Power distribution network hidden point fast positioning method and system based on unmanned aerial vehicle cooperation

Abstract

The present application relates to the technical field of power distribution network fault diagnosis, in particular to a power distribution network hidden danger point rapid positioning method and system based on unmanned aerial vehicle cooperation, comprising: obtaining initial spatial position information of a target power distribution network line, and generating a cruise path of a first group of unmanned aerial vehicles based on the initial spatial position information; controlling the first group of unmanned aerial vehicles to fly along the cruise path and collect physical field characteristic information in real time; constructing a physical field distribution gradient map based on the physical field characteristic information collected by each unmanned aerial vehicle in the first group and its corresponding spatial coordinates; identifying an abnormal gradient area and determining the spatial coordinates of a suspected hidden danger point; collecting images of the suspected hidden danger point based on the spatial coordinates of the suspected hidden danger point; comparing the visual image set with the suspected hidden danger point, and if there is a difference in the comparison result, the suspected hidden danger point is determined as the final hidden danger point. By collecting air micro-flow disturbance data and spatial electrostatic field distortion data, weak physical field changes generated in the early stage of the fault can be captured.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault diagnosis technology, specifically to a method and system for rapid location of potential hazards in power distribution networks based on UAV collaboration. Background Technology

[0002] As the end-point of the power system directly facing users, the distribution network has a wide distribution and a complex environment, making it prone to early, minor faults due to insulator contamination, broken cable strands, damaged conductors, and foreign objects adhering to the wires. If these faults are not detected and addressed in a timely manner, they may gradually develop into serious accidents such as short circuits and line breaks, affecting the reliability of power supply.

[0003] Currently, the inspection of power distribution lines mainly relies on manual inspection, helicopter inspection, and drone inspection equipped with visible light or infrared thermal imaging equipment. Drone visual inspection is a relatively common technical means. It identifies obvious physical deformation or foreign object attachment by taking pictures of the line and using image processing technology. For early and weak faults, they have not yet produced obvious visual characteristics and are difficult to detect by conventional visual means. Visual detection is easily affected by environmental factors such as lighting, weather, and obstruction, resulting in missed detection and false detection. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a method and system for rapid location of potential hazards in power distribution networks based on UAV collaboration.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for rapid location of hidden danger points in power distribution networks based on UAV collaboration, comprising the following steps: S1. Obtain the initial spatial location information of the target power distribution network line, and generate the cruise path of the first group of UAVs based on the initial spatial location information; S2. Control the first group of drones to fly along the cruise path, and during the flight, collect physical field characteristic information in a preset space range around the power distribution line in real time through the sensor array mounted on each drone. The physical field characteristic information includes air micro-flow disturbance data and spatial electrostatic field distortion data. S3. Based on the physical feature information and corresponding spatial coordinates collected by each UAV in the first group, construct a physical field distribution gradient map of the area covered by the inspection path; S4. Identify abnormal gradient regions in the physical field distribution gradient map, and determine the spatial coordinates of suspected hidden danger points based on the center point of the abnormal gradient region. S5. Based on the spatial coordinates of the suspected potential hazard points, dispatch a second group of drones equipped with visual acquisition sensors to fly to the target area and acquire multi-angle, multi-spectral images of the suspected potential hazard points and their surroundings to obtain a set of visual images. S6. Compare the visual image set with the pre-stored standard state images of suspected hazard points, identify physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, determine the suspected hazard point as the final hazard point and generate a hazard report containing location and image information.

[0006] In a preferred embodiment, in step S1, the tower coordinates, conductor sag point coordinates, and line turning point coordinates of the target distribution network line are read from the ground control station and the coordinate data are combined into a discrete point sequence describing the spatial direction of the entire line. The ground control station sorts the discrete point sequence according to the actual connection order of the line, forming a continuous spatial polyline, and uses the spatial polyline as the initial spatial location information of the target distribution network line; Based on the initial spatial location information of the conductor, the ground control station extends a preset first horizontal safety distance to both sides along the direction of the line extension, with the conductor as the center line. At the same time, it extends a preset first vertical safety distance upward from the highest point of the conductor and a preset second vertical safety distance downward from the lowest point of the conductor, forming a three-dimensional rectangular columnar corridor. The ground control station extracts the coordinates of the two endpoints of each traverse segment from the initial spatial position information and calculates the spatial straight line equation of each traverse segment; Along the direction of the conductor, at the third vertical offset distance directly above the conductor, generate an aerial trajectory line that is completely parallel to and equidistant from the spatial direction of the conductor. The third vertical offset distance is less than the first vertical safety distance and is a fixed value. At the turning point of two adjacent conductor segments, a spatial arc transition method is used to smoothly connect the aerial trajectory line to form a continuous and smooth three-dimensional spatial curve. The three-dimensional spatial curve is used as the reference cruise line of the first group of UAVs. The ground control station obtains the formation position number of each drone in the first group of drones. The formation position includes the following position along the route and the staggered position to the left and right perpendicular to the route. Using each point on the baseline cruise line as a formation reference point, and based on the formation position number of each UAV, the offset vector of each UAV relative to the formation reference point is calculated in a plane perpendicular to the baseline cruise line according to the preset first lateral spacing and first longitudinal spacing. The offset vector is superimposed onto the spatial coordinates of the formation reference point to obtain the expected spatial coordinates of each UAV at each formation reference point. Connect all desired spatial coordinates in chronological order to form the cruise path of each drone, and combine the cruise paths of all drones to form the formation cruise path of the first group of drones. The ground control station uploads the cruise path of each UAV in the form of waypoint sequence to the onboard computer of the corresponding UAV via a wireless data link; Each waypoint contains three-dimensional spatial coordinates, the expected arrival time, and the allowable position error range. After receiving the waypoint sequence, the UAV's flight controller executes autonomous flight tasks according to the waypoint order and corrects its own position in real time during the flight to ensure that the deviation between the actual flight trajectory and the preset cruise path is always controlled within the allowable position error range.

[0007] In a preferred embodiment, in S2, after the ground control station completes the loading of the formation cruise path, take-off and formation flight commands are simultaneously sent to the first group of UAVs via a wireless data link. Each drone autonomously flies to the starting waypoint according to its pre-stored waypoint sequence and adjusts its attitude so that the nose is tangential to the cruise path. Once all UAVs have reached the starting waypoint and formed the preset formation, the ground control station issues a cruise start command. Each UAV maintains formation flight along its own cruise path, while the onboard computer sends a start data acquisition command to the onboard sensor array. The sensor array includes a biomimetic microfluidic sensing array and an electrostatic field disturbance sensing probe. Both are started synchronously according to a preset first sampling frequency to begin real-time sensing of physical field changes within a preset spatial range around the drone. The biomimetic microfluidic sensing array consists of multiple miniature airflow sensing units arranged in a preset geometric layout. Each sensing unit converts the flow of air molecules across its surface into an electrical signal output. When there is partial discharge or overheating in the power distribution network line, the discharge or high temperature causes the surrounding air molecules to ionize and generate directional flow, forming air micro-flow disturbance. During the flight of the drone, each sensitive unit in the biomimetic microfluidic sensing array independently outputs a voltage value that is proportional to the airflow velocity; The onboard computer reads the output voltage values ​​of all sensitive units in the biomimetic microfluidic sensing array at the same time, forming a multi-dimensional raw data vector; The computer uses the average value of the output voltage of all sensitive units as a reference value, and subtracts the reference value from the output voltage of each sensitive unit to obtain the differential output value of each sensitive unit. The computer sorts the differential output values ​​according to their coordinate positions based on the geometric coordinates of each sensitive unit in the array, forming a spatial difference matrix; The computer finds the two sensitive cells with the largest difference value in the spatial difference matrix, calculates the direction of the line connecting the two sensitive cells, and uses the direction of the line as the overall direction of the airflow. The maximum difference value is multiplied by a preset first scaling factor to obtain the instantaneous velocity value of the airflow. The computer statistically analyzes the changes in airflow direction and velocity over time within a preset first time window. The changes are then divided by a preset second scaling factor to obtain the turbulence information of the airflow. The electrostatic field disturbance sensing probe consists of one or more metal electrodes with the electrode surface exposed to the air. When a drone flies over an area with insulator contamination, broken cable strands, or damaged wires, the potential hazard point generates local electric field distortion, causing a change in the electrostatic field strength in the surrounding space. The electrostatic field disturbance sensing probe on the UAV senses the spatial potential through electrodes, converts the sensed potential value into an analog voltage signal output, and the onboard computer continuously collects the analog voltage signal at a preset second sampling frequency to obtain a potential sequence that changes over time. The airborne computer acquires the potential change sequence within a preset second time window, takes the difference between the maximum and minimum values ​​in the sequence as the potential peak value, and divides the potential peak value by a preset third scaling coefficient to obtain the spatial electric field intensity value within the time window. The computer plots each sampled value in the potential change sequence into a waveform in chronological order, forming a potential change waveform. The computer performs baseline drift removal on the waveform, takes the average value of the entire waveform as the DC component, and subtracts the DC component from each sampled value to obtain the AC fluctuation component. The computer extracts the number of zero-crossing points, the number of peaks, and the steepness of the rising and falling edges of the AC wave components, and combines these features into distorted waveform information.

[0008] In a preferred embodiment, in S3, the ground control station receives data points from all UAVs in the first group, wherein each data point includes three-dimensional spatial coordinates, air microfluidic disturbance data, and spatial electrostatic field distortion data. The air microfluidic disturbance data includes airflow direction, airflow speed, and turbulence information, and the spatial electrostatic field distortion data includes spatial electric field strength and distortion waveform information. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding air microfluidic disturbance data from the data points. The ground control station divides the entire cruise path coverage area into a uniform three-dimensional grid according to a preset first spatial step size. Each grid cell is a cube voxel. For each grid cell, the ground control station collects the air microfluidic disturbance data of all data points falling into the grid cell. The arithmetic mean of the collected airflow velocity values ​​is taken as the representative airflow velocity of the grid cell. The representative airflow direction value is obtained by vector averaging the collected airflow direction value. The arithmetic mean of the collected turbulence values ​​is taken as the representative turbulence of the grid cell. For blank grid cells where no data points fall, the ground control station uses a spatial interpolation method to search outward from the blank grid cell with the pre-set first search radius for the already assigned grid cells. The representative airflow velocity, representative airflow direction, and representative turbulence of these already assigned grid cells are estimated by weighting inversely proportionally to the distance and then filled into the blank grid cell. After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional hybrid field map, which is used as the first microfluidic field map. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding spatial electrostatic field distortion data from the data points. The ground control station uses the same three-dimensional grid division method as the one used to establish the first microflow field map to divide the entire cruise path coverage area into three-dimensional grids of the same size. For each grid cell, the ground control station collects the spatial electric field intensity values ​​of all data points falling within the grid cell, and takes the arithmetic mean as the representative electric field intensity of the grid cell. At the same time, it collects the distorted waveform information of all data points falling within the grid cell, and takes the arithmetic mean of the number of zero crossings, the number of peaks, and the steepness of the waveform in the distorted waveform information to obtain a set of representative distortion characteristic values. For blank grid cells, the ground control station uses the same spatial interpolation method as the first microflow field map to search for neighboring already assigned grid cells with a preset second search radius, and calculates the representative electric field intensity and representative distortion characteristic value of the blank grid cell by distance inverse weighted interpolation. After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional scalar field map, which is used as the first electric field map. The ground control station spatially registers the first microfluidic field map and the first electric field map: the ground control station checks whether the grid origin, grid step size and grid range of the two field maps are completely consistent. If there is any inconsistency, the field map with the larger coverage area is used as the reference, and the field map with the smaller coverage area is extended by boundary. The grid cells of the extended part are filled by copying the values ​​of the neighboring grid cells outward to ensure that the two field maps have the same grid topology. The ground control station creates a new 3D grid with the same grid topology as the registered field map, serving as the framework for the integrated physical field distribution gradient map; For each grid cell in the new three-dimensional grid, the ground control station reads the airflow velocity, airflow direction and turbulence intensity of the grid cell from the first microflow field map, and reads the spatial electric field intensity and distortion characteristic value of the grid cell from the first electric field map. The data are combined into a multi-dimensional feature vector and stored in the grid cell. The ground control station calculates the comprehensive gradient value of each grid cell: first, it calculates the sum of the absolute values ​​of the differences in airflow velocity between the grid cell and its multiple neighboring grid cells to obtain the first gradient component; The second gradient component is obtained by summing the absolute values ​​of the differences in spatial electric field intensity between grid cells and their adjacent grid cells. Finally, the first gradient component and the second gradient component are weighted and summed according to the preset first weight and second weight to obtain the comprehensive gradient value of the grid cell. The combined gradient values ​​of all grid cells are arranged according to spatial coordinates to form a three-dimensional gradient value distribution map, which is then used as a multi-dimensional combined physical field distribution gradient map.

