Internet-of-things unmanned aerial vehicle intelligent AI electric power line defect identification system

Abstract

According to the Internet of Things unmanned aerial vehicle intelligent AI electric power line defect identification system provided by the invention, dispersed unmanned aerial vehicles, machine nests, personnel, processes and data are integrated into a whole through a unified machine nest management platform, and a modern power grid inspection operation and maintenance new system integrating data acquisition, intelligent analysis, business management and decision support is successfully constructed; the safety, the efficiency and the intelligent level of power grid operation are obviously improved; the system comprises an unmanned aerial vehicle, a nest and a nest management platform, the nest management platform and the nest both adopt an MQTT protocol to carry out instruction control and state updating on the unmanned aerial vehicle, and the nest management platform is in wireless communication with the nest; the nest management platform comprises a task management unit, an unmanned aerial vehicle inspection monitoring unit, a defect management unit, a data center unit and a personnel and authority management unit.

Description

Technical Field

[0001] This invention relates to the field of power line defect identification technology, specifically to an IoT-enabled drone-based intelligent AI power line defect identification system. Background Technology

[0002] With the iteration of drone technology and the increasing maturity of AI technology, the power industry has begun to explore the use of drones for inspections to address the challenges of harsh external conditions, severe tree obstruction, time-consuming and labor-intensive maintenance of traditional power distribution lines, and the inability to directly assess potential defects. This has alleviated the pressure of manual inspections to some extent. However, existing drone inspection solutions still have significant limitations:

[0003] 1. Low level of automation: Most applications still rely on the "manual operation of drones, manual shooting, and manual analysis" model. Pilots need to be on-site, and operational efficiency is limited by the pilot's skills and physical strength, making large-scale, high-frequency automated inspections impossible.

[0004] 2. Disconnect between "drone nest" and "platform": Although automatic drone nests (charging compartments) have emerged to enable automatic take-off, landing and charging of drones, the functions of the nest management platform are often relatively simple, limited to basic flight control, status monitoring and data retrieval, lacking an integrated, full-process visualized operation platform for operation and maintenance business.

[0005] 3. Defect identification relies on manual labor, resulting in a significant efficiency bottleneck: After the inspection, professionals need to spend a lot of time viewing and sifting through a massive amount of images and videos to visually identify defects. This process is tedious, time-consuming, and prone to missed or misjudged defects due to fatigue, becoming a bottleneck in the overall inspection process.

[0006] 4. Severe data silos: Flight mission data, raw inspection data, defect reports, and processing work orders are scattered across different systems or files, without forming a unified data center. This hinders information flow between inspection, diagnosis, and maintenance stages, making it difficult to achieve closed-loop management of defects throughout the entire process of "discovery-reporting-review-processing-verification".

[0007] Against this backdrop, we developed this "IoT-enabled UAV Intelligent AI Power Line Defect Identification System." Its core objective is not merely to implement single-point technology applications, but to build a unified platform suitable for remote, visual operation by maintenance personnel. This platform integrates data collection, intelligent analysis, business management, and decision support across the entire process, ultimately creating a digital, visual, and intelligent modern power grid inspection and maintenance system. Summary of the Invention

[0008] This invention aims to address the technical deficiencies of existing technologies by providing an IoT-based intelligent AI power line defect identification system using drones. This system integrates dispersed drones, drone nests, personnel, processes, and data into a unified drone nest management platform, successfully constructing a modern power grid inspection and maintenance system that integrates "data collection, intelligent analysis, business management, and decision support," significantly improving the safety, efficiency, and intelligence level of power grid operation.

[0009] This invention discloses an IoT-enabled intelligent AI power line defect identification system for drones, comprising a drone, a drone housing, and a housing management platform. Both the housing management platform and the drone housing use the MQTT protocol for command control and status updates of the drone, and the housing management platform communicates wirelessly with the drone housing. The housing management platform includes a task management unit, a drone inspection and monitoring unit, a defect management unit, a data center unit, and a personnel and access control unit.

[0010] The task management unit is used for intelligent scheduling throughout the entire task lifecycle;

[0011] The drone inspection and monitoring unit is used for real-time status perception and anomaly intervention;

[0012] The defect management unit is used for defect identification, classification, and closed-loop processing.

[0013] Data center units are used for data storage, analysis, and mining;

[0014] The personnel and access control unit is used for identity authentication and access control.

[0015] Furthermore,

[0016] The task management unit includes a task generation and parameter parsing module, a task allocation and path optimization module, a task execution monitoring and dynamic adjustment module, and a task acceptance and archiving module.

[0017] The task generation and parameter parsing module takes into account the inspection area, task type, time window, and equipment type set by the user. It extracts the equipment ledger within the target area from the database using a GIS coordinate-based regional boundary matching algorithm. This ledger includes tower locations, line routes, and historical defect points. Combined with the drone's endurance parameters and nest deployment location, it calculates the minimum inspection unit.

[0018] The task allocation and path optimization module initializes and generates a random path set covering all target towers, starting from the tower nest, and then uses a fitness function and iterative optimization to perform task allocation and path optimization.

[0019] The task execution monitoring and dynamic adjustment module uses a threshold-triggered adjustment algorithm based on real-time input of the drone's location, remaining battery power, number of completed poles, and sudden weather events: if the remaining battery power is less than the battery power required for return based on the real-time location, an "emergency return" is automatically triggered, and incomplete tasks are marked as "pending inspection"; if no-fly weather occurs, the task is suspended and enters a "waiting queue", and is reordered according to priority after the weather recovers.

