Transformer substation three-dimensional flood prevention monitoring system and method based on multi-source heterogeneous data fusion

Abstract

The invention belongs to the technical field of flood prevention monitoring, and discloses a transformer substation three-dimensional flood prevention monitoring system and method based on multi-source heterogeneous data fusion. Comprising the steps of collecting oblique photography and laser radar data of a transformer substation, and constructing a three-dimensional model of the transformer substation; acquiring associated data such as associated data of equipment and meteorological associated data, and importing the associated data to a corresponding position of the three-dimensional model; performing analysis based on the associated data of the equipment, and performing real-time alarm monitoring and early warning monitoring on the equipment of the transformer substation; performing graded flood prevention early warning on the basis of analysis of meteorological associated data and associated data of the equipment; by integrating the state data of the equipment and the meteorological associated data for comprehensive analysis, the one-sidedness defect of single-dimensional flood prevention risk assessment is overcome, multi-dimensional risk collaborative quantification of equipment hidden dangers and meteorological factors is realized, and early warning has higher perspectiveness.

Description

Technical Field

[0001] This invention relates to the field of flood control monitoring technology, and more specifically, to a three-dimensional flood control monitoring system and method for substations that integrates multi-source heterogeneous data. Background Technology

[0002] As the core hub of the power system, the safe and stable operation of substations is directly related to the reliability of regional power supply. However, the risks of heavy rainfall and waterlogging during the flood season can easily lead to equipment insulation failure, short circuit tripping, or even power outages. Therefore, flood prevention monitoring is one of the key aspects of substation operation and maintenance management.

[0003] Currently, flood prevention monitoring of substations requires the construction of a 3D model. Existing model construction methods often rely on point cloud data from lidar. While lidar can relatively completely reflect the geometric structure of a substation, it lacks the ability to represent features such as texture, markings, and rust marks. These details are not displayed, resulting in an inaccurate 3D model. Furthermore, current technologies often treat real-time equipment status monitoring and regional flood risk assessment as independent modules, without linking equipment anomalies with surrounding meteorological and terrain risks. This can easily overlook hidden risks such as related equipment failures leading to flood control facility malfunctions or deteriorating weather trends causing water accumulation in the equipment area, resulting in delayed or missed early warnings.

[0004] In view of this, the present invention proposes a three-dimensional flood control monitoring system and method for substations based on multi-source heterogeneous data fusion to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: A three-dimensional flood control monitoring method for substations using multi-source heterogeneous data fusion, the method comprising: Collect three-dimensional spatial perception data of the substation, and construct a three-dimensional model of the substation based on the analysis of the three-dimensional spatial perception data; Collect relevant data related to flood control at substations and import the relevant data into the corresponding locations in the 3D model. The relevant data includes equipment-related data and meteorological data. Based on the analysis of the equipment's own associated data, real-time alarm monitoring and early warning monitoring are carried out on each piece of equipment in the substation. The substation is divided into multiple grid units. Based on the analysis of the equipment's own correlation data and meteorological correlation data within each grid unit, flood control risk values ​​are generated, and graded flood control early warnings are carried out based on the flood control risk values.

[0006] Furthermore, the method for constructing a three-dimensional model of a substation is as follows: Three-dimensional spatial perception data includes oblique photogrammetry image data, oblique photogrammetry point cloud data, and lidar point cloud data; Based on the equipment structure of the substation, the substation is divided into multiple blocks; a three-dimensional coordinate system is constructed, and each block has its own coordinate position in the three-dimensional coordinate system; oblique photography point cloud data and lidar point cloud data of each block are acquired and collectively referred to as point cloud data; the point cloud data is analyzed to determine whether there is occlusion in each block; blocks with occlusion are marked as occluded blocks, and blocks without occlusion are marked as unoccluded blocks; the oblique photography point cloud data of the occluded blocks are filled, and a three-dimensional model is constructed based on the filled point cloud data of each occluded area and the point cloud data of each unoccluded block.

[0007] Furthermore, the method for determining whether each block is obstructed is as follows: Based on oblique photogrammetry point cloud data, the oblique photogrammetry point cloud coverage value, oblique photogrammetry normal vector, and oblique photogrammetry elevation value of each block are obtained. Based on LiDAR point cloud data, the LiDAR point cloud coverage value, LiDAR normal vector, and LiDAR elevation value are obtained. The point cloud coverage difference value of each block is calculated based on the oblique photogrammetry point cloud coverage value and the LiDAR point cloud coverage value. The normal vector deviation of each block is calculated based on the oblique photogrammetry normal vector and the LiDAR normal vector. The elevation difference value of each block is calculated based on the oblique photogrammetry elevation value and the LiDAR elevation value. After normalization, the point cloud coverage difference value, normal vector deviation, and elevation difference value are weighted and accumulated to obtain the difference index of each block. The difference index of each block is compared with the preset difference index judgment threshold for each block. If the difference index is less than or equal to the difference index judgment threshold, it is determined that the corresponding block has no occlusion; otherwise, the oblique photogrammetry image data of the corresponding block is retrieved for viewpoint backtracking verification. If the verification passes, it is determined that the block has no occlusion; otherwise, it is determined that the block has occlusion.

[0008] Furthermore, the method for filling the oblique photographic point cloud data of the occluded area is as follows: Unobstructed blocks adjacent to obstructed blocks are designated as adjacent safe blocks. LiDAR point cloud data of these adjacent safe blocks are acquired and labeled as adjacent LiDAR point cloud data. Multi-granularity semantic segmentation is performed on the adjacent LiDAR point cloud data to identify and label equipment components within the adjacent safe blocks. An equipment topology graph is established, where nodes represent equipment components, and edges represent connections, spatial constraints, and functional dependencies. Simultaneously, a substation equipment knowledge base is constructed, storing standard design parameters and maintenance history data for equipment components. Multi-dimensional features of equipment components are extracted from adjacent safe blocks, forming multi-dimensional feature vectors. After normalizing the feature vectors of each dimension, a feature fusion network is used to fuse the feature vectors of each dimension into a unified feature representation vector. The representation vectors are dimensionally compressed to obtain the device semantic DNA codes of device components in adjacent security blocks. Based on the device topology map and multi-dimensional feature vectors, a set of filling constraint rules for device components is generated. The device semantic DNA codes of device components in adjacent security blocks are used as the standard semantic DNA codes of corresponding device components in occluded blocks. From the LiDAR point cloud data of occluded blocks aligned with the same geographic coordinate system as the oblique photogrammetry point cloud data, valid point clouds that match the standard semantic DNA codes and satisfy the filling constraint rule set are selected and filled into the oblique photogrammetry point cloud data of the occluded blocks. Based on the standard design parameters and maintenance history data of device components in adjacent security blocks in the device knowledge base, the filled oblique photogrammetry point cloud data is optimized, and a 3D model is constructed based on the optimized oblique photogrammetry point cloud data.

[0009] Furthermore, the method for real-time alarm monitoring of various equipment in the substation is as follows: Acquire various self-related data of each device in the substation and import the self-related data into the corresponding position of each device in the 3D model. The self-related data includes self-directed data and self-indirected data. Acquire the self-directed data of each device in real time and compare the self-directed data with the preset self-directed data allowable working range. When the self-directed data is not within the corresponding self-directed data allowable working range, it is determined that the device is abnormal and a real-time alarm is triggered. When the directly related data is within the corresponding allowed working range, multiple indirectly related data related to the device are acquired. Each indirectly related data item has a corresponding allowed working range. The number of indirectly related data items that are not within the corresponding allowed working range is counted to obtain the indirectly related data trigger value. The indirectly related data trigger value is compared with the preset indirectly related data trigger threshold. When the indirectly related data trigger value is greater than the indirectly related data trigger threshold, the device is judged to be abnormal and a real-time alarm is triggered.

