Substation unmanned aerial vehicle autonomous inspection method and system

Abstract

The invention provides an unmanned aerial vehicle autonomous inspection method and system for a transformer substation. The core content of the unmanned aerial vehicle autonomous inspection method comprises seven steps of task pre-planning, flight path generation, BIM model generation, edge point extraction, inspection path generation, inspection point planning and inspection task execution. The steps and links are combined with various advanced technologies such as LiDAR scanning, SLAM technology, reinforcement learning, genetic algorithm, ant colony algorithm and edge detection, efficient, intelligent and safe inspection of the transformer substation is realized, and powerful support is provided for operation and maintenance management of the transformer substation. Wherein the obstacle avoidance path is autonomously planned through the reinforcement learning dynamic training obstacle avoidance model, dynamic obstacle avoidance and path optimization are realized, and the flight risk is reduced. And autonomous flight and task execution in a complex environment are supported. And real-time mapping and positioning of the unmanned aerial vehicle are realized by using the SLAM technology, and the preprocessing dependence is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent operation and maintenance of power facilities, and specifically to a substation drone inspection method integrating artificial intelligence, multi-sensor fusion and autonomous control. Background Art

[0002] With the development of drone technology, the use of drones to replace manual inspections of transmission lines has become increasingly common. Currently, substation inspections face the following key challenges: High dependence on manual labor: Traditional inspections require climbing equipment, resulting in low efficiency and high risk; Insufficient intelligence in individual drones: Existing drones rely on preset routes and are unable to dynamically avoid unexpected obstacles (such as birds and temporary equipment); High rates of missed defects: Single visible light inspections struggle to detect hidden defects such as insulator discharge and internal heating; and Data siloing: Inspection results are not integrated with the full lifecycle management of the equipment.

[0003] Therefore, a substation UAV autonomous inspection method is designed to solve the problems of poor autonomy of UAVs and insufficient utilization of inspection data in the complex environment of substations. Summary of the Invention

[0004] In view of this, it is necessary to provide a method for autonomous substation drone inspection. This method uses reinforcement learning to dynamically train an obstacle avoidance model to autonomously plan an obstacle avoidance path, achieving dynamic obstacle avoidance and path optimization to reduce flight risks. It supports autonomous flight and mission execution in complex environments. It uses SLAM technology to achieve real-time drone mapping and positioning, reducing reliance on preprocessing. This method supports autonomous flight and mission execution in complex environments. It uses SLAM technology to achieve real-time drone mapping and positioning, reducing reliance on preprocessing. It supports autonomous flight and mission execution in complex environments. It uses SLAM technology to achieve real-time drone mapping and positioning, reducing reliance on preprocessing. It is easy to promote and apply, and its operation is relatively simple, making it suitable for promotion and application in a wider range of scenarios.

[0005] The present application provides a method and system for autonomous inspection of substations by drones, the core contents of which include seven steps: task pre-planning, flight path generation, BIM model generation, edge point extraction, inspection path generation, inspection point planning, and inspection task execution. These steps combine a variety of advanced technologies, such as LiDAR scanning, SLAM technology, reinforcement learning, genetic algorithm, ant colony algorithm, edge detection, etc., and combine a variety of advanced technologies to achieve efficient, intelligent and safe substation inspections, providing strong support for substation operation and maintenance management. Among them, the obstacle avoidance model is dynamically trained through reinforcement learning to autonomously plan the obstacle avoidance path, realize dynamic obstacle avoidance and path optimization, and reduce flight risks. Support autonomous flight and task execution in complex environments. Use SLAM technology to achieve real-time mapping and positioning of drones, reducing preprocessing dependence.

[0006] In a first aspect, an embodiment of the present application provides a method for autonomous inspection of a substation by a drone, the method comprising:

[0007] S1: Mission Pre-Planning: Generate a high-precision 3D map of the inspection area through preliminary LiDAR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land according to the preset preliminary routes and mission markings;

[0008] S2: Flight Path Generation: After inputting the inspection target, the UAV inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data through the dynamic training of the obstacle avoidance model through reinforcement learning, and dynamically adjust the route according to the flight distance, power limit and obstacle conditions;

[0009] S3: BIM model generation: After reaching the key target point, a BIM model of the inspection target is constructed using 3D laser scanning technology. SLAM technology is used to achieve real-time drone mapping and positioning, reducing pre-processing dependency. A BIM incremental update mechanism is also established to scan only the changed areas.

[0010] S4: Edge point extraction: extracting the edge contour of the BIM model by combining Canny edge detection with Hough transform;

[0011] S5: Inspection path generation: Genetic algorithm combined with ant colony algorithm and the Pareto optimality concept are used to optimize the inspection path design to improve inspection efficiency and coverage;

[0012] S6: Inspection point planning: Plan inspection points evenly based on the inspection route to achieve comprehensive coverage of inspection targets;

[0013] S7: Inspection mission execution: The drone performs inspection tasks according to the planned inspection routes and inspection points, adjusts the shooting angle, focal length, and exposure parameters, collects image data in real time and transmits it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

[0014] Optionally, in an implementation of the first aspect of the present invention, S1: mission pre-planning: generating a high-precision three-dimensional map of the inspection area through preliminary LiDAR scanning, marking key target points according to the mission, presetting a preliminary route, and automatically taking off, operating, and landing according to the preset preliminary route and mission markings, including:

[0015] LiDAR collects centimeter-level precision 3D terrain and target feature information of the transmission line corridor and surrounding environment, generating a 3D map of the inspection area.