[0020] The task acceptance and archiving module uses the inspection data and path deviation rate uploaded by the UAV to check whether there is at least one clear photo of the key parts of each tower. The clarity of the clear photo is judged by the image entropy value. An entropy value > 50 is considered qualified. If the path deviation rate is < 5% and the data is complete, it is marked as "task completed"; otherwise, it is marked as "re-inspection required" and a supplementary inspection task is automatically generated.

[0021] Furthermore,

[0022] The UAV inspection and monitoring unit includes a real-time data acquisition and analysis module, a flight status monitoring and risk warning module, and an abnormal intervention and control command generation module.

[0023] The real-time data acquisition and analysis module takes the status data transmitted back by the UAV every second as input. The status data includes GPS coordinates, altitude, speed, battery level, attitude angle, and camera parameters. It uses data cleaning and standardization algorithms to remove outliers, converts the attitude angle into three-dimensional spatial coordinates, and converts the battery level into remaining flight time. The remaining flight time is based on the energy consumption model of the current speed and load: time = remaining battery level / average current.

[0024] The flight status monitoring and risk warning module uses a multi-dimensional threshold comparison algorithm to provide risk warnings.

[0025] If the warning level is "emergency", the abnormal intervention and control command generation module will automatically generate control commands through the emergency return-to-home algorithm and the hovering waiting algorithm. The emergency return-to-home algorithm calculates the shortest return-to-home path and sends the "one-click return-to-home" command to the UAV through encrypted transmission via the MQTT protocol. The hovering waiting algorithm sends a "hover" command if it is temporarily unable to return to home, and at the same time notifies the nest to prepare for emergency reception.

[0026] Furthermore,

[0027] The defect management unit includes a defect identification and location module, a defect classification and risk assessment module, a defect dispatch and progress tracking module, and a defect re-inspection and closed-loop verification module.

[0028] The defect identification and localization module takes equipment images or videos captured by a drone as input, and performs preprocessing, model inference, and spatial localization based on deep learning-based target detection and segmentation algorithms. The preprocessing includes image dehazing, image stabilization, and cropping. The model inference uses an improved YOLOv8 model to output the defect category, location coordinates, and confidence score. A confidence score > 0.8 is considered a valid identification. Spatial localization combines the GPS coordinates and camera angles taken by the drone to map the image defects to the actual equipment location.

[0029] The defect classification and risk assessment module classifies defects by inputting defect category, equipment importance, and environmental factors, and uses a weighted scoring classification algorithm.

[0030] The defect dispatching and progress tracking module uses an intelligent dispatching algorithm to dispatch defects and uses a state transition algorithm to record defects whose status has expired and have not been processed, and automatically escalates the reminders.

[0031] The defect re-inspection and closed-loop verification module inputs processed on-site photos and UAV re-inspection images, and uses a defect elimination verification algorithm to compare the image features before and after processing. If the defect features disappear and the re-inspection confidence level is <0.2, it is determined that "the defect has been eliminated" and the closed-loop is archived; otherwise, it is marked as "processing invalid" and a new order is dispatched.

[0032] Furthermore,

[0033] The data center unit includes a data access and standardized storage module, a historical data trend analysis module, and a data value mining and application module.

[0034] The data access and standardized storage module takes in multi-source data, including UAV images or videos, sensor temperature and humidity data, mission logs, and defect records. It uses data classification and indexing algorithms to build a three-level index based on "time + device ID + data type", links it to a distributed database, performs fragmentation processing on unstructured data, and extracts key frames for separate marking.

[0035] The historical data trend analysis module, by inputting historical defect records and environmental data of a certain device, uses an LSTM neural network to learn the correlation between defect occurrence and environmental factors, predicts the defect risk value for the next 6 months, and outputs a "high / medium / low" risk level.

[0036] The data value mining and application module uses an association rule mining algorithm to analyze the correlation of defects in multiple devices. If the confidence level is greater than 70%, the association rule is output. The data visualization uses a heat map generation algorithm to map the defect density onto a GIS map, intuitively displaying high-risk areas.

[0037] Furthermore,

[0038] The personnel and access control unit includes a user authentication module, an access allocation and dynamic adjustment module, and an operation log auditing and risk warning module.

[0039] The user authentication module uses a multi-factor authentication algorithm, including password verification and abnormal login detection, to input user login information. The password verification uses SHA-256 encryption for storage and is compared during login. The abnormal login detection uses a behavior baseline algorithm. If the login IP does not match the commonly used IP or the login time deviates from the usual time period, secondary verification is triggered.

[0040] The permission allocation and dynamic adjustment module is based on the RBAC permission mapping algorithm. It presets roles, and each role is bound to a set of permissions. The permission granularity control is achieved through a bit mapping algorithm, which maps each operation to a bit. User permissions are stored as binary numbers. Dynamic adjustment is achieved when a user's role changes. The algorithm automatically triggers the "role unbinding - new role binding" process and updates the permission bits synchronously.

[0041] The operation log audit and risk warning module records all user operations through an abnormal operation identification algorithm, identifies abnormal behavior through frequent pattern mining, and scores risks based on the sensitivity level and degree of abnormality of the operation. If the score is greater than 80, an audit warning is triggered, and the administrator is notified to check.