[0010] Furthermore, the method for early warning monitoring of various equipment in the substation is as follows: When it is determined that there is no abnormality in the equipment, the equipment is recorded as the target equipment. Based on the position of the target equipment in the 3D model, a network of equipment associations with the target equipment is constructed, and the equipment in the network of equipment associations that are associated with the target equipment is recorded as the associated equipment. The system acquires the direct correlation data of the target device under time series T, constructs the direct correlation data function of the target device, and simultaneously acquires the direct correlation data of each associated device under the corresponding time series T in the device association network, constructing the direct correlation data function of each associated device. Each target device and each associated device has a corresponding ideal direct correlation data function. A Cartesian coordinate system is established with time as the x-axis and direct correlation data as the y-axis. In the Cartesian coordinate system, under time series T, the area enclosed by the direct correlation data function of the target device and its corresponding ideal direct correlation data function is recorded as the first area. The first area is normalized to obtain the first state coefficient. The area enclosed by the direct correlation data function of each associated device and its corresponding ideal direct correlation data function is recorded as the second area. The second areas of each associated device are weighted, accumulated, and then normalized to obtain the second state coefficient. The first and second state coefficients are weighted and accumulated to obtain the total state coefficient. The state coefficient is compared with a preset state coefficient threshold. When the total state coefficient is greater than the state coefficient threshold, it is determined that the device has an abnormal risk, and an early warning is issued.

[0011] Furthermore, the method for generating flood control risk values ​​is as follows: Topographic data of the substation and its surrounding area are obtained from a 3D model. The topographic data is divided into multiple grid cells using a digital elevation model (DEM) to form a DEM grid. Each grid cell carries corresponding meteorological data and equipment-related data. A spatial graph is constructed using the grid cells in the DEM grid as nodes and spatial adjacency relationships as edges. In the spatial graph, based on the spatial adjacency relationships of each grid cell, any grid cell is selected as the target grid cell, and the grid cells upstream of the target grid cell are designated as influencing grid cells. The adjacency weights between the target grid cell and each influencing grid cell are obtained based on the Euclidean distance between each influencing grid cell and the target grid cell. A first risk value is generated based on the equipment-related data of the grid cell, and a second risk value is generated based on the meteorological data of the grid cell. The first risk value and the second risk value are added together to obtain the flood control risk value.

[0012] Furthermore, the methods for obtaining the first risk value and the second risk value are as follows: Based on the analysis results of the corresponding device's own associated data within the grid cell, the total state coefficient of each device is extracted. The average of the total state coefficients of each device is calculated to obtain the device state value of the grid cell. The device state values ​​of each influencing grid cell are multiplied by the adjacency weight and then summed to obtain the adjacent device state value. The device state value of the target grid cell and the adjacent state values ​​are weighted, summed, and normalized to obtain the first risk value. Meteorological correlation data for each grid cell is acquired and organized into a time input sequence at a fixed time step to obtain the time input sequence for each grid cell. The time input sequence is then input into an LSTM network, where the gating mechanism captures the risk trend of the grid cell in the time dimension. The risk trend at the last time step in the LSTM network is taken as the temporal risk value of the grid cell. The temporal risk values ​​of each influencing grid cell are multiplied by the adjacency weights and then summed to obtain the adjacency temporal risk value. Finally, the temporal risk value of the target grid cell and the adjacency temporal risk values ​​are weighted, summed, and normalized to obtain the second risk value.

[0013] Furthermore, the method for tiered flood warning based on flood risk values ​​is as follows: Flood warnings include four levels: high-risk warning, medium-low-risk warning, low-risk warning, and zero-risk warning. Each warning level has a corresponding flood risk value range. When the flood risk value falls within the range corresponding to the low-risk warning, a low-risk warning is issued. When the flood risk value falls within the range corresponding to the medium-risk warning, a medium-risk warning is issued. When the flood risk value falls within the range corresponding to the high-risk warning, a high-risk warning is issued. When the flood risk value falls within the range corresponding to the zero-risk warning, no warning is issued.

[0014] A three-dimensional flood control monitoring system for substations based on multi-source heterogeneous data fusion. The system includes: The 3D model building module is used to collect 3D spatial perception data of the substation and build a 3D model of the substation based on the analysis of the 3D spatial perception data. The multi-source data acquisition module is used to collect relevant data related to flood control in substations and import the relevant data into the corresponding location of the 3D model. The relevant data includes equipment-related data and meteorological data. The equipment monitoring and early warning module is used to analyze the equipment's own associated data and perform real-time alarm monitoring and early warning monitoring for each piece of equipment in the substation. The flood control monitoring and early warning module is used to divide the substation into multiple grid units, analyze the data related to the equipment in each grid unit and the meteorological data, generate flood control risk values, and conduct graded flood control early warning based on the flood control risk values.

[0015] The technical effects and advantages of the multi-source heterogeneous data fusion substation three-dimensional flood control monitoring system and method of the present invention are as follows: This invention uses LiDAR technology and oblique photogrammetry to construct a three-dimensional model. The LiDAR point cloud data can fill the structural gaps in the oblique photogrammetry, while retaining the texture information of the oblique photogrammetry and mapping it onto the completed blocks. This results in a three-dimensional model that has both the complete geometric structure guaranteed by LiDAR and the realistic texture appearance guaranteed by oblique photogrammetry, thus more accurately constructing the most realistic three-dimensional model of the substation.

[0016] This invention divides the substation terrain into grid cells and, based on the elevation difference and Euclidean distance analysis of upstream influencing grid cells, performs spatial correlation risk weighting on each grid cell to obtain corresponding adjacency weights. This enables precise capture of flood control risk correlations between different areas, thereby improving flood control accuracy. At the same time, by integrating and comprehensively analyzing equipment status data and meteorological correlation data, it solves the one-sidedness of assessing flood control risk from a single dimension, achieving multi-dimensional risk synergistic quantification of equipment hazards and meteorological factors, making early warning more forward-looking. Attached Figure Description

[0017] Figure 1 This is a flowchart of a three-dimensional flood control monitoring method for substations based on multi-source heterogeneous data fusion, according to the present invention. Figure 2 This is a block diagram of a three-dimensional flood control monitoring system for substations based on multi-source heterogeneous data fusion, according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown in the figure, this embodiment discloses a three-dimensional flood control monitoring method for substations based on multi-source heterogeneous data fusion. The method mainly includes: Collect three-dimensional spatial perception data of the substation, and construct a three-dimensional model of the substation based on the analysis of the three-dimensional spatial perception data; Collect relevant data related to flood control at substations and import the relevant data into the corresponding locations in the 3D model. The relevant data includes equipment-related data and meteorological data. Based on the analysis of the equipment's own associated data, real-time alarm monitoring and early warning monitoring are carried out on each piece of equipment in the substation. The substation is divided into multiple grid units. Based on the analysis of the equipment's own correlation data and meteorological correlation data within each grid unit, flood control risk values ​​are generated, and graded flood control early warnings are carried out based on the flood control risk values.

