A power transmission tower jumper form full-area laser radar measurement system

Abstract

The present application relates to the technical field of power transmission line detection, and discloses a kind of transmission tower jumper form full-area laser radar measurement system.The system includes point cloud acquisition module, implicit space conversion module, jumper form generation module and multi-dimensional difference analysis module.Point cloud acquisition module obtains the original three-dimensional point cloud data set of transmission tower jumper by full-area laser radar scanning;Implicit space conversion module converts the original three-dimensional point cloud data set into low-dimensional implicit space vector set;Jumper form generation module generates jumper form evolution track based on the implicit space vector corresponding to original three-dimensional point cloud and jumper design parameter;Multi-dimensional difference analysis module calculates spatial position difference, curvature variation difference and amplitude fluctuation difference according to the current implicit space vector corresponding to current scanning point cloud and the implicit space vector on jumper form evolution track.The system can realize full-area measurement, is not affected by environment, and improves the comprehensiveness and accuracy of jumper form measurement.

Description

Technical Field

[0001] This invention relates to the field of power transmission line detection technology, specifically to a full-area lidar measurement system for power transmission tower jumper configuration. Background Technology

[0002] As a key component connecting different transmission towers, the condition of transmission tower jumpers directly affects the safe and stable operation of transmission lines. During long-term operation, jumpers are affected by various factors such as the natural environment and mechanical stress, which can easily lead to deformation, displacement, and other problems. If these problems are not detected and addressed in time, they may cause line faults or even large-scale power outages.

[0003] Currently, the measurement of jumper configurations on transmission towers mainly relies on traditional manual inspections and partially automated testing methods. Manual inspections typically require workers to climb the transmission towers and use tools such as rulers and total stations to make measurements. This method is not only inefficient, but also limited by the skill level of the workers and environmental conditions, making it difficult to guarantee measurement accuracy. It also poses high safety risks, especially in inclement weather conditions, where the feasibility of manual inspections is significantly reduced.

[0004] Existing automated inspection methods partially employ machine vision technology, acquiring jumper wire images via cameras and then performing morphological analysis. However, this method is susceptible to factors such as lighting conditions and weather. In overcast or foggy weather, image quality deteriorates, leading to reduced measurement accuracy or even malfunction. Furthermore, machine vision technology struggles to acquire the three-dimensional spatial information of jumper wires, making it difficult to comprehensively and accurately analyze complex jumper wire morphological changes.

[0005] Some measurement systems use partial scanning, only acquiring point cloud data for a portion of the jumper area, failing to achieve full-area coverage. This limits the understanding of the overall jumper morphology, potentially missing key deformation areas and affecting the assessment of the jumper's operational status. Furthermore, these systems lack effective methods for data processing and analysis to transform and utilize the raw point cloud data, making it difficult to generate the evolution trajectory of the jumper morphology or to conduct in-depth analysis of jumper morphological changes from multiple dimensions, resulting in an incomplete and inaccurate assessment of the jumper's condition. Summary of the Invention

[0006] The purpose of this invention is to provide a full-area lidar measurement system for the jumper configuration of power transmission towers, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A full-area lidar measurement system for jumper morphology of power transmission towers includes: a point cloud acquisition module, a hidden space conversion module, a jumper morphology generation module, and a multi-dimensional difference analysis module;

[0009] The point cloud acquisition module obtains the original three-dimensional point cloud data set of the power transmission tower jumper wires through full-area lidar scanning.

[0010] The latent space transformation module converts the original three-dimensional point cloud data set into a low-dimensional latent space vector set.

[0011] The jumper shape generation module generates the jumper shape evolution trajectory based on the latent space vector corresponding to the original three-dimensional point cloud data and the jumper design parameters.

[0012] The multi-dimensional difference analysis module calculates the spatial position deviation of the jumper, the jumper curvature change offset index, and the jumper amplitude similarity based on the latent space vector corresponding to the current scanned 3D point cloud data and the latent space vector on the jumper morphology evolution trajectory.

[0013] Preferably, the latent space transformation module includes: a point cloud encoder and a feature decoder;

[0014] The point cloud encoder converts the original 3D point cloud data set into discrete feature vectors;

[0015] The feature decoder reconstructs discrete feature vectors into a set of latent space vectors;

[0016] Before running, the latent space transformation module needs to be trained through an adversarial training network. The training process includes: inputting historical jumper 3D point cloud data samples to the point cloud encoder to generate discrete feature vectors, reconstructing latent space vectors through the feature decoder, evaluating the quality of the reconstructed latent space vectors by the discriminator, and optimizing the network parameters of the latent space transformation module by minimizing the loss function of reconstruction loss and adversarial loss.

[0017] Preferably, the jumper shape generation module applies noise perturbation to the latent space vector, uses jumper design parameters as conditional input, and concatenates them with the latent space vector as conditional constraints. The conditional constraints guide noise removal and generate the jumper shape evolution trajectory.

[0018] The jumper pattern generation module optimizes its parameters by maximizing the log-likelihood estimate of the pattern evolution trajectory.

[0019] Preferably, the multi-dimensional difference analysis module includes: a spatial deviation calculation unit, a curvature offset detection unit, and an amplitude similarity evaluation unit;

[0020] The spatial deviation calculation unit is used to align the current hidden space vector with the hidden space vector on the morphological evolution trajectory and calculate the spatial position deviation of the jumper.

[0021] The curvature offset detection unit extracts point cloud curvature features from the current three-dimensional point cloud data and the morphological evolution trajectory, and calculates the jumper curvature change offset index.

[0022] The amplitude similarity evaluation unit is used to extract amplitude spectrum features from the current three-dimensional point cloud data and morphological evolution trajectory, and calculate the jumper amplitude similarity.

[0023] Preferably, the execution process of the spatial deviation calculation unit includes: determining the target trajectory point on the jumper morphology evolution trajectory that is closest to the current latent space vector, and calculating the jumper spatial position deviation based on the target trajectory point;

[0024] The curvature offset detection unit extracts the actual curvature energy of the current three-dimensional point cloud data through a multi-scale curvature decomposition algorithm, and adjusts the theoretical curvature energy of the jumper in the initial state according to the morphological parameters corresponding to each latent space vector on the jumper morphology evolution trajectory. It calculates the ratio of theoretical curvature energy to actual curvature energy density as the jumper curvature change offset index.

[0025] The amplitude similarity evaluation unit uses a point set distance algorithm to match the phase error between the set of amplitude mutation points in the current three-dimensional point cloud data and the set of amplitude mutation points on the jumper morphology evolution trajectory, and uses this error as the jumper amplitude similarity.

