Unmanned aerial vehicle group autonomous investigation and decision-making method and system for uninterruptible operation of distribution network

Abstract

The invention discloses an unmanned aerial vehicle group autonomous investigation and decision-making method and system for distribution network uninterruptible operation, and mainly relates to the field of distribution network uninterruptible operation. The whole process from task issuing to report generation does not need personnel to intervene in a field investigation link; through multi-sensor fusion acquisition and AI processing, an objective and quantitative data basis is provided for safety analysis; a physical site is quickly and automatically converted into interactive and analyzable digital twin bodies, so that the system has the capability of dealing with environmental changes, and can learn and optimize the own investigation strategy through historical data, thereby realizing the spanning from automation to intelligence. The method has the beneficial effects that the technical problems that the flexibility is insufficient and the subsequent data processing amount is relatively large during investigation of the existing unmanned aerial vehicle can be solved, the flexibility of a route in the investigation process is improved, and the difficulty and the workload of subsequent data processing are reduced.

Description

Technical Field

[0001] This invention relates to the field of live-line work in power distribution networks, specifically to a method and system for autonomous surveying and decision-making by unmanned aerial vehicle (UAV) swarms in live-line work in power distribution networks. Background Technology

[0002] With the expansion of power distribution network scale and the increasing requirements for power supply reliability, the traditional manual on-site survey mode can no longer meet the needs of the development of live-line work. Its inherent defects are mainly reflected in: Inefficiency and heavy burden on grassroots units. Surveyors must travel long distances to each work site, which is time-consuming and laborious. Especially in areas with multiple work sites or complex geographical environments, efficiency is extremely low, resulting in insufficient capacity of grassroots units.

[0003] The process is highly subjective, making accuracy difficult to guarantee. Determining safe distances and assessing equipment status heavily relies on the experience and visual observation of surveyors, lacking objective and quantitative data support. This can easily lead to misjudgments and create potential safety hazards.

[0004] Incomplete information records make it difficult to trace and analyze the incident. Traditional methods of recording information through photos and notes result in fragmented and unstructured information that is difficult to accurately and completely reconstruct the full picture of the scene, which is not conducive to subsequent solution discussions, accident retrospective analysis, and long-term trend analysis.

[0005] The current level of intelligence in drone surveying technology is insufficient. Although there have been attempts to use drones for surveying, most still require on-site operation by pilots, or can only execute simple preset flight paths, and cannot cope with dynamic changes on site (such as temporary obstacles). Furthermore, the collected data requires a large amount of manual processing in the later stages, failing to form a closed loop of "collection-analysis-decision". Summary of the Invention

[0006] The purpose of this invention is to provide an autonomous survey and decision-making method and system for unmanned aerial vehicle (UAV) swarms in power distribution network uninterrupted operation. It can solve the technical problems of insufficient flexibility and large amount of subsequent data processing in existing UAV surveys, improve the flexibility of routes during the survey process, process and update data in real time, and reduce the difficulty and workload of subsequent data processing.

[0007] To achieve the above objectives, the present invention employs the following technical solution: The autonomous survey and decision-making method for unmanned aerial vehicle (UAV) swarms for live-line power distribution work includes the following steps: S1, the central dispatch platform receives the reconnaissance mission and assigns it to the target UAV nest according to the mission location information; S2, the drone swarm within the target drone nest adopts a hybrid collaborative decision-making method to determine the final collaborative scheme for executing the reconnaissance task; S3 generates an initial twin copy of the work site in the digital twin platform based on historical data and plans the drone baseline reconnaissance route; S4, the UAV performs autonomous flight reconnaissance according to the benchmark reconnaissance route, and collects on-site data and senses environmental changes in real time during the flight; S5, perform virtual-real interaction and dynamic matching between real-time perception data and the initial twin copy; if a significant environmental difference is detected, trigger local obstacle avoidance and global mission update of the UAV. S6: The drone returns to base and transmits all data back. The platform then builds a high-precision digital twin model of the work site based on the transmitted data. S7. In the digital twin model, a safety rule engine is embedded to conduct a spatiotemporal four-dimensional dynamic safety risk assessment oriented towards the operation process. S8 automatically generates a structured survey report containing a list of risk points and a safe operation plan based on the risk assessment results; S9 aggregates and updates the global AI model based on the local AI model training results of each drone nest through federated learning.

[0008] Furthermore, the hybrid collaborative decision-making method in step S2 includes: S21, the central dispatch platform performs centralized planning at the task level, dividing macro task packages and assigning them to one or more UAV nests; S22, within the assigned drone swarm, a distributed negotiation mechanism is used to autonomously elect a leader drone. S23, the lead drone presides over a voting or bidding process based on a consensus mechanism to determine the sub-tasks and collaborative routes to be executed by each drone in the fleet.

[0009] Furthermore, the virtual-real interaction and dynamic matching in step S5 include: S51, the drone constructs a real-time perception twin of the local environment in flight based on SLAM technology; S52, perform rapid matching and difference detection between the real-time sensing twin and the initial twin copy; S53: If a significant difference is detected, the drone will first autonomously perform local path replanning to complete real-time obstacle avoidance. S54 simultaneously sends the difference information and local models back to the central platform, which dynamically updates the global digital twin model and replans routes for the collaborative fleet.

[0010] Furthermore, the spatiotemporal four-dimensional dynamic security risk assessment in step S7 includes: S71, Import the 3D model of the work equipment into the digital twin model; S72, simulating the complete motion process of the work implement when performing the work task; S73, with time as the fourth dimension, continuously samples and calculates the instantaneous electrical distance between the moving parts and the live equipment along the entire working trajectory; S74, based on a preset safety rule engine, dynamically determines whether there is any instantaneous distance below the safety threshold at any moment and identifies crossing risk points.

[0011] Furthermore, the safety rule engine supports dynamic risk thresholds, which are functions related to the movement speed of the operating equipment, specifically expressed as: minimum safe distance = basic safe distance + k × movement speed, where k is a weighting coefficient.

[0012] Furthermore, step S8 is followed by step S10: converting the generated safety operation plan into augmented reality (AR) visual instructions and sending them to the AR terminals of on-site workers to overlay and display virtual safety boundaries and operation guidance information in the real work field of vision.

[0013] Furthermore, the federated learning method in step S9 specifically includes: S91: Each drone nest uses locally stored survey data to train the local AI vision model and obtain local model parameters. S92: The central platform collects local model parameters from each drone nest and aggregates them through a federated averaging algorithm to generate an optimized global AI model. S93, the global AI model is distributed to each drone nest to achieve continuous collaborative evolution of the drone swarm's AI capabilities.

[0014] The autonomous survey and decision-making system for unmanned aerial vehicle (UAV) swarms used for live-line power distribution work includes: The central dispatch platform is used for task management, global collaborative decision-making, digital twin modeling, and deep analysis. At least one drone nest for storing, charging, and maintaining drones; A drone swarm consists of at least two drones equipped with multiple sensors and AI processing modules, used to perform autonomous reconnaissance missions. The network communication module is used to establish a reliable data transmission link between the central dispatch platform, the UAV nest, and the UAV swarm.

[0015] Furthermore, the central dispatch platform specifically includes: The task management unit is used to receive, parse, and distribute exploration tasks; A digital twin engine for building, rendering, and updating high-precision 3D models of the work site; The rules engine, with embedded power safety procedures, is used to perform dynamic safety risk assessments. Federated learning servers are used to aggregate model parameters and distribute global AI models.

[0016] Furthermore, the drone nest also includes a self-diagnostic unit for monitoring the battery health, propeller wear, and sensor performance of the drones inside the nest, and for providing predictive maintenance reminders.

[0017] Furthermore, the AI ​​processing module on the drone preloads the power grid GIS / CIM data of the target area before flight, which is used to achieve semantic-level real-time recognition and refined survey guidance based on prior knowledge.

[0018] Furthermore, it also includes an augmented reality (AR) terminal, used to receive and display AR visual instructions issued by the central dispatch platform, and to overlay and fuse the safety operation plan in the digital twin model with the real physical site.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Improve efficiency: Enable "one-click survey", shortening the original manual survey of several hours to a fully automated operation of ten or tens of minutes, greatly relieving the pressure on the grassroots level.

[0020] 2. Ensuring safety: By leveraging AI vision, digital twins, and rule engines, safety judgments are shifted from experience-driven to data-driven, enabling the early detection and elimination of risks in the virtual world and preventing accidents at their source.

[0021] 3. Scientific and Visualized Decision-Making: Provides intuitive 3D scenes and quantitative analysis data, making the formulation and review of work plans based on evidence; AR technology accurately empowers on-site operations with digital decision-making, reducing human error.

[0022] 4. System Adaptability and Intelligence: Technologies such as collaborative task allocation, dynamic path planning, and equipment health management enable the system to have the ability to self-optimize and self-maintain, and its level of intelligence is far higher than that of simple automated equipment.

[0023] 5. High value of data assets: The resulting structured database and digital twin model are core assets for building a digital power grid, and can be used for more value-added applications such as planning and design, condition assessment, and disaster simulation. Attached Figure Description

[0024] Appendix Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0025] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0026] Reference Figure 1 The present invention describes an autonomous survey and decision-making method and system for unmanned aerial vehicle (UAV) swarms used in power distribution network live-line work. The autonomous survey and decision-making method for unmanned aerial vehicle (UAV) swarms used in power distribution network live-line work includes the following steps: S1, the central dispatch platform receives the reconnaissance mission and assigns it to the target UAV nest according to the mission location information; S2, the drone swarm within the target drone nest adopts a hybrid collaborative decision-making method to determine the final collaborative scheme for executing the reconnaissance task; S3 generates an initial twin copy of the work site in the digital twin platform based on historical data and plans the drone baseline reconnaissance route; S4, the UAV performs autonomous flight reconnaissance according to the benchmark reconnaissance route, and collects on-site data and senses environmental changes in real time during the flight; S5, perform virtual-real interaction and dynamic matching between real-time perception data and the initial twin copy; if a significant environmental difference is detected, trigger local obstacle avoidance and global mission update of the UAV. S6: The drone returns to base and transmits all data back. The platform then builds a high-precision digital twin model of the work site based on the transmitted data. S7. In the digital twin model, a safety rule engine is embedded to conduct a spatiotemporal four-dimensional dynamic safety risk assessment oriented towards the operation process. S8 automatically generates a structured survey report containing a list of risk points and a safe operation plan based on the risk assessment results; S9 aggregates and updates the global AI model based on the local AI model training results of each drone nest through federated learning.

[0027] Preferably, the hybrid collaborative decision-making method in step S2 includes: S21, the central dispatch platform performs centralized planning at the task level, dividing macro task packages and assigning them to one or more UAV nests; S22, within the assigned drone swarm, a distributed negotiation mechanism is used to autonomously elect a leader drone. S23, the lead drone presides over a voting or bidding process based on a consensus mechanism to determine the sub-tasks and collaborative routes to be executed by each drone in the fleet.

[0028] Preferably, the virtual-real interaction and dynamic matching in step S5 include: S51, the drone constructs a real-time perception twin of the local environment in flight based on SLAM technology; S52, perform rapid matching and difference detection between the real-time sensing twin and the initial twin copy; S53: If a significant difference is detected, the drone will first autonomously perform local path replanning to complete real-time obstacle avoidance. S54 simultaneously sends the difference information and local models back to the central platform, which dynamically updates the global digital twin model and replans routes for the collaborative fleet.

[0029] Preferably, the spatiotemporal four-dimensional dynamic security risk assessment in step S7 includes: S71, Import the 3D model of the work equipment into the digital twin model; S72, simulating the complete motion process of the work implement when performing the work task; S73, with time as the fourth dimension, continuously samples and calculates the instantaneous electrical distance between the moving parts and the live equipment along the entire working trajectory; S74, based on a preset safety rule engine, dynamically determines whether there is any instantaneous distance below the safety threshold at any moment and identifies crossing risk points.

[0030] Preferably, the safety rule engine supports dynamic risk thresholds, which are functions related to the movement speed of the operating equipment, specifically expressed as: minimum safe distance = basic safe distance + k × movement speed, where k is a weighting coefficient.

[0031] Preferably, step S8 is followed by step S10: converting the generated safety operation plan into augmented reality (AR) visual instructions and sending them to the AR terminals of on-site workers to overlay and display virtual safety boundaries and operation guidance information in the real work field of view.

[0032] Preferably, the federated learning method in step S9 specifically includes: S91: Each drone nest uses locally stored survey data to train the local AI vision model and obtain local model parameters. S92: The central platform collects local model parameters from each drone nest and aggregates them through a federated averaging algorithm to generate an optimized global AI model. S93, the global AI model is distributed to each drone nest to achieve continuous collaborative evolution of the drone swarm's AI capabilities.

[0033] The autonomous survey and decision-making system for unmanned aerial vehicle (UAV) swarms used for live-line power distribution work includes: The central dispatch platform is used for task management, global collaborative decision-making, digital twin modeling, and deep analysis. At least one drone nest for storing, charging, and maintaining drones; A drone swarm consists of at least two drones equipped with multiple sensors and AI processing modules, used to perform autonomous reconnaissance missions. The network communication module is used to establish a reliable data transmission link between the central dispatch platform, the UAV nest, and the UAV swarm.

[0034] Preferably, the central dispatch platform specifically includes: The task management unit is used to receive, parse, and distribute exploration tasks; A digital twin engine for building, rendering, and updating high-precision 3D models of the work site; The rules engine, with embedded power safety procedures, is used to perform dynamic safety risk assessments. Federated learning servers are used to aggregate model parameters and distribute global AI models.

[0035] Preferably, the drone nest also includes a self-diagnostic unit for monitoring the battery health, propeller wear, and sensor performance of the drones inside the nest, and for providing predictive maintenance reminders.

[0036] Preferably, the AI ​​processing module on the drone preloads the power grid GIS / CIM data of the target area before flight, so as to realize semantic-level real-time recognition and refined survey guidance based on prior knowledge.

[0037] Preferably, it also includes an augmented reality (AR) terminal, used to receive and display AR visual instructions issued by the central dispatch platform, and to overlay and fuse the safety operation plan in the digital twin model with the real physical site.

[0038] The core architecture of this system consists of a perception and execution layer (drone swarm and nest), a network transmission layer, a platform processing layer (digital twin and AI analysis), and an application decision layer. Its innovative technologies are detailed below: 1. Collaborative exploration and task allocation technology based on swarm intelligence The system combines centralized dispatch with autonomous negotiation: a central dispatch unit is established to receive reconnaissance tasks across the entire region. For regionalized multi-tasking, the dispatch unit does not simply assign tasks, but broadcasts the task packages to multiple UAV nests in the relevant areas.

[0039] Task allocation based on a market auction mechanism: Drones within each drone cluster act as intelligent agents, bidding to the central scheduling unit based on their location, battery power, and payload capacity to calculate the "cost" of completing their sub-tasks. The scheduling unit allocates tasks based on global optimization principles (such as the shortest total time and lowest total energy consumption), achieving efficient group collaboration.

[0040] Dynamic route reconfiguration: During collaborative missions, if a drone needs to return to base early due to unforeseen circumstances (such as strong winds), it can share its unfinished reconnaissance sub-tasks and collected data with other nearby drones in real time, allowing the latter to take over and complete the tasks, thus achieving dynamic and seamless handover of missions.

[0041] 2. AI visual real-time analysis and guidance technology that deeply integrates prior knowledge Visual enhancement based on power grid GIS / CIM model: The drone's AI vision module is preloaded with power grid geographic information system (GIS) and public information model (CIM) data of the target area before takeoff, and the approximate location of the towers, equipment type and voltage level are known.

[0042] Semantic-level real-time recognition and guidance: During flight, AI not only identifies obstacles but also performs semantic recognition, such as: "10kV drain line identified, close-up of the clamp needs to be captured." Subsequently, AI automatically generates a detailed local flight path, guiding the drone to capture the designated component at the optimal angle and distance, ensuring the purposefulness and high quality of data collection.

[0043] Immediate reporting of abnormal conditions: If the AI ​​identifies obvious equipment defects during the inspection process (such as broken insulators or severe corrosion of fittings), it can immediately trigger a high-level alarm and transmit close-up images and location information back in real time to remind monitoring personnel to pay priority attention.

[0044] 3. Digital twin deep analysis technology with physics engine and rule engine Embedded automated safety verification rules: The digital twin platform not only incorporates general power safety regulations, but also allows for customized safety rule engines based on the specific operational requirements of local companies. For example, a rule could be defined as: "For work near 110kV lines, the minimum air gap between the boom and live parts shall not be less than 1.5 meters."

[0045] Operation simulation with integrated physics engine: The platform integrates a lightweight physics engine. When users simulate placing large machinery such as insulated bucket trucks and cranes in the twin, the system can simulate the vehicle's supporting, extending, and rotating actions, and calculate the dynamic distance changes between its motion envelope and surrounding live equipment in real time, providing early warnings for possible collisions or insufficient safety distances.

