An ice thickness monitoring method based on a multi-functional sensor of a power transmission line

Abstract

The application discloses a kind of ice thickness monitoring method based on power transmission line multifunctional sensor, it is related to power transmission line technical field.The method steps include: the power transmission line wire of no icing state is shot, the obtained image is converted into three-dimensional point cloud and is carried out cylinder fitting, obtains no icing wire diameter;Ice-covered wire is shot and is converted into ice-covered wire point cloud, gets wire fusion point cloud by adaptive weighted fusion method, calculates ice equivalent diameter by poisson surface reconstruction method, obtains first ice thickness in combination with no icing wire diameter;The height sequence of ice-covered wire is obtained to calculate actual sag variation, and the theoretical sag variation is obtained in combination with the first ice thickness and the no icing wire diameter;Difference index is calculated by comparing the two, the first ice thickness is corrected by the index, and finally the second ice thickness is obtained.

Description

Technical Field

[0001] This invention relates to the field of power transmission line technology, specifically to a method for monitoring icing thickness based on a multifunctional sensor for power transmission lines. Background Technology

[0002] With the continuous expansion of the power grid and the increasing frequency of extreme weather events, icing on transmission lines has become a core hidden danger threatening the safe and stable operation of the power grid. Icing on conductors significantly increases the vertical load on the lines, causing abnormal sag and potentially leading to serious accidents such as phase-to-phase flashover, conductor breakage, and even tower collapse. Therefore, accurate and real-time monitoring of the icing thickness on transmission lines is a key technical requirement for ensuring power supply reliability.

[0003] In traditional ice thickness detection, there is a multi-view point cloud fusion measurement method. This method deploys multiple cameras to capture images of the icy conductor from different angles, converts the collected images into three-dimensional point cloud data, then uses a fixed weight allocation strategy to fuse the multi-view point clouds, and then obtains a three-dimensional model of the icy conductor through surface reconstruction. Finally, the ice thickness is calculated by combining the reference diameter of the uniced conductor.

[0004] However, point cloud data from different perspectives have inherent differences in density, registration accuracy, and visual coverage. Traditional methods use fixed-weight fusion, which cannot dynamically adjust the weight allocation according to the real-time quality of point clouds from each perspective. As a result, the fused point cloud is difficult to accurately reproduce the complete outline of the icing conductor, ultimately affecting the accuracy of icing thickness measurement. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for monitoring icing thickness based on a multifunctional sensor for power transmission lines, thereby resolving the problems existing in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring icing thickness based on a multifunctional sensor for transmission lines, comprising the following steps: Step S1: Take a picture of the conductor in the transmission line in the state of no ice to obtain an image of the conductor without ice; perform three-dimensional point cloud transformation on the image of the conductor without ice and perform cylinder fitting to obtain the diameter of the conductor without ice. Step S2: Take a picture of the ice-covered conductor to obtain an image of the ice-covered conductor, and convert it into an ice-covered conductor point cloud; perform three-dimensional point cloud fusion on the ice-covered conductor point cloud using an adaptive weighted fusion method to obtain a conductor fused point cloud; perform surface reconstruction on the conductor fused point cloud using the Poisson surface reconstruction method to calculate the ice-covered equivalent diameter; combine the ice-covered equivalent diameter and the diameter of the un-ice-covered conductor to calculate the first ice thickness. Step S3: Obtain the height sequence of the icy conductor, and calculate the actual sag change based on the icy conductor height sequence. Combine the first icing thickness and the diameter of the uniced conductor to calculate the theoretical sag change. Step S4: Calculate the difference index by comparing the theoretical sag change with the actual sag change; correct the first icing thickness using the difference index to obtain the second icing thickness.

[0007] Preferably, the step of photographing the conductors in the transmission line under ice-free conditions to obtain images of the ice-free conductors includes the following specific steps: During the non-icing period of the transmission line, the lowest point of sag of the conductor span between two towers is selected as the measurement point. An integrated monitoring device is installed at this measurement point, with the BeiDou positioning module fixed inside the device. Three sets of binocular cameras are arranged on the monitoring device: the first set of binocular cameras is installed at the bottom of the device; the second set of binocular cameras is installed at one end of the device along the conductor axis of the transmission line, with its optical axis perpendicular to the conductor axis in the horizontal plane and tilted downwards. The first set of cameras points to the bottom left area of ​​the conductor, covering the lower left side of the conductor; the second set of cameras is installed symmetrically with the second set of cameras, with the optical axis direction symmetrical to the second set of cameras, covering the lower right side of the conductor. During the non-icing period of the transmission line, the three sets of binocular cameras on the monitoring device simultaneously capture images of the non-iced conductor. The non-iced conductor images include the bottom binocular image pair captured by the first set of binocular cameras, the bottom left binocular image pair captured by the second set of binocular cameras, and the bottom right binocular image pair captured by the third set of binocular cameras.

[0008] Preferably, the step of converting the image of the ice-free conductor into a 3D point cloud and performing cylinder fitting to obtain the diameter of the ice-free conductor includes the following steps: Stereo correction is performed on the synchronously acquired bottom binocular image pairs, bottom left binocular image pairs, and bottom right binocular image pairs. For each pair of binocular images, the disparity between the left and right images in each pair is calculated. , , Represents the x-coordinate of the left image. Let x be the x-coordinate of the right image. Based on the focal length calibrated for the left camera in the binocular camera system and the baseline distance Distance between the principal point and the binocular camera, the pixel coordinates (u,v) of the left image are converted into three-dimensional points (X,Y,Z) in the binocular camera coordinate system:

[0009]

[0010]

[0011] Where X, Y, and Z are the three-dimensional coordinates of the point cloud. This indicates the focal length of the left camera in the X-direction of a binocular camera system. This indicates the focal length of the left camera in the Y direction within a binocular camera system. Here, denoted as parallax, Distance is the baseline distance between the two cameras, and u,v are the pixel coordinates of the left image. After converting pixel coordinates into three-dimensional coordinates, we obtain the three-dimensional point cloud data of each set of binocular cameras and filter out the point cloud clusters that represent the main body of the conductor. A cylindrical model is fitted to the point cloud clusters belonging to the main body of the conductor. The cylindrical model is represented as (C, r, axis), where C is a point on the axis of the cylinder, axis is the unit direction vector of the axis of the cylinder, and r is the radius of the cylinder. The optimal cylinder radius r is obtained by minimizing the sum of squared distances from the point cloud to the cylinder surface using the least squares method. 2r is then used as the reference diameter to obtain the reference diameter for the left visual view. and right visual reference diameter The formula for the sum of squares of the distances from the point cloud to the cylindrical surface is as follows:

[0012] in, Let be the three-dimensional coordinate vector of the j-th point in the point cloud, C be the point on the axis of the cylinder, axis be the unit direction vector of the axis of the cylinder, and r be the radius of the cylinder. The diameter of the icing-free conductor is calculated using the left and right visual reference diameters.