[0020] Through the above scheme, this application first constructs a 3D model of the substation by collecting 3D spatial perception data, including oblique photography and LiDAR data. For example, a hexacopter UAV is used to conduct oblique photography of the substation, while ground-based LiDAR scans the substation terrain, main transformers, cable trenches, drainage pumps, and other facilities. Through image matching and point cloud fusion analysis, a centimeter-level 3D model of the substation, including equipment spatial location and terrain elevation, is constructed. If necessary, the model can be optimized for substations of key interest, such as disassembling the component levels of key equipment (transformers, switchgear, etc.) on the 3D model and collecting the structural parameters and material properties of each component. The system maps material properties (such as the waterproof rating of the insulation layer) and connection relationships (such as the fixing method of the wiring terminals) one by one into the 3D model. This accurately reproduces the structure, materials, and connection details of key equipment, allowing for a more intuitive understanding of the actual situation of key equipment and enabling more accurate decision-making. Then, it collects relevant data related to flood control in the substation and imports this data into the corresponding locations in the 3D model. This data includes equipment-specific data and meteorological data for easy viewing by maintenance personnel. Further analysis of the equipment-specific data enables real-time alarm and early warning monitoring of various substation devices. This allows for real-time analysis and early warning analysis based on the equipment's own data to determine if there are any abnormalities or risks, enabling timely handling and ensuring the equipment is in good working condition, thus guaranteeing normal flood control. Finally, the data obtained from the 3D model divides the substation into multiple grid units. Based on the analysis of equipment-specific and meteorological data within each grid unit, flood risk values ​​are generated. Based on these flood risk values, graded flood warnings are issued. This approach not only analyzes the equipment's own condition but also analyzes meteorological data such as precipitation and duration to accurately predict flood risks. The application assesses the flood control capabilities of grid units to provide corresponding risk warnings and improve operation and maintenance response efficiency. Furthermore, to further verify the flood control risk assessment results and enhance the scientific nature of flood control decisions, this application can also introduce a disaster simulation model. Based on the three-dimensional model and grid unit division results, simulations of natural disaster scenarios such as rainstorms are conducted. By simulating the impact of disasters on each grid unit of the substation, the impact of equipment under different disaster conditions and the overall flood control performance are evaluated. This allows for a comprehensive assessment of the substation's overall flood control capabilities during the flood season, and the formulation or optimization of corresponding flood control measures to further strengthen the substation's flood control construction.

[0021] The method for constructing a 3D model of a substation is as follows: 3D spatial perception data includes oblique photogrammetry image data, oblique photogrammetry point cloud data, and lidar point cloud data; based on the equipment structure of the substation, the substation is divided into multiple blocks; a 3D coordinate system is constructed, with each block corresponding to its own coordinate position in the 3D coordinate system; oblique photogrammetry point cloud data and lidar point cloud data of each block are acquired and collectively referred to as point cloud data; based on the point cloud data, it is analyzed to determine whether there is occlusion in each block, and blocks with occlusion are marked as occluded blocks, while blocks without occlusion are marked as unoccluded blocks; the oblique photogrammetry point cloud data of the occluded blocks are filled, and a 3D model is constructed based on the filled point cloud data of each occluded area and the point cloud data of each unoccluded block; The method for determining whether each block has occlusion is as follows: Based on oblique photogrammetry point cloud data, obtain the oblique photogrammetry point cloud coverage value, oblique photogrammetry normal vector, and oblique photogrammetry elevation value of each block; based on LiDAR point cloud data, obtain the LiDAR point cloud coverage value, LiDAR normal vector, and LiDAR elevation value; calculate the point cloud coverage difference value of each block based on the oblique photogrammetry point cloud coverage value and the LiDAR point cloud coverage value, calculate the normal vector deviation of each block based on the oblique photogrammetry normal vector and the LiDAR normal vector, and calculate the elevation difference value of each block based on the oblique photogrammetry elevation value and the LiDAR elevation value; after normalizing the point cloud coverage difference value, normal vector deviation, and elevation difference value, assign weights and sum them to obtain the difference index of each block; compare the difference index of each block with the preset difference index judgment threshold of each block; if the difference index is less than or equal to the difference index judgment threshold, it is determined that the corresponding block does not have occlusion; otherwise, the oblique photogrammetry image data of the corresponding block is retrieved for viewpoint backtracking verification; if the verification passes, it is determined that the block does not have occlusion; otherwise, it is determined that the block has occlusion. The method for filling the oblique photogrammetric point cloud data of occluded blocks is as follows: Unoccluded blocks adjacent to occluded blocks are designated as adjacent safe blocks. LiDAR point cloud data of adjacent safe blocks are acquired and labeled as adjacent LiDAR point cloud data. Multi-granularity semantic segmentation is performed on the adjacent LiDAR point cloud data to identify and label the equipment components of adjacent safe blocks. An equipment topology graph is established, where nodes represent equipment components, and edges represent connection relationships, spatial constraints, and functional dependencies. Simultaneously, a substation equipment knowledge base is constructed, storing standard design parameters and maintenance history data for equipment components. Multi-dimensional features of equipment components are extracted from adjacent safe blocks to form multi-dimensional feature vectors. After normalizing the dimensional feature vectors of each dimension, a feature fusion network is used to fuse the dimensional feature vectors of each dimension into a unified vector. The feature representation vectors are used to compress the dimensions of the feature representation vectors to obtain the device semantic DNA codes of device components in adjacent security blocks. Based on the device topology map and multi-dimensional feature vectors, a set of filling constraint rules for device components is generated. The device semantic DNA codes of device components in adjacent security blocks are used as the standard semantic DNA codes of device components in occluded blocks. From the lidar point cloud data of occluded blocks aligned with the same geographic coordinate system as the oblique photogrammetry point cloud data, valid point clouds that match the standard semantic DNA codes and satisfy the filling constraint rule set are selected and filled into the oblique photogrammetry point cloud data of the occluded blocks. Based on the standard design parameters and maintenance history data of device components in adjacent security blocks in the device knowledge base, the filled oblique photogrammetry point cloud data is optimized, and a three-dimensional model is constructed based on the optimized oblique photogrammetry point cloud data.

[0022] The above scheme provides a specific method for constructing a 3D model of a substation. First, based on the substation's equipment structure, the substation is divided into multiple blocks, each corresponding to a specific coordinate position in the 3D coordinate system. Since oblique photogrammetry point cloud data is mostly generated through stereo matching of multi-view images, although it can provide realistic texture information, it can only acquire point cloud data of the equipment surface and cannot penetrate the equipment's interior. LiDAR point cloud data, on the other hand, is mainly generated by actively emitting laser pulses and receiving echoes; it can acquire the equipment's true geometric contour information but lacks the ability to represent equipment texture. Therefore, in order to accurately determine the blocks... To determine if an area is occluded, firstly, based on oblique photogrammetry point cloud data, obtain the oblique photogrammetry point cloud coverage value, oblique photogrammetry normal vector, and oblique photogrammetry elevation value for each block. Secondly, based on LiDAR point cloud data, obtain the LiDAR point cloud coverage value, LiDAR normal vector, and LiDAR elevation value. Specifically, the point cloud coverage value for each block can be obtained by statistically analyzing the number of point clouds from both oblique photogrammetry and LiDAR data. The elevation value and normal vector are then calculated by fitting a local point cloud to a plane. Finally, based on the oblique photogrammetry and LiDAR point cloud coverage values, calculate the point cloud coverage difference value for each block. The point cloud coverage difference value = (LiDAR point cloud coverage value + LiDAR normal vector ... Coverage value (point cloud coverage value from oblique photogrammetry) / LiDAR point cloud coverage value represents the difference in point cloud coverage between LiDAR and oblique photogrammetry. A smaller difference in coverage indicates higher consistency in coverage and a lower likelihood of occlusion. Normal vector deviation is calculated for each block based on the oblique photogrammetry and LiDAR normal vectors. Normal vector deviation is an indicator used to compare the differences in surface normal vector directions between two point cloud datasets in the same area. In 3D point clouds, normal vectors describe the surface orientation; therefore, in unoccluded areas, the two point clouds should have similar normal vectors, while in occluded areas… Due to missing or mismatched point cloud data, normal vectors may exhibit significant differences. Therefore, an average normal vector is derived from the oblique photogrammetry point cloud data and LiDAR point cloud data of the block, and the angle between these two normal vectors is calculated. The smaller the angle, the lower the likelihood of occlusion. Then, based on the oblique photogrammetry elevation values ​​and LiDAR elevation values, the elevation difference value for each block is calculated. The elevation difference value quantifies the degree of difference between the oblique photogrammetry point cloud data and the LiDAR point cloud data in the vertical direction. The smaller the value, the stronger the elevation consistency between the two point clouds, and the higher the likelihood that the block is unoccluded. The elevation difference value is obtained by first obtaining... and The absolute value of the difference between them is denoted as the first difference value, and then... and The absolute value of the difference between the first and second differences is denoted as the second difference value. The elevation difference value is obtained by adding the first and second difference values. Therefore, after normalizing the point cloud coverage difference value, normal vector deviation, and elevation difference value, they are weighted and accumulated. The weights of these three values ​​can be determined based on their correlation with the occlusion phenomenon. For example, a higher weight can be given to the point cloud coverage difference value (e.g., 0.5), followed by the normal vector deviation value (e.g., 0.3), and a lower weight to the elevation difference value (e.g., 0.2), resulting in the difference index for each block. A comprehensive analysis of these three values ​​can more accurately indicate whether occlusion exists in each block. The smaller the difference index, the lower the probability of occlusion. Therefore, by statistically analyzing the difference index samples of occluded blocks in the same scene and taking the upper limit of the 95% confidence interval of their distribution, a difference index judgment threshold is proposed for each block. The difference index of each block is compared with the preset difference index judgment threshold for the corresponding block. If the difference index is higher... If the difference index is less than or equal to the threshold for judgment, the block is judged to be occluded; otherwise, it indicates that occlusion may exist and further judgment is required. The further judgment method is to retrieve the oblique photogrammetry image data of the corresponding block for viewpoint backtracking verification. For example, retrieve the shooting equipment parameters (such as camera pose and shooting angle) corresponding to the oblique photogrammetry image to reconstruct the original viewpoint of the block at the time of shooting and observe whether there is an object occlusion in the image from that viewpoint. If there is no occlusion, the verification is passed; otherwise, if there is occlusion, the verification is failed, and the block is judged to be occluded. By integrating multiple dimensions such as point cloud coverage, normal vector, and elevation value to comprehensively analyze and calculate the difference index, the misjudgment problem of existing single index judgment being easily affected by factors such as object reflectivity is reduced, and the accuracy of occlusion judgment is improved. At the same time, by combining the threshold of each block and viewpoint backtracking verification for further judgment, the accuracy and scene adaptability of occlusion judgment are improved, while also taking into account the judgment precision and efficiency.

[0023] After determining that an area is occluded, the point cloud data of the occluded area needs to be filled in. Specifically, the non-occluded areas adjacent to the occluded area are recorded as adjacent safe areas. The LiDAR point cloud data of the adjacent safe areas is obtained and marked as adjacent LiDAR point cloud data. Multi-granularity semantic segmentation is performed on the adjacent LiDAR point cloud data to identify and label the equipment components of the adjacent safe areas. An equipment topology graph is established, where nodes represent equipment components and edges represent connections, spatial constraints, and functional dependencies. At the same time, a substation equipment knowledge base is constructed, which stores the standard design parameters and maintenance history data of equipment components. The equipment topology graph is established by mapping nodes to equipment components and edges to... The graph structure corresponding to the connection relationships, spatial constraints, and functional dependencies between equipment components is constructed using common techniques in the field of semantic modeling of point clouds for industrial equipment, which will not be elaborated upon here. For example, if the equipment component identified and labeled after multi-granularity semantic segmentation is a 10kV switchgear-door panel-metal insulation composite material, the graph can be constructed by labeling the cabinet, door panel, busbar, etc., at nodes, and labeling the connection relationship between the door panel and the cabinet, the spatial constraints of the busbar inside the cabinet, etc. Similarly, a substation equipment knowledge base is built in advance, which contains the standard design parameters and maintenance history data of each equipment component, such as the standard dimensions of the 10kV switchgear and historical maintenance pictures and records. Then, from adjacent safety zones... The multi-dimensional features of equipment components are extracted from the block, and multi-dimensional feature vectors are constructed. These multi-dimensional features include geometric features (such as size parameters, shape description, surface curvature, and normal vectors), topological features (such as connectivity, adjacency relationships, and hierarchical structure), functional features (such as voltage level, current capacity, and safety distance requirements), and material features (such as reflectivity, thermal conductivity, and aging characteristics). Each of these multi-dimensional features is then used to construct a dimensional feature vector for its respective dimension. For example, a geometric dimensional feature vector is constructed based on parameters such as size parameters, shape description, surface curvature, and normal vectors; a topological dimensional feature vector is constructed based on connectivity, adjacency relationships, and hierarchical structure; and a dimensional feature vector is constructed based on voltage level, current capacity, and safety distance requirements. Functional dimension feature vectors are constructed based on distance requirements, and material dimension feature vectors are constructed based on reflectivity, thermal conductivity, and aging characteristics. Then, the dimensional feature vectors of each dimension in the multi-dimensional feature vectors are normalized separately. A feature fusion network is then used to fuse the feature vectors of each dimension into a unified feature representation vector. Finally, the feature representation vector is dimensionally compressed to obtain a unique device semantic DNA code. The dimensionality compression of the feature representation vector can be achieved using Principal Component Analysis (PCA), a method commonly used in industrial feature dimensionality reduction, to extract the principal components of the feature representation vector to reduce dimensionality. The compressed low-dimensional vector is then mapped to a unique string-based code to obtain the device semantic DNA code.For example, the feature representation vector (a 128-dimensional vector integrating geometric, topological, functional, and material features) of a 10kV switchgear door panel in an adjacent safety block is compressed into a 16-dimensional principal component vector using PCA. Then, the values ​​of each dimension of the vector are converted into a unique string of 10kV switchgear-door panel-0.85_0.23_0.17_… (16-dimensional values) according to the rule of equipment type-component name-principal component vector value, thus obtaining the equipment semantic DNA encoding for that component. Based on the equipment topology map and multi-dimensional feature vectors, a set of filling constraint rules for each equipment component is generated. The filling constraint rule set for each equipment component can be used to generate rules based on the equipment topology map. The connection or hierarchical relationship of components, combined with geometric, functional, and material features from multiple dimensions, is used to extract spatial constraints, structural matching, and functional compliance conditions between components, and then organize them into a set of rules. This is a common technique in the field of semantic point cloud modeling for industrial equipment, and will not be elaborated on here. For example, for the filling constraint rule set of a 10kV switchgear door panel obstruction block, the generation logic is to extract the hierarchical relationship of the door panel as an outer layer component of the cabinet from the equipment topology map, combine the door panel size (geometric feature), the busbar safety distance (functional feature), and the metal reflection (material feature) to organize the corresponding filling constraint rule set; then, the equipment semantic DN of adjacent safety block equipment components is used to form a set of rules. The A-code serves as the standard semantic DNA code for the equipment components corresponding to the occluded area. Valid point clouds matching the standard semantic DNA code and satisfying the filling constraint rule set are selected from the LiDAR point cloud data of the occluded area, aligned to the same geographic coordinate system as the oblique photogrammetry point cloud data. These valid point clouds are then filled into the oblique photogrammetry point cloud data of the occluded area. Based on the standard design parameters and maintenance history data of the corresponding equipment components in the equipment knowledge base, the filled oblique photogrammetry point cloud data is optimized, and a 3D model is constructed based on the optimized oblique photogrammetry point cloud data. For example, the color and rust marks of the equipment are extracted from images of the corresponding equipment in the historical maintenance records of the door panel. Details such as appearance are filled in to obtain filled point cloud data, making the point cloud data more realistic. Finally, when there is no occlusion, a 3D model is constructed based on the oblique photogrammetry point cloud data and LiDAR point cloud data in each block. For blocks with occlusion, a 3D model is constructed based on the optimized oblique photogrammetry point cloud data. In this way, LiDAR point cloud data can fill in the structural gaps of oblique photogrammetry, while preserving the texture information of oblique photogrammetry and mapping it to the completed blocks. This ensures that the constructed 3D model has both the complete geometric structure guaranteed by LiDAR and the realistic texture appearance guaranteed by oblique photogrammetry, thus more accurately constructing a realistic 3D model of the substation.