[0026] Preferably, it also includes a key point topology association module; the key point topology association module constructs a three-dimensional difference matrix by considering the jumper spatial position deviation, jumper curvature offset index and jumper amplitude similarity, and inputs it into the jumper spatial topology network;

[0027] The jumper spatial topology network is constructed based on the jumper structure parameters of the transmission tower to establish node connection relationships, and the spatial distance constraints between nodes are marked.

[0028] Preferably, the key point topology association module performs anomaly propagation deduction:

[0029] The jumper morphology deviation attenuation factor is calculated based on the spatial distance between nodes. Cross-node association features are captured through a multi-head attention mechanism. The path simulation algorithm is used to deduce the diffusion path of jumper morphology deviation in the jumper spatial topology network.

[0030] The frequency of deviations occurring at each node during the anomaly propagation simulation is statistically analyzed, and an anomaly probability distribution map for each node is generated by combining the spatial distance between nodes.

[0031] Preferably, it also includes a dynamic measurement strategy module; the dynamic measurement strategy module configures scanning parameters according to the anomaly probability distribution map of the nodes: enabling a high-resolution scanning mode for high-probability anomaly areas, and applying multi-band spatial sampling perturbation to adjacent nodes of high-probability anomaly areas.

[0032] Preferably, the dynamic measurement strategy module applies multi-band spatial sampling perturbations to adjacent nodes in high-probability anomaly regions, including:

[0033] Multi-wavelength test signals are injected using a controllable laser source, and point cloud response data for each wavelength band are recorded.

[0034] Theoretical point cloud response data for each band is calculated based on jumper space topology network;

[0035] Compare the Euclidean distance between the point cloud response data and the theoretical point cloud response data for each band. If the Euclidean distance exceeds the set abnormal offset threshold, it is determined to be an abnormal offset.

[0036] Preferably, it also includes a point cloud stability analysis module; the point cloud stability analysis module performs box plot cleaning on the historical point cloud dataset and calculates the point cloud density fluctuation factor, curvature stability factor and amplitude dispersion factor;

[0037] A three-dimensional feature space for point clouds is constructed using point cloud density fluctuation factor, curvature stability factor, and amplitude discrepancy factor. A spatial search algorithm is used to find the optimal stability region in the three-dimensional feature space, determine the optimal scanning interval, and trigger periodic point cloud acquisition based on the optimal scanning interval.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] By setting up a point cloud acquisition module, it is possible to use lidar technology to scan the entire area of ​​the power transmission tower jumper wires and obtain a complete set of original three-dimensional point cloud data. Compared with the traditional local scanning method, this full-area measurement method can fully cover all parts of the jumper wires without missing any area's morphological information, thus providing a complete data foundation for subsequent analysis and helping to understand the overall morphological condition of the jumper wires more comprehensively.

[0040] The latent space transformation module can convert a large and complex set of original 3D point cloud data into a set of low-dimensional latent space vectors. This transformation process simplifies the complexity of the data, making the high-dimensional point cloud data that was originally difficult to process and analyze directly more concise and orderly. This makes it easier for subsequent modules to process and mine the data efficiently. At the same time, it reduces the pressure of data storage and transmission, and improves the overall operating efficiency of the full-area lidar measurement system for power transmission tower jumper morphology.

[0041] The jumper morphology generation module generates a jumper morphology evolution trajectory based on the initial latent space vector corresponding to the original 3D point cloud data and the jumper design parameters. This trajectory can intuitively reflect the morphological change process of the jumper from its initial state to its current state, helping to trace the history of jumper morphological changes and understand the trends and patterns of morphological changes. Through this trajectory, the morphological characteristics of the jumper at different stages can be clearly seen, thus better grasping the internal logic of jumper morphological changes and providing a powerful reference for understanding the stress conditions and deformation causes of the jumper.

[0042] The multi-dimensional difference analysis module can calculate spatial position differences, curvature change differences, and amplitude fluctuation differences based on the current latent space vector corresponding to the current scanned point cloud and the latent space vector on the jumper morphology evolution trajectory. This multi-dimensional difference analysis overcomes the limitations of traditional single-dimensional analysis, providing a more comprehensive and detailed reflection of jumper morphology changes. Spatial position differences reflect the jumper's displacement in space, curvature change differences reflect changes in the jumper's bending degree, and amplitude fluctuation differences help understand changes in the jumper's vibration. Through this multi-dimensional difference information, abnormal changes in jumper morphology can be identified more accurately, potential problems can be detected in a timely manner, and a more comprehensive and accurate assessment of the jumper's operational status can be made.

[0043] This invention presents a full-area lidar measurement system for transmission tower jumper configurations. LiDAR technology itself is unaffected by environmental factors such as lighting conditions and weather, enabling the system to operate stably in various complex environments. This ensures the continuity and reliability of measurement work and avoids the measurement failures caused by environmental factors inherent in traditional machine vision technology. Furthermore, the entire system automates the entire process from data acquisition, conversion, processing to analysis, reducing manual intervention, minimizing the impact of human factors on measurement results, and improving the objectivity and consistency of measurements. Attached Figure Description

[0044] Figure 1 This is a timing diagram of the full-area lidar measurement system for the jumper configuration of transmission towers according to the present invention.

[0045] Figure 2 A flowchart illustrating the operation of the jumper pattern generation module;

[0046] Figure 3 A flowchart illustrating the operation of the multidimensional difference analysis module;

[0047] Figure 4 A flowchart illustrating the anomaly propagation in the key point topology association module;

[0048] Figure 5 This is a flowchart of the multi-band sampling disturbance for the dynamic measurement strategy module. Detailed Implementation

[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 This invention provides a full-area lidar measurement system for jumper morphology of power transmission towers, comprising: a point cloud acquisition module, a hidden space conversion module, a jumper morphology generation module, and a multi-dimensional difference analysis module;

[0051] The point cloud acquisition module scans the entire area of ​​the power transmission tower jumpers using LiDAR equipment deployed throughout the tower to obtain a raw 3D point cloud dataset. This module employs a high-precision LiDAR sensor, covering the entire area of ​​the power transmission tower jumpers at a preset scanning angle and resolution, ensuring the integrity and accuracy of the raw 3D point cloud data.

[0052] The latent space transformation module receives the original 3D point cloud data set and transforms it into a low-dimensional latent space vector set through dimensionality reduction processing in order to extract the core features of the original 3D point cloud data.