[0042] Furthermore,

[0043] The fitness function integrates the minimization of total path length, obstacle avoidance, battery life, and task priority; the iterative optimization uses selection, crossover, and mutation operations to select the optimal path and assign it to idle drones; the real-time status of the drones is obtained through the drone status monitoring module, and the real-time status of the drones includes idle, charging, and faulty.

[0044] Furthermore,

[0045] The multi-dimensional threshold comparison algorithm includes location risk, battery risk, and attitude risk. The location risk is calculated by the distance between the real-time coordinates of the UAV and the electronic fence. If the distance is less than the safety threshold, a "deviation from flight path warning" is triggered, and the location of the nearest obstacle is displayed. The battery risk is triggered by the remaining flight time being less than 1.2 times the return time, which triggers a "low battery warning". The attitude risk is triggered by the pitch angle or roll angle being greater than 15°, or the heading angle change being greater than 30° / s, which triggers an "abnormal attitude warning".

[0046] Furthermore,

[0047] The weighted scoring and grading algorithm first presets a base score based on the severity of the defect, then adjusts the weights by weighting the equipment importance and the environment, and finally grades the defect according to the grading criteria. The grading criteria are as follows: a total score > 90 points is a "critical defect" and will be processed within 24 hours; 60-90 points is a "serious defect" and will be processed within 7 days; and < 60 points is a "general defect" and will be processed within 30 days.

[0048] Furthermore,

[0049] The intelligent dispatching algorithm calculates the matching degree based on the skill tags, geographical location, and current workload of the handlers. The matching degree is calculated as: skill matching × 0.6 + distance × 0.3 + workload × 0.1. The algorithm then automatically assigns the task to the optimal handler.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] The implementation of this system has fundamentally transformed the traditional power line inspection operation mode, constructing a new inspection and maintenance system that integrates automation, intelligence, and visualization. Its beneficial effects are comprehensive and multi-layered, specifically reflected in:

[0052] 1. Improve power grid operation safety

[0053] Accurate identification: The AI ​​defect identification model can detect many early and subtle defects that are difficult for the human eye to detect (such as micro cracks and pin-level defects), thus achieving "prevention before problems occur";

[0054] Anomaly Intervention: The drone inspection and monitoring unit can perceive the drone status and inspection environment in real time, and provide timely alarms and remote intervention for flight anomalies, sudden weather changes, etc., to avoid secondary accidents;

[0055] Closed-loop management: The defect management unit ensures that every discovered defect can be tracked, processed, and verified, forming a closed loop, eliminating the management loophole of "only checking but not fixing", and directly improving the safety level of the line itself.

[0056] 2. Improve inspection efficiency

[0057] Fully automated operation: Through intelligent scheduling of the drone nest and task management unit, the drone can achieve autonomous take-off, inspection, landing, charging and data transmission. Pilots no longer need to travel long distances to the site to operate the drone. One person can monitor multiple drones and manage the inspection tasks of multiple drone nests.

[0058] Furthermore, AI real-time / near real-time analysis has changed the traditional method of manual review of massive amounts of images. AI algorithms can complete the automatic analysis, defect identification and classification of inspection images within minutes or even seconds, improving analysis efficiency by tens or even hundreds of times.

[0059] The closed-loop management efficiency is improved. The defect management unit makes the entire process of defect "identification, classification, reporting, dispatching, processing and review" online and closed-loop, clarifies the responsible persons and processing time limits, and greatly shortens the cycle from defect discovery to defect elimination, thereby improving the overall efficiency of operation and maintenance management.

[0060] 3. Achieve digitalization and intelligentization of operation and maintenance management.

[0061] The task management unit realizes intelligent scheduling of the entire task lifecycle. All inspection plans, execution processes, and result data are clearly visible, realizing the standardization and transparency of operation and maintenance management. The data center unit transforms the scattered and unstructured inspection data into unified, analyzable, and mineable digital assets. The AI ​​model is continuously optimized in continuous data training. Operation and maintenance personnel can intuitively understand the panoramic view of the health status of the entire power grid through the platform, transforming from a "passive response" mode based on experience to a "proactive prediction and intelligent decision-making" mode based on data.

[0062] This system has built a modern power grid inspection and maintenance system that is data-driven, automatically executed, and intelligently decision-making, comprehensively improving the safety, efficiency, and intelligence of power grid operation and maintenance. Attached Figure Description

[0063] Figure 1 This is a framework diagram of the nest management platform of the present invention;

[0064] Figure 2 This is the interface of the machine nest management platform of the present invention;

[0065] Figure 3 This is the data center unit interface of the nest management platform of the present invention;

[0066] Figure 4 This is the interface of the drone inspection and monitoring unit of the drone nest management platform of the present invention;

[0067] Figure 5 This is the interface of the flight status monitoring and risk warning module of the nest management platform of the present invention. Detailed Implementation

[0068] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0069] like Figures 1-5 As shown, it illustrates a specific embodiment of the present invention:

[0070] like Figures 1-5As shown, the IoT drone intelligent AI power line defect identification system disclosed in this invention includes a drone, a drone housing, and a drone housing management platform. Both the drone housing management platform and the drone housing use the MQTT protocol to control the drone and update its status. The drone housing management platform and the drone housing communicate wirelessly. The drone housing management platform includes a task management unit, a drone inspection and monitoring unit, a defect management unit, a data center unit, and a personnel and access control unit.

