AI-based power line simulation inspection efficiency optimization method and system

Abstract

The invention discloses an AI-based power line simulation inspection efficiency optimization method and system, and the method comprises the following steps: carrying out the line inspection of a power line through an image collection device carried by an unmanned plane, obtaining a line image, carrying out the recognition of the line image through a lightweight image recognition model, obtaining a fault point of the power line, and carrying out the optimization of the line simulation inspection efficiency of the power line. According to the obtained fault points, performing path planning by adopting an improved grey wolf optimization algorithm to obtain a detection route, flying along the detection route through a multi-modal data acquisition device carried by the unmanned aerial vehicle, acquiring multi-modal data of each fault point, judging the acquired multi-modal data through a fault diagnosis model to obtain a fault category probability, and determining the fault category probability according to the fault category probability. And inputting the fault point position and the fault category probability into the three-dimensional model of the power line to realize visual display of the fault category probability. According to the method, the lightweight image recognition model, the multi-modal fusion analysis and the three-dimensional model of the power line are combined, so that the inspection efficiency and the fault diagnosis precision are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power line imitation line inspection efficiency, and particularly to an AI-based power line imitation line inspection efficiency optimization method and system. Background Art

[0002] In the operation and maintenance of power systems, the inspection of power lines is an important means to ensure power supply safety and reliability. However, traditional inspection methods are inefficient, risky, and have limited coverage, making it difficult to meet the growing operation and maintenance needs of modern power grids. In recent years, with the maturity of unmanned aerial vehicle (UAV) technology, artificial intelligence, and three-dimensional reconstruction technology, the intelligent and digital power line inspection mode has gradually become the industry development trend.

[0003] UAV inspection has the characteristics of high efficiency, safety, and wide coverage. It can complete full-line inspections in high-risk areas and collect multi-modal data affecting the health status of equipment. However, UAV inspection still faces some challenges during deployment, such as how to quickly detect fault points in massive line data, how to optimize flight routes to reduce inspection time and energy consumption, how to deeply integrate multi-modal data to improve the accuracy of fault diagnosis, and how to visually present fault information through three-dimensional visualization to assist operation and maintenance decisions. Summary of the Invention

[0004] To solve the problems in related technologies, the present application provides an AI-based power line imitation line inspection efficiency optimization method and system, which solves the problems mentioned in the background art.

[0005] To achieve the above object, the present invention provides the following technical solution: An AI-based power line imitation line inspection efficiency optimization method, comprising the following steps: Step S1: Use an image acquisition device carried by a UAV to inspect the power line and obtain images along the line; Step S2: Identify the obtained images along the line through a lightweight image recognition model to obtain the fault points of the power line; Step S3: According to the obtained fault points, use an improved grey wolf optimization algorithm for path planning to obtain the detection route; Step S4: Fly along the detection route through a multi-modal data acquisition device carried by the UAV to collect multi-modal data of each fault point; Step S5: Discriminate the collected multi-modal data through a fault diagnosis model to obtain the probability of the fault category; Step S6: Input the fault point location and the probability of the fault category into the three-dimensional model of the power line to realize the visual display of the probability of the fault category.

[0006] Furthermore, the specific process of obtaining images along the line is as follows: The power line is divided into M×N rectangular grids, and each rectangular grid represents an inspection area; M represents the number of rows; N represents the number of columns; Let the boundary coordinates of the M×N rectangular grids be ( , , ) to ( , , ); represents the minimum coordinate of the rectangular grid in the axis direction; represents the minimum coordinate of the rectangular grid in the axis direction; represents the minimum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; The parameters of the image acquisition device include the image acquisition width , the image acquisition height , and the UAV flight height ; The size of the rectangular grid is ([[]] , ), indicating; ; In the formula, represents the width of the rectangular grid; represents the height of the rectangular grid; is the overlap coefficient; The inspection path of the inspection area starts from the predefined starting point of the divided rectangular grid, and visits the center points of the rectangular grids in the order of the spiral path to form an inspection path set; and respectively represent the axis and axis coordinates of the starting point, indicating: ; In the formula, represents the inspection path set; represents the axis coordinate of the center point of the th rectangular grid in the inspection path set; represents the axis coordinate of the center point of the th rectangular grid in the inspection path set; The spiral path is a spiral curve centered on the starting point : ; In the formula, and respectively represent the axis coordinate and axis coordinate of the point on the spiral curve at the th step; and respectively represent the axis coordinate increment and axis coordinate increment; represents the direction control factor; and respectively represent the axis coordinate and axis coordinate of the point on the spiral curve at the th step; represents rounding down; Stop when the inspection path accesses more than all rectangular grids; Calculate the points on the spiral curve based on the improved spiral scanning algorithm to generate a flight path, and the UAV carries an image acquisition device to fly along the flight path, and the inspection area is photographed through the image acquisition device to obtain images along the line.

