Full-automatic electric power safety line patrol method and system based on laser point cloud

Abstract

The invention discloses a laser point cloud-based full-automatic electric power safety line patrol method and system, and relates to the field of electric power patrol. Three-dimensional point cloud data of an electric power transmission and distribution system are acquired; processing the point cloud data to obtain preprocessed point cloud data; inputting the preprocessed point cloud data into an improved RandLA-Net model for point-by-point classification, and outputting semantic segmentation results of a wire, an insulator, an electric tower and vegetation; constructing a complete scene model of the power transmission and distribution system according to the semantically segmented point cloud data; and based on the complete scene model of the power transmission and distribution system, completing power safety line patrol. Through the technical scheme provided by the invention, potential safety hazards can be found in time, and corresponding measures are taken for repair and prevention.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electric power inspection, in particular to a full-automatic electric power safety line inspection method and system based on laser point cloud. BACKGROUND

[0002] At present, optical remote sensing unmanned aerial vehicle line inspection (for example: visible light line inspection, infrared light line inspection) in the electric power scene analyzes the obtained two-dimensional image data, which has limitations such as lack of depth information, limited view angle, and inability to directly model in three dimensions. However, three-dimensional point cloud data can (accurately measure distance, height, volume, etc., can obtain full three-dimensional information in a single scan, without the need for multiple shootings) perfectly solve the problem. Especially in the electric power system, for example, in the long-distance line inspection of overhead transmission lines, laser point cloud data can obtain global information and directly display on the display in front of the electric power workers, and it is a three-dimensional space with high precision, high density. This can more conveniently complete the line inspection work and improve the line inspection efficiency. SUMMARY

[0003] The purpose of the present application is to overcome the shortcomings of the prior art and provide a full-automatic electric power safety line inspection method based on laser point cloud, which comprises the following steps: Step one, obtaining three-dimensional point cloud data of the electric power transmission and distribution system; Step two, removing outliers from the point cloud data through statistical filtering, performing voxel downsampling with a preset resolution, fitting and segmenting the conductor and the tower based on RANSAC, removing environmental noise based on density clustering, and optimizing the surface details of the equipment using morphological filtering to obtain preprocessed point cloud data; Step three, inputting the preprocessed point cloud data into an improved RandLA-Net model for point-by-point classification, and outputting the semantic segmentation results of the conductor, insulator, tower and vegetation; Step four, performing plane segmentation and voxelization on the tower point cloud data in the semantic segmented point cloud data to generate an LOD model; performing RANSAC cylindrical fitting on the conductor point cloud data and calculating the sag through B-spline; performing surface defect detection after loose wave reconstruction of the insulator point cloud; repairing the broken part of the broken conductor point cloud through RANSAC line fitting; removing discrete points with a density lower than a threshold from the vegetation point cloud; smoothing the boundary by using erosion-expansion operation on the residual noise on the tower surface; and constructing a complete scene model of the electric power transmission and distribution system; Step five, based on the complete scene model of the electric power transmission and distribution system, calculating the minimum distance from the conductor to the vegetation, marking the tree barrier risk whose distance exceeds the safety threshold; calculating the conductor sag and comparing it with the design threshold to judge the risk and give an early warning; constructing a tower topology graph to check the connection integrity and completing the electric power safety line inspection.

[0004] Furthermore, the improved RandLA-Net model includes: optimizing the shallow layer sampling rate, adopting multi-scale feature fusion for small targets, and improving the attention mechanism; The shallow sampling rate is optimized: RandLA-Net default layer-by-layer sampling rate , define the sampling attenuation amplitude as Δ, for the l The sampling rate of the layer Adjusted to: Multi-scale feature fusion is used for small targets: the original point features fraw and geometric features including curvature, normal vector, density fgeo Splicing to obtain the enhanced feature vector fenhanced : fenhanced = Concat ( fraw , fgeo ) Improved attention mechanism: assign independent learnable query transformation matrices to conductors, poles, and vegetation categories and bond change matrix , through the transformation matrix and bond change matrix Mapping the point feature vector to a category-specific subspace, Softmax normalized to a probability distribution is: in For the The feature vector of a point, For the The feature vector of a point, is the feature dimension.

[0005] Furthermore, the method of repairing the broken portion of the broken wire point cloud by RANSAC linear fitting includes: The equations of the lines connecting the conductors and the pylons are given in polar coordinate form: Each iteration randomly selects 2 points and calculates the line parameters , defining point The distance to the line is the error: like , is the preset threshold, then the point is an interior point; According to the ratio of internal points w and confidence probability p according to Determine the number of iterations k; For the multiple straight lines obtained by fitting, the endpoint spacing and direction consistency calculations are performed on adjacent line segments, and two line segments with short endpoint spacing and small angle are merged into one straight line.