[0009] In a preferred embodiment, in step S4, the ground control station reads a multi-dimensional integrated physical field distribution gradient map. The gradient map consists of multiple grid cells arranged in a three-dimensional grid, and each grid cell stores a comprehensive gradient value. The ground control station traverses all grid cells and uses the comprehensive gradient value stored in each grid cell as the field strength gradient magnitude value of the grid cell. The ground control station determines the field strength gradient direction based on the trend of airflow velocity change between each grid cell and its neighboring grid cells: find the neighboring grid cell with the largest comprehensive gradient value in the six directions around each grid cell, calculate the direction vector from the current grid cell to the neighboring cell, normalize the direction vector and use it as the field strength gradient direction of the current grid cell. If the comprehensive gradient values ​​of multiple neighboring cells are equal and all are the maximum, then the composite direction of multiple direction vectors is used as the field strength gradient direction. The ground control station obtains the pre-stored field strength gradient magnitude threshold, traverses all grid cells again, marks grid cells with field strength gradient magnitude greater than or equal to the preset threshold as candidate abnormal grid cells, and marks grid cells with field strength gradient magnitude less than the preset threshold as normal grid cells. The ground control station uses a connected component analysis algorithm to merge all adjacent candidate abnormal grid cells into a connected component. For each connected region, the ground control station checks whether the total number of grid cells contained in the connected region is greater than the preset minimum region area threshold. If it is greater, the connected region is marked as a preliminary abnormal region and the spatial coordinate list of all grid cells contained in the preliminary abnormal region is recorded. If it is less than, the connected region is excluded. For each initially selected anomaly region, the ground control station obtains the field intensity gradient direction of all grid cells in the region, calculates the average direction of all field intensity gradient directions in the region, and uses the average direction as the region's reference direction. For each grid cell within the region, the ground control station calculates the angle between the field strength gradient direction and the regional reference direction; If the included angle of grid cells in a region that exceed the preset consistency ratio threshold is less than the preset direction deviation threshold, then the initially selected abnormal region is marked as a candidate abnormal gradient region. The ground control station further judges the convergence characteristics of the candidate abnormal gradient region: selects all grid cells located at the edge of the region within the candidate abnormal gradient region, calculates whether the gradient direction of each edge grid cell points into the region, and if the edge grid cells in the region that exceed the preset convergence ratio threshold meet the judgment condition of pointing inward, then the candidate abnormal gradient region is determined as an abnormal gradient region. For each connected region identified as an anomalous gradient region, the ground control station extracts the coordinates of the center point of all grid cells on the region boundary as the coordinates of the edge point. The boundary grid cell is a grid cell in the anomalous gradient region that has at least one adjacent grid cell that does not belong to the anomalous gradient region. The ground control station collects the coordinates of the center points of all boundary grid cells to form a set of edge point coordinates. The ground control station uses the rotating caliper method or the exhaustive search method to find the smallest volume geometry in three-dimensional space that can include all edge point coordinates. The smallest volume geometry is pre-selected as a cuboid, sphere or ellipsoid. The ground control station calculates the coordinates of the geometric center point based on the type of the smallest circumscribed geometric figure: if the smallest circumscribed geometric figure is a cuboid, then the average of the minimum and maximum values ​​of the cuboid in the three coordinate axes is taken, and the three-dimensional coordinates formed by the three average values ​​are used as the coordinates of the geometric center point of the cuboid. If the smallest circumscribed geometric figure is a sphere, then the coordinates of the center of the sphere are used as the coordinates of the geometric center point; if the smallest circumscribed geometric figure is an ellipsoid, then the center points corresponding to the lengths of the three semi-axis of the ellipsoid are calculated as the coordinates of the geometric center point, and the ground control station determines the coordinates of the geometric center point as the spatial coordinates of the first suspected hidden danger point. The ground control station records the spatial coordinates of the first suspected hidden danger point in memory in the form of three-dimensional coordinates, and stores them in association with the identifier, size and average gradient magnitude of the abnormal gradient region corresponding to the coordinates. If multiple abnormal gradient regions are identified within the coverage area of ​​the same cruise path, the ground control station will sequentially perform the above steps for each abnormal gradient region to generate multiple spatial coordinates of first suspected hidden danger points.

[0010] In a preferred embodiment, in step S5, after determining the spatial coordinates of the first suspected potential hazard point, the ground control station selects a second group of drones equipped with visual acquisition sensors from the drone library; the ground control station sends take-off commands to each drone in the second group and uploads the spatial coordinates of the first suspected potential hazard point as the target waypoint; The second group of drones took off to a safe altitude according to their own takeoff procedure, and then flew to a hovering point near the spatial coordinates of the first suspected hidden danger point in a fast flight mode with a speed higher than that of the first group of drones. The hovering point is located at the first hovering distance directly in front of the spatial coordinates of the first suspected hidden danger point, and the height of the hovering point is consistent with the height of the spatial coordinates of the first suspected hidden danger point; Once all the drones in the second group have reached their respective hovering points, each drone sends a ready signal to the ground control station. After receiving the readiness signals from all UAVs, the ground control station sends a low-speed data acquisition command to the second group of UAVs, limiting the maximum flight speed of each UAV to one-third of the cruising speed of the first group of UAVs. The ground control station uploads the speed limit parameters to the flight controller of each UAV via a wireless link. The flight controller then automatically controls the actual flight speed within the speed limit value during subsequent flights. The ground control station generates a detailed data collection path for each UAV based on the pre-selected trajectory type. The trajectory type is either a spiral flight trajectory or a figure-eight flight trajectory. If it is a spiral flight trajectory, the ground control station takes the spatial coordinates of the first suspected hidden danger point as the center, uses the starting radius as the horizontal distance between the spiral starting point and the center point, uses the radius decay step as the amount of radius reduction for each flight circle, and uses the vertical ascent step as the amount of altitude increase for each flight circle. Starting from the starting point, a series of continuous spatial points are generated in a counterclockwise or clockwise direction. The spatial points are connected in sequence to form a three-dimensional spiral line from the outside to the inside, from the bottom to the top, or from the top to the bottom. If the flight path is a figure-eight shape, the ground control station uses the spatial coordinates of the first suspected hidden danger point as the center, the first horizontal span as the width of a single loop, and the second horizontal span as the interval between two loops to generate a closed curve in the horizontal plane with two tangent circular loops alternating. The curve passes through the offset point directly above or below the center point. The ground control station discretizes the generated trajectory into a series of waypoints. Each waypoint contains three-dimensional spatial coordinates and the desired nose direction, which always points to the spatial coordinates of the first suspected potential hazard point. The ground control station uploads the waypoint sequence to the flight controllers of the second group of UAVs. The second group of drones departed from the hovering point and flew in sequence according to the received waypoint sequence, while keeping the nose of the drone always pointing to the spatial coordinates of the first suspected potential hazard point; During flight, the visual acquisition sensors on each drone continuously capture images according to the image acquisition frequency; The visual acquisition sensor includes a high-resolution visible light camera and a multispectral camera. The multispectral camera can simultaneously acquire images in the infrared, near-infrared, or ultraviolet bands. While acquiring images, each drone binds and stores the shooting time corresponding to each image, the drone's own three-dimensional spatial coordinates, shooting attitude angle, and camera parameters as metadata and image data. Shooting attitude angles include pitch angle, roll angle, and yaw angle; camera parameters include focal length, aperture, and exposure time. After completing the entire flight path, each drone in the second group automatically flew back to the return point and landed. During or after landing, each drone uploaded all the images it had collected and their corresponding metadata to the ground control station via a wireless data link. The ground control station receives image data from multiple drones, classifies and organizes it according to shooting time, drone number and shooting location to form a visual image set. The ground control station temporarily stores the visual image set in local memory and creates an index to associate each image with the corresponding spatial coordinates and shooting parameters. If a drone malfunctions during the data acquisition process or the image quality does not meet the clarity requirements, the ground control station will dispatch a backup drone to re-execute the data acquisition task of the drone until images from all perspectives are successfully acquired.

[0011] In a preferred embodiment, in step S6, the ground control station acquires pre-stored standard status images of the first suspected hidden danger point; the standard status images are multi-angle, multi-spectral images pre-captured by a drone under the healthy condition of the power distribution network lines, and are classified and stored in the database of the ground control station according to spatial coordinates and location type. Based on the spatial coordinates of the first suspected hidden danger point, the ground control station retrieves the standard state image set corresponding to the spatial coordinates from the database. The standard state image set and the visual image set have the same parameter settings in terms of shooting angle, spectral band and imaging resolution. The ground control station preprocesses each image in the visual imagery set, including converting the images from their original format to a uniform pixel matrix, converting the images to grayscale, and performing histogram equalization on the images. The ground control station extracts edge density features, texture statistics features, and local binary pattern features for each image. The edge density features are obtained by calculating the ratio of the number of edge points where pixel values ​​change abruptly to the total number of pixels in the image. Texture statistical features are obtained by statistically analyzing the frequency distribution of pixel values ​​in the image and the difference distribution between adjacent pixels. Local binary pattern features are obtained by comparing the grayscale relationship between each pixel and its surrounding neighboring pixels, encoding it into a binary value, and then statistically analyzing its histogram. The ground control station arranges the feature values ​​extracted from all images under the same spectral band in a fixed order to form a sub-vector. Then, it splices the sub-vectors of different spectral bands together in a preset spectral order to form the first feature vector. The first feature vector is then normalized so that the values ​​of all dimensions fall between zero and one. Perform the same preprocessing and feature extraction operations on each image in the pre-stored standard state image set to obtain the second feature vector, and perform the same normalization process on the second feature vector so that the value of each dimension of the second feature vector also falls between zero and one. The first feature vector and the second feature vector have the same number of dimensions and the feature meaning of each dimension corresponds one-to-one. The ground control station calculates the cosine similarity between the first and second eigenvectors as follows: multiply the value of each dimension in the first eigenvector by the corresponding value in the second eigenvector to obtain the product value of each dimension, add the product values ​​of all dimensions to obtain the sum of the numerators, calculate the sum of the squares of the values ​​of each dimension in the first eigenvector and take the square root to obtain the magnitude of the first vector, calculate the sum of the squares of the values ​​of each dimension in the second eigenvector and take the square root to obtain the magnitude of the second vector, multiply the magnitude of the first vector by the magnitude of the second vector to obtain the product of the denominators, and divide the sum of the numerators by the product of the denominators to obtain the cosine similarity value. The ground control station obtains a pre-stored matching threshold and compares the cosine similarity value with the matching threshold. If the cosine similarity is greater than or equal to the matching threshold, it is determined that the first suspected hazard point does not exhibit physical deformation or foreign object attachment characteristics. The first suspected hazard point is marked as a false alarm and removed from the list to be reviewed. If the cosine similarity is lower than the matching threshold, it is determined that physical deformation or foreign object attachment characteristics exist. The first suspected hazard point is identified as the final hazard point. The ground control station generates a hazard report, which includes the spatial coordinates of the final hazard point, the hazard type, the similarity value, multiple images of the defect selected from the visual image set, and a comparison image of the corresponding viewpoint extracted from the standard state image. The ground control station stores the hazard report in a local database and sends it to the terminal device of the maintenance personnel via a wireless network.

[0012] This invention also provides a rapid location system for potential hazards in power distribution networks based on drone collaboration, comprising: Cruise path generation module: acquires the initial spatial location information of the target power distribution network line, and generates the cruise path of the first group of UAVs based on the initial spatial location information; Feature information acquisition module: controls the first group of drones to fly along the cruise path, and during the flight, collects physical field feature information in real time within a preset space range around the power distribution line through the sensor array mounted on each drone. The physical field feature information includes air micro-flow disturbance data and spatial electrostatic field distortion data. Gradient map construction module: Based on the physical feature information and corresponding spatial coordinates collected by each UAV in the first group, construct the physical field distribution gradient map of the inspection path coverage area; Hazard identification module: Identifies abnormal gradient regions in the physical field distribution gradient map, and determines the spatial coordinates of suspected hazard points based on the center point of the abnormal gradient region; Hazardous point data acquisition module: Based on the spatial coordinates of the suspected hazard points, a second group of UAVs equipped with visual acquisition sensors is dispatched to fly to the target area to acquire multi-angle, multi-spectral images of the suspected hazard points and their surroundings, thereby obtaining a set of visual images; Hazard location module: Compares the set of visual images with the pre-stored standard state images of suspected hazard points, identifies physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, the suspected hazard point is determined as the final hazard point and a hazard report containing location and image information is generated.

[0013] The beneficial effects of this invention are: by collecting air micro-flow disturbance data and spatial electrostatic field distortion data, this invention can keenly capture the weak physical field changes generated in the early stage of faults such as partial discharge, overheating, insulator contamination, and cable strand breakage, and can be perceived before the hidden dangers form obvious visual characteristics, thus realizing early warning of distribution network faults. This invention constructs a physical field gradient map through multi-UAV collaboration, achieving high positioning accuracy. The first group of UAVs flies synchronously along a formation cruise path to acquire densely distributed physical field data. By constructing microfluidic field maps and electric field maps and fusing them with feature layers, a multi-dimensional comprehensive physical field distribution gradient map is generated. Through gradient magnitude filtering, directional consistency analysis, and convergence feature recognition, the center of abnormal gradient regions can be identified, thereby obtaining the spatial coordinates of suspected potential hazard points with high confidence. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0017] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0018] like Figure 1 This embodiment provides a method for rapid location of potential hazards in power distribution networks based on UAV collaboration, including the following steps: S1. Obtain the initial spatial location information of the target power distribution network line, and generate the cruise path of the first group of UAVs based on the initial spatial location information; Furthermore, in S1, the tower coordinates, conductor sag point coordinates, and line turning point coordinates of the target distribution network line are read from the ground control station and the coordinate data are combined into a discrete point sequence describing the spatial direction of the entire line. The ground control station sorts the discrete point sequence according to the actual connection order of the line, forming a continuous spatial polyline, and uses the spatial polyline as the initial spatial location information of the target distribution network line; Based on the initial spatial location information of the conductor, the ground control station extends a preset first horizontal safety distance to both sides along the direction of the line extension, with the conductor as the center line. At the same time, it extends a preset first vertical safety distance upward from the highest point of the conductor and a preset second vertical safety distance downward from the lowest point of the conductor, forming a three-dimensional rectangular columnar corridor. The ground control station extracts the coordinates of the two endpoints of each traverse segment from the initial spatial position information and calculates the spatial straight line equation of each traverse segment; Along the direction of the conductor, at the third vertical offset distance directly above the conductor, generate an aerial trajectory line that is completely parallel to and equidistant from the spatial direction of the conductor. The third vertical offset distance is less than the first vertical safety distance and is a fixed value. At the turning point of two adjacent conductor segments, a spatial arc transition method is used to smoothly connect the aerial trajectory line to form a continuous and smooth three-dimensional spatial curve. The three-dimensional spatial curve is used as the reference cruise line of the first group of UAVs. The ground control station obtains the formation position number of each drone in the first group of drones. The formation position includes the following position along the route and the staggered position to the left and right perpendicular to the route. Using each point on the baseline cruise line as a formation reference point, and based on the formation position number of each UAV, the offset vector of each UAV relative to the formation reference point is calculated in a plane perpendicular to the baseline cruise line according to the preset first lateral spacing and first longitudinal spacing. The offset vector is superimposed onto the spatial coordinates of the formation reference point to obtain the expected spatial coordinates of each UAV at each formation reference point. Connect all desired spatial coordinates in chronological order to form the cruise path of each drone, and combine the cruise paths of all drones to form the formation cruise path of the first group of drones. The ground control station uploads the cruise path of each UAV in the form of waypoint sequence to the onboard computer of the corresponding UAV via a wireless data link; Each waypoint contains three-dimensional spatial coordinates, the expected arrival time, and the allowable position error range. After receiving the waypoint sequence, the UAV's flight controller executes autonomous flight tasks according to the waypoint order and corrects its own position in real time during the flight to ensure that the deviation between the actual flight trajectory and the preset cruise path is always controlled within the allowable position error range.

[0019] The first horizontal safety distance is preset based on the voltage level and electromagnetic field influence range of the power distribution network line, while the first vertical safety distance and the second vertical safety distance are preset based on the electrical safety clearance between the UAV flight altitude and the conductor. The first horizontal safety distance refers to the interval distance extended to the left and right in two horizontal directions perpendicular to the direction of the conductor, with the conductor as the center line. The interval distance is used to ensure that the UAV maintains sufficient electrical safety clearance and collision avoidance space with the live line in the horizontal direction. The first horizontal safety distance is preset and stored in the ground control station according to the voltage level of the target power distribution line and the range of electromagnetic field influence generated by the voltage level. For example, different fixed distance values ​​are corresponding to different voltage levels (such as 10kV and 35kV lines). The first vertical safety distance is the interval distance that extends vertically upwards from the highest point of the conductor (i.e., the highest point of the conductor sag, usually located near the tower suspension point). The interval distance is used to limit the maximum allowable altitude of the drone flight and prevent the drone from touching the conductor due to insufficient height when passing over it. The first vertical safety distance is preset according to the electrical safety clearance required between the drone flight altitude and the conductor. The electrical safety clearance takes into account the discharge risk that the drone body may cause and the insulation requirements of the line voltage level. Its value is obtained by looking up the minimum safety distance of the corresponding voltage level in the power industry safety regulations and adding a preset margin value. The second vertical safety distance is the interval distance extending vertically downwards from the lowest point of the conductor (i.e., the lowest point of the conductor's sag between the two towers). This interval distance is used to limit the minimum permissible altitude for drone flight, ensuring that the drone will not snag on the conductor or be interfered with by ground objects below when passing under the conductor due to excessive altitude. The second vertical safety distance is preset based on the electrical safety clearance between the drone and the live line, combined with the minimum permissible value of the conductor's distance from the ground. In actual operation, the ground control station will call the maximum conductor sag value given in the design parameters of the target line and set the second vertical safety distance to a value greater than the sum of the sag value and the safety clearance, thereby providing sufficient safety margin for the drone.

[0020] It should be noted that the specific calculation steps for the offset vector are as follows: The ground control station obtains the formation position number of each drone in the first group of drones. Each formation position number uniquely corresponds to a relative position within the formation. The relative position includes two dimensions: the first dimension is the forward and backward following position along the line direction, and the second dimension is the left and right offset position perpendicular to the line direction. The following position is represented by integers. For example, a negative number indicates that the formation reference point is behind the formation reference point, a positive number indicates that the formation reference point is in front of the formation reference point, and zero indicates that the formation reference point is aligned. The left and right offset positions are also represented by integers. A negative number indicates that the formation reference point is to the left of the reference cruise line, a positive number indicates that the formation reference point is to the right of the reference cruise line, and zero indicates that the formation reference point is aligned. The ground control station acquires a pre-set first lateral spacing and a first longitudinal spacing. The first lateral spacing is defined as the distance between two adjacent UAVs in the left-right offset direction in a plane perpendicular to the reference cruise line. The first longitudinal spacing is defined as the distance between two adjacent UAVs in the front-back direction along the tangent of the reference cruise line. For a specific drone in the first group of drones, the ground control station reads the drone's preset formation position number, parses out the preceding and following position values ​​and the left and right offset position values, multiplies the preceding and following position values ​​by the first longitudinal spacing to obtain a longitudinal offset, and multiplies the left and right offset position values ​​by the first lateral spacing to obtain a lateral offset. The ground control station takes the current formation reference point on the baseline cruise line, obtains the spatial coordinates of the point, as well as the tangent direction vector and normal direction vector of the baseline cruise line at that point. The tangent direction vector points to the forward direction of the line, and the normal direction vector is perpendicular to the tangent direction and lies in the horizontal plane, pointing to the right side of the baseline cruise line. The vertical offset is projected along the direction of the tangent vector to generate a vertical offset component, and the horizontal offset is projected along the direction of the normal vector to generate a horizontal offset component. The direction of the vertical offset component is determined by the sign of the values ​​of the preceding and following positions: when the value is positive, the vertical offset component points in the direction of the tangent vector, and when the value is negative, the vertical offset component points in the opposite direction of the tangent vector. The direction of the horizontal offset component is determined by the sign of the values ​​of the left and right offset positions: when the value is positive, the horizontal offset component points in the direction of the normal vector, and when the value is negative, the horizontal offset component points in the opposite direction of the normal vector. The longitudinal offset component and the lateral offset component are vector-added to obtain a composite vector. The composite vector is the offset vector of a specific UAV relative to the formation reference point. The magnitude of the offset vector is determined by the longitudinal offset and the lateral offset. The direction of the offset vector is located in a plane perpendicular to the reference cruise line and is determined by the composite direction of the longitudinal offset component and the lateral offset component. The ground control station repeatedly performs the offset vector acquisition step for each UAV in the first group of UAVs to obtain the offset vector of each UAV relative to the formation reference point. All offset vectors are based on the spatial coordinates of the formation reference point and are used to subsequently superimpose and generate the desired spatial coordinates of each UAV.

[0021] S2. Control the first group of drones to fly along the cruise path, and during the flight, collect physical field characteristic information in a preset space range around the power distribution line in real time through the sensor array mounted on each drone. The physical field characteristic information includes air micro-flow disturbance data and spatial electrostatic field distortion data. Furthermore, in S2, after the ground control station completes the loading of the formation cruise path, it simultaneously sends take-off and formation flight commands to the first group of UAVs via a wireless data link. Each drone autonomously flies to the starting waypoint according to its pre-stored waypoint sequence and adjusts its attitude so that the nose is tangential to the cruise path. Once all UAVs have reached the starting waypoint and formed the preset formation, the ground control station issues a cruise start command. Each UAV maintains formation flight along its own cruise path, while the onboard computer sends a start data acquisition command to the onboard sensor array. The sensor array includes a biomimetic microfluidic sensing array and an electrostatic field disturbance sensing probe. Both are started synchronously according to a preset first sampling frequency to begin real-time sensing of physical field changes within a preset spatial range around the drone. The biomimetic microfluidic sensing array consists of multiple miniature airflow sensing units arranged in a preset geometric layout. Each sensing unit converts the flow of air molecules across its surface into an electrical signal output. When there is partial discharge or overheating in the power distribution network line, the discharge or high temperature causes the surrounding air molecules to ionize and generate directional flow, forming air micro-flow disturbance. During the flight of the drone, each sensitive unit in the biomimetic microfluidic sensing array independently outputs a voltage value that is proportional to the airflow velocity; The onboard computer reads the output voltage values ​​of all sensitive units in the biomimetic microfluidic sensing array at the same time, forming a multi-dimensional raw data vector; The computer uses the average value of the output voltage of all sensitive units as a reference value, and subtracts the reference value from the output voltage of each sensitive unit to obtain the differential output value of each sensitive unit. The computer sorts the differential output values ​​according to their coordinate positions based on the geometric coordinates of each sensitive unit in the array, forming a spatial difference matrix; The computer finds the two sensitive cells with the largest difference value in the spatial difference matrix, calculates the direction of the line connecting the two sensitive cells, and uses the direction of the line as the overall direction of the airflow. The maximum difference value is multiplied by a preset first scaling factor to obtain the instantaneous velocity value of the airflow. The computer statistically analyzes the changes in airflow direction and velocity over time within a preset first time window. The changes are then divided by a preset second scaling factor to obtain the turbulence information of the airflow. The electrostatic field disturbance sensing probe consists of one or more metal electrodes with the electrode surface exposed to the air. When a drone flies over an area with insulator contamination, broken cable strands, or damaged wires, the potential hazard point generates local electric field distortion, causing a change in the electrostatic field strength in the surrounding space. The electrostatic field disturbance sensing probe on the UAV senses the spatial potential through electrodes, converts the sensed potential value into an analog voltage signal output, and the onboard computer continuously collects the analog voltage signal at a preset second sampling frequency to obtain a potential sequence that changes over time. The airborne computer acquires the potential change sequence within a preset second time window, takes the difference between the maximum and minimum values ​​in the sequence as the potential peak value, and divides the potential peak value by a preset third scaling coefficient to obtain the spatial electric field intensity value within the time window. The computer plots each sampled value in the potential change sequence into a waveform in chronological order, forming a potential change waveform. The computer performs baseline drift removal on the waveform, takes the average value of the entire waveform as the DC component, and subtracts the DC component from each sampled value to obtain the AC fluctuation component. The computer extracts the number of zero-crossing points, the number of peaks, and the steepness of the rising and falling edges of the AC wave component, and combines these features into distorted waveform information.

[0022] It should be noted that the specific steps for calculating the spatial difference matrix are as follows: The airborne computer acquires the preset geometric layout information of the biomimetic microfluidic sensing array. The preset geometric layout information is stored in the non-volatile memory of the airborne computer in advance, including the unique number of each micro airflow sensing unit in the array and the two-dimensional geometric coordinates of the unit in the array plane. The two-dimensional geometric coordinates take the geometric center of the array as the origin, the horizontal direction as the first coordinate axis, and the vertical direction as the second coordinate axis. The coordinates of each unit are represented by a pair of numerical values ​​with positive and negative signs. The airborne computer associates the differential output values ​​of all sensitive units in the biomimetic microfluidic sensing array at the same time with the unique number and two-dimensional geometric coordinates of the corresponding unit. Each differential output value is bound to the sensitive unit that produced the value, forming a set of triplet data. The triplet data includes the unique number of the sensitive unit, the two-dimensional geometric coordinates of the sensitive unit, and the differential output value of the unit. The onboard computer determines the minimum and maximum coordinate values ​​of the entire array in the horizontal direction, as well as the minimum and maximum coordinate values ​​in the vertical direction, based on the two-dimensional geometric coordinates of all sensitive units. Using the minimum and maximum coordinate values ​​in the horizontal direction as boundaries and a preset first coordinate interval as a step size, the horizontal direction is divided into multiple horizontal intervals. Using the minimum and maximum coordinate values ​​in the vertical direction as boundaries and the preset second coordinate interval as the step size, the vertical direction is divided into multiple vertical intervals. The first and second coordinate intervals are preset according to the average spacing between adjacent sensitive units in the array, so that each horizontal and vertical interval contains at most one sensitive unit. The onboard computer creates a blank two-dimensional table. The number of rows in the table is equal to the number of intervals in the vertical direction, and the number of columns is equal to the number of intervals in the horizontal direction. Each row corresponds to a vertical interval, arranged from top to bottom in ascending order of vertical coordinates; each column corresponds to a horizontal interval, arranged from left to right in ascending order of horizontal coordinates. The two-dimensional table is the framework of the spatial difference matrix to be filled. The onboard computer traverses the triplet data of each sensitive unit. For each sensitive unit, it extracts the horizontal and vertical coordinate values ​​from its two-dimensional geometric coordinates, determines which horizontal interval the horizontal coordinate value falls into, and which vertical interval the vertical coordinate value falls into, and determines the row and column numbers of the sensitive unit in the two-dimensional table based on the intervals they fall into. Then, it fills the differential output value of the sensitive unit into the cells corresponding to the row and column numbers in the two-dimensional table. For blank cells in a two-dimensional table that are not filled with any sensitive units, the onboard computer sets the cell value to a preset null value marker. The null value marker is a special value that is different from all valid differential output values, such as negative zero or a fixed large negative number that is outside the normal differential output range, used to indicate that there are no sensitive units at that position. After completing the triplet data traversal and filling of all sensitive units, the airborne computer obtains a complete two-dimensional table. Each non-empty cell in the table stores the differential output value of the sensitive unit at the corresponding position, and each empty cell stores the empty value marker. The two-dimensional table is the spatial difference matrix. The row order of the spatial difference matrix corresponds to the arrangement order of the sensitive units in the vertical direction, and the column order corresponds to the arrangement order of the sensitive units in the horizontal direction. This maps the discrete differential output values ​​into a regular matrix form according to geometric coordinates, which is convenient for finding the two sensitive units with the largest difference value and calculating the connection direction.

[0023] The preset first sampling frequency refers to the number of times data is collected per second when the bionic microfluidic sensing array and the electrostatic field disturbance sensing probe are started synchronously. The preset first sampling frequency is set according to the highest possible change frequency of the air microfluidic disturbance and spatial electrostatic field distortion signal generated by the weak fault in the power distribution network. The highest signal frequency is multiplied by the preset oversampling factor (e.g., 4 times or 8 times) according to the Nyquist sampling theorem, and is stored in advance in the configuration file of the airborne computer. The preset first scaling coefficient is used to convert the maximum differential output value in the biomimetic microfluidic sensing array into the instantaneous velocity value of the airflow, and the airflow velocity value corresponding to the unit differential output value (e.g., how many meters per second per volt). In a laboratory environment, the biomimetic microfluidic sensing array is placed in a standard airflow field with known velocity, and the linear relationship between the array output differential value and the actual airflow velocity is measured. The conversion slope is obtained through least squares fitting, and the conversion slope is used as the first scaling coefficient to be pre-calibrated and stored in the onboard computer. The preset first time window is used to statistically analyze the changes in airflow direction and speed over time to calculate turbulence intensity. It is the length of continuous data captured during turbulence intensity analysis. The preset first time window takes a typical value of the duration (e.g., 0.5 to 2 seconds) as the length of the first time window based on the duration characteristics of the micro-flow disturbances generated by the hidden points in the power distribution network, and is preset. The preset second scaling coefficient is used to convert the change amplitude of airflow direction and velocity into turbulence value. It is the turbulence increment corresponding to a unit change amplitude. The preset second scaling coefficient is calibrated experimentally by measuring the change amplitude in an airflow field that generates known turbulence, dividing the change amplitude by the actual turbulence to obtain the second scaling coefficient, and storing it in the onboard computer. The preset second sampling frequency is the sampling rate of the analog voltage signal output by the electrostatic field disturbance sensing probe, which is the number of potential values ​​collected per second. The preset second sampling frequency is determined according to the upper limit of the high-frequency component of the local electric field distortion signal of the distribution network (e.g., the frequency corresponding to the steep pulse leading edge generated by partial discharge), according to the sampling theorem, and a sampling rate of not less than twice the upper limit frequency is determined, and an additional engineering margin is added (e.g., multiplied by 2 to 4 times), and finally preset to a fixed value. The preset second time window is used to extract the maximum value, minimum value and waveform features in the potential change sequence. It is the length of time for continuous data acquisition when performing spatial electric field strength and distorted waveform analysis. The preset second time window is set based on an integer multiple of the power frequency cycle of the distribution network to ensure that the window contains the complete power frequency electric field change law, while also capturing transient distortion events in a timely manner. The preset third scaling coefficient is used to convert the peak-to-peak potential value into a spatial electric field strength value. It is the electric field strength value corresponding to a unit potential difference. The preset third scaling coefficient is obtained by measuring the corresponding peak-to-peak potential value using an electrostatic field disturbance sensing probe in a uniform electric field environment with a known standard electric field strength. The known electric field strength is divided by the measured peak-to-peak value to obtain the third scaling coefficient. After repeated testing and averaging, the average value is stored in the onboard computer.

