A monitoring method, device, equipment and storage medium for overhead transmission lines

Abstract

The present invention provides a monitoring method, device, equipment and storage medium for overhead transmission lines, which belongs to the field of transmission line maintenance. The method includes: constructing an electric power element model and an unmanned aerial vehicle simulation model, obtaining real-time data of the target overhead transmission line through the unmanned aerial vehicle simulation model, and locating the fault position in the electric power element model according to the abnormal data when there is abnormal data in the real-time data. The tree building process of KD-Tree is optimized by an adaptive segmentation strategy, and the point cloud data is rendered by the optimized KD-Tree to construct the electric power element model, which can reduce the depth of the tree, improve the search efficiency, the accuracy of intelligent recognition and the response speed, and realize accurate analysis of the line status; the real-time operation data of the target overhead transmission line is mapped to the electric power element model through the unmanned aerial vehicle simulation model, and the location where the fault may exist is located according to the abnormal data when there is abnormal data, thereby improving the efficiency of line monitoring and maintenance.

Description

Technical Field

[0001] The present invention belongs to the field of power transmission line maintenance, and in particular relates to a monitoring method, device, equipment and storage medium for an overhead power transmission line. Background Art

[0002] With the rapid development of the economy, the scale of power supply is constantly expanding. Transmission lines are an important part of the power system. Their safe and stable operation is crucial to ensuring the reliability of power supply. However, with the continuous expansion of the scale of the power system and the complexity of the grid structure, transmission lines are facing more and more challenges and problems. The reliability and safety of transmission lines have become the focus of attention. For example, transmission lines often have equipment failures, power load imbalance, and ground short circuits. The occurrence of these problems will seriously affect the operating efficiency of the power system and the stability of power supply. Especially in the area near the thermal power plant, the distribution of transmission lines is relatively dense, and the time and money costs required for traditional manual inspection methods are relatively high. Furthermore, in the drone-assisted power grid inspection method, considering that the trajectory optimization of drones is relatively complex and the flight inertia is large, drones are only suitable for single-line and long-distance transmission lines. It is not suitable for full coverage inspection of dense transmission lines. Therefore, real-time monitoring and accurate identification of possible problems and abnormal conditions in transmission lines are of great significance for the safe operation and intelligent management of power systems.

[0003] The existing technology establishes a virtual scene of a UAV digital twin model and a standard model of the inspection target. During operation, the UAV synchronizes its own position information to the UAV digital twin model in real time through the Beidou communication satellite; then the target turning angle of the UAV is calculated based on the position shape of the UAV virtual model and the position of the target to be detected in the virtual scene, and the target turning angle is sent to the UAV. After receiving the turning angle data, the UAV is driven to stably operate toward the area to be inspected to achieve the purpose of real-time optimization of the UAV camera orientation; then the UAV sends the video image of the detected transmission line and other data to the platform layer, the platform layer structures the transmission line video, and performs design modeling and analysis based on the transmission line; finally, the processed data is integrated to establish a full-factor static scene and a data-driven model, and finally a real-time updated "digital twin" is established to support decision-making on various activities in the life cycle of the transmission line.

[0004] However, due to insufficient data processing or model training caused by environmental complexity, existing technologies still have problems such as low accuracy, slow response speed, and insufficient adaptability to complex environments in terms of intelligent line identification and target positioning, which limits the efficiency and intelligence level of line monitoring and maintenance. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a monitoring method, device, equipment and storage medium for an overhead transmission line.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, the present application provides a method for monitoring an overhead transmission line, the method comprising:

[0008] Acquire image data and point cloud data of a target overhead transmission line, and construct a digital twin model based on the point cloud data; the digital twin model includes a power element model and a UAV simulation model; use an adaptive segmentation strategy to optimize the KD-Tree construction process, render the point cloud data through the optimized KD-Tree, and construct the power element model; plan a path for the UAV through the UAV simulation model, and map the real-time data of the target overhead transmission line to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead transmission line; monitor the real-time data, and when there is abnormal data in the real-time data, locate the fault position in the power element model according to the abnormal data.

[0009] Optionally, the process of optimizing the tree construction of KD-Tree by using the adaptive segmentation strategy includes:

[0010] An axis is selected for segmentation according to the distribution of the data set of the point cloud data, and then the position of the segmentation point is adjusted according to the density and distribution of the data set to segment the data set.

[0011] Optionally, rendering the point cloud data by using the optimized KD-Tree includes:

[0012] For each leaf node of the KD-Tree, the point cloud subset stored therein is represented in the form of Gaussian distribution, and a group of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling points are discrete representations of point cloud data; starting from the root node, traverse the KD-Tree, and determine the Gaussian weight of each sampling point in the leaf node by the following formula;

[0013]

[0014] Wherein, e is a natural constant, d represents the distance from the pixel position of the sampling point to the Gaussian center point, and σ represents the standard deviation of the leaf node where the sampling point is located; according to the Gaussian weight of each sampling point in the leaf node, the interpolation result of the Gaussian function in the leaf node is determined by the following formula:

[0015]

[0016] Where N represents the number of sampling points in the leaf node, w i is the Gaussian weight at the i-th sampling point, v i is the value at the i-th sampling point; and Gaussian weighted interpolation operation is performed on all leaf nodes according to the interpolation result.