[0071] The task management unit is used for intelligent scheduling throughout the entire task lifecycle;

[0072] The drone inspection and monitoring unit is used for real-time status perception and anomaly intervention;

[0073] The defect management unit is used for defect identification, classification, and closed-loop processing.

[0074] Data center units are used for data storage, analysis, and mining;

[0075] The personnel and access control unit is used for identity authentication and access control.

[0076] Specifically,

[0077] The task management unit includes a task generation and parameter parsing module, a task allocation and path optimization module, and a task allocation and path optimization module.

[0078] The module includes a task execution monitoring and dynamic adjustment module, and a task acceptance and archiving module.

[0079] The task generation and parameter parsing module takes into account the inspection area, task type, time window, and equipment type set by the user. It extracts the equipment ledger within the target area from the database using a GIS coordinate-based regional boundary matching algorithm. This ledger includes tower locations, line routes, and historical defect points. Combined with the drone's endurance parameters and nest deployment location, it calculates the minimum inspection unit.

[0080] The task allocation and path optimization module initializes and generates a random path set covering all target towers, starting from the tower nest, and then uses a fitness function and iterative optimization to perform task allocation and path optimization.

[0081] The task execution monitoring and dynamic adjustment module uses a threshold-triggered adjustment algorithm based on real-time input of the drone's location, remaining battery power, number of completed poles, and sudden weather events: if the remaining battery power is less than the battery power required for return based on the real-time location, an "emergency return" is automatically triggered, and incomplete tasks are marked as "pending inspection"; if no-fly weather occurs, the task is suspended and enters a "waiting queue", and is reordered according to priority after the weather recovers.

[0082] The task acceptance and archiving module uses the inspection data and path deviation rate uploaded by the UAV to check whether there is at least one clear photo of the key parts of each tower. The clarity of the clear photo is judged by the image entropy value. An entropy value > 50 is considered qualified. If the path deviation rate is < 5% and the data is complete, it is marked as "task completed"; otherwise, it is marked as "re-inspection required" and a supplementary inspection task is automatically generated.

[0083] The task management unit is a full-lifecycle intelligent scheduling algorithm process. Users define the inspection area, task type, time window, and equipment type, such as the designated inspection area ("220kV Donghong Line #1-#50 towers"), task type (routine inspection / special inspection), time window (e.g., "2025-10-28 08:00-12:00"), and equipment type (UAV model / nest number). This unit has the following advantages:

[0084] Precise planning: Based on GIS, equipment ledgers and drone endurance, intelligent calculation of the "smallest inspection unit" ensures precise matching of tasks and resources, improving the scientific nature of planning and resource utilization from the source.

[0085] Path optimization: Using intelligent algorithms for task allocation and path planning, it can automatically calculate the optimal flight path covering all target towers, significantly improving inspection efficiency and reducing flight energy consumption.

[0086] Intelligent and highly safe operation: It has the ability to make real-time dynamic adjustments and can automatically trigger emergency return or mission pause / resumption through intelligent judgment of thresholds such as power and weather, which greatly ensures the safety of operation.

[0087] Standardized and closed-loop acceptance: Quantitative indicators such as image entropy are introduced to automatically accept data quality and are automatically linked to task status. Re-inspection instructions are automatically generated for non-conforming tasks, forming a "discovery-review" quality closed loop, ensuring the validity of inspection data and the final completion rate of tasks.

[0088] Specifically,

[0089] The UAV inspection and monitoring unit includes a real-time data acquisition and analysis module, a flight status monitoring and risk warning module, and an abnormal intervention and control command generation module.

[0090] The real-time data acquisition and analysis module takes the status data transmitted back by the UAV every second as input. The status data includes GPS coordinates, altitude, speed, battery level, attitude angle, and camera parameters. It uses data cleaning and standardization algorithms to remove outliers, converts the attitude angle into three-dimensional spatial coordinates, and converts the battery level into remaining flight time. The remaining flight time is based on the energy consumption model of the current speed and load: time = remaining battery level / average current.

[0091] The flight status monitoring and risk warning module uses a multi-dimensional threshold comparison algorithm to provide risk warnings.

[0092] If the warning level is "emergency", the abnormal intervention and control command generation module will automatically generate control commands through the emergency return-to-home algorithm and the hovering waiting algorithm. The emergency return-to-home algorithm calculates the shortest return-to-home path and sends the "one-click return-to-home" command to the UAV through encrypted transmission via the MQTT protocol. The hovering waiting algorithm sends a "hover" command if it is temporarily unable to return to home, and at the same time notifies the nest to prepare for emergency reception.

[0093] Real-time data acquisition removes outliers. Outliers such as GPS jumps greater than 10m are identified as signal interference and are filled using the average of the first 3 seconds. The UAV inspection and monitoring unit constructs a closed-loop control system of real-time perception, intelligent early warning, and automatic intervention. This unit has the following advantages:

[0094] Precise Situational Awareness: Through data cleaning and standardization, raw data (such as attitude angles and battery power) are transformed into intuitive and decision-making spatial location and remaining flight time, providing monitoring personnel with a unified, accurate, and highly readable global operational view.

[0095] Risk warning proactive: Employing a multi-dimensional threshold comparison algorithm, it can automatically diagnose flight status 24 / 7, enabling early detection and graded warning of potential risks (such as low battery or abnormal attitude), transforming passive response into proactive prevention.

[0096] Automated and highly secure emergency response: In emergency situations, the system can bypass human delays and automatically execute the optimal emergency commands calculated beforehand (such as emergency return to base or hovering), greatly improving the efficiency and reliability of emergency response.