[0016] Generate preliminary routes using 3D map data based on mission requirements and terrain characteristics;

[0017] Using AI algorithms, key features in 3D point clouds are quickly extracted, key inspection target points are marked on 3D maps, and then linked to records in the power grid asset database.

[0018] According to the preset routes and mission markers, the drone can take off autonomously, perform inspection tasks along the planned route and land automatically.

[0019] Optionally, in an implementation of the first aspect of the present invention, S2: Flight Path Generation: Inputting an inspection target, the UAV inspection system can autonomously plan an obstacle avoidance path based on preset tasks and real-time environmental data by dynamically training an obstacle avoidance model through reinforcement learning, including:

[0020] The agent drone continuously interacts with the environment and obtains the optimal value function q of the state s through the feedback or rewards given by the environment, so as to continuously optimize the state-action to obtain the optimal strategy f. The problem of finding the best strategy is transformed into finding the maximum value of the action-state value function generated under all strategies.

[0021] The training process is a path planning algorithm based on reinforcement learning, which allows the drone to learn through trial and error and obtain rewards through continuous interaction with the environment. The optimal value function q(s,a) and the optimal strategy f(s|a) are expressed as follows:

[0022]

[0023] Where a represents the agent point state.

[0024] Optionally, in an implementation of the first aspect of the present invention, the reinforcement learning-based path planning algorithm allows the drone to learn by trial and error through continuous interaction with the environment and obtain rewards, including:

[0025] With the drone as the center, three spherical areas with preset obstacle avoidance radii of R1, R2, and R3 are preset, where R1>R2>R3;

[0026] When obstacles are within different preset obstacle avoidance radius of the drone, different collision avoidance strategies are implemented, specifically:

[0027] Combining traction and repulsion, the gravitational force generated by the target point and the repulsive force generated by the obstacle on the drone are converted into rewards or penalties obtained by the drone after performing actions in the state. The optimized reward function is:

[0028]

[0029] Among them, d max Indicates the maximum distance between the starting point of the agent and the target. represents the distance between the agent point and the target at time t, represents the distance between the agent point and the obstacle at time t, and D represents the distance between the drone and the obstacle.

[0030] Optionally, in an implementation of the first aspect of the present invention, S3: BIM model generation: After reaching the key target point, a BIM model of the inspection target is constructed using three-dimensional laser scanning technology, and SLAM technology is used to achieve real-time drone mapping and positioning, reducing preprocessing dependence, and establishing a BIM incremental update mechanism to scan only the changed areas, including:

[0031] Use LiDAR to scan inspection targets and generate high-density point cloud data;

[0032] Collect the position and attitude data of the drone through IMU;

[0033] Collect image data through a camera;

[0034] Utilizing SLAM algorithms to optimize and process the point cloud data, position and posture data, and image data to establish an accurate BIM model;

[0035] The optimization process includes:

[0036] Adjust the position of each point in the point cloud data and minimize the global error to obtain a more stable and accurate three-dimensional model. The error formula e i for:

[0037] e i =z i -h(p i ),

[0038] Among them, z i represents the observed data, p i Indicates the position of each point, h(p i ) represents the observed data predicted by the model;

[0039] The optimization function to minimize the global error is:

[0040]

[0041] Where n represents the number of point cloud data;

[0042] Adjust the position of each point to minimize the error;

[0043] Based on the historical BIM model and current scan data, the position changes of deformation feature points are extracted through feature point matching and comparative analysis, thereby realizing incremental updates of the BIM model.

[0044] Optionally, in an implementation of the first aspect of the present invention, the step S4: edge point extraction: extracting the edge contour of the BIM model by using Canny edge detection combined with Hough transform identification, includes:

[0045] Obtain the point cloud dataset of the BIM model;

[0046] By setting two high and low thresholds, the edge points are divided into strong edges and weak edges. Strong edges are directly retained, while weak edges are retained based on their connection relationship with strong edges, thus obtaining an edge point set.

[0047] Map the edge points detected by Canny to Hough space, where each edge point corresponds to a straight line;

[0048] Count the likelihood of the line in the accumulator matrix;

[0049] By setting a threshold, significant line candidates are screened out and the line features in the image are finally extracted;

[0050] Connect isolated edge points into continuous edge lines.

[0051] Optionally, in an implementation of the first aspect of the present invention, S6: Inspection point planning: planning inspection points based on the average of the inspection paths to achieve full coverage of the inspection targets, including:

[0052] Calculating the total length of the inspection path according to the edge line;

[0053] Calculate the number of inspection points based on the horizontal or vertical range of the camera and the total length of the inspection path;

[0054] The inspection points are divided according to the number of the inspection points.

[0055] In a second aspect, an embodiment of the present application provides a substation drone autonomous inspection system, which is applied to the substation drone autonomous inspection method according to any one of claims 1 to 7, and is characterized by comprising:

[0056] The mission pre-planning module is used to generate a high-precision 3D map of the inspection area through preliminary LiDAR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land according to the preset preliminary routes and mission markings;

[0057] The flight path generation module is used to input inspection targets. Based on preset tasks and real-time environmental data, the drone inspection system can autonomously plan obstacle avoidance paths through dynamic training of obstacle avoidance models through reinforcement learning, and dynamically adjust the route according to flight distance, power limit, and obstacle conditions.