[0017] Optionally, the UAV simulation model is used to perform path planning for the UAV, including:

[0018] The UAV simulation model plans the path of the UAV through a graph convolutional recurrent attention network, and the graph convolutional recurrent attention network includes a convolutional neural network, a recurrent neural network and an attention mechanism; the convolutional neural network extracts a feature map through the image data, the recurrent neural network obtains time series features according to the feature map, and the attention mechanism generates a context vector according to the time series features. Finally, the graph convolution layer in the graph convolutional recurrent attention network generates node features at each position in the path according to the context vector, and determines the UAV path according to the node features.

[0019] Optionally, the performing path planning for the drone through a graph convolutional recurrent attention network includes:

[0020] The convolutional neural network extracts the feature map F through the image data:

[0021] F = CNN(X)

[0022] Among them, X is the image data, and F is the high-level feature map finally output by the pooling layer of the convolutional neural network;

[0023] The recurrent neural network processes the feature graph F to obtain the time series feature {h 1 ,h 2 ,...,h n};

[0024] The attention mechanism determines the attention weight of the hidden state at each time step based on the time series characteristics:

[0025] e i =v T tanh(W e h i +b e )

[0026] Among them, v T Represents the attention score at time step i, e i is the attention score v at time step i T The transpose of the weight vector, tanh is the hyperbolic tangent activation function, W e The weight matrix is ​​a trainable parameter used to transform the input hidden state h iLinear transformation into a new feature representation, h i is the hidden state at time step i, b e is the bias vector;

[0027] s i =softmax(f(e i ))

[0028] Among them, s i is the attention weight of node i, f(e i ) is a function of the characteristics of node i and the environmental conditions;

[0029] Finally, the context vector c is generated:

[0030]

[0031] Among them, t represents the length of the time step;

[0032] The graph convolution layer models the drone path planning problem as a graph structure and uses the context vector c obtained from the attention mechanism as the initial feature vector in the graph structure:

[0033] H 0 =c

[0034] Among them, H 0 is the initial node feature matrix;

[0035] And using the spatial relationship and terrain information between nodes, the feature representation of each node is extracted:

[0036]

[0037] Among them, H (l+1) is the node feature matrix of the l+1th layer, K is the number of multi-scale feature maps, α k is the weight of the feature map, is the adjacency matrix of the kth feature graph, W (l) is the weight matrix, is the degree matrix and σ is the activation function.

[0038] Optionally, the digital twin model also includes an inspection scenario model; through computer graphics technology, based on real geographic data and power engineering design drawings, a three-dimensional scenario model of the power transmission scenario is constructed, and according to environmental factors, parameters under different scenarios are preset to construct an inspection scenario model to provide inspection simulation and testing in a real environment.

[0039] Optionally, the digital twin model also includes a data acquisition model, through which operating data in the inspection scenario model is collected in real time.

[0040] In a second aspect, the present application provides a monitoring device for an overhead transmission line, the device comprising:

[0041] An acquisition module, used to acquire image data and point cloud data of a target overhead transmission line;

[0042] A construction module is used to construct a digital twin model according to the point cloud data; the digital twin model includes a power element model and a UAV simulation model; an adaptive segmentation strategy is used to optimize the KD-Tree construction process, and the point cloud data is rendered by the optimized KD-Tree to construct the power element model;

[0043] The inspection module is used to plan the path of the drone through the drone simulation model, and map the real-time data of the target overhead transmission line to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead transmission line; the real-time data is monitored, and when there is abnormal data in the real-time data, the fault position is located in the power element model according to the abnormal data.

[0044] Optionally, the construction module is further used to select an axis for segmentation according to the distribution of the data set of the point cloud data, and then adjust the position of the segmentation point according to the density and distribution of the data set to segment the data set.

[0045] Optionally, the construction module is further used to represent the point cloud subset stored in each leaf node of the KD-Tree in the form of Gaussian distribution, and a group of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling points are discrete representations of the point cloud data; starting from the root node, traverse the KD-Tree, and determine the Gaussian weight of each sampling point in the leaf node by the following formula;

[0046]

[0047] Wherein, e is a natural constant, d represents the distance from the pixel position of the sampling point to the Gaussian center point, and σ represents the standard deviation of the leaf node where the sampling point is located; according to the Gaussian weight of each sampling point in the leaf node, the interpolation result of the Gaussian function in the leaf node is determined by the following formula:

[0048]

[0049] Where N represents the number of sampling points in the leaf node, w i is the Gaussian weight at the i-th sampling point, v i is the value at the i-th sampling point; and Gaussian weighted interpolation operation is performed on all leaf nodes according to the interpolation result.

[0050] Optionally, the inspection module is also used to control the UAV simulation model to plan the path of the UAV through a graph convolutional recurrent attention network, and the graph convolutional recurrent attention network includes a convolutional neural network, a recurrent neural network and an attention mechanism; the convolutional neural network extracts a feature map through the image data, the recurrent neural network obtains time series features based on the feature map, and the attention mechanism generates a context vector based on the time series features. Finally, the graph convolution layer in the graph convolutional recurrent attention network generates node features for each position in the path based on the context vector, and the UAV path is determined based on the node features.

[0051] Optionally, the inspection module is further used to extract a feature map F according to the image data through the convolutional neural network:

[0052] F = CNN(X)

[0053] Among them, X is the image data, and F is the high-level feature map finally output by the pooling layer of the convolutional neural network;

[0054] The recurrent neural network processes the feature graph F to obtain the time series feature {h 1 ,h 2 ,...,h n};

[0055] The attention mechanism determines the attention weight of the hidden state at each time step based on the time series characteristics:

[0056] e i =v T tanh(W e h i +b e )

[0057] Among them, v T Represents the attention score at time step i, e i is the attention score v at time step i T The transpose of the weight vector, tanh is the hyperbolic tangent activation function, W e The weight matrix is ​​a trainable parameter used to transform the input hidden state h i Linear transformation into a new feature representation, h i is the hidden state at time step i, b e is the bias vector;

[0058] s i =softmax(f(e i ))

[0059] Among them, s i is the attention weight of node i, f(e i ) is a function of the characteristics of node i and the environmental conditions;

[0060] Finally, the context vector c is generated:

[0061]

[0062] Among them, t represents the length of the time step;

[0063] The graph convolution layer models the drone path planning problem as a graph structure and uses the context vector c obtained from the attention mechanism as the initial feature vector in the graph structure:

[0064] H 0 =c

[0065] Among them, H 0 is the initial node feature matrix;

[0066] And using the spatial relationship and terrain information between nodes, the feature representation of each node is extracted:

[0067]

[0068] Among them, H (l+1) is the node feature matrix of the l+1th layer, K is the number of multi-scale feature maps, α k is the weight of the feature map, is the adjacency matrix of the kth feature graph, W (l) is the weight matrix, is the degree matrix and σ is the activation function.

[0069] Optionally, the digital twin model also includes an inspection scenario model, and the construction module is also used to construct a three-dimensional scenario model of the power transmission scenario based on real geographic data and power engineering design drawings through computer graphics technology, and preset parameters under different scenarios according to environmental factors to construct an inspection scenario model to provide inspection simulation and testing in a real environment.

[0070] Optionally, the digital twin model also includes a data acquisition model, and the inspection module is further used to collect operating data in the inspection scenario model in real time through the data acquisition model.

[0071] In a third aspect, the present application provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned overhead transmission line monitoring method is implemented.

[0072] In a fourth aspect, the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for monitoring overhead transmission lines when executing the program.

[0073] The disturbance suppression method of the provided quadrotor UAV system has the following beneficial effects:

[0074] The KD-Tree building process is optimized through an adaptive segmentation strategy, and the point cloud data is rendered through the optimized KD-Tree to construct the power element model, which represents the model of power equipment and lines; in this way, the power equipment and lines of the target overhead transmission line are digitally modeled, which can reduce the depth of the tree, improve search efficiency, accuracy of intelligent recognition and response speed, and convert large-scale point cloud data into continuous density distribution through the optimized KD-Tree to achieve accurate analysis of the line status; the drone path is planned through the drone simulation model to adapt to complex environmental changes, and the real-time operation data of the target overhead transmission line is mapped to the power element model through the drone simulation model, and the location of possible faults is located according to the abnormal data in the presence of abnormal data, thereby improving the efficiency of line monitoring and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 A schematic flow chart of a method for monitoring an overhead power transmission line provided in the present application according to an exemplary embodiment.

[0077] Figure 2 This is a block diagram of a monitoring device for an overhead power transmission line provided in accordance with an exemplary embodiment of the present application.

[0078] Figure 3 A schematic diagram of a computer device for implementing a monitoring method for overhead power transmission lines provided in the present application. DETAILED DESCRIPTION

[0079] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0080] The present disclosure is described below in conjunction with specific embodiments.

[0081] Figure 1 is a flow chart of a monitoring method for an overhead transmission line provided by the present disclosure according to an exemplary embodiment, such as Figure 1 As shown, including:

[0082] S101. Construct a digital twin model of the target overhead transmission line.

[0083] Specifically, the image data and point cloud data of the target overhead transmission line can be obtained first, and a digital twin model can be constructed based on the point cloud data.

[0084] For example, high-resolution camera equipment can be used to capture images from multiple angles and perform three-dimensional reconstruction to generate point cloud data. Image acquisition can generally be performed using a high-resolution camera equipped with a drone, and a digital twin model of the target overhead transmission line is constructed based on the collected point cloud data.

[0085] The digital twin model includes: inspection scene model, power element model, drone simulation model, data acquisition model, feature extraction model. The drone simulation model includes target detection model and inspection operation model.

[0086] The inspection scene model is used to simulate complex power transmission scenes. Specifically, through computer graphics technology, based on real geographic data and power engineering design drawings, a three-dimensional scene model of the power transmission scene is constructed, and according to environmental factors, parameters in different scenes are preset to construct the inspection scene model to provide inspection simulation and testing in a real environment. The power transmission scene includes three-dimensional scene models of various lines, towers, insulators and other elements, and considers factors such as terrain and meteorological conditions, and presets parameters in different scenes, such as weather conditions, terrain complexity, etc., to provide inspection simulation and testing in a real environment.

[0087] The power element model is used to accurately digitally model power equipment and lines. Data is collected using sensors carried by drones, and a Gaussian point cloud rendering method based on KD-Tree is used to render large-scale point clouds to generate the power element model.

[0088] The UAV simulation model is used to virtually simulate the flight and operation of the UAV, simulate the flight conditions and working scenarios in different environments, and simulate the flight conditions and working scenarios of the UAV in different environments based on the principles of flight mechanics and control algorithms, taking into account the UAV's dynamic characteristics, flight attitude control, sensor data acquisition, etc., to verify the reliability and performance of the system.

[0089] The data acquisition model is used to collect the operating data in the inspection scene model in real time; based on the inspection process of the UAV simulation model, various operating data are collected in real time, including images, videos and sensor data, providing a basis for subsequent analysis and processing.

[0090] The feature extraction model is used to extract useful feature information from the collected data for tasks such as target recognition, fault detection and state assessment.

[0091] The target detection model is used to identify abnormal data in the operating data in the power transmission scenario, including line anomalies, equipment defects, etc., and locate the fault location that generates the abnormal data so as to discover and handle the problem in time.

[0092] The inspection operation model is used to plan and execute the inspection operation of the drone, including path planning, route optimization, operation strategy, etc. Path planning and route optimization are performed based on the target information output by the target detection model and the location information provided by the perception and positioning module to ensure comprehensive coverage and effective monitoring of power lines.

[0093] In one embodiment, an adaptive segmentation strategy can be used to optimize the KD-Tree tree building process, and the point cloud data is rendered by the optimized KD-Tree to build the power element model. KD-Tree is a data structure for efficiently organizing multidimensional data, which can accelerate search and query operations. Combining KD-Tree with Gaussian point cloud rendering technology can improve efficiency when rendering large-scale point clouds. At the same time, a KD-Tree construction method with an adaptive segmentation strategy is introduced in the tree building process to improve the efficiency and quality of point cloud data processing. The specific steps are as follows:

[0094] S1. Construct a KD-Tree. For point cloud data, select an axis in the data set for segmentation, usually the x, y or z axis, and divide the data set into two subsets. Continue segmenting for each subset, and gradually segment the point cloud data set into a tree-structured KD-Tree. Repeat this process until the number of points in each subset reaches a certain threshold or the depth of the tree reaches a predetermined maximum depth. This application uses an adaptive segmentation strategy to optimize the KD-Tree tree building process and dynamically adjusts the segmentation method according to the characteristics of the data set. Specifically, select an axis for segmentation based on the data set distribution of the point cloud data, and then adjust the position of the segmentation point based on the density and distribution of the data set to segment the data set to reduce the depth of the tree and improve search efficiency.

[0095] S2. Use Gaussian distribution to represent the point cloud subset. For each leaf node of KD-Tree, the point cloud subset stored in it is represented in the form of Gaussian distribution. Each leaf node contains a set of sampling points, which are collectively represented as a Gaussian function. The specific method is: calculate the centroid of all points in the subset as the center point P of the Gaussian function c , and the standard deviation σ of the Gaussian function is determined by density weighting. This representation method converts the originally discrete point cloud data into a continuous density distribution, which is convenient for subsequent efficient rendering.

[0096] S3. Traverse the KD-Tree and perform Gaussian weighted interpolation operations on all leaf nodes.

[0097] When rendering, start from the root node and traverse the KD-Tree. For each non-leaf node, check the distance between the spatial area represented by the node and the current pixel position to determine whether it is necessary to further traverse the child nodes of the node. When reaching a leaf node, calculate the Gaussian function weight: for each leaf node of the KD-Tree, the point cloud subset stored in it is represented in the form of a Gaussian distribution, and a set of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling point is a discrete representation of the point cloud data; for the Gaussian function in the leaf node, given its parameter center point P c , standard deviation σ and pixel position p in the rendered image, calculate the distance d from this pixel position to the Gaussian center point. Use the Gaussian weighted interpolation formula to calculate the contribution of the Gaussian function to the pixel.

[0098] Starting from the root node, traverse the KD-Tree and determine the Gaussian weight of each sampling point in the leaf node using the following formula;

[0099]

[0100] Among them, e is a natural constant, d represents the distance from the pixel position of the sampling point to the Gaussian center point, and σ represents the standard deviation of the leaf node where the sampling point is located;

[0101] Superposition Gaussian function: Perform Gaussian weighted interpolation operation according to the Gaussian weight of each sampling point in the leaf node. Specifically, the Gaussian weight w of each sampling point position in the leaf node obtained by the above calculation is added to i and the value v of each sampling point i Multiply them together and perform weighted averaging on all sampling points. The interpolation result of the Gaussian function in the leaf node is determined by the following formula:

[0102]

[0103] Where N represents the number of sampling points in the leaf node, w i is the Gaussian weight at the i-th sampling point, v i is the value at the i-th sampling point, such as a color value or a density value. By taking a weighted average of all sampling points, the interpolation result of the Gaussian function in the leaf node is obtained, and then a Gaussian weighted interpolation operation is performed on all leaf nodes according to the interpolation result.

[0104] In this way, the present invention can transform large-scale point cloud data into continuous density distribution with the help of KD-tree 3D Gaussian method, and realize accurate analysis of line status through Gaussian weighted interpolation operation.

[0105] S102. Map the operating data of the target overhead transmission line to the digital twin model in real time.

[0106] Specifically, the operation data of the target overhead transmission line is acquired in real time through the UAV simulation model, and the operation data is mapped to the power element model.

[0107] In one embodiment, the drone is path-planned through a drone simulation model, and the real-time data of the target overhead transmission line is mapped to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead transmission line, including images, videos, and sensor data. The drone simulation model has the ability to perform high-performance model rendering and modeling and to achieve association between model characteristics to meet the high-fidelity requirements of system-environment interaction. The module can map the real drone system-related modules to the physical simulation engine of Unity through data transmission, realize effective simulation in a high-fidelity environment, and can interact with the simulation sensors to achieve more realistic algorithm and inspection task testing. Based on the original drone simulation model, deep learning technology is introduced to optimize the drone's path planning and enhance its environmental perception capabilities.

[0108] The path planning steps of this deep learning are as follows:

[0109] S1. Data collection and processing: Collect a large amount of drone flight data from real scenarios and convert it into a format suitable for deep learning model processing. Data preprocessing includes technical means such as noise removal and data enhancement to improve the generalization ability of the model.

[0110] S2. Construction of deep learning model: In view of the limitations of traditional deep learning models in path planning optimization, this application proposes a new deep learning model, which combines the structures of RNN (Recurrent Neural Network), CNN (Convol utional Neural Networks) and attention mechanism, and introduces the idea of ​​GNN (Graph Neural Network), and innovatively generates a graph convolutional recurrent attention network. The graph convolutional recurrent attention network is an end-to-end deep learning architecture designed to solve the problems of drone path planning and environmental perception. The network combines GCN, RNN and attention mechanism, and has the ability to automatically learn and understand the impact of different terrain and weather conditions on path planning, and can dynamically adjust the path planning strategy to adapt to different environments. In this application, the drone simulation model plans the path of the drone through a graph convolutional recurrent attention network, which includes a convolutional neural network, a recurrent neural network and an attention mechanism; the convolutional neural network extracts a feature map through the image data, the recurrent neural network obtains time series features based on the feature map, and the attention mechanism generates a context vector based on the time series features. Finally, the graph convolution layer in the graph convolutional recurrent attention network generates node features for each position in the path based on the context vector, and determines the drone path based on the node features.

[0111] Specifically, the convolutional neural network extracts the feature map F through image data:

[0112] F = CNN(X)

[0113] Among them, X is the image data, and F is the high-level feature map finally output by the pooling layer of the convolutional neural network;

[0114] The recurrent neural network processes the feature graph F to obtain the time series feature {h 1 ,h 2 ,...,h n};

[0115] The attention mechanism determines the attention weight of the hidden state at each time step based on the time series characteristics:

[0116] e i =v T tanh(W e h i +b e )

[0117] Among them, v T Represents the attention score at time step i, e i is the attention score v at time step iT The transpose of the weight vector, tanh is the hyperbolic tangent activation function, W e The weight matrix is ​​a trainable parameter used to transform the input hidden state h i Linear transformation into a new feature representation, h i is the hidden state at time step i, b e is the bias vector;

[0118] s i =softmax(f(e i ))

[0119] Among them, s i is the attention weight of node i, f(e i ) is a function of the characteristics of node i and the environmental conditions;

[0120] Finally, the context vector c is generated:

[0121]

[0122] Among them, t represents the length of the time step;

[0123] The graph convolution layer models the drone path planning problem as a graph structure and uses the context vector c obtained from the attention mechanism as the initial feature vector in the graph structure:

[0124] H 0 =c

[0125] Among them, H 0 is the initial node feature matrix;

[0126] And using the spatial relationship and terrain information between nodes, the feature representation of each node is extracted:

[0127]

[0128] Among them, H (l+1) is the node feature matrix of the l+1th layer, K is the number of multi-scale feature maps, α k is the weight of the feature map, is the adjacency matrix of the kth feature graph, W (l) is the weight matrix, is the degree matrix and σ is the activation function.

[0129] In this way, the graph convolution recurrent attention network adopted by the present invention combines graph convolution, recurrent neural network and attention mechanism, which can automatically learn and understand the impact of different terrain and weather conditions on path planning, and adjust the path planning strategy in real time.

[0130] The inspection scene model uses advanced digital twin technology, which combines multi-source heterogeneous 3D data processing and rendering technologies such as BIM / GIM, laser point cloud, oblique photography, 3DMax, and uses the real-time 3D application development tool Unity3D to build a high-fidelity physical engine.

[0131] S103: Monitor the real-time data and issue an alarm if there is abnormal data in the real-time data.

[0132] Among them, the alarm information may include the precise positioning information and processing plan of the abnormal data.

[0133] Specifically, the real-time data can be monitored, and when there is abnormal data in the real-time data, the fault location can be located in the power element model according to the abnormal data.

[0134] In addition, the processing scheme may be a pre-set processing scheme for related faults, or a processing scheme for historical faults, which will not be described in detail here.

[0135] In this way, the safety and reliability of the line are improved through timely and effective early warning information. Maintenance personnel can carry out targeted maintenance work based on the positioning information and processing solutions in the alarm information, which improves the efficiency and accuracy of maintenance work, avoids unnecessary waste of manpower and material resources, effectively reduces maintenance costs and improves maintenance efficiency.

[0136] By adopting the above method, the tree building process of KD-Tree is optimized through an adaptive segmentation strategy, and the point cloud data is rendered through the optimized KD-Tree to construct the power element model, which represents the model of power equipment and lines; in this way, the power equipment and lines of the target overhead transmission line are digitally modeled, which can reduce the depth of the tree, improve the search efficiency, the accuracy of intelligent recognition and the response speed, and convert large-scale point cloud data into a continuous density distribution through the optimized KD-Tree to achieve accurate analysis of the line status; the drone path is planned through the drone simulation model to adapt to complex environmental changes, and the real-time operation data of the target overhead transmission line is mapped to the power element model through the drone simulation model, and the location of possible faults is located according to the abnormal data in the presence of abnormal data, thereby improving the efficiency of line monitoring and maintenance.

[0137] Figure 2 The present application provides a monitoring device for an overhead power transmission line according to an exemplary embodiment, the device comprising:

[0138] An acquisition module 201 is used to acquire image data and point cloud data of a target overhead transmission line;

[0139] A construction module 202 is used to construct a digital twin model according to the point cloud data; the digital twin model includes a power element model and a UAV simulation model; an adaptive segmentation strategy is used to optimize the KD-Tree construction process, and the point cloud data is rendered by the optimized KD-Tree to construct the power element model;

[0140] The inspection module 203 is used to plan the path of the drone through the drone simulation model, and map the real-time data of the target overhead transmission line to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead transmission line, including images, videos and sensor data; the real-time data is monitored, and when there is abnormal data in the real-time data, the fault position is located in the power element model according to the abnormal data.

[0141] Optionally, the construction module 202 is further configured to select an axis for segmentation according to the distribution of the point cloud data set, and then adjust the position of the segmentation point according to the density and distribution of the data set to segment the data set.

[0142] Optionally, the construction module 202 is further used to represent the point cloud subset stored in each leaf node of the KD-Tree in the form of Gaussian distribution, and a group of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling point is a discrete representation of the point cloud data; starting from the root node, traverse the KD-Tree, and determine the Gaussian weight of each sampling point in the leaf node by the following formula;

[0143]

[0144] Wherein, e is a natural constant, d represents the distance from the pixel position of the sampling point to the Gaussian center point, and σ represents the standard deviation of the leaf node where the sampling point is located; according to the Gaussian weight of each sampling point in the leaf node, the interpolation result of the Gaussian function in the leaf node is determined by the following formula:

[0145]

[0146] Where N represents the number of sampling points in the leaf node, w i is the Gaussian weight at the i-th sampling point, v i is the value at the i-th sampling point; Gaussian weighted interpolation operation is performed on all leaf nodes according to the interpolation result.

[0147] Optionally, the inspection module 203 is also used to control the UAV simulation model to plan the path of the UAV through a graph convolutional recurrent attention network, which includes a convolutional neural network, a recurrent neural network and an attention mechanism; the convolutional neural network extracts a feature map through the image data, the recurrent neural network obtains time series features based on the feature map, the attention mechanism generates a context vector based on the time series features, and finally, the graph convolution layer in the graph convolutional recurrent attention network generates node features for each position in the path based on the context vector, and determines the UAV path based on the node features.

[0148] Optionally, the inspection module 203 is further configured to extract a feature graph F according to the image data through the convolutional neural network:

[0149] F = CNN(X)

[0150] Among them, X is the image data, and F is the high-level feature map finally output by the pooling layer of the convolutional neural network;

[0151] The recurrent neural network processes the feature graph F to obtain the time series feature {h 1 ,h 2 ,...,h n};

[0152] The attention mechanism determines the attention weight of the hidden state at each time step based on the time series characteristics:

[0153] e i =v T tanh(W e h i +b e )

[0154] Among them, v T Represents the attention score at time step i, e i is the attention score v at time step i T The transpose of the weight vector, tanh is the hyperbolic tangent activation function, W e The weight matrix is ​​a trainable parameter used to transform the input hidden state h i Linear transformation into a new feature representation, h i is the hidden state at time step i, b e is the bias vector;

[0155] s i =softmax(f(e i ))

[0156] Among them, s i is the attention weight of node i, f(e i ) is a function of the characteristics of node i and the environmental conditions;

[0157] Finally, the context vector c is generated:

[0158]

[0159] Among them, t represents the length of the time step;

[0160] The graph convolution layer models the drone path planning problem as a graph structure and uses the context vector c obtained from the attention mechanism as the initial feature vector in the graph structure:

[0161] H 0 =c

[0162] Among them, H 0 is the initial node feature matrix;

[0163] And using the spatial relationship and terrain information between nodes, the feature representation of each node is extracted:

[0164]

[0165] Among them, H (l+1) is the node feature matrix of the l+1th layer, K is the number of multi-scale feature maps, α k is the weight of the feature map, is the adjacency matrix of the kth feature graph, W (l) is the weight matrix, is the degree matrix and σ is the activation function.

[0166] Optionally, the digital twin model also includes an inspection scenario model. The construction module 202 is also used to construct a three-dimensional scenario model of the power transmission scenario based on real geographic data and power engineering design drawings through computer graphics technology, and preset parameters under different scenarios according to environmental factors to construct an inspection scenario model to provide inspection simulation and testing in a real environment.

[0167] Optionally, the digital twin model also includes a data acquisition model, and the inspection module 203 is also used to collect operating data in the inspection scenario model in real time through the data acquisition model.

[0168] The above-mentioned device is used to optimize the tree building process of KD-Tree through an adaptive segmentation strategy, and the point cloud data is rendered through the optimized KD-Tree to construct the power element model, which represents the model of power equipment and lines; in this way, the power equipment and lines of the target overhead transmission line are digitally modeled, which can reduce the depth of the tree, improve the search efficiency, the accuracy of intelligent recognition and the response speed, and convert large-scale point cloud data into a continuous density distribution through the optimized KD-Tree to achieve accurate analysis of the line status; the drone path is planned through the drone simulation model to adapt to complex environmental changes, and the real-time operation data of the target overhead transmission line is mapped to the power element model through the drone simulation model, and the location of possible faults is located according to the abnormal data in the presence of abnormal data, thereby improving the efficiency of line monitoring and maintenance.

[0169] The present application also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A monitoring method for overhead transmission lines is provided.

[0170] This application also provides Figure 3 The structural diagram of the computer device shown in FIG. Figure 3 As shown in the figure, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A monitoring method for overhead transmission lines is provided.

[0171] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented 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.

[0172] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0173] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0174] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0175] It should be noted that the above-described specific implementation methods can enable those skilled in the art to more fully understand the invention, but do not limit the invention in any way. Therefore, although the present specification and embodiments have described the invention in detail, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the protection scope of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

Claims

1. A method for monitoring an overhead power transmission line, characterized in that: The method comprises: Acquire image data and point cloud data of a target overhead transmission line, and construct a digital twin model based on the point cloud data; the digital twin model includes a power element model and a UAV simulation model; Adopting an adaptive segmentation strategy to optimize the KD-Tree building process, rendering the point cloud data through the optimized KD-Tree, and constructing the power element model; The UAV is path-planned by the UAV simulation model, and the real-time data of the target overhead power transmission line is mapped to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead power transmission line; Monitoring the real-time data, and if there is abnormal data in the real-time data, locating the fault position in the power element model according to the abnormal data; The power factor model adopts an adaptive segmentation strategy to optimize the KD-Tree tree building process, which includes: Selecting an axis for segmentation according to the distribution of the data set of the point cloud data, and then adjusting the position of the segmentation point according to the density and distribution of the data set to segment the data set; The rendering of the point cloud data by using the optimized KD-Tree comprises: For each leaf node of the KD-Tree, the point cloud subset stored therein is represented in the form of Gaussian distribution, and a set of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling points are a discrete representation of the point cloud data; Starting from the root node, traverse the KD-Tree and determine the Gaussian weight of each sampling point in the leaf node by the following formula; in, is a natural constant, Characterizes the distance from the pixel position of the sampling point to the Gaussian center point, σ Characterizes the standard deviation of the leaf node where the sampling point is located; According to the Gaussian weight of each sampling point in the leaf node, the interpolation result of the Gaussian function in the leaf node is determined by the following formula: in, Indicates the number of sampling points in the leaf node, It is Gaussian weights at sampling points, It is The value at each sampling point; Performing a Gaussian weighted interpolation operation on all leaf nodes according to the interpolation result; The UAV simulation model is used to plan the path of the UAV, including: The UAV simulation model plans the path of the UAV through a graph convolutional recurrent attention network, which includes a convolutional neural network, a recurrent neural network and an attention mechanism; The convolutional neural network extracts feature maps through the image data, the recurrent neural network obtains time series features based on the feature maps, the attention mechanism generates a context vector based on the time series features, and finally the graph convolution layer in the graph convolution recurrent attention network generates node features for each position in the path based on the context vector, and determines the drone path based on the node features.