[0007] Furthermore, the lightweight image recognition model is constructed based on the YOLOv8 model; Insert the CBAM model after the C2f module in the backbone network of the YOLOv8 model; The CBAM model includes channel attention and spatial attention; The processing process of channel attention is as follows: Extract the global statistical information of each channel in the image along the line through global average pooling and max pooling, and send the extracted global statistical information of each channel into a multi-layer perceptron for fusion to obtain the channel attention weight of the image along the line, expressed as: ; In the formula, represents the image along the line, represents the channel attention weight of the image along the line; represents the activation function; represents the global average pooling operation on the image along the line; represents the max pooling operation on the image along the line; represents the multi-layer perceptron; The processing process of spatial attention is as follows: After obtaining the global statistical information of each extracted channel through concatenation operation, a 7×7 convolution operation is used to extract the concatenated result of the global statistical information, and the spatial attention weights of the along-line image are obtained, which is expressed as: ; In the formula, represents the spatial attention weights of the along-line image; represents the 7×7 convolution operation; After adjusting the input along-line image through channel attention and spatial attention, it is fused with the initial along-line image F, which is expressed as: ; In the formula, represents the updated overall convolutional feature map, that is, the fault point of the along-line image; represents element-wise multiplication.

[0008] Furthermore, combining the RTK-GPS positioning of the UAV and the parameters of the image acquisition device, the parameter matrix and the distortion coefficient D=(k1,k2,p1,p2) are used to correct the fault point of the along-line image ; and respectively represent the pixel axis coordinate and pixel axis coordinate of the fault point of the along-line image; the corrected fault point coordinates of the along-line image are obtained, which is expressed as: ; In the formula, and respectively represent the pixel axis coordinate and pixel axis coordinate of the corrected fault point of the along-line image; represents the first radial distortion coefficient; represents the second radial distortion coefficient; represents the first tangential distortion coefficient; represents the second tangential distortion coefficient; and respectively represent the pixel axis coordinate and pixel axis coordinate of the center point; where is the radius; represents the radial distance from the center of the fault point position of the along-line image to the pixel point; The corrected fault point coordinates of the along-line image are mapped to the geographical coordinates , that is, the fault point of the power line; The perspective transformation homography matrix H is expressed as: 。

[0009] Furthermore, the grey wolf optimization algorithm is improved. The specific process is as follows: Define a multi-objective optimization problem, which includes the total path length, risk cost, and energy consumption, and is expressed as: ; ; ; In the formula, represents minimizing the objective function; represents the total path length; represents the risk cost; represents the energy consumption; represents the transpose operation; represents the number of fault points; and respectively represent the axis coordinate and axis coordinate of the th fault point; and respectively represent the axis coordinate and axis coordinate of the th fault point; represents the function of taking the minimum value; represents the distance from the current th fault point to the nearest obstacle; represents the UAV safety distance threshold; Introduce a dynamic weight and inertia factor to improve the grey wolf optimization algorithm, and define a dynamic position update mechanism: ; ; In the formula, represents the distance vector between the grey wolf and the target point; represents the position of the grey wolf at the th iteration, representing the current solution; represents the position of the grey wolf at the th iteration; represents the dynamic weight; represents the optimal solution of the grey wolf position at the th iteration; represents the amplitude of the dynamic adjustment of the grey wolf position update at the th iteration; represents the dynamic adjustment for the The approximation strength of the optimal solution of the gray wolf position in the i-th iteration; represents the natural constant; represents the attenuation coefficient; represents the current time variable; represents the maximum number of iterations; and at the i-th iteration of the gray wolf position introduce the velocity term of the particle swarm optimization algorithm: ; In the formula, represents the velocity vector of the current solution point obtained in the i-th iteration; represents the velocity vector of the current solution point obtained in the j-th iteration; represents the global optimal point; represents the inertia factor; and respectively represent the individual learning factor and the social learning factor; represents the first random number; represents the second random number.

[0010] Furthermore, the multimodal data acquisition device includes an optical camera, an infrared thermal imager, and a LiDAR sensor; The unmanned aerial vehicle carries the multimodal data acquisition device and flies along the detection route. The fault points on the detection route are collected through the optical camera, the infrared thermal imager, and the LiDAR sensor respectively, and visible light images, infrared images, and LiDAR point clouds are obtained respectively; Align the coordinate systems of the optical camera and the LiDAR sensor, and use external parameter calibration to calculate the rotation matrix and the translation matrix, expressed as: ; In the formula, and respectively represent the coordinate information collected by the optical camera and the LiDAR sensor; represents the rotation matrix, estimated through calibration; represents the translation matrix, time-aligned; Synchronize the visible light image, the infrared image, and the LiDAR point cloud through timestamp calibration, and use linear interpolation to align the time resolutions of the visible light image, the infrared image, and the LiDAR point cloud to obtain the multimodal data of each fault point.