[0058] The BIM model generation module is used to construct a BIM model of the inspection target through 3D laser scanning technology after reaching the key target point. It uses SLAM technology to achieve real-time mapping and positioning of the drone, reducing the dependence on preprocessing, and establishes a BIM incremental update mechanism to scan only the changed areas;

[0059] An edge point extraction module is used to extract the edge contour of the BIM model by combining Canny edge detection with Hough transform;

[0060] The inspection path generation module is used to optimize the inspection path design by combining genetic algorithm with ant colony algorithm and introducing the Pareto optimal concept to improve inspection efficiency and coverage;

[0061] Inspection point planning module, used to plan inspection points according to the inspection route to achieve comprehensive coverage of inspection targets;

[0062] The inspection task execution module is used for drones to perform inspection tasks according to the planned inspection routes and inspection points, adjust the shooting angle, focal length, and exposure parameters, collect image data in real time and transmit it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

[0063] In a third aspect, an embodiment of the present application provides an electronic device, characterized by including:

[0064] processor;

[0065] a memory for storing processor-executable instructions;

[0066] Among them, the processor is configured to implement the substation drone autonomous inspection method as described in the first aspect when executing the instructions.

[0067] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a program, and the program instructs a device to execute the substation drone autonomous inspection method as described in the first aspect.

[0068] The present application provides a method and system for autonomous inspection of substations by drones, including task pre-planning: generating a high-precision three-dimensional map of the inspection area through preliminary LiDAR scanning, marking key target points according to the task, presetting a preliminary route, and automatically taking off, operating, and landing according to the preset preliminary route and task marking; flight path generation: inputting the inspection target, the drone inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data through reinforcement learning dynamic training obstacle avoidance model, and dynamically adjust the route according to the flight distance, power limit and obstacle situation; BIM model generation: after arriving at the key target point, constructing the BIM model of the inspection target through three-dimensional laser scanning technology, and using SLAM technology to realize real-time construction of drones Mapping and positioning, reducing pre-processing dependence, and establishing a BIM incremental update mechanism to only scan changed areas; edge point extraction: extracting the edge contour of the BIM model through Canny edge detection combined with Hough transform; inspection path generation: using genetic algorithm combined with ant colony algorithm, and introducing the Pareto optimal concept, the inspection path is optimized to improve inspection efficiency and coverage; inspection point planning: planning inspection points according to the average inspection path to achieve comprehensive coverage of inspection targets; inspection task execution: the drone performs inspection tasks according to the planned inspection path and inspection points, adjusts the shooting angle, focal length, exposure parameters, collects image data in real time and transmits it back to the ground station, and the intelligent analysis subsystem can intelligently identify the data and generate reports.

[0069] Beneficial effects:

[0070] (1) It combines a variety of advanced technologies, such as LiDAR scanning, SLAM technology, reinforcement learning, genetic algorithm and ant colony algorithm, to improve inspection efficiency, coverage and safety.

[0071] (2) Through reinforcement learning, the obstacle avoidance model is dynamically trained to autonomously plan obstacle avoidance paths, achieving dynamic obstacle avoidance and path optimization to reduce flight risks. It supports autonomous flight and mission execution in complex environments.

[0072] (3) Use SLAM technology to achieve real-time mapping and positioning of drones, reducing reliance on preprocessing.

[0073] (4) It is easy to promote and apply, and the operation is relatively simple, making it suitable for promotion and application in more application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A flow chart of a method for autonomous substation inspection by drones provided in one embodiment of the present application.

[0075] Figure 2 A schematic diagram of three spherical areas with preset obstacle avoidance radii of R1, R2, and R3 provided in an embodiment of the present application.

[0076] Figure 3 Schematic diagram of a model showing the traction force of a drone and the repulsive force exerted on the drone by obstacles provided in one embodiment of the present application.

[0077] Figure 4 A schematic diagram of a module of a substation drone autonomous inspection system provided by one embodiment.

[0078] Figure 5 A schematic diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0079] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0080] It should be noted that, in the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art in the art to which this application relates. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0081] It should be noted that, in the embodiments of the present application, words such as "first" and "second" are only used for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order. Features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way.

[0082] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0083] Example 1

[0084] The present application provides a method and system for autonomous inspection of substations by drones, the core contents of which include seven steps: task pre-planning, flight path generation, BIM model generation, edge point extraction, inspection path generation, inspection point planning, and inspection task execution. These steps combine a variety of advanced technologies, such as LiDAR scanning, SLAM technology, reinforcement learning, genetic algorithm, ant colony algorithm, edge detection, etc., and combine a variety of advanced technologies to achieve efficient, intelligent and safe substation inspections, providing strong support for substation operation and maintenance management. Among them, the obstacle avoidance model is dynamically trained through reinforcement learning to autonomously plan the obstacle avoidance path, realize dynamic obstacle avoidance and path optimization, and reduce flight risks. Support autonomous flight and task execution in complex environments. Use SLAM technology to achieve real-time mapping and positioning of drones, reducing preprocessing dependence.

[0085] Figure 1 This is a flow chart of the autonomous inspection method for substations provided by drones in one embodiment of the present application. Figure 1 As shown, a substation drone autonomous inspection method includes:

[0086] S1: Mission Pre-Planning: Generate a high-precision 3D map of the inspection area through preliminary LiDAR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land based on the preset preliminary routes and mission markers.

[0087] Specifically, S1: Mission Pre-Planning: Generate a high-precision three-dimensional map of the inspection area through preliminary LiDAR scanning, mark key target points according to the mission, preset a preliminary route, and automatically take off, operate, and land according to the preset preliminary route and mission markings, including:

[0088] LiDAR collects centimeter-level precision 3D terrain and target feature information of the transmission line corridor and surrounding environment, generating a 3D map of the inspection area.

[0089] Generate preliminary routes using 3D map data based on mission requirements and terrain characteristics;

[0090] Using AI algorithms, key features in 3D point clouds are quickly extracted, key inspection target points are marked on 3D maps, and then linked to records in the power grid asset database.