[0046] Multiple solution comparison and optimization suggestions: The system supports saving multiple simulated operation solutions and can perform quantitative comparisons from multiple dimensions such as "safety", "operability" and "efficiency" to help the operation manager select the optimal solution.

[0047] 4. Automatic generation of survey reports based on digital twins and AR on-site augmentation technology Automatic generation of structured reports: Based on the analysis results, the system automatically generates a structured survey report that includes key screenshots of the 3D model, a list of risk point annotations, a safety distance verification result table, a summary of the recommended work plan, and links to simulation videos. The report can be directly used for work permit approval.

[0048] AR-assisted on-site operations: The generated work plan can be exported to AR (Augmented Reality) glasses. After wearing AR glasses, on-site workers can overlay virtual guidance information derived from the digital twin model onto their real field of vision, such as the ideal swing path of the crane boom and the safety boundary lines that need to be maintained, achieving seamless integration between the digital plan and the physical site, and improving work accuracy and safety.

[0049] Example: A municipal power supply company needs to conduct a site survey before carrying out live-line replacement of pole-mounted switches on poles #15 to #18 of the 10kV Cuihua line. The specific work steps are as follows: Task assignment: The person in charge of the work submits the survey task through the PC and delineates the survey area (towers #15-#18).

[0050] Intelligent Scheduling: The central scheduling system determines that "Nest A" is responsible for the task based on its location. Nest A contains two UAVs (UAV-1 and UAV-2). The system uses an auction algorithm to assign UAV-1 to be responsible for the global scanning and detailed reconnaissance of poles #15-#16, and UAV-2 to be responsible for poles #17-#18.

[0051] Autonomous and collaborative exploration: The two drones took off autonomously in turn. When UAV-1 flew to pole #15, the AI ​​detected a switch on the pole and automatically triggered a fine scan to take pictures of the switch body, leads, and adjacent equipment.

[0052] During the journey, UAV-2 detected a temporary construction crane near pole #17 and used its AI vision to dynamically replan its route, safely detouring around it.

[0053] Twin Reconstruction and Deep Analysis: After the data is transmitted back, the platform generates a digital twin model of the section with centimeter-level precision within 30 minutes.

[0054] The system automatically invoked the rules engine to simulate placing the insulated boom truck model at a predetermined position next to pole #16 and simulating the boom raising to the switch. The physics engine calculations revealed that when the boom rotated to a certain angle, the distance between it and the adjacent #15 pole conductor was only 0.9 meters, below the safety regulation requirement of 1.0 meter. The system highlighted and flashed this risk point in the model as a warning.

[0055] Solution generation and AR assistance: The system generated a report that clearly pointed out the risks and recommended that the location of the boom truck be slightly adjusted 2 meters eastward. The report included a simulation video.

[0056] On the day of the operation, on-site personnel used AR glasses to see the virtual markers of the best parking spaces recommended by the system and the safe movement envelope of the bucket arm. They strictly followed the visual guidance to complete the operation, effectively avoiding safety risks.

Claims

1. An unmanned aerial vehicle group autonomous surveying and decision-making method for power distribution network non-power-off operation, characterized in that, The method comprises the following steps: S1, a central scheduling platform receives a survey task and distributes it to a target UAV nest according to task location information; S2, a UAV swarm in the target UAV nest determines a final coordination scheme for executing the survey task by using a hybrid coordination decision method; S3, an initial twin copy of a work site is generated in a digital twin platform based on historical data, and a reference survey route for a UAV is planned; S4, the UAV performs autonomous flight survey according to the reference survey route, and collects real-time site data and senses environmental changes during flight; S5, real-time sensing data is interacted with and dynamically matched with the initial twin copy, and if a major environmental difference is detected, the UAV triggers local obstacle avoidance and global task update; S6, the UAV returns and returns all data, and the platform constructs a high-precision digital twin model of the work site based on the returned data; S7, in the digital twin model, a safety rule engine is embedded to perform four-dimensional dynamic safety risk assessment in time and space for the work process; S8, based on the risk assessment result, a structured survey report containing a risk point list and a safe work scheme is automatically generated; S9, based on the local AI model training result of each UAV nest, the global AI model is aggregated and updated through federated learning.

2. The method of claim 1, wherein, The hybrid coordination decision method in step S2 comprises: S21, the central scheduling platform performs centralized planning at the task level, divides and assigns a macro task package to one or more UAV nests; S22, within the assigned UAV swarm, a lead UAV is autonomously elected by the swarm through a distributed negotiation mechanism; S23, the lead UAV hosts a voting or bidding process based on a consensus mechanism to determine the sub-tasks and coordination routes for each UAV in the swarm.

3. The method of claim 1, wherein, The virtual-real interaction and dynamic matching in step S5 comprises: S51, the UAV constructs a real-time sensing twin of the local environment in flight based on SLAM technology; S52, the real-time sensing twin is quickly matched with the initial twin copy for difference detection; S53, if a major difference is detected, the UAV first autonomously re-plans the local path to complete real-time obstacle avoidance; S54, at the same time, the difference information and local model are sent back to the central platform, which dynamically updates the global digital twin model and re-plans the route for the coordination swarm.

4. The method of claim 1, wherein, The four-dimensional dynamic safety risk assessment in time and space in step S7 comprises: S71, a three-dimensional model of the work machine is imported into the digital twin model; S72, the complete motion process of the work machine during task execution is simulated; S73, the instantaneous electrical distance between the moving parts and the live equipment is continuously sampled and calculated along the entire work trajectory with time as the fourth dimension; S74, according to the preset safety rule engine, it is dynamically judged whether there is any instantaneous distance below the safety threshold at any time, and the crossing risk points are identified.

5. The method of claim 4, wherein the method further comprises: The safety rule engine supports a dynamic risk threshold, which is a function related to the motion speed of the work machine, specifically represented as: minimum safety distance = basic safety distance + k × motion speed, where k is the weight coefficient.

6. The method of claim 1, wherein the method further comprises: The step S8 is followed by a step S10 of converting the generated safety operation scheme into augmented reality (AR) visual instructions and issuing the AR visual instructions to an AR terminal of an on-site operator for superimposed display of a virtual safety boundary and operation guidance information in a real operation field of view.

7. The method of claim 1, wherein, The federated learning manner in step S9 specifically includes: S91, each drone nest trains a local AI vision model using locally stored survey data to obtain local model parameters; S92, the central platform collects local model parameters from each drone nest and aggregates them to generate an optimized global AI model through a federated averaging algorithm; S93, the global AI model is distributed to each drone nest to achieve continuous co-evolution of swarm AI capabilities.

8. An unmanned aerial vehicle group autonomous surveying and decision-making system for power distribution network non-power-off operation, characterized in that, Comprise: a central scheduling platform for task management, global collaborative decision-making, digital twin modeling, and deep analysis; at least one drone nest for storing, charging, and maintaining drones; a drone swarm composed of at least two drones equipped with various sensors and AI processing modules for performing autonomous survey tasks; a network communication module for establishing a reliable data transmission link between the central scheduling platform, the drone nest, and the drone swarm.

9. The autonomous surveying and decision-making system for the unmanned aerial vehicle group for power grid non-stop operation according to claim 8, characterized in that, The central scheduling platform specifically includes: a task management unit for receiving, analyzing, and distributing survey tasks; a digital twin engine for building, rendering, and updating high-precision three-dimensional models of the work site; a rule engine embedded with power safety regulations for performing dynamic safety risk assessment; a federated learning server for aggregating model parameters and distributing global AI models.

10. The autonomous surveying and decision-making system for the unmanned aerial vehicle group for power grid non-stop operation according to claim 8, characterized in that: The drone nest also includes a self-diagnosis unit for monitoring the battery health status, propeller wear, and sensor performance of the drones in the nest and providing predictive maintenance reminders.

11. The autonomous surveying and decision-making system for the unmanned aerial vehicle group for power grid non-stop operation according to claim 8, characterized in that: The AI processing module carried by the drone preloads the target area's power grid GIS / CIM data before flight to enable semantic-level real-time recognition and refined survey guidance based on prior knowledge.

12. The autonomous surveying and decision-making system for the unmanned aerial vehicle group for power grid non-stop operation according to claim 8, characterized in that: It also includes an augmented reality (AR) terminal for receiving and displaying AR visual instructions issued by the central scheduling platform, superimposing and fusing the safety operation scheme in the digital twin model with the real physical site.

Images