[0011] Furthermore, the fault diagnosis model includes an EfficientNetV2 model and a Point Transformer model; Extract the visible light image features and infrared image features using the EfficientNetV2 model: F RGB =EfficientNetV2(I RGB ); F IR =EfficientNetV2(I IR ); In the formula, F RGB represents the visible light image features; I RGB represents the input visible light image; F IR represents the infrared image features; I IR represents the input infrared image; The processing process of the Point Transformer model is as follows: Process the LiDAR point cloud using the self-attention mechanism combined with positional encoding to obtain the features of the LiDAR point cloud, expressed as: ; ; In the formula, represents the features of the LiDAR point cloud; represents the activation function; and respectively represent the query matrix and the key matrix; represents the dimension of the key matrix; represents the positional encoding; represents the coordinates of the features of the LiDAR point cloud; represents the coordinates of the features of the LiDAR point cloud; represents the value matrix; represents the LiDAR point cloud features after the th layer of processing; and respectively represent the query matrix and the key matrix of the th layer; represents the value matrix of the th layer; is the relative positional encoding of the distance difference between the coordinates of the features of the LiDAR point cloud and the coordinates of the features of the LiDAR point cloud; Adopt a dynamic gating network to adaptively adjust the weights RGB of the visible light image features F IR , the infrared image features F and the LiDAR point cloud features , and according to the weights Obtain the unified feature after fusion , which means: ; ; In the formula, and are the fully connected weight and bias respectively; represents the concatenation operation; represents the th individually extracted feature after being processed by the dynamic gating network; represents the weight obtained after being processed by the dynamic gating network; The unified feature after fusion After passing through the fully connected layer, the Softmax classifier is used to output the probability of the fault category, which means: ; In the formula, represents the classifier outputting the probability of the th fault category probability; and represent the weight and bias of the classifier respectively.

[0012] Furthermore, an AI-based power line inspection efficiency optimization system for imitation line inspection, which is applied to the above-mentioned AI-based power line inspection efficiency optimization method for imitation line inspection, includes: An image acquisition module, which is used to perform line inspection on the power line through the image acquisition device carried by the unmanned aerial vehicle to obtain the images along the line; An identification module, which is used to identify the images along the line through a lightweight image recognition model to obtain the fault points of the power line; A detection planning module, which is used to perform path planning by using an improved grey wolf optimization algorithm according to the obtained fault points to obtain the detection route; A detection planning module, which is used to fly along the detection route through the multi-modal data acquisition device carried by the unmanned aerial vehicle to collect the multi-modal data of each fault point; A fault point data acquisition module, which is used to discriminate the collected multi-modal data through a fault diagnosis model to obtain the probability of the fault category; A visualization display module, which is used to input the fault point position and the probability of the fault category into the 3D model of the power line to realize the visualization display of the probability of the fault category.

[0013] Furthermore, a computer storage medium includes: The computer storage medium stores multiple instructions, and the instructions include a processor, and the processor is applied to the above-mentioned AI-based power line inspection efficiency optimization method for imitation line inspection.

[0014] Compared with the existing technologies, the present invention has the following beneficial effects: (1) By combining a lightweight image recognition model, multi-modal fusion analysis, and 3D reconstruction technology, the present invention supports efficient and accurate inspection by drones. By improving the Grey Wolf Optimization algorithm, it plans the optimal fine inspection route. Using the NeRF model, it combines the multi-modal data of each fault point with the 3D model of the power line, providing intuitive and dynamic fault visualization for maintenance personnel, not only effectively improving the inspection efficiency and fault diagnosis accuracy, but also significantly reducing the maintenance cost and risk.

[0015] (2) Through the improved rectangular grid division and spiral scanning path design, the present invention effectively improves the inspection efficiency and ensures the coverage integrity and quality of image acquisition along the line, while supporting the inspection tasks of power lines in complex terrains.