[0006] Furthermore, removing discrete points with a density lower than a threshold from the vegetation point cloud includes: Assume that any point in the semantic point cloud : in is the indicator function, r is the neighborhood radius; set the density threshold ρth ,when ρ ( pi )< ρth Remove the point.

[0007] Furthermore, the following formula is used to calculate the conductor sag: in L is the distance between the highest points of the two ends of the wire, H is the horizontal tension of the conductor, w is the weight per unit length of the conductor.

[0008] A fully automatic power safety line patrol system based on laser point cloud applies the fully automatic power safety line patrol method based on laser point cloud, including: a data processing module, a data acquisition module, a preprocessing module, a semantic segmentation module, a scene recovery module, and a spatial analysis module; the data acquisition module, preprocessing module, semantic segmentation module, scene recovery module, and spatial analysis module are respectively connected to the data processing module.

[0009] The beneficial effects of this invention are: using drones to replace line patrol personnel to patrol the power transmission and distribution system, obtaining three-dimensional point cloud data, first pre-processing the point cloud data, then using deep learning methods to perform semantic segmentation on the data, then performing scene restoration, and finally analyzing the spatial relationships between various elements in the power scene to monitor the status of power lines and detect anomalies. This provides a basis for power planning and operation and maintenance decision-making, promptly identifying potential safety hazards and taking appropriate measures to repair and prevent them. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 图1 This is a flowchart of a fully automatic power safety line inspection method based on laser point cloud; 图2 This is a schematic diagram of the implementation process of the fully automatic power safety line inspection method based on laser point cloud. DETAILED DESCRIPTION

[0011] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0012] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0013] like 图1 As shown in the figure, the fully automatic power safety line inspection method based on laser point cloud includes the following steps: Step 1: Obtain three-dimensional point cloud data of the power transmission and distribution system; Step 2: Statistical filtering is used to remove outliers from the point cloud data, and voxel downsampling is performed at a preset resolution. RANSAC fitting is used to segment the conductors and towers, density clustering is used to remove environmental noise, and morphological filtering is used to optimize the surface details of the equipment to obtain preprocessed point cloud data. Step 3: Input the preprocessed point cloud data into the improved RandLA-Net model for point-by-point classification, and output the semantic segmentation results of conductors, insulators, poles and vegetation; Step 4: Plane segmentation and voxelization are performed on the power tower point cloud data in the semantically segmented point cloud data to generate an LOD model. RANSAC cylindrical fitting is used on the conductor point cloud data, followed by B-spline striping to calculate sag. Surface defect detection is performed on the insulator point cloud after loose wave reconstruction. The broken conductor point cloud is repaired using RANSAC linear fitting. Discrete points with a density below the threshold are removed from the vegetation point cloud. The residual noise on the tower surface is smoothed using an erosion-dilation operation. Finally, a complete scenario model of the power transmission and distribution system is constructed. Step 5: Based on the complete scenario model of the power transmission and distribution system, the minimum distance from the conductor to the vegetation is calculated, and the tree obstacle risk exceeding the safety threshold is marked; the conductor sag is calculated and compared with the design threshold to determine the risk and issue an early warning; a tower topology map is constructed to verify the connection integrity and complete the power line safety inspection.

[0014] The improved RandLA-Net model includes: optimizing the shallow layer sampling rate, adopting multi-scale feature fusion for small targets, and improving the attention mechanism; The shallow sampling rate is optimized: RandLA-Net default layer-by-layer sampling rate , define the sampling attenuation amplitude as Δ, for the l The sampling rate of the layer Adjusted to: Multi-scale feature fusion is used for small targets: the original point features fraw and geometric features including curvature, normal vector, density fgeo Splicing to obtain the enhanced feature vector fenhanced : fenhanced = Concat ( fraw , fgeo ) Improved attention mechanism: assign independent learnable query transformation matrices to conductors, poles, and vegetation categories and bond change matrix , through the transformation matrix and bond change matrix Mapping the point feature vector to a category-specific subspace, Softmax normalized to a probability distribution is: in For the The feature vector of a point, For the The feature vector of a point, is the feature dimension.

[0015] The method of repairing the broken part of the broken wire point cloud by RANSAC linear fitting includes: The equations of the lines connecting the conductors and the pylons are given in polar coordinate form: Each iteration randomly selects 2 points and calculates the line parameters , defining point The distance to the line is the error: like , is the preset threshold, then the point is an interior point; According to the ratio of internal points w and confidence probability p according to Determine the number of iterations k ; For the multiple straight lines obtained by fitting, the endpoint spacing and direction consistency calculations are performed on adjacent line segments, and two line segments with short endpoint spacing and small angle are merged into one straight line.

[0016] The step of removing discrete points with a density lower than a threshold from the vegetation point cloud includes: Assume that any point in the semantic point cloud : in is the indicator function, r is the neighborhood radius; set the density threshold ρth ,when ρ ( pi )< ρth Remove the point.

[0017] The calculation of conductor sag adopts the following formula: in L is the distance between the highest points of the two ends of the wire, H is the horizontal tension of the conductor, w is the weight per unit length of the conductor.

[0018] A fully automatic power safety line patrol system based on laser point cloud applies the fully automatic power safety line patrol method based on laser point cloud, including: a data processing module, a data acquisition module, a preprocessing module, a semantic segmentation module, a scene recovery module, and a spatial analysis module; the data acquisition module, preprocessing module, semantic segmentation module, scene recovery module, and spatial analysis module are respectively connected to the data processing module.

[0019] Specifically, a fully automated transmission line inspection method based on the semantic spatial construction and analysis of laser point clouds has been invented. This method is primarily used for high-precision, comprehensive monitoring and analysis of power equipment, improving the safety and reliability of the power system. Power workers can directly view equipment status and information on a display, addressing the limitations of traditional manual line inspection, which requires a large amount of labor and is difficult to inspect in complex terrain. The present invention involves point cloud registration technology, point cloud semantic segmentation technology, and transmission line monitoring technology.

[0020] Applications for this system include transmission line inspection (generating 3D models of power corridors using laser point clouds to detect conductor sag, insulator damage, and conductor tension) and substation inspection (generating 3D models of substations using laser point clouds to analyze equipment spacing and structural deformation). First, a drone equipped with a lidar acquires 3D point cloud data of the transmission line in a single flight. Statistical filtering is then applied to the data to remove significant outliers. Downsampling is performed at a resolution of 0.1 meters using the center voxel. RANSAC is then used to segment the conductors and towers. Density-based clustering is applied to the remaining point cloud data to remove ambient noise. Finally, morphological filtering is applied to the equipment surfaces to optimize detail. Next, the processed data is semantically segmented using an improved RandLA-Net model, predicting the type point by point. Conductors, insulators, towers, and vegetation are visualized using different colors. Scene reconstruction then performs planar segmentation and voxelization on the segmented tower point cloud data to generate an LOD (level of detail) model. RANSAC cylindrical fitting and B-spline striping are performed on the conductor point cloud data to calculate sag. Loose wave reconstruction is performed on the insulator point cloud for surface defect detection. After semantic segmentation and scene restoration, the original point cloud data is converted into quantifiable, actionable, and decision-making data to facilitate subsequent analysis. Finally, the spatial relationships of the laser point cloud scene are constructed and analyzed. The minimum distance from the conductor to the vegetation is calculated point by point. Tree obstacles exceeding the safety threshold are marked. Conductor sag is calculated and compared to the design threshold to determine risk and issue warnings. A tower topology map is constructed to verify connection integrity.

[0021] like 图2 As shown, this invention investigates fully automated line inspection using laser point clouds. To achieve a system that can be applied accurately, precise semantic segmentation, efficient geometric calculation algorithms, and a business-specific rule base are required. By optimizing random sampling, improving local feature aggregation, and refining the attention mechanism within RandLA-Net, the model performs multi-scale feature fusion to enhance its ability to capture small objects such as slender conductors. Edge point weights are added to the loss function to improve segmentation accuracy at the conductor-insulator junction. This results in accurate semantic segmentation results. For regular shapes like conductors and towers, model parameters (such as conductor radius) are preset to reduce iterations. The FLANN library is used to accelerate KD tree queries and improve distance calculation efficiency. Rules for tree barrier distance (≥5 meters) and conductor sag error (±10 cm) are set according to the power industry standard "Operational Regulations for Overhead Transmission Lines," defining pole-to-conductor connection rules. Safety standards for different voltage levels are differentiated and differentiated rules are configured. By collaboratively advancing data accuracy, algorithm efficiency, and business-specific rules, a complete closed-loop laser point cloud processing system can be achieved, enabling precise application across the entire system.

[0022] Based on research progress both domestically and internationally, the accuracy of the system needs to be improved due to factors such as scene texture, number of objects, and structural complexity, which have a certain impact on the acquisition of 3D point cloud data. This invention focuses on solving the problems of semantic segmentation, scene restoration, and spatial construction and analysis.

[0023] Semantic segmentation, due to the variability and diversity of power transmission scenarios, presents a challenge in understanding the scene's content while providing the required feature information to ensure the stability and accuracy of the entire system. Semantic segmentation methods can classify a point cloud within the scene, segmenting the target point cloud into its corresponding category and outputting the semantic segmentation result.

[0024] Adjust the model structure and parameters of the pre-processed point cloud data: Because the wires in the power scene are sparse, RandLA-Net defaults to layer-by-layer sampling rate =[4,4,4,4,2], the key points of the wire will be lost prematurely, so we choose to reduce the shallow sampling rate, and the mathematical expression is (1). is the sampling attenuation amplitude.

[0025] (1) If you use =2 as an example, then =[2,2,4,4,2]( =[4-2,4-2,4,4,2]). In this way, the number of samples in the first and second layers is increased from 25% to 50%, while the sampling rates of the third, fourth, and fifth layers remain unchanged.

[0026] For small objects such as conductors and insulators, 10 points are retained to achieve sampling optimization for power scenarios. At least k=10 key points are retained. First, 50% of the points are retained by reducing shallow sampling. If the number of sampling points is less than 10, we will sort them by density features and then resample them to 10 points. Let the point cloud set be P and the key object area set be: , then for each region , sampling must meet the following requirements: (2) Indicates a point Whether to be retained (1 for retention, 0 for discard).

[0027] By increasing the search radius for thin wires, we can capture continuity over longer distances. The current search radius is , then the current domain search radius is within the original search radius Increase times the length of the wire .

[0028] (3) When local features are aggregated, use Display to add geometric features to help distinguish between traverses and vegetation. and geometric features including curvature, normal vector, density Splicing is performed to obtain the enhanced feature vector .

[0029] (4) For different categories such as wires and towers, use independent attention weights. For each category (wires, towers, vegetation, etc.) Assigning independent learnable query transformation matrices and bond change matrix .pass and Will (No. eigenvectors of points), (No. The feature vector of each point is mapped to a category-specific subspace. The dot product is used to measure and In category The similarity is calculated by the correlation strength under To prevent gradient explosion, Softmax is normalized to probability distribution: (5) In LocalSpatialEncoding (LSE), conductor axial direction encoding is added to enhance the modeling of conductor orientation. This improvement in local features and attention mechanisms can be better applied in power scenarios.

[0030] For scene restoration, we first separate the labeled semantic point cloud data by category. We then perform RANSAC line fitting on wires and pylons to repair broken parts.

[0031] RANSAC line fitting process: The linear equations of the wires and towers are in polar coordinate form to avoid the problem of infinite slope of vertical lines: (6) Each iteration randomly selects 2 points and calculates the line parameters .Definition point The distance to the line is the error: like (preset threshold), then the point is an interior point. By the interior point ratio and confidence probability Decide: ,in is the estimated proportion of inliers.

[0032] Break repair process: Run RANSAC multiple times on the semantic point cloud, remove the inliers of the fitted segments until no significant segments remain, and obtain multiple straight lines. Then, calculate the endpoint spacing and direction consistency of adjacent segments. Merge two segments with a short endpoint spacing and a small angle into a single line.

[0033] For the segmented vegetation noise, remove the discrete points with density lower than the threshold. Suppose any point in the semantic point cloud , its local density is the radius The number of points in the neighborhood, where is an indicator function, which takes 1 when the distance condition is met and 0 otherwise.

[0034] (7) Setting density thresholds , directly remove low-density points, when Removed when.

[0035] Finally, the erosion-dilation operation is used to smooth the edges of the tower's surface to remove any residual noise. This results in a complete scene model that conforms to the power topology scenario.

[0036] Semantic space construction and analysis: After obtaining a complete model of the power topology scene, the point cloud is aligned with the designed CAD model through ICP. When calculating the safe distance between the wire and the vegetation, the distance between the wire and the vegetation is calculated point by point. The KD tree can accelerate the query and visually mark the risk points. Suppose a point on the vegetation is and , calculate the distance according to formula (5).

[0037] (8) When measuring the vertical arc, the span is determined by measuring the distance between the highest points at both ends of the wire. , measure the horizontal tension of the conductor , is the weight per unit length of the conductor. The sag is measured according to formula (6).

[0038] (9) Compare the measured sag with the designed sag and issue an early warning when it exceeds 10%.

Claims

1. A fully automatic power safety line inspection method based on laser point cloud, characterized by: The steps include: Step 1: Obtain three-dimensional point cloud data of the power transmission and distribution system; Step 2: Statistical filtering is used to remove outliers from the point cloud data, and voxel downsampling is performed at a preset resolution. RANSAC fitting is used to segment the conductors and towers, density clustering is used to remove environmental noise, and morphological filtering is used to optimize the surface details of the equipment to obtain preprocessed point cloud data. Step 3: Input the preprocessed point cloud data into the improved RandLA-Net model for point-by-point classification, and output the semantic segmentation results of conductors, insulators, poles and vegetation; Step 4: Plane segmentation and voxelization are performed on the power tower point cloud data in the semantically segmented point cloud data to generate an LOD model. RANSAC cylindrical fitting is used on the conductor point cloud data, followed by B-spline striping to calculate sag. Surface defect detection is performed on the insulator point cloud after loose wave reconstruction. The broken conductor point cloud is repaired using RANSAC linear fitting. Discrete points with a density below a threshold are removed from the vegetation point cloud. The residual noise on the tower surface is smoothed using an erosion-dilation operation. Finally, a complete scenario model of the power transmission and distribution system is constructed. Step 5: Based on the complete scenario model of the power transmission and distribution system, the minimum distance from the conductor to the vegetation is calculated, and tree obstacle risks exceeding the safety threshold are marked; Calculate conductor sag and compare it with the design threshold to determine risks and issue warnings; Construct tower topology maps to verify connection integrity and complete power line safety inspections.

2. The fully automatic power line safety inspection method based on laser point cloud according to claim 1 is characterized in that: The improved RandLA-Net model includes: optimizing the shallow layer sampling rate, adopting multi-scale feature fusion for small targets, and improving the attention mechanism; The shallow sampling rate is optimized: RandLA-Net default layer-by-layer sampling rate , define the sampling attenuation amplitude as Δ, for the l The sampling rate of the layer Adjusted to: Multi-scale feature fusion is used for small targets: the original point features fraw and geometric features including curvature, normal vector, density fgeo Splicing to obtain the enhanced feature vector fenhanced : fenhanced = Concat ( fraw , fgeo ) Improved attention mechanism: assign independent learnable query transformation matrices to conductors, poles, and vegetation categories and bond change matrix , through the transformation matrix and bond change matrix Mapping the point feature vector to a category-specific subspace, Softmax normalized to a probability distribution is: in For the The feature vector of a point, For the The feature vector of a point, is the feature dimension.

3. The fully automatic power line inspection method based on laser point cloud according to claim 1 is characterized in that: The method of repairing the broken part of the broken wire point cloud by RANSAC linear fitting includes: The equations of the lines connecting the conductors and the pylons are given in polar coordinate form: Each iteration randomly selects 2 points and calculates the line parameters , defining point The distance to the line is the error: like , is the preset threshold, then the point is an interior point; According to the ratio of internal points w and confidence probability p according to Determine the number of iterations k ; For the multiple straight lines obtained by fitting, the endpoint spacing and direction consistency calculations are performed on adjacent line segments, and two line segments with short endpoint spacing and small angle are merged into one straight line.

4. The fully automatic power line inspection method based on laser point cloud according to claim 1 is characterized in that: The step of removing discrete points with a density lower than a threshold from the vegetation point cloud includes: Assume that any point in the semantic point cloud : in is the indicator function, r is the neighborhood radius; set the density threshold ρth ,when ρ ( pi )< ρth Remove the point.

5. The fully automatic power line inspection method based on laser point cloud according to claim 1 is characterized in that: The calculation of conductor sag adopts the following formula: in L is the distance between the highest points of the two ends of the wire, H is the horizontal tension of the conductor, w is the weight per unit length of the conductor.

6. The fully automatic power safety line inspection system based on laser point cloud is characterized by: The fully automatic power safety line inspection method based on laser point cloud described in any one of claims 1 to 5 includes: a data processing module, a data acquisition module, a preprocessing module, a semantic segmentation module, a scene recovery module, and a spatial analysis module; the data acquisition module, preprocessing module, semantic segmentation module, scene recovery module, and spatial analysis module are respectively connected to the data processing module.

Images