Point cloud-based power model parameter backfilling method and system

Abstract

The present application relates to the technical field of point cloud data processing, and more particularly to a power model parameter backfilling method and system based on point cloud. The method comprises the following steps: obtaining tower pole point cloud data and constructing a tower pole model to obtain the tower pole model; identifying residual focus of the tower pole model to obtain residual focus data; performing electromagnetic field back propagation according to the residual focus data to obtain tower pole parameter missing data; obtaining tower pole non-integrity data according to the tower pole parameter missing data; obtaining implicit missing data from the tower pole non-integrity data; obtaining tower pole missing data from the implicit missing data; performing reverse inference of tower height according to the tower pole missing data to obtain tower height inference data; performing electromagnetic field fitting optimization on the tower height inference data to obtain tower height fitting data for parameter backfilling operation. The present application has the characteristics of information fusion and structure-physical coupling optimization, and can effectively improve the automation completion accuracy of key parameters (such as tower height) in the tower pole model.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, and in particular to a method and system for backfilling parameters of a power model based on point clouds. Background Technology

[0002] With the advancement of the State Grid's digital transformation, digital twin modeling and structural parameter standardization of transmission lines have become crucial foundations for ensuring operational safety and intelligent maintenance. Currently, in the actual surveying and modeling of transmission lines, point cloud data of tower structures are commonly obtained using UAV laser scanning or multi-view image reconstruction. Although point cloud data has high spatial reconstruction capabilities, it is often affected by factors such as tower top component obstruction, tower base terrain interference, and meteorological limitations, resulting in problems such as incomplete structural height, missing crossarms, and blurred key components, which seriously affect the accuracy of subsequent models and the reliability of structural safety analysis. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a method and system for backfilling power model parameters based on point clouds, thereby resolving at least one of the aforementioned technical issues.

[0004] This application provides a method for backfilling parameters of a power model based on point clouds, the method comprising:

[0005] S1. Obtain tower point cloud data and construct a tower model based on the tower point cloud data to obtain the tower model;

[0006] S2. Perform residual focusing identification on the tower model to obtain residual focusing data; perform electromagnetic field back-engineering based on the residual focusing data to obtain missing tower parameter data; perform tower incompleteness identification based on the missing tower parameter data to obtain tower incompleteness data; perform implicit missing data supplementation on the tower incompleteness data to obtain implicit missing data; perform temporal structure degradation processing based on the implicit missing data to obtain tower missing data.

[0007] S3. Based on the missing tower data, reverse inference of tower height is performed to obtain the inferred tower height data;

[0008] S4. Perform electromagnetic field fitting optimization on the tower height estimation data to obtain tower height fitting data for parameter backfilling.

[0009] This invention features multi-source information fusion and structural-physical coupling optimization, effectively improving the accuracy of automated parameter completion for key parameters (such as tower height) in tower models. By introducing residual focusing identification and electromagnetic field back-inference mechanisms, it can accurately locate missing tower parameter regions even when point cloud data is incomplete or structurally ambiguous. Furthermore, by combining incompleteness identification and implicit missing parameter completion strategies, it enhances the reasoning ability for weakly expressed structures. Through temporal structural degradation modeling, it achieves trend prediction of long-term evolutionary losses. In the tower height inference stage, a tension-geometric constraint model is introduced, and combined with electromagnetic field simulation for physical consistency optimization, effectively improving the engineering credibility and simulation adaptability of the backfilled parameters.

[0010] Optionally, the residual focusing identification includes:

[0011] The tower-line coupling path is constructed on the tower model to obtain the tower-line coupling path data;

[0012] Residual template matching is performed based on the tower-line coupling path data to obtain residual matching data;

[0013] Pseudo-missing data is obtained by performing pseudo-missing data stripping based on residual matching data;

[0014] Residual plots are generated from the pseudo-missing data to obtain residual plot data;

[0015] Focused regions are extracted from the residual map data to obtain focused residual data.

[0016] This invention introduces a tower-line coupling path construction and residual template matching mechanism to effectively identify local deviation areas in the tower model caused by structural inconsistencies or data occlusion. Compared to traditional point cloud density or geometric integrity judgment methods, this method starts from the spatial coordination relationship between the tower and the conductor, constructs a refined residual index system, and improves the accuracy of residual identification by removing non-structural anomalies through pseudo-missing data stripping. The generation of the residual map realizes the quantitative expression of spatial distribution characteristics, which facilitates the focused extraction and location of high-confidence anomaly regions, thereby providing accurate target areas for subsequent parameter backfilling.

[0017] Optionally, the electromagnetic field backtracking includes:

[0018] Electromagnetic field data of the residual region is obtained by acquiring electromagnetic field data through a magnetic field sensor pre-installed on the terminal based on the residual focusing data.

[0019] A tower-shaped electromagnetic field simulation model is constructed based on the electromagnetic data of the residual region.

[0020] High-sensitivity feature extraction was performed on the tower-shaped electromagnetic field simulation model to obtain high-sensitivity feature data;

[0021] The missing structural parameters of tower parameters are inferred from highly sensitive feature data.

[0022] This invention combines the residual focusing region with the electromagnetic data collected by the terminal magnetic field sensor to achieve the perception of the physical field response in structurally abnormal areas. Compared with traditional methods that rely on point cloud geometric features for missing data assessment, this method introduces a tower-type electromagnetic field simulation model construction mechanism, which can accurately simulate the electromagnetic field distribution characteristics under different structural parameters and improve the coupling modeling capability between structure and field. Through a highly sensitive feature extraction step, the structural parameter dimensions that have the greatest impact on electromagnetic field changes are effectively identified, enabling quantitative inference of the missing tower parameters.

[0023] Optionally, the high-sensitivity feature extraction includes:

[0024] Parameter-field mapping was performed on the tower-shaped electromagnetic field simulation model to obtain parameter-field mapping data;

[0025] Local sensitivity data is obtained by calculating the local sensitivity based on the parameter field mapping data;

[0026] A full-space sensitivity map is constructed based on local sensitivity data to obtain full-space sensitivity map data;

[0027] High response regions are extracted from the full-space sensitivity map data to obtain high response region data. The high response regions are obtained by clustering the full-space sensitivity map data and sorting them according to the number of data in the clusters, taking the top 5 or 10.

[0028] Parameter-region correspondences are extracted from high-response region data to obtain highly sensitive feature data.

[0029] This invention establishes a quantitative relationship between structural parameter variations and spatial electromagnetic response by performing parameter-field mapping on a tower-type electromagnetic field simulation model, overcoming the limitations of traditional geometric decoupling models in assessing the influence of structural features. Through local sensitivity calculations, the response intensity of each parameter to local electromagnetic field disturbances can be accurately evaluated. A full-space sensitivity map is constructed, representing the global distribution of electromagnetic response to structural disturbances, which facilitates the comprehensive identification of key sensitive regions within the structure. High-response regions are extracted using clustering and sorted according to the number within each cluster, ensuring the extracted regions are spatially significant and representative. By establishing a set of highly sensitive features through parameter-region correspondence, not only are the most significant structural parameters affecting the electromagnetic response identified, but physical criteria are also provided for the inverse estimation of key parameters such as tower height.

[0030] Optionally, the tower incompleteness identification includes:

[0031] The missing alignment data is obtained by aligning the missing tower parameter data with the preset tower function diagram data;

[0032] Geometric continuity features and topological connectivity features are extracted from the missing alignment data to obtain geometric continuity feature data and topological connectivity feature data, respectively.

[0033] Graph attention structure processing is performed based on geometric continuity feature data and topological connectivity feature data to obtain structure graph data;

[0034] Incomplete area mining is performed based on the structural diagram data to obtain incomplete tower data.

[0035] This invention combines missing structural parameter information with pre-defined tower functional diagram data to achieve precise alignment between the structural model and the standard functional configuration, thereby effectively identifying potential abnormal modules. By extracting geometric continuity and topological connectivity features from the alignment results, it systematically captures fractures, misalignments, or connection anomalies in the spatial morphology of the tower structure. A graph attention mechanism is introduced for structural graph modeling and feature enhancement, enabling the model to focus on local structural anomaly areas, improving the robustness and accuracy of non-integrity region identification. Through the non-integrity region mining process, high-confidence regions of structural integrity failure are output, achieving intelligent perception of weak structures and hidden defects in the tower.

[0036] Optionally, the implicit missing complement includes:

[0037] Implicit verification was performed on the non-integrity data of the tower, resulting in implicit verification data;

[0038] Attribute expansion is performed on the implicit verification data to obtain implicit augmented data;

[0039] Implicit inference is performed based on implicit augmented data to obtain implicit missing data.

[0040] This invention addresses the problem of implicit structural deficiencies that are difficult to identify in traditional point cloud modeling by proposing a multi-level reasoning mechanism based on incomplete results. By implicitly checking the incomplete data of tower structures, regions that are not explicitly labeled but may have structural anomalies can be identified. Attribute expansion is performed by combining the semantics of the tower structure with prior parameters to supplement missing semantic labels and structural parameters in the original model, improving the completeness and contextual consistency of the model's expression. Through an implicit reasoning module, the indirect identification and logical completion of missing structures are achieved using a rule base, structural commonalities, and a contextual graph, effectively overcoming the shortcomings of traditional methods in identifying weakly expressive regions.

