Power distribution network fine inspection and live working cooperative unmanned aerial vehicle system

Abstract

The application provides a power distribution network fine inspection and live working cooperative unmanned aerial vehicle system, relates to the power distribution network operation and maintenance technical field, and comprises an unmanned aerial vehicle inspection subsystem, a multi-sensor fusion defect identification module, a task planning module, a safety control module, a closed-loop iteration module and a digital twin visualization management and control platform.The unmanned aerial vehicle inspection subsystem adopts a non-vector space control strategy, constructs a conductor dynamic tubular domain model, and calculates the state deviation of collected images and pre-stored images in real time through a distance scalar.The application realizes the integration of self-adaptive tracking and fine defect identification of the inspection target, breaks through the limitation of traditional feature positioning, adopts the core idea of direct tracking without feature positioning, constructs a conductor dynamic tubular domain model, calculates the image state deviation in real time through a Hausdorff distance scalar, effectively improves the reliability and positioning accuracy of power distribution network equipment tracking, and adapts to the target tracking demand in complex inspection environments.

Description

Technical Field

[0001] This invention relates to the field of power distribution network operation and maintenance technology, and in particular to a collaborative unmanned aerial vehicle (UAV) system for refined inspection and live-line work of power distribution networks. Background Technology

[0002] In existing technologies, distribution network inspection and live-line work are mostly conducted in a separate operation mode. Inspection mainly relies on manual inspection or traditional drone inspection. Traditional drone inspection often uses vector space control strategies, relying on feature localization to achieve target tracking, and using single or a few sensors for defect detection. Live-line work, on the other hand, requires manual operation of drones or work equipment, combined with manual visual judgment of safe distances, and manual verification of work quality after the operation. Meanwhile, some distribution network operation and maintenance systems have introduced digital twin technology, but it is mostly used only for 3D modeling and status display of equipment, without deep integration with inspection and work processes, and lacks an effective closed-loop update mechanism. Currently, the coordination between drone inspection and live-line work is poor, and task planning relies heavily on preset routes, resulting in insufficient adaptability and difficulty in meeting the needs of scenarios with frequent distribution network line switching and modification. Specifically, traditional inspection target tracking relies on feature-based localization, which is susceptible to environmental interference, resulting in low tracking reliability and insufficient positioning accuracy, making it difficult to accurately track conductors and distribution network equipment. Defect identification uses a single sensor, limiting the types of defects that can be identified and resulting in low accuracy, leading to problems such as "unclear visibility and inaccurate judgment," requiring secondary manual analysis. Furthermore, traditional laser point cloud modeling is costly and time-consuming. Task planning lacks intelligent optimization, resource allocation is unreasonable, and drones are prone to returning to base prematurely due to insufficient power. High-risk areas have insufficient inspection frequency, and multi-drone network collaboration efficiency is low. Live-line working safety distances rely on manual visual judgment, resulting in large errors and poor safety, and lacking effective dynamic control and emergency protection mechanisms. Inspection and live-line working lack a closed-loop iterative mechanism, post-operation quality verification is not timely, digital twin model updates are lagging, and there is a data gap of "heavy inspection, light feedback," making it difficult to achieve the transformation of distribution network operation and maintenance from "fault repair" to "predictive maintenance." These shortcomings seriously restrict the level of refinement and intelligence in distribution network operation and maintenance. Therefore, this invention proposes a collaborative drone system for refined inspection and live-line working of distribution networks to solve the problems existing in the prior art. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a collaborative UAV system for refined inspection and live-line work in power distribution networks. This system integrates adaptive tracking of inspection targets with refined defect identification, overcoming the limitations of traditional feature-based localization. It adopts a core approach of direct tracking without feature-based localization, constructing a dynamic tubular domain model of conductors and calculating image state deviation in real time using Hausdorff distance scalars. This effectively improves the reliability and positioning accuracy of power distribution network equipment tracking, increasing the tracking reliability to over 99.99% and improving positioning accuracy by more than two times compared to traditional technologies. It solves the problems of traditional tracking relying on feature-based localization and being susceptible to environmental interference, adapting to target tracking needs in complex inspection environments.

[0004] To achieve the objectives of this invention, the following technical solution is employed: a collaborative UAV system for refined inspection and live-line work in power distribution networks, comprising a UAV inspection subsystem, a multi-sensor fusion defect identification module, a task planning module, a safety control module, a closed-loop iteration module, and a digital twin visualization management platform. The UAV inspection subsystem employs a non-vector space control strategy, constructs a dynamic tubular domain model of the conductor, and calculates the state deviation between the acquired image and the pre-stored image in real time using distance scalars to achieve precise tracking of power distribution network equipment. The multi-sensor fusion defect identification module integrates multiple sensors and combines AI visual navigation to achieve power distribution network equipment identification and subtle defect detection. The task planning module generates inspection and live-line work task plans based on a unified cognitive model of the power distribution network and heuristic optimization algorithms. The safety control module achieves dynamic management and control of the safety distance for live-line work through electric field induction and fuzzy control; the closed-loop iterative module is used to complete the equipment status comparison and incremental update of the digital twin model after the operation; the digital twin visualization management and control platform realizes the digital and visual management and control of the entire process of power distribution network operation and maintenance, and the modules work together to complete the collaborative tasks of refined inspection of power distribution network and live-line work.

[0005] Further improvements are made in that the non-vector space control strategy of the UAV inspection subsystem adopts the core idea of ​​direct tracking without feature localization, and the distance scalar is calculated using the Hausdorff distance, the calculation formula of which is: ; Where dH(X,Y) is the Hausdorff distance scalar, which is the state deviation between the acquired image and the pre-stored image; X is the set of pixels in the real-time acquired image; Y is the set of pixels in the pre-stored standard image; x is any pixel in set X; y is any pixel in set Y; d(x,y) is the Euclidean distance between pixel x and pixel y; sup represents the supremum, which is the maximum value in the set; inf represents the infrem, which is the minimum value in the set.

[0006] Further improvements are made in the following aspects: The dynamic tubular domain model of the conductor is dynamically adjusted based on the real-time position, diameter and running attitude of the conductor. The radius of the tubular domain is 3-5 times the diameter of the conductor. An adaptive meshing method is adopted, and the mesh size dynamically changes with the distance between the conductor and the UAV. The closer the distance, the finer the mesh division, ensuring tracking accuracy.

[0007] Further improvements include: the multi-sensor fusion defect identification module integrates a high-definition visible light camera, an infrared thermal imaging camera, a lidar, and a partial discharge sensor, and is equipped with the TINY AI visual navigation system to achieve accurate identification of more than 9 types of power distribution network equipment; defect identification adopts a combination of image frame splitting processing, feature point matching, and temperature anomaly analysis, and the feature point matching calculation formula is: ; Where Sij is the matching similarity between the i-th acquired image feature point and the j-th standard defect feature point; n is the number of dimensions of the feature point; xik is the k-th dimensional feature value of the i-th acquired image feature point; and xjk is the k-th dimensional feature value of the j-th standard defect feature point. The average value of all dimensions of the feature points of the i-th acquired image feature point; The average of all dimensional eigenvalues ​​for the j-th standard defect feature point; when When the match is successful, it is identified as a defect of the corresponding type.

[0008] Further improvements include: the multi-sensor fusion defect identification module automatically identifies more than 12 types of minor defects, and the generated structured defect report includes the defect location, type, severity, discovery time, line and equipment number information, and simultaneously marks the defect level, providing accurate data support for live-line work and adapting to scenarios with frequent switching and modification of distribution network lines.

[0009] A further improvement lies in the following: the unified cognitive model of the distribution network constructed by the task planning module integrates a fault propagation knowledge graph, and generates the optimal inspection and operation task planning scheme through a heuristic optimization algorithm. The objective function of the heuristic optimization algorithm is: ; Where F is the objective function value of the task planning, and the smaller the value, the better the planning scheme; These are the weighting coefficients for inspection route length, drone energy consumption, and operation time, respectively. L represents the total route length for UAV inspection and operation, in km; E represents the total energy consumption of the UAV in completing the task, in kWh; T represents the total time to complete the inspection and operation task, in h; Meanwhile, the algorithm constraints are as follows: , This is the maximum energy consumption for drone operation, preventing drones from returning to base prematurely due to insufficient power.

[0010] Further improvements include: the task planning module automatically increases the frequency of patrols in special areas such as wildfire risk zones and areas prone to external damage, with the frequency of patrols being 2-3 times that of regular areas, thus tilting resources towards high-risk scenarios; at the same time, it supports collaborative operations of multiple drones, allocating inspection and operation resources according to the endurance, load status and location of each drone.

[0011] Further improvements include: the safety control module incorporates an electric field strength sensor and a proximity induction array, dynamically calculates the minimum safe operating distance based on voltage level and real-time electric field gradient, and employs a fuzzy PID control algorithm to achieve distance regulation. The calculation formula for the output parameters of the fuzzy PID controller is as follows: , , ; in, These are the proportional coefficient, integral coefficient, and derivative coefficient of the fuzzy PID controller, respectively. These are the initial proportional coefficient, initial integral coefficient, and initial derivative coefficient of the PID controller, respectively. These are the proportional coefficient correction, integral coefficient correction, and differential coefficient correction obtained from fuzzy logic reasoning, respectively; when a sudden change in electric field or a decrease in insulation resistance is detected, emergency back-off and host traction protection actions are automatically triggered.

[0012] A further improvement is made in that: the closed-loop iterative module compares the differences in equipment state before and after the operation using a semantic segmentation algorithm. The formula for calculating the semantic segmentation similarity is: ; Where IoU is the intersection-union ratio of the equipment images before and after the operation, i.e., the state similarity; A is the semantic segmentation region of the equipment image after the operation; B is the semantic segmentation region of the equipment image before the operation. This is the intersection region of A and B, i.e., the region where the state remains unchanged; The region is the union of A and B, i.e., the overall equipment region. When IoU < 0.9, it is determined that the post-operation state has changed effectively, and an operation quality assessment report is automatically generated and a local incremental update of the digital twin model is triggered to achieve a closed-loop iteration of defect identification, operation execution, effect verification, and model correction.