[0016] (3) Through the features of visible light images, infrared images, and LiDAR point clouds, combined with a dynamic gating network, the present invention performs efficient and robust fault diagnosis. By fusing the features of visible light images, infrared images, and LiDAR point clouds, the diagnosis accuracy is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0018] As Figure 1 shown, the present invention provides a technical solution: an AI-based method for optimizing the inspection efficiency of power line imitation lines, including the following steps: Step S1: Use the image acquisition device carried by the drone to inspect the power line and obtain the images along the line; Step S2: Use the lightweight image recognition model to identify the images along the line and obtain the fault points of the power line; Step S3: According to the obtained fault points, use the improved Grey Wolf Optimization algorithm for path planning to obtain the detection route; Step S4: Fly the drone equipped with a multi-modal data acquisition device along the detection route to collect the multi-modal data of each fault point; Step S5: Use the fault diagnosis model to discriminate the collected multi-modal data and obtain the fault category probability; Step S6: Input the fault point position and the fault category probability into the 3D model of the power line to realize the visual display of the fault category probability.

[0019] Among them, the specific process of obtaining the images along the line is: The power line is divided into M×N rectangular grids, and each rectangular grid represents an inspection area to ensure that no inspection area is missed during the inspection task of an unmanned aerial vehicle on a line; M represents the number of rows; N represents the number of columns; Let the boundary coordinates of the M×N rectangular grids be ( , , ) to ( , , ); represents the minimum coordinate of the rectangular grid in the axis direction; represents the minimum coordinate of the rectangular grid in the axis direction; represents the minimum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; represents the maximum coordinate of the rectangular grid in the axis direction; The parameters of the image acquisition device include the image acquisition width , the image acquisition height , and the flight height of the unmanned aerial vehicle ; The size of the rectangular grid is ( , ), which means; ; In the formula, represents the width of the rectangular grid; represents the height of the rectangular grid; is the overlap coefficient; The inspection path of the inspection area starts from the predefined starting point of the divided rectangular grid, and visits the center points of the rectangular grids in the order of the spiral path to form an inspection path set; and respectively represent the axis and axis coordinates of the starting point, indicating: ; In the formula, represents the inspection path set; represents the -th axis coordinate of the center point of the rectangular grid in the inspection path set; represents the The axial coordinates of the center points of a rectangular grid; The spiral path is a spiral curve centered on the starting point as follows: ; Wherein, and respectively represent the axial coordinates of the points on the spiral curve at the th step and the axial coordinates; and respectively represent the axial coordinate increment and the axial coordinate increment; represents the direction control factor; and respectively represent the axial coordinates of the points on the spiral curve at the th step and the axial coordinates; represents rounding down; Stop when the inspection path accesses all rectangular grids; calculate the points on the spiral curve based on the improved spiral scanning algorithm to generate a flight path, and the UAV carries an image acquisition device to fly along the flight path, and the inspection area is photographed through the image acquisition device to obtain images along the line.

[0020] Among them, the lightweight image recognition model is constructed based on the YOLOv8 model; the CBAM model is inserted after the C2f module of the backbone network of the YOLOv8 model, and the CBAM model includes channel attention and spatial attention; enhance the detection ability for small targets (such as broken strands of wires); The processing process of channel attention is as follows: Extract the global statistical information of each channel in the image along the line through global average pooling and max pooling, and send the extracted global statistical information of each channel into a multi-layer perceptron for fusion to obtain the channel attention weight of the image along the line, expressed as: ; Wherein, represents the image along the line, represents the channel attention weight of the image along the line; represents the activation function; represents the global average pooling operation on the image along the line; represents the max pooling operation on the image along the line; represents the multi-layer perceptron; Finally, the channel attention weight of the image along the line will weightedly adjust each channel of the image along the line to strengthen the key features; The processing process of spatial attention is as follows: After obtaining the global statistical information splicing result by splicing the global statistical information of each extracted channel, a 7×7 convolution operation is used to extract the global statistical information splicing result, and the spatial attention weight of the along-line image is obtained, which is expressed as: ; In the formula, represents the spatial attention weight of the along-line image; represents the 7×7 convolution operation; After adjusting the input along-line image through channel attention and spatial attention, it is further fused with the initial along-line image F, which is expressed as: ; In the formula, represents the updated overall convolutional feature map, that is, the fault point of the along-line image; represents element-wise multiplication; And the Manhattan distance L1-norm is used to perform channel pruning on the YOLOv8 model to remove redundant convolutional kernels; the number of parameters of the YOLOv8 model is reduced from 25.9M of the original YOLOv8 model to 8.7M. Finally, the pruned YOLOv8 model is converted from the FP32 model to the INT8 format, and the inference speed is increased by 2.3 times and the power consumption is reduced by 40%.