[0097] Efficient Operation and Maintenance Management: This unit has transformed the mode from "manually monitoring the screen" to "intelligent system monitoring with key personnel focus", which greatly reduces the workload of monitoring personnel and allows a single person to monitor the inspection operations of multiple drones at the same time, thus improving the overall operation and maintenance efficiency.

[0098] Specifically,

[0099] The defect management unit includes a defect identification and location module, a defect classification and risk assessment module, a defect dispatch and progress tracking module, and a defect re-inspection and closed-loop verification module.

[0100] The defect identification and localization module takes equipment images or videos captured by a drone as input, and performs preprocessing, model inference, and spatial localization based on deep learning-based target detection and segmentation algorithms. The preprocessing includes dehazing (for hazy weather), image stabilization (correcting blur caused by drone shaking), and cropping (focusing on the equipment area and eliminating background interference). The model inference uses an improved YOLOv8 model (the training set contains 100,000+ power defect samples: missing bolt pins, broken insulators, broken conductor strands, etc.), and outputs the defect category (e.g., "missing bolt pin"), location coordinates (pixel box in the image), and confidence score (0~1). A confidence score > 0.8 is considered a valid identification. The spatial localization combines the GPS coordinates and camera angle (calculated using IMU data) taken by the drone to map the image defect to the actual equipment location (e.g., "220kV Donghong Line #15 tower A-phase insulator").

[0101] The defect classification and risk assessment module classifies defects by inputting defect category, equipment importance (e.g., main line > branch line), and environmental factors (e.g., corrosion defect risk is higher in coastal areas). It uses a weighted scoring classification algorithm to classify defects.

[0102] The defect dispatching and progress tracking module uses an intelligent dispatching algorithm to dispatch defects. Through a state transition algorithm, the defect status changes from newly discovered → processing → repaired → accepted. Defects that have not been processed within the time limit are automatically upgraded and reminded. If a serious defect exceeds 3 days, an early warning is pushed to the management personnel.

[0103] The defect re-inspection and closed-loop verification module takes processed on-site photos and UAV re-inspection images as input, and uses a defect elimination verification algorithm to compare the image features before and after processing (such as pixel changes in the defect area). If the defect features disappear (such as the bolt pin has been installed) and the re-inspection confidence level is <0.2, it is determined that "the defect has been eliminated" and the closed-loop is archived; otherwise, it is marked as "processing invalid" and a new order is dispatched.

[0104] This defect management unit achieves the following results by constructing an intelligent, closed-loop defect handling process:

[0105] More accurate identification and more precise positioning: The improved AI model, trained on massive samples and combined with image preprocessing technology, significantly improves the accuracy and robustness of defect identification in complex environments (such as fog and shaking); through spatial positioning technology, it can accurately map two-dimensional image defects to specific equipment locations, providing clear guidance for on-site maintenance.

[0106] More scientific rating and more controllable risk: The weighted rating is based on a comprehensive consideration of defect type, equipment importance and environmental factors, which changes the traditional model of relying solely on defect type. This makes risk assessment more forward-looking, can prioritize high-risk defects, optimize resource allocation and improve the safety level of power grid operation.

[0107] More efficient processes and comprehensive management: Through intelligent task dispatching and status tracking, automatic task allocation and full monitoring of the processing process are achieved; the automatic timeout reminder mechanism ensures that every defect is responded to in a timely manner, avoiding task backlog and management blind spots, and improving operation and maintenance management efficiency.

[0108] A more rigorous closed loop ensures quality: The introduction of an automatic re-inspection and verification algorithm based on image comparison uses quantitative standards to determine whether defects have been truly eliminated, ensuring the effectiveness of the processing work and forming a rigorous "discovery-processing-verification" quality closed loop, eliminating "false elimination" and recurring problems.

[0109] Specifically,

[0110] The data center unit includes a data access and standardized storage module, a historical data trend analysis module, and a data value mining and application module.

[0111] The data access and standardized storage module takes multi-source data as input, including UAV images or videos, sensor temperature and humidity data, task logs, and defect records. It uses data classification and indexing algorithms to build a three-level index (e.g., "20251028_#15 tower_image data") based on "time + device ID + data type", and associates it with a distributed database (e.g., HBase for storing images, MySQL for storing structured data). It also segments unstructured data (each segment is 5 minutes long) and extracts keyframes (frames containing defects) and marks them separately.

[0112] The historical data trend analysis module, by inputting historical defect records of a certain device (such as the number of times "bolts loosened" on tower #15 in the past 3 years) and environmental data (such as annual average humidity), uses an LSTM neural network to learn the correlation between defect occurrence and environmental factors (such as the probability of bolt loosening increases by 15% for every 10% increase in humidity), predicts the defect risk value for the next 6 months, and outputs a "high / medium / low" risk level.

[0113] The data value mining and application module uses an association rule mining algorithm to analyze the correlation of defects in multiple devices. If the confidence level is greater than 70%, the association rule is output ("The tilting of the tower may cause uneven stress on the insulator and damage"). The data visualization uses a heat map generation algorithm to map the defect density (number of defects per unit kilometer) to a GIS map, intuitively displaying high-risk areas (such as areas with high incidence of external force damage in urban-rural fringe areas).