[0013] Preferably, the calculation of the diameter of the ice-free conductor using the left visual reference diameter and the right visual reference diameter specifically involves:

[0014] in, The diameter of the non-icing conductor. The diameter is the left visual reference diameter. The diameter is the right visual reference diameter.

[0015] Preferably, the step of performing three-dimensional point cloud fusion on the icing guideline point cloud using an adaptive weighted fusion method to obtain the fused guideline point cloud includes the following steps: Based on the rotation matrix and translation vector from three perspectives, the reconstructed left, right, and bottom point clouds from the left, right, and bottom binocular cameras are transformed to a unified world coordinate system:

[0016] in, The coordinates of a point in the world coordinate system. Let be the rotation matrix under the cam-th viewpoint. Let be the translation vector at the cam-th viewpoint. Let be the coordinates of the point in the cam-th viewpoint coordinate system; After converting the point cloud into the same world coordinate system, the iterative nearest point algorithm is used for point cloud registration; After point cloud registration, the point cloud quality evaluation index is calculated for each viewpoint. The calculation formula is as follows:

[0017] in, Let be the point cloud quality evaluation index from the cam-th viewpoint. Let be the point cloud density under the cam-th view. Let be the registration error under the cam-th viewpoint. Let be the visual coverage of the cam-th viewpoint. The point cloud density weighting coefficient. These are the registration error weighting coefficients. This is the visual coverage weighting coefficient; Arrange the point cloud quality evaluation indexes (mechanical energy) from three perspectives in ascending order: , It is the first value in the point cloud quality evaluation index. It is the second value in the point cloud quality evaluation index. The median is the third value in the point cloud quality evaluation index. : Calculate the upper quantile : Calculate the lower quantile : Calculate the quantile dispersion : ; Calculate the anomaly detection threshold : = * , This is the abnormal adjustment coefficient; Calculate the anomaly factor of the i-th viewpoint:

[0018] in, Let be the anomaly factor of the i-th viewpoint. The median of the point cloud quality evaluation index. This is the threshold for anomaly detection; Establish mutual support relationships among the three cameras, and measure the point cloud similarity for any two viewpoints:

[0019] in, For the point cloud similarity measure between cam1 and cam2, The effective number of point clouds for viewpoint cam1. This represents the number of points in the overlapping region. Relationship matrix construction:

[0020] in, The relationship index between cam1 and cam2 is... This is the consistency threshold; The support level for each perspective is:

[0021] in, For the support of the cam1 vision, is the abnormality factor for the second visual image; Based on the point cloud quality evaluation index, support, and preset baseline weights, calculate the view weight for each viewpoint:

[0022] in, Let be the view weight under the cam-th view. The preset baseline weights for the i-th viewpoint are: The support level for the cam-th viewpoint; Based on the viewpoint weights of each viewpoint, point cloud weighted fusion is performed to obtain a fused point cloud.

[0023] Preferably, the point cloud registration using the iterative nearest point algorithm includes the following steps: The iterative nearest point algorithm includes a source point cloud and a target point cloud. For each point in the source point cloud, the algorithm searches for its closest corresponding point in the target point cloud by Euclidean distance, thus obtaining a point cloud pair. The algorithm calculates the distance between all point cloud pairs and removes invalid corresponding point cloud pairs whose distance is greater than a preset threshold, thus obtaining a set of valid point cloud pairs. Based on the remaining set of valid point cloud pairs, the algorithm uses singular value decomposition to obtain the optimal rigid body transformation matrix, thereby achieving point cloud registration.

[0024] Preferably, the calculation of the first icing thickness by combining the icing equivalent diameter and the diameter of the uniced conductor includes the following specific steps: Poisson surface reconstruction is performed on the fused point cloud. A continuous surface model is generated by solving the Poisson equation, and then segments are cut at fixed intervals along the conductor axis on the continuous surface model. For each cross-section, calculate the area A and the initial equivalent diameter:

[0025] in, Indicates the first The initial equivalent diameter of each cross section, For the first The area of ​​each cross section; Calculate the minimum circumcircle diameter for each cross-section. and the diameter of the largest circumcircle ,when When the diameter is >0.3, the equivalent diameter of icing is : ;when When ≤0.3, the equivalent diameter of icing = ; Calculate the overall equivalent diameter:

[0026] in, For the overall equivalent diameter, For the first The equivalent diameter of the ice-covered cross section For the first The root mean square error of the fit of each cross section; Based on the overall equivalent diameter and the diameter of the non-icing conductor Calculate the first icing thickness:

[0027] in, The first icing thickness, For the overall equivalent diameter, This refers to the diameter of the non-icing conductor.

[0028] Preferably, the calculation of the actual sag change based on the icing conductor height sequence includes the following specific steps: The BeiDou positioning module obtains the elevation value Alt(t) of the conductor at time t and calculates the actual sag change at time t:

[0029] in, Let be the actual change in sag at time t. As the reference elevation, Let be the elevation at time t.

[0030] Preferably, the calculation of the theoretical sag change by combining the first icing thickness and the diameter of the uniced conductor includes the following specific steps: Using the conductor mechanical model, the theoretical change in sag is calculated:

[0031] in, This represents the theoretical change in sag. The sag conversion factor is... The diameter of the non-icing conductor. Indicates the first icing thickness. This is the correction factor for non-uniform icing.

[0032] A multifunctional sensor for power transmission lines includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0033] This invention provides a method for monitoring ice thickness based on a multifunctional sensor for power transmission lines, involving machine learning and deep learning technologies, which has the following beneficial effects: (1) The adaptive weighted fusion algorithm integrates multi-view point cloud data from three sets of binocular cameras and dynamically adjusts the weight of each viewpoint by combining point cloud density, registration accuracy and visual coverage. This effectively compensates for the blind spots and imaging distortion problems of a single viewpoint, achieves high-fidelity reconstruction of the three-dimensional contour of the ice-covered conductor, significantly improves the accuracy of calculating the equivalent diameter of the ice-covered conductor, and provides reliable basic data support for subsequent thickness measurement.

[0034] (2) By combining the first icing thickness with the diameter of the uniced conductor, the theoretical sag change was calculated, and a correlation model between the icing geometry and the line mechanical response was established, accurately quantifying the influence of icing load on conductor sag. This theoretical value provides an important physical reference benchmark for the actual sag change, creating key conditions for subsequently judging the rationality of visual measurement data and identifying potential observation biases.

[0035] (3) Adaptive correction of the first icing thickness using the difference index can effectively eliminate errors caused by noise interference and missed detection of blind icing in visual measurements. The corrected second icing thickness is closer to the actual state of conductor icing, greatly improving the reliability and accuracy of the thickness data and providing high-quality observation input for subsequent icing trend prediction and risk warning. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating the steps of a method for monitoring ice thickness based on a multifunctional sensor for power transmission lines proposed in this invention. Figure 2 This is a step hierarchy diagram of obtaining the first icing thickness in a method for monitoring icing thickness based on a multi-functional sensor for transmission lines proposed in this invention; Figure 3 This is a step hierarchy diagram of obtaining the difference index in a method for monitoring icing thickness based on a multifunctional sensor for transmission lines proposed in this invention; Figure 4 This is a schematic diagram of a multifunctional sensor for power transmission lines proposed in this invention; Figure 5 This is another schematic diagram of a multifunctional sensor for power transmission lines proposed in this invention; Figure 6 This is another schematic diagram of a multifunctional sensor for power transmission lines proposed in this invention; Figure 7 This is a schematic diagram showing the usage status of a multifunctional sensor for power transmission lines proposed in this invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1: Please see Figures 1-3 The present invention provides a technical solution: a method for monitoring ice thickness based on a multifunctional sensor for power transmission lines.