[0041] Optionally, the temporal structure degradation processing includes:

[0042] The tower time-series structure evolution diagram is constructed based on the implicit missing data to obtain the tower time-series structure evolution diagram data;

[0043] Structural degradation trajectory data is obtained by extracting the structural degradation trajectory data based on the tower's time-series structural evolution diagram data.

[0044] Structural degradation prediction is performed based on structural degradation trajectory data to obtain tower missing data.

[0045] This invention constructs a time-series structural evolution diagram of towers based on implicitly missing data. This diagram not only reflects the spatial topological changes of the tower structure under historical operating conditions but also identifies structural continuity weakening characteristics caused by long-term operation, environmental stress, or maintenance intervention. By extracting the transition trajectories of node states in the evolution diagram, potential degradation paths and key evolution nodes can be effectively captured, enhancing the forward-looking identification capability of potential failure areas of the towers. The structural degradation prediction integrates time series modeling and graph structure reasoning, ensuring that the missing data inference results not only possess spatial structural consistency but also exhibit temporal evolutionary rationality, effectively improving the stability and continuity of model parameter backfilling.

[0046] Optionally, S3 includes:

[0047] The adjacent tower height trend is calculated based on the missing tower data to obtain adjacent tower height trend data;

[0048] The structural difference index is calculated based on the adjacent tower height trend data to obtain the structural difference index data;

[0049] Based on the structural difference index data, the tower height is optimized for tension balance, and the tower height inference data is obtained.

[0050] This invention combines the height trend of adjacent towers with structural stability analysis, effectively improving the accuracy and physical consistency of missing tower height estimation. By calculating the height trend of adjacent towers based on missing tower data, a preliminary linear or nonlinear fitting model of height variation can be constructed to capture the spatial variation pattern of tower height. Based on the structural difference index calculation, the geometric and topological differences between the current tower and adjacent structures in terms of height, connection method, and crossarm position are quantified, enhancing the robustness of structural inference. A tension balance constraint model is introduced, incorporating the relationship between conductor tension and span into the optimization process, ensuring that the estimated tower height is not only geometrically reasonable but also meets the physical and mechanical constraints of actual operation.

[0051] Optionally, S4 includes:

[0052] A geometric model of the tower height is constructed based on the tower height inference data, and the tower height geometric model is obtained;

[0053] Electromagnetic field calculations were performed on the geometric model of the tower height to obtain electromagnetic field data for the tower height.

[0054] Real-time electromagnetic field data is acquired, and an error field is constructed based on the real-time electromagnetic field data and the electromagnetic field data of the tower height to obtain error field data.

[0055] Based on the error field data, residual optimization is performed on the tower height inference data to obtain tower height fitting data for parameter backfilling.

[0056] This invention significantly enhances the physical consistency and practical adaptability of the estimated tower height by fusing tower height estimation data with electromagnetic field simulation results. Through the construction of a tower height geometric model and electromagnetic field simulation calculations, a mapping relationship between structural parameters and field response is established. Subsequently, real-time electromagnetic field data is introduced, and an error field is constructed based on the simulation results to achieve a spatially quantifiable expression of the structure-field response difference. Residual optimization is performed based on the error field data to dynamically adjust the estimated tower height value, ensuring optimal matching under electromagnetic constraints. Introducing electrical measured data feedback into the geometric parameter optimization process provides good engineering interpretability and parameter correctability, effectively bridging the error between structural modeling and actual operating conditions.

[0057] Optionally, this application also provides a point cloud-based power model parameter backfilling system for performing the point cloud-based power model parameter backfilling method described above, wherein the point cloud-based power model parameter backfilling system includes:

[0058] The point cloud modeling and structure reconstruction module is used to acquire tower point cloud data and construct tower models based on the tower point cloud data to obtain tower models.

[0059] The missing focus and incompleteness identification module is used to perform residual focus identification on the tower model to obtain residual focus data; perform electromagnetic field back-engineering based on the residual focus data to obtain missing tower parameter data; perform tower incompleteness identification based on the missing tower parameter data to obtain tower incompleteness data; perform implicit missing data supplementation on the tower incompleteness data to obtain implicit missing data; and perform temporal structure degradation processing based on the implicit missing data to obtain tower missing data.

[0060] The tower height inference and structural parameter regression module is used to infer the tower height in reverse based on the missing tower data, and obtain the tower height inference data.

[0061] The electromagnetic fitting optimization and parameter backfilling module is used to optimize the electromagnetic field fitting of the tower height estimation data to obtain tower height fitting data for parameter backfilling.

[0062] The purpose of this invention is to construct a basic tower structure model through point cloud modeling in step S1, providing geometric support for subsequent analysis. In stage S2, abnormal structural regions are first located through tower-line coupling analysis and residual focusing identification. Then, an electromagnetic field back-inference mechanism is introduced to establish a coupling relationship between the structure and electromagnetic response, accurately inferring the type and location of missing parameters. Based on this, potential defect areas in the structural evolution process are effectively characterized through incompleteness identification, implicit missingness supplementation, and temporal degradation processing, ensuring the spatial integrity and temporal consistency of the missing parameter determination. In S3, the tower height is inferred in reverse by combining the structural difference index and the tension balance model, and parameter regression is performed by integrating geometric trends and mechanical constraints. In S4, electromagnetic field fitting optimization is used to match the simulated field distribution with sensor data to correct the tower height estimation results, achieving consistent calibration of the structure-field response. Attached Figure Description

[0063] Other features, objects, and advantages of this application will become more apparent from the following detailed description of the non-limiting embodiments, taken with reference to the accompanying drawings:

[0064] Figure 1 A flowchart illustrating the steps of a point cloud-based power model parameter backfilling method according to an embodiment is shown.

[0065] Figure 2 A flowchart illustrating the steps of a residual focusing identification method according to an embodiment is shown.

[0066] Figure 3 A flowchart illustrating the steps of a method for reverse inference of tower height according to one embodiment is shown.

[0067] Figure 4 A flowchart illustrating the steps of a method for fitting and optimizing the electromagnetic field of a tower height according to an embodiment is shown.

[0068] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0069] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0070] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. Functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0071] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0072] Please see Figures 1 to 4 This application provides a method for backfilling parameters of a power model based on point clouds, the method comprising:

[0073] S1. Obtain tower point cloud data and construct a tower model based on the tower point cloud data to obtain the tower model;

[0074] Specifically, a multi-view UAV system equipped with a LiDAR sensor performs a flight path to collect high-density point cloud data along the power transmission line. Combined with attitude calculation and GNSS positioning information, a raw point cloud dataset with three-dimensional spatial coordinates (x, y, z) and reflection intensity information is generated. A random sample consensus algorithm is used to fit the ground plane, and a height threshold filtering strategy is employed to remove ground points and background noise points. From the remaining point cloud, a density-based spatial clustering algorithm (such as DBSCAN) is applied to extract high-density clusters corresponding to the tower structure as the target point set. For each tower cluster, principal direction fitting is performed, specifically using principal component analysis to calculate the first principal component direction vector of the cluster as an estimate of the tower axis. A preliminary skeleton path for the tower axis is constructed through projection or truncation. Based on the distribution characteristics of point clouds along the main axis (such as the Z-axis), the tower structure is divided into typical structural areas such as the tower base section, tower body section, crossarm section, and tower top section. A semantic tree model of the tower structure is constructed. The structural semantic tree consists of several structural unit nodes. The nodes are established in a parent-child hierarchical relationship based on physical connection or component function, which is used to express the segmentation logic of the tower body, the position of components, and related information.

[0075] S2. Perform residual focusing identification on the tower model to obtain residual focusing data; perform electromagnetic field back-engineering based on the residual focusing data to obtain missing tower parameter data; perform tower incompleteness identification based on the missing tower parameter data to obtain tower incompleteness data; perform implicit missing data supplementation on the tower incompleteness data to obtain implicit missing data; perform temporal structure degradation processing based on the implicit missing data to obtain tower missing data.

[0076] Specifically, the locations of the conductor suspension points in the tower model are extracted. Based on the spatial geometric relationship between the suspension points and the tower top node, the coupling path between the conductor path and the tower structure is calculated. The calculation of the coupling path includes indicators such as the Euclidean distance between points and the angle between the conductor inclination angle and the tower's main axis. The key structural points in the tower model (such as the tower base, crossarms, and tower top) are registered and aligned with the corresponding points in the standard tower template. The residual vector between the actual observed coordinates and the template coordinates is calculated point by point, representing the geometric deviation of each key point. A threshold judgment is applied to the above residual vectors. If the residual value of a point is less than a set threshold (e.g., 10 cm), it is marked as a "non-missing" area. Simultaneously, combined with point cloud occlusion angle and visibility judgment, false alarm areas caused by viewpoint occlusion and scanning blind spots are excluded, and truly suspected missing structures are extracted. The residual information of each point in the point cloud is mapped into a visual color heatmap, where the color depth represents the magnitude of the deviation, thus intuitively displaying areas of concentrated structural anomalies in three-dimensional space. Density-based spatial clustering algorithms (such as DBSCAN) are used to perform cluster analysis on high-bias points in the residual map. The clusters are sorted according to the number of points in each cluster and the mean residual. The top 5 to 10 most significant residual clusters are selected as the key analysis areas to form a residual focus dataset.