[0114] This data center unit, through systematic data governance and intelligent analysis, transforms massive and chaotic raw data into systematic and easily searchable data through unified access, classification, indexing, and distributed storage of multi-source heterogeneous data. This lays the foundation for subsequent analysis and applications, greatly improving data management efficiency and availability. The forward-looking operation and maintenance decision-making based on LSTM model historical trend analysis reveals the inherent patterns between defects and environmental factors, enabling a shift towards pre-emptive prediction. This makes operation and maintenance strategies more forward-looking, prioritizing high-risk equipment and effectively preventing failures before they occur. Utilizing association rule mining, it can discover deep causal relationships hidden among defects in multiple devices, forming knowledge and experience that can guide design and operation and maintenance. Combined with GIS heatmap visualization, the data analysis results are presented intuitively, quickly locating high-risk areas and providing strong decision support for power grid planning and risk prevention.

[0115] Specifically,

[0116] The personnel and access control unit includes a user authentication module, an access allocation and dynamic adjustment module, and an operation log auditing and risk warning module.

[0117] The user authentication module uses a multi-factor authentication algorithm, including password verification and abnormal login detection, to accept user login information (such as username and password). The password verification uses SHA-256 encryption for storage and is compared during login (e.g., salting, where the salt value is a random string to prevent rainbow table cracking). The abnormal login detection uses a behavior baseline algorithm. If the login IP does not match the commonly used IP or the login time deviates from the usual time period (e.g., 3 AM), a secondary verification is triggered, such as an SMS verification code.

[0118] The permission allocation and dynamic adjustment module is based on the RBAC (Role-Based Access Control) permission mapping algorithm. Preset roles include administrator, inspector, and auditor. Each role is bound to a set of permissions; for example, an administrator can delete tasks, while an inspector can only view tasks. Permission granularity control uses a bit mapping algorithm, assigning each operation, such as "view defects" or "assign tasks," to a single bit (0 = no permission, 1 = permission). User permissions are stored as binary numbers, such as 1011 indicating permissions 1, 2, and 4. Dynamic adjustment occurs when a user's role changes, automatically triggering a "role unbinding - new role binding" process and synchronously updating the permission bits.

[0119] The operation log auditing and risk warning module records all user operations, such as "deleting defective records" and "modifying task time," through an abnormal operation identification algorithm. It identifies abnormal behavior by mining frequent patterns, such as a user deleting 10 records in one hour, which far exceeds the normal operation frequency. Based on the operation sensitivity level, such as data deletion = 10 points, query = 1 point, and the degree of abnormality, a risk score is given. If the score is greater than 80 points, an audit warning is triggered, and the administrator is notified to check.

[0120] This personnel and access control unit constructs a security system integrating identity authentication, access control, and behavior auditing. Employing multi-factor authentication and abnormal login detection, it builds a dynamic account security defense line, effectively preventing account theft and unauthorized access, significantly improving system access security. Access management is precise and efficient: based on the RBAC model and bit mapping technology, it achieves refined and automated access control, ensuring clear responsibilities for different user roles; automatic synchronization of permissions when roles change significantly reduces management complexity and the risk of human error. Through full log recording and abnormal operation identification algorithms, it achieves full auditability of user behavior and proactively detects internal risks (such as malicious or accidental operations), providing real-time early warning and post-event investigation capabilities for system security and data integrity, meeting compliance requirements.

[0121] Specifically,

[0122] The fitness function integrates minimizing the total path length, obstacle avoidance, battery life, and task priority, such as adding 30% weight to special patrol tasks. The iterative optimization uses selection, crossover, and mutation operations to select the optimal path, such as "Nest → #1 Tower → #8 Tower → ... → #50 Tower → Nest", and assigns it to idle drones. The real-time status of the drones is obtained through the nest status monitoring module, and the real-time status of the drones includes idle, charging, and faulty.

[0123] The fitness function comprehensively considers multiple constraints such as path length, obstacle avoidance, battery life, and task priority (e.g., special patrol tasks have a weight of +30%). Through selection, crossover, and mutation operations in iterative optimization, it can select the flight path with the lowest overall cost and highest efficiency at the global level, rather than simply pursuing the shortest distance. The optimized path is dynamically allocated to drones in an "idle" state, and the real-time status of the drones (idle, charging, faulty) is used as the key input for allocation. This achieves accurate and real-time matching between task requirements and drone resources, effectively improving the attendance rate and utilization rate of the entire drone fleet. This algorithm can not only plan the optimal path, but also intelligently distinguish task priorities, ensuring that important tasks such as special patrols are given priority. At the same time, by considering battery life and obstacle avoidance, most flight risks are avoided in advance at the planning level, significantly improving the success rate and safety of task execution.

[0124] Specifically,

[0125] The multi-dimensional threshold comparison algorithm includes location risk, battery risk, and attitude risk. The location risk is calculated by determining the distance between the UAV's real-time coordinates and the electronic fence (no-fly zone / obstacle). If the distance is less than a safety threshold, such as 50m, a "deviation from flight path warning" is triggered, and the location of the nearest obstacle is displayed. The battery risk is triggered by the remaining flight time being less than 1.2 × the return time required, which triggers a "low battery warning". The attitude risk is triggered by the pitch or roll angle being greater than 15°, which may lead to loss of control, or the heading angle changing abruptly by more than 30° / s, which triggers an "abnormal attitude warning" that may be affected by strong winds.

[0126] By comprehensively monitoring three key dimensions—position, battery level, and attitude—in real time, a comprehensive and multi-layered three-dimensional risk perception system has been constructed, covering major risk types from airspace incursions and energy depletion to flight instability, greatly improving the comprehensiveness of risk identification. Employing thresholds with safety margins (such as a 1.2x battery level coefficient and a 15° attitude angle), the system has shifted from "critical alarm" to "early warning," reserving valuable decision-making and response time for the system or operator, transforming passive handling into proactive prevention.

[0127] Specifically,

[0128] The weighted scoring and grading algorithm first presets a base score based on the severity of the defect, such as broken conductor strand = 100 points, loose bolt = 50 points. Then, it adjusts the weights by weighting the equipment importance and the environment. Finally, it grades the defect according to the grading standards, such as 1.5 for main lines and 1.2 for coastal / high humidity areas. The grading standards are as follows: a total score > 90 points is a "critical defect" and must be handled within 24 hours; 60-90 points is a "serious defect" and must be handled within 7 days; and < 60 points is a "general defect" and must be handled within 30 days.

[0129] This weighted scoring and grading algorithm, through quantitative evaluation and multi-factor fusion, calculates using preset base scores and explicit mathematical formulas. It eliminates the subjectivity and experience dependence of traditional manual grading, ensuring consistency of grading results among different defects and different assessors. It directly links the grading results (total score) with explicit processing time limits, effectively prioritizing the handling of high-risk defects, optimizing resource allocation, thereby improving the overall efficiency of hazard elimination and ensuring the safe and stable operation of the power grid.

[0130] Specifically,

[0131] The intelligent dispatching algorithm calculates the matching degree based on the skill tags of the handlers, such as "proficient in insulator replacement", geographical location, proximity to the defect point, and current workload. The matching degree is calculated as: skill matching × 0.6 + distance × 0.3 + load × 0.1. The algorithm then automatically assigns the task to the optimal handler.

[0132] By comprehensively calculating skill matching, geographical proximity, and current workload, tasks can be quickly and automatically assigned to the most suitable personnel. This significantly shortens the communication and decision-making time of traditional manual task assignment and accelerates defect response. The algorithm weight settings (skill 0.6, distance 0.3, workload 0.1) transform task assignment decisions from a model that relies on the administrator's personal experience to an objective, quantitative algorithm based on clear rules. This reduces subjective arbitrariness and ensures the fairness and scientific nature of task allocation, making it a key link in realizing the intelligent upgrade of operation and maintenance management.

[0133] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention. These changes involve related technologies well known to those skilled in the art, and all of them fall within the protection scope of the present invention.

[0134] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. An IoT-based drone-based intelligent AI power line defect identification system, characterized in that: The system includes a drone, a drone nest, and a drone nest management platform. Both the drone nest management platform and the drone nest use the MQTT protocol for command control and status updates. The drone nest management platform and the drone nest communicate wirelessly. The drone nest management platform includes a task management unit, a drone inspection and monitoring unit, a defect management unit, a data center unit, and a personnel and access control unit. The task management unit is used for intelligent scheduling throughout the entire task lifecycle; The drone inspection and monitoring unit is used for real-time status perception and anomaly intervention; The defect management unit is used for defect identification, classification, and closed-loop processing. Data center units are used for data storage, analysis, and mining; The personnel and access control unit is used for identity authentication and access control.

2. The IoT-based drone-based intelligent AI power line defect identification system according to claim 1, characterized in that, The task management unit includes a task generation and parameter parsing module, a task allocation and path optimization module, a task execution monitoring and dynamic adjustment module, and a task acceptance and archiving module. The task generation and parameter parsing module takes into account the inspection area, task type, time window, and equipment type set by the user. It extracts the equipment ledger within the target area from the database using a GIS coordinate-based regional boundary matching algorithm. This ledger includes tower locations, line routes, and historical defect points. Combined with the drone's endurance parameters and nest deployment location, it calculates the minimum inspection unit. The task allocation and path optimization module initializes and generates a random path set covering all target towers, starting from the tower nest, and then uses a fitness function and iterative optimization to perform task allocation and path optimization. The task execution monitoring and dynamic adjustment module uses a threshold-triggered adjustment algorithm based on real-time input of the drone's location, remaining battery power, number of completed poles, and sudden weather events: if the remaining battery power is less than the battery power required for return based on the real-time location, an "emergency return" is automatically triggered, and incomplete tasks are marked as "pending inspection"; if no-fly weather occurs, the task is suspended and enters a "waiting queue", and is reordered according to priority after the weather recovers. The task acceptance and archiving module uses the inspection data and path deviation rate uploaded by the UAV to check whether there is at least one clear photo of the key parts of each tower. The clarity of the clear photo is judged by the image entropy value. An entropy value > 50 is considered qualified. If the path deviation rate is < 5% and the data is complete, it is marked as "task completed"; otherwise, it is marked as "re-inspection required" and a supplementary inspection task is automatically generated.

3. The IoT-based drone intelligent AI power line defect identification system according to claim 2, characterized in that, The UAV inspection and monitoring unit includes a real-time data acquisition and analysis module, a flight status monitoring and risk warning module, and an abnormal intervention and control command generation module. The real-time data acquisition and analysis module takes the status data transmitted back by the UAV every second as input. The status data includes GPS coordinates, altitude, speed, battery level, attitude angle, and camera parameters. It uses data cleaning and standardization algorithms to remove outliers, converts the attitude angle into three-dimensional spatial coordinates, and converts the battery level into remaining flight time. The remaining flight time is based on the energy consumption model of the current speed and load: time = remaining battery level / average current. The flight status monitoring and risk warning module uses a multi-dimensional threshold comparison algorithm to provide risk warnings. If the warning level is "emergency", the abnormal intervention and control command generation module will automatically generate control commands through the emergency return-to-home algorithm and the hovering waiting algorithm. The emergency return-to-home algorithm calculates the shortest return-to-home path and sends the "one-click return-to-home" command to the UAV through encrypted transmission via the MQTT protocol. The hovering waiting algorithm sends a "hover" command if it is temporarily unable to return to home, and at the same time notifies the nest to prepare for emergency reception.

4. The IoT-based drone intelligent AI power line defect identification system according to claim 3, characterized in that, The defect management unit includes a defect identification and location module, a defect classification and risk assessment module, a defect dispatch and progress tracking module, and a defect re-inspection and closed-loop verification module. The defect identification and localization module takes equipment images or videos captured by a drone as input, and performs preprocessing, model inference, and spatial localization based on deep learning-based target detection and segmentation algorithms. The preprocessing includes image dehazing, image stabilization, and cropping. The model inference uses an improved YOLOv8 model to output the defect category, location coordinates, and confidence score. A confidence score > 0.8 is considered a valid identification. Spatial localization combines the GPS coordinates and camera angles taken by the drone to map the image defects to the actual equipment location. The defect classification and risk assessment module classifies defects by inputting defect category, equipment importance, and environmental factors, and uses a weighted scoring classification algorithm. The defect dispatching and progress tracking module uses an intelligent dispatching algorithm to dispatch defects and uses a state transition algorithm to record defects whose status has expired and have not been processed, and automatically escalates the reminders. The defect re-inspection and closed-loop verification module inputs processed on-site photos and UAV re-inspection images, and uses a defect elimination verification algorithm to compare the image features before and after processing. If the defect features disappear and the re-inspection confidence level is <0.2, it is determined that "the defect has been eliminated" and the closed-loop is archived; otherwise, it is marked as "processing invalid" and a new order is dispatched.

5. The IoT-based drone intelligent AI power line defect identification system according to claim 4, characterized in that, The data center unit includes a data access and standardized storage module, a historical data trend analysis module, and a data value mining and application module. The data access and standardized storage module takes in multi-source data, including UAV images or videos, sensor temperature and humidity data, mission logs, and defect records. It uses data classification and indexing algorithms to build a three-level index based on "time + device ID + data type", links it to a distributed database, performs fragmentation processing on unstructured data, and extracts key frames for separate marking. The historical data trend analysis module, by inputting historical defect records and environmental data of a certain device, uses an LSTM neural network to learn the correlation between defect occurrence and environmental factors, predicts the defect risk value for the next 6 months, and outputs a "high / medium / low" risk level. The data value mining and application module uses an association rule mining algorithm to analyze the correlation of defects in multiple devices. If the confidence level is greater than 70%, the association rule is output. The data visualization uses a heat map generation algorithm to map the defect density onto a GIS map, intuitively displaying high-risk areas.

6. The IoT-based drone intelligent AI power line defect identification system according to claim 5, characterized in that, The personnel and access control unit includes a user authentication module, an access allocation and dynamic adjustment module, and an operation log auditing and risk warning module. The user authentication module uses a multi-factor authentication algorithm, including password verification and abnormal login detection, to input user login information. The password verification uses SHA-256 encryption for storage and is compared during login. The abnormal login detection uses a behavior baseline algorithm. If the login IP does not match the commonly used IP or the login time deviates from the usual time period, secondary verification is triggered. The permission allocation and dynamic adjustment module is based on the RBAC permission mapping algorithm, with preset roles and each role bound to a set of permissions; The granular control of permissions uses a bit mapping algorithm to assign each operation to a bit, and user permissions are stored as binary numbers; dynamic adjustment is made, and when a user's role changes, the algorithm automatically triggers the "role unbinding - new role binding" process and updates the permission bits synchronously; The operation log auditing and risk warning module records all user operations through an abnormal operation identification algorithm and identifies abnormal behavior through frequent pattern mining. Risk scores are assigned based on the sensitivity level and degree of abnormality of the operation. Scores greater than 80 points trigger an audit alert and notify the administrator to conduct a review.

7. The IoT-based drone intelligent AI power line defect identification system according to claim 2, characterized in that, The fitness function integrates the minimization of total path length, obstacle avoidance, battery life, and task priority; the iterative optimization uses selection, crossover, and mutation operations to select the optimal path and assign it to idle drones; the real-time status of the drones is obtained through the drone status monitoring module, and the real-time status of the drones includes idle, charging, and faulty.

8. The IoT-based drone intelligent AI power line defect identification system according to claim 3, characterized in that, The multi-dimensional threshold comparison algorithm includes location risk, battery risk, and attitude risk. The location risk is calculated by measuring the distance between the drone's real-time coordinates and the electronic fence. If the distance is less than the safety threshold, a "deviation from flight path warning" is triggered, and the location of the nearest obstacle is displayed. The battery risk is triggered by a remaining flight time < 1.2 × return time, which triggers a "low battery warning"; the attitude risk is triggered by a pitch or roll angle > 15°, or a sudden change in heading angle > 30° / s, which triggers an "attitude anomaly warning".

9. The IoT-based drone intelligent AI power line defect identification system according to claim 4, characