipment feature vectors and calculating the vector quality for the equipment positioning results to obtain the equipment feature quality scores, and performing threshold evaluation and marking of the supplementary acquisition area on the equipment feature quality scores to generate a supplementary acquisition requirement map; performing temporal correlation on the supplementary acquisition requirement map to obtain state sequence data, organizing the features of the state sequence data to construct a temporal feature library of electrical equipment in the substation to be inspected, and performing spatial mapping and division of the marked area on the supplementary acquisition requirement map to generate a marked map of the area to be supplemented for acquisition.

[0097] Furthermore, in practical applications, visible light images are collected by fixed high-definition cameras to obtain the appearance status and opening / closing position information of the equipment. At the same time, an infrared thermal imager is used to obtain the temperature distribution image of the equipment to monitor abnormal heating conditions. The partial discharge monitoring device of ultra-high frequency is used to collect the discharge distribution image to evaluate the insulation status of the equipment, as well as the historical inspection feature data corresponding to the historical acquisition data and historical images collected by various equipment, etc., to form the fixed inspection image data. Then, the brightness histogram of the image is calculated, and the brightness distribution characteristics in the image are analyzed. Then, based on the histogram distribution, the image is subjected to adaptive equalization processing. By adjusting the pixel brightness distribution, the image contrast is enhanced to generate a brightness equalized inspection image. The multi-scale decomposition algorithm is used. Through Gaussian filtering and downsampling operations at different scales, the brightness equalized inspection image is decomposed into an image pyramid containing different levels of details, so as to obtain multiple scales of brightness equalized inspection image layers. Then, gradient calculations are performed on each scale layer respectively to extract the edge and contour information in the image. At the same time, texture feature operators such as local binary pattern are used to extract the texture features of the image, and these features are organized according to the scale to generate multi-scale image feature data containing edge, contour and texture information. Based on these feature data, image segmentation is carried out. Algorithms such as region growing or watershed are used to divide the image into multiple candidate regions, and combined with the predefined electrical equipment feature template, equipment feature matching and positioning are carried out on these regions. Finally, the precise positioning results of various electrical equipment in the substation are obtained. Then, based on the equipment positioning results, feature vectors of each equipment area are extracted, including multi-dimensional information such as shape features, texture features, and color features. By calculating indicators such as the similarity between these feature vectors and the standard template, the stability and distinguishability of the features, a score reflecting the feature quality is obtained, and these feature quality scores are compared with a preset threshold. The areas below the threshold are marked, and these areas usually indicate positions with poor image quality or insufficient feature extraction and need to be supplemented with acquisition, thus generating a supplementary acquisition requirement map. Then, for the marked areas in the supplementary acquisition requirement map, by comparing the image data collected at different time points, the change features of the equipment status are extracted, and these time series features are organized according to the equipment type and location to construct a complete electrical equipment time series feature library. The supplementary acquisition requirement map is projected onto the three-dimensional space model of the substation to determine the specific spatial position and range of the areas that need to be supplemented with acquisition, and these areas are reasonably divided according to the equipment distribution and environmental constraints. Finally, a marked map of the areas to be supplemented with acquisition for guiding subsequent UAV supplementary inspections is generated.

[0098] Step S304: Obtain the original point cloud data corresponding to multiple inspection angles of the marked map of the areas to be supplemented with acquisition, and perform spatial registration and status annotation on the original point cloud data based on the corresponding electrical equipment feature marks in the time series feature library to obtain high-precision spatial reference data.

[0099] Among them, the distribution of electrical equipment in the area to be supplemented and collected is statistically analyzed to obtain equipment density data, and the threshold of the equipment density data is calculated to generate regional complexity data; the regional complexity data is clustered to obtain regional hierarchy data, and the scanning parameters of the regional hierarchy data are configured to generate a hierarchical scanning plan; for each scanning layer in the hierarchical scanning plan, the laser scanning parameters are set to obtain a scanning control sequence, and based on the scanning control sequence, the preset laser scanning device is used to perform hierarchical execution on the substation to be inspected to generate original point cloud data; the original point cloud data is subjected to multiple echo calculations to obtain point cloud intensity data, and the point cloud intensity data is numerically calibrated to generate point cloud data with intensity information; the geometric features of the original point cloud data are extracted to obtain inspection space feature data, and the boundary fitting calculation of the space feature data is performed to generate equipment boundary data; based on the corresponding electrical equipment feature markers in the time series feature library, the similarity of the electrical equipment contour features of the equipment boundary data is calculated to obtain a set of corresponding feature points, and the rigid transformation matrix of the set of corresponding feature points is solved to generate feature registration parameters; based on the feature registration parameters, the spatial transformation calculation of the point cloud data in adjacent sub-regions of the original point cloud data is performed to obtain a regional stitching result, and the overlapping region error minimization calculation of the regional stitching result is performed to generate a complete inspection point cloud model; the statistical outlier calculation of the complete point cloud model is performed to obtain a point cloud density distribution map, and the voxel filtering of the point cloud density distribution map is performed to generate reduced point cloud data; the world coordinate system mapping calculation of the reduced point cloud data is performed to obtain absolute coordinate point cloud, and the electrical equipment type and position annotation of the absolute coordinate point cloud are performed to generate high-precision spatial reference data.

[0100] In practical applications, first, the distribution of electrical equipment within the marked area is calculated by parameters such as the number of devices per unit area, the device spacing, and the device spatial distribution density, etc., to obtain data reflecting the density of equipment distribution within the area. Based on the device type, size, and safety distance requirements, by setting different density thresholds, the area is divided into different levels such as high density, medium density, and low density, so as to generate complexity data characterizing the scanning difficulty of the area. Then, algorithms such as hierarchical clustering or K-means (K-means clustering) are used to cluster the area complexity data, classify areas with similar complexity characteristics, form scanning areas at different levels to obtain area level data, and based on these level data, corresponding scanning parameters are configured for each level, including scanning resolution, sampling density, scanning angle, and overlap degree, etc. At the same time, considering the equipment height distribution and safety distance requirements, a hierarchical scanning scheme containing multiple scanning layers is finally generated. Then, based on the hierarchical scanning scheme, specific laser scanning parameters are set for each scanning layer, including laser emission frequency, scanning speed, angular resolution, and ranging range, etc., and these parameters are organized into a scanning control sequence according to the scanning order to control the high-precision laser scanning device carried, and according to the parameters set in the scanning control sequence, the to-be-supplemented acquisition area in the substation is scanned layer by layer, that is, the laser scanning device emits laser beams and receives reflected signals, and by measuring the flight time and return intensity of the laser signals, raw point cloud data containing spatial position information is generated. Then, multiple echo analyses are performed on the acquired raw point cloud data. By processing multiple echo signals returned by each laser pulse, the intensity characteristics of the point cloud are extracted to obtain point cloud intensity data reflecting the reflection characteristics of the target, and calibration processing is performed on these intensity data. By eliminating the interference of factors such as distance attenuation and atmospheric influence, the normalization of the intensity value is realized, and finally, raw point cloud data with reliable intensity information is generated.

[0101] Secondly, geometric feature extraction is performed on the original point cloud data. That is, by calculating feature parameters such as the local curvature, normal vector, and principal direction of the point cloud, and combining the spatial distribution characteristics and topological relationships of the point cloud, the inspection space feature data reflecting the three-dimensional structural characteristics of the electrical equipment is extracted. Then, boundary detection and fitting calculations are performed on these spatial feature data. Algorithms such as region growing and RANSAC (Random Sample Consensus) are used to identify basic geometric shapes such as planes and cylindrical surfaces in the point cloud, and the complete device boundary data is obtained through boundary tracing and surface fitting. Furthermore, the feature markers of the corresponding electrical equipment are called from the time series feature library. These feature markers contain the standard geometric model and key feature point information of the device. Then, these feature markers are feature-matched with the extracted device boundary data. That is, by calculating the similarity of local feature descriptors, the corresponding relationship between boundary points is established, a set of corresponding feature points is generated, and based on these corresponding point pairs, the rigid transformation matrix is solved using the least squares method or robust estimation method to obtain the feature registration parameters describing the point cloud registration relationship. Furthermore, the point clouds in adjacent scan sub-regions of the original point cloud data are unified into the same coordinate system through rotation and translation operations, and the preliminary regional stitching result is obtained. The registration error in the overlapping region is continuously adjusted by the iterative closest point algorithm to minimize the point pair distance in the overlapping region, and finally, an accurate complete inspection point cloud model is generated. Furthermore, statistical analysis is performed on the complete point cloud model, the local point cloud density distribution is calculated, and abnormal points are identified. Outliers are detected by setting density thresholds and distance thresholds, and a density distribution map reflecting the point cloud quality is generated. Based on this distribution map, the voxel filtering method is used to downsample and streamline the point cloud. By dividing the space into regular grids and retaining the representative points in each grid, data redundancy is reduced while maintaining the model accuracy, and compact streamlined point cloud data is obtained. Furthermore, the streamlined point cloud data is transformed from the local coordinate system to the world coordinate system. Through control point matching and coordinate transformation calculations, the corresponding relationship between the point cloud data and the actual geographical space is established, and point cloud data with absolute coordinates is obtained. Based on the previously established feature correspondence relationship, the electrical equipment type of each region in the point cloud is identified and the spatial position is marked, and finally, high-precision spatial reference data containing both accurate geometric information and device semantic annotations is generated.

[0102] Step S306: Based on the time series feature library, perform spatio-temporal feature matching on the high-precision spatial reference data and construct the device constraints of the electrical equipment to generate a three-dimensional semantic model of the substation to be inspected. Then, detect the visual dead zones corresponding to the fixed acquisition terminals in the three-dimensional semantic model and perform multi-robot path planning to obtain the visual dead zones to be inspected and the collaborative inspection paths.

[0103] Among them, in this embodiment, a three-dimensional space coordinate mapping is performed on the time series feature library to obtain feature space position data, and a distance metric calculation is performed on the feature space position data and high-precision space reference data to generate valid feature pairs; the electrical equipment boundary contour is extracted from the valid feature pairs and three-dimensional voxel reconstruction is performed to generate a device three-dimensional model, and the relative distance and azimuth angle are calculated for the device three-dimensional model to obtain a device distance matrix; hierarchical clustering of the device distance matrix is performed to generate a spatial structure tree, and electrical terminal identification and loop tracing calculation are performed on the device nodes in the spatial structure tree to generate an electrical topology diagram; the operating parameters of the device nodes in the electrical topology diagram are measured and time series sorting is performed to generate a state change sequence, and threshold comparison and pattern recognition are performed on the state change sequence to obtain a state discrimination relation; importance calculation is performed on the state discrimination relation to generate state evaluation system data, and multi-modal feature combination and semantic annotation calculation are performed on the device three-dimensional model, electrical topology diagram, and state evaluation system data to generate a three-dimensional semantic model, that is, the electrical equipment type in the multi-modal feature set is identified to obtain device category data, and functional attribute annotation is performed on the device category data to generate device semantic labels; the electrical connection relationship of the device semantic labels is analyzed to obtain loop composition data, and the working state of the loop composition data is annotated to generate loop semantic labels; the device semantic labels and loop semantic labels are hierarchically organized to obtain a semantic relationship tree, and constraint rules are defined for the semantic relationship tree to generate a three-dimensional semantic model; the perspective visibility of each electrical device in the three-dimensional semantic model is calculated to obtain device visual occlusion data, and regional union and intersection operations are performed on the device visual occlusion data to generate a visual dead zone to be inspected; the power supply range of each electrical device in the three-dimensional semantic model is identified and the load level is calculated to obtain device weight data, and a sorting operation of preset multiple inspection factors is performed on the device weight data to generate an inspection order list; the visual dead zone to be inspected is rasterized and ray traced to obtain visible area data, and the optimal inspection perspective is calculated for the visible area data to generate an effective observation point set; the inspection path length of the effective observation point set is estimated and evenly distributed for multi-robot inspection to generate a multi-robot collaborative control sequence, and path interpolation and curve fitting are performed on the collaborative control sequence to obtain continuous route data; obstacle avoidance calculation is performed on the continuous route data to generate a safe route map, and the flight distance and turning angle of the safe route map are calculated and the preset inspection path parameters are adjusted for the second time to generate a collaborative inspection path.

[0104] In practical applications, first, the electrical equipment features stored in the time-series feature library are subjected to three-dimensional space coordinate mapping. That is, feature point projection and coordinate transformation are realized through the projection transformation matrix P, and two-dimensional image features are converted into the three-dimensional space. The projection transformation matrix is: P = K[R|t], where K is the internal parameter matrix, R is the rotation matrix, and t is the translation vector. For the feature point x, the mapping relationship of its three-dimensional space coordinate X is: x = PX. Here, K includes the focal length and the principal point coordinates, and R and t describe the camera pose (for example, for the feature points (u, v) of the transformer bushing, its position (X, Y, Z) in the world coordinate system can be obtained through this transformation), so as to obtain the spatial position data of the features. By calculating the Euclidean distance between feature points and the similarity of feature descriptors, the corresponding relationship between two-dimensional features and three-dimensional point clouds is established, and the feature corresponding point pairs with high confidence are screened out, thus generating effective feature pairs. Furthermore, based on these effective feature pairs, the boundary contour information of the electrical equipment is extracted, that is, the external contour of the equipment is obtained through edge detection and contour tracking algorithms. At the same time, the voxelization method is used to convert the point cloud data into a regular three-dimensional grid structure, and on this basis, three-dimensional reconstruction is carried out to generate a complete three-dimensional model of the equipment, and the relative spatial relationship between the models is calculated, including parameters such as the Euclidean distance, projection distance, and relative azimuth angle between equipment, to construct a complete distance matrix between equipment, which is used to describe the spatial distribution characteristics of each equipment in the substation. Furthermore, the hierarchical clustering algorithm is adopted. Through the bottom-up clustering process, the equipment with similar spatial positions and related functions is organized into a hierarchical tree structure, generating a structure tree reflecting the spatial organization relationship of the equipment, and the electrical terminals of each equipment node in the structure tree are identified. By analyzing the connection characteristics and terminal shapes of the equipment, the positions and types of electrical connection points are identified, and circuit tracing is carried out along these connection points, and finally a complete electrical topology diagram is constructed to reflect the electrical connection relationship between equipment. Furthermore, the operating parameters of each equipment node in the electrical topology diagram are collected and measured, including key parameters such as voltage, current, and temperature, and these parameters are arranged in chronological order to construct a time series of the equipment operating state. Through threshold analysis and pattern recognition of these sequences, the characteristic patterns reflecting the change rules of the equipment operating state are extracted, and the relational expressions for state discrimination are established to evaluate the operating conditions of the equipment. Furthermore, the importance weights of these state discrimination relational expressions are calculated, considering factors such as equipment type, operating environment, and historical faults, to generate the complete data of the state evaluation system. Furthermore, feature fusion is carried out on the equipment three-dimensional model, electrical topology diagram, and state evaluation data. Through multi-modal feature extraction and combination, a feature set comprehensively reflecting the equipment characteristics is obtained to identify the electrical equipment type for the equipment category data. The specific category of each equipment is determined through pattern matching and classification algorithms, and functional attribute labels are added to it to generate detailed equipment semantic labels;Furthermore, analyze the electrical connection relationships among devices, determine the composition structures and connection methods of each circuit, and label the working states of the circuits based on real-time operation parameters to generate circuit semantic tags. Furthermore, hierarchically organize the device semantic tags and circuit semantic tags according to functional and spatial relationships to construct a semantic relationship tree, and define corresponding constraint rules for the semantic relationship tree based on the operation specifications and safety requirements of the power system, and finally generate a complete three-dimensional semantic model integrating geometric information, electrical characteristics, and operation states.

[0105] Secondly, conduct visibility analysis for each electrical device in the model. By calculating the field of view angle and observation distance of the fixed acquisition terminal, and combining the spatial positions and occlusion relationships of the devices, determine the visibility of each device under the existing monitoring perspectives to obtain detailed visual occlusion data. Furthermore, perform regional union and intersection operations on these occlusion data, and finally determine the range of visual dead zones that need to be supplemented for inspection by merging adjacent occlusion regions and eliminating overlapping parts. Furthermore, through the line-of-sight visibility evaluation function, track the connection relationships in the electrical topology diagram for the electrical devices in the three-dimensional semantic model. Among them, the line-of-sight visibility evaluation function is:

[0106] ;

[0107] Among them, represents the visibility score from the observation point p to the target point q, is the weight coefficient of the i-th occluder, is the occlusion metric function, is the observation distance, and λ is the distance attenuation coefficient (for example, when observing a certain circuit breaker, if there is a transformer blocking in front, and wi = 0.8, Oi = 0.6, the distance attenuation λ = 0.1, and the observation distance is 10 meters, then the visibility score is approximately 0.48, indicating a large occlusion in this perspective). To determine the power supply coverage range of each device, and at the same time, through the multi-factor comprehensive evaluation function, combine the rated capacity and actual load level of the device to calculate the weight data reflecting the importance of the device. Among them, the multi-factor comprehensive evaluation function is:

[0108] ;

[0109] Among them, W(e) is the comprehensive weight of device e, is the power supply importance index, is the power supply range coefficient, is the load level index, is the load capacity coefficient, is the reliability index, Let \(K\) be the maintenance coefficient, and \(\alpha\), \(\beta\), \(\gamma\) be the weight coefficients of each group of indicators (for example, for a main transformer, if its power supply range covers important users (\(P = 0.9\), \(S = 0.8\)), has a relatively high load level (\(L = 0.85\), \(C = 0.7\)), and strict reliability requirements (\(R = 0.95\), \(M = 0.9\)), then its comprehensive weight can reach more than \(0.85\)); then comprehensively evaluate these weight data with preset inspection factors (such as equipment failure rate, maintenance cycle, operating status, etc.), generate a reasonable inspection order list through multi-factor weighted ranking; then conduct spatial division on the determined visual dead zone, discretize the continuous space through rasterization processing, and use the ray tracing algorithm to analyze the line-of-sight accessibility from each grid point to the target device, so as to obtain detailed visible area data, and determine the viewing angle position that can obtain the best observation effect through a multi-objective optimization function, where the multi-objective optimization function is:

[0110] ;

[0111] where \(x\) is the candidate observation point, is the target point, is the observation distance, is the observation angle, \(h(x)\) is the height constraint of the observation point, and \(w_1\), \(w_2\), \(w_3\) are weight coefficients (for example, when planning the observation point of a certain circuit breaker, if \(w_1 = 0.4\), \(w_2 = 0.3\), \(w_3 = 0.3\) are set, the system will select the optimal observation position under the constraints of a 10-meter best observation distance, a 45-degree best observation angle, and a safe flight height). By comprehensively considering factors such as observation distance, viewing angle range, and imaging quality, finally generate a set of effective observation point sets; then assign inspection tasks to these observation points, estimate the path length and flight time between each observation point, combine the performance parameters and working status of multiple UAVs, achieve balanced task allocation, generate a multi-UAV collaborative control sequence, and perform path interpolation and curve smoothing processing on these discrete control points, and generate smooth and continuous route data through Bessel curve or spline function fitting; then optimize the safety of the planned route, that is, analyze the obstacle distribution in the 3D semantic model, combine the equipment safety distance requirements, dynamically adjust the route, generate a preliminary safe route map, calculate the flight distance and turning angle in the route, and then make a fine adjustment to the route according to the preset inspection parameters (such as flight speed, turning radius, obstacle avoidance distance, etc.), and finally generate a collaborative inspection path that meets the inspection requirements and ensures flight safety, providing an accurate navigation basis for the subsequent actual inspection task execution.