[0077] A magnetic field sensor array is deployed within the residual focusing region to collect electromagnetic data such as electric field strength (E) and magnetic induction intensity (B), constructing a mapping table between spatial points and electromagnetic responses as the observation basis for inversion calculations. Electromagnetic simulation tools (such as COMSOL Multiphysics or ANSYS Maxwell) are used to construct an electromagnetic field model for the corresponding tower type. Model parameters include input conditions such as tower height, conductor suspension point location, voltage level, and current intensity, forming a causal model of adjustable structural parameters and simulated electromagnetic response. A small perturbation of ±5% is applied to the set of tower structural parameters (e.g., tower height, tower tilt angle, and three-dimensional coordinates of suspension points) to simulate changes in electromagnetic response. The absolute values ​​of the partial derivatives of the electric field response with respect to each structural parameter are calculated to obtain the sensitivity value of the electric field to structural changes. A complete sensitivity distribution map is formed in three-dimensional space, and the top 10% of the response amplitudes are selected as high-sensitivity regions for structural electromagnetic coupling. A structural inversion optimization problem is constructed, using the sum of squared differences between simulated and measured electric field data as the objective function to solve for the optimal structural parameters that minimize the error between the two. This optimization can be achieved using the Particle Swarm Optimization (PSO) algorithm or a quasi-Newton method (such as L-BFGS), and the output optimal structural parameters are the missing data to be filled in for the tower.

[0078] The current tower structure division results are compared with the preset standard structural function map to determine whether there are any missing or abnormal connections of key components (such as crossarms, suspension points, and insulator strings). A graph structural model is established based on the tower structure, where nodes represent structural components and edges represent physical connections. The model analyzes for structural breaks, weak connections, or non-closed loops. A graph attention mechanism model (such as GAT) is introduced into the structural graph to calculate the importance score of each structural node in terms of network attention; low-attention regions are considered potentially incomplete regions with unclear functions or insufficient information. Combining the functional alignment results, topology analysis results, and graph attention scores, structural incompleteness labels are generated and output as input data for implicit missing data completion.

[0079] For structural segments in incomplete regions, structural code rules are matched. For example, a certain type of steel tower should typically include two layers of crossarms. If the current detected structure does not meet this rule, it is marked as potentially missing. Using structural semantic tags and functional graphs, it is determined whether the currently missing region should have specific functional components (such as mounting points or surge arresters at voltage levels). If missing, a "missing, presumed, to be completed" tag is added. Completion is performed by combining knowledge graph-based tower type logical reasoning with structural statistical information of adjacent towers. For example, if the mounting point height of two adjacent towers is 45 meters, this tower can be presumed to have the same mounting point height. Point cloud data collected from historical periods of the towers (such as T1 to TN) is acquired and spatially aligned using a unified coordinate system to ensure consistency of temporal structural positions. For each structural segment, its temporal state trajectory is constructed, including three main dimensions: spatial position offset, local point cloud density change, and the degree of abrupt change in the point cloud normal vector distribution. The study determines whether patterns such as continuous subsidence, structural sparsity, or disordered normal orientation exist in the evolutionary trajectory to identify trends of structural aging, damage, or gradual regression. Structural segments exhibiting significant degradation trends are defined as "degraded missing segments," and the output is uniformly presented as tower missing data.

[0080] S3. Based on the missing tower data, reverse inference of tower height is performed to obtain the inferred tower height data;

[0081] Specifically, based on the height information of the two adjacent towers before and after the current tower, an initial estimate of its tower height is calculated. This value is obtained by taking the arithmetic mean of the heights of adjacent towers to approximate the height trend of the current tower within the overall line. The structural difference index between the current tower and its adjacent towers is calculated. This index, obtained by averaging the differences between multiple structural parameters (such as hanging point height, crossarm arrangement, number of tower segments, etc.), reflects the degree of similarity in structural continuity. The smaller the structural difference, the more reliable the average inference; the larger the structural difference, the more necessary optimization and correction are required. (Based on pre-defined template matching of other point cloud data corresponding to missing data,) A physical tension balance model based on the conductor tension-tower height coupling relationship is constructed. The theoretical tower height is calculated based on the conductor unit weight, tower spacing, and tension parameters. ,in This represents the current tower height. The tower height is the height of the adjacent tower. This refers to the unit weight of the conductor (weight per unit length). This refers to the horizontal span between the current tower and its adjacent tower. The tension (force) of the conductor is used; and the error between the actual observed conductor shape and the model conductor shape is used as the optimization objective. The least squares optimization method is used to deduce the optimal tower height estimate.

[0082] S4. Perform electromagnetic field fitting optimization on the tower height estimation data to obtain tower height fitting data for parameter backfilling.

[0083] Specifically, a three-dimensional tower structure model based on the inferred tower height is constructed, and electromagnetic field simulation calculations are performed in conjunction with the corresponding electrical parameters to obtain spatial electromagnetic distribution data. Measured electromagnetic data is obtained by electromagnetic field sensors deployed on-site. The measured data and simulation data are correlated and analyzed to construct an electromagnetic field error field, which is used to characterize the influence of structural errors on the electromagnetic field response. A tower height correction and optimization model is established based on the principle of minimum fitting error, and the tower height parameters are iteratively adjusted through optimization algorithms to make the simulation results match the measured results as closely as possible. The corrected tower height fitting value is output as the input for tower parameter backfilling.

[0084] Optionally, the residual focusing identification includes:

[0085] S21. Construct the tower-line coupling path for the tower model to obtain tower-line coupling path data;

[0086] Specifically, the input data includes 3D point cloud data of the tower structure and a set of spatial coordinates related to conductor suspension points or connection points. Based on the tower point cloud data, density clustering algorithms or vertical hierarchical segmentation methods are used to extract the crossarm region or conductor suspension point region in the tower. The conductor suspension points adjacent to the tower structure are extracted from the conductor point cloud or connection point set and denoted as conductor suspension point position vectors. The top point or suspension point position in the tower model is connected to the conductor suspension points to construct the tower-line coupling path. The spatial Euclidean distance of the path is calculated. ,in For the first The spatial Euclidean distance of the coupling path of the tower line. This refers to the position vector (three-dimensional coordinates) of the hanging point in the tower model. Given the anchor point position vectors (3D coordinates) of adjacent conductors; calculate the angle between the tower direction and the conductor direction: ,in For the first The directional angle of the coupling path of the tower line, It is a cosine function. Let be the unit normal vector at the end of the tower. Let be the unit normal vector at the conductor suspension point. This yields the tower-line coupling path data set Π, containing all paths and their corresponding Euclidean distances and orientation angles.

[0087] S22. Perform residual template matching based on the tower-line coupling path data to obtain residual matching data;

[0088] Specifically, from a pre-defined standard tower template library, a reference tower structure template most similar to the current tower structure is extracted, such as a straight-line tower or a tension tower. A spatial registration algorithm (such as the Iterative Closest Point (ICP) algorithm or the Procrustes alignment method) is used to align the current tower model to the coordinate system of the standard tower template, ensuring consistency of path coordinates. For each tower line coupling path, the length difference and direction angle difference between it and the standard template path are calculated to construct a residual index. The calculation formula is as follows: ,in For the first The residual index value of the path, The observation distance of the current path. This is a reference length for the standard path. This is a weighting factor for the difference in orientation angles, with a value ranging from 0.1 to 0.3. The observation direction angle of the current path. The reference orientation angle is used for the standard path. The residual values ​​of each path are compared with a set threshold (e.g., 0.5 meters) to identify paths whose deviations exceed the allowable range. A residual matching dataset is obtained, defined as the set of all paths whose residual values ​​are greater than the threshold.

[0089] S23. Perform pseudo-missing data stripping based on residual matching data to obtain pseudo-missing data stripping data;

[0090] Specifically, from the residual matching data set obtained in step S22, all residual values ​​greater than a set threshold are extracted. The paths are used as a candidate set of abnormal paths. For each abnormal path, the following three pseudo-missing criteria are evaluated sequentially: the local point cloud density within the path region is not less than 80% of the average point cloud density in the tower region; the path ends with structurally continuous boundary features, judged by whether the curvature change at the end point meets the continuity standard; and the path ends with a clear conductor connection structure, indicating that the actual connection has been established. If any two of the above three conditions are met, the path is determined to be a "pseudo-missing path". This path is then separated from the residual data set and stored separately.