2. The method according to claim 1, characterized in that The path planning of the drone through the graph convolutional recurrent attention network includes: The convolutional neural network extracts feature maps from image data : in, is the image data, It is the high-level feature map finally output by the pooling layer of the convolutional neural network; Recurrent Neural Network to Feature Map Processing to obtain time series features ; The attention mechanism determines the attention weight of the hidden state at each time step based on the time series characteristics: in, Characterizing time steps The attention score, is the time step Attention score The transpose of the weight vector, is the hyperbolic tangent activation function, The weight matrix is ​​a trainable parameter that transforms the hidden state of the input Linear transformation into a new feature representation, is the time step The hidden state of is the bias vector; in, Is a node The attention weight, Is a node The characteristics of the function and environmental conditions; Finally, the context vector is generated : in, The step size representing the time step; The graph convolution layer models the drone path planning problem as a graph structure and converts the context vector obtained from the attention mechanism into As the initial eigenvector in the graph structure: in, is the initial node feature matrix; And using the spatial relationship and terrain information between nodes, the feature representation of each node is extracted: in, For the The node feature matrix of the layer, is the number of multi-scale feature maps, is the weight of the feature map, It is The adjacency matrix of the feature graph is is the weight matrix, is the degree matrix, is the activation function.

3. The method according to claim 1, characterized in that The digital twin model also includes an inspection scenario model; through computer graphics technology, based on real geographic data and power engineering design drawings, a three-dimensional scenario model of the power transmission scenario is constructed, and according to environmental factors, parameters under different scenarios are preset to construct an inspection scenario model to provide inspection simulation and testing in a real environment.

4. The method according to claim 3, characterized in that The digital twin model also includes a data acquisition model, through which the operating data in the inspection scenario model is collected in real time.

5. A monitoring device for an overhead power transmission line, characterized in that: The device comprises: An acquisition module, used to acquire image data and point cloud data of a target overhead transmission line; A construction module is used to construct a digital twin model according to the point cloud data; the digital twin model includes a power element model and a UAV simulation model; an adaptive segmentation strategy is used to optimize the KD-Tree construction process, and the point cloud data is rendered by the optimized KD-Tree to construct the power element model; The inspection module is used to plan the path of the UAV through the UAV simulation model, and map the real-time data of the target overhead power transmission line to the power element model according to the planned path; the real-time data represents the real-time operation data of the target overhead power transmission line; monitor the real-time data, and when there is abnormal data in the real-time data, locate the fault position in the power element model according to the abnormal data; The construction module is also used to select an axis for segmentation according to the distribution of the data set of the point cloud data, and then adjust the position of the segmentation point according to the density and distribution of the data set to segment the data set; For each leaf node of the KD-Tree, the point cloud subset stored therein is represented in the form of Gaussian distribution, and a set of sampling points contained in each leaf node are collectively represented as a Gaussian function; the sampling points are a discrete representation of the point cloud data; Starting from the root node, traverse the KD-Tree and determine the Gaussian weight of each sampling point in the leaf node by the following formula; in, is a natural constant, Represents the distance from the pixel position of the sampling point to the Gaussian center point, and σ represents the standard deviation of the leaf node where the sampling point is located; According to the Gaussian weight of each sampling point in the leaf node, the interpolation result of the Gaussian function in the leaf node is determined by the following formula: in, Indicates the number of sampling points in the leaf node, It is Gaussian weights at sampling points, It is The value at each sampling point; Performing a Gaussian weighted interpolation operation on all leaf nodes according to the interpolation result; The inspection module is also used to plan the path of the drone using a drone simulation model through a graph convolutional recurrent attention network, wherein the graph convolutional recurrent attention network includes a convolutional neural network, a recurrent neural network and an attention mechanism; The convolutional neural network extracts feature maps through the image data, the recurrent neural network obtains time series features based on the feature maps, the attention mechanism generates a context vector based on the time series features, and finally the graph convolution layer in the graph convolution recurrent attention network generates node features for each position in the path based on the context vector, and determines the drone path based on the node features.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of claims 1 to 4 is implemented.

Images