[0040] Step S1: Take a picture of the conductor in the transmission line under the condition of no ice to obtain an image of the conductor without ice; perform three-dimensional point cloud transformation on the image of the conductor without ice and perform cylinder fitting to obtain the diameter of the conductor without ice.

[0041] During the non-icing period of the transmission line, based on the line design parameters (such as span, elevation difference, and conductor type), mechanical calculations are first used to determine a region (such as a conductor segment of a certain length near the center of the span) between two adjacent towers near the theoretical lowest point of sag. Then, an integrated monitoring device is initially installed in this region and enters the "lowest point calibration mode": the BeiDou positioning module within the integrated monitoring device is controlled to move along the conductor in this region to at least five evenly distributed points, continuously collecting elevation data at each point for at least 10 minutes. The average elevation of each point is calculated, and the point with the smallest average elevation is the actual lowest point of sag. Finally, the monitoring device is securely installed at this point, and the final three-dimensional coordinates (longitude, latitude, and elevation) of this point are recorded as a reference. The high-precision BeiDou positioning module is fixed inside the device, ensuring that its antenna phase center is rigidly fixed to the conductor without relative displacement, thus allowing the BeiDou-positioned coordinates to accurately reflect the actual spatial position of the conductor at that point. To achieve multi-dimensional monitoring of the conductor icing status and the environment beneath the line, three sets of wide-angle binocular cameras are deployed on the integrated device: the first set is installed at the bottom of the device with its lens pointing vertically downwards. Its main function is to monitor construction activities or environmental changes in the ground area directly below the conductor and to acquire images of the area under the conductor to supplement icing information; the second set of cameras is installed at one end of the device along the conductor's axis, with its optical axis perpendicular to the conductor's axis in the horizontal plane and tilted downwards. Angle (e.g.) The first set of cameras (10°-15°) points towards the bottom left side of the conductor to ensure its core field of view covers the lower left side of the conductor. The second set of cameras is installed symmetrically with the second set, with its optical axis symmetrical to the second set, covering the lower right side of the conductor. Since all three sets of cameras use wide-angle lenses, the fields of view of the second and third sets of cameras covering the bottom left and right sides of the conductor overlap with the edge portion of the first set of bottom cameras' field of view that covers the side of the conductor. This overlap will provide spatial constraints for subsequent registration of 3D point cloud data.

[0042] It should be noted that the three sets of binocular cameras are powered by inductors, allowing them to operate 24 hours a day without interruption. These binocular cameras can monitor the surrounding environment, preventing power line faults caused by nearby construction and providing early warnings.

[0043] Under favorable weather conditions (visibility ≥ 5 km, wind speed ≤ 3 m / s, relative humidity ≤ 85%) and when the conductor surface is clean and ice-free (no visible stains or frost), data acquisition is triggered synchronously. Images of the ice-free conductors in the transmission line are captured: three sets of binocular cameras on the integrated control device simultaneously capture images. The first set of cameras acquires bottom binocular image pairs, while the second and third sets acquire bottom left and bottom right binocular image pairs, respectively. Simultaneously, a high-precision BeiDou positioning module integrated within the same device continuously acquires and records data at a specific frequency. The elevation of each measurement point.

[0044] The synchronously acquired bottom stereo image pairs, bottom left stereo image pairs, bottom right stereo image pairs, and BeiDou positioning data are processed. For each stereo image pair, stereo correction is first performed to align the corresponding epipolar lines of the left and right images (a stereo image pair includes both the left and right images) in each pair. Then, a semi-global matching algorithm is used to find matching points for each pixel in the left image on the corresponding row of the right image, and the disparity is calculated. , , Represents the x-coordinate of the left image. The x-coordinate of the right image is determined based on the internal parameters (focal length) calibrated for the left camera in the binocular camera system. , and the main point , The baseline distance Distance between the left and right cameras (the horizontal distance between the optical centers of the left and right cameras in the binocular camera unit) is used to convert the pixel coordinates (u,v) of the left image into three-dimensional points (X,Y,Z) in the camera coordinate system based on parallax and baseline distance.

[0045]

[0046]

[0047] Where X, Y, and Z are the three-dimensional coordinates of the point cloud. This indicates the focal length of the left camera in the X-direction of a binocular camera system. This indicates the focal length of the left camera in the Y direction within a binocular camera system. Here, denoted as parallax, Distance is the baseline distance between the two cameras, and u,v are the pixel coordinates of the left image.

[0048] It should be noted that parallax ∈[ min, max], The minimum value can be 5 pixels, and the maximum value can be 50 pixels. Matching points that exceed the parallax range are considered invalid and are discarded. Only when the value of Parallax is positive will the pixel be included in the 3D coordinate calculation. If Parallax ≤ 0, the pixel will be discarded and no point cloud data will be generated.

[0049] It should be noted that in 3D coordinate transformation, only the pixel coordinates of the left image are used because the parallax itself already contains the key information of the corresponding matching point in the right image. This is achieved through the formula... The calculated disparity value essentially establishes the geometric relationship between corresponding pixels in the left and right images. Therefore, by combining the intrinsic parameters of the left camera and the baseline distance, a single pixel in the left image and its disparity are sufficient to uniquely determine the position of that point in three-dimensional space using triangulation. This is a standard and efficient calculation method that avoids redundant calculations using the coordinates of the right image.

[0050] After converting pixel coordinates to 3D coordinates, 3D point cloud data for each set of binocular cameras is obtained. To accurately extract the main conductor from the point cloud containing complex backgrounds such as the ground, towers, and insulators, since the camera's field of view is set to the conductor, large objects such as towers and insulators in the background are outside the field of view. The noise in the point cloud mainly comes from the environmental background (such as the ground and sky). By using RANSAC-based plane fitting and Euclidean clustering algorithms, noise can be effectively removed, and the point cloud clusters representing the main conductor can be accurately extracted. First, RANSAC-based plane fitting is performed on the 3D point cloud data: the algorithm randomly samples three points to generate a candidate plane model, calculates the distance from all points in the point cloud to the plane, and selects points with a distance less than a threshold Threshold, where Threshold = 2 * Threshold ∈ [3cm, 8cm], to avoid processing errors caused by thresholds that are too small or too large. Points with the highest overall noise standard deviation of the point cloud (calculated by statistically analyzing the dispersion of the point cloud on the initial fitted plane) are identified as "inliers." After multiple iterations (e.g., 1000), the planar model with the most inliers is selected, and its corresponding inliers (usually corresponding to the ground or tower, etc., as the main planar background) are removed as overall noise, resulting in a non-planar point cloud dominated by foreground objects such as conductors and insulator strings. Subsequently, Euclidean clustering is performed on the non-planar point cloud, starting from any point in the point cloud and searching within its radius (radius calculation: Radius = ...). +5cm, of which For each conductor (with a known diameter), all neighboring points are identified and grouped into the same cluster. Then, the search continues recursively from these newly added points until no new neighboring points are added, thus forming a complete object point cloud cluster. Finally, based on prior knowledge (such as the conductor point cloud cluster having the longest size and the most points), the point cloud cluster representing the conductor's main body is selected from all clustering results. (If the conductor point cloud count is less than 60% of the total point cloud after clustering, the Radius is adjusted to 1.2*Radius, and re-clustering is performed.)

[0051] A cylindrical model is fitted to the point cloud clusters belonging to the main body of the conductor. The cylindrical model can be represented as (C, r, axis), where C is a point on the axis of the cylinder, axis is the unit direction vector of the axis of the cylinder, and r is the radius of the cylinder. The optimal cylinder radius r is obtained by minimizing the sum of squared distances from the point cloud to the cylinder surface using the least squares method. Using 2r as the reference diameter, the left visual reference diameter is obtained. and right visual reference diameter The formula for the sum of squares of the distances from the point cloud to the cylindrical surface is as follows:

[0052] in, Let be the three-dimensional coordinate vector of the j-th point in the point cloud, C be the point on the axis of the cylinder, axis be the unit direction vector of the axis of the cylinder, and r be the radius of the cylinder.

[0053] It should be noted that although 3D reconstruction and point cloud generation are performed on the bottom binocular image pairs in the baseline data processing, the main purpose is to provide a spatial reference for subsequent point cloud fusion, rather than to calculate the diameter of the conductor. This is mainly due to the following two reasons: First, the optical axis of the bottom camera is vertically downward, and its imaging center area is concentrated on the ground environment. The observation of the bottom of the conductor is at the edge of the field of view, resulting in significant distortion. The diameter directly fitted from this perspective is insufficiently accurate and unreliable. Second, the optical axes of the left and right cameras are specifically pointed to the sides of the bottom of the conductor, which can more accurately capture the complete lower side contour of the conductor. The diameter calculated from these cameras already has sufficient geometric representation capabilities. Therefore, to avoid introducing errors and optimize computational efficiency, the diameter calculation relies only on the data from the side cameras.

[0054] Diameter based on left visual reference and right visual reference diameter Calculate the diameter of the non-icing conductor , = .

[0055] It should be noted that, in the calculation of the baseline diameter without icing, although there is a blind spot in the top view of the conductor, the point cloud coverage of more than 180 degrees provided by the left and right views is sufficient to accurately reconstruct the conductor diameter using cylinder fitting. During the icing monitoring phase, an adaptive weighted fusion algorithm can be used to comprehensively utilize multi-view data to compensate for local blind spots, achieving a complete reconstruction of the ice outline.

[0056] Simultaneously, the measurement points collected by the BeiDou positioning module are continuously... coordinate data points {( , , | i=1,2,... The process involves processing the data. A sliding window weighted average algorithm is used to suppress random noise, and the baseline elevation is calculated. :

[0057] in, As the reference elevation, This represents the elevation of the i-th coordinate data point. The precision weight is for the i-th data point. This represents the number of coordinate data points.

[0058] It should be noted that the precision weights The specific value is determined by both the satellite elevation angle and the signal-to-noise ratio. The calculation formula is as follows: = * ,in Let be the satellite's elevation angle at the i-th sampling time. Let be the carrier phase signal-to-noise ratio observation value at the i-th sampling time. This is a preset signal-to-noise ratio reference value (usually 45dB-Hz). The formula is obtained through... This significantly reduces the weight of low-elevation-angle satellites to suppress atmospheric delay errors; through This involves increasing the weight of high signal-to-noise ratio observations to suppress random observation noise. Ultimately, this is achieved by applying weights to all... Perform normalization processing to ensure This enables the calculation of the optimal weighted average based on the observation quality.

[0059] Ultimately, the diameter of the ice-free conductor... The reference elevation obtained from BeiDou data After association, it is stored in the benchmark database.

[0060] Step S2: Take a picture of the icy conductor to obtain an image of the icy conductor, and convert it into an icy conductor point cloud; perform three-dimensional point cloud fusion on the icy conductor point cloud using an adaptive weighted fusion method to obtain a conductor fused point cloud; perform surface reconstruction on the conductor fused point cloud using the Poisson surface reconstruction method to calculate the equivalent diameter of the icy conductor, and calculate the first icing thickness by combining the equivalent diameter of the icy conductor and the diameter of the uniced conductor.

[0061] After the monitoring unit enters normal operation, the periodic data acquisition and processing procedure under icing conditions is initiated. A dynamic time interval mechanism is adopted initially. t=5 minutes, calculate the ice thickness growth rate IceGrow after every 3 data collections, IceGrow= If IceGrow ≥ 0.1 mm / min (rapid ice removal), t is adjusted to 2 minutes; if IceGrow ≤ 0.02 mm / min (slow icing), The time interval t is adjusted to 10 minutes, and the synchronization accuracy must meet the requirement that the time difference between the camera and the BeiDou module is ≤50ms, which is calibrated through the internal clock of the device. At each sampling time t, three sets of binocular cameras and the high-precision BeiDou positioning module are synchronously triggered. The first set (bottom) of cameras synchronously acquires the bottom binocular image pairs; the second set (left) and the third set (right) of binocular cameras acquire the binocular image pairs of the left and right areas of the bottom of the pointing guide, respectively. At the same time, the BeiDou positioning module initiates the BeiDou elevation data stability verification through real-time dynamic differential positioning technology, continuously acquiring 10 elevation points. Only those with a standard deviation ≤3mm can be used for subsequent calculations; otherwise, outliers are removed and re-acquired, and finally, the elevation Alt(t) of the measured point at sampling time t is output.

[0062] The next step is the 3D reconstruction of the icing contour. Using 3D point cloud data simultaneously acquired and reconstructed from left, right, and bottom cameras (i.e., left, right, and bottom point clouds), a 3D model of the icing surface of the conductor is constructed through spatial registration and fusion algorithms. First, based on the camera extrinsic parameters (including rotation matrix and translation vector) of the binocular cameras for each set of visual views (left, right, and bottom), the left, right, and bottom point clouds are transformed from their respective coordinate systems to a unified world coordinate system, laying the spatial foundation for subsequent fusion. The transformation formula is as follows:

[0063] in, The coordinates of a point in the world coordinate system. Let be the rotation matrix under the cam-th viewpoint. Let be the translation vector at the cam-th viewpoint. Let be the coordinates of the point in the cam-th viewpoint coordinate system.

[0064] The next step is the point cloud registration stage. Since all three cameras use wide-angle lenses, there is an overlap in the point clouds under their three views.

[0065] In the overlapping regions of the point clouds, a keypoint extraction algorithm based on Inherent Shape Features (ISS) is used. This algorithm filters locations with significant geometric features (such as corners and boundary points) by calculating the eigenvalues ​​of the three-dimensional covariance matrix of the points. A fast point feature histogram descriptor is calculated for each keypoint; this 33-dimensional vector encodes its local geometric features by combining the relative deviations of the point's normal vector. Subsequently, feature matching and mismatch removal are performed using the Random Sample Consensus Algorithm (RANSAC): three pairs of matching points are randomly selected to calculate the candidate transformation matrix, and the number of inliers satisfying the transformation is counted (point cloud pairs whose distance after transformation is less than an inlier threshold, where the inlier threshold = 2 * ...). , The standard deviation of the overall noise of the point cloud is calculated by statistically analyzing the dispersion of the point cloud. (The number of iterations is given.) =ceil( ), where p=0.99, representing the expected registration success rate. The estimated noise rate of the point cloud is set to 0.3; The transformation with the most interior points is then selected as the optimal initial transformation matrix. Based on the initial transformation provided by the coarse registration, the iterative nearest-point algorithm is used for precise registration. This algorithm iteratively executes the following steps until the convergence condition is met (the root mean square error change between adjacent iterations is less than 1%). And the root mean square error ≤ 0.5mm, or 50 iterations: First, for each point in the source point cloud, search for its closest Euclidean distance counterpart in the target point cloud to establish a preliminary point cloud pair correspondence (for example, when registering the left and bottom point clouds, the left point cloud can be defined as the source point cloud, which needs to be transformed and aligned, while the bottom point cloud can be defined as the target point cloud, serving as the alignment reference); then, to suppress mismatch interference, calculate the distance between all point cloud pairs and remove invalid corresponding point cloud pairs with a distance greater than a set threshold (for example, calculate the distance between all point cloud pairs, take the median distance, and remove corresponding points with a distance greater than 3 times the median); the root mean square error output after fine registration must be < 1mm, otherwise the registration is considered a failure, and the coarse registration + fine registration process is re-executed; finally, based on the remaining set of valid point cloud pairs, solve for the optimal rigid body transformation matrix that minimizes the overall alignment error using singular value decomposition. This minimizes the objective function, which is:

[0066] in, It is the j-th point in the source point cloud. In the target point cloud The nearest corresponding point It represents the number of valid corresponding point cloud pairs.

[0067] After registration, an adaptive weighted fusion algorithm is used to optimize and fuse the point clouds from multiple views. First, the point cloud quality evaluation index for each view is calculated using the following formula:

[0068] in, Let be the point cloud quality evaluation index from the cam-th viewpoint. Let be the point cloud density under the cam-th view. Let be the registration error under the cam-th viewpoint. Let be the visual coverage of the cam-th viewpoint. The point cloud density weighting coefficient. These are the registration error weighting coefficients. This is the visual coverage weighting coefficient.

[0069] It should be noted that, , and The value of must satisfy the normalization condition. + + =1, its setting is based on a clear physical meaning and optimization objective: to assign point cloud density. Highest weight (recommended) =0.5), because it directly determines the detail and completeness of the 3D reconstruction of the icy surface; the reciprocal of the registration error Secondary weight (recommended) =0.3), used to prioritize the accuracy of multi-view point cloud spatial registration and effectively suppress "ghosting"; visual coverage The weight is relatively the lowest (recommended) =0.2), which has a relatively limited contribution to the overall accuracy under the premise of multi-view complementarity. This weight allocation (0.5,0.3,0.2) reflects the fusion strategy of "prioritizing detailed accuracy, ensuring spatial consistency, and supplementing the observation range", which aims to guide the algorithm to output a high-fidelity and high-reliability 3D model of the icing of the conductor.

[0070] Point cloud density This represents the number density of valid points in the point cloud at that viewpoint (e.g., points per square meter). Higher density provides a more detailed description of the shape. The calculation formula is... Visual coverage Registration error indicates the completeness of the observation of the traverse surface from that viewpoint, usually expressed as the angle of the traverse outline covered by the point cloud (e.g., in radians or percentages). It is the root mean square error of the point cloud at this viewpoint after ICP registration.

[0071] Arrange the point cloud quality evaluation indices from the three perspectives in ascending order: Calculate the median : Calculate the upper quantile : Calculate the lower quantile : Quantile dispersion : .

[0072] Set anomaly detection threshold: = * ( This is the abnormal adjustment coefficient. =1.2+0.3*(1- ),in The average visual coverage of the three viewpoints. The lower the temperature (e.g., on foggy days), The larger the value, the greater the tolerance for anomaly detection. Calculate the anomaly severity factor for the cam-th viewpoint:

[0073] in, Let be the anomaly factor of the i-th viewpoint. The median of the point cloud quality evaluation index. This is the threshold for anomaly detection.

[0074] Establish mutual support relationships among the three cameras, and measure the point cloud similarity for any two viewpoints:

[0075] in, For the point cloud similarity measure between cam1 and cam2, The effective number of point clouds for viewpoint cam1. This represents the number of points in the overlapping area.

[0076] Relationship matrix construction:

[0077] in, The relationship index for cam1 and cam2, and the consistency threshold. , =0.2+0.1*( ), The average point cloud density is given by the three viewpoints. The higher, The larger the value, the higher the similarity requirement can be. ∈[0.2,0.4].

[0078] The support level for each perspective is:

[0079] in, For the support of the cam1 vision, is the abnormality factor of the second vision.

[0080] Based on the point cloud quality evaluation index, support, and preset benchmark weights, the view weights for each viewpoint are calculated:

[0081] in, Let be the view weight under the cam-th view. The preset baseline weights for the i-th viewpoint are: denoted as the support level for the cam-th viewpoint.

[0082] It should be noted that the preset baseline weights for the cam-th viewpoint... Left-side perspective weight The main contribution is to the accuracy of the left side profile of the guide, with a weighting of 0.4; the right side view weight... The main contribution is to the accuracy of the right-side profile of the guide, with a weighting of 0.4; the bottom view weight is also significant. The main contribution is to bottom icing detection, with a weighting of 0.2.

[0083] Based on the calculated visual weights, point clouds from three viewpoints are fused. The fusion process employs a voxel-based resampling method. First, the point clouds from each viewpoint are uniformly sampled to the same density. Then, in overlapping regions, point clouds are optimized according to their weights: viewpoints with higher visual weights retain more points, while viewpoints with lower visual weights are appropriately thinned out. The final result is a fused point cloud with uniform density. This lays the foundation for subsequent surface reconstruction.

[0084] Normal vector estimation is performed on the fused point cloud (using the k-nearest neighbor method, k=10) to obtain the unit normal vector of each point. The gradient field V is the distribution function of the normal vectors of all points in space, i.e., V(p) = Where p is any point in space, Let j be the j-th point in the point cloud. For smoothing coefficients, =0.1* ( (where is the fitting radius of the fused point cloud), the root mean square error of the normal vector estimation must be ≤0.1 rad, otherwise re-estimation is required.

[0085] For fused point clouds To reconstruct the Poisson surface, the Poisson equation is solved. ,in For the Laplace operator, we obtain the indicator function. The isosurfaces are extracted, where V is the gradient field of the point cloud. =0 generates a continuous surface model S(t). On the continuous surface model, along the conductor axis at intervals (interval = max(5mm, min(15mm, 2*...). )), The ice thickness is estimated based on the initial point cloud. = - (The thinner the ice layer and the smaller the intervals, the better to capture local details.) There are several cross-sections. For each cross-section, calculate its area A (obtained from the point cloud model), and then convert this area to the area of ​​a circle to deduce its initial equivalent diameter.

[0086] in, Indicates the first The initial equivalent diameter of each cross section, For the first The area of ​​a cross-section.

[0087] Calculate the minimum circumcircle diameter for each cross-section. and the diameter of the largest circumcircle ,like If the diameter is greater than 0.3, the cross-section is determined to have severely uneven icing. Therefore, the equivalent diameter of the icing is... Reduce the impact of localized protrusions; if ≤0.3, no correction needed, directly take the equivalent diameter of icing. = .

[0088] The overall equivalent diameter is calculated using a weighted average method:

[0089] in, For the overall equivalent diameter, For the first The equivalent diameter of the ice-covered cross section For the first The root mean square error of the fit for each cross section.

[0090] Based on the overall equivalent diameter and the diameter of the non-icing conductor Calculate the first icing thickness:

[0091] in, The first icing thickness, For the overall equivalent diameter, This refers to the diameter of the non-icing conductor.

[0092] like < If the measurement is deemed abnormal, the average of the previous three valid first thickness measurements will be taken as the current measurement. And alert; if ≥ Then calculate the icing thickness normally. Record the local maximum thickness corresponding to the equivalent diameter of all cross-sections. .

[0093] It should be noted that the difference in diameter calculation methods between steps S1 and S2 stems from the fundamental difference in the surface morphology and measurement target of the conductor under ice-free baseline and ice-covered monitoring conditions. In the ice-free state (S1), the conductor surface is clean, geometrically regular, and shaped like a standard cylinder. Furthermore, an absolute baseline unaffected by environmental interference is established at this point. The optical axes of the two cameras are precisely pointed to the lower side of the conductor, and their observation data can completely and accurately capture the conductor's inherent contour. Therefore, a simple arithmetic mean of the diameter from the left and right perspectives is used. = This simplified processing is sufficient to establish a baseline diameter, optimizing computational efficiency while ensuring the purity of the baseline data. However, under icing conditions, ice accumulation exhibits significant non-uniformity and randomness, creating blind spots in visual monitoring (such as the top and bottom of the conductor), leading to biased or even severely distorted observation data from a single perspective. In this situation, multi-view point cloud fusion technology is necessary. This leverages the complementarity of data from the left, right, and bottom perspectives, and adaptively weights the data based on its quality to reconstruct the true conductor outline encased in irregular ice in three-dimensional space, thereby calculating the equivalent diameter that reflects the overall ice load. This shift from "simple averaging" to "complex fusion" demonstrates the method's adaptive optimization capability for different operating conditions and is a design feature ensuring accuracy and robustness throughout the entire process from baseline to monitoring.

[0094] Step S3: Obtain the height sequence of the icing conductor, and calculate the actual sag change based on the icing conductor height sequence. Combine the first icing thickness and the diameter of the unicing conductor to calculate the theoretical sag change.

[0095] The elevation value Alt(t) is directly read from the BeiDou positioning data and compared with the benchmark elevation of the guide wire in the benchmark database under icy conditions. By performing difference calculations, the actual change in sag at time t is obtained:

[0096] in, Let be the actual change in sag at time t. As the reference elevation, Let be the elevation at time t.

[0097] A conductor mechanics model based on the parabolic assumption is adopted. After icing, the load per unit length of the conductor increases, leading to an increase in sag:

[0098] in, This represents the theoretical change in sag. The sag conversion factor is... The diameter of the non-icing conductor. Indicates the first icing thickness. This is a correction factor for non-uniform icing. = .

[0099] It should be noted that the sag conversion coefficient... Based on the elastic modulus E of the conductor (known and determined by the conductor type) and cross-sectional area (A= * ),pass (t)= + Calculate the real-time horizontal stress after icing, where The increment of icing weight per unit length of conductor = *g*[ *( - )], The horizontal stress of the conductor when there is no ice is represented, and L is the span (the horizontal distance between adjacent towers); the real-time horizontal stress is... Substituting (t) into the sag formula, (t)= ,in Let g be the density of ice and g be the acceleration due to gravity. Real-time dynamic update of (t), with constraints: (t)≤ (Maximum allowable stress on the conductor, design parameters), if (t)> It is directly identified as a high-risk state, without the need for subsequent difference index calculation, and an early warning is triggered directly.

[0100] Step S4: Calculate the difference index by comparing the theoretical sag change with the actual sag change; correct the first icing thickness using the difference index to obtain the second icing thickness.

[0101] A physical consistency analysis is conducted based on the theoretical and actual sag changes:

[0102] in, This represents the difference index at time t. Let be the theoretical change in sag at time t. Let be the actual change in sag at time t.

[0103] Set a reasonableness threshold θtol (θtol=0.1+0.05*min(1, ),in For the first icing thickness, when ≤5mm (thin icing), θtol=0.15 (larger deviation allowed); when >5mm (thick icing), θtol=0.1 (strict deviation control); θtol∈[0.08,0.2], to avoid [causing ice buildup]. (If the threshold is too small or too large, it will exceed a reasonable range). When ε(t) > θtol, a significant physical inconsistency is considered to have occurred, triggering an adaptive correction mechanism. When a significant physical inconsistency occurs, and > In this case, the actual sag change detected mechanically exceeds the range of sag change detected by the point cloud. This indicates the existence of a visual monitoring blind spot, and that icing is developing within this blind spot. The most common blind spot is the bottom of the conductor, as these areas are blind spots for side cameras.

[0104] Increase the fusion weight of the bottom camera, as it is crucial for detecting bottom icing:

[0105] in, The corrected bottom weights, The corrected left-side weights, This is the gain coefficient.

[0106] After obtaining the corrected bottom weights, the corrected bottom weights, the original left weights, and the original right weights need to be normalized to ensure that the sum of all weights is 1, thus obtaining the final weights used for point cloud fusion.

[0107] When significant physical inconsistencies occur, and < At times, the sag change measured by point cloud exceeded the sag change reflected by mechanical measurements. This may be because the point cloud calculation method overestimates the ice thickness, for example, misclassifying shadows, dirt, frost, or oddly shaped icicles on the conductor as uniform icing. Side cameras are more susceptible to this type of interference; therefore, reducing the fusion weights of the side cameras (left and right) can suppress potential visual noise.