[0112] Step S308: Based on the collaborative inspection path, control the target UAV to inspect the visual dead zone to be inspected, obtain the supplementary inspection images, and perform feature state matching and multi-source feature fusion on the supplementary inspection images and the fixed inspection images to obtain a multi-dimensional feature state map.

[0113] Among them, extract the edge and texture features of electrical equipment and calculate local descriptors for the supplementary inspection images to generate device feature description data. Based on the historical inspection feature data corresponding to the fixed inspection images, perform nearest neighbor search and ratio test on the device feature description data to obtain a set of feature point pairs; verify the consistency of the inspection space for the set of feature point pairs to generate valid feature matching pairs, and perform time window division and sequence comparison on the valid feature matching pairs to obtain state change data, and determine the fixed inspection state and extract the time sequence of the time sequence feature library to obtain the historical state sequence and historical feature trend; perform pattern recognition on the state change data and the historical state sequence to generate a change feature map, and based on the historical feature trend, perform multi-source feature fusion on the change feature map to obtain feature fusion data; calculate the credibility and select features for the feature fusion data to generate a weighted feature map, and based on the historical inspection feature data, perform inspection space mapping and inspection parameter association on the weighted feature map to generate a state feature space; perform time sequence correlation calculation and dimension transformation on the state feature space to obtain a key state vector, and perform feature time sequence organization and spatial registration on the key state vector to generate a multi-dimensional feature state map.

[0114] In practical applications, based on the generated collaborative inspection path, multiple drones are controlled to perform collaborative inspections on visual dead zones according to a predetermined route, and supplementary image data is collected through the equipped high-definition cameras, infrared thermal imagers, and ultra-high frequency partial discharge monitoring devices. Furthermore, edge detection and texture feature extraction are performed on the obtained supplementary inspection images, that is, by calculating feature operators such as image gradients and local binary patterns, and at the same time using methods such as SIFT (Scale-Invariant Feature Transform) or SURF (Speeded-Up Robust Features) to calculate local feature descriptors, generating comprehensive device feature description data, and using the historical feature data corresponding to fixed inspection images for fixed inspections, searching for similar points in the feature space through the nearest neighbor search algorithm, and screening reliable matching points through a ratio test to establish a set of feature point pairs. Furthermore, spatial consistency verification is performed on the set of feature point pairs, and incorrect matches are removed through the RANSAC algorithm to ensure the accuracy of feature matching, obtaining reliable effective feature matching pairs, and performing temporal analysis on these matching pairs, segmenting the feature changes through setting a time window, and performing sequence comparison analysis to obtain data reflecting the device state changes. At the same time, historical state information and change trends of fixed inspections are extracted from the temporal feature library for subsequent state evaluation. Furthermore, a pattern recognition algorithm is used to analyze the state change data, that is, by identifying the change patterns and abnormal features of the device state, generating a feature map reflecting the state changes, and fusing the multi-source image features collected by the fixed terminal and the drones based on the historical feature trends.

[0115] Moreover, further through a data fusion algorithm at the feature level, considering the characteristics of different sensors and data reliability comprehensively, fused feature data is generated. Furthermore, credibility evaluation is performed on the feature fusion data, by calculating indicators such as the stability, distinctiveness, and temporal consistency of the features, assigning weight coefficients to different features, and accordingly selecting the most representative features to generate a weighted feature map, and mapping the weighted feature map into the inspection space based on the historical inspection feature data, and establishing an association with the device operation parameters to construct a state feature space containing spatial information and state features. Furthermore, for the state feature space, by calculating the temporal correlation between features and using dimensionality reduction methods such as principal component analysis to extract key features, obtaining key state vectors reflecting the device state, organizing these state vectors in chronological order, and performing precise registration with the spatial position information, finally generating a multi-dimensional feature state map integrating time, space, and state information.

[0116] Step S310, extract the electrical equipment state features from the multi-dimensional feature state map to obtain a device operation feature set, and perform threshold statistics on the device operation feature set to generate abnormal marker data.

[0117] Step S312: Calculate the electrical change quantity and the time series curve for the abnormally marked data, generate an abnormal change curve, and based on the preset operating parameters of the electrical equipment, conduct parameter comparison and deviation statistics on the abnormal change curve to obtain the operating deviation data.

[0118] Step S314: Divide the grading boundaries for the operating deviation data, generate the equipment status evaluation result, and conduct time series reconstruction and failure mode identification on the equipment status evaluation result to generate the equipment trend prediction diagram.

[0119] Step S316: Extract the risk factors and calculate the hazard degree for the trend prediction diagram, generate the risk warning diagram, and based on the risk warning diagram, integrate the inspection results of the electrical equipment to generate the substation inspection result.

[0120] Among them, extract the electrical equipment status characteristics from the multi-dimensional feature status diagram to obtain the equipment operation feature set, conduct threshold statistics on the equipment operation feature set to generate abnormally marked data; calculate the electrical change quantity and the time series curve for the abnormally marked data to generate an abnormal change curve, and based on the preset operating parameters of the electrical equipment, conduct parameter comparison and deviation statistics on the abnormal change curve to obtain the operating deviation data; divide the grading boundaries for the operating deviation data to generate the equipment status evaluation result, and conduct time series reconstruction and failure mode identification on the equipment status evaluation result to generate the equipment trend prediction diagram; extract the risk factors and calculate the hazard degree for the trend prediction diagram to generate the risk warning diagram, and based on the risk warning diagram, integrate the inspection results of the electrical equipment to generate the substation inspection result.