[0091] According to the preset routes and mission markers, the drone can take off autonomously, perform inspection tasks along the planned route and land automatically.

[0092] It is understandable that in this embodiment, laser radar (LiDAR) technology is widely used in inspection tasks to obtain centimeter-level precision three-dimensional terrain and target feature information of the transmission line corridor and the surrounding environment. These data can generate high-precision three-dimensional maps, providing basic support for subsequent route planning. On the basis of the three-dimensional map, combined with the task requirements and terrain characteristics, a preliminary route is generated using route planning software or algorithm. For example, it is mentioned that route planning mainly adopts a method based on three-dimensional point cloud data, by extracting the spatial information of the inspection target points and generating routes in combination with the inspection requirements. In addition, it is also pointed out that the original point cloud is classified by the point cloud filtering algorithm, and the tower components are marked, and then the route planning algorithm is used to generate the initial route.

[0093] It is understandable that in this embodiment, laser radar (LiDAR) technology is widely used in inspection tasks to obtain centimeter-level precision three-dimensional terrain and target feature information of the transmission line corridor and the surrounding environment. These data can generate high-precision three-dimensional maps, providing basic support for subsequent route planning. On the basis of the three-dimensional map, combined with the task requirements and terrain characteristics, a preliminary route is generated using route planning software or algorithms. For example, a method based on three-dimensional point cloud data is used to extract the spatial information of the inspection target points and generate a route in combination with the inspection requirements. The original point cloud can also be classified by a point cloud filtering algorithm, and the tower components can be marked, and then the route planning algorithm can be used to generate the initial route.

[0094] AI algorithms are used to quickly extract key features (such as towers, conductors, and insulators) from 3D point clouds and mark key inspection target points on a 3D map. These target points are linked to records in the grid asset database to ensure the accuracy of inspections.

[0095] Based on pre-set routes and mission markers, the drone can autonomously take off, perform inspections along the planned route, and land automatically. Using offline 3D point clouds, the drone can autonomously plan its flight path and perform functions such as autonomous takeoff and landing, and automatic photography. It can also utilize artificial intelligence technology for route planning, obstacle avoidance, and autonomous takeoff and landing.

[0096] LiDAR collects high-precision 3D maps, combined with AI algorithms to extract key features and generate preliminary routes, ultimately enabling autonomous drone inspections. This process, encompassing data collection, route planning, target point marking, and autonomous flight, demonstrates the efficiency and intelligence of drone intelligent inspection technology.

[0097] S2: Flight path generation: After the inspection target is input, the drone inspection system can autonomously plan the obstacle avoidance path based on the preset tasks and real-time environmental data, through the dynamic training of the obstacle avoidance model through reinforcement learning, and dynamically adjust the route according to the flight distance, power limit and obstacle conditions.

[0098] It can be understood that in this embodiment, the inspection target and environmental data are input, including preset tasks and real-time data, wherein the preset tasks are combined with the three-dimensional map generated by LiDAR to mark key target points (such as poles and wires) and are associated with the power grid asset database. Real-time data, such as dynamic obstacle information (such as flying birds and temporary buildings) obtained through visual sensors or LiDAR, is input into the UAV reinforcement learning dynamic training obstacle avoidance model.

[0099] The framework of the dynamic training obstacle avoidance model for reinforcement learning can be a hierarchical deep reinforcement learning (DRL) model: the bottom layer processes real-time obstacle avoidance (such as DDPG), and the top layer coordinates global track point tracking; it can also be a recurrent double Q network (DDQN-GRU): processing time series information to enhance the adaptability of some observable environments.

[0100] Specifically, S2: Flight Path Generation: Inputting the inspection target, the UAV inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data by dynamically training the obstacle avoidance model through reinforcement learning, including:

[0101] The agent drone continuously interacts with the environment and obtains the optimal value function q of the state s through the feedback or rewards given by the environment, so as to continuously optimize the state-action to obtain the optimal strategy f. The problem of finding the best strategy is transformed into finding the maximum value of the action-state value function generated under all strategies.

[0102] The training process is a path planning algorithm based on reinforcement learning, which allows the drone to learn through trial and error and obtain rewards through continuous interaction with the environment. The optimal value function q(s,a) and the optimal strategy f(s|a) are expressed as follows:

[0103]

[0104] Among them, a represents the agent point state.

[0105] Figure 2 A schematic diagram of a spherical area with three preset obstacle avoidance radii of R1, R2, and R3 provided in an embodiment of the present application. Specifically, Figure 2 As shown, the reinforcement learning-based path planning algorithm allows the drone to learn through trial and error through continuous interaction with the environment and obtain rewards including:

[0106] With the drone as the center, three spherical areas with preset obstacle avoidance radii of R1, R2, and R3 are preset, where R1>R2>R3;

[0107] When the obstacle is within the different preset obstacle avoidance radius of the drone, different anti-collision avoidance strategies are implemented. Figure 3This is a schematic diagram of a model of the traction force of a drone and the repulsive force generated by obstacles on the drone provided in one embodiment of the present application. Figure 3 As shown, specifically:

[0108] Combining traction and repulsion, the gravitational force generated by the target point and the repulsive force generated by the obstacle on the drone are converted into rewards or penalties obtained by the drone after performing actions in the state. The optimized reward function is:

[0109]

[0110] Among them, d max Indicates the maximum distance between the starting point of the agent and the target. represents the distance between the agent point and the target at time t, represents the distance between the agent point and the obstacle at time t, and D represents the distance between the drone and the obstacle.

[0111] Specifically, such as Figure 3 As shown, when the obstacle is within the safe area or no obstacle is detected, that is, D ≥ R2 or R2 ≤ D ≤ R1, the drone is only affected by the gravity generated by the target point, and the reward function is

[0112] When the obstacle is within the range of R3-R2, the drone is affected by the repulsive force of the obstacle and the traction force of the drone. The reward function decreases as the distance between the agent and the obstacle increases. The reward function consists of two parts:

[0113] When the obstacle is within the R3 range, the risk of collision between the UAV and the obstacle is high, and the UAV must adopt a collision avoidance strategy, setting the reward function to

[0114] The route is dynamically adjusted based on the comprehensive flight distance, power consumption, and successful approach to the target point. Specifically, a multi-objective optimization method can be set to set different reward weights for comprehensive flight distance, power consumption, successful approach to the target point, etc. for dynamic training and path planning.

[0115] S3: BIM model generation: After reaching the key target point, the BIM model of the inspection target is constructed through 3D laser scanning technology. SLAM technology is used to achieve real-time mapping and positioning of the drone, reducing pre-processing dependence, and establishing a BIM incremental update mechanism to only scan the changed areas.

[0116] Specifically, SLAM (Simultaneous Localization and Mapping) technology is the core method for achieving real-time mapping and positioning for drones. It allows drones in unknown environments to collect environmental information through sensors (such as lidar and IMU), simultaneously estimate their own position and attitude, and construct a map of the environment. SLAM technology can significantly reduce reliance on preprocessing and improve inspection efficiency. For example, in indoor environments or where GPS signals are limited, SLAM technology can use lidar data to generate accurate point cloud maps and achieve high-precision positioning and mapping through algorithm optimization.

[0117] Point cloud data generated based on 3D laser scanning and SLAM technology can be further converted into geometric BIM objects. These geometric objects can be used to create building information models (BIM) and, combined with historical image data for comprehensive analysis, extract deformation feature points, thereby enabling ground displacement deformation measurement. In addition, through an incremental update mechanism, only the changed area is scanned instead of the entire target area, which can greatly improve data processing efficiency and reduce computing costs. This mechanism is particularly suitable for scenarios where BIM models need to be updated regularly, such as building maintenance or construction monitoring.

[0118] Drones equipped with LiDAR and SLAM systems can perform inspection tasks in complex environments. For example, during power line inspections, drones use SLAM technology to accurately measure transmission lines and generate three-dimensional maps to assist in fault detection.

[0119] It is understood that in this embodiment, the S3: BIM model generation: after reaching the key target point, the BIM model of the inspection target is constructed through 3D laser scanning technology, and the SLAM technology is used to achieve real-time drone mapping and positioning, reducing the dependence on preprocessing, and establishing a BIM incremental update mechanism, scanning only the changed areas, including:

[0120] Use LiDAR to scan inspection targets and generate high-density point cloud data;

[0121] Collect the position and attitude data of the drone through IMU;

[0122] Collect image data through a camera;

[0123] Utilizing SLAM algorithms to optimize and process the point cloud data, position and posture data, and image data to establish an accurate BIM model;

[0124] The optimization process includes:

[0125] Adjust the position of each point in the point cloud data and minimize the global error to obtain a more stable and accurate three-dimensional model. The error formula e i for:

[0126] e i =z i -h(p i ),

[0127] Among them, z i represents the observed data, p i Indicates the position of each point, h(p i ) represents the observed data predicted by the model;

[0128] The optimization function to minimize the global error is:

[0129]

[0130] Where n represents the number of point cloud data;

[0131] Adjust the position of each point to minimize the error;

[0132] Based on the historical BIM model and current scan data, the position changes of deformation feature points are extracted through feature point matching and comparative analysis, thereby realizing incremental updates of the BIM model.

[0133] S4: Edge point extraction: The edge contour of the BIM model is extracted by combining Canny edge detection with Hough transform.

[0134] Specifically, Canny edge detection is a multi-stage algorithm used to accurately extract edges from images. Its main steps include: Gaussian filtering: Smoothing the image with a Gaussian kernel to reduce noise interference. Gradient calculation: Calculating the gradient amplitude and direction of the image to determine the potential edge location. Non-maximum suppression: Filtering edge candidate points along the gradient direction, removing non-maximum points, and retaining possible edges. Dual threshold detection: Using high and low thresholds to further filter candidate edge points, strong edge points are directly retained, while weak edge points are retained based on whether they are connected to strong edges.

[0135] It is understandable that, in this embodiment, since the BIM model has been pre-built in the early stage, there is no need to go through the Gaussian filtering step.

[0136] The Hough transform is a geometric feature detection method primarily used to detect geometric shapes such as lines and circles in images. Its basic concept is to map points in image space to parameter space and detect specific geometric shapes by counting peaks in the parameter space. For line detection, the Hough transform converts the line equation ρ = xcosθ + ysinθ into parameter space and calculates the cumulative value of each parameter combination to find the line's parameters.

[0137] Specifically, the S4: edge point extraction: extracting the edge contour of the BIM model by combining Canny edge detection with Hough transform recognition, including:

[0138] Obtain the point cloud dataset of the BIM model;

[0139] By setting two high and low thresholds, the edge points are divided into strong edges and weak edges. Strong edges are directly retained, while weak edges are retained based on their connection relationship with strong edges, thus obtaining an edge point set.

[0140] Map the edge points detected by Canny to Hough space, where each edge point corresponds to a straight line;

[0141] Count the likelihood of the line in the accumulator matrix;

[0142] By setting a threshold, significant line candidates are screened out and the line features in the image are finally extracted;

[0143] Connect isolated edge points into continuous edge lines.