[0024] The method for real-time alarm monitoring of various equipment in a substation is as follows: Various self-related data of each substation equipment are acquired and imported into the corresponding positions of each equipment in a 3D model. Self-related data includes direct and indirect self-related data. Direct self-related data of each equipment is acquired in real-time and compared with a preset allowable working range. If the direct self-related data is not within the corresponding allowable working range, an anomaly is detected, and a real-time alarm is triggered. If the direct self-related data is within the corresponding allowable working range, multiple indirect self-related data related to the equipment are acquired. Each indirect self-related data item has a corresponding allowable working range. The number of indirect self-related data items that are not within the corresponding allowable working range is counted to obtain the indirect self-related data trigger value. This trigger value is compared with a preset indirect self-related data trigger threshold. If the trigger value is greater than the trigger threshold, an anomaly is detected, and a real-time alarm is triggered.

[0025] The method for early warning monitoring of various equipment in a substation is as follows: When it is determined that there is no abnormality in the equipment, the equipment is recorded as the target equipment. Based on the position of the target equipment in the 3D model, an equipment association network is constructed that is associated with the target equipment. Equipment in the equipment association network that is associated with the target equipment is recorded as associated equipment. The self-direct association data of the target equipment under time series T is obtained, and the self-direct association data function of the target equipment is constructed. At the same time, the self-direct association data of each associated equipment under the corresponding time series T in the equipment association network is obtained, and the self-direct association data function of each associated equipment is constructed. The target equipment and each associated equipment are all set with corresponding ideal self-direct association data functions. A right angle is established with time as the x-axis and self-direct association data as the y-axis. In a Cartesian coordinate system, under a time series T, the area enclosed by the target device's directly correlated data function and its corresponding ideal directly correlated data function is denoted as the first area. The first area is normalized to obtain the first state coefficient. The area enclosed by the directly correlated data function and its corresponding ideal directly correlated data function of each associated device is denoted as the second area. The second areas of each associated device are weighted, accumulated, and then normalized to obtain the second state coefficient. The first and second state coefficients are weighted and accumulated to obtain the total state coefficient. The state coefficient is compared with a preset state coefficient threshold. When the total state coefficient is greater than the state coefficient threshold, it is determined that the device has an abnormal risk, and an early warning is issued.

[0026] The above technical solution provides a specific method for real-time alarm monitoring and early warning monitoring of various equipment in a substation. First, real-time alarm monitoring of the equipment involves acquiring various self-correlated data of each piece of equipment in the substation. For example, if the equipment is a water pump, its self-correlated data may include temperature, pressure, vibration amplitude, etc. If the equipment is a transformer, its self-correlated data may include winding temperature, oil temperature, oil pressure, load current, ambient temperature, etc. This self-correlated data is then imported into the corresponding location of each piece of equipment in a 3D model for easy viewing by maintenance personnel. The self-correlated data includes both direct and indirect self-correlated data. To increase detection accuracy, both direct and indirect self-correlated data are acquired. The data is all collected over a short period of time. For example, when monitoring water pump pressure, pressure is the directly correlated data, while temperature and vibration amplitude are indirectly correlated data. Similarly, when monitoring transformer winding temperature, winding temperature is the directly correlated data, while oil temperature, oil pressure, load current, and ambient temperature are indirectly correlated data. Real-time acquisition of the directly correlated data from each device is performed, comparing it with a preset allowable operating range. This preset allowable operating range can be determined based on historical directly correlated data during normal operation, combined with the expertise of personnel in the field. When the directly correlated data is not within the corresponding range... If the directly related data is within the allowed working range, an anomaly is detected in the device, and a real-time alarm is triggered. If the directly related data is within its corresponding allowed working range, further determination is made. At this point, multiple indirectly related data related to the device are acquired. Each indirectly related data item has a corresponding allowed working range, which can be determined based on historical indirectly related data during normal operation and the expertise of personnel in the field. The number of indirectly related data items that are not within their corresponding allowed working ranges is counted and recorded as the number of items exceeding the allowed working range. The trigger point for the indirectly related data is then determined. The trigger value for self-indirectly related data is the ratio of the number of items exceeding the threshold to the total number of self-indirectly related data items (the total number of self-indirectly related data items acquired). This trigger value is then compared to a preset trigger threshold. The trigger threshold for each device can be determined based on the professional knowledge and experience of those skilled in the art. When the trigger value exceeds the trigger threshold, it indicates that the device is malfunctioning, and a real-time alarm is triggered. For example, if the winding temperature of a transformer is 88℃ (within the allowable range), it indicates that the transformer is normal. At this time, the cooling system oil pressure is 0.25MPa (below 0.With an operating pressure of 3MPa (out of range), an ambient temperature of 45℃ (above 40℃, out of range), and a load current of 1100A (within the allowable range), and the self-indirectly correlated data trigger threshold set to 1 / 3, if the self-indirectly correlated data trigger value is 2 / 3, it indicates that the equipment is malfunctioning, an anomaly is detected, and a real-time alarm is triggered. First, the direct correlated data of the equipment itself is used to determine if there is an obvious anomaly, and then the indirect correlated data is used to supplement the monitoring of hidden risks. This effectively solves the problem of easy missed detections with single data points and improves the accuracy of real-time alarms. When equipment is judged to be normal, although there is no real-time fault, there may be anomaly risks. For example, although the equipment has not triggered the alarm threshold, it is constantly near the alarm threshold and slowly approaching it, indicating a high risk of anomaly. Therefore, when the equipment is judged to be normal, it is recorded as the target equipment. Based on the position of the target equipment in the 3D model, a network of equipment associations with the target equipment is constructed. Equipment in the network that is associated with the target equipment is recorded as associated equipment. The criteria for determining association can be obtained from the actual operational relationships of substation equipment, such as functional dependencies, physical connections, and topological hierarchy. For example, transformers and cooling fans are functionally associated (cooling fans...). The device (for transformer cooling) is physically connected to the load switchgear (the switchgear controls the load switching of the transformer) and topologically connected to the voltage transformer (the transformer collects voltage data from the transformer). It acquires the direct correlation data of the target device under time series T, constructs its direct correlation data function, and simultaneously acquires the direct correlation data of each associated device under the corresponding time series T in the device correlation network, constructing its direct correlation data function for each associated device. The target device and each associated device have corresponding ideal direct correlation data functions. The ideal direct correlation data functions set for each device can be determined based on the device's rated design parameters and long-term stable operating history data under standard operating conditions. The following operational standards are proposed for the equipment industry: A Cartesian coordinate system is established with time as the x-axis and its directly correlated data as the y-axis. In this system, under time series T, the area enclosed by the target equipment's directly correlated data function and its corresponding ideal directly correlated data function is denoted as the first area. The first area is then normalized to obtain the first state coefficient. This first state coefficient indicates the degree of deviation between the target equipment's directly correlated data and its ideal directly correlated data under time series conditions; a larger value indicates a higher risk. The area enclosed by the directly correlated data function and its corresponding ideal directly correlated data function for each correlated equipment is denoted as the second area. The second areas for each correlated equipment are then weighted and accumulated. The data is then normalized to obtain the second state coefficient. The weights of each associated device can be determined autonomously based on the degree of influence of each associated device on the target device; the higher the degree of influence, the higher the weight. The second state coefficient indicates the comprehensive deviation between the direct correlation data of the devices associated with the target device in the time series and the ideal data; the larger the value, the higher the risk. Finally, the first and second state coefficients are weighted and summed. The weights of the first and second state coefficients can be determined according to the actual situation. Since the first state coefficient is obtained from the direct correlation data of the target device, its weight should be higher than that of the second state coefficient. For example, if the weight of the first state coefficient is 0.7, the weight of the second state coefficient is 0.3. Obtain the overall state coefficient and compare it with a preset state coefficient threshold. The preset state coefficient threshold for each device can be independently determined based on the statistical average of the overall state coefficients before historical faults, combined with the equipment's operational safety redundancy requirements. When the overall state coefficient exceeds the state coefficient threshold, it is determined that the device has an abnormal risk, and an early warning is issued. By combining this method with the analysis of the deviation between the device's and its associated devices' time-series data and ideal data, early warnings can be issued before anomalies occur, thus identifying slowly developing risks in advance and making the warnings more forward-looking. Furthermore, by coordinating with the networking of associated devices and the collaborative monitoring of target devices, coupled risks can be covered, improving the accuracy of the early warnings.