[0021] Among them, combining the RTK-GPS positioning of the drone and the parameters of the image acquisition device, the parameter matrix and the distortion coefficient D=(k1,k2,p1,p2) are used to correct the fault point of the along-line image ; and respectively represent the pixels of the fault point of the along-line image axis coordinate and pixel axis coordinate; the corrected fault point coordinates of the along-line image are obtained , which is expressed as: ; In the formula, and respectively represent the pixels of the corrected fault point of the along-line image axis coordinate and pixel axis coordinate; represents the first radial distortion coefficient; represents the second radial distortion coefficient; represents the first tangential distortion coefficient; represents the second tangential distortion coefficient; and respectively represent the pixels of the center point Axis coordinates and pixels Axis coordinates; where is the radius; represents the radial distance from the center of the fault point position on the along-line image to the pixel point; The corrected fault point coordinates on the along-line image are mapped to geographical coordinates through the perspective transformation homography matrix H , that is, the fault point of the power line; The perspective transformation homography matrix H, which means: ; ; Among them, H is a 3×3 perspective transformation homography matrix, which is solved by the RANSAC algorithm.

[0022] Among them, the improved grey wolf optimization algorithm, the specific process is: Define the multi-objective optimization problem, which includes the total path length, risk cost and energy consumption: ; ; ; In the formula, represents minimizing the objective function; represents the total path length; represents the risk cost; represents the energy consumption; represents the transpose operation; represents the number of fault points; and respectively represent the axis coordinate and axis coordinate of the th fault point; and respectively represent the axis coordinate and axis coordinate of the th fault point; represents the minimum value function; represents the distance from the current th fault point to the nearest obstacle; represents the UAV safety distance threshold; Introduce dynamic weights and inertia factors to improve the grey wolf optimization algorithm, enhance the multi-objective optimization problem ability, and define the dynamic position update mechanism: ; ; In the formula, represents the distance vector between the grey wolf and the target point; Denotes the position of the grey wolf at the -th iteration, representing the current solution; Denotes the position of the grey wolf at the -th iteration; Denotes the dynamic weight; Denotes the optimal solution of the grey wolf position at the -th iteration; Denotes the amplitude of the update of the grey wolf position at the -th iteration with dynamic adjustment; Denotes the approximation strength of the dynamic adjustment to the optimal solution of the grey wolf position at the -th iteration; Denotes the natural constant; Denotes the attenuation coefficient; Denotes the current time variable; Denotes the maximum number of iterations; And introduce the velocity term of the particle swarm optimization algorithm at the grey wolf position at the -th iteration: ; ; In the formula, Denotes the velocity vector of the current solution point obtained at the -th iteration; Denotes the velocity vector of the current solution point obtained at the -th iteration; Denotes the global optimal point; Denotes the inertia factor; And respectively denote the individual learning factor and the social learning factor; Denotes the first random number; Denotes the second random number.

[0023] Among them, the multi-modal data acquisition device includes an optical camera, an infrared thermal imager, and a LiDAR sensor; The unmanned aerial vehicle carries the multi-modal data acquisition device and flies along the detection route. The fault points on the detection route are collected through the optical camera, the infrared thermal imager, and the LiDAR sensor respectively, and visible light images, infrared images, and LiDAR point clouds are obtained respectively; Align the coordinate systems of the optical camera and the LiDAR sensor, and use external parameter calibration to calculate the rotation matrix and the translation matrix, denoted as: ; In the formula, and respectively denote the coordinate information collected by the optical camera and the LiDAR sensor; Denotes the rotation matrix, estimated through calibration; Represents a translation matrix for time alignment; Synchronize visible light images, infrared images, and LiDAR point clouds through timestamp calibration, and use linear interpolation to align the time resolutions of visible light images, infrared images, and LiDAR point clouds to obtain multi-modal data for each fault point; Use the low-latency protocol MQTT to transfer the collected multi-modal data between the drone and the control center.

[0024] Among them, the fault diagnosis model includes an EfficientNetV2 model and a Point Transformer model; Use the EfficientNetV2 model to extract visible light image features and infrared image features: F RGB = EfficientNetV2(I RGB ); F IR = EfficientNetV2(I IR ); In the formula, F RGB represents visible light image features; I RGB represents the input visible light image; F IR represents infrared image features; I IR represents the input infrared image; The processing process of the Point Transformer model is as follows: Use the self-attention mechanism combined with position encoding to process the LiDAR point cloud to obtain the features of the LiDAR point cloud, expressed as: ; ; In the formula, represents the features of the LiDAR point cloud; represents the activation function; and represent the query matrix and the key matrix respectively; represents the dimension of the key matrix; represents the position encoding; represents the features of the LiDAR point cloud coordinates; represents the features of the LiDAR point cloud coordinates; represents the value matrix; represents the LiDAR point cloud features after processing in the layer; and represent the query matrix and the key matrix of the layer respectively; Represents the value matrix of the layer; Is the feature of the LiDAR point cloud Of the coordinates and the feature of the LiDAR point cloud Relative position encoding of the distance difference between the coordinates; Adopts a dynamic gating network to adaptively adjust the visible light image feature F RGB , infrared image feature F IR And the feature of the LiDAR point cloud Weights , and according to the weights Obtain the fused unified feature , expressed as: ; ; In the formula, And Are the fully connected weights and biases respectively; Represents the concatenation operation; Represents the th individually extracted feature after being processed by the dynamic gating network; Represents the weights obtained after being processed by the dynamic gating network; Fused unified feature After passing through the fully connected layer, use the Softmax classifier to output the probability of the fault category, expressed as: ; In the formula, Represents the output of the classifier th fault category probability; And Represent the weights and biases of the classifier respectively; Is the fault category (such as "insulator damage", "wire break", etc.).