[0144] Specifically, in practical applications, Canny edge detection is combined with Hough transform. First, edge points in the image are extracted using Canny edge detection, and then these edge points are input into Hough transform for further processing. The specific steps are as follows:

[0145] Extract edge points: All edge points in the BIM model image are extracted using the Canny algorithm. Parameter space mapping: The extracted edge points are mapped to the Hough parameter space, and the cumulative value of each parameter combination is counted. Peak detection: Peaks are found in the Hough parameter space, which correspond to straight line features in the image. Line extraction: The straight line contours of the model are extracted based on the results of the Hough transform, and optimized in combination with the edge points detected by Canny to ensure that the extracted contours are accurate and coherent. The Canny algorithm can effectively suppress noise and retain important features, while the Hough transform can accurately detect straight line features. Through dual threshold screening and parameter space statistics, this method is highly robust to noise and image quality. The edge contours of the BIM model can be efficiently extracted by combining Canny edge detection with the Hough transform. This method combines the advantages of both technologies, ensuring the accuracy of edge extraction while improving robustness to noise.

[0146] S5: Inspection path generation: Genetic algorithm is combined with ant colony algorithm, and the Pareto optimal concept is introduced to optimize the design of inspection paths to improve inspection efficiency and coverage.

[0147] Specifically, the genetic algorithm is an optimization algorithm based on the principles of natural selection and genetics. It generates new solutions through operations such as crossover and mutation. It has strong global search capabilities, but is prone to getting stuck in local optima and has slow convergence. The ant colony algorithm, on the other hand, is a heuristic optimization algorithm that mimics the pheromone mechanism used in ant foraging behavior, enabling it to quickly find high-quality solutions. However, low initial pheromone concentrations can lead to slow convergence. Combining the two can overcome their respective shortcomings. For example, the genetic algorithm is used for global search, while the ant colony algorithm is used for local optimization, thereby improving the efficiency and quality of path planning.

[0148] Pareto optimality is a multi-objective optimization method that aims to find a set of solutions such that improving one objective does not degrade other objectives. This method is particularly suitable for inspection tasks that require simultaneous consideration of multiple objectives (such as path length, inspection time, and energy consumption). Using Pareto optimality, decision makers can select the optimal solution set based on their actual needs, enabling more flexible path planning.

[0149] A genetic algorithm is used to generate an initial set of paths, and new path solutions are generated through crossover and mutation operations. The paths generated by the genetic algorithm serve as the initial search paths for the ant colony algorithm, which then performs local optimization on these paths to improve their quality. During the optimization process, the Pareto optimality theory is introduced to decompose the multi-objective optimization problem into multiple single-objective optimization problems, which are then solved separately and ultimately integrated into a set of Pareto optimal solutions.

[0150] This inspection path generation method, which combines a genetic algorithm with an ant colony algorithm and incorporates the concept of Pareto optimality, is an efficient and flexible solution. It not only improves inspection efficiency and coverage, but also meets the requirements of multi-objective optimization, providing strong support for inspection tasks in complex environments.

[0151] S6: Inspection point planning: Plan inspection points evenly based on the inspection route to achieve comprehensive coverage of inspection targets.

[0152] It is understood that, in this embodiment, the S6: Inspection point planning: planning inspection points according to the average inspection path to achieve full coverage of the inspection target, includes:

[0153] Calculating the total length of the inspection path according to the edge line;

[0154] Calculate the number of inspection points based on the horizontal or vertical range of the camera and the total length of the inspection path;

[0155] The inspection points are divided according to the number of the inspection points.

[0156] Specifically, first, the total length of the inspection path needs to be calculated based on the edge line of the inspection path. This step is the basis for the entire planning and provides a basis for the subsequent calculation of the number of inspection points. According to the shooting range of the camera (horizontal or vertical range) and the total length of the inspection path, the number of inspection points that need to be set is calculated. Parameters such as shooting distance and angle will directly affect the distribution of points. The path point optimization planning requires a coverage rate of 100%, so the calculation of the number of points needs to fully consider the coverage range. Finally, based on the calculated number of inspection points, the inspection path is divided into several inspection points. These points should be evenly distributed throughout the inspection path to ensure full coverage of the target area.

[0157] S7: Inspection mission execution: The drone performs inspection tasks according to the planned inspection routes and inspection points, adjusts the shooting angle, focal length, and exposure parameters, collects image data in real time and transmits it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

[0158] As you can understand, in this embodiment, the data received by the ground station is processed by the intelligent analysis module, which uses artificial intelligence algorithms to identify and analyze images, such as detecting temperature anomalies and crack expansion, and generate inspection reports. The AI module can automatically analyze anomalies in the data and issue alarms, while also combining historical data to predict equipment failures and provide maintenance recommendations. Furthermore, the intelligent analysis system can generate visual reports that display key information from the inspection process.

[0159] Example 2

[0160] like Figure 4 As shown, the present application provides a substation drone autonomous inspection system, which is applied to the substation drone autonomous inspection method as described in Example 1, including: a task pre-planning module 11, a flight path generation module 12, a BIM model generation module 13, an edge point extraction module 14, an inspection path generation module 15, an inspection point planning module 16, and an inspection task execution module 17.