[0027] The method for generating flood control risk values ​​is as follows: Topographic data of the substation and its surrounding area is obtained from a 3D model. This topographic data is then divided into multiple grid cells using a digital elevation model (DEM) to form a DEM grid. Each grid cell carries corresponding meteorological data and equipment-related data. A spatial graph is constructed using the grid cells as nodes and spatial adjacency relationships as edges. In the spatial graph, based on the spatial adjacency relationships of each grid cell, any grid cell is selected as the target grid cell. The grid cells upstream of the target grid cell are designated as influencing grid cells. The Euclidean distance between each influencing grid cell and the target grid cell, as well as the elevation difference between them, are normalized to obtain a first influence factor and a second influence factor. The adjacency weights between the target grid cell and each influencing grid cell are calculated based on the second and first influence factors. A first risk value is generated based on the equipment-related data of the grid cells, and a second risk value is generated based on the meteorological data of the grid cells. The first and second risk values ​​are then added together to obtain the flood control risk value. The methods for obtaining the first and second risk values ​​are as follows: Based on the analysis results of the corresponding equipment's own correlation data within the grid cell, the total state coefficient of each equipment is extracted, and the average of the total state coefficients of each equipment is calculated to obtain the equipment state value of the grid cell; the equipment state values ​​of each influencing grid cell are multiplied by the adjacency weights and then summed to obtain the adjacent equipment state values; the equipment state values ​​of the target grid cell and the adjacent state values ​​are weighted, summed, and normalized to obtain the first risk value; meteorological correlation data of each grid cell are obtained, and the meteorological correlation data are organized into time input sequences according to a fixed time step to obtain the time input sequences of each grid cell; the time input sequences are input into the LSTM network, and the risk trend of the grid cell in the time dimension is captured through the gating mechanism in the LSTM network; the risk trend of the last time step in the LSTM network is taken as the temporal risk value of the grid cell; the temporal risk values ​​of each influencing grid cell are multiplied by the adjacency weights and then summed to obtain the adjacent temporal risk value; the temporal risk value of the target grid cell and the adjacent temporal risk values ​​are weighted, summed, and normalized to obtain the second risk value.

[0028] The above technical solution provides a specific method for generating flood control risk values. First, topographic data of the substation and its surrounding area is obtained from a 3D model. This topographic data is obtained using existing 3D model point cloud data extraction methods. For example, the topographic area of ​​the substation and its surrounding area is first defined in the 3D model of the substation, and then the 3D coordinate data (X and Y correspond to planar position, Z corresponds to elevation information) of the topographic area point cloud within this area is extracted. This coordinate data containing planar position and elevation information constitutes the topographic data. The topographic data is then divided into multiple grid cells using a digital elevation model (DEM) to form a DEM grid. Each grid cell carries corresponding meteorological data and corresponding equipment-related data. Related data could include rainfall, duration of rainfall, water level at equipment location, etc. A spatial graph is constructed using grid cells in the digital elevation model (DEM) grid as nodes and spatial adjacency relationships as edges. Within the spatial graph, based on the spatial adjacency relationships of each grid cell, any grid cell is selected as the target grid cell. Grid cells upstream of the target grid cell (those with elevation values ​​higher than the target grid cell in the spatial graph are designated as upstream) are designated as influencing grid cells. The Euclidean distance between each influencing grid cell and the target grid cell (obtained from their coordinates) and the elevation difference between them are used to determine the influence of each grid cell. The elevation of the target grid cell is obtained by subtracting its elevation value from the elevation value of the affected grid cell. After normalizing the Euclidean distance and elevation difference, the first and second influence factors are obtained, respectively. Based on the second and first influence factors, the adjacency weights of the target grid cell and each affected grid cell are calculated. The first influence factor is obtained through the Euclidean distance between the affected and target grid cells. A greater Euclidean distance indicates a more significant attenuation of risk transmission between the target and affected grid cells. The second influence factor represents the elevation difference; a larger elevation difference indicates a faster water spread and stronger transmission intensity. Therefore, the adjacency weights of the target and affected grid cells are calculated based on the second and first influence factors. The adjacency weights of the target grid cell and each influencing grid cell are calculated as follows: the second influencing factor is divided by the first influencing factor and then normalized to obtain the adjacency weights of the target grid cell and each influencing grid cell; by dividing the substation terrain into grid cells, and based on the elevation difference and Euclidean distance analysis of the upstream influencing grid cells, spatial correlation risk weighting is performed on each grid cell to obtain the corresponding adjacency weights, which can accurately capture the flood control risk correlation between different areas, thereby improving the accuracy of flood control; then, the total state coefficient of each device is extracted from the corresponding device correlation data within the grid cell, and the average of the total state coefficients of each device is calculated to obtain the device state value of the grid cell;The device status values ​​of each influencing grid cell are multiplied by their adjacency weights and then summed to obtain the adjacent device status value. The status value of the target grid cell and the adjacent status values ​​are then weighted, summed, and normalized to obtain the first risk value. Here, the weight of the status value is higher than that of the adjacent status value; for example, the weight of the status value is 0.7, and the weight of the adjacent status value is 0.3. The larger the first risk value, the worse the condition of the grid cell, and therefore the worse the flood control function, and the higher the risk. Then, meteorological correlation data for each grid cell, such as rainfall, rainfall duration, and water level, are obtained. This meteorological correlation data is then organized into a time input sequence according to a fixed time step to obtain... The time input sequence for each grid cell is obtained by acquiring meteorological correlation data for the most recent hour, for example, with a time step of 10 minutes. This time input sequence is then fed into an LSTM network. The gating mechanism within the LSTM network captures the risk trend of the grid cell in the time dimension. The risk trend of the last time step in the LSTM network is taken as the temporal risk value of the grid cell. The LSTM uses a forget gate, input gate, and output gate to remember the time-series data such as rainfall, water level, and rainfall duration in the sequence, thus obtaining the risk trend in the time dimension. This is a common approach in meteorological time series analysis, and LSTM is used for time series analysis. Trend capture and feature extraction of data are mature technologies, which will not be described in detail here. For example, LSTM networks are used to extract the changing trends of rainfall, water accumulation, and rainfall duration sequences. A comprehensive risk trend is obtained through integrated analysis. The risk trend of the last time step is taken as the temporal risk value of the grid cell. The larger the temporal risk value, the higher the degree of risk. Finally, the temporal risk values ​​of each influencing grid cell are multiplied by the adjacency weights and summed to obtain the adjacency temporal risk value. The temporal risk value of the target grid cell and the adjacency temporal risk values ​​are summed and normalized to obtain the second risk value. The first risk value and the second risk value are then added together. This method obtains flood control risk values; by using an LSTM network to capture trends in the time series of meteorological data, it overcomes the limitation of relying solely on instantaneous meteorological data to identify slowly developing flood control risks in advance. It effectively extracts key temporal risk characteristics such as increasing rainfall intensity, thereby improving flood control early warning efficiency. Furthermore, this method integrates equipment status data with meteorological data and quantifies them into first and second risk values, overcoming the one-sidedness of assessing flood control risk from a single dimension. It achieves multi-dimensional risk synergistic quantification of equipment hazards and meteorological factors, facilitating rapid risk level determination by maintenance personnel and significantly improving the accuracy of substation flood control risk assessment and maintenance response efficiency.

[0029] The method for tiered flood control early warning based on flood risk values ​​is as follows: Flood control early warning includes four levels: high-risk warning, medium-low-risk warning, low-risk warning, and zero-risk warning. Each warning level has a corresponding flood risk value range. When the flood risk value falls within the range corresponding to the low-risk warning, a low-risk warning is issued; when it falls within the range corresponding to the medium-risk warning, a medium-risk warning is issued; when it falls within the range corresponding to the high-risk warning, a high-risk warning is issued; and when it falls within the range corresponding to the zero-risk warning, no warning is issued.

[0030] The above solution provides a specific method for tiered flood warning based on flood risk values. First, flood warnings are divided into four levels: high-risk, medium-low-risk, low-risk, and zero-risk. Different alarm messages are set for each level to allow maintenance personnel to quickly understand the severity of the flood risk. For example, a zero-risk warning is a normal situation and does not require a warning; only background recording is needed, such as generating a log. A low-risk warning indicates a minor risk and can be implemented with a light alert and enhanced inspection. For example, a light yellow pop-up window appears to maintenance personnel, and the inspection app pushes a reminder to the maintenance team, adjusting the inspection frequency for that area from once every 2 hours to once every hour, and highlighting areas prone to waterlogging. A medium-low-risk warning provides a moderate alert and requires standby. For example, a red pop-up window appears to maintenance personnel, with increased frequency, accompanied by real-time monitoring footage of the grid unit. Maintenance personnel are also notified to prepare appropriate flood control equipment and be on standby. Awaiting notification; When a high-risk warning is issued, indicating a high level of risk, an audible and visual alarm will be activated, along with the emergency response plan. This includes continuous flashing red lights and voice alerts on-site, while automatically activating backup flood control measures for waterproofing. Each warning level has a corresponding flood risk value range. These ranges can be set by statistically analyzing the risk value distribution across different risk levels in historical flood events, combined with the safety redundancy requirements of the corresponding grid units. A low-risk warning is issued when the flood risk value falls within the range corresponding to a low-risk warning; a medium-risk warning is issued when it falls within the range corresponding to a medium-risk warning; a high-risk warning is issued when it falls within the range corresponding to a high-risk warning; and no warning is issued when the flood risk value falls within the range corresponding to a zero-risk warning. This method facilitates rapid risk assessment by maintenance personnel, significantly improving the accuracy of substation flood risk assessment and maintenance response efficiency.

[0031] In one embodiment, a three-dimensional flood control monitoring system for substations, incorporating multi-source heterogeneous data, is disclosed, such as... Figure 2 As shown, the system includes: The 3D model building module is used to collect 3D spatial perception data of the substation and build a 3D model of the substation based on the analysis of the 3D spatial perception data. The multi-source data acquisition module is used to collect relevant data related to flood control in substations and import the relevant data into the corresponding location of the 3D model. The relevant data includes equipment-related data and meteorological data. The equipment monitoring and early warning module is used to analyze the equipment's own associated data and perform real-time alarm monitoring and early warning monitoring for each piece of equipment in the substation. The flood control monitoring and early warning module is used to divide the substation into multiple grid units, analyze the data related to the equipment in each grid unit and the meteorological data, generate flood control risk values, and conduct graded flood control early warning based on the flood control risk values.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0033] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0034] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A three-dimensional flood control monitoring method for substations using multi-source heterogeneous data fusion, characterized in that, The methods include: Collect three-dimensional spatial perception data of the substation, and construct a three-dimensional model of the substation based on the analysis of the three-dimensional spatial perception data; Collect relevant data related to flood control at substations and import the relevant data into the corresponding locations in the 3D model. The relevant data includes equipment-related data and meteorological data. Based on the analysis of the equipment's own associated data, real-time alarm monitoring and early warning monitoring are carried out on each piece of equipment in the substation. The substation is divided into multiple grid units. Based on the analysis of the equipment's own correlation data and meteorological correlation data within each grid unit, flood control risk values ​​are generated, and graded flood control early warnings are carried out based on the flood control risk values.

2. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The method for constructing a 3D model of a substation is as follows: Three-dimensional spatial perception data includes oblique photogrammetry image data, oblique photogrammetry point cloud data, and lidar point cloud data; Based on the equipment structure of the substation, the substation is divided into multiple blocks; a three-dimensional coordinate system is constructed, and each block has its own coordinate position in the three-dimensional coordinate system; oblique photography point cloud data and lidar point cloud data of each block are acquired and collectively referred to as point cloud data; the point cloud data is analyzed to determine whether there is occlusion in each block. Blocks that are obscured are marked as obscured blocks, and blocks that are not obscured are marked as unobscured blocks; The oblique photographic point cloud data of the occluded areas are filled, and a 3D model is constructed based on the filled point cloud data of each occluded area and the point cloud data of each unoccluded area.

3. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 2, characterized in that, The method for determining whether each block is obstructed is as follows: Based on oblique photogrammetry point cloud data, the oblique photogrammetry point cloud coverage value, oblique photogrammetry normal vector, and oblique photogrammetry elevation value of each block are obtained; based on lidar point cloud data, the lidar point cloud coverage value, lidar normal vector, and lidar elevation value are obtained; based on the oblique photogrammetry point cloud coverage value and the lidar point cloud coverage value, the point cloud coverage difference value of each block is calculated; based on the oblique photogrammetry normal vector and the lidar normal vector, the normal vector deviation of each block is calculated; based on the oblique photogrammetry elevation value and the lidar elevation value, the elevation difference value of each block is calculated. After normalizing the point cloud coverage difference value, normal vector deviation value, and elevation difference value, the difference index of each block is obtained by weighting and summing them. The difference index of each block is then compared with the preset difference index judgment threshold of each block. If the difference index is less than or equal to the difference index judgment threshold, it is determined that there is no occlusion in the corresponding block; otherwise, the oblique photogrammetric image data of the corresponding block is retrieved for viewpoint backtracking verification. If the verification passes, the block is determined to be unobstructed; otherwise, the block is determined to be obstructed.

4. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 2, characterized in that, The method for filling in the oblique photographic point cloud data of occluded areas is as follows: Unobstructed blocks adjacent to obstructed blocks are designated as adjacent safe blocks. LiDAR point cloud data of these adjacent safe blocks are acquired and labeled as adjacent LiDAR point cloud data. Multi-granularity semantic segmentation is performed on the adjacent LiDAR point cloud data to identify and label equipment components within the adjacent safe blocks. An equipment topology graph is established, where nodes represent equipment components, and edges represent connection relationships, spatial constraints, and functional dependencies. Simultaneously, a substation equipment knowledge base is constructed, storing standard design parameters and maintenance history data for equipment components. Multi-dimensional features of equipment components are extracted from adjacent safe blocks, forming multi-dimensional feature vectors. After normalizing the dimensional feature vectors of each dimension, a feature fusion network is used to fuse the dimensional feature vectors of each dimension into a unified feature representation vector. The feature representation vector is dimensionally compressed to obtain the device semantic DNA code of the device component corresponding to the adjacent security block; based on the device topology map and multi-dimensional feature vector, a set of filling constraint rules for the device component is generated; the device semantic DNA code of the device component of the adjacent security block is used as the standard semantic DNA code of the device component corresponding to the occluded block; from the lidar point cloud data of the occluded block aligned to the same geographic coordinate system as the oblique photogrammetry point cloud data, the effective point clouds that match the standard semantic DNA code and satisfy the filling constraint rule set are selected, and the effective point clouds are filled into the oblique photogrammetry point cloud data of the occluded block; Based on the standard design parameters of equipment components in adjacent security blocks in the equipment knowledge base and maintenance history data, the filled oblique photogrammetry point cloud data is optimized, and a 3D model is constructed based on the optimized oblique photogrammetry point cloud data.

5. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The method for real-time alarm monitoring of various equipment in a substation is as follows: Acquire various self-related data of each device in the substation and import the self-related data into the corresponding position of each device in the 3D model. The self-related data includes self-directed data and self-indirected data. Acquire the self-directed data of each device in real time and compare the self-directed data with the preset self-directed data allowable working range. When the self-directed data is not within the corresponding self-directed data allowable working range, it is determined that the device is abnormal and a real-time alarm is triggered. When the directly related data is within the corresponding allowed working range, multiple indirectly related data related to the device are acquired. Each indirectly related data item has a corresponding allowed working range. The number of indirectly related data items that are not within the corresponding allowed working range is counted to obtain the indirectly related data trigger value. The indirectly related data trigger value is compared with the preset indirectly related data trigger threshold. When the indirectly related data trigger value is greater than the indirectly related data trigger threshold, the device is judged to be abnormal and a real-time alarm is triggered.

6. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 5, characterized in that, The method for early warning monitoring of various equipment in a substation is as follows: When it is determined that there is no abnormality in the equipment, the equipment is recorded as the target equipment. Based on the position of the target equipment in the 3D model, a network of equipment associations with the target equipment is constructed, and the equipment in the network of equipment associations that are associated with the target equipment is recorded as the associated equipment. The system acquires the direct correlation data of the target device under time series T, constructs the direct correlation data function of the target device, and simultaneously acquires the direct correlation data of each associated device under the corresponding time series T in the device association network, constructing the direct correlation data function of each associated device. Each target device and each associated device has a corresponding ideal direct correlation data function. A Cartesian coordinate system is established with time as the x-axis and direct correlation data as the y-axis. In the Cartesian coordinate system, under time series T, the area enclosed by the direct correlation data function of the target device and its corresponding ideal direct correlation data function is recorded as the first area. The first area is normalized to obtain the first state coefficient. The area enclosed by the direct correlation data function of each associated device and its corresponding ideal direct correlation data function is recorded as the second area. The second areas of each associated device are weighted, accumulated, and then normalized to obtain the second state coefficient. The first and second state coefficients are weighted and accumulated to obtain the total state coefficient. The state coefficient is compared with a preset state coefficient threshold. When the total state coefficient is greater than the state coefficient threshold, it is determined that the device has an abnormal risk, and an early warning is issued.

7. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The method for generating flood control risk values ​​is as follows: The terrain data of the substation and its surroundings are obtained from the 3D model. The terrain data is divided into multiple grid cells using a digital elevation model (DEM) to form a DEM grid. Each grid cell carries corresponding meteorological data and corresponding equipment-related data. A spatial graph is constructed using the grid cells in the DEM grid as nodes and spatial adjacency relationships as edges. In the spatial graph, based on the spatial adjacency relationship of each grid cell, any grid cell is selected as the target grid cell, and each grid cell upstream of the target grid cell is recorded as the influencing grid cell. Based on the Euclidean distance between each influencing grid cell and the target grid cell, the adjacency weight between the target grid cell and each influencing grid cell is obtained. A first risk value is generated based on the device's own associated data within the grid cells, and a second risk value is generated based on the meteorological associated data within the grid cells. The first and second risk values ​​are then added together to obtain the flood control risk value.

8. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 7, characterized in that, The methods for obtaining the first and second risk values ​​are as follows: Based on the analysis results of the corresponding device's own associated data within the grid cell, the total state coefficient of each device is extracted. The average of the total state coefficients of each device is calculated to obtain the device state value of the grid cell. The device state values ​​of each influencing grid cell are multiplied by the adjacency weight and then summed to obtain the adjacent device state value. The device state value of the target grid cell and the adjacent state values ​​are weighted, summed, and normalized to obtain the first risk value. Meteorological correlation data for each grid cell is acquired and organized into a time input sequence at a fixed time step to obtain the time input sequence for each grid cell. The time input sequence is then input into an LSTM network, where the gating mechanism captures the risk trend of the grid cell in the time dimension. The risk trend at the last time step in the LSTM network is taken as the temporal risk value of the grid cell. The temporal risk values ​​of each influencing grid cell are multiplied by the adjacency weights and then summed to obtain the adjacency temporal risk value. Finally, the temporal risk value of the target grid cell and the adjacency temporal risk values ​​are weighted, summed, and normalized to obtain the second risk value.

9. The method for three-dimensional flood control monitoring of substations based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The method for tiered flood control early warning based on flood risk values ​​is as follows: The flood warning system includes four levels: high-risk warning, medium-low-risk warning, low-risk warning, and zero-risk warning. Each warning level has a corresponding range of flood risk values. A low-risk warning is issued when the flood risk value falls within the range corresponding to a low-risk warning; a medium-risk warning is issued when the flood risk value falls within the range corresponding to a medium-risk warning; a high-risk warning is issued when the flood risk value falls within the range corresponding to a high-risk warning; and no warning is issued when the flood risk value falls within the range corresponding to a zero-risk warning.

10. A three-dimensional flood control monitoring system for substations based on multi-source heterogeneous data fusion, implementing the three-dimensional flood control monitoring method for substations based on multi-source heterogeneous data fusion as described in any one of claims 1-9, characterized in that the system... include: The 3D model building module is used to collect 3D spatial perception data of the substation and build a 3D model of the substation based on the analysis of the 3D spatial perception data. The multi-source data acquisition module is used to collect relevant data related to flood control in substations and import the relevant data into the corresponding location of the 3D model. The relevant data includes equipment-related data and meteorological data. The equipment monitoring and early warning module is used to analyze the equipment's own associated data and perform real-time alarm monitoring and early warning monitoring for each piece of equipment in the substation. The flood control monitoring and early warning module is used to divide the substation into multiple grid units, analyze the data related to the equipment in each grid unit and the meteorological data, generate flood control risk values, and conduct graded flood control early warning based on the flood control risk values.

Images