[0025] Among them, to realize the visualization display of the fault category probability, the specific process is: Construct a neural network with a radiation field through the NeRF model. Use the visible light image combined with the image acquisition device parameters through the neural network with a radiation field to generate a 3D model of the power line in a specific fault area; introduce the LiDAR point cloud to supplement the geometric structure information and constrain the geometric field of the reconstructed 3D model of the power line; Combine the infrared image with the 3D model of the power line; through texture mapping or color coding, superimpose the infrared temperature information of the infrared image on the 3D model of the power line to realize heat map visualization and generate a high-fidelity, multi-modal superimposed 3D model of the power line; Add the fault point location and fault category probability to the 3D model of the power line to achieve visual display of the fault category probability; Export the 3D model of the power line generated by the NeRF model to the.glb or.obj format, including the fused texture of visible light images and thermal imaging, and use the Three.js model to achieve front-end interaction.

[0026] Among them, an AI-based power line imitation line inspection efficiency optimization system is applied to the above-mentioned AI-based power line imitation line inspection efficiency optimization method, including: An image acquisition module, which is used to perform line inspection on the power line through an image acquisition device carried by a drone to obtain images along the line; An identification module, which is used to identify the images along the line through a lightweight image recognition model to obtain the fault points of the power line; A detection planning module, which is used to perform path planning by using an improved grey wolf optimization algorithm according to the obtained fault points to obtain a detection route; A detection planning module, which is used to fly along the detection route through a multi-modal data acquisition device carried by a drone to collect multi-modal data of each fault point; A fault point data acquisition module, which is used to discriminate the collected multi-modal data through a fault diagnosis model to obtain the fault category probability; A visualization display module, which is used to input the fault point location and fault category probability into the 3D model of the power line to achieve visual display of the fault category probability.

[0027] A computer storage medium stores multiple instructions, and the instructions are suitable for being loaded and executed by a processor to perform the above-mentioned AI-based power line imitation line inspection efficiency optimization method.

[0028] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based method for optimizing the inspection efficiency of power line imitation inspection, characterized in that, Including the following steps: Step S1: Use an image acquisition device carried by a drone to patrol the power line and obtain images along the line; Step S2: Use a lightweight image recognition model to identify the images along the line and obtain the fault points of the power line; Step S3: According to the obtained fault points, use an improved grey wolf optimization algorithm for path planning to obtain the detection route; Step S4: Fly along the detection route by a drone carrying a multi-modal data acquisition device to collect multi-modal data of each fault point; Step S5: Use a fault diagnosis model to discriminate the collected multi-modal data and obtain the fault category probability; Step S6: Input the fault point location and the fault category probability into the 3D model of the power line to realize the visual display of the fault category probability.

2. The method for optimizing the inspection efficiency of an AI-based power line imitation line inspection according to claim 1, characterized in that: The specific process of obtaining the images along the line is as follows: Divide the power line into M×N rectangular grids, and each rectangular grid represents a patrol area; M represents the number of rows; N represents the number of columns; Let the boundary coordinates of an \(M\times N\) rectangular grid be from ( , , ) to ( , , ); denote the minimum coordinate of the rectangular grid in the axis direction; denote the minimum coordinate of the rectangular grid in the axis direction; denote the minimum coordinate of the rectangular grid in the axis direction; denote the maximum coordinate of the rectangular grid in the axis direction; denote the maximum coordinate of the rectangular grid in the axis direction; denote the maximum coordinate of the rectangular grid in the axis direction; The parameters of the image acquisition device include the image acquisition width , the image acquisition height , the flight height of the drone ; The rectangular grid size is ( , ), indicating; ； In the formula, represents the width of the rectangular grid; represents the height of the rectangular grid; is the overlap coefficient; The inspection path of the inspection area starts from the predefined starting point of the divided rectangular grid and visits the center points of the rectangular grid in the order of the spiral path to form a set of inspection paths; and respectively represent the axis and axis coordinates of the starting point, indicating: ； In the formula, represents the inspection path set; represents the -axis coordinate of the center point of the th rectangular grid in the inspection path set; represents the -axis coordinate of the center point of the th rectangular grid in the inspection path set; The spiral path starts from the starting point and is a spiral curve centered thereon: ； Wherein, and respectively represent the x-axis coordinate and y-axis coordinate of the point on the spiral curve at the n-th step; and respectively represent the increment of the x-axis coordinate and y-axis coordinate; and respectively represent the x-axis coordinate and y-axis coordinate of the point on the spiral curve at the n-th step; represents rounding down; Stop when the patrol path visits all the rectangular grids; calculate the points on the spiral curve based on the improved spiral scanning algorithm to generate the flight path, and the drone carrying the image acquisition device flies along the flight path, and the image acquisition device takes pictures of the patrol area to obtain the images along the line.

3. An AI-based method for optimizing the inspection efficiency of power line tracing, as claimed in claim 2, wherein: The lightweight image recognition model is constructed based on the YOLOv8 model; the CBAM model is inserted after the C2f module of the backbone network of the YOLOv8 model; the CBAM model includes channel attention and spatial attention; The processing process of channel attention is as follows: Extract the global statistical information of each channel in the image along the line through global average pooling and max pooling, and send the extracted global statistical information of each channel into a multi-layer perceptron for fusion to obtain the channel attention weight of the image along the line, expressed as: ； In the formula, represents the image along the line, represents the channel attention weight of the image along the line; represents the activation function; represents the global average pooling operation on the image along the line; represents the max pooling operation on the image along the line; represents the multi-layer perceptron; The processing process of spatial attention is as follows: After obtaining the global statistical information splicing result through the splicing operation of the extracted global statistical information of each channel, then use a 7×7 convolution operation to extract the global statistical information splicing result to obtain the spatial attention weight of the image along the line, expressed as: ； wherein, represents the spatial attention weight of the image along the line; represents a 7×7 convolution operation; After adjusting the input image along the line through channel attention and spatial attention, it is fused with the initial image F along the line, expressed as: ； In the formula, represents updating the overall convolutional feature map, that is, the fault point of the line image; represents element-wise multiplication.

4. An AI-based method for optimizing the inspection efficiency of power line imitation lines according to claim 3, characterized in that: Combined with the RTK-GPS positioning of the drone and the parameters of the image acquisition device, use the parameters of the image acquisition device to construct a parameter matrix and the distortion coefficients D=(k1,k2,p1,p2) to correct the fault points of the images along the line ; and respectively represent the pixel axis coordinate and pixel axis coordinate of the fault point of the image along the line; obtain the corrected coordinate of the fault point of the image along the line , which means: ； In the formula, and respectively represent the pixel axis coordinate and pixel axis coordinate of the fault point in the image along the line; represents the first radial distortion coefficient; represents the second radial distortion coefficient; represents the first tangential distortion coefficient; represents the second tangential distortion coefficient; and respectively represent the pixel axis coordinate and pixel axis coordinate of the center point; where is the radius; represents the radial distance from the center of the position of the fault point in the image along the line to the pixel point; The corrected fault point coordinates of the image along the line are mapped to geographical coordinates through the perspective transformation homography matrix H , that is, the fault point of the power line; ​ The perspective transformation homography matrix H, expressed as: 。 5. The method for optimizing the inspection efficiency of an AI-based power line imitation line inspection according to claim 4, characterized in that: The improved grey wolf optimization algorithm, the specific process is as follows: Define a multi-objective optimization problem, and the multi-objective optimization problem includes the total path length, risk cost and energy consumption, expressed as: ； ； ； In the formula, represents minimizing the objective function; represents the total path length; represents the risk cost; represents the energy consumption; represents the transpose operation; represents the number of fault points; and respectively represent the -axis coordinate and -axis coordinate of the th fault point; and respectively represent the -axis coordinate and -axis coordinate of the th fault point; represents the function of taking the minimum value; represents the distance from the th current fault point to the nearest obstacle; represents the UAV safety distance threshold; Introduce dynamic weights and inertia factors to improve the grey wolf optimization algorithm, and define a dynamic position update mechanism: ； ； Wherein, represents the distance vector between the grey wolf and the target point; represents the position of the grey wolf at the -th iteration, representing the current solution; represents the position of the grey wolf at the -th iteration; represents the dynamic weight; represents the optimal solution of the grey wolf position at the -th iteration; represents the amplitude of the update of the grey wolf position adjusted dynamically at the -th iteration; represents the approximation strength of the optimal solution of the grey wolf position adjusted dynamically for the -th iteration; represents the natural constant; represents the attenuation coefficient; represents the current time variable; represents the maximum number of iterations; And introduce the velocity term of the particle swarm algorithm into the gray wolf position at the th iteration: ​ ； In the formula, represents the velocity vector of the current solution point obtained in the -th iteration; represents the velocity vector of the current solution point obtained in the -th iteration; represents the global optimal point; represents the inertia factor; and represent the individual learning factor and the social learning factor respectively; represents the first random number; represents the second random number.