[0091] S24. Generate residual plots from the pseudo-missing data to obtain residual plot data;

[0092] Specifically, the residual value on each electromagnetic path is mapped back to the starting position of its respective path, which is typically the spatial coordinate point of the connection point between the line and the tower body. This mapping is used to establish the spatial association between the residual and the tower body. Taking each point in the tower point cloud as the target, the Euclidean distance between it and all residual data points is calculated, and the residual values ​​are diffused and superimposed using a Gaussian kernel function. The residual weight value of each point is calculated as follows: the residual weight is the sum of the response contributions of multiple residual source points to the target point, expressed as: ,in Point the cloud midpoint for the tower. The residual heat value, For path order items, For the natural index term, For the first The residual values ​​on the path, For the first The coordinates of the starting point of each path. The width parameter of the Gaussian kernel function controls the spatial influence range, with a value ranging from 0.3 meters to 0.5 meters. For the first The residual values ​​along each path are calculated. All point cloud points are assigned corresponding residual heat weights and presented using color coding to form a structured residual map layer. This layer includes color-coded point cloud data: using color to represent the residual heat level of points; and a heat voxel map: mapping the residual heat values ​​to a regular voxel grid to form a volumetric heatmap. The output is structured residual map data.

[0093] S25. Extract the focused region from the residual map data to obtain the residual focused data.

[0094] Specifically, based on high residual points in the residual heatmap data, a density-based clustering algorithm (such as DBSCAN) is used to extract spatial residual clusters. The clustering parameters are set as follows: minimum number of points within a cluster is 20; neighborhood radius is 0.3 meters. All extracted residual clusters are sorted in descending order of the average residual intensity of their internal points to identify key areas of significant residual aggregation. From the sorting results, the top k residual clusters (k ranges from 5 to 10) with the highest residual values ​​are selected as focused structural anomaly regions. For each focused cluster, its geometric bounding range (which can be the minimum bounding box or a 3D convex hull) is calculated, and the average residual value within the cluster is calculated as its residual characteristic index.

[0095] Optionally, the electromagnetic field backtracking includes:

[0096] Electromagnetic field data of the residual region is obtained by acquiring electromagnetic field data through a magnetic field sensor pre-installed on the terminal based on the residual focusing data.

[0097] Specifically, based on the positioning results of the residual focusing area, the system collects electromagnetic response data of the area through electromagnetic field sensor devices pre-deployed at the terminal to obtain actual electromagnetic field observation data. The electromagnetic field sensors include, but are not limited to, Hall effect sensors and three-dimensional electric field antennas, which are respectively set at key positions of the tower structure, including the conductor hanging point area, the area below the crossarm structure, and the vertical projection point of the conductor on the ground. The sensor sampling frequency is set between 1 Hz and 10 Hz, and the sampling time interval is less than 0.5 seconds. The data collected by the system includes the following: (1) electric field strength value, calibrated according to three-dimensional spatial coordinates (x, y, z), in volts per meter (V / m); (2) magnetic induction intensity value, also calibrated according to three-dimensional coordinate position, in Tesla (T); (3) timestamp information of the acquisition time, used for time synchronization with spatial position information.

[0098] A tower-shaped electromagnetic field simulation model is constructed based on the electromagnetic data of the residual region.

[0099] Specifically, the input data for the simulation model includes (1) three-dimensional point cloud structure data from the residual focusing region; and (2) electrical parameter information of the transmission system, such as the rated voltage, current value, conductor material (e.g., LGJ-240 type), and phase line arrangement structure of the conductors. In the geometric modeling stage, the system converts the tower point cloud data into a CAD geometric model, preferably using the STL format. For the missing structural parts in the point cloud data, preliminary geometric filling can be performed based on the residual focusing results. Subsequently, in the physical parameter setting stage, the system sets the voltage level (e.g., 220kV, 500kV), conductor parameters, and dielectric properties according to the actual application scenario. The tower body is set as an ideal conductor, the surrounding environment medium is set as air, and its dielectric constant is taken as the vacuum dielectric constant (approximately). Regarding boundary conditions, the ground area is set as a ground boundary condition to simulate actual grounding conditions. The simulation process can select a frequency domain solver or a quasi-static electromagnetic field solver to adapt to the electromagnetic response simulation requirements under different frequency ranges. The simulation results output by the model include (1) the electric field distribution function, which represents the electric field intensity at the three-dimensional coordinates (x,y,z), in volts per meter (V / m); and (2) the magnetic field distribution function, which represents the magnetic induction intensity at the corresponding coordinate point, in tesla (T).

[0100] High-sensitivity feature extraction was performed on the tower-shaped electromagnetic field simulation model to obtain high-sensitivity feature data;

[0101] Specifically, the structural parameters set include the total tower height (in meters), the crossarm tilt angle (in degrees), and the spatial position (three-dimensional coordinates) of the conductor suspension points within the structure. To assess the impact of parameter disturbances, a set of disturbance parameters is defined, where each parameter... The system varies the parameter around its initial value with a perturbation amplitude (preferably 5% to 10% of the original value), forming a perturbation range. For each parameter, the system calculates the difference in electric field response at the spatial point (x, y, z) before and after the perturbation, and calculates the sensitivity value based on the following formula: ,in Let be the sensitivity value of the k-th parameter at the point (x, y, z). To the disturbance parameters The electric field response intensity under the condition, For parameters Set the electric field response intensity at this spatial point, in volts per meter (V / m). Let the parameter perturbation amplitude be . The original value is 5%-10%. To extract the influence of the dominant response, the maximum sensitivity value of each spatial point under all parameter perturbations is calculated: The system according to Based on the distribution results, a three-dimensional sensitivity voxel map is constructed, dividing the simulation space into discrete voxel units and assigning a sensitivity score to each voxel. High-sensitivity voxel regions with scores in the top 10% are selected, and spatial clustering algorithms are used to identify and form high-sensitivity region clusters. For each high-sensitivity cluster, the system analyzes its main controlled parameters and calculates the average sensitivity index within the cluster. A high-sensitivity feature dataset is thus formed, with each item being a triplet.

[0102] The missing structural parameters of tower parameters are inferred from highly sensitive feature data.

[0103] Specifically, the objective function for the fitting error of the electric field response is constructed, defined as follows for the observation area. For all spatial points (x, y, z) within the range, calculate the difference between the measured electric field intensity and the corresponding simulated electric field at each point, and then sum the squared differences in a weighted manner to form the overall error loss function. The formula is as follows: ,in This is the electric field response error loss function, reflecting the fitting error between the measured and simulated results. For the spatial points belonging to the observation area, This is the measured electric field strength value. Given structural parameters The electric field strength value obtained from the simulation is as follows: The set of structural parameters to be estimated may include (1) the total height of the tower (unit: meters); (2) the position vector of the conductor suspension point in space (three-dimensional coordinate representation); (3) the angle between the crossarm and the vertical direction (unit: degrees), etc. During parameter optimization, if... For optimization algorithms with low dimensionality and differentiable objective functions, quasi-Newton type optimization algorithms (such as L-BFGS) are preferred. If the parameter space is complex, containing discontinuous variables or constraints, particle swarm optimization (PSO) or Bayesian optimization methods can be used to improve the global optimum search capability. Parameter values ​​must meet engineering boundary conditions. The optimal parameter estimates are obtained through the optimization process, thus improving the objective function. The solution that achieves the minimum parameters: , To find the parameter solution operator that minimizes the objective function in the structural parameter space, the system organizes the above estimation results into a set of missing tower structural parameter data.

[0104] Optionally, the high-sensitivity feature extraction includes:

[0105] Parameter-field mapping was performed on the tower-shaped electromagnetic field simulation model to obtain parameter-field mapping data;

[0106] Specifically, a set of input parameters for the tower structure is defined, including but not limited to multiple geometric and structural characteristic variables such as tower height, crossarm angle, suspension point position, and crossarm length. The sub-parameters in this parameter set are represented as follows: tower height (vertical height from the base to the top); crossarm angle (angle between the crossarm and the main shaft axis); suspension point position (three-dimensional spatial position of the guide wire suspension point relative to the tower body); crossarm length (horizontal projection length between the endpoints of the crossarm); and other structural parameters are extended according to the actual tower type. For each of the above structural parameters, perturbation rules are set to simulate its possible variations. Each parameter is subjected to a 5% to 10% relative perturbation within its physically feasible range, and 5 to 7 numerical points are evenly selected within this perturbation range to form multiple sets of parameter perturbation combinations. Each set of combinations constitutes an independent simulation input vector. The above parameter perturbation combinations are sequentially input into the tower electromagnetic field simulation model. This simulation model can be constructed based on a three-dimensional finite element electromagnetic solution framework (such as COMSOL Multiphysics or ANSYS Maxwell), and the spatial distribution of electric and magnetic fields is solved under each set of input parameters. The output is the field response function, which contains the distribution values ​​of the electric field intensity and magnetic induction intensity in the three-dimensional coordinate space corresponding to each perturbation group, denoted as the electric field distribution E(i)(x,y,z): representing the value of the electric field distribution in the first three-dimensional coordinate space. Given a set of parameters, the electric field intensity at a spatial point (x,y,z) is given; the magnetic field distribution B(i)(x,y,z) represents the magnetic induction intensity under the same conditions. All parameter perturbation sets are paired with their corresponding electromagnetic field responses to establish a parameter-field mapping dataset. The output of this mapping dataset is a set of mapping pairs, where each pair consists of a structural parameter vector and a corresponding field distribution diagram, representing the explicit mapping relationship between structural perturbations and field responses.

[0107] Local sensitivity data is obtained by calculating the local sensitivity based on the parameter field mapping data;

[0108] Specifically, perturbation rules for structural parameters are defined. For each structural parameter variable (such as tower height, hanging point position, crossarm angle, etc.), a small perturbation is applied based on the original value, with the perturbation amplitude set to 5% of the original value (e.g., multiplying the original value by 1.05). For each structural parameter, the corresponding electric field distribution data is obtained from the electromagnetic simulation model before and after the perturbation. The sensitivity response is calculated using the finite partition method. In the electric field simulation data, for each spatial coordinate point (x, y, z), the change in electric field intensity before and after the perturbation is calculated and divided by the perturbation amplitude of that parameter to obtain the sensitivity value of that spatial point for that structural parameter. In other words, local sensitivity refers to the rate of change of field response caused by a unit parameter perturbation. The above operations are performed sequentially on all structural parameters, and a sensitivity matrix is ​​established at each spatial point. Finally, a local sensitivity dataset can be output, where each element represents the electric field sensitivity response of a certain structural parameter at a certain spatial location. The output format is a set of mappings of structural parameters and sensitivity fields. That is, for each structural parameter, a corresponding three-dimensional sensitivity distribution map is formed; each distribution map represents the sensitivity value in spatial coordinates (x, y, z).

[0109] A full-space sensitivity map is constructed based on local sensitivity data to obtain full-space sensitivity map data;

[0110] Specifically, multi-parameter sensitivity fusion processing is performed. For the local sensitivity data corresponding to each structural parameter obtained in the previous processing step, there are multiple sensitivity values ​​corresponding to each structural parameter at each three-dimensional spatial point. The following two fusion strategies are adopted: at each spatial location, the maximum value among all the sensitivity values ​​corresponding to the structural parameters is taken as the sensitivity representation of that location. Alternatively, a weighted average is performed on the sensitivity values ​​of all structural parameters. The weight corresponding to each parameter can be preset based on engineering experience or its physical importance in the structural function. For example, the weight of the tower height parameter can be slightly higher than that of secondary parameters such as the crossarm angle. This fusion method can take into account the influence of multiple factors and improve the engineering interpretability of the sensitivity field. The system performs three-dimensional spatial reconstruction processing. The spatial sensitivity values ​​obtained after the above fusion are mapped to a unified voxel mesh structure. This voxel mesh is a three-dimensional discrete structure used to divide spatial regions. Each voxel unit records the fused sensitivity value at its corresponding location, forming a continuous three-dimensional sensitivity layer. Full-space sensitivity map data is generated. The output data result is a sensitivity set in the form of a three-dimensional point set, where each data point contains its spatial coordinates and the sensitivity value at that location.

[0111] High response regions are extracted from the full-space sensitivity map data to obtain high response region data. The high response regions are obtained by clustering the full-space sensitivity map data and sorting them according to the number of data in the clusters, taking the top 5 or 10.

[0112] Specifically, the system sorts all sensitivity points in the full-space sensitivity map. This point set consists of the three-dimensional spatial sensitivity data obtained in the previous steps, with each spatial location corresponding to a sensitivity value. The system sorts the sensitivity values ​​of all points in descending order and sets a sensitivity threshold (e.g., selecting the top 10% of high-value points). Spatial clustering is then performed on the selected high-sensitivity point set. The density-based spatial clustering method (DBSCAN) is used to cluster and identify the high-sensitivity point cloud. During the clustering process, the following key parameters are set: the minimum number of points within a cluster is set to 30; the spatial radius threshold is set to 0.5 meters, indicating that points within this distance range can be considered as neighboring points within the same cluster. Based on these parameters, the system performs spatial clustering analysis on the high-response point set, obtaining several spatial clusters. All obtained clusters are sorted according to the number of points they contain, and the top five or ten clusters with the most points are selected as high-response regions. The system outputs high-response region data results, including triplet information for each cluster: cluster identifier, number of points within the cluster, and average sensitivity value within the cluster. The output format is as follows: High-response region dataset = {(cluster ID, number of points, average sensitivity value)}.

[0113] Parameter-region correspondences are extracted from high-response region data to obtain highly sensitive feature data.

[0114] Specifically, for each extracted high-response region cluster, all three-dimensional spatial points within it are traversed, and the total sensitivity response of all points in the cluster under different structural parameters is statistically analyzed. Specifically, for any parameter in the set of structural parameters, the cumulative value of the sensitivity function corresponding to that parameter across all points within the cluster is calculated. That is, the total sensitivity response of the structural parameter to all points in the cluster is calculated. Then, among all structural parameters, the set of parameters that maximizes this cumulative sensitivity is identified as the dominant structural variable in the electromagnetic response of the current cluster. This process can be described as selecting the parameter with the largest cumulative sensitivity value in the cluster from all candidate parameters as the dominant response parameter for that region. A correspondence is established between the dominant parameter and the high-response region, forming a triplet mapping structure of "parameter-spatial region-response intensity". This structure binds each high-response cluster to its dominant structural parameter and average sensitivity value, indicating that the electromagnetic field response within the cluster is mainly affected by changes in this structural parameter. The system outputs a high-sensitivity feature dataset.

[0115] Optionally, the tower incompleteness identification includes:

[0116] The missing alignment data is obtained by aligning the missing tower parameter data with the preset tower function diagram data;

[0117] Specifically, the input data includes missing tower parameter data, referring to structural attribute parameters that were not fully acquired during modeling or data acquisition, such as crossarm length and conductor suspension point positions; and a preset tower functional diagram, a standardized structural atlas covering key component nodes (such as tower feet, crossarm centers, insulator suspension points, etc.) and their connection relationships for a specific tower type, used to guide point cloud structure alignment. During the alignment process, for each structural node in the functional diagram, the system searches for the point with the smallest Euclidean distance in the 3D point cloud modeling result (PointCloud), completing the node position alignment. This matching process is defined as: ,in These are candidate locations in the point cloud. This is a minimum value search operation that searches for the point closest to the reference point in the point cloud. Let be the coordinate vector of any sampling point in the point cloud. This refers to a functional node, i.e., a standard node in the tower functional diagram. If the calculated matching distance... If the value exceeds a preset threshold, or if the point cloud at that location has issues such as occlusion or sparse density, the system marks the structural node as a potentially missing node and performs [further actions]. and Assign a value of Null. The system outputs a set of structure alignment results, represented as several triples ( , , ),in This is a standard node in the tower functional diagram; The point cloud location points that are matched with it; The alignment error between the two is expressed in meters.

[0118] Geometric continuity features and topological connectivity features are extracted from the missing alignment data to obtain geometric continuity feature data and topological connectivity feature data, respectively.

[0119] Specifically, for node pairs with connection relationships in the structural function diagram ( The system extracts the unit normal vector at each node based on the reconstructed geometry in the point cloud model. and And calculate the cosine of the included angle, defining the geometric continuity index as follows: ,in As a geometric continuity index, it reflects the degree of change in the connection direction between structural segments. The larger the value, the more obvious the angular abrupt change between the connecting segments, that is, the worse the geometric continuity. For vectors and The cosine of the angle between them, For nodes The unit normal vector at a given location represents the local spatial orientation of the node's structure. For nodes The unit normal vector at the location. The system constructs a structural connected graph G=(V,E), where V represents the set of nodes in the functional graph and E represents the set of connecting edges. By comparing with the connection information obtained from point cloud reconstruction, the system extracts the following topological feature indicators: (1) Node degree difference index , ,in Nodes in the functional graph The degree of connection, (2) Topological break marker: For edges that exist in the functional graph but are missing in the point cloud connection ( The system marks such edges as “broken edges” to indicate that the topological incompleteness may be caused by actual missing or occluded connections in the structure.

[0120] Graph attention structure processing is performed based on geometric continuity feature data and topological connectivity feature data to obtain structure graph data;

[0121] Specifically, the system uses each node in the structural alignment result as a vertex V in the graph and sets its input feature vector. It consists of the following: Structural alignment error; : Nodal degree difference; lc: Local curvature of the node; : Geometric continuity index of the connection direction with adjacent nodes. The edge connection relationship E between nodes is generated based on the following conditions: (1) the three-dimensional Euclidean distance between nodes is less than 1.5 meters; (2) or the connection edge of the node pair already exists in the functional graph. The system uses the graph attention network (GAT) structure to perform feature aggregation learning on the above graph. For each pair of connected nodes, its attention weight is calculated and defined as follows: ,in For nodes Its neighboring nodes Attention weights For nodes Neighbor set Normalization is performed. This is a rectified activation function with leakage (used for non-linear processing in attention score calculation). The trainable parameter vector (weight vector transpose) in the attention scoring function. This is the feature transformation weight matrix (used for linear transformation of node features). For nodes The input feature vector, For nodes The input feature vector. Based on the above attention weights, the system performs weighted aggregation on the neighbor features of each node to obtain structured aggregated data: , For nodes The updated feature vector (structured aggregated data). The activation function is used to introduce non-linear feature representations. To represent nodes The set of all adjacent nodes, For nodes The set of all adjacent nodes, For nodes Its neighboring nodes Attention weights This is the feature transformation weight matrix (used for linear transformation of node features). For nodes The input feature vector.