[0121] In the embodiments of the present application, by acquiring the inspection images of the fixed acquisition terminal, performing multi-dimensional feature extraction and marking of the areas to be supplemented on them, constructing a time-series feature library of electrical equipment, and then performing spatial registration and status annotation on the multi-angle point cloud data based on this feature library to generate high-precision spatial reference data; then generating a three-dimensional semantic model through spatio-temporal feature matching and equipment constraint construction, and performing visual dead zone detection and multi-robot path planning based on this model; subsequently controlling the drone to perform supplementary inspection on the visual dead zone, and performing feature matching and fusion on the acquired supplementary images and the fixed inspection images; finally performing anomaly detection and risk assessment on the fused multi-dimensional feature status map to output the inspection results. Through multi-stage data processing and feature analysis, the problem of visual limitations of fixed terminals in substation inspections is solved. Especially in the aspects of electrical equipment status recognition and spatial information acquisition, the collaborative advantages of fixed terminals and drones are fully utilized, effectively improving the inspection efficiency; and by adopting multi-source data fusion and three-dimensional semantic modeling strategies, not only the information complementarity between fixed terminals and drones is realized, but also the reliability of inspection results is enhanced; in addition, through feature space reconstruction and multi-robot collaboration, the abnormal status of equipment is accurately identified, thus overall realizing the efficient and precise inspection of substation equipment; in addition, state features are extracted from the multi-dimensional feature status map, and features such as the appearance status, temperature distribution, and insulation status of the equipment are extracted from visible light images, infrared thermal images, and partial discharge images respectively. At the same time, combined with the electrical parameters (such as voltage, current, power factor, etc.) and mechanical features (such as switch position, knife switch angle, etc.) of the equipment, a complete set of equipment operation features is formed, and then statistical analysis is performed on these features. By extracting the equipment features corresponding to the substation, the normal operation ranges and alarm thresholds of different types of features are set, and the abnormal status exceeding the threshold is marked to generate abnormal marking data including the abnormal position and type; based on the operation deviation data, grading criteria are set according to the equipment type and operation requirements, and the equipment status is divided into different levels such as normal, attention, warning, and danger to generate a comprehensive status assessment result. Then, time series analysis is performed on these assessment results. By reconstructing the historical state change process and combining the preset equipment failure mode library, the potential fault development trend is identified, and a trend prediction map for predicting the future state of the equipment is generated.

[0122] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this document, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0123] Based on the same inventive concept, an embodiment of the present application further provides a substation inspection device for implementing the above-mentioned substation inspection method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the substation inspection device provided below can refer to the limitations on the substation inspection method in the above text, and will not be repeated here.

[0124] In an exemplary embodiment, as Figure 4 shown, a substation inspection device is provided, including: a region marking module 401, a status annotation module 402, a model acquisition module 403, a path acquisition module 404, and a result acquisition module 405, where:

[0125] The region marking module 401 is configured to extract features from the inspection image of the substation to be inspected, obtain a time-series feature library of electrical equipment in the substation to be inspected, and mark the area to be supplemented for acquisition in the inspection image to obtain an area-to-be-supplemented acquisition mark map; wherein, the time-series feature library includes feature marks of electrical equipment; the area-to-be-supplemented acquisition mark map represents the original point cloud data of multiple inspection angles of electrical equipment.

[0126] The status annotation module 402 is configured to perform spatial registration and status annotation on the original point cloud data according to the feature marks of the electrical equipment included in the time-series feature library to obtain spatial reference data.

[0127] The model acquisition module 403 is configured to perform feature matching and construction on the spatial reference data based on the time-series feature library to obtain a three-dimensional semantic model of the substation to be inspected.

[0128] The path acquisition module 404 is configured to detect visual dead zones and plan paths for the three-dimensional semantic model to obtain the visual dead zones to be inspected and the collaborative inspection paths of the substation to be inspected.

[0129] The result acquisition module 405 is configured to control the target UAV to perform inspection on the visually dead zone to be inspected based on the collaborative inspection path, obtain supplementary inspection images, and fuse the supplementary inspection images and the inspection images to obtain the inspection result of the substation to be inspected.

[0130] Further, in one embodiment, the area marking module 401 is further configured to obtain multi-scale luminance equalized inspection image layers based on the inspection images; obtain the positioning results of the equipment in the substation to be inspected based on the multi-scale luminance equalized inspection image layers; extract feature vectors and calculate vector quality for the positioning results to obtain equipment feature quality scores, obtain a supplementary acquisition requirement map based on the equipment feature quality scores; and obtain a time series feature library of the electrical equipment in the substation to be inspected and a map of marked areas to be supplemented based on the supplementary acquisition requirement map.

[0131] Further, in one embodiment, the status annotation module 402 is further configured to obtain feature registration parameters based on the original point cloud data and the feature markings of the electrical equipment included in the time series feature library; perform spatial transformation calculations on the point cloud data of adjacent sub-regions in the original point cloud data based on the feature registration parameters until an inspection point cloud model is obtained; perform statistical outlier calculations on the point cloud model to obtain a point cloud density distribution map, and obtain filtered point cloud data based on the point cloud density distribution map; obtain coordinate point clouds based on the filtered point cloud data, and annotate the coordinate point clouds to obtain spatial reference data.

[0132] Further, in one embodiment, the model acquisition module 403 is further configured to obtain feature spatial position data based on the time series feature library, and process the feature spatial position data and the spatial reference data to obtain feature pairs; obtain a three-dimensional model of the equipment based on the feature pairs, and obtain a distance matrix between equipment associated with the substation to be inspected according to the three-dimensional model of the equipment; obtain an electrical topology diagram based on the distance matrix between equipment, and obtain a status discrimination relation formula based on the electrical topology diagram; and obtain a three-dimensional semantic model associated with the substation to be inspected based on the status discrimination relation formula, the three-dimensional model of the equipment, and the electrical topology diagram.

[0133] Further, in one embodiment, the path acquisition module 404 is further configured to obtain visual occlusion data corresponding to each electrical equipment in the substation to be inspected based on the three-dimensional semantic model, and obtain the visually dead zone to be inspected based on the visual occlusion data; obtain an observation point set based on the visually dead zone to be inspected, and obtain continuous flight path data according to the observation point set; perform obstacle avoidance calculations in the substation to be inspected based on the continuous flight path data to obtain a safe flight path map, and process the safe flight path map to obtain a collaborative inspection path for the substation to be inspected.

[0134] Further, in one embodiment, the result acquisition module 405 is further configured to perform feature state matching and multi-source feature fusion on the patrol supplement image and the patrol image to obtain a multi-dimensional feature state map; extract the state features of the electrical equipment in the multi-dimensional feature state map to obtain a device operation feature set, and obtain device anomaly marking data based on the device operation feature set; generate a device anomaly change curve based on the device anomaly marking data, and analyze the parameters in the device anomaly change curve based on the preset electrical equipment operation parameters to obtain device operation deviation data; perform grading boundary division on the device operation deviation data to obtain a device state evaluation result, and obtain a device trend prediction map based on the device state evaluation result; extract the risk factors in the device trend prediction map, and calculate the harm degree of the device trend prediction map to obtain a risk warning map; and integrate the patrol results of the electrical equipment based on the risk warning map to obtain the patrol results of the substation to be patrolled.

[0135] Each module in the above substation patrol device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0136] In an exemplary embodiment, Figure 5 FIG. is a schematic structural diagram of a substation patrol provided by an embodiment of the present application. The substation patrol device 500 may vary greatly due to configuration or performance differences, and may include one or more processors (central processing units, CPU) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 for storing application programs 533 or data 532 (for example, one or more mass storage devices). Among them, the memory 520 and the storage media 530 can be transient storage or persistent storage. The program stored in the storage media 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations for the substation patrol device 500 based on multi-machine cooperation. Further, the processor 510 can be set to communicate with the storage media 530 and execute a series of instruction operations in the storage media 530 on the substation patrol device 500.

[0137] The substation patrol device 500 may further include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input / output interfaces 560, and / or one or more operating systems 531. Those skilled in the art can understand, Figure 5The structure of the substation inspection equipment shown does not constitute a limitation on the substation inspection equipment, and it may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0138] Those skilled in the art can understand that Figure 5 the structure shown in [the figure] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer equipment to which the solution of this application is applied. The specific computer equipment may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0139] In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0140] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0141] In an embodiment, a computer program product is provided, which includes a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0143] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0144] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0145] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A substation inspection method, characterized in that: The method comprises: Extracting features from the inspection image of the substation to be inspected to obtain a time series feature library of electrical equipment in the substation to be inspected, and marking the area to be supplemented for collection on the inspection image to obtain a marking map of the area to be supplemented for collection; wherein the time series feature library includes feature markings of the electrical equipment; and the marking map of the area to be supplemented for collection represents original point cloud data of multiple inspection angles of the electrical equipment; According to the characteristic marks of the electrical equipment included in the time series characteristic library, spatial registration and state annotation are performed on the original point cloud data to obtain spatial reference data; Performing feature matching and construction on the spatial reference data based on the time series feature library to obtain a three-dimensional semantic model of the substation to be inspected; Performing visual blind spot detection and path planning on the three-dimensional semantic model to obtain the visual blind spot to be inspected and the collaborative inspection path of the substation to be inspected; The target UAV is controlled to inspect the visual blind zone to be inspected based on the collaborative inspection path to obtain an inspection supplementary image, and the inspection supplementary image and the inspection image are integrated to obtain an inspection result of the substation to be inspected.