[0108]

[0109]

[0110] in, The corrected bottom weights, The corrected left-side weights, This is the gain coefficient. To reduce the coefficient, The corrected right-hand weights, This represents the difference index at time t.

[0111] The adjusted weights (corrected bottom weights, corrected left-side weights, and corrected right-side weights) are then normalized to ensure that the sum of all weights is 1, thus obtaining the final weights used for point cloud fusion.

[0112] It should be noted that the gain coefficient and reduce coefficient These are two independent configurable parameters, whose values ​​jointly determine the sensitivity and stability of the system's response to the mechanical-visual difference; the recommended baseline value is [value missing]. It is 0.5. A value of 0.5 achieves a robust balance between correction efficiency and oscillation suppression, and this baseline value can be used as the initial parameter of the system. Based on this, both can be independently tuned within the range of 0.3 to 1.0: for more decisive suppression of specific viewing angle noise (such as side-view lighting interference), the value can be set to... > (like =0.7, =0.4); if a more sensitive response is needed to the blind zone icing implied by mechanical feedback, it can be set to > (like =0.7, =0.4), thereby achieving asymmetric fine calibration for specific scenarios.

[0113] Using the adjusted final weights as the return step, a new weighted fusion is performed on the point cloud data at the current moment to obtain the corrected fused point cloud. Based on the corrected fused point cloud, surface reconstruction and thickness extraction are performed again to obtain the second icing thickness after sag-visual consistency correction. .

[0114] Step S5: Collect the second ice thickness sequence within a preset time period and input it into the gray prediction model based on dynamic gray action amount. Output the predicted ice thickness value, classify the risk level based on the predicted ice thickness value and issue an early warning, thereby realizing the detection of ice thickness.

[0115] Corrected thickness value The specific process for inputting new observations into the gray prediction model based on dynamic gray action is as follows: The obtained time series of sag changes { } as the original sequence : , Let be the ice thickness at the k-th sampling time, K be the total number of sampling times, and k be the index of the sampling time. For the original sequence... Perform an accumulation generation to obtain the accumulation sequence. , .

[0116] Establish the whitening differential equation of the improved DGM(2,1) model, and introduce the linear time-varying gray action to optimize the model's adaptability:

[0117] Where a is the development coefficient, bk+c is the linear time-varying ash action amount, b is the rate of change of the ash action amount, and c is the baseline value of the ash action amount. This represents a cumulative sequence.

[0118] Solving the parameter series using the least squares method ,satisfy: The matrix B and vector Y are defined as follows:

[0119]

[0120] in It is a first-order cumulative difference operator that satisfies , Optimize parameters for the gray derivative.

[0121] After obtaining the parameter estimates, a single cumulative sequence is generated. The time response function is:

[0122] in, This represents the predicted value of a cumulative sequence at time k+1. , The constants of the time response function are to be determined by fitting the accumulated sequence observations using the least squares method.

[0123] By performing cumulative subtraction on the accumulated prediction sequence, the predicted value of the icing thickness is obtained:

[0124] in, This represents the predicted change in icing thickness at time k+1. This represents the predicted value of a cumulative sequence at time k.

[0125] Take the predicted icing thickness for the next window of time (e.g., window=15, each window is 3 minutes long), and calculate the icing risk value based on the predicted icing thickness:

[0126] Where FX is the icing risk value, and min() is the function to find the minimum value. This represents the maximum predicted icing thickness within the time window. To preset a dangerous thickness threshold, This represents the average thickness growth rate over the time window. This is a preset threshold for the rate of dangerous growth.

[0127] After obtaining the predicted icing thickness based on the grey prediction model, the predicted value is classified into risk levels according to preset safe operation thresholds. Specifically, when the icing risk value is below the attention threshold, the risk level is determined to be "normal," and monitoring continues without triggering an alarm; when the icing risk value exceeds the attention threshold but does not reach the warning threshold, the level is upgraded to "attention," and the system provides a prompt on the monitoring interface; when the icing risk value exceeds the warning threshold but does not reach the danger threshold, the level is set to "warning," and an alarm message is automatically sent to the monitoring center, prompting increased attention; when the icing risk value exceeds the danger threshold, the risk level is upgraded to the highest level, "dangerous," at which point the highest level warning is immediately triggered, and an emergency notification is automatically sent to the designated operation and maintenance personnel. This prediction-based hierarchical early warning mechanism realizes a closed loop from "trend prediction" to "risk decision-making," transforming traditional passive alarms into proactive early warnings, providing a critical time window for inspection scheduling and de-icing operations, thereby significantly improving the initiative and timeliness of transmission line anti-icing and disaster reduction.

[0128] Example 2: Please see Figures 1-7The present invention provides a technical solution: a multifunctional sensor for power transmission lines, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0129] Specifically, the aforementioned multi-functional sensor for power transmission lines includes: a housing 1, and a camera 2 mounted on the housing 1. A photovoltaic panel 3 is also mounted on the top of the housing 1. Inside the housing 1 are sensors, a communication module, a memory, a processor, and a computer program stored in the memory. The number and types of sensors, and the number and types of cameras 2, can be designed according to actual needs.

[0130] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0131] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring icing thickness based on a multifunctional sensor for power transmission lines, characterized in that: Includes the following steps: Step S1: Take a picture of the conductor in the transmission line in the state of no ice to obtain an image of the conductor without ice; perform three-dimensional point cloud transformation on the image of the conductor without ice and perform cylinder fitting to obtain the diameter of the conductor without ice. Step S2: Take a picture of the ice-covered conductor to obtain an image of the ice-covered conductor, and convert it into an ice-covered conductor point cloud; perform three-dimensional point cloud fusion on the ice-covered conductor point cloud using an adaptive weighted fusion method to obtain a conductor fused point cloud; perform surface reconstruction on the conductor fused point cloud using the Poisson surface reconstruction method to calculate the ice-covered equivalent diameter; combine the ice-covered equivalent diameter and the diameter of the un-ice-covered conductor to calculate the first ice thickness. Step S3: Obtain the height sequence of the icy conductor, and calculate the actual sag change based on the icy conductor height sequence. Combine the first icing thickness and the diameter of the uniced conductor to calculate the theoretical sag change. Step S4: Calculate the difference index by comparing the theoretical sag change with the actual sag change; correct the first icing thickness using the difference index to obtain the second icing thickness.

2. The method for monitoring icing thickness based on a multi-functional sensor for transmission lines according to claim 1, characterized in that: The process of photographing the conductors of a transmission line in an ice-free state to obtain images of the ice-free conductors includes the following specific steps: During the non-icing period of the transmission line, the measurement point of the conductor between two towers is determined, and an integrated monitoring device is installed at the measurement point, with the Beidou positioning module fixed inside the detection device; Three sets of binocular cameras are installed on the monitoring device: the first set of binocular cameras is installed at the bottom of the device; the second set of binocular cameras is installed at one end of the device along the conductor axis of the transmission line, with its optical axis perpendicular to the conductor axis in the horizontal plane and tilted downwards. The first set of cameras points to the bottom left area of ​​the conductor, covering the lower left side of the conductor; the second set of cameras is installed symmetrically with the second set of cameras, with the optical axis direction symmetrical to the second set of cameras, covering the lower right side of the conductor. During the non-icing period of the transmission line, the three sets of binocular cameras on the monitoring device simultaneously capture images of the non-iced conductor. The non-iced conductor images include the bottom binocular image pair captured by the first set of binocular cameras, the bottom left binocular image pair captured by the second set of binocular cameras, and the bottom right binocular image pair captured by the third set of binocular cameras.

3. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 2, characterized in that: The process of converting the image of the ice-free conductor into a 3D point cloud and fitting it to a cylinder to obtain the diameter of the ice-free conductor includes the following steps: Stereo correction is performed on the synchronously acquired bottom binocular image pairs, bottom left binocular image pairs, and bottom right binocular image pairs. For each pair of binocular images, the disparity between the left and right images in each pair is calculated. , , Represents the x-coordinate of the left image. Given the x-coordinate of the right image, based on the focal length calibrated for the left camera in the binocular camera system and the baseline distance Distance between the principal point and the binocular camera, the pixel coordinates (u,v) of the left image are converted into three-dimensional points (X,Y,Z) in the binocular camera coordinate system. After converting pixel coordinates into three-dimensional coordinates, we obtain the three-dimensional point cloud data of each set of binocular cameras and filter out the point cloud clusters that represent the main body of the conductor. A cylindrical model is fitted to the point cloud clusters belonging to the main body of the conductor. The cylindrical model is represented as (C, r, axis), where C is a point on the axis of the cylinder, axis is the unit direction vector of the axis of the cylinder, and r is the radius of the cylinder. The optimal cylinder radius r is obtained by minimizing the sum of squared distances from the point cloud to the cylinder surface using the least squares method. 2r is then used as the reference diameter to obtain the reference diameter for the left visual view. and right visual reference diameter The formula for the sum of squares of the distances from the point cloud to the cylindrical surface is as follows: ； in, Let be the three-dimensional coordinate vector of the j-th point in the point cloud, C be the point on the axis of the cylinder, axis be the unit direction vector of the axis of the cylinder, and r be the radius of the cylinder. The diameter of the icing-free conductor is calculated using the left and right visual reference diameters.

4. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 3, characterized in that: The calculation of the diameter of the ice-free conductor using the left and right visual reference diameters is as follows: ； in, The diameter of the non-icing conductor. The diameter is the left visual reference diameter. The diameter is the right visual reference diameter.

5. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 4, characterized in that: The method of performing three-dimensional point cloud fusion on the icing guideline point cloud using an adaptive weighted fusion method to obtain the fused guideline point cloud includes the following steps: Based on the rotation matrix and translation vector from three perspectives, the reconstructed left point cloud, right point cloud and bottom point cloud from the left binocular camera, right binocular camera and bottom binocular camera are transformed into a unified world coordinate system. After converting the point cloud into the same world coordinate system, the iterative nearest point algorithm is used for point cloud registration; After point cloud registration, the point cloud quality evaluation index is calculated for each viewpoint. The calculation formula is as follows: ； in, Let be the point cloud quality evaluation index from the cam-th viewpoint. Let be the point cloud density under the cam-th view. Let be the registration error under the cam-th viewpoint. Let be the visual coverage of the cam-th viewpoint. The point cloud density weighting coefficient. These are the registration error weighting coefficients. This is the visual coverage weighting coefficient; Calculate the anomaly factor of the i-th viewpoint: ； in, Let be the anomaly factor of the i-th viewpoint. The median of the point cloud quality evaluation index. This is the threshold for anomaly detection; Establish mutual support relationships among the three cameras, and measure the point cloud similarity for any two viewpoints: ； in, For the point cloud similarity measure between cam1 and cam2, The effective number of point clouds for viewpoint cam1. This represents the number of points in the overlapping region. The support level for each perspective is: ； in, For the support of the cam1 vision, The relationship index between cam1 and cam2 is... is the abnormality factor for the second visual image; Based on the point cloud quality evaluation index, support, and preset baseline weights, calculate the view weight for each viewpoint: ； in, Let be the view weight under the cam-th view. The preset baseline weights for the i-th viewpoint are: The support level for the cam-th viewpoint; Based on the viewpoint weights of each viewpoint, point cloud weighted fusion is performed to obtain a fused point cloud.

6. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 5, characterized in that: The point cloud registration using the iterative nearest point algorithm includes the following steps: The iterative nearest point algorithm includes a source point cloud and a target point cloud. For each point in the source point cloud, the algorithm searches for its closest corresponding point in the target point cloud by Euclidean distance, thus obtaining a point cloud pair. The algorithm calculates the distance between all point cloud pairs and removes invalid corresponding point cloud pairs whose distance is greater than a preset threshold, thus obtaining a set of valid point cloud pairs. Based on the remaining set of valid point cloud pairs, the algorithm uses singular value decomposition to obtain the optimal rigid body transformation matrix, thereby achieving point cloud registration.

7. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 6, characterized in that: The calculation of the first icing thickness by combining the equivalent diameter of the iced conductor and the diameter of the uniced conductor includes the following specific steps: Poisson surface reconstruction is performed on the fused point cloud. A continuous surface model is generated by solving the Poisson equation, and then segments are cut at fixed intervals along the conductor axis on the continuous surface model. For each cross-section, calculate the area A and the initial equivalent diameter: ； in, Indicates the first The initial equivalent diameter of each cross section, For the first The area of ​​each cross section; Calculate the minimum circumcircle diameter for each cross-section. and the diameter of the largest circumcircle ,when When the diameter is >0.3, the equivalent diameter of icing is : ;when When ≤0.3, the equivalent diameter of icing = ; Calculate the overall equivalent diameter: ； in, For the overall equivalent diameter, For the first The equivalent diameter of the ice-covered cross section For the first The root mean square error of the fit of each cross section; Based on the overall equivalent diameter and the diameter of the non-icing conductor Calculate the first icing thickness: ； in, The first icing thickness, For the overall equivalent diameter, This refers to the diameter of the non-icing conductor.

8. The method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 7, characterized in that: The calculation of the actual sag change based on the icy conductor height sequence includes the following specific steps: The BeiDou positioning module obtains the elevation value Alt(t) of the conductor at time t and calculates the actual sag change at time t: ； in, Let be the actual change in sag at time t. As the reference elevation, Let be the elevation at time t.

9. A method for monitoring icing thickness based on a multifunctional sensor for transmission lines according to claim 8, characterized in that: The theoretical sag change is calculated by combining the first icing thickness and the diameter of the uniced conductor, including the following specific steps: Using the conductor mechanical model, the theoretical change in sag is calculated: ； in, This represents the theoretical change in sag. The sag conversion factor is... The diameter of the non-icing conductor. Indicates the first icing thickness. This is the correction factor for non-uniform icing.

10. A multifunctional sensor for power transmission lines, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-9.

Images