[0013] Further improvements include: the digital twin visualization management platform constructs a high-precision digital twin model of power distribution network equipment based on LiDAR point clouds and inspection images, integrating five major functions: inspection data management, defect diagnosis and analysis, task scheduling, work process monitoring, and maintenance report generation; it supports collaborative operation management of multiple drones; and it combines AI algorithms to achieve health status assessment, fault prediction, and maintenance strategy optimization for power distribution network equipment. The health status assessment value is: , Among them, H1 is the equipment operating parameter compliance rate, H2 is the defect rectification completion rate, and H3 is the historical operating reliability score. The value of H ranges from 0 to 1, and the larger the value, the better the health status of the equipment.

[0014] The beneficial effects of this invention are as follows: 1. This invention integrates adaptive tracking of inspection targets with refined defect identification, breaking through the limitations of traditional feature-based localization. It adopts the core idea of ​​direct tracking without feature localization, constructs a dynamic tubular domain model of the conductor, and calculates the image state deviation in real time using Hausdorff distance scalar, effectively improving the reliability and positioning accuracy of power distribution equipment tracking. The tracking reliability rate is increased to over 99.99%, and the positioning accuracy is more than twice that of traditional technologies. It solves the problems of traditional tracking relying on feature localization and being susceptible to environmental interference, adapting to the target tracking needs in complex inspection environments. At the same time, the adaptive adjustment of the dynamic tubular domain model further ensures the stability of tracking and avoids the lag and errors of traditional tracking methods.

[0015] 2. This invention optimizes the multi-sensor fusion defect identification module, integrating a high-definition visible light camera, an infrared thermal imaging camera, a lidar, and a partial discharge sensor. Combined with the TINY AI visual navigation system, it achieves accurate identification of more than 9 types of distribution network equipment, with a comprehensive identification rate of over 95%. It can automatically identify more than 12 types of minor defects, generate structured defect reports, and mark defect levels without the need for manual secondary analysis. This solves the problems of unclear visibility and inaccurate judgment in traditional inspections, significantly improving the level of inspection precision. At the same time, it avoids the drawbacks of high cost and long cycle of traditional laser point cloud modeling. It can complete autonomous inspections without the need for preset flight paths, perfectly adapting to scenarios with frequent changes in distribution network lines, reducing operation and maintenance costs, and improving inspection efficiency. It is especially suitable for distribution network inspections in complex terrains such as mountainous and forested areas.

[0016] 3. This invention constructs an intelligent task planning mechanism. Based on the unified cognitive model of the power distribution network and the fault propagation knowledge graph, it combines a heuristic optimization algorithm to generate the optimal inspection and operation task planning scheme, rationally allocates drone resources, avoids drones returning to base prematurely due to insufficient power, ensures the stability of multi-drone networking and inspection efficiency, and automatically increases the frequency of inspections for special areas such as wildfire risk areas and sections prone to external damage, realizing the tilting of resources towards high-risk scenarios, effectively improving the ability to investigate hidden dangers in high-risk areas, reducing the fault rate of the power distribution network. At the same time, the task planning algorithm can be adjusted in real time according to the line switching situation, further improving the adaptability of the system and solving the problem of insufficient flexibility of traditional preset routes.

[0017] 4. This invention achieves dynamic and precise control of the safety distance for live-line work. The system has a built-in electric field strength sensor and proximity induction array, combined with a fuzzy PID control algorithm, to dynamically calculate and maintain the minimum safe working distance based on the voltage level and real-time electric field gradient. Rapid feed is allowed outside the safe distance, and the system switches to a high-damping position control mode after entering the safe distance threshold. When a sudden change in electric field or a decrease in insulation resistance is detected, emergency retraction and host traction protection actions are automatically triggered, completely replacing manual visual distance judgment, greatly improving the safety of live-line work, reducing the safety risks to workers, and solving the problems of low accuracy and poor safety of traditional live-line work safety distance control. It is suitable for live-line work requirements of different voltage levels such as 10kV and 20kV.

[0018] 5. This invention constructs a closed-loop iterative mechanism for defect identification, job execution, effect verification, and model correction. After the job is completed, sensors are automatically invoked for secondary imaging and thermal imaging comparison. The differences in equipment status are compared through semantic segmentation algorithms, and a job quality assessment report is generated and synchronously written into the digital twin, triggering incremental updates of the local model. This fills the data gap in the existing system, which emphasizes inspection but neglects feedback. The accuracy of the digital twin is continuously optimized as the frequency of jobs increases. At the same time, the digital twin visualization management and control platform realizes digital and visualized management and control of the entire distribution network operation and maintenance process. Combined with AI algorithms, it realizes equipment health status assessment, fault prediction, and operation and maintenance strategy optimization, promoting the transformation of the distribution network from emergency repair to predictive maintenance, improving the intelligent level of distribution network operation and maintenance management, and providing reliable decision support for intelligent operation and maintenance of the distribution network. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the composition of the present invention. Detailed Implementation

[0020] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0021] Example 1 according to Figure 1 As shown, this embodiment proposes a collaborative UAV system for refined inspection and live-line work in power distribution networks. This system enables precise tracking and refined defect identification of power distribution equipment. The UAV inspection subsystem employs a non-vector space control strategy, abandoning the traditional feature-based localization mode. The core idea is "direct tracking without feature localization." First, a dynamic tubular domain model of the conductor is constructed. The model is dynamically adjusted based on the conductor's real-time position, diameter, and operating attitude. The radius of the tubular domain is set to four times the conductor's diameter. An adaptive mesh generation method is used, with the mesh size dynamically changing with the distance between the conductor and the UAV: ​​when the distance between the UAV and the conductor is greater than 5m, the mesh size is set to 0.5m × 0.5m; when the distance is between 2 and 5m, the mesh size is set to 0.2m × 0.2m; and when the distance is less than 2m, the mesh size is set to 0.1m × 0.1m. The closer the distance, the finer the mesh generation, ensuring tracking accuracy.

[0022] During the tracking process, high-definition visible light cameras mounted on the drone acquire real-time images of the power lines and distribution network equipment. These images are then compared with pre-stored standard equipment images. The Hausdorff distance scalar is used to calculate the state deviation between the two images in real time. The calculation formula is as follows: Where X is the set of pixels in the real-time acquired image, Y is the set of pixels in the pre-stored standard image, and d(x,y) is the Euclidean distance between the two points, this formula can accurately quantify image deviation, thereby adjusting the UAV's flight attitude and position to achieve real-time and accurate tracking of power lines and distribution network equipment. Testing showed that this tracking method has a reliability rate of 99.995%, and its positioning accuracy is 2.3 times higher than traditional feature-based positioning technology, effectively solving the problems of traditional tracking being susceptible to environmental interference and lacking accuracy.

[0023] Simultaneously, the multi-sensor fusion defect identification module works in tandem, integrating a high-definition visible light camera, an infrared thermal imaging camera, a lidar, and a partial discharge sensor. It is equipped with the TINY AI visual navigation system, which employs lightweight artificial intelligence algorithms to eliminate reliance on real-time differential positioning (RTK) signals, enabling autonomous inspection even in signal-free areas. The high-definition visible light camera captures external defects, the infrared thermal imaging camera detects abnormal temperatures (such as overheated connectors), the lidar acquires 3D dimensional information, and the partial discharge sensor detects abnormal discharges. The multi-sensor data is fused using a combination of Naive Bayes classification and DS evidence theory to improve the accuracy of defect identification.

[0024] In the defect identification process, the acquired images are first split into frames, and feature points of each frame are extracted. Then, a feature point matching formula is used. Perform matching, when the similarity is... When the temperature exceeds a preset threshold (the threshold varies depending on the device, e.g., 70°C for conductor joints), it is identified as a temperature anomaly defect. This embodiment can accurately identify nine types of distribution network equipment (conductors, insulators, towers, joints, circuit breakers, disconnect switches, transformers, surge arresters, and distribution boxes), achieving a comprehensive identification rate of 95.8%. It can automatically identify 12 types of minor defects, such as insulator damage, broken conductor strands, overheated joints, and partial discharge, generating a structured defect report containing information such as defect location, type, and severity, and simultaneously marking the defect level, eliminating the need for manual secondary analysis.

[0025] Example 2 according to Figure 1 As shown, this embodiment proposes a collaborative UAV system for refined inspection and live-line work of power distribution networks. Based on Embodiment 1, the multi-sensor fusion defect identification module is optimized to further improve the accuracy and efficiency of defect identification. First, the acquisition parameters of each sensor are optimized: the resolution of the high-definition visible light camera is set to 4K (3840×2160) and the frame rate is set to 30fps to ensure that the subtle details of the equipment are captured; the temperature measurement range of the infrared thermal imaging camera is -20℃ to 150℃, and the temperature measurement accuracy is ±0.5℃, which can accurately capture the minute temperature changes of the equipment, especially suitable for detecting hidden defects such as overheating of joints; the ranging range of the lidar is 0.5m to 100m, and the ranging accuracy is ±1cm, which can accurately obtain the three-dimensional size information of the equipment and assist in judging defects such as deformation and damage of the equipment; the detection frequency range of the partial discharge sensor is 10kHz to 1MHz, which can accurately detect the partial discharge signal of the equipment and avoid missing discharge defects.

[0026] Secondly, the multi-sensor data fusion algorithm was optimized by adopting a weighted fusion strategy. Weights were assigned to different sensors based on their reliability in different defect identification scenarios: In the identification of visual defects (such as insulator damage and broken conductor strands), the weight of the high-definition visible light camera was set to 0.4, the weight of the lidar to 0.3, and the weights of the infrared thermal imaging camera and the partial discharge sensor to 0.15 each; in the identification of temperature anomaly defects (such as overheated joints), the weight of the infrared thermal imaging camera was set to 0.5, the weight of the high-definition visible light camera to 0.2, and the weights of the lidar and the partial discharge sensor to 0.15 each; in the identification of partial discharge defects, the weight of the partial discharge sensor was set to 0.6, and the weights of the high-definition visible light camera and the infrared thermal imaging camera to 0.2 each. This weighted fusion effectively improved the accuracy of defect identification and avoided misjudgments by a single sensor.

[0027] Furthermore, a self-learning function for defect identification has been added. The system can automatically record defect data for each identification, including defect images, feature parameters, and identification results. Through machine learning algorithms, the feature point matching model and defect judgment threshold are continuously optimized. With increased usage frequency, the accuracy of defect identification can be further improved. Simultaneously, the generation logic of structured defect reports has been optimized, adding on-site photos of defects and predictions of defect development trends to the reports, providing more comprehensive and accurate data support for live-line work. In this embodiment, the identification rate of minor defects reaches 94.2%, an improvement of 1.3 percentage points compared to Embodiment 1, and the defect misjudgment rate is reduced to below 0.5%, significantly improving the precision of inspections. It also avoids the drawbacks of high cost and long cycle of traditional laser point cloud modeling, enabling autonomous inspections without pre-setting flight paths. This makes it suitable for scenarios with frequent changes in distribution network lines, especially applicable to distribution network inspections in complex regions such as Xinjiang and mountainous areas.

[0028] Example 3 according to Figure 1 As shown in the figure, this embodiment proposes a collaborative UAV system for refined inspection and live-line work in power distribution networks. This system enables intelligent planning of inspection and live-line work tasks. First, a unified cognitive model of the power distribution network is constructed. This model integrates information such as power distribution network equipment parameters, line topology, historical defect data, fault records, and meteorological data. Combined with a fault propagation knowledge graph, it identifies potential defect risk information. The fault propagation knowledge graph, based on the fault types in the power distribution network (such as short circuits, grounding, equipment damage, etc.), constructs a relationship network including fault propagation paths, impact ranges, and associated equipment. This allows for accurate identification of potential defect risk points and risk levels (general, moderate, severe).

[0029] The task planning module employs a heuristic optimization algorithm, aiming to achieve the shortest route, lowest energy consumption, and shortest time, to generate the optimal inspection and operation task planning scheme. The objective function is: ,in L is the total route length, E is the total energy consumption, and T is the total time. The constraints are as follows: ( (The maximum power consumption for the drone's flight is set to 5 kWh in this embodiment). The algorithm first prioritizes the inspection areas based on the risk level identified by the unified cognitive model of the power distribution network. High-risk areas (such as areas with serious defects, wildfire risk areas, and areas prone to external damage) have the highest priority, followed by medium-risk areas, and low-risk areas have the lowest priority.

[0030] Different inspection strategies are planned for areas with different priorities: the number of inspections in high-risk areas is set at 2.5 times that of regular areas, and the inspection routes are densely planned to ensure that every risk point is covered; the number of inspections in medium-risk areas is set at a regular frequency, with routes balancing efficiency and coverage; and the number of inspections in low-risk areas is set at a sparse frequency, prioritizing inspection resources for high-risk areas. At the same time, based on the currently available drone resources (including the drone's endurance, load status, and location), inspection and operation tasks are rationally allocated to avoid overloading or insufficient endurance of a single drone.

[0031] In this embodiment, five drones are configured to operate collaboratively, with three used for routine inspections and two used for high-risk area inspections and live-line work support. When the energy consumption of a particular drone reaches... When the drone reaches 80% completion, the system automatically plans a return route and schedules other idle drones to take over the unfinished tasks, preventing drones from returning prematurely due to insufficient power. Testing showed that this task planning method improved inspection efficiency by 35%, reduced drone return rate by 80%, and improved the stability of multi-drone networks by 40%. It effectively solves the problems of unreasonable resource allocation and low efficiency in traditional task planning, while also being suitable for scenarios with frequent power grid line changes. After a line change, the system can regenerate the optimal inspection route within 10 minutes without manual intervention.

[0032] Example 4 according to Figure 1 As shown in the figure, this embodiment proposes a collaborative UAV system for refined inspection of power distribution networks and live-line work, which is used to realize dynamic control of safe distance during live-line work. The system has a built-in electric field strength sensor and a proximity induction array. The electric field strength sensor has a detection range of 0~100kV / m and a detection accuracy of ±0.1kV / m. The proximity induction array consists of 8 sensing units, which are evenly distributed around the working unit, and can detect the distance and electric field gradient between the working unit and the live equipment from all directions.

[0033] When the submachine gun approaches the live equipment, the system first presets different minimum safe working distances based on the voltage level of the live equipment (covering two common voltage levels, 10kV and 20kV, in this embodiment): the minimum safe working distance for 10kV voltage level is set to 0.7m, and the minimum safe working distance for 20kV voltage level is set to 1.0m. Then, the system collects electric field gradient data in real time using an electric field strength sensor, and combines this with the detection data from the proximity induction array to dynamically calculate the actual distance between the submachine gun and the live equipment. A fuzzy PID control algorithm is used to achieve distance regulation. The calculation formula for the output parameters of the fuzzy PID controller is as follows: , , .

[0034] In this embodiment, the initial parameters of the PID controller are set as follows: Kp0=5.0, Ki0=0.1, Kd0=0.5. The input quantities for the fuzzy logic inference are "the deviation between the actual distance and the minimum safe distance" and "the rate of change of the deviation", and the output quantity is... The fuzzy word set is set to {zero, small, medium, large}, and the quantization factor is determined by empirical formulas. When the actual distance between the submachine and the live equipment is greater than the minimum safe distance, the fuzzy PID controller outputs a larger Kp and a smaller Kd, allowing the submachine to feed quickly and improve work efficiency. When the actual distance enters the safe distance threshold (minimum safe distance + 0.2m), the controller switches to high-damping position control mode, increasing Kd and decreasing Kp, so that the submachine approaches slowly, ensuring distance control accuracy. When a sudden change in electric field is detected (electric field strength change rate greater than 5kV / m·s) or a decrease in insulation resistance (insulation resistance less than 100MΩ), the system automatically triggers emergency retraction and main unit traction protection actions. The submachine retreats urgently at a speed of 0.5m / s, while the main unit pulls the submachine away from the live equipment, completely replacing manual visual distance judgment.

[0035] Tests have shown that this safety distance control method achieves a control accuracy of ±0.1m and an emergency retraction response time of ≤0.5s. It can effectively avoid safety hazards during live-line work, improve the safety of live-line work, and is suitable for various live-line work scenarios such as replacing insulators and tightening joints. It solves the problems of low accuracy and poor safety of traditional live-line work safety distance control.

[0036] Example 5 according to Figure 1 As shown, this embodiment proposes a collaborative UAV system for refined inspection and live-line work of power distribution networks. This system is used to achieve closed-loop iteration after operation and visualized management and control of the entire power distribution network operation and maintenance process. First, a high-precision digital twin model of power distribution network equipment is constructed. Based on lidar point cloud and inspection images, a three-dimensional reconstruction algorithm is used to restore the three-dimensional structure, operating status and surrounding environment of the power distribution network equipment. The accuracy of the model reaches ±0.5cm, which can realize three-dimensional visualization of equipment and real-time mapping of status. The synchronization update cycle between the digital twin model and the actual equipment status is set to 10 minutes.

[0037] After completing live-line work (such as replacing insulators), the closed-loop iterative module automatically calls the high-definition visible light camera and infrared thermal imaging camera of the UAV inspection subsystem to perform secondary imaging of the work point, acquiring equipment images after the work, comparing them with equipment images before the work, and calculating the state similarity between the two through a semantic segmentation algorithm. The calculation formula is as follows: Where A is the semantic segmentation region of the equipment image after the operation, and B is the semantic segmentation region of the equipment image before the operation. When IoU < 0.9, it is determined that the status has changed effectively after the operation, and the system automatically generates an operation quality assessment report. The report includes comparison images of the equipment before and after the operation, status change analysis, operation quality score, etc. The operation quality score adopts a 10-point scale, with a score ≥ 8 points being qualified and ≥ 9 points being excellent.

[0038] The work quality assessment report is simultaneously written into the distribution network digital twin, triggering a local model incremental update. Only the model data of the work area is updated, without the need to remodel the entire digital twin model, improving update efficiency by 40%. This allows the accuracy of the digital twin to continuously optimize as the frequency of work increases. Meanwhile, the digital twin visualization and control platform integrates five major functions: inspection data management, defect diagnosis and analysis, work task scheduling, work process monitoring, and maintenance report generation. It supports collaborative operation control of multiple drones and adopts a B / S architecture, accessible via computers, mobile phones, and other terminals.

[0039] The platform combines AI algorithms to assess the health status of distribution network equipment, predict faults, and optimize operation and maintenance strategies. The health status assessment value H = 0.3H1 + 0.4H2 + 0.3H3, where H1 is the equipment operating parameter compliance rate, H2 is the defect rectification completion rate, and H3 is the historical operational reliability score. When H < 0.6, the equipment is considered to be in poor health, and the system automatically issues an early warning and generates targeted operation and maintenance strategies; when H < 0.6, the equipment is considered to be in poor health. When the equipment is deemed to be in a generally healthy state, the system will remind the user to perform a regular inspection; when When the equipment is deemed to be in good health, the regular inspection frequency is maintained. In this embodiment, the accuracy of the digital twin model increases to over 98.5% with the frequency of operations. The platform can realize digital and visual management of the entire distribution network operation and maintenance process, promote the transformation of the distribution network from "fault repair" to "predictive maintenance", and improve the efficiency of operation and maintenance management.

[0040] Validation data: To verify the effectiveness and superiority of this invention, 50 10kV and 20kV distribution network lines with a total length of 120km were selected in a certain region, covering different terrains such as mountainous areas, plains, and urban areas. These lines included 860 sets of various distribution network equipment such as conductors, insulators, and towers, among which 42 defects were found (including 28 minor defects). Five drones described in this invention were deployed for a 30-day field test, which was compared with the traditional drone inspection + manual live-line operation mode. The test results are shown in the table below: Test results show that the present invention is superior to the traditional mode in terms of distribution network equipment tracking, defect identification, inspection efficiency, live-line work safety, and digital twin adaptation. All performance indicators meet the design requirements, and the operation and maintenance costs are significantly reduced. It is suitable for scenarios with frequent switching and modification of distribution network lines and can effectively solve many problems existing in the operation and maintenance of traditional distribution networks.

[0041] This collaborative UAV system for refined inspection and live-line work of power distribution networks integrates adaptive tracking of inspection targets with refined defect identification. Breaking through the limitations of traditional feature-based localization, it adopts a core approach of direct tracking without feature-based localization. It constructs a dynamic tubular domain model of conductors and calculates image state deviation in real time using Hausdorff distance scalars, effectively improving the reliability and positioning accuracy of power distribution network equipment tracking. The tracking reliability rate is increased to over 99.99%, and the positioning accuracy is more than twice that of traditional technologies. It solves the problems of traditional tracking relying on feature-based localization and being susceptible to environmental interference, adapting to target tracking needs in complex inspection environments. Furthermore, the adaptive adjustment of the dynamic tubular domain model further ensures tracking stability, avoiding the lag and errors of traditional tracking methods. This invention optimizes the multi-sensor fusion defect identification module, integrating a high-definition visible light camera, an infrared thermal imaging camera, a lidar, and a partial discharge sensor. Combined with the TINY AI visual navigation system, it achieves accurate identification of more than 9 types of distribution network equipment, with a comprehensive identification rate of over 95%. It can automatically identify more than 12 types of minor defects, generate structured defect reports, and mark defect levels without the need for secondary manual analysis. This solves the problems of unclear visibility and inaccurate judgment in traditional inspections, significantly improving the level of inspection precision. At the same time, it avoids the drawbacks of high cost and long cycle of traditional laser point cloud modeling. It can complete autonomous inspections without the need for preset flight paths, perfectly adapting to scenarios with frequent changes in distribution network lines, reducing operation and maintenance costs, and improving inspection efficiency. It is especially suitable for distribution network inspections in complex terrains such as mountainous and forested areas. This invention constructs an intelligent task planning mechanism. Based on a unified cognitive model of the power distribution network and a fault propagation knowledge graph, it combines a heuristic optimization algorithm to generate the optimal inspection and operation task planning scheme, rationally allocates drone resources, avoids drones returning to base prematurely due to insufficient power, ensures the stability of multi-drone networking and inspection efficiency, and automatically increases the frequency of inspections in special areas such as wildfire risk areas and sections prone to external damage, realizing the allocation of resources to high-risk scenarios. This effectively improves the ability to investigate hidden dangers in high-risk areas and reduces the fault rate of the power distribution network. At the same time, the task planning algorithm can be adjusted in real time according to line switching, further improving the system's adaptability and solving the problem of insufficient flexibility of traditional preset routes. This invention achieves dynamic and precise control of the safety distance for live-line work. The system incorporates an electric field strength sensor and a proximity induction array, combined with a fuzzy PID control algorithm. It dynamically calculates and maintains the minimum safe working distance based on the voltage level and real-time electric field gradient. Rapid feed is allowed outside the safe distance, and the system switches to a high-damping position control mode after entering the safe distance threshold. Furthermore, it automatically triggers emergency retraction and main unit traction protection actions when a sudden change in electric field or a decrease in insulation resistance is detected. This completely replaces manual visual distance judgment, significantly improving the safety of live-line work, reducing the safety risks to workers, and solving the problems of low accuracy and poor safety in traditional live-line work safety distance control. It is suitable for live-line work requirements at different voltage levels such as 10kV and 20kV.This invention constructs a closed-loop iterative mechanism encompassing defect identification, job execution, effect verification, and model correction. After a job is completed, sensors are automatically invoked for secondary imaging and thermal imaging comparison. Semantic segmentation algorithms are used to compare equipment status differences, generating a job quality assessment report which is simultaneously written into the digital twin, triggering incremental updates to the local model. This fills the data gap in existing systems that emphasize inspection over feedback, allowing the accuracy of the digital twin to continuously improve with increasing job frequency. Simultaneously, the digital twin visualization management platform enables digital and visualized management of the entire distribution network operation and maintenance process. Combined with AI algorithms, it achieves equipment health status assessment, fault prediction, and operation and maintenance strategy optimization, promoting the transformation of the distribution network from emergency repair to predictive maintenance, improving the intelligence level of distribution network operation and maintenance management, and providing reliable decision support for intelligent operation and maintenance of the distribution network.

[0042] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A collaborative UAV system for refined inspection and live-line work in power distribution networks, comprising a UAV inspection subsystem, a multi-sensor fusion defect identification module, a task planning module, a safety control module, a closed-loop iteration module, and a digital twin visualization management and control platform, characterized in that: The UAV inspection subsystem adopts a non-vector space control strategy, constructs a dynamic tubular domain model of the conductor, and calculates the state deviation between the acquired image and the pre-stored image in real time through distance scalar calculation to achieve accurate tracking of distribution network equipment; the multi-sensor fusion defect identification module integrates multiple sensors and combines AI visual navigation to achieve distribution network equipment identification and detection of minor defects; the task planning module generates inspection and live-line operation task plans based on the unified cognitive model of the distribution network and heuristic optimization algorithms. The safety control module achieves dynamic management and control of the safety distance for live-line work through electric field induction and fuzzy control; the closed-loop iterative module is used to complete the equipment status comparison and incremental update of the digital twin model after the operation; the digital twin visualization management and control platform realizes the digital and visual management and control of the entire process of power distribution network operation and maintenance, and the modules work together to complete the collaborative tasks of refined inspection of power distribution network and live-line work.

2. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The non-vector space control strategy of the UAV inspection subsystem adopts the core idea of ​​direct tracking without feature localization. The distance scalar is calculated using the Hausdorff distance, and its calculation formula is as follows: ； Where dH(X,Y) is the Hausdorff distance scalar, which is the state deviation between the acquired image and the pre-stored image; X is the set of pixels in the real-time acquired image; Y is the set of pixels in the pre-stored standard image; x is any pixel in set X; y is any pixel in set Y; d(x,y) is the Euclidean distance between pixel x and pixel y; sup represents the supremum, which is the maximum value in the set; inf represents the infrem, which is the minimum value in the set.

3. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 2, characterized in that: The dynamic tubular domain model of the conductor is dynamically adjusted based on the real-time position, diameter, and running attitude of the conductor. The radius of the tubular domain is 3-5 times the diameter of the conductor. An adaptive mesh generation method is adopted, and the mesh size dynamically changes with the distance between the conductor and the UAV. The closer the distance, the finer the mesh generation, ensuring tracking accuracy.

4. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The multi-sensor fusion defect identification module integrates a high-definition visible light camera, an infrared thermal imaging camera, a lidar, and a partial discharge sensor, and is equipped with the TINY AI visual navigation system to achieve accurate identification of more than nine types of power distribution network equipment. Defect identification employs a combination of image frame splitting processing, feature point matching, and temperature anomaly analysis. The feature point matching calculation formula is as follows: ； Where Sij is the matching similarity between the i-th acquired image feature point and the j-th standard defect feature point; n is the number of dimensions of the feature point; xik is the k-th dimensional feature value of the i-th acquired image feature point; and xjk is the k-th dimensional feature value of the j-th standard defect feature point. The average value of all dimensions of the feature points of the i-th acquired image feature point; The average of all dimensional eigenvalues ​​for the j-th standard defect feature point; when When the match is successful, it is identified as a defect of the corresponding type.

5. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 4, characterized in that: The multi-sensor fusion defect identification module automatically identifies more than 12 types of minor defects. The generated structured defect report includes the defect location, type, severity, discovery time, line and equipment number information, and simultaneously marks the defect level, providing accurate data support for live-line work and adapting to scenarios with frequent changes in distribution network lines.

6. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The task planning module constructs a unified cognitive model of the distribution network that integrates a fault propagation knowledge graph. It then generates the optimal inspection and operation task planning scheme through a heuristic optimization algorithm. The objective function of the heuristic optimization algorithm is: ； Where F is the objective function value of the task planning, and the smaller the value, the better the planning scheme; These are the weighting coefficients for inspection route length, drone energy consumption, and operation time, respectively. L represents the total route length for UAV inspection and operation, in km; E represents the total energy consumption of the UAV in completing the task, in kWh; T represents the total time to complete the inspection and operation task, in h; Meanwhile, the algorithm constraints are as follows: , This is the maximum energy consumption for drone operation, preventing drones from returning to base prematurely due to insufficient power.

7. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 6, characterized in that: The task planning module automatically increases the frequency of patrols in special areas such as wildfire risk zones and areas prone to external damage. The frequency of patrols after encryption is 2-3 times that of regular areas, thus tilting resources towards high-risk scenarios. At the same time, it supports collaborative operations of multiple drones, allocating inspection and operation resources according to the endurance, load status and location of each drone.

8. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The safety control module incorporates an electric field strength sensor and a proximity induction array. Based on voltage level and real-time electric field gradient, it dynamically calculates the minimum safe operating distance and employs a fuzzy PID control algorithm to achieve distance regulation. The calculation formula for the output parameters of the fuzzy PID controller is as follows: 、 、 ； in, These are the proportional coefficient, integral coefficient, and derivative coefficient of the fuzzy PID controller, respectively. These are the initial proportional coefficient, initial integral coefficient, and initial derivative coefficient of the PID controller, respectively. These are the proportional coefficient correction, integral coefficient correction, and differential coefficient correction obtained from fuzzy logic reasoning, respectively; when a sudden change in electric field or a decrease in insulation resistance is detected, emergency back-off and host traction protection actions are automatically triggered.

9. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The closed-loop iterative module compares the differences in equipment status before and after the operation using a semantic segmentation algorithm. The semantic segmentation similarity calculation formula is as follows: ； Where IoU is the intersection-union ratio of the equipment images before and after the operation, i.e., the state similarity; A is the semantic segmentation region of the equipment image after the operation; B is the semantic segmentation region of the equipment image before the operation. This is the intersection region of A and B, i.e., the region where the state remains unchanged; The region is the union of A and B, i.e., the overall equipment region. When IoU < 0.9, it is determined that the post-operation state has changed effectively, and an operation quality assessment report is automatically generated and a local incremental update of the digital twin model is triggered to achieve a closed-loop iteration of defect identification, operation execution, effect verification, and model correction.

10. The collaborative UAV system for refined inspection and live-line work of power distribution networks according to claim 1, characterized in that: The digital twin visualization and control platform constructs a high-precision digital twin model of power distribution network equipment based on LiDAR point clouds and inspection images. It integrates five major functions: inspection data management, defect diagnosis and analysis, task scheduling, work process monitoring, and maintenance report generation. It supports collaborative operation control of multiple drones. Combined with AI algorithms, it achieves health status assessment, fault prediction, and maintenance strategy optimization for power distribution network equipment. The health status assessment value is: ， Among them, H1 is the equipment operating parameter compliance rate, H2 is the defect rectification completion rate, and H3 is the historical operating reliability score. The value of H ranges from 0 to 1, and the larger the value, the better the health status of the equipment.

Images