6. The method for optimizing the inspection efficiency of an AI-based power line imitation line inspection according to claim 5, wherein: The multi-modal data acquisition device includes an optical camera, an infrared thermal imager and a LiDAR sensor; The drone carrying the multi-modal data acquisition device flies along the detection route, and the optical camera, infrared thermal imager and LiDAR sensor are used to collect the fault points on the detection route respectively, and visible light images, infrared images and LiDAR point clouds are obtained respectively; Align the coordinate systems of the optical camera and the LiDAR sensor, and use external parameter calibration to calculate the rotation matrix and translation matrix, expressed as: ； In the formula, and respectively represent the coordinate information collected by the optical camera and the LiDAR sensor; represents the rotation matrix, estimated through calibration; represents the translation matrix, time-aligned; Synchronize visible light images, infrared images, and LiDAR point clouds through timestamp calibration, and use linear interpolation to align the time resolutions of visible light images, infrared images, and LiDAR point clouds to obtain multi-modal data of each fault point.

7. An AI-based method for optimizing the inspection efficiency of power line imitation inspection according to claim 6, characterized in that: The fault diagnosis model includes an EfficientNetV2 model and a Point Transformer model; Use the EfficientNetV2 model to extract visible light image features and infrared image features: F RGB =EfficientNetV2(I RGB ); F IR =EfficientNetV2(I IR ); In the formula, F RGB represents the visible light image feature; I RGB represents the input visible light image; F IR represents the infrared image feature; I IR represents the input infrared image; The processing process of the Point Transformer model is as follows: Use the self-attention mechanism combined with position encoding to process the LiDAR point cloud to obtain the features of the LiDAR point cloud, which are expressed as: ； ； In the formula, represents the feature of the LiDAR point cloud; represents the activation function; and represent the query matrix and the key matrix respectively; represents the dimension of the key matrix; represents the position encoding; represents the feature of the LiDAR point cloud coordinates; represents the feature of the LiDAR point cloud coordinates; represents the value matrix; represents the LiDAR point cloud feature after processing in the th layer; and represent the query matrix and the key matrix in the th layer respectively; represents the value matrix in the th layer; is the relative position encoding of the distance difference between the coordinates of the feature of the LiDAR point cloud and the coordinates of the feature of the LiDAR point cloud; Adopting dynamic gating network to adaptively adjust visible light image feature F RGB , infrared image features F IR and the characteristics of LiDAR point cloud Weight , and according to the weight Get the unified features after fusion ,express: ； ； In the formula, and are the fully-connected weights and biases respectively; represents the concatenation operation; represents the th individually extracted feature after the processing of the dynamic gating network; represents the weight obtained after the processing of the dynamic gating network. Fused unified features After passing through the fully connected layer, the Softmax classifier is used to output the probability of the fault category, which is expressed as: ； In the formula, represents the probability of the classifier outputting the th fault category ; and respectively represent the weight and bias of the classifier.

8. An AI-based power line imitation line inspection efficiency optimization system, which is applied to an AI-based power line imitation line inspection efficiency optimization method as described in any one of claims 1-7, and is characterized in that, Include: An image acquisition module for patrolling the power line through an image acquisition device carried by a drone to obtain images along the line; An identification module for identifying the images along the line through a lightweight image recognition model to obtain the fault points of the power line; A detection planning module for performing path planning using an improved grey wolf optimization algorithm according to the obtained fault points to obtain a detection route; A detection planning module for flying along the detection route through a multi-modal data acquisition device carried by a drone to collect multi-modal data of each fault point; A fault point data acquisition module for discriminating the collected multi-modal data through a fault diagnosis model to obtain the probability of the fault category; A visualization display module for inputting the fault point position and the probability of the fault category into the 3D model of the power line to realize the visualization display of the probability of the fault category.

9. A computer storage medium, characterized in that, Include: A computer storage medium stores multiple instructions, and the instructions include a processor, and the processor is applied to an AI-based power line imitation line patrol efficiency optimization method according to any one of claims 1-7.

Images