[0053] The jumper morphology generation module generates a jumper morphology evolution trajectory based on the latent space vector corresponding to the original 3D point cloud data and jumper design parameters such as jumper length, tension, and environmental conditions. This jumper morphology evolution trajectory describes the sequence of morphological changes of the jumper under different conditions.

[0054] The multi-dimensional difference analysis module compares and analyzes the latent space vectors corresponding to the currently scanned 3D point cloud data with the latent space vectors on the trajectory of the jumper shape evolution, calculates the differences in spatial position, curvature change, and amplitude fluctuation, and thus assesses the deviation of the jumper shape.

[0055] In one embodiment of the present invention, the 3D point cloud data is first preprocessed by voxelization to convert it into a regular mesh structure, facilitating processing by the latent space transformation module. The latent space transformation module consists of a point cloud encoder and a feature decoder. The point cloud encoder uses a deep neural network structure to process the voxelized 3D point cloud data, progressively extracting the spatial features of the 3D point cloud data. Specifically: the convolutional layer of the point cloud encoder performs convolution operations on the voxelized 3D point cloud data, capturing local geometric features of the 3D point cloud data by sliding the convolution kernel; activation functions such as ReLU are embedded in the convolutional layer to enhance the nonlinear fitting ability of the point cloud encoder; the pooling layer of the point cloud encoder aggregates the local geometric features output by the convolutional layer, retaining key information while reducing the data dimensionality, achieving preliminary integration of global information; the fully connected layer at the end of the encoder further maps the features aggregated by the pooling layer into discrete feature vectors of fixed length, completing the dimensionality compression and feature encoding of the high-dimensional point cloud data. The feature decoder receives the discrete feature vectors and reconstructs them into low-dimensional latent space vectors through deconvolution layers and upsampling operations. The network structure of the feature decoder is symmetrical to the network structure of the point cloud encoder, thereby progressively recovering the key features of the 3D point cloud data. Because the reconstruction process of the feature decoder focuses on preserving the topological structure and geometric properties of the 3D point cloud data, the set of latent space vectors forms a compact representation of the 3D point cloud data.

[0056] The latent space transformation module is trained using an adversarial training network framework. Training uses historical jumper 3D point cloud data samples as input. The point cloud encoder converts the samples into discrete feature vectors, and the feature decoder attempts to reconstruct the latent space vectors. Simultaneously, a discriminator network participates in training; its structure includes multiple convolutional layers to evaluate the distributional consistency between the reconstructed and ground truth latent space vectors. The training process optimizes the network parameters of the latent space transformation module by minimizing the reconstruction loss and adversarial loss functions. The reconstruction loss is calculated by determining the mean squared error between the reconstructed and ground truth latent space vectors, while the adversarial loss is based on the output probability of the discriminator network. Training is iterative, updating the weight parameters of the encoder, feature decoder, and discriminator network using backpropagation until the loss functions converge and the discriminator network can no longer distinguish between the reconstructed and ground truth latent space vectors.

[0057] The point cloud encoder and feature decoder need to achieve lossless feature extraction to ensure that the latent space vector retains the complete information of the original 3D point cloud data. Specifically, the point cloud encoder adopts voxelization preprocessing and a residual convolutional network structure. First, the original 3D point cloud data is converted into a regular voxel grid to unify the input format. Then, features are extracted by stacking residual convolutional blocks. Residual connections avoid gradient vanishing during deep network training by directly mapping paths, thus preserving the subtle geometric features of the original 3D point cloud data. The feature decoder adopts a deconvolutional and residual connection structure symmetrical to the encoder. The deconvolutional layers gradually restore the feature map resolution, and the residual connections fuse the feature maps output by each layer of the encoder with the corresponding layers of the decoder to reconstruct the topological structure and spatial location information of the point cloud. During the training phase, by minimizing the reconstruction loss, the point cloud encoder is forced to remove only redundant information and retain the core features. The feature decoder accurately restores the core features, ultimately achieving the preservation of the complete information of the original 3D point cloud data in the latent space vector. Adversarial training ensures the robustness and generalization ability of latent space vector transformation. This is achieved through a three-way game mechanism involving a point cloud encoder, feature decoder, and discriminator network. The discriminator network employs a multi-layer convolutional neural network structure, taking both real and reconstructed latent space vectors as input. After feature extraction through convolutional layers, it outputs binary classification probabilities via fully connected layers, determining whether the input is a real or reconstructed latent space vector. Training employs an alternating optimization strategy: first, the parameters of the point cloud encoder and feature decoder are fixed, and the discriminator network is trained to maximize the cross-entropy loss that distinguishes between real and reconstructed latent space vectors; then, the discriminator network parameters are fixed again, and the point cloud encoder and feature decoder are trained to minimize the reconstruction loss and adversarial loss. This process iterates until all three losses converge. Through adversarial training, the point cloud encoder and feature decoder learn universally applicable features from jumper samples, rather than relying on redundant features specific to certain samples. This allows for stable transformation even when faced with new jumper samples or slight noise interference, improving the robustness and generalization ability of latent space vector transformation.

[0058] In one embodiment of the present invention, such as Figure 2 The jumper morphology generation module operates on the latent space vector domain. Starting with the latent space vector corresponding to the original 3D point cloud data of the jumper, it applies Gaussian-distributed noise perturbation. The noise level increases with the time step, gradually masking the characteristics of the latent space vector. Jumper design parameters are used as input conditions, including jumper length, tension coefficient, and environmental load parameters. These jumper design parameters are converted into vector form through an embedding layer and concatenated with the latent space vector corresponding to the original 3D point cloud data. This serves as a condition constraint for the jumper morphology generation module. The condition constraint guides the noise removal process, ensuring that the evolution trajectory of the generated jumper morphology conforms to physical laws.

[0059] In this embodiment, the noise removal process adopts the diffusion model principle. By predicting the added noise and gradually subtracting it, the jumper morphology sequence is recovered from a purely noisy state. The specific implementation process of noise removal guided by conditional constraints is as follows: the jumper design parameters, including jumper length, tension coefficient, and environmental load parameters, are converted into a fixed-dimensional parameter vector through an embedding layer. This parameter vector is then concatenated with the initial latent space vector to form a fused latent vector with conditional information, which serves as the initial input for morphological evolution. A U-Net neural network is used to perform noise prediction. This network includes an encoder, a decoder, and skip connections. The encoder extracts conditional and morphological features from the fused latent vector through convolutional and pooling layers. The decoder gradually recovers feature details through deconvolutional and upsampling layers. Skip connections fuse the feature maps of each layer of the encoder with the corresponding layers of the decoder to retain local information. The U-Net neural network takes the current noisy latent vector and the corresponding fused latent vector as input and outputs the predicted value of the added noise in this step. Subsequently, the denoised latent space vector is obtained by subtracting the predicted noise from the current noisy latent vector according to the update rule. This process is repeated multiple times to generate a sequence from the noisy state to the true morphology, i.e., the jumper morphology evolution trajectory. By forcibly introducing conditional constraints, the U-Net structure's neural network associates the correspondence between physical parameters and morphology when learning noise prediction, ensuring that the generated trajectory conforms to the morphological change law of the jumper under specific tension and length constraints, and avoiding deformation that violates physical limits.

[0060] The training of the jumper pattern generation module focuses on maximizing the log-likelihood estimate of the pattern evolution trajectory. The specific formula for the log-likelihood estimate of the pattern evolution trajectory is as follows:

[0061]

[0062] in, Let be the log-likelihood function of the morphological evolution trajectory; All learnable parameters in the jumper pattern generation module, including: convolutional kernel weights and fully connected layer weights, etc. Let the expected value be the mathematical expectation of all input samples. For latent space vectors, For the applied Gaussian noise, This represents the total number of time steps in the noise removal process. for go through The complete noisy latent space vector obtained after step noise perturbation; In order to learnable parameters Below, from fully noisy latent vectors Reverse engineering to recover the hidden space vector The conditional probability is calculated by assuming the logarithm of the probability of recovering a latent space vector from a noisy state. Maximizing this value allows the jumper pattern generation module to learn a probability distribution that conforms to the natural evolution of jumper patterns, ensuring that the generated jumper pattern trajectory is statistically close to the real pattern change process.

[0063] The training data includes historical jumper pattern evolution sequences and their corresponding design parameters. All learnable parameters in the jumper pattern generation module are optimized using a gradient ascent algorithm. The training process involves sampling multiple noisy trajectories, calculating the likelihood value of each trajectory, and updating the jumper pattern generation module through backpropagation. Iterative training enables the jumper pattern generation module to generate jumper pattern evolution trajectories that conform to the physical characteristics of jumpers.

[0064] The noise perturbation and removal processes in the jumper morphology generation module need to be precisely controlled to produce a smooth and physically reasonable sequence of jumper morphology evolution trajectories. This precise control is achieved through conditional constraints and training optimization: In the noise perturbation stage, based on the gradual characteristics of jumper physical deformation, a sequence is set where the noise variance increases linearly with the time step. Low-variance noise is applied at the initial time step to preserve the initial morphological features of the jumper, and the noise variance is gradually increased with the step size to ensure that the perturbation process conforms to the gradual nature of the jumper morphology's natural changes. In the noise removal stage, a fixed-step iterative denoising mechanism is used, with the denoising amplitude of each step strictly matching the noise intensity predicted by the network to avoid abrupt changes in the jumper morphology evolution trajectory due to excessive denoising amplitude. Simultaneously, during the training of the U-Net neural network, a physical constraint loss term is introduced, using the jumper design parameters as a regularization condition to constrain the update range of the latent space vector. In addition, a smoothness loss is added during the training process, which calculates the Euclidean distance between the latent space vectors of adjacent time steps. By minimizing the smoothness loss, it is ensured that the differences between adjacent shapes in the generated jumper shape evolution trajectory sequence are within a physically reasonable range, ultimately producing a smooth shape evolution trajectory that conforms to the mechanical properties of jumpers.

[0065] In one embodiment of the present invention, such as Figure 3 The multi-dimensional difference analysis module consists of a spatial deviation calculation unit, a curvature offset detection unit, and an amplitude similarity evaluation unit. The spatial deviation calculation unit is used to compare the spatial position difference between the current latent space vector and the latent space vector on the morphological evolution trajectory; the curvature offset detection unit is used to compare the curvature change difference between the current 3D point cloud data and the morphological evolution trajectory; and the amplitude similarity evaluation unit is used to compare the amplitude fluctuation difference between the current 3D point cloud data and the morphological evolution trajectory.

[0066] The spatial deviation calculation unit first locates the target trajectory point closest to the current latent space vector on the jumper morphology evolution trajectory using a nearest neighbor search algorithm. This search process is based on the Euclidean distance metric between vectors, selecting the trajectory point with the smallest distance as the target trajectory point. After obtaining the index of the target trajectory point, the spatial deviation calculation unit calculates the spatial position deviation between the current latent space vector and the target trajectory point based on the index. This calculation is based on the difference analysis of the three-dimensional coordinate system, comparing the coordinates of the current three-dimensional point cloud data with the coordinates of the target trajectory point point by point, and finally outputting the root mean square value of the spatial position deviation of all current three-dimensional point cloud data as the jumper spatial position deviation.

[0067] The curvature offset detection unit uses a multi-scale curvature decomposition algorithm to process the current 3D point cloud data. First, it calculates the curvature value of each point using a local surface fitting method to establish a curvature energy distribution map. From the curvature energy distribution map, it obtains the curvature energy of different frequency bands as the actual curvature energy. The multi-scale curvature analysis uses wavelet transform technology to decompose the curvature signal on the curvature energy distribution map into different frequency bands, paying particular attention to preset sensitive frequency bands. These frequency bands usually correspond to the key change features of the jumper shape. When decomposing the curvature signal into different frequency bands, the derivation of the theoretical curvature energy is first based on the jumper design parameters. The theoretical curvature distribution and corresponding theoretical curvature energy of the jumper in the initial state are obtained through structural mechanics calculations. Then, combined with the morphological parameters corresponding to each latent space vector on the jumper shape evolution trajectory, including the force state and spatial attitude at the trajectory point, the theoretical curvature energy in the initial state is adapted and adjusted to obtain the theoretical curvature energy corresponding to each time point on the jumper shape evolution trajectory. The theoretical curvature energy is directly related to the trajectory points on the jumper shape evolution trajectory, and its value dynamically matches the evolution of the shape on the jumper shape evolution trajectory. For each sensitive frequency band, the curvature offset detection unit calculates the curvature energy density ratio by comparing the actual curvature energy with the corresponding theoretical curvature energy. This ratio serves as the jumper curvature offset index, reflecting the degree of deviation between the measured curvature energy and the theoretical curvature energy. The greater the deviation of the ratio from 1, the more significant the curvature offset.

[0068] The calculation process for the curvature-energy-density ratio in this invention is as follows:

[0069]

[0070] in, Indicates sensitive frequency bands The ratio of curvature energy density under these conditions and These represent the sensitive frequency bands. The upper and lower limits of frequency, This represents the theoretical curvature energy of the curvature signal at frequency f. This represents the curvature energy distribution at frequency f.

[0071] The amplitude similarity evaluation unit first converts the spatial distribution of the current 3D point cloud data into a frequency domain representation using Fourier transform, extracting amplitude spectral features. A gradient detection algorithm is used to identify amplitude abrupt change points, locating frequency points with drastic changes in the amplitude spectral features. A point set distance algorithm is used to match theoretically expected amplitude abrupt change points with the current amplitude abrupt change points. The theoretically expected amplitude abrupt change points are directly related to points on the jumper morphology evolution trajectory, and are determined by inversely calculating amplitude features from each latent space vector on the jumper morphology evolution trajectory. Specifically, the process for determining theoretically expected amplitude abrupt change points is as follows: The jumper morphology generation module generates a jumper morphology evolution trajectory based on the initial latent space vector corresponding to the original 3D point cloud and jumper design parameters. Each latent space vector on this trajectory corresponds to a specific jumper morphology. Frequency domain analysis is performed on the jumper morphology corresponding to each latent space vector on the trajectory to extract amplitude features under each morphology. From these amplitude features, the locations of amplitude abrupt changes are located, and these locations constitute the set of theoretically expected amplitude abrupt change points. The maximum and minimum distances between this set and the current set of amplitude abrupt change points are then calculated using Hausdorff distance to achieve matching. The phase error of the amplitude abrupt change points is calculated based on the positional deviation of the abrupt change points on the frequency axis, and the overall phase error index is obtained by weighted averaging, which serves as the jumper amplitude similarity.

[0072] Spatial deviation calculation is performed first, providing a basic spatial alignment reference for subsequent analysis. The curvature offset detection unit uses the spatial alignment results to extract local curvature features, ensuring spatial consistency in curvature comparison. Amplitude similarity assessment is performed last; its analysis is based on the completion of spatial and curvature feature processing. The output results of all units are normalized to eliminate dimensional differences, facilitating subsequent comprehensive evaluation.

[0073] In terms of data stream processing, the spatial deviation calculation unit, curvature offset detection unit, and amplitude similarity evaluation unit adopt a parallel computing architecture. The spatial deviation calculation unit deploys a fast nearest neighbor search algorithm based on KD-trees to optimize the processing efficiency of large-scale 3D point cloud data. The curvature offset detection unit implements multi-threaded wavelet transform to process curvature analysis at multiple scales simultaneously. The amplitude similarity evaluation unit adopts a frequency domain block processing strategy to improve the computational performance of Fourier transform. The entire multi-dimensional difference analysis module is equipped with a result verification mechanism, which ensures the reliability of the analysis results by cross-checking the consistency of the outputs of each unit. Specifically, the spatial position deviation of the jumper output by the spatial deviation calculation unit, the curvature offset index of the jumper output by the curvature offset detection unit, and the amplitude similarity of the jumper output by the amplitude similarity evaluation unit are correlated and mapped to establish the correspondence between the three and the spatial coordinates of the jumper. Secondly, if an abnormal spatial position deviation occurs at a certain spatial coordinate, it is necessary to simultaneously check whether the curvature offset index and amplitude similarity at that coordinate also show abnormalities. If the abnormal trends of the three are consistent, the analysis results are considered reliable. If only a single dimension is abnormal, the point cloud data reprocessing process is triggered to re-execute the latent space transformation and multi-dimensional difference calculation. By comparing the secondary calculation results with the initial results, the analysis results are verified and corrected.

[0074] The configuration parameters of the multi-dimensional difference analysis module are adjusted according to the jumper type and application environment to ensure that the multi-dimensional difference analysis module can adapt to the needs of various actual measurement scenarios:

[0075] The selection of sensitive frequency bands is determined based on the material properties and typical vibration modes of the jumper wires: the natural vibration frequency range of the jumper wires is obtained through structural dynamics calculations based on the material properties of the jumper wires, such as elastic modulus, Poisson's ratio, density, and structural parameters such as cross-sectional moment of inertia and length; then, based on the curvature change signal of the jumper wires under normal operating conditions in historical point cloud data, the main vibration frequency bands are extracted using spectrum analysis, and the intersection operation is performed between the natural vibration frequency range and the main vibration frequency bands. The determined intersection frequency band is the sensitive frequency band.

[0076] The search radius for spatial position deviation calculation is dynamically adjusted according to the point cloud density: the current scanned 3D point cloud data is voxelized, and the number of points in a unit voxel is counted to determine the point cloud density; if the point cloud density is high, the search radius is set to 1 times the voxel size, and if the point cloud density is low, the search radius is set to 2 times the voxel size, so as to achieve the adaptation of the search radius and the point cloud density.

[0077] The frequency resolution for amplitude similarity assessment is adaptively set based on the sampling rate and jumper length: the Nyquist frequency is calculated based on the LiDAR sampling rate, and the minimum frequency interval to be resolved is determined based on the jumper length and common vibration wavelengths; the frequency band is divided according to the minimum frequency interval with the Nyquist frequency as the upper limit, thereby determining the frequency resolution. The higher the sampling rate, the longer the jumper length, and the smaller the minimum frequency interval, the higher the frequency resolution.

[0078] In one embodiment of the invention, a key point topology association module is further included, which receives the output results from the multi-dimensional difference analysis module, including data in three dimensions: jumper spatial position deviation, jumper curvature offset index, and jumper amplitude similarity. These data are integrated to construct a three-dimensional difference matrix, where each dimension corresponds to a difference type. The three-dimensional difference matrix assigns corresponding weight coefficients based on the importance of different difference types, and the comprehensive difference degree at a specific location on the jumper is obtained through a weighted fusion method.

[0079] The constructed 3D difference matrix is ​​input into the jumper spatial topology network, which is built based on the specific structural parameters of the transmission tower, including jumper suspension point locations, supporting rod geometric relationships, and connection node distribution. Nodes in the jumper spatial topology network represent key measurement points on the jumpers, and the connections between nodes are determined based on actual physical connections and spatial proximity principles. Each node is labeled with spatial distance constraint parameters, including the actual physical distance between nodes and the maximum allowable deviation tolerance range. The topology of the jumper spatial topology network is represented using graph theory, with nodes as vertices and connections as edges. The edge weights reflect the spatial distance and mechanical connection strength between nodes.

[0080] like Figure 4 The key point is that the topology association module performs anomaly propagation analysis:

[0081] First, the morphological deviation attenuation factor is calculated based on the spatial distance between nodes. The spatial distance is calculated using the three-dimensional Euclidean distance formula. The morphological deviation attenuation factor characterizes the degree of reduction in the propagation of morphological deviation between nodes, and its calculation is based on the following formula:

[0082]

[0083] in, This represents the morphological deviation attenuation factor from node i to node j. This represents the spatial distance between node i and node j. This is the attenuation coefficient, determined based on the jumper material properties and environmental conditions. The morphological deviation attenuation factor ensures that nodes farther apart are less affected by morphological deviations.

[0084] Secondly, a multi-head attention mechanism is employed to capture cross-node association features. This mechanism simultaneously calculates multiple attention weight matrices, each focusing on different aspects of node relationships, including spatial proximity, deviation similarity, and mechanical correlation. The attention weights are calculated based on the correlation between node feature vectors using a dot-product attention function. The outputs of multiple attention heads are concatenated and linearly transformed to form a comprehensive representation of node association features. Specifically, the multi-head attention mechanism sets three independent attention heads corresponding to spatial proximity, deviation similarity, and mechanical correlation, respectively. Each attention head independently calculates its attention weight matrix based on the jumper space topology network and multi-dimensional difference analysis results. For spatial proximity, the corresponding attention head uses the three-dimensional spatial coordinates of nodes in the topology network as input features and generates a weight matrix by calculating the spatial distance relationship between nodes. Nodes that are closer in distance receive higher attention weights in this matrix, thus focusing on the spatial location associations between nodes. For deviation similarity, the attention head integrates the spatial position deviation, curvature offset index, and amplitude similarity output from the multi-dimensional difference analysis module into a node deviation feature vector. A weight matrix is ​​generated by calculating the correlation between different node deviation feature vectors; nodes with more similar deviation features have higher weights, thus capturing the correlation of deviation patterns. For mechanical correlation, the attention head uses the mechanical correlation parameters between nodes in the topology network. These parameters are determined based on transmission tower structural parameters such as jumper suspension point positions, supporting member geometric relationships, and jumper design parameters. A weight matrix is ​​generated by quantifying the mechanical transmission strength between nodes; nodes with closer mechanical correlations have higher weights, thus focusing on the mechanical interaction relationships between nodes. The outputs of the three attention heads are concatenated and linearly transformed to form a cross-node correlation feature that comprehensively reflects the three types of node relationships.

[0085] The path simulation algorithm is responsible for deducing the propagation path of morphological deviations in the topological network. It employs a random walk model, starting from high-deviation nodes and calculating the transition probability based on node connection weights and attenuation factors. Specifically, the determination of high-deviation nodes is based on node morphological data measured by lidar. The actual measured parameters of each node are compared with preset jumper standard morphological parameters, and nodes deviating from the standard parameter threshold are selected as candidate high-deviation nodes. This process is directly related to the calculation results of the multi-head attention mechanism. Candidate nodes need to be further verified by a weight matrix of focusing deviation similarity in the multi-head attention mechanism. This weight matrix calculates the correlation degree of deviation features between the candidate node and surrounding nodes, eliminating false deviation nodes caused by local noise. Finally, high-deviation nodes with significant deviation features and deviation correlations with surrounding nodes are determined as the starting point of the random walk. The random walk model starts from this node and performs path sampling based on the feature correlation between nodes to achieve accurate traversal of the jumper morphological deviation region. The simulation process records the propagation path and intensity changes of the deviation between nodes, obtaining statistically significant propagation patterns through multiple Monte Carlo simulations. Each simulation considers the degree of deviation of the current node and the reception sensitivity of adjacent nodes. Specifically, the calculation of node connection weights is based on the node connection relationships and structural parameters of the jumper space topology network. The physical connection relationships between nodes are determined according to the transmission tower structural parameters. Then, combined with the spatial distance constraints and mechanical correlation characteristics between nodes, nodes with physical connections are assigned higher weights, while nodes without direct connections but spatially adjacent are assigned lower weights, thus obtaining the node connection weights. The calculation of transition probabilities is related to the cross-node correlation features captured by the multi-head attention mechanism. These cross-node correlation features are used to correct the node connection weights. The correlation weight for that dimension is obtained by calculating the correlation of the node feature vectors under the corresponding dimension. Then, the correlation weights output by all attention heads are concatenated and integrated into a comprehensive cross-node correlation feature through a linear transformation. The transition probability, based on the corrected node connection weights and combined with a morphological deviation attenuation factor, yields the transition probability of each node propagating to its neighboring nodes. The propagation direction is determined by selecting the neighboring node with the highest transition probability, thus determining the path for preferential deviation diffusion. The calculated transition probability is used to determine the likelihood of a deviation propagating from the current node to its neighboring nodes in a random walk model. The magnitude of the transition probability determines the priority of neighboring nodes receiving the deviation from the current node. Based on the transition probability, the propagation direction of the deviation in the topology network can be clearly determined, thereby accurately predicting the propagation path and intensity changes of the deviation.

[0086] Statistical analysis of the frequency of deviations occurring at each node during the simulation process is a crucial step in the analysis. A counter is used to record the number of times each node is traversed by the deviation propagation path, along with the intensity of the deviation at each node. Frequency statistics employ a sliding time window method to distinguish between short-term, sudden deviations and long-term, persistent deviation patterns. Combined with the spatial distance parameters of the nodes, the anomaly probability value for each node is calculated.

[0087] The final keypoint anomaly probability distribution map is generated. This map is based on the topological network nodes, with probability values ​​indicating the anomaly risk level of each node. Probability calculations are based on the frequency, intensity, and spatial location parameters of node deviations. A Bayesian probability model uses the frequency and intensity of deviations as prior information and the spatial distance between a node and high-anomaly-probability nodes as likelihood information. This information is then integrated through Bayesian probability model calculations. Nodes with high deviation frequency, high deviation intensity, and proximity to high-anomaly nodes have higher anomaly probabilities, thus obtaining the anomaly probability for each keypoint. The keypoint anomaly probability distribution map is visualized as a heatmap, with different color depths representing different probability levels for easy identification of high-risk areas.

[0088] The construction of the three-dimensional difference matrix needs to precisely correspond to the node locations in the network topology. The calculation of the attenuation factor needs to be updated in real time to adapt to environmental changes. The parameters of the multi-head attention mechanism need to be dynamically adjusted according to the network topology. The number of path simulations needs to ensure statistical significance; typically, a sufficient number of simulations are required to obtain stable results. The calculation of anomaly probabilities needs to be recalibrated periodically to ensure accuracy.

[0089] The key point anomaly probability distribution map identifies high-risk areas requiring focused attention, which may need more detailed inspection or preventative measures. The projected path information helps understand the propagation patterns of deviations, providing a theoretical basis for jumper maintenance. All output data undergoes standardization to ensure compatibility with other modules and consistency in data exchange.

[0090] In one embodiment of the present invention, such as Figure 5 It also includes a dynamic measurement strategy module, which dynamically adjusts the scanning parameter configuration of the LiDAR system based on the key point anomaly probability distribution map generated by the key point topology association module. The following is an illustration using a specific example:

[0091] Suppose that the critical point anomaly probability distribution map of a transmission tower jumper shows high anomaly probabilities in areas A7 and B3, at 0.87 and 0.92 respectively, while the probabilities in adjacent areas C5 and D2 are 0.45 and 0.38. Based on this critical point anomaly probability distribution, the dynamic measurement strategy module first enables a high-resolution scanning mode for high-probability anomaly areas. For area A7, the LiDAR scanning resolution is increased from 0.5 degrees in standard mode to 0.1 degrees, and the sampling frequency is increased from 100Hz to 500Hz. Simultaneously, the point cloud acquisition time for this area is extended from the default 2 seconds to 5 seconds to ensure sufficiently dense point cloud data is obtained. Area B3 uses the same parameter configuration, but based on its spatial location characteristics, an additional vertical scanning angle range is added. Node C5 is an adjacent node of A7, and node D2 is an adjacent node of B3. Multi-wavelength test signals are injected through a controllable laser source. The process adopts a wavelength switching mechanism, switching between three bands of 1064nm, 1550nm and 905nm in a specific time sequence. The illumination duration of each wavelength is set to 100 milliseconds, with a cooling cycle of 50 milliseconds to apply multi-band spatial sampling perturbation and simultaneously record the point cloud response data under each band.

[0092] Point cloud response data based on jumper spatial topology network computation theory: An optical reflection model is established based on the design parameters of the transmission tower and the material properties of the jumpers. This model considers the reflectivity differences of lasers at different wavelengths, the optical properties of the jumper surface material, and the influence of ambient lighting conditions. For each test wavelength, the optical reflection model generates the corresponding expected point cloud distribution, including point cloud density distribution, reflection intensity distribution, and three-dimensional coordinate features.

[0093] Calculate the Euclidean distance between the measured point cloud response data and the theoretical point cloud response data: Compare the coordinate difference distribution between the measured and theoretical point cloud response data individually for each test wavelength. The Euclidean distance is calculated using a weighted average method, assigning higher weights to high curvature regions and lower weights to flat regions. When the calculated Euclidean distance exceeds a set abnormal offset threshold, it is determined that there is an abnormal offset in that region.

[0094] Referring to Table 1, the test results of multi-band spatial sampling disturbances are presented. Based on the test results, enhanced monitoring mode is activated for areas identified as abnormal. For the abnormalities at wavelengths of 1064nm and 1550nm at nodes A7 and B3, the scanning frequency in these areas is increased, and the scanning interval is shortened from the usual 60 minutes to 15 minutes. At the same time, a data verification mechanism is triggered, and the raw point cloud data is sent to the quality control unit for further analysis.

[0095] Table 1: Multi-band spatial sampling disturbance response data

[0096]

[0097] During the execution of the dynamic measurement strategy module, changes in environmental conditions are monitored in real time. When wind speed exceeds a threshold or temperature changes drastically, the sampling time parameters of the multi-frequency bands are automatically adjusted to extend the sampling duration and ensure data stability. The module also records the equipment and environmental parameters for each scan, establishing a scan log database for subsequent trend analysis and optimization.

[0098] The dynamic measurement strategy module enables focused monitoring of high-risk areas by dynamically adjusting the scanning strategy. At the same time, it improves the accuracy and reliability of anomaly identification by effectively combining multi-band test signals. Furthermore, it can adaptively adjust the monitoring intensity based on real-time detection results, optimizing resource allocation while ensuring monitoring effectiveness.

[0099] In one embodiment of the present invention, it further includes: a point cloud stability analysis module, used for quality assessment of historical point cloud datasets and optimization of acquisition strategies.

[0100] First, access the historical point cloud data stored in the historical point cloud database. This data is arranged in time series and contains historical point cloud datasets obtained from multiple scan jobs. Each dataset has metadata such as timestamps, environmental parameter records, and device configuration information.

[0101] The box plot cleaning process is performed on historical point cloud datasets. This cleaning method identifies outlier data points based on statistical distribution characteristics: it calculates statistical feature values ​​for each historical point cloud dataset, including the upper and lower quartiles and interquartile range of the coordinate values, sets an outlier judgment threshold, and typically marks data points exceeding a certain multiple of the interquartile range as outliers. The cleaning process uses a sliding window mechanism to process historical point cloud data in segments according to time sequence, ensuring that the temporal characteristics of the point cloud are preserved. The cleaned data is reorganized into a clean point cloud set, while retaining the location information logs of removed points.

[0102] The point cloud density fluctuation factor is calculated based on a clean point cloud set and is used to quantify the consistency of point cloud distribution. The point cloud space of the clean point cloud set is divided into a regular voxel grid, the point number distribution within each voxel is statistically analyzed, the variance and coefficient of variation of all voxel points are calculated, and the point cloud density fluctuation factor is obtained by weighted averaging in combination with the density change trend over time. This factor can reflect the stability and uniformity of spatial coverage during point cloud acquisition.

[0103] The calculation of curvature stability factor requires extracting the geometric features of the clean point cloud set, using a local surface fitting algorithm to calculate the curvature value of each data point, analyzing the time series data of curvature values ​​using the moving window method, calculating the ratio of the standard deviation to the mean of the curvature values, and considering the skewness and kurtosis characteristics of the curvature distribution. By combining these statistics, the curvature stability factor is obtained, which can characterize the stability of the geometric features of the jumper morphology.

[0104] The calculation of amplitude discretization factor focuses on the vibration characteristics of clean point cloud sets. The amplitude spectrum characteristics of the point cloud are extracted through frequency domain analysis, the amplitude values ​​of the main frequency components are calculated, the distribution of amplitude values ​​in the time series is analyzed, and the coefficient of variation and range ratio of amplitude values ​​are calculated. At the same time, the influence of environmental factors on amplitude is considered. After eliminating environmental interference through regression analysis, a pure amplitude discretization factor is obtained, which can reflect the stability of jumper vibration behavior.

[0105] Point cloud density fluctuation factor, curvature stability factor, and amplitude dispersion factor are used to construct the three-dimensional feature space of the point cloud. Data points from historical scanning operations are mapped to this three-dimensional feature space to form a data point distribution. A spatial search algorithm is used to find the optimal stability region, which corresponds to the coordinate range where all three factor values ​​are low. The search process considers the density distribution and clustering characteristics of the data points to ensure that the found optimization region is statistically significant.

[0106] The optimal scanning interval is determined based on the search results in the feature space. The time interval parameters corresponding to the data points in the optimal stability region are analyzed, and the distribution characteristics of these time intervals are statistically analyzed. Cluster analysis is used to identify the most common time interval values. Combined with the physical characteristics of the jumper and environmental conditions, the final optimal scanning interval is determined. This optimal scanning interval balances the data acquisition frequency and resource consumption.

[0107] Periodic point cloud acquisition is triggered based on the optimal scanning interval. A timer scheduler is configured to automatically start scanning jobs at predetermined time intervals. Environmental condition parameters are checked before each scan, and the scanning time is automatically adjusted if the environmental parameters exceed the preset range. Point cloud quality is monitored in real time during scanning, and an exception handling process is immediately triggered if data quality degradation is detected. All scan job records are stored in the system log for subsequent analysis and optimization.

[0108] The point cloud stability analysis module also includes a continuous learning mechanism, which periodically recalculates the 3D feature space, updates the distribution of data points in the 3D feature space, and re-evaluates the optimal scanning interval based on the new data. This adapts to the effects of environmental changes and equipment aging, ensuring that the scanning strategy always remains in optimal condition.

[0109] The implementation of the point cloud stability analysis module emphasizes data integrity and traceability. All intermediate calculation results are stored with timestamps and version information. Anomalies recorded during the cleaning process are used for subsequent analysis of the causes of point cloud instability. The calculation process in the 3D feature space is fully reproducible, and all parameter settings and algorithm selections are recorded in detail in the configuration file. Periodically collected scan data undergoes quality inspection before being stored in the database to ensure data reliability.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A full-area lidar measurement system for the jumper configuration of transmission towers, characterized in that, include: Point cloud acquisition module, latent space transformation module, jumper pattern generation module, multi-dimensional difference analysis module; The point cloud acquisition module obtains the original three-dimensional point cloud data set of the power transmission tower jumper wires through full-area lidar scanning. The latent space transformation module converts the original three-dimensional point cloud data set into a low-dimensional latent space vector set. The jumper shape generation module generates the jumper shape evolution trajectory based on the initial latent space vector corresponding to the initial jumper point cloud and the jumper design parameters. The multi-dimensional difference analysis module calculates spatial position differences, curvature change differences, and amplitude fluctuation differences based on the current hidden space vector corresponding to the current scanned point cloud and the hidden space vector on the jumper morphology evolution trajectory. The multi-dimensional difference analysis module includes a spatial deviation calculation unit, a curvature offset detection unit, and an amplitude similarity evaluation unit. The spatial deviation calculation unit aligns the current latent space vector with the latent space vector on the morphological evolution trajectory and calculates the spatial position deviation. The curvature offset detection unit extracts point cloud curvature features and calculates the curvature change offset index; The amplitude similarity evaluation unit matches the amplitude distribution characteristics of the point cloud and quantifies the similarity. The execution process of the spatial deviation calculation unit includes: determining the index of the target trajectory point that is closest to the current latent space vector on the jumper morphology evolution trajectory, and calculating the spatial position deviation based on the target trajectory point index; The curvature offset detection unit extracts the curvature energy distribution of a preset sensitive frequency band through a multi-scale curvature decomposition algorithm and calculates the ratio of theoretical to measured curvature energy density. The amplitude similarity evaluation unit uses a point set distance algorithm to match the phase error at amplitude abrupt change points.

2. The full-area lidar measurement system for jumper configuration of transmission towers according to claim 1, characterized in that, The latent space transformation module includes a point cloud encoder and a feature decoder; The point cloud encoder decomposes the original 3D point cloud data set into discrete feature vectors; The feature decoder reconstructs discrete feature vectors into a set of latent space vectors; Before running, the latent space transformation module needs to be trained through an adversarial training network. The training process includes: inputting historical jumper point cloud samples to the point cloud encoder to generate discrete feature vectors, reconstructing the latent space vectors through the feature decoder, and having the discriminator evaluate the reconstruction quality and update the network parameters until convergence.

3. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 2, characterized in that, The jumper shape generation module applies noise perturbation to the initial latent space vector, uses the initial jumper point cloud and jumper design parameters as constraints, and generates the jumper shape evolution trajectory through the noise removal process. The jumper pattern generation module needs to optimize network parameters by maximizing the log-likelihood estimate of the pattern evolution trajectory.

4. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 3, characterized in that, It also includes a key point topology association module; the key point topology association module constructs a three-dimensional difference matrix by using spatial position deviation, curvature offset index and amplitude similarity, and inputs it into the jumper space topology network; The jumper space topology network is constructed based on the structural parameters of the transmission tower to establish node connection relationships and to mark the spatial distance constraints between nodes.

5. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 4, characterized in that, The key point topology association module performs anomaly propagation inference: it calculates the morphological deviation attenuation factor based on the spatial distance between nodes, captures cross-node association features through a multi-head attention mechanism, and uses a path simulation algorithm to infer the diffusion path of morphological deviation in the topology network. The frequency of deviations at each node during the simulation process is statistically analyzed, and a probability distribution map of key point anomalies is generated by combining the spatial distance parameters.

6. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 5, characterized in that, It also includes a dynamic measurement strategy module; the dynamic measurement strategy module configures scanning parameters according to the key point anomaly probability distribution map: enabling high-resolution scanning mode for high-probability anomaly areas; and applying multi-band spatial sampling perturbation to adjacent nodes.

7. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 6, characterized in that, The dynamic measurement strategy module performs multi-band spatial sampling perturbation by injecting multi-wavelength test signals through a controllable laser source; The theoretical point cloud response is calculated based on the jumper space topology network; the Euclidean distance between the measured point cloud response and the theoretical point cloud response is compared to determine the abnormal offset.

8. A full-area lidar measurement system for jumper configuration of transmission towers according to claim 7, characterized in that, It also includes a point cloud stability analysis module; the point cloud stability analysis module performs box plot cleaning on the historical point cloud data set and calculates the point cloud density fluctuation factor, curvature stability factor and amplitude dispersion factor. A three-dimensional search is performed in the point cloud feature space using three factors as an index to determine the optimal scanning interval; periodic point cloud acquisition is triggered based on the optimal scanning interval.

Images