2. The method according to claim 1, characterized in that: The method of extracting features from the inspection image of the substation to be inspected to obtain a time series feature library of electrical equipment in the substation to be inspected, and marking the area to be supplemented for collection on the inspection image to obtain a marking map of the area to be supplemented for collection, includes: Based on the inspection image, a multi-scale brightness balanced inspection image layer is obtained; Based on the multi-scale brightness balanced inspection image layer, obtaining the positioning result of the equipment in the substation to be inspected; Extracting feature vectors and calculating vector quality on the positioning results to obtain a device feature quality score, and obtaining a supplementary acquisition requirement graph based on the device feature quality score; Based on the supplementary collection requirement diagram, a time series feature library of the electrical equipment in the substation to be inspected and a marking diagram of the area to be supplemented with collection are obtained.

3. The method according to claim 1, characterized in that: The step of performing spatial registration and state labeling on the original point cloud data according to the feature mark of the electrical equipment included in the time series feature library to obtain spatial reference data includes: Obtaining feature registration parameters based on the original point cloud data and feature markers of the electrical equipment included in the time series feature library; Based on the feature registration parameters, performing spatial transformation calculation on the point cloud data of adjacent sub-areas in the original point cloud data until an inspection point cloud model is obtained; Calculating statistical outliers on the point cloud model to obtain a point cloud density distribution map, and obtaining filtered point cloud data based on the point cloud density distribution map; A coordinate point cloud is obtained based on the filtered point cloud data, and the coordinate point cloud is annotated to obtain the spatial reference data.

4. The method according to claim 1, characterized in that: The performing feature matching and construction on the spatial reference data based on the time series feature library to obtain the three-dimensional semantic model of the substation to be inspected includes: Obtaining feature space position data based on the time series feature library, and processing the feature space position data and the spatial reference data to obtain feature pairs; Obtaining a three-dimensional model of the equipment based on the feature pair, and obtaining a distance matrix between equipments associated with the substation to be inspected according to the three-dimensional model of the equipment; According to the distance matrix between the devices, an electrical topology diagram is obtained, and based on the electrical topology diagram, a state discrimination relational expression is obtained; Based on the state discrimination relational expression, the equipment three-dimensional model and the electrical topology diagram, a three-dimensional semantic model associated with the substation to be inspected is obtained.

5. The method according to claim 1, characterized in that The detecting of visual dead zones and path planning of the three-dimensional semantic model to obtain visual dead zones and collaborative inspection paths of the substation to be inspected includes: Based on the three-dimensional semantic model, visual occlusion data corresponding to each electrical device in the substation to be inspected is obtained, and based on the visual occlusion data, the visual dead zone to be inspected is obtained; Based on the visual dead zone to be inspected, an observation point set is obtained, and continuous route data is obtained according to the observation point set; Obstacle avoidance calculation is performed in the substation to be inspected based on the continuous route data to obtain a safe route map, and the safe route map is processed to obtain a collaborative inspection path for the substation to be inspected.

6. The method according to claim 5, characterized in that The fusing the inspection supplementary image and the inspection image to obtain the inspection result of the substation to be inspected includes: Performing feature state matching and multi-source feature fusion on the inspection supplementary image and the inspection image to obtain a multi-dimensional feature state diagram; Extracting the state characteristics of the electrical equipment in the multidimensional characteristic state diagram to obtain a device operation characteristic set, and obtaining device abnormality marking data based on the device operation characteristic set; Generate an equipment abnormality change curve based on the equipment abnormality mark data, and analyze the parameters in the equipment abnormality change curve based on preset electrical equipment operation parameters to obtain equipment operation deviation data; The equipment operation deviation data is divided into graded boundaries to obtain equipment status evaluation results, and based on the equipment status evaluation results, an equipment trend prediction diagram is obtained; Extracting risk factors from the equipment trend prediction graph, and calculating the degree of harm of the equipment trend prediction graph to obtain a risk warning graph; Based on the risk warning map, the inspection results of the electrical equipment are integrated to obtain the inspection results of the substation to be inspected.

7. A substation inspection device, characterized in that: The device comprises: The area marking module is used to extract features from the inspection image of the substation to be inspected, obtain a time series feature library of electrical equipment in the substation to be inspected, and mark the inspection image for the area to be supplemented with acquisition, to obtain a marking map of the area to be supplemented with acquisition; wherein the time series feature library includes the feature markings of the electrical equipment; the marking map of the area to be supplemented with acquisition represents the original point cloud data of multiple inspection angles of the electrical equipment; A state labeling module, used to perform spatial registration and state labeling on the original point cloud data according to the feature marks of the electrical equipment included in the time series feature library to obtain spatial reference data; A model acquisition module, used for performing feature matching and construction on the spatial reference data based on the time series feature library to obtain a three-dimensional semantic model of the substation to be inspected; A path acquisition module, used to detect visual dead zones and plan paths for the three-dimensional semantic model, and obtain visual dead zones and collaborative inspection paths for the substation to be inspected; The result acquisition module is used to control the target UAV to inspect the visual blind zone to be inspected based on the collaborative inspection path, obtain a supplementary inspection image, and fuse the supplementary inspection image with the inspection image to obtain the inspection result of the substation to be inspected.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images