[0024] S3. Based on the physical feature information and corresponding spatial coordinates collected by each UAV in the first group, construct a physical field distribution gradient map of the area covered by the inspection path; Furthermore, in S3, the ground control station receives data points from all UAVs in the first group, where each data point includes three-dimensional spatial coordinates, air microfluidic disturbance data, and spatial electrostatic field distortion data. The air microfluidic disturbance data includes airflow direction, airflow speed, and turbulence information, and the spatial electrostatic field distortion data includes spatial electric field strength and distortion waveform information. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding air microfluidic disturbance data from the data points. The ground control station divides the entire cruise path coverage area into a uniform three-dimensional grid according to a preset first spatial step size. Each grid cell is a cube voxel. For each grid cell, the ground control station collects the air microfluidic disturbance data of all data points falling into the grid cell. The arithmetic mean of the collected airflow velocity values ​​is taken as the representative airflow velocity of the grid cell. The representative airflow direction value is obtained by vector averaging the collected airflow direction value. The arithmetic mean of the collected turbulence values ​​is taken as the representative turbulence of the grid cell. For blank grid cells where no data points fall, the ground control station uses a spatial interpolation method to search outward from the blank grid cell with the pre-set first search radius for the already assigned grid cells. The representative airflow velocity, representative airflow direction, and representative turbulence of these already assigned grid cells are estimated by weighting inversely proportionally to the distance and then filled into the blank grid cell. After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional hybrid field map, which is used as the first microfluidic field map. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding spatial electrostatic field distortion data from the data points. The ground control station uses the same three-dimensional grid division method as the one used to establish the first microflow field map to divide the entire cruise path coverage area into three-dimensional grids of the same size. For each grid cell, the ground control station collects the spatial electric field intensity values ​​of all data points falling within the grid cell, and takes the arithmetic mean as the representative electric field intensity of the grid cell. At the same time, it collects the distorted waveform information of all data points falling within the grid cell, and takes the arithmetic mean of the number of zero crossings, the number of peaks, and the steepness of the waveform in the distorted waveform information to obtain a set of representative distortion characteristic values. For blank grid cells, the ground control station uses the same spatial interpolation method as the first microflow field map to search for neighboring already assigned grid cells with a preset second search radius, and calculates the representative electric field intensity and representative distortion characteristic value of the blank grid cell by distance inverse weighted interpolation. After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional scalar field map, which is used as the first electric field map. The ground control station spatially registers the first microfluidic field map and the first electric field map: the ground control station checks whether the grid origin, grid step size and grid range of the two field maps are completely consistent. If there is any inconsistency, the field map with the larger coverage area is used as the reference, and the field map with the smaller coverage area is extended by boundary. The grid cells of the extended part are filled by copying the values ​​of the neighboring grid cells outward to ensure that the two field maps have the same grid topology. The ground control station creates a new 3D grid with the same grid topology as the registered field map, serving as the framework for the integrated physical field distribution gradient map; For each grid cell in the new three-dimensional grid, the ground control station reads the airflow velocity, airflow direction and turbulence intensity of the grid cell from the first microflow field map, and reads the spatial electric field intensity and distortion characteristic value of the grid cell from the first electric field map. The data are combined into a multi-dimensional feature vector and stored in the grid cell. The ground control station calculates the comprehensive gradient value of each grid cell: first, it calculates the sum of the absolute values ​​of the differences in airflow velocity between the grid cell and its multiple neighboring grid cells to obtain the first gradient component; The second gradient component is obtained by summing the absolute values ​​of the differences in spatial electric field intensity between grid cells and their adjacent grid cells. Finally, the first gradient component and the second gradient component are weighted and summed according to the preset first weight and second weight to obtain the comprehensive gradient value of the grid cell. The combined gradient values ​​of all grid cells are arranged according to spatial coordinates to form a three-dimensional gradient value distribution map, which is then used as a multi-dimensional combined physical field distribution gradient map.

[0025] It should be noted that the specific steps for calculating the representative electric field strength and the representative distortion eigenvalue are as follows: A1. The ground control station obtains the three-dimensional spatial coordinates of the center point of the blank grid cell that needs to be interpolated. The center point coordinates are calculated from the row, column, and layer index of the blank grid cell and the side length of the grid cell (i.e., the preset first spatial step). The first-dimensional coordinate is obtained by multiplying the row index of the grid cell by the first spatial step, the second-dimensional coordinate is obtained by multiplying the column index by the first spatial step, and the third-dimensional coordinate is obtained by multiplying the layer index by the first spatial step. A2. The ground control station reads the preset second search radius. The second search radius is a preset distance value that is greater than twice the first spatial step size and less than five times the first spatial step size. It is used to limit the spatial range of searching for neighboring assigned grid cells. The second search radius is obtained by selecting a radius value that can ensure a sufficient number of neighboring grid cells are searched while avoiding the introduction of interference from too far-away cells, based on the size of the grid cells and the continuous change characteristics of the physical field in space. This radius value is then stored in the configuration file of the ground control station. A3. The ground control station defines a spherical search space with the center point coordinates of the blank grid cell as the center and the preset second search radius as the radius. The ground control station traverses all the assigned grid cells in the entire first electric field diagram (i.e., the grid cells that have obtained the electric field strength and distortion characteristic values ​​through direct measurement or previous interpolation). It calculates the spatial straight-line distance between the center point coordinates of each assigned grid cell and the center point coordinates of the blank grid cell. If the spatial straight-line distance is less than or equal to the second search radius, the assigned grid cell is marked as a candidate neighbor cell, and the center point coordinates, the representative electric field strength value, and a set of representative distortion characteristic values ​​(including the average number of zero crossings, the average number of peaks, and the average waveform steepness) of the candidate neighbor cell are recorded. A4. The ground control station counts the total number of candidate neighboring cells. If the total number of candidate neighboring cells is zero, that is, there are no assigned grid cells in the spherical search space, the ground control station expands the preset second search radius to twice the original value and repeats A3 until at least one candidate neighboring cell is found. If there are still no candidate neighboring cells after expanding three times, the ground control station sets the representative electric field intensity of the blank grid cell to the preset default electric field intensity value, sets the number of zero crossings, the number of peaks, and the waveform steepness in the representative distortion characteristic value to zero, and ends the calculation of the blank grid cell. A5. For each candidate neighboring cell found, the ground control station calculates the reciprocal of the spatial straight-line distance between the candidate neighboring cell and the blank grid cell. The reciprocal is used as the initial weight value of the candidate neighboring cell. The smaller the spatial straight-line distance, the larger its reciprocal, indicating that the candidate neighboring cell contributes more to the blank grid cell. A6. The ground control station adds up the initial weight values ​​of all candidate neighboring units to obtain a total weight value. For each candidate neighboring unit, the ground control station divides its initial weight value by the total weight value to obtain the normalized weight value of the candidate neighboring unit. The sum of the normalized weight values ​​is equal to one. A7. The ground control station calculates the representative electric field intensity of the blank grid cell. The ground control station multiplies the representative electric field intensity of each candidate neighbor cell by the normalized weight value of that cell to obtain the weighted electric field intensity contribution value of each candidate neighbor cell. The weighted electric field intensity contribution values ​​of all candidate neighbor cells are added together, and the sum is the representative electric field intensity of the blank grid cell. A8. The ground control station calculates the number of zero-crossings component in the representative distortion eigenvalue of the blank grid cell. The ground control station multiplies the average number of zero-crossings of each candidate neighboring cell by the normalized weight value of that cell to obtain the weighted contribution value of the number of zero-crossings of each candidate neighboring cell. The weighted contribution values ​​of the number of zero-crossings of all candidate neighboring cells are added together, and the sum is the representative number of zero-crossings of the blank grid cell.

[0026] A9. The ground control station calculates the peak count component and waveform steepness component in the representative distortion feature value in the same way as in step 8. For the peak count component, the average peak count of each candidate neighboring unit is multiplied by the normalized weight value and then summed to obtain the representative peak count. For the waveform steepness component, the average waveform steepness of each candidate neighboring unit is multiplied by the normalized weight value and then summed to obtain the representative waveform steepness.

[0027] A10. The ground control station combines the calculated representative electric field strength, the number of representative zero crossings, the number of representative peaks, and the representative waveform steepness as the complete representative distortion characteristic value of the blank grid cell. Among them, the number of representative zero crossings, the number of representative peaks, and the representative waveform steepness together constitute a set of representative distortion characteristic values. At this point, the ground control station has completed the interpolation calculation of the blank grid cell and obtained the representative electric field strength and representative distortion characteristic value of the grid cell.

[0028] The preset first spatial step size refers to the side length of each cube voxel (i.e., grid cell) when the ground control station divides the coverage area of ​​the cruise path into a three-dimensional grid. It is the spatial discretization sampling interval of the physical field data. The preset first spatial step size is based on the spatial variation gradient characteristics of the physical field (air microfluidic field and electrostatic field) around the power distribution line. It takes one-quarter to one-eighth of the gradient feature length as the initial value of the first spatial step size. At the same time, considering the spatial density of the data points collected by the UAV (i.e. the average distance between adjacent data points), the first spatial step size is set to a value slightly larger than the average distance to ensure that at least one data point can fall in each grid cell. The preset first search radius is used to perform spatial interpolation on blank grid cells when establishing the first microfluidic field map. It is a spherical radius that searches for neighboring assigned grid cells in three-dimensional space with the center of the blank grid cell as the center. The preset first search radius is set according to the multiple relationship of the first spatial step size, usually two to three times the first spatial step size, so that the search range can cover a complete neighborhood around the blank grid cell (including adjacent grid cells in the six directions of up, down, left, right, front and back), while not including grid cells that are too far away in the interpolation calculation and smoothing out the true gradient changes of the local physical field. The preset first and second weights are used to calculate the comprehensive gradient value of each grid cell in the comprehensive physical field distribution gradient map. The first weight corresponds to the contribution ratio of the airflow velocity gradient component (i.e., the first gradient component) in the comprehensive gradient, and the second weight corresponds to the contribution ratio of the spatial electric field intensity gradient component (i.e., the second gradient component) in the comprehensive gradient. The sum of the two is equal to one. The preset first and second weights are experimentally calibrated according to the sensitivity of the target distribution network line to air micro-flow disturbances and spatial electrostatic field distortions based on typical fault types. For faults mainly caused by partial discharge and overheating, the air micro-flow disturbance signal is more sensitive, so the first weight is set to a value greater than the second weight (e.g., 0.7 to 0.3). For faults mainly caused by insulator contamination and broken cable strands, the spatial electrostatic field distortion signal is more sensitive. Therefore, the second weight is set to a value greater than the first weight. For comprehensive faults or unknown fault types, they can be set to be equal. The ground control station pre-selects a set of weight values ​​according to the target fault type of the inspection task and loads them before the task begins.

[0029] S4. Identify abnormal gradient regions in the physical field distribution gradient map, and determine the spatial coordinates of suspected hidden danger points based on the center point of the abnormal gradient region. Furthermore, in S4, the ground control station reads a multi-dimensional integrated physical field distribution gradient map. The gradient map consists of multiple grid cells arranged in a three-dimensional grid, and each grid cell stores a comprehensive gradient value. The ground control station traverses all grid cells and uses the comprehensive gradient value stored in each grid cell as the field strength gradient magnitude value of the grid cell. The ground control station determines the field strength gradient direction based on the trend of airflow velocity change between each grid cell and its neighboring grid cells: find the neighboring grid cell with the largest comprehensive gradient value in the six directions around each grid cell, calculate the direction vector from the current grid cell to the neighboring cell, normalize the direction vector and use it as the field strength gradient direction of the current grid cell. If the comprehensive gradient values ​​of multiple neighboring cells are equal and all are the maximum, then the composite direction of multiple direction vectors is used as the field strength gradient direction. The ground control station obtains the pre-stored field strength gradient magnitude threshold, traverses all grid cells again, marks grid cells with field strength gradient magnitude greater than or equal to the preset threshold as candidate abnormal grid cells, and marks grid cells with field strength gradient magnitude less than the preset threshold as normal grid cells. The ground control station uses a connected component analysis algorithm to merge all adjacent candidate abnormal grid cells into a connected component. For each connected region, the ground control station checks whether the total number of grid cells contained in the connected region is greater than the preset minimum region area threshold. If it is greater, the connected region is marked as a preliminary abnormal region and the spatial coordinate list of all grid cells contained in the preliminary abnormal region is recorded. If it is less than, the connected region is excluded. For each initially selected anomaly region, the ground control station obtains the field intensity gradient direction of all grid cells in the region, calculates the average direction of all field intensity gradient directions in the region, and uses the average direction as the region's reference direction. For each grid cell within the region, the ground control station calculates the angle between the field strength gradient direction and the regional reference direction; If the included angle of grid cells in a region that exceed the preset consistency ratio threshold is less than the preset direction deviation threshold, then the initially selected abnormal region is marked as a candidate abnormal gradient region. The ground control station further judges the convergence characteristics of the candidate abnormal gradient region: selects all grid cells located at the edge of the region within the candidate abnormal gradient region, calculates whether the gradient direction of each edge grid cell points into the region, and if the edge grid cells in the region that exceed the preset convergence ratio threshold meet the judgment condition of pointing inward, then the candidate abnormal gradient region is determined as an abnormal gradient region. For each connected region identified as an anomalous gradient region, the ground control station extracts the coordinates of the center point of all grid cells on the region boundary as the coordinates of the edge point. The boundary grid cell is a grid cell in the anomalous gradient region that has at least one adjacent grid cell that does not belong to the anomalous gradient region. The ground control station collects the coordinates of the center points of all boundary grid cells to form a set of edge point coordinates. The ground control station uses the rotating caliper method or the exhaustive search method to find the smallest volume geometry in three-dimensional space that can include all edge point coordinates. The smallest volume geometry is pre-selected as a cuboid, sphere or ellipsoid. The ground control station calculates the coordinates of the geometric center point based on the type of the smallest circumscribed geometric figure: if the smallest circumscribed geometric figure is a cuboid, then the average of the minimum and maximum values ​​of the cuboid in the three coordinate axes is taken, and the three-dimensional coordinates formed by the three average values ​​are used as the coordinates of the geometric center point of the cuboid. If the smallest circumscribed geometric figure is a sphere, then the coordinates of the center of the sphere are used as the coordinates of the geometric center point; if the smallest circumscribed geometric figure is an ellipsoid, then the center points corresponding to the lengths of the three semi-axis of the ellipsoid are calculated as the coordinates of the geometric center point, and the ground control station determines the coordinates of the geometric center point as the spatial coordinates of the first suspected hidden danger point. The ground control station records the spatial coordinates of the first suspected hidden danger point in memory in the form of three-dimensional coordinates, and stores them in association with the identifier, size and average gradient magnitude of the abnormal gradient region corresponding to the coordinates. If multiple abnormal gradient regions are identified within the coverage area of ​​the same cruise path, the ground control station will sequentially perform the above steps for each abnormal gradient region to generate multiple spatial coordinates of first suspected hidden danger points.