[0161] It can be understood that in this embodiment, the mission pre-planning module 11 is used to generate a high-precision three-dimensional map of the inspection area through preliminary LiDAR scanning, mark key target points according to the mission, preset a preliminary route, and automatically take off, operate, and land according to the preset preliminary route and mission marking;

[0162] It can be understood that in this embodiment, the flight path generation module 12 is used to input the inspection target, and the UAV inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data by dynamically training the obstacle avoidance model through reinforcement learning, and dynamically adjust the route according to the flight distance, power limit and obstacle conditions;

[0163] It can be understood that in this embodiment, the BIM model generation module 13 is used to construct a BIM model of the inspection target through three-dimensional laser scanning technology after reaching the key target point, use SLAM technology to achieve real-time mapping and positioning of the drone, reduce pre-processing dependence, and establish a BIM incremental update mechanism to scan only the changed areas;

[0164] It can be understood that, in this embodiment, the edge point extraction module 14 is used to extract the edge contour of the BIM model by combining Canny edge detection with Hough transform;

[0165] It is understood that, in this embodiment, the inspection path generation module 15 is used to optimize the inspection path by using a genetic algorithm combined with an ant colony algorithm and introducing the Pareto optimality concept to improve inspection efficiency and coverage.

[0166] It can be understood that, in this embodiment, the inspection point planning module 16 is used to plan inspection points according to the average inspection path to achieve full coverage of the inspection target;

[0167] It can be understood that in this embodiment, the inspection task execution module 17 is used for the UAV to perform inspection tasks according to the planned inspection path and inspection points, adjust the shooting angle, focal length, and exposure parameters, collect image data in real time and transmit it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

[0168] Figure 5 This is an electronic device provided by an embodiment of the present application. Figure 5 As shown, the electronic device includes at least the following parts: a processor 101 and a memory 100 , a communication interface 103 , and a bus 102 .

[0169] In the embodiment of the present application, the memory 100 is used to store instructions executable by the processor 101. The processor 101 is configured to execute the instructions to implement the following Figure 4 The equipment module shown is for autonomous drone inspection of substations.

[0170] In an embodiment of the present application, a computer-readable storage medium includes instructions, and the instructions instruct a device to execute the method of the first aspect. For example, the instructions instruct the device to execute Figure 1 The method is shown in the process steps.

[0171] The program running in the electronic device involved in one embodiment of the present application can be a program that controls a central processing unit (CPU) and the like to realize the functions of the above-mentioned embodiment involved in one embodiment of the present invention (a program that enables a computer to function). Then, the information processed by these devices is temporarily stored in a random access memory (RAM) during its processing, and then stored in various ROMs such as read-only memory (Flash ROM) and hard disk drive (HDD), and is read, modified, and written by the CPU as needed.

[0172] It should be noted that a portion of the electronic device of the above embodiment may also be implemented by a computer. In this case, a program for implementing the control function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer and executed.

[0173] It should be noted that the "computer" mentioned here refers to a computer built into an electronic device, employing hardware including an operating system (OS) and peripheral devices. Furthermore, "computer-readable recording medium" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computers.

[0174] Furthermore, "computer-readable recording media" may include: media that dynamically store programs for a short period of time, such as communication lines when transmitting programs via networks such as the Internet or communication lines such as telephone lines; and media that store programs for a fixed period of time, such as volatile memory within computers acting as servers or clients in this context. Furthermore, the aforementioned program may be a program for implementing a portion of the aforementioned functions, or a program that can achieve the aforementioned functions by combining with a program already stored in a computer.

[0175] Furthermore, the electronic device in the above-described embodiments can also be implemented as a collection (device group) consisting of multiple devices. Each device constituting the device group may have a portion or all of the functions or functional blocks of the electronic device in the above-described embodiments. A device group only needs to have all the functions or functional blocks of the electronic device.

[0176] Those skilled in the art should recognize that the above embodiments are merely intended to illustrate the present application and are not intended to limit the present application. As long as they are within the spirit of the present application, appropriate changes and modifications to the above embodiments are within the scope of protection claimed in the present application.

Claims

1. A substation drone autonomous inspection method, characterized in that: The method comprises: S1: Mission Pre-Planning: Generate a high-precision 3D map of the inspection area through preliminary LiD AR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land according to the preset preliminary routes and mission markings; S2: Flight Path Generation: After inputting the inspection target, the UAV inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data through the dynamic training of the obstacle avoidance model through reinforcement learning, and dynamically adjust the route according to the flight distance, power limit and obstacle conditions; S 3: BIM model generation: After reaching the key target point, a BIM model of the inspection target is constructed using 3D laser scanning technology. SLAM technology is used to achieve real-time mapping and positioning of the drone, reducing the reliance on preprocessing. A BIM incremental update mechanism is also established to scan only the changed areas. S 4: Edge point extraction: extracting the edge contour of the BIM model by using Canny edge detection combined with Hough transform; S5: Inspection path generation: Genetic algorithm combined with ant colony algorithm and the Pareto optimal concept are introduced to optimize the inspection path design to improve inspection efficiency and coverage; S 6: Inspection point planning: Plan inspection points evenly according to the inspection route to achieve comprehensive coverage of inspection targets; S 7: Inspection mission execution: The drone performs inspection tasks according to the planned inspection routes and inspection points, adjusts the shooting angle, focal length, and exposure parameters, collects image data in real time and transmits it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

2. The substation autonomous inspection method using a drone according to claim 1, characterized in that: S1: Mission Pre-Planning: Generate a high-precision 3D map of the inspection area through preliminary LiD AR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land according to the preset preliminary routes and mission markings, including: LiDAR collects centimeter-level precision 3D terrain and target feature information of the transmission line corridor and surrounding environment, generating a 3D map of the inspection area. Generate preliminary routes using 3D map data based on mission requirements and terrain characteristics; Using AI algorithms, key features in 3D point clouds are quickly extracted, key inspection target points are marked on 3D maps, and then linked to records in the power grid asset database. According to the preset routes and mission markers, the drone can take off autonomously, perform inspection tasks along the planned route and land automatically.

3. The substation autonomous inspection method using a drone according to claim 2, characterized in that: S2: Flight Path Generation: Input the inspection target. The UAV inspection system can autonomously plan the obstacle avoidance path based on the preset task and real-time environmental data by dynamically training the obstacle avoidance model through reinforcement learning, including: The agent drone continuously interacts with the environment and obtains the optimal value function q of the state s through the feedback or rewards given by the environment, so as to continuously optimize the state-action to obtain the optimal strategy f. The problem of finding the best strategy is transformed into finding the maximum value of the action-state value function generated under all strategies. The training process is a path planning algorithm based on reinforcement learning, which allows the drone to learn through trial and error and obtain rewards through continuous interaction with the environment. The optimal value function q(s,a) and the optimal strategy f(s|a) are expressed as follows: Where a represents the agent point state.

4. The substation autonomous inspection method using a drone according to claim 3 is characterized in that: The reinforcement learning-based path planning algorithm allows the drone to learn through trial and error through continuous interaction with the environment and obtain rewards including: With the drone as the center, three spherical areas with preset obstacle avoidance radii of R1, R2, and R3 are preset, where R1>R2>R3; When obstacles are within different preset obstacle avoidance radius of the drone, different collision avoidance strategies are implemented, specifically: Combining traction and repulsion, the gravitational force generated by the drone's target point and the repulsive force generated by the obstacle on the drone are converted into rewards or penalties after the drone performs an action in the state. The optimized reward function is: Among them, d max Indicates the maximum distance between the starting point of the agent and the target. represents the distance between the agent point and the target at time t, represents the distance between the agent point and the obstacle at time t, and D represents the distance between the drone and the obstacle.

5. The substation autonomous inspection method using a drone according to claim 1, characterized in that: S3: BIM model generation: After reaching the key target point, a BIM model of the inspection target is constructed using 3D laser scanning technology. SLAM technology is used to achieve real-time drone mapping and positioning, reducing pre-processing dependency. A BIM incremental update mechanism is also established, scanning only the changed areas, including: Use LiDAR to scan inspection targets and generate high-density point cloud data; Collect the position and attitude data of the drone through IMU; Collect image data through a camera; Utilizing SLAM algorithms to optimize and process the point cloud data, position and posture data, and image data to establish an accurate BIM model; The optimization process includes: Adjust the position of each point in the point cloud data and minimize the global error to obtain a more stable and accurate three-dimensional model. The error formula e i for: e i =z i -h(p i ), Among them, z i represents the observed data, p i Indicates the position of each point, h(p i ) represents the observed data predicted by the model; The optimization function to minimize the global error is: Where n represents the number of point cloud data; Adjust the position of each point to minimize the error; Based on the historical BIM model and current scan data, the position changes of deformation feature points are extracted through feature point matching and comparative analysis, thereby realizing incremental updates of the BIM model.

6. The substation autonomous inspection method using a drone according to claim 4, characterized in that: S4: Edge point extraction: extracting the edge contour of the BIM model through Canny edge detection combined with Hough transform recognition, including: Obtain the point cloud dataset of the BIM model; By setting two high and low thresholds, the edge points are divided into strong edges and weak edges. Strong edges are directly retained, while weak edges are retained based on their connection relationship with strong edges, thus obtaining an edge point set. Map the edge points detected by Canny to Hough space, where each edge point corresponds to a straight line; Count the likelihood of the line in the accumulator matrix; By setting a threshold, significant line candidates are screened out and the line features in the image are finally extracted; Connect isolated edge points into continuous edge lines.

7. The substation autonomous inspection method using a drone according to claim 6, characterized in that: S6: Inspection point planning: Planning inspection points based on the average inspection route to achieve comprehensive coverage of inspection targets, including: Calculating the total length of the inspection path according to the edge line; Calculate the number of inspection points based on the horizontal or vertical range of the camera and the total length of the inspection path; The inspection points are divided according to the number of the inspection points.

8. A substation drone autonomous inspection system, applied to the substation drone autonomous inspection method according to any one of claims 1 to 7, characterized in that: include: The mission pre-planning module is used to generate a high-precision 3D map of the inspection area through preliminary LiD AR scanning, mark key target points according to the mission, preset preliminary routes, and automatically take off, operate, and land according to the preset preliminary routes and mission markings; The flight path generation module is used to input inspection targets. Based on preset tasks and real-time environmental data, the drone inspection system can autonomously plan obstacle avoidance paths through dynamic training of obstacle avoidance models through reinforcement learning, and dynamically adjust the route according to flight distance, power limit, and obstacle conditions. The BIM model generation module is used to construct a BIM model of the inspection target through 3D laser scanning technology after reaching the key target point. It uses SLAM technology to achieve real-time mapping and positioning of the drone, reducing the dependence on preprocessing, and establishes a BIM incremental update mechanism to scan only the changed areas; An edge point extraction module is used to extract the edge contour of the BIM model by using Canny edge detection combined with Hough transform; Inspection path generation module, which uses genetic algorithm combined with ant colony algorithm and introduces Pareto optimality concept to optimize the inspection path to improve inspection efficiency and coverage; Inspection point planning module, used to plan inspection points according to the inspection route to achieve comprehensive coverage of inspection targets; The inspection task execution module is used for drones to perform inspection tasks according to the planned inspection routes and inspection points, adjust the shooting angle, focal length, and exposure parameters, collect image data in real time and transmit it back to the ground station. The intelligent analysis subsystem can intelligently identify the data and generate reports.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the substation drone autonomous inspection method as described in any one of claims 1 to 7 when executing the instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, and the program instructs a device to execute the substation drone autonomous inspection method according to any one of claims 1 to 7.

Images