[0122] Incomplete area mining is performed based on the structural diagram data to obtain incomplete tower data.

[0123] Specifically, the aggregate score of each node in the structural graph reflects its degree of integrity based on geometric continuity and topological connectivity. The system sets a structural integrity score threshold, preferably within the range of [0,1], for example, 0.6. If a node's score is below this threshold, it is marked as a potentially structurally incomplete node. Based on the above determination, the system performs graph clustering on all marked incomplete nodes to form spatially correlated clusters of incomplete nodes. Clustering methods may include label propagation algorithms or spectral clustering algorithms to identify structural regions that are locally continuous but globally missing. After clustering, the system further determines whether each node cluster contains key regions in the structural functional graph, such as crossarm structures, conductor connection points, and main pole segments. If a cluster is determined to belong to a key component region, it will be given higher priority and will be a key target for repair or modeling completion.

[0124] Optionally, the implicit missing complement includes:

[0125] Implicit verification was performed on the non-integrity data of the tower, resulting in implicit verification data;

[0126] Specifically, the dataset of non-incomplete structural regions of the tower obtained in the previous step is denoted as non-incomplete data. A pre-defined standard semantic model of the tower structure defines the structural composition of the standard tower type, the spatial topological connections between components, and functional dependency paths, such as logical rules like "the main pole segment should be connected below the crossarm component" and "the hanging point component should be equipped with an insulator component." Three types of rules are used to investigate implicit missing components at the semantic level: 1. For each non-incomplete region, if a downstream functional component (such as an insulator) is observed to exist, but its upstream dependent component (such as a hanging point) is missing, then the component is determined to be "implicitly missing." This logical relationship is modeled based on functional dependency paths. 2. If two structural components should have a spatial connection in the standard model (such as between upper and lower main poles), but continuous data points are missing in the middle area of ​​the point cloud, and their vertical spacing is within a reasonable tolerance range (e.g., between 0.5 meters and 3 meters), then there may be an "implicit component break" at that location. 3. Map the point cloud topology of incomplete regions to the functional node model in the standard semantic graph. If there is a mapping interruption or a node fails to effectively connect to upstream or downstream functional units, the structural region is marked as a "latent candidate region". The system outputs a latent verification dataset, where each data item includes the corresponding structural unit identifier; the determined latent missing type (e.g., missing hanging point, broken crossarm, etc.); and semantic evidence supporting the determination (e.g., broken connection between upper and lower components, missing functional node mapping, etc.). The output structure can be represented as a set of triples: Latent Missing Candidate Dataset = {(structural unit number, missing type, criterion source)}.

[0127] Attribute expansion is performed on the implicit verification data to obtain implicit augmented data;

[0128] Specifically, for each candidate region of a component marked as implicitly missing, the system constructs its corresponding structural attribute vector. This attribute vector includes at least three-dimensional spatial coordinates (x, y, z), representing the spatial location of the candidate component in the tower model; preliminary inference of the component type, such as crossarm, hanging point, or connecting segment; the types of upper and lower level structural units directly adjacent to the candidate component in space, used to assist in determining their upstream and downstream semantic dependencies; the spatial distance to adjacent components, used to infer whether the component conforms to the standard layout logic; and the index number of the standard tower template to which it belongs, used for structural matching and geometric constraint binding with the reference model. This attribute vector expresses the spatial, semantic, and structural inference information of the implicit component to the greatest extent possible in the absence of explicit observation data. Based on the specific type of the missing component, the system adopts differentiated geometric and semantic expansion methods, specifically including the system extracting information on the upper and lower support points of the implicit candidate region (such as the extension endpoint of the main pole, the connection position of the insulator, etc.), and using the regular geometric shape of the crossarm in the standard tower structural template (such as lateral extension, symmetrical layout, etc.) to fit the component axis. Then, based on the template, possible endpoint ranges are generated to define the crossarm replacement path. The system performs spatial sampling based on the known area of ​​the crossarm axis, according to the standard hanging point spacing (e.g., equidistant 1.2 meters or 1.5 meters). These equidistant points are then used as candidate areas for potential hanging points, thus constructing a set of implicit extended areas for hanging points. The system combines a standard structural semantic graph model to list the possible locations and connection methods of each type of component within the structural hierarchy. Based on existing point cloud and structural graph information, it constructs all enumerable completion locations and configuration candidates. The system outputs an implicit enhanced attribute dataset for subsequent completion inference. Each data item represents a structural attribute vector of a implicitly missing component, including its spatial location, type inference, adjacent upper and lower structures, and standard constraint mapping relationships.

[0129] Implicit inference is performed based on implicit augmented data to obtain implicit missing data.

[0130] Specifically, the system first constructs a graph structure representation of all implicit enhancement candidate components. The graph structure is defined as follows: a set of graph nodes represents all candidate components; a set of graph edges represents the structural connections or spatial adjacencies between components; and a set of node attributes contains the attribute vectors of each component, such as 3D position, inferred component type, adjacency relationship, spatial spacing, and template index. Based on the above graph structure input, a graph neural network model (e.g., Graph Convolutional Network (GCN) or Graph Attention Network (GAT)) is used for training or transfer learning to predict whether each candidate component should actually exist. The output of the graph neural network model is a probability score, representing the confidence level of a candidate component as a real component. The system can use a threshold (e.g., confidence level greater than 0.85) for explicit labeling, identifying it as a latent missing component. For example, if the neural network output probability of a certain node component is 0.92, then that component will be labeled as a "latent missing component" and included in the completion process. Without introducing a neural network model, the system can use structural rule matching and spatial relationship patterns for explicit reasoning and judgment. This method, based on component dependency constraints and spatial layout rules in the standard tower structure model, executes the following logic: If an insulator component exists approximately 2.8 meters above the main pole segment, but no connecting component point cloud observations are available for that location, the system infers that a hanging point component should exist at that location; if the spatial distance from the main pole to both the left and right crossarms is approximately 1.5 meters (tolerance range ±10 cm), but there are no structural connecting segments, the system infers that a tie rod component may be missing. Through the above semantic rule matching method, the system can perform explicit missing component judgments without relying on a training model. The output is a set of implicitly missing components, including the spatial location, structural type, and inference confidence score for each component.

[0131] Optionally, the temporal structure degradation processing includes:

[0132] The tower time-series structure evolution diagram is constructed based on the implicit missing data to obtain the tower time-series structure evolution diagram data;

[0133] Specifically, the input data includes the following two categories: (1) implicit missing data, which refers to records of certain structural components missing in the data due to occlusion, wear or damage during point cloud acquisition at different times, used to implicitly identify the structural degradation process; (2) component semantic information, including the component category (such as crossarm, main rod, insulator, etc.), its position coordinates in three-dimensional space, and the connection topology relationship with other components, as the basis for semantic matching and component tracking. For each point cloud acquisition time, the system extracts the tower structure state at that time point, the contents of which include the spatial coordinate information and existence status marker of each component (1 indicates existence, 0 indicates missing). This sequence is used to generate multi-time period structural state snapshots. Based on the structural state snapshots at consecutive time points, the system constructs a time-series structural evolution diagram of the tower. , For the set of nodes in the structural evolution graph, each node Representation of components At the point of time The structural state. For each edge, there is a set of edges representing structural state changes. Connecting components At consecutive time points and The state indicates the state change process of the component. The output tower time-series structural evolution diagram data is a set of triplets.

[0134] Structural degradation trajectory data is obtained by extracting the structural degradation trajectory data based on the tower's time-series structural evolution diagram data.

[0135] Specifically, the system targets each component Extract its structural state change sequence at multiple time points. ,in =1: Indicates that the component is in During this period, the status is newly added (missing becomes existing); =0: indicates that the state has not changed; =-1: Indicates that the component is missing or degraded at this stage. If consecutive -1 values ​​appear in the component state sequence (e.g., [0,0,-1,-1,-1]), it can be determined that the component has a gradual degradation trend, reflecting its continuous damage or occlusion risk over multiple time periods. To extract common degradation behaviors, the system performs trajectory similarity analysis on all component state sequences. The Dynamic Time Warping (DTW) algorithm is used to calculate the time series distance between different components, forming a similarity matrix: Based on the similarity results above, K-means or hierarchical clustering methods are used to analyze the state sequences. Clustering is performed to form several degenerate trajectory clusters. Each trajectory cluster represents a set of components with similar degradation patterns over time.

[0136] Structural degradation prediction is performed based on structural degradation trajectory data to obtain tower missing data.

[0137] Specifically, for each type of structural degradation trajectory cluster, the system uses the central trajectory c_k of the cluster as a representative to construct an evolution prediction model of the structural state. The model may include (1) a probabilistic time model such as a Markov chain, establishing a state transition matrix T, and defining the state transition probability at any time as: ,in The current state (existing / missing). For the state at the next moment, Its transition probability. (2) Temporal neural network model (LSTM), the input is the historical state sequence of the component. The output is a sequence of predicted states for the next n time steps: Each predicted value ∈{1,0,-1}, For the future State prediction data for each time step The value of is 1…n, and the system outputs the confidence probability. Used for risk assessment. The system determines the potential missing risk of components based on the prediction results. A component is marked as a missing candidate if the following conditions are met: All satisfy: , where k is the set prediction time window length (e.g., 3 to 5 time steps). To predict the confidence level of missing states, the system outputs a set of tower missing prediction data.

[0138] Optionally, S3 includes:

[0139] S31. Calculate the height trend of adjacent towers based on the missing tower data to obtain the height trend data of adjacent towers;

[0140] Specifically, the system acquires the following input data items: the tower height parameters to be completed in the current tower model; the complete point cloud models of two or more towers adjacent to the current tower and their known tower height values, denoted as the tower height value of the previous tower and the tower height value of the next tower; and the spatial orientation information of the transmission line, including electrical and structural factors such as the line path direction, terrain slope changes, span distance, and suspension point height. To effectively predict the current tower height, the system establishes an adjacent tower height trend model based on the following methods: A local tower height sequence is constructed by selecting the tower height values ​​at two locations before and after the current tower (the selectable range is the two towers before and after the current tower). Trend estimation is performed based on the local sequence, with fitting methods including but not limited to simple weighted average, linear regression, and local spline interpolation to handle nonlinear variations. If the terrain of the location has significant undulations, the system introduces a terrain elevation difference factor as a correction term to improve the accuracy of trend prediction. If the cable types connected to adjacent towers differ, the system can normalize the structural parameters of different tower types to a unified electrical equivalent tower height, ensuring the consistency and comparability of the model predictions. The estimated trend tower height value of the current tower is output as the basic reference data for reverse inference of tower height.

[0141] S32. Calculate the structural difference index based on the adjacent tower height trend data to obtain the structural difference index data;

[0142] Specifically, the system requires input data including point cloud parameters of the current tower structure, such as the height of the crossarm center, the height of the bottom platform, and the spatial coordinates of insulator components; and the estimated tower height trend obtained from the previous step, i.e., the reference tower height fitted to adjacent structures. The system combines multiple dimensions of structural difference factors to construct a structural difference index. This index mainly reflects the degree of deviation between the current tower structure and the trend tower type in terms of geometric height, component density, and spatial configuration of key nodes. Specifically, it includes: First, a height deviation term, representing the absolute difference between the measurable structural height in the actual point cloud and the trend tower height. The actual structural height refers to the height of the point cloud range from the bottom of the tower to the top of the crossarm. Second, a vertical density variation term, representing the change in point cloud density gradient along the Z-axis, i.e., the rate of change in the number of component point clouds per unit height, reflecting structural continuity and component stacking patterns. Third, a topological offset term, representing the degree of spatial offset of key structural nodes such as crossarms and diagonal braces relative to their expected positions in the standard template, calculated as relative offset distance or directional angle error. The three difference factors mentioned above will be combined in a weighted manner to form the structural difference index. The mathematical representation of the difference index is as follows: ,in It is the structural difference index (which measures the overall geometric and topological difference between the actual structure and the trend structure). This is the weighting coefficient for the height deviation term, with a value of 0.5. This represents the actual point cloud structure height (the vertical height of the point cloud from the bottom of the tower to the top of the crossarm). To fit the tower height trend estimate (the reference tower height estimated from the adjacent tower heights). This is the weighting coefficient for the point cloud density variation term, with a value of 0.3. This represents the vertical point cloud density gradient change (the rate of change of the number of points per unit height). This is the weighting coefficient for the topology offset term of the critical node, with a value of 0.2. This refers to the topological offset (such as the relative offset or angular deviation between the spatial position of key nodes like crossarms and diagonal braces and the standard template). The system outputs a standardized structural difference index, whose numerical range can be linearly normalized to the interval [0,1].

[0143] S33. Optimize the tower height for tension balance based on the structural difference index data to obtain the tower height inference data.

[0144] Specifically, in the actual erection of transmission lines, the conductors between adjacent towers must maintain a state of tension balance. Since changes in tower height directly affect the conductor suspension angle, and thus the overall mechanical stability, the current tower height should satisfy the following tension balance relationship: current tower position tension... Tangent angle ≈ tension between adjacent tower sites The tangent angle, where the tension angle can be approximated as the arctangent function of the ratio of the height difference between two adjacent towers to their horizontal span, is used as the optimization objective. To deduce the optimal tower height from the known height and span information of adjacent towers, the system inputs the known heights of the towers on both sides; the corresponding horizontal span; the calculated structural difference index; and the current tower height. The system uses the current tower height as a variable and constructs the following objective function: minimizing the weighted sum of the absolute value of the tension difference between the two sides and the structural difference index. This objective function balances physical tension symmetry and point cloud structure consistency, avoiding deviations from the actual structure in the optimization results. The tension value can be calculated based on height and span using an electrical simulation model or mechanical formulas; the structural difference index is introduced as a regularization term, with its influence adjusted by weighting coefficients. To efficiently solve the above nonlinear optimization problem, the system can employ optimization algorithms such as Particle Swarm Optimization (PSO), gradient descent, or Newton-Raphson iteration. During the optimization process, the tower height search range can be set to a symmetrical interval of the trend tower height value, such as from the trend estimate minus a certain height deviation to the addition of that deviation, ensuring that the search space is reasonable and does not deviate from the actual engineering range. The tower height value output by the optimization is the reverse inferred tower height of the current tower, denoted as the inferred tower height value.

[0145] Optionally, S4 includes:

[0146] S41. Construct a geometric model of the tower height based on the tower height inference data to obtain the tower height geometric model;

[0147] Specifically, the construction of the tower height geometric model relies on the following input data: the inferred tower height, representing the estimated total height of the target tower, obtained by the structural parameter inversion module, in meters; and structural geometric parameter information extracted from point cloud data, including the reference position of the tower feet in three-dimensional space, the local geometric dimensions of the components, the inclination parameters of each structural segment, and the outer contour shape. The system uses a standard tower structure template as the basic geometric framework and adjusts the proportions of its main tower segments to ensure that the model height matches the inferred value. The specific processing is as follows: Let the original height of the standard tower be... Calculate the scaling factor of the main segment. : , To determine the final inferred tower height, a scaling factor is used to linearly stretch the tapered and contour sections within the main tower region of the tower template, while maintaining the positions of crossarms and tower feet unchanged or adapting them according to rules. After the geometric model is constructed, the system further constructs its structural topology model. , where the set of nodes This represents the connection points of various structural components, including main rod connection nodes, crossarm support points, cable splicing points, etc. Each node records its three-dimensional spatial coordinate information; edge set Each edge represents the connection between two components, constructed according to standard tower design rules to maintain structural topological consistency. This topology model also retains spatial annotations for key functional locations, such as conductor attachment points and signal feed points.

[0148] S42. Perform electromagnetic field calculations on the geometric model of the tower height to obtain electromagnetic field data for the tower height;

[0149] Specifically, the electromagnetic field simulation aims to model the radiated electromagnetic response of transmission towers under actual operating conditions. The main outputs include the spatial three-dimensional electric field intensity distribution E(x,y,z), in volts per meter (V / m); and the spatial three-dimensional magnetic induction vector B(x,y,z), in tesla (T). Solution methods and parameter settings: The system uses the boundary element method (BEM) or the finite element method (FEM) to numerically solve the tower structure and cable system. Input parameters include conductor current values, such as the current per phase conductor in a 220kV transmission line (approximately 400–600 amperes); cable path information, including the spatial coordinates, height, and path curvature of the conductor suspension points; and environmental medium parameters, such as the relative permittivity of air and the ground reflection boundary model (ideal conductor / hybrid boundary, etc.). The system sets the solution domain as follows: horizontally, it extends outward from the center of the tower with a radius of 30 meters; vertically, it extends upward from the base of the tower to a range 1.2 times the tower height. Within this space, an equally spaced three-dimensional simulation mesh is constructed for discrete electromagnetic field calculation. The electromagnetic simulation results output by the system include the following fields: electric field strength E_i(x,y,z): representing the electric field strength at the mesh point (x,y,z); magnetic field strength B_i(x,y,z): representing the magnetic induction vector at the same location.

[0150] S43. Obtain real-time electromagnetic field data, and construct an error field based on the real-time electromagnetic field data and the tower height electromagnetic field data to obtain error field data.

[0151] Specifically, the system deploys multiple sets of electromagnetic field sensors at key locations around the tower, including electric field sensors (such as spherical electric field probes) to collect the E field strength (unit: V / m); and magnetic field sensors (such as Hall probes and induction coils) to collect the B field vector components (unit: T). Each measurement point j ∈ measurement point set J corresponds to a set of measured data pairs (E_j^real, B_j^real). All data are timestamped during acquisition, and the system automatically removes data affected by external transient interference (such as lightning discharge, sudden arcing, etc.). The system performs error calculations at the intersection of the simulated field F_sim and the measured point set J. For each spatial coordinate (x, y, z), the magnitude errors of the electric and magnetic fields are calculated separately, defined as follows: Electric field error: ,in The electric field error (the vector norm of the difference between the measured and simulated electric field modulus, in V / m) is the error in electric field. The measured electric field intensity vector is... The simulated electric field intensity vector; magnetic field error: ,in The magnetic field error (the vector norm of the difference between the measured and simulated magnetic field modulus, in T) The measured magnetic flux density vector. This simulates the magnetic flux density vector. Error calculation is based on the vector norm form, reflecting the absolute deviation between the measured and simulated values, and possesses physical interpretability. The system organizes all spatial point error information into a three-dimensional error field tensor.

[0152] S44. Based on the error field data, perform residual optimization on the tower height inference data to obtain tower height fitting data for parameter backfilling.

[0153] Specifically, the system constructs a total field error function L(H) with the tower height H as the variable, which is used to measure the difference between the electromagnetic simulation results and the measured data under this height condition. ,in The electromagnetic total field error function (the objective function with tower height as the variable). The set of simulation sampling points represents the set of spatial locations involved in error calculation. This is the electric field error weighting coefficient (used to adjust the contribution of the electric field residual to the total error). The electric field error term represents the error at height. Below, the error value between the simulated electric field and the measured electric field at point (x,y,z) is given. This is the magnetic induction intensity error weighting coefficient (used to adjust the contribution of the magnetic field residual to the total error). This is the magnetic field error term, representing the error at height. Below, the error value between the simulated magnetic induction intensity and the measured value at point (x,y,z) is given. H is the tower height variable. Depending on the nonlinearity of the objective function, the system selects different optimization methods: for functions with complex shapes and multiple local extrema, the system preferentially uses particle swarm optimization (PSO) or simulated annealing to ensure the search for the globally optimal tower height; if the function L(H) is approximately convex in the parameter neighborhood, the system uses gradient descent for iterative updates, with the update formula as follows: ,in To optimize the updated tower height value (the current estimate obtained after gradient descent or heuristic algorithm solution), For the tower height variable, The learning rate is the step size parameter in gradient descent iterations, controlling the update magnitude in each step. Let H be the derivative of the error function with respect to H. The system sets the following convergence criteria: when the error decrease ΔL between two consecutive iterations is less than the set minimum tolerance (e.g., 0.01), the optimization is considered to have converged; or the calculation is terminated when the number of iterations reaches the upper limit (e.g., 50 times) to avoid resource exhaustion. The output optimization result is the fitted tower height value, which serves as the geometric correction parameter based on the error inversion mechanism. If the system needs to backfill and update the tower structure model, it adjusts the height ratio of the main tower segment in the geometric modeling module based on the fitted tower height value, and marks this segment as the electromagnetic fitting backfill area in the structural model.

[0154] Optionally, this application also provides a point cloud-based power model parameter backfilling system for performing the point cloud-based power model parameter backfilling method described above, wherein the point cloud-based power model parameter backfilling system includes:

[0155] The point cloud modeling and structure reconstruction module is used to acquire tower point cloud data and construct tower models based on the tower point cloud data to obtain tower models.

[0156] The missing focus and incompleteness identification module is used to perform residual focus identification on the tower model to obtain residual focus data; perform electromagnetic field back-engineering based on the residual focus data to obtain missing tower parameter data; perform tower incompleteness identification based on the missing tower parameter data to obtain tower incompleteness data; perform implicit missing data supplementation on the tower incompleteness data to obtain implicit missing data; and perform temporal structure degradation processing based on the implicit missing data to obtain tower missing data.

[0157] The tower height inference and structural parameter regression module is used to infer the tower height in reverse based on the missing tower data, and obtain the tower height inference data.

[0158] The electromagnetic fitting optimization and parameter backfilling module is used to optimize the electromagnetic field fitting of the tower height estimation data to obtain tower height fitting data for parameter backfilling.

[0159] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended application documents rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the application documents be incorporated into the invention.

[0160] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A point cloud based power model parameter backfilling method, characterized in that, The method comprises: S1, acquiring tower pole point cloud data, and constructing a tower pole model according to the tower pole point cloud data to obtain the tower pole model; S2, performing residual focus identification on the tower pole model to obtain residual focus data; performing electromagnetic field back propagation according to the residual focus data to obtain tower pole parameter missing data; performing tower pole non-integrity identification according to the tower pole parameter missing data to obtain tower pole non-integrity data; performing implicit missing supplement on the tower pole non-integrity data to obtain implicit missing data; and performing time sequence structure degradation processing according to the implicit missing data to obtain tower pole missing data; S3, performing tower height reverse inference according to the tower pole missing data to obtain tower height inference data; S4, performing electromagnetic field fitting optimization on the tower height inference data to obtain tower height fitting data for parameter back filling operation; The electromagnetic field back propagation comprises: According to the residual focus data, electromagnetic field data is collected through a magnetic field sensor pre-installed in the terminal to obtain residual area electromagnetic data; a tower type electromagnetic field simulation model is constructed according to the residual area electromagnetic data to obtain the tower type electromagnetic field simulation model; high sensitivity feature data is obtained by extracting high sensitivity features from the tower type electromagnetic field simulation model; and tower pole parameter missing data is obtained by inferring the structure parameter missing amount from the high sensitivity feature data; The time sequence structure degradation processing comprises: According to the implicit missing data, a tower pole time sequence structure evolution graph is constructed to obtain tower pole time sequence structure evolution graph data; structure degradation track data is obtained by extracting structure degradation tracks according to the tower pole time sequence structure evolution graph data; and tower pole missing data is obtained by predicting the structure degradation according to the structure degradation track data.

2. The method of claim 1, wherein, The residual focus identification comprises: A tower line coupling path is constructed for the tower pole model to obtain tower line coupling path data; Residual template matching is performed according to the tower line coupling path data to obtain residual matching data; Pseudo missing peeling is performed according to the residual matching data to obtain pseudo missing peeling data; Residual graph data is obtained by generating a residual graph for the pseudo missing peeling data; Focusing area extraction is performed on the residual graph data to obtain residual focus data.

3. The method of claim 1, wherein, The high sensitivity feature extraction comprises: Parameter-field mapping is performed on the tower type electromagnetic field simulation model to obtain parameter-field mapping data; Local sensitivity data is obtained by calculating local sensitivity according to the parameter-field mapping data; Full space sensitivity map data is obtained by constructing a full space sensitivity map according to the local sensitivity data; High response area data is obtained by extracting a high response area from the full space sensitivity map data, wherein the high response area is obtained by clustering calculation on the full space sensitivity map data and sorting according to the number of data in the cluster, and the first 5 or 10 number of processing processes are taken; Parameter-region correspondence relationship extraction is performed according to the high response area data to obtain high sensitivity feature data.

4. The method of claim 1, wherein, The tower pole non-integrity identification comprises: The missing alignment data is obtained by aligning the tower pole parameter missing data and pre-set tower pole function graph data; Geometric continuity feature data and topological connectivity feature data are obtained by respectively extracting geometric continuity features and topological connectivity features from the missing alignment data; According to the geometric continuity feature data and the topological connectivity feature data, structure graph data is obtained by performing graph attention structure processing; According to the structure graph data, non-complete region mining is performed to obtain tower pole non-completeness data.

5. The method of claim 1, wherein, The implicit missing supplement includes: The implicit checking is performed on the tower pole non-completeness data to obtain implicit checking data; According to the implicit checking data, attribute expansion is performed to obtain implicit enhancement data; According to the implicit enhancement data, implicit reasoning is performed to obtain implicit missing data.

6. The method of claim 1, wherein, S3 includes: According to the tower pole missing data, adjacent tower height trend calculation is performed to obtain adjacent tower height trend data; According to the adjacent tower height trend data, structure difference index calculation is performed to obtain structure difference index data; According to the structure difference index data, tension balance tower height optimization is performed to obtain tower height inference data.

7. The method according to claim 1, characterized in that S4 It includes: The tower height geometric model is constructed on the tower height inference data to obtain the tower height geometric model; The electromagnetic field calculation is performed on the tower height geometric model to obtain the tower height electromagnetic field data; Real-time electromagnetic field data is obtained, and error field construction is performed according to the real-time electromagnetic field data and the tower height electromagnetic field data to obtain error field data; The residual error optimization is performed on the tower height inference data according to the error field data to obtain the tower height fitting data for parameter backfill operation.

8. A point cloud based power model parameter backfilling system, characterized in that, The point cloud-based power model parameter backfill method as claimed in claim 1 is executed, and the point cloud-based power model parameter backfill system includes: The point cloud modeling and structure reconstruction module is used to obtain tower pole point cloud data, and construct a tower pole model according to the tower pole point cloud data to obtain the tower pole model; The missing focus and non-completeness identification module is used to identify residual error focus of the tower pole model to obtain residual error focus data; perform electromagnetic field back propagation according to the residual error focus data to obtain tower pole parameter missing data; identify tower pole non-completeness according to the tower pole parameter missing data to obtain tower pole non-completeness data; perform implicit missing supplement on the tower pole non-completeness data to obtain implicit missing data; and perform time sequence structure degradation processing according to the implicit missing data to obtain tower pole missing data; The tower height reasoning and structure parameter regression module is used to perform tower height reverse inference according to the tower pole missing data to obtain tower height inference data; The electromagnetic fitting optimization and parameter backfill module is used to perform electromagnetic field fitting optimization on the tower height inference data to obtain tower height fitting data for parameter backfill operation.

Images