[0030] It should be noted that the specific steps for calculating the direction vector from the current grid cell to the adjacent cell are as follows: The ground control station obtains the index position of the currently processed grid cell in the 3D grid. Each grid cell is uniquely determined by the row index, column index, and layer index. The row index corresponds to the first coordinate axis direction, the column index corresponds to the second coordinate axis direction, and the layer index corresponds to the third coordinate axis direction. The ground control station also obtains the 3D spatial coordinates of the center point of the grid cell. The 3D spatial coordinates are calculated by multiplying the index position of the grid cell by the preset first spatial step size. The ground control station identifies six neighboring grid cells around the current grid cell. These six neighboring grid cells are located in the following six directions of the current grid cell: the first direction is where the row index increases by one (corresponding to the positive direction of the first coordinate axis), the second direction is where the row index decreases by one (corresponding to the negative direction of the first coordinate axis), the third direction is where the column index increases by one (corresponding to the positive direction of the second coordinate axis), the fourth direction is where the column index decreases by one (corresponding to the negative direction of the second coordinate axis), the fifth direction is where the layer index increases by one (corresponding to the positive direction of the third coordinate axis), and the sixth direction is where the layer index decreases by one (corresponding to the negative direction of the third coordinate axis). For each direction, the ground control station calculates the index of the neighboring grid cells based on the index of the current grid cell and checks whether the neighboring grid cells are within the boundary range of the 3D grid (i.e., the row index, column index, and layer index are all not less than zero and not greater than the maximum index value in their respective directions). If a neighboring grid cell in a certain direction exceeds the grid boundary, that direction is marked as invalid and will not be included in subsequent comparisons. The ground control station reads the integrated gradient value stored in each effective adjacent grid cell from the integrated physical field distribution gradient map. The integrated gradient value is a scalar value stored in each grid cell. The ground control station records the index position of each effective adjacent grid cell and its corresponding integrated gradient value. The ground control station compares the combined gradient values ​​of all valid adjacent grid cells to find the maximum value. The comparison process is as follows: The ground control station takes the combined gradient value of the first valid adjacent grid cell as the temporary maximum value and records the index position of that cell as the temporary maximum cell. It then iterates through each of the remaining valid adjacent grid cells and compares the combined gradient value of the currently traversed cell with the temporary maximum value. If the combined gradient value of the current cell is greater than the temporary maximum value, the temporary maximum value is updated to the combined gradient value of the current cell, and the temporary maximum cell is updated to the index of the current cell. If the comprehensive gradient value of the current cell is equal to the temporary maximum value, the original temporary maximum cell is temporarily retained. At the same time, the case where there are multiple equal maximum values ​​is recorded. After traversing all valid adjacent cells, the ground control station obtains the maximum value of the comprehensive gradient value and the corresponding index of one or more adjacent cells. The ground control station handles cases where multiple adjacent cells have the same maximum integrated gradient value. If only one adjacent cell has an integrated gradient value equal to the maximum value, then that adjacent cell is determined as the final target adjacent cell. If multiple adjacent cells have integrated gradient values ​​equal to the maximum value, the ground control station calculates the spatial straight-line distance between the center point coordinates of the current grid cell and the center point coordinates of each adjacent cell with the maximum value, and selects the adjacent cell with the smallest spatial straight-line distance as the final target adjacent cell. If the spatial straight-line distances are also equal, then an adjacent cell is selected as the final target adjacent cell according to a preset priority order (e.g., the first direction takes precedence over the second direction, the second direction takes precedence over the third direction, and so on). The ground control station acquires the three-dimensional spatial coordinates of the center point of the current grid cell, which is recorded as the first coordinate. At the same time, it acquires the three-dimensional spatial coordinates of the center point of the finally determined target adjacent cell, which is recorded as the second coordinate. The center point coordinate is calculated as follows: multiply the row index of the grid cell by the preset first spatial step to obtain the first dimension coordinate value, multiply the column index by the first spatial step to obtain the second dimension coordinate value, and multiply the layer index by the first spatial step to obtain the third dimension coordinate value. The coordinate values ​​of the three dimensions together constitute the three-dimensional spatial coordinates of the center point. The ground control station calculates the direction vector from the current grid cell to the target's adjacent cell. The calculation process for the direction vector is as follows: subtract the first dimension coordinate value from the first dimension coordinate value in the second coordinate system to obtain a first difference value. Subtracting the second-dimensional coordinate value from the second-dimensional coordinate value in the first coordinate system yields a second difference; subtracting the third-dimensional coordinate value from the third-dimensional coordinate value in the first coordinate system yields a third difference. The three values, consisting of the first difference, the second difference, and the third difference, together form a three-dimensional direction vector, which represents the spatial direction and distance from the center of the current grid cell to the center of the target adjacent cell. The ground control station normalizes the direction vector obtained in step seven. The normalization process is as follows: First, calculate the magnitude of the direction vector by squaring the first difference, the second difference, and the third difference, then adding them together, and finally taking the square root of the sum to obtain the magnitude value. Dividing the first difference by the magnitude gives the normalized first component. Dividing the second difference by the magnitude gives the normalized second component. Dividing the third difference by the magnitude gives the normalized third component. The three normalized components form a unit direction vector of length 1. The direction of the unit direction vector is the same as the original direction vector, but the length is scaled to a unit length. The ground control station uses this unit direction vector as the final field strength gradient direction of the current grid cell.

[0031] The ground control station uses a connected component analysis algorithm to merge all adjacent candidate anomaly grid cells into a single connected component. The specific steps are as follows: The ground control station first creates a three-dimensional marker array with the same grid topology as the integrated physical field distribution gradient map. Each element in the marker array corresponds one-to-one with a grid cell and is used to record the connected region number to which the grid cell belongs. The ground control station sets the initial value of all elements in the marker array to zero, indicating that they have not been visited by any connected region. At the same time, the ground control station initializes a connected region counter and sets the initial value of the counter to zero. The ground control station traverses all grid cells in the entire 3D mesh in ascending order of row index, in ascending order of column index within each row, and in ascending order of layer index within each layer. For each traversed grid cell, the ground control station first checks whether the grid cell has been marked as a candidate abnormal grid cell. If the current grid cell is not a candidate abnormal grid cell, the ground control station skips the grid cell and continues to traverse the next grid cell. If the current grid cell is a candidate abnormal grid cell, the ground control station further checks whether the element value corresponding to the grid cell in the marker array is zero. If the element value is not zero, it means that the grid cell has been assigned to a certain connected region, and the ground control station skips the grid cell. If the element value is zero, the ground control station determines that it has found the starting grid cell of a new connected region. At this time, the ground control station increments the connected region counter by one to obtain a new connected region number, and assigns the new connected region number to the element corresponding to the current grid cell in the marker array. The ground control station expands the connected region using a breadth-first search approach. It creates a first-in, first-out queue, adding the index position of the current grid cell (including row, column, and layer indices) to the queue. The ground control station repeats the following operations until the queue is empty: Take the index position of a grid cell from the head of the queue and record it as the current processing unit; The ground control station checks the grid cells in the six adjacent directions surrounding the current processing unit (i.e., row index increases by one, row index decreases by one, column index increases by one, column index decreases by one, layer index increases by one, layer index decreases by one); For each direction, the ground control station first determines whether the adjacent grid cells in that direction are within the boundary of the 3D grid. If they are outside the boundary, the direction is skipped. If an adjacent grid cell is within the boundary, the ground control station checks whether that adjacent grid cell has been marked as a candidate anomalous grid cell. If the adjacent grid cell is not a candidate anomalous grid cell, skip that direction; If the adjacent grid cell is a candidate anomalous grid cell, the ground control station checks whether the value of the element corresponding to the adjacent grid cell in the marker array is zero; If the element value is zero, the index of the adjacent grid cell is placed at the tail of the queue, and the current connected region number is assigned to the element corresponding to the adjacent grid cell in the marker array; if the element value is not zero, the adjacent grid cell is skipped. When the queue is empty, it means that all candidate abnormal grid cells that can be reached from the starting grid cell through adjacency have been visited and marked as the current connected region number. These grid cells marked with the same number together constitute a connected region. The ground control station records the list of index positions of all grid cells contained in the connected region and stores it in association with the connected region number. The ground control station continues to traverse the remaining unvisited grid cells in the 3D mesh, repeating the above process until all grid cells have been processed. Finally, the ground control station obtains several connected regions, each consisting of a set of mutually adjacent (i.e. connected by six-neighborhood paths) candidate anomalous grid cells, and each connected region has a unique number.

[0032] The specific steps for the ground control station to calculate the average direction of all field intensity gradient directions within the region are as follows. The ground control station first obtains the field intensity gradient direction of all grid cells in the current preliminary anomaly area. The field intensity gradient direction of each grid cell has been calculated as a unit direction vector. The unit direction vector consists of three components, which correspond to the components of the first coordinate axis direction, the second coordinate axis direction, and the third coordinate axis direction, respectively, and the sum of the squares of the three components is equal to one. The ground control station creates three accumulators, named First Component Accumulator, Second Component Accumulator, and Third Component Accumulator, and sets the initial value of all three accumulators to zero. At the same time, the ground control station creates a counter to record the total number of grid cells in the area and sets the initial value of the counter to zero. The ground control station traverses every grid cell within the region. For each traversed grid cell, the ground control station reads the three component values ​​of the field strength gradient direction of that grid cell. The ground control station adds the first component value to the first component accumulator, the second component value to the second component accumulator, and the third component value to the third component accumulator. After processing each grid cell, the ground control station increments the counter value by one. After traversing all grid cells in the region, the ground control station obtains the final values ​​of the three accumulators, which are denoted as the first component sum, the second component sum, and the third component sum, respectively, as well as the total number of grid cells in the region. The ground control station checks if the total number is zero. If the total number is zero (i.e., there are no grid cells in the area), the ground control station cannot calculate the average direction. In this case, the average direction is set to a preset default direction vector, such as a unit direction vector with the first component being positive 1, the second component being zero, and the third component being zero, and the calculation ends. If the total number is not zero, the ground control station first calculates the three components of the average vector. Specifically, the ground control station divides the sum of the first components by the total number to obtain the first average component, divides the sum of the second components by the total number to obtain the second average component, and divides the sum of the third components by the total number to obtain the third average component. The first average component, the second average component, and the third average component together form a three-dimensional vector, called the average direction vector. However, the magnitude of this vector is not necessarily equal to one. The ground control station normalizes the average direction vector to make it a unit direction vector, which serves as the average direction of all field strength gradient directions in the final region. The specific steps of the normalization process are as follows: First, calculate the magnitude of the average direction vector by squaring the first, second, and third average components respectively, adding the three squared values, and then taking the square root of the sum to obtain the magnitude value. Then, divide the first average component by the magnitude to obtain the normalized first component, divide the second average component by the magnitude to obtain the normalized second component, and divide the third average component by the magnitude to obtain the normalized third component. The ground control station checks whether the normalized modulus is zero. If the modulus is zero (i.e., the first average component, the second average component, and the third average component are all zero), it indicates that the field intensity gradient directions of all grid cells in the region completely cancel each other out. At this time, the average direction is set to the preset default direction vector.

[0033] If the modulus is not zero, the unit vector formed by the normalized first, second, and third components is the average direction to be sought. The ground control station uses this average direction as the regional reference direction for calculating the angle between the gradient direction of each grid cell in the region and the regional reference direction in subsequent steps.

[0034] The specific steps for calculating the angle between the field gradient direction and the regional reference direction are as follows: The ground control station first obtains the field strength gradient direction of the current grid cell. The field strength gradient direction has been represented as a unit direction vector in the previous step, consisting of three components, corresponding to the components of the first coordinate axis direction, the second coordinate axis direction, and the third coordinate axis direction, respectively, and the sum of the squares of the three components is equal to one. At the same time, the ground control station obtains the regional reference direction of the current preliminary anomaly area. This direction is also represented as a unit direction vector, consisting of three components, and the sum of the squares of the three components is equal to one. The ground control station calculates the dot product between the field strength gradient direction and the regional reference direction. The calculation process is as follows: multiply the first component of the field strength gradient direction with the first component of the regional reference direction to obtain a first product; multiply the second component of the field strength gradient direction with the second component of the regional reference direction to obtain a second product; multiply the third component of the field strength gradient direction with the third component of the regional reference direction to obtain a third product; add the first, second, and third products to obtain the dot product. Since both direction vectors are unit vectors, the dot product is equal to the cosine of the angle between the two directions. The ground control station checks the range of the dot product value. Due to possible floating-point rounding errors during the calculation process, the dot product value may slightly exceed the range of -1 to +1. If the dot product value is greater than +1, the ground control station will truncate the dot product value to +1. If the dot product value is less than -1, the ground control station will truncate the dot product value to -1. The ground control station calculates the included angle based on the dot product value. The included angle is obtained through inverse cosine transformation. The specific logic is as follows: if the dot product value is positive 1, the included angle is 0 degrees, indicating that the two directions are exactly the same; if the dot product value is negative 1, the included angle is 180 degrees, indicating that the two directions are completely opposite; if the dot product value is zero, the included angle is 90 degrees, indicating that the two directions are perpendicular to each other. For other dot product values ​​between -1 and +1, the ground control station obtains the corresponding angle value by looking up the inverse cosine mapping table pre-stored in memory. The mapping table divides the dot product values ​​in the range of -1 to +1 into multiple sub-intervals at equal intervals. Each sub-interval corresponds to a preset angle value. The ground control station reads the corresponding angle value as an approximate value of the included angle based on the sub-interval in which the dot product value falls. If higher precision is required, the ground control station can use a linear interpolation method: determine the left and right boundary points of the sub-interval in which the dot product value is located and their corresponding left and right angle values, and then calculate an angle value proportionally between the left and right angle values ​​based on the relative position of the dot product value in the sub-interval. The ground control station records the final included angle value as the angle numerical value.

[0035] The specific steps for calculating whether the gradient direction of each edge grid cell points into the interior of the region are as follows: The ground control station obtains the geometric center coordinates of the current abnormal gradient region. The center coordinates are the spatial coordinates of the first suspected hidden danger point. Alternatively, the average value of the center coordinates of all grid cells in the abnormal gradient region is recalculated as the center coordinates of the region. The ground control station marks the center coordinates of this center as the center of the region. The ground control station obtains the coordinates of the center point of the edge grid cell currently being judged, and records them as the edge point coordinates. The edge grid cell is the grid cell that has been determined to be on the boundary of the abnormal gradient region. That is, the grid cell belongs to the abnormal gradient region and there is at least one adjacent grid cell that does not belong to the abnormal gradient region. The ground control station calculates the direction vector from the edge point coordinates to the center of the region. Specifically, it subtracts the first-dimensional coordinates of the edge point coordinates from the first-dimensional coordinates of the region center to obtain a first difference; it subtracts the second-dimensional coordinates of the edge point coordinates from the second-dimensional coordinates of the region center to obtain a second difference; and it subtracts the third-dimensional coordinates of the edge point coordinates from the third-dimensional coordinates of the region center to obtain a third difference. The three-dimensional vector composed of the first, second, and third differences is the direction vector from the edge point to the center of the region. The ground control station normalizes the direction vector pointing from the edge point to the center of the area to obtain a unit direction vector. The normalization steps are as follows: First, calculate the magnitude of the direction vector by squaring the first difference, the second difference, and the third difference, adding them together, and then taking the square root of the sum. Then, divide the first difference by the magnitude to obtain the first component, divide the second difference by the magnitude to obtain the second component, and divide the third difference by the magnitude to obtain the third component. These three components constitute a unit direction vector with a length of 1, denoted as the internal pointing vector. The ground control station obtains the field intensity gradient direction of the edge grid cells. The field intensity gradient direction has been calculated as a unit direction vector, denoted as the gradient direction vector. The ground control station calculates the angle between the gradient direction vector and the inner pointing vector using the following steps: multiply the first component of the gradient direction vector by the first component of the inner pointing vector to obtain the first product; multiply the second component of the gradient direction vector by the second component of the inner pointing vector to obtain the second product; multiply the third component of the gradient direction vector by the third component of the inner pointing vector to obtain the third product; add the first, second, and third products to obtain the dot product value; restrict the dot product value to the interval between -1 and +1; and obtain the angle value (in degrees) through inverse cosine mapping or table lookup. The ground control station determines whether the included angle is less than a preset inward pointing angle threshold, which is usually 90 degrees. If the included angle is less than 90 degrees, it is determined that the gradient direction of the edge grid cell points inward. If the included angle is greater than or equal to 90 degrees, it is determined that it does not point inward. The ground control station records the determination result (true or false). The specific steps for the ground control station to find the smallest volume geometry in three-dimensional space that can encompass the coordinates of all edge points, using either the rotating caliper method or the exhaustive search method, are as follows: Rotary chuck method The ground control station acquires the coordinates of all edge points to form a three-dimensional point set. The ground control station calculates the three-dimensional convex hull of this point set. The calculation method of the convex hull is as follows: First, find the point with the smallest first coordinate axis direction in the point set as the starting point. Sort all points according to polar angle order. Then, use the Graham scan method or incremental method to construct the faces and edges of the convex hull in turn. Finally, a convex polyhedron composed of several triangular faces is obtained. The convex polyhedron contains all edge points and is convex. The ground control station treats each face on the convex hull as a candidate bottom face. For each candidate bottom face, the ground control station calculates the normal direction of the bottom face and projects all points on the convex hull onto a plane with the normal direction as the normal, thus obtaining a two-dimensional point set. The ground control station uses the two-dimensional rotating caliper method on the projection plane to calculate the minimum area bounding rectangle of the two-dimensional point set. The specific steps are as follows: calculate the convex hull (two-dimensional) of the projection point set, find an edge on the convex hull as the starting edge, rotate the edge and update the heel point at the same time, record the area of ​​the bounding rectangle under each rotation angle, and take the rectangle with the smallest area. The ground control station constructs a cuboid in three-dimensional space based on the minimum area rectangle and the normal direction of the bottom surface: the normal direction of the bottom surface is used as the height direction of the cuboid, the two sides of the rectangle are used as the other two directions of the cuboid, the range of the rectangle on the projection plane determines the bottom size of the cuboid, and the projection range of the points on the convex hull on the normal direction of the bottom surface determines the height of the cuboid. The ground control station calculates the volume of the cuboid (i.e., the area of ​​the base multiplied by the height). After traversing all candidate bases, the ground control station compares the volumes of all candidate cuboids, selects the cuboid with the smallest volume as the minimum volume geometry, and outputs the coordinates of the center point of the cuboid, the direction vectors of the three axes, and the half-lengths of the three axes. Exhaustive search method The ground control station acquires the coordinates of all edge points, calculates the minimum and maximum values ​​of the point set in the three coordinate axes, and obtains an initial axis-aligned bounding box. The three faces of the bounding box are parallel to the coordinate plane, and the volume of the axis-aligned bounding box is calculated. The ground control station sets a search step size angle, for example, a preset angle interval of five degrees. The ground control station enumerates all possible rotational attitudes in three-dimensional space, specifically: around the first coordinate axis from zero to 180 degrees, increasing by step size; around the second coordinate axis from zero to 180 degrees, increasing by step size; around the third coordinate axis from zero to 180 degrees, increasing by step size. Each combination of three rotational angles represents a spatial orientation. For each spatial orientation, the ground control station performs coordinate transformation on all edge point coordinates according to that orientation, that is, rotates the original coordinates to a new coordinate system based on that orientation. In the new coordinate system, the ground control station calculates the minimum and maximum values ​​in the three coordinate axis directions of all transformed points, thereby obtaining an axis-aligned bounding box, and calculates the volume of the bounding box (i.e., the product of the side lengths in the three directions). The ground control station traverses all orientation combinations, records the minimum volume obtained and its corresponding orientation. If a volume of zero is found during the search process (i.e. all edge points are coplanar or collinear), the volume is considered invalid and the search continues. After enumerating all orientations, the ground control station selects the bounding box with the smallest volume as the minimum volume geometry. The orientation angle, center point coordinates (in the original coordinate system), and side lengths in three directions of this bounding box are output. If more accurate results are needed, a second refinement search can be performed near the orientation corresponding to the minimum volume, reducing the search step size (e.g., one degree or 0.5 degrees). The third and fourth steps are repeated until the step size is less than the preset accuracy threshold. The ground control station outputs corresponding parameters based on the pre-selected geometric shape type (cubic prism, sphere, or ellipsoid). For a sphere, the exhaustive search method can be simplified to: calculating the distance from the coordinates of all edge points to the center of the point set, taking the maximum distance as the radius, and using the center of the point set as the center of the sphere. For an ellipsoid, principal component analysis can be used to obtain the three principal axis directions first, and then the minimum semi-axis length can be searched along each principal axis direction.

[0036] The pre-stored field strength gradient magnitude threshold is a critical value used to determine whether the comprehensive gradient value of a grid cell has reached an abnormal level. It is the statistical upper limit of the physical field gradient magnitude of a normal line. Under the condition that the distribution network line is healthy and fault-free, the first group of UAVs is controlled to fly along the same cruise path multiple times to collect and calculate a large amount of comprehensive gradient magnitude data under normal conditions. Statistical analysis is performed on these data to calculate their mean and standard deviation. The mean plus three times the standard deviation is used as the field strength gradient magnitude threshold, which is pre-stored in the configuration file of the ground control station. The pre-stored field strength gradient magnitude threshold can be set according to the historical data of different line sections. The preset threshold is the same as the aforementioned field strength gradient magnitude threshold. Both have the same meaning and are used to distinguish grid cells into candidate anomalies and normal ones. The preset minimum area threshold refers to the minimum number of grid cells required for a connected region to be identified as a preliminary anomaly region. This means excluding false anomaly regions caused by a single or a few isolated noise points. The pre-stored field strength gradient magnitude threshold is determined based on the continuous characteristics of the physical field spatial distribution and the size of the grid cells (i.e., the preset first spatial step size). It is the number of grid cells that can cover the minimum spatial range affected by a typical physical field anomaly source. It is usually set to four or eight to ensure that the actual spatial range corresponding to this area is greater than the relevant length of the sensor measurement noise. The specific value is determined by statistical analysis of the actual size of the anomaly region in multiple field tests and is pre-stored in the ground control station. The preset consistency ratio threshold refers to the lower limit of the proportion of grid cells in a preliminary anomaly area where the angle between the field strength gradient direction and the regional reference direction is less than the direction deviation threshold. It is used to determine whether the gradient direction in the area has statistical consistency. The preset consistency ratio threshold is based on the statistical analysis of the physical field gradient direction distribution characteristics generated by a large number of real distribution network faults. It takes an empirical value that can distinguish between regions with single-direction gradient changes (such as physical field linear boundaries) and random direction noise regions. It is usually set to 70%. This value is preset during system initialization and can be adjusted according to the line environment. The preset convergence ratio threshold refers to the lower limit of the proportion of edge grid cells pointing inward within a candidate anomaly gradient region. It is used to determine whether the gradient direction shows a feature of converging from the surroundings to the center, thereby confirming that the region corresponds to the real physical field anomaly source. The preset convergence ratio threshold is obtained by simulating different types of distribution network faults (such as partial discharge and insulator contamination) in the laboratory, measuring the distribution of physical field gradient directions around the fault point, and statistically analyzing the proportion of edge grid cells whose gradient direction points to the fault center. The minimum effective value of this proportion is taken as the convergence ratio threshold. This value is pre-stored in the ground control station and can be configured according to the identification sensitivity requirements of different fault types.

[0037] S5. Based on the spatial coordinates of the suspected potential hazard points, dispatch a second group of drones equipped with visual acquisition sensors to fly to the target area and acquire multi-angle, multi-spectral images of the suspected potential hazard points and their surroundings to obtain a set of visual images. Furthermore, in S5, after determining the spatial coordinates of the first suspected hidden danger point, the ground control station selects a second group of drones equipped with visual acquisition sensors from the drone library; the ground control station sends take-off commands to each drone in the second group and uploads the spatial coordinates of the first suspected hidden danger point as the target waypoint; The second group of drones took off to a safe altitude according to their own takeoff procedure, and then flew to a hovering point near the spatial coordinates of the first suspected hidden danger point in a fast flight mode with a speed higher than that of the first group of drones. The hovering point is located at the first hovering distance directly in front of the spatial coordinates of the first suspected hidden danger point, and the height of the hovering point is consistent with the height of the spatial coordinates of the first suspected hidden danger point; Once all the drones in the second group have reached their respective hovering points, each drone sends a ready signal to the ground control station. After receiving the readiness signals from all UAVs, the ground control station sends a low-speed data acquisition command to the second group of UAVs, limiting the maximum flight speed of each UAV to one-third of the cruising speed of the first group of UAVs. The ground control station uploads the speed limit parameters to the flight controller of each UAV via a wireless link. The flight controller then automatically controls the actual flight speed within the speed limit value during subsequent flights. The ground control station generates a detailed data collection path for each UAV based on the pre-selected trajectory type. The trajectory type is either a spiral flight trajectory or a figure-eight flight trajectory. If it is a spiral flight trajectory, the ground control station takes the spatial coordinates of the first suspected hidden danger point as the center, uses the starting radius as the horizontal distance between the spiral starting point and the center point, uses the radius decay step as the amount of radius reduction for each flight circle, and uses the vertical ascent step as the amount of altitude increase for each flight circle. Starting from the starting point, a series of continuous spatial points are generated in a counterclockwise or clockwise direction. The spatial points are connected in sequence to form a three-dimensional spiral line from the outside to the inside, from the bottom to the top, or from the top to the bottom. If the flight path is a figure-eight shape, the ground control station uses the spatial coordinates of the first suspected hidden danger point as the center, the first horizontal span as the width of a single loop, and the second horizontal span as the interval between two loops to generate a closed curve in the horizontal plane with two tangent circular loops alternating. The curve passes through the offset point directly above or below the center point. The ground control station discretizes the generated trajectory into a series of waypoints. Each waypoint contains three-dimensional spatial coordinates and the desired nose direction, which always points to the spatial coordinates of the first suspected potential hazard point. The ground control station uploads the waypoint sequence to the flight controllers of the second group of UAVs. The second group of drones departed from the hovering point and flew in sequence according to the received waypoint sequence, while keeping the nose of the drone always pointing to the spatial coordinates of the first suspected potential hazard point; During flight, the visual acquisition sensors on each drone continuously capture images according to the image acquisition frequency; The visual acquisition sensor includes a high-resolution visible light camera and a multispectral camera. The multispectral camera can simultaneously acquire images in the infrared, near-infrared, or ultraviolet bands. While acquiring images, each drone binds and stores the shooting time corresponding to each image, the drone's own three-dimensional spatial coordinates, shooting attitude angle, and camera parameters as metadata along with the image data. Shooting attitude angles include pitch angle, roll angle, and yaw angle; camera parameters include focal length, aperture, and exposure time. After completing the entire flight path, each drone in the second group automatically flew back to the return point and landed. During or after landing, each drone uploaded all the images it had collected and their corresponding metadata to the ground control station via a wireless data link. The ground control station receives image data from multiple drones, classifies and organizes it according to shooting time, drone number and shooting location to form a visual image set. The ground control station temporarily stores the visual image set in local memory and creates an index to associate each image with the corresponding spatial coordinates and shooting parameters. If a drone malfunctions during the data acquisition process or the image quality does not meet the clarity requirements, the ground control station will dispatch a backup drone to re-execute the data acquisition task of the drone until images from all perspectives are successfully acquired.

[0038] S6. Compare the visual image set with the pre-stored standard state images of suspected hazard points, identify physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, determine the suspected hazard point as the final hazard point and generate a hazard report containing location and image information.

[0039] Furthermore, in S6, the ground control station acquires the standard status image of the first suspected hidden danger point that has been stored in advance; the standard status image is a multi-angle, multi-spectral image of the power distribution network line taken in advance by a drone under the healthy condition, and is stored in the database of the ground control station according to spatial coordinates and location type. Based on the spatial coordinates of the first suspected hidden danger point, the ground control station retrieves the standard state image set corresponding to the spatial coordinates from the database. The standard state image set and the visual image set have the same parameter settings in terms of shooting angle, spectral band and imaging resolution. The ground control station preprocesses each image in the visual imagery set, including converting the images from their original format to a uniform pixel matrix, converting the images to grayscale, and performing histogram equalization on the images. The ground control station extracts edge density features, texture statistics features, and local binary pattern features for each image. The edge density features are obtained by calculating the ratio of the number of edge points where pixel values ​​change abruptly to the total number of pixels in the image. Texture statistical features are obtained by statistically analyzing the frequency distribution of pixel values ​​in the image and the difference distribution between adjacent pixels. Local binary pattern features are obtained by comparing the grayscale relationship between each pixel and its surrounding neighboring pixels, encoding it into a binary value, and then statistically analyzing its histogram. The ground control station arranges the feature values ​​extracted from all images under the same spectral band in a fixed order to form a sub-vector. Then, it splices the sub-vectors of different spectral bands together in a preset spectral order to form the first feature vector. The first feature vector is then normalized so that the values ​​of all dimensions fall between zero and one. Perform the same preprocessing and feature extraction operations on each image in the pre-stored standard state image set to obtain the second feature vector, and perform the same normalization process on the second feature vector so that the value of each dimension of the second feature vector also falls between zero and one. The first feature vector and the second feature vector have the same number of dimensions and the feature meaning of each dimension corresponds one-to-one. The ground control station calculates the cosine similarity between the first and second eigenvectors as follows: multiply the value of each dimension in the first eigenvector by the corresponding value in the second eigenvector to obtain the product value of each dimension, add the product values ​​of all dimensions to obtain the sum of the numerators, calculate the sum of the squares of the values ​​of each dimension in the first eigenvector and take the square root to obtain the magnitude of the first vector, calculate the sum of the squares of the values ​​of each dimension in the second eigenvector and take the square root to obtain the magnitude of the second vector, multiply the magnitude of the first vector by the magnitude of the second vector to obtain the product of the denominators, and divide the sum of the numerators by the product of the denominators to obtain the cosine similarity value. The ground control station obtains a pre-stored matching threshold and compares the cosine similarity value with the matching threshold. If the cosine similarity is greater than or equal to the matching threshold, it is determined that the first suspected hazard point does not exhibit physical deformation or foreign object attachment characteristics. The first suspected hazard point is marked as a false alarm and removed from the list to be reviewed. If the cosine similarity is lower than the matching threshold, it is determined that physical deformation or foreign object attachment characteristics exist. The first suspected hazard point is identified as the final hazard point. The ground control station generates a hazard report, which includes the spatial coordinates of the final hazard point, the hazard type, the similarity value, multiple images of the defect selected from the visual image set, and a comparison image of the corresponding viewpoint extracted from the standard state image. The ground control station stores the hazard report in a local database and sends it to the terminal device of the maintenance personnel via a wireless network.

[0040] like Figure 2 This embodiment provides: a rapid location system for potential hazards in power distribution networks based on drone collaboration, including: Cruise path generation module: acquires the initial spatial location information of the target power distribution network line, and generates the cruise path of the first group of UAVs based on the initial spatial location information; Feature information acquisition module: controls the first group of drones to fly along the cruise path, and during the flight, collects physical field feature information in real time within a preset space range around the power distribution line through the sensor array mounted on each drone. The physical field feature information includes air micro-flow disturbance data and spatial electrostatic field distortion data. Gradient map construction module: Based on the physical feature information and corresponding spatial coordinates collected by each UAV in the first group, construct the physical field distribution gradient map of the inspection path coverage area; Hazard identification module: Identifies abnormal gradient regions in the physical field distribution gradient map, and determines the spatial coordinates of suspected hazard points based on the center point of the abnormal gradient region; Hazardous point data acquisition module: Based on the spatial coordinates of the suspected hazard points, a second group of UAVs equipped with visual acquisition sensors is dispatched to fly to the target area to acquire multi-angle, multi-spectral images of the suspected hazard points and their surroundings, thereby obtaining a set of visual images; Hazard location module: Compares the set of visual images with the pre-stored standard state images of suspected hazard points, identifies physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, the suspected hazard point is determined as the final hazard point and a hazard report containing location and image information is generated.

[0041] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0043] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0046] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0047] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for rapid location of potential hazards in power distribution networks based on UAV collaboration, characterized in that, Includes the following steps: S1. Obtain the initial spatial location information of the target power distribution network line, and generate the cruise path of the first group of UAVs based on the initial spatial location information; S2. Control the first group of drones to fly along the cruise path, and during the flight, collect physical field characteristic information in a preset space range around the power distribution line in real time through the sensor array mounted on each drone. The physical field characteristic information includes air micro-flow disturbance data and spatial electrostatic field distortion data. The ground control station receives data points from all UAVs in the first group. Each data point contains three-dimensional spatial coordinates, air microfluidic disturbance data, and spatial electrostatic field distortion data. The air microfluidic disturbance data includes airflow direction, airflow speed, and turbulence information, while the spatial electrostatic field distortion data includes spatial electric field strength and distortion waveform information. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding air microfluidic disturbance data from the data points. The ground control station divides the entire cruise path coverage area into a uniform three-dimensional grid according to a preset first spatial step size. Each grid cell is a cube voxel. For each grid cell, the ground control station collects the air microfluidic disturbance data of all data points falling into the grid cell. The arithmetic mean of the collected airflow velocity values ​​is taken as the representative airflow velocity of the grid cell. The representative airflow direction value is obtained by vector averaging the collected airflow direction value. The arithmetic mean of the collected turbulence values ​​is taken as the representative turbulence of the grid cell. For blank grid cells where no data points fall, the ground control station uses a spatial interpolation method to search outward from the blank grid cell with the pre-set first search radius for the already assigned grid cells. The representative airflow velocity, representative airflow direction, and representative turbulence of these already assigned grid cells are estimated by weighting inversely proportionally to the distance and then filled into the blank grid cell. S3. Based on the physical field feature information and corresponding spatial coordinates collected by each UAV in the first group, construct a physical field distribution gradient map of the inspection path coverage area; After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional hybrid field map, which is used as the first microfluidic field map. The ground control station extracts the three-dimensional spatial coordinates of all data points and the corresponding spatial electrostatic field distortion data from the data points. The ground control station uses the same three-dimensional grid division method as the one used to establish the first microflow field map to divide the entire cruise path coverage area into three-dimensional grids of the same size. For each grid cell, the ground control station collects the spatial electric field intensity values ​​of all data points falling within the grid cell, and takes the arithmetic mean as the representative electric field intensity of the grid cell. At the same time, it collects the distorted waveform information of all data points falling within the grid cell, and takes the arithmetic mean of the number of zero crossings, the number of peaks, and the steepness of the waveform in the distorted waveform information to obtain a set of representative distortion characteristic values. For blank grid cells, the ground control station uses the same spatial interpolation method as the first microflow field map to search for neighboring already assigned grid cells with a preset second search radius, and calculates the representative electric field intensity and representative distortion characteristic value of the blank grid cell by distance inverse weighted interpolation. After assigning values ​​to all grid cells, the ground control station obtains a three-dimensional scalar field map, which is used as the first electric field map. The ground control station spatially registers the first microfluidic field map and the first electric field map: the ground control station checks whether the grid origin, grid step size and grid range of the two field maps are completely consistent. If there is any inconsistency, the field map with the larger coverage area is used as the reference, and the field map with the smaller coverage area is extended by boundary. The grid cells of the extended part are filled by copying the values ​​of the neighboring grid cells outward to ensure that the two field maps have the same grid topology. The ground control station creates a new 3D grid with the same grid topology as the registered field map, serving as the framework for the integrated physical field distribution gradient map; For each grid cell in the new three-dimensional grid, the ground control station reads the airflow velocity, airflow direction and turbulence intensity of the grid cell from the first microflow field map, and reads the spatial electric field intensity and distortion characteristic value of the grid cell from the first electric field map. The data are combined into a multi-dimensional feature vector and stored in the grid cell. The ground control station calculates the comprehensive gradient value of each grid cell: first, it calculates the sum of the absolute values ​​of the differences in airflow velocity between the grid cell and its multiple neighboring grid cells to obtain the first gradient component; The second gradient component is obtained by summing the absolute values ​​of the differences in spatial electric field intensity between grid cells and their adjacent grid cells. Finally, the first gradient component and the second gradient component are weighted and summed according to the preset first weight and second weight to obtain the comprehensive gradient value of the grid cell. The combined gradient values ​​of all grid cells are arranged according to spatial coordinates to form a three-dimensional gradient value distribution map, which is then used as a multi-dimensional combined physical field distribution gradient map. S4. Identify abnormal gradient regions in the physical field distribution gradient map, and determine the spatial coordinates of suspected hidden danger points based on the center point of the abnormal gradient region. S5. Based on the spatial coordinates of the suspected potential hazard points, dispatch a second group of drones equipped with visual acquisition sensors to fly to the target area and acquire multi-angle, multi-spectral images of the suspected potential hazard points and their surroundings to obtain a set of visual images. S6. Compare the visual image set with the pre-stored standard state images of suspected hazard points, identify physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, determine the suspected hazard point as the final hazard point and generate a hazard report containing location and image information.

2. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration as described in claim 1, characterized in that, In S1, the tower coordinates, conductor sag point coordinates, and line turning point coordinates of the target distribution network line are read from the ground control station and the coordinate data are combined into a discrete point sequence describing the spatial direction of the entire line. The ground control station sorts the discrete point sequence according to the actual connection order of the line, forming a continuous spatial polyline, and uses the spatial polyline as the initial spatial location information of the target distribution network line; Based on the initial spatial location information of the conductor, the ground control station extends a preset first horizontal safety distance to both sides along the direction of the line extension, with the conductor as the center line. At the same time, it extends a preset first vertical safety distance upward from the highest point of the conductor and a preset second vertical safety distance downward from the lowest point of the conductor, forming a three-dimensional rectangular columnar corridor. The ground control station extracts the coordinates of the two endpoints of each traverse segment from the initial spatial position information and calculates the spatial straight line equation of each traverse segment; Along the direction of the conductor, at the third vertical offset distance directly above the conductor, generate an aerial trajectory line that is completely parallel to and equidistant from the spatial direction of the conductor. The third vertical offset distance is less than the first vertical safety distance and is a fixed value. At the turning point of two adjacent conductor segments, a spatial arc transition method is used to smoothly connect the aerial trajectory line to form a continuous and smooth three-dimensional spatial curve. The three-dimensional spatial curve is used as the reference cruise line of the first group of UAVs. The ground control station obtains the formation position number of each drone in the first group of drones. The formation position includes the following position along the route and the staggered position to the left and right perpendicular to the route. Using each point on the baseline cruise line as a formation reference point, and based on the formation position number of each UAV, the offset vector of each UAV relative to the formation reference point is calculated in a plane perpendicular to the baseline cruise line according to the preset first lateral spacing and first longitudinal spacing. The offset vector is superimposed onto the spatial coordinates of the formation reference point to obtain the expected spatial coordinates of each UAV at each formation reference point. Connect all desired spatial coordinates in chronological order to form the cruise path of each drone, and combine the cruise paths of all drones to form the formation cruise path of the first group of drones. The ground control station uploads the cruise path of each UAV in the form of waypoint sequence to the onboard computer of the corresponding UAV via a wireless data link; Each waypoint contains three-dimensional spatial coordinates, the expected arrival time, and the allowable position error range. After receiving the waypoint sequence, the UAV's flight controller executes autonomous flight tasks according to the waypoint order and corrects its own position in real time during the flight to ensure that the deviation between the actual flight trajectory and the preset cruise path is always controlled within the allowable position error range.

3. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration as described in claim 1, characterized in that, In S2, after the ground control station completes the loading of the formation cruise path, take-off and formation flight commands are sent simultaneously to the first group of UAVs via a wireless data link. Each drone autonomously flies to the starting waypoint according to its pre-stored waypoint sequence and adjusts its attitude so that the nose is tangential to the cruise path. Once all UAVs have reached the starting waypoint and formed the preset formation, the ground control station issues a cruise start command. Each UAV maintains formation flight along its own cruise path, while the onboard computer sends a start data acquisition command to the onboard sensor array. The sensor array includes a biomimetic microfluidic sensing array and an electrostatic field disturbance sensing probe. Both are started synchronously according to a preset first sampling frequency to begin real-time sensing of physical field changes within a preset spatial range around the drone. The biomimetic microfluidic sensing array consists of multiple miniature airflow sensing units arranged in a preset geometric layout. Each sensing unit converts the flow of air molecules across its surface into an electrical signal output. When there is partial discharge or overheating in the power distribution network line, the discharge or high temperature causes the surrounding air molecules to ionize and generate directional flow, forming air micro-flow disturbance. During the flight of the drone, each sensitive unit in the biomimetic microfluidic sensing array independently outputs a voltage value that is proportional to the airflow velocity; The onboard computer reads the output voltage values ​​of all sensitive units in the biomimetic microfluidic sensing array at the same time, forming a multi-dimensional raw data vector; The computer uses the average value of the output voltage of all sensitive units as a reference value, and subtracts the reference value from the output voltage of each sensitive unit to obtain the differential output value of each sensitive unit. The computer sorts the differential output values ​​according to their coordinate positions based on the geometric coordinates of each sensitive unit in the array, forming a spatial difference matrix.

4. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration according to claim 3, characterized in that, The computer finds the two sensitive cells with the largest difference value in the spatial difference matrix, calculates the direction of the line connecting the two sensitive cells, and uses the direction of the line as the overall direction of the airflow. The maximum difference value is multiplied by a preset first scaling factor to obtain the instantaneous velocity value of the airflow. The computer statistically analyzes the changes in airflow direction and velocity over time within a preset first time window. The changes are then divided by a preset second scaling factor to obtain the turbulence information of the airflow. The electrostatic field disturbance sensing probe consists of one or more metal electrodes with the electrode surface exposed to the air. When a drone flies over an area with insulator contamination, broken cable strands, or damaged wires, the potential hazard point generates local electric field distortion, causing a change in the electrostatic field strength in the surrounding space. The electrostatic field disturbance sensing probe on the UAV senses the spatial potential through electrodes, converts the sensed potential value into an analog voltage signal output, and the onboard computer continuously collects the analog voltage signal at a preset second sampling frequency to obtain a potential sequence that changes over time. The airborne computer acquires the potential change sequence within a preset second time window, takes the difference between the maximum and minimum values ​​in the sequence as the potential peak value, and divides the potential peak value by a preset third scaling coefficient to obtain the spatial electric field intensity value within the time window. The computer plots each sampled value in the potential change sequence into a waveform in chronological order, forming a potential change waveform. The computer performs baseline drift removal on the waveform, takes the average value of the entire waveform as the DC component, and subtracts the DC component from each sampled value to obtain the AC fluctuation component. The computer extracts the number of zero-crossing points, the number of peaks, and the steepness of the rising and falling edges of the AC wave component, and combines these features into distorted waveform information.

5. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration according to claim 1, characterized in that, In S4, the ground control station reads a multi-dimensional integrated physical field distribution gradient map. The gradient map consists of multiple grid cells arranged in a three-dimensional grid, and each grid cell stores a comprehensive gradient value. The ground control station traverses all grid cells and uses the comprehensive gradient value stored in each grid cell as the field strength gradient magnitude value of the grid cell; The ground control station determines the field strength gradient direction based on the trend of airflow velocity change between each grid cell and its neighboring grid cells: find the neighboring grid cell with the largest comprehensive gradient value in the six directions around each grid cell, calculate the direction vector from the current grid cell to the neighboring cell, normalize the direction vector and use it as the field strength gradient direction of the current grid cell. If the comprehensive gradient values ​​of multiple neighboring cells are equal and all are the maximum, then the composite direction of multiple direction vectors is used as the field strength gradient direction. The ground control station obtains the pre-stored field strength gradient magnitude threshold, traverses all grid cells again, marks grid cells with field strength gradient magnitude greater than or equal to the preset threshold as candidate abnormal grid cells, and marks grid cells with field strength gradient magnitude less than the preset threshold as normal grid cells. The ground control station uses a connected component analysis algorithm to merge all adjacent candidate abnormal grid cells into a connected component. For each connected region, the ground control station checks whether the total number of grid cells contained in the connected region is greater than the preset minimum region area threshold. If it is greater, the connected region is marked as a preliminary abnormal region and the spatial coordinate list of all grid cells contained in the preliminary abnormal region is recorded. If it is less than, the connected region is excluded. For each initially selected anomaly region, the ground control station obtains the field intensity gradient direction of all grid cells in the region, calculates the average direction of all field intensity gradient directions in the region, and uses the average direction as the region's reference direction. For each grid cell within the region, the ground control station calculates the angle between the field strength gradient direction and the regional reference direction; If the included angle of grid cells in a region that exceed the preset consistency ratio threshold is less than the preset direction deviation threshold, then the initially selected abnormal region is marked as a candidate abnormal gradient region. The ground control station further judges the convergence characteristics of the candidate abnormal gradient region: selects all grid cells located at the edge of the region within the candidate abnormal gradient region, calculates whether the gradient direction of each edge grid cell points into the region, and if the edge grid cells in the region that exceed the preset convergence ratio threshold meet the judgment condition of pointing inward, then the candidate abnormal gradient region is determined as an abnormal gradient region. For each connected region identified as an anomalous gradient region, the ground control station extracts the coordinates of the center point of all grid cells on the region boundary as the coordinates of the edge point. The boundary grid cell is a grid cell in the anomalous gradient region that has at least one adjacent grid cell that does not belong to the anomalous gradient region. The ground control station collects the coordinates of the center points of all boundary grid cells to form a set of edge point coordinates. The ground control station uses the rotating caliper method or the exhaustive search method to find the smallest volume geometry in three-dimensional space that can include all edge point coordinates. The smallest volume geometry is pre-selected as a cuboid, sphere or ellipsoid. The ground control station calculates the coordinates of the geometric center point based on the type of the smallest circumscribed geometric figure: if the smallest circumscribed geometric figure is a cuboid, then the average of the minimum and maximum values ​​of the cuboid in the three coordinate axes is taken, and the three-dimensional coordinates formed by the three average values ​​are used as the coordinates of the geometric center point of the cuboid. If the smallest circumscribed geometric figure is a sphere, then the coordinates of the center of the sphere are used as the coordinates of the geometric center point; if the smallest circumscribed geometric figure is an ellipsoid, then the center points corresponding to the lengths of the three semi-axis of the ellipsoid are calculated as the coordinates of the geometric center point, and the ground control station determines the coordinates of the geometric center point as the spatial coordinates of the first suspected hidden danger point. The ground control station records the spatial coordinates of the first suspected hidden danger point in memory in the form of three-dimensional coordinates, and stores them in association with the identifier, size and average gradient magnitude of the abnormal gradient region corresponding to the coordinates. If multiple abnormal gradient regions are identified within the coverage area of ​​the same cruise path, the ground control station will sequentially perform the above steps for each abnormal gradient region to generate multiple spatial coordinates of first suspected hidden danger points.

6. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration according to claim 1, characterized in that, In step S5, after determining the spatial coordinates of the first suspected potential hazard point, the ground control station selects a second group of drones equipped with visual acquisition sensors from the drone library; the ground control station sends take-off commands to each drone in the second group and uploads the spatial coordinates of the first suspected potential hazard point as the target waypoint; The second group of drones took off to a safe altitude according to their own takeoff procedure, and then flew to a hovering point near the spatial coordinates of the first suspected hidden danger point in a fast flight mode with a speed higher than that of the first group of drones. The hovering point is located at the first hovering distance directly in front of the spatial coordinates of the first suspected hidden danger point, and the height of the hovering point is consistent with the height of the spatial coordinates of the first suspected hidden danger point; Once all the drones in the second group have reached their respective hovering points, each drone sends a ready signal to the ground control station. After receiving the readiness signals from all UAVs, the ground control station sends a low-speed data acquisition command to the second group of UAVs, limiting the maximum flight speed of each UAV to one-third of the cruising speed of the first group of UAVs. The ground control station uploads the speed limit parameters to the flight controller of each UAV via a wireless link. The flight controller then automatically controls the actual flight speed within the speed limit value during subsequent flights. The ground control station generates a detailed data collection path for each UAV based on the pre-selected trajectory type. The trajectory type is either a spiral flight trajectory or a figure-eight flight trajectory. If it is a spiral flight trajectory, the ground control station takes the spatial coordinates of the first suspected hidden danger point as the center, uses the starting radius as the horizontal distance between the spiral starting point and the center point, uses the radius decay step as the amount of radius reduction for each flight circle, and uses the vertical ascent step as the amount of altitude increase for each flight circle. Starting from the starting point, a series of continuous spatial points are generated in a counterclockwise or clockwise direction. The spatial points are connected in sequence to form a three-dimensional spiral line from the outside to the inside, from the bottom to the top, or from the top to the bottom. If the flight path is a figure-eight shape, the ground control station uses the spatial coordinates of the first suspected hidden danger point as the center, the first horizontal span as the width of a single loop, and the second horizontal span as the interval between two loops to generate a closed curve in the horizontal plane with two tangent circular loops alternating. The curve passes through the offset point directly above or below the center point. The ground control station discretizes the generated trajectory into a series of waypoints. Each waypoint contains three-dimensional spatial coordinates and the desired nose direction, which always points to the spatial coordinates of the first suspected potential hazard point. The ground control station uploads the waypoint sequence to the flight controllers of the second group of UAVs. The second group of drones departed from the hovering point and flew in sequence according to the received waypoint sequence, while keeping the nose of the drone always pointing to the spatial coordinates of the first suspected potential hazard point; During flight, the visual acquisition sensors on each drone continuously capture images according to the image acquisition frequency; The visual acquisition sensor includes a high-resolution visible light camera and a multispectral camera. The multispectral camera can simultaneously acquire images in the infrared, near-infrared, or ultraviolet bands. While acquiring images, each drone binds and stores the shooting time corresponding to each image, the drone's own three-dimensional spatial coordinates, shooting attitude angle, and camera parameters as metadata along with the image data. Shooting attitude angles include pitch angle, roll angle, and yaw angle; camera parameters include focal length, aperture, and exposure time. After completing the entire flight path, each drone in the second group automatically flew back to the return point and landed. During or after landing, each drone uploaded all the images it had collected and their corresponding metadata to the ground control station via a wireless data link. The ground control station receives image data from multiple drones, classifies and organizes it according to shooting time, drone number and shooting location to form a visual image set. The ground control station temporarily stores the visual image set in local memory and creates an index to associate each image with the corresponding spatial coordinates and shooting parameters. If a drone malfunctions during the data acquisition process or the image quality does not meet the clarity requirements, the ground control station will dispatch a backup drone to re-execute the data acquisition task of the drone until images from all perspectives are successfully acquired.

7. The method for rapid location of potential hazards in power distribution networks based on UAV collaboration according to claim 1, characterized in that, In step S6, the ground control station acquires the standard status image of the first suspected hidden danger point that has been stored in advance. The standard status image is a multi-angle, multi-spectral image of the power distribution network line taken in advance by a drone when the line is in good condition. It is classified and stored in the database of the ground control station according to spatial coordinates and location type. Based on the spatial coordinates of the first suspected hidden danger point, the ground control station retrieves the standard state image set corresponding to the spatial coordinates from the database. The standard state image set and the visual image set have the same parameter settings in terms of shooting angle, spectral band and imaging resolution. The ground control station preprocesses each image in the visual imagery set, including converting the images from their original format to a uniform pixel matrix, converting the images to grayscale, and performing histogram equalization on the images. The ground control station extracts edge density features, texture statistics features, and local binary pattern features for each image. The edge density features are obtained by calculating the ratio of the number of edge points where pixel values ​​change abruptly to the total number of pixels in the image. Texture statistical features are obtained by statistically analyzing the frequency distribution of pixel values ​​in the image and the difference distribution between adjacent pixels. Local binary pattern features are obtained by comparing the grayscale relationship between each pixel and its surrounding neighboring pixels, encoding it into a binary value, and then statistically analyzing its histogram. The ground control station arranges the feature values ​​extracted from all images under the same spectral band in a fixed order to form a sub-vector. Then, it splices the sub-vectors of different spectral bands together in a preset spectral order to form the first feature vector. The first feature vector is then normalized so that the values ​​of all dimensions fall between zero and one. Perform the same preprocessing and feature extraction operations on each image in the pre-stored standard state image set to obtain the second feature vector, and perform the same normalization process on the second feature vector so that the value of each dimension of the second feature vector also falls between zero and one. The first feature vector and the second feature vector have the same number of dimensions and the feature meaning of each dimension corresponds one-to-one. The ground control station calculates the cosine similarity between the first and second eigenvectors as follows: multiply the value of each dimension in the first eigenvector by the corresponding value in the second eigenvector to obtain the product value of each dimension, add the product values ​​of all dimensions to obtain the sum of the numerators, calculate the sum of the squares of the values ​​of each dimension in the first eigenvector and take the square root to obtain the magnitude of the first vector, calculate the sum of the squares of the values ​​of each dimension in the second eigenvector and take the square root to obtain the magnitude of the second vector, multiply the magnitude of the first vector by the magnitude of the second vector to obtain the product of the denominators, and divide the sum of the numerators by the product of the denominators to obtain the cosine similarity value. The ground control station obtains a pre-stored matching threshold and compares the cosine similarity value with the matching threshold. If the cosine similarity is greater than or equal to the matching threshold, it is determined that the first suspected hazard point does not exhibit physical deformation or foreign object attachment characteristics. The first suspected hazard point is marked as a false alarm and removed from the list to be reviewed. If the cosine similarity is lower than the matching threshold, it is determined that physical deformation or foreign object attachment characteristics exist. The first suspected hazard point is identified as the final hazard point. The ground control station generates a hazard report, which includes the spatial coordinates of the final hazard point, the hazard type, the similarity value, multiple images of the defect selected from the visual image set, and a comparison image of the corresponding viewpoint extracted from the standard state image. The ground control station stores the hazard report in a local database and sends it to the terminal device of the maintenance personnel via a wireless network.

8. A rapid location system for potential hazards in a power distribution network based on UAV collaboration is used to execute the rapid location method for potential hazards in a power distribution network based on UAV collaboration as described in any one of claims 1-7, characterized in that, include: Cruise path generation module: acquires the initial spatial location information of the target power distribution network line, and generates the cruise path of the first group of UAVs based on the initial spatial location information; Feature information acquisition module: controls the first group of drones to fly along the cruise path, and during the flight, collects physical field feature information in real time within a preset space range around the power distribution line through the sensor array mounted on each drone. The physical field feature information includes air micro-flow disturbance data and spatial electrostatic field distortion data. Gradient map construction module: Based on the physical feature information and corresponding spatial coordinates collected by each UAV in the first group, construct the physical field distribution gradient map of the inspection path coverage area; Hazard identification module: Identifies abnormal gradient regions in the physical field distribution gradient map, and determines the spatial coordinates of suspected hazard points based on the center point of the abnormal gradient region; Hazardous point data acquisition module: Based on the spatial coordinates of the suspected hazard points, a second group of UAVs equipped with visual acquisition sensors is dispatched to fly to the target area to acquire multi-angle, multi-spectral images of the suspected hazard points and their surroundings, thereby obtaining a set of visual images; Hazard location module: Compares the set of visual images with the pre-stored standard state images of suspected hazard points, identifies physical deformation or foreign object attachment characteristics in the images, and if there is a difference in the comparison results, the suspected hazard point is determined as the final hazard point and a hazard report containing location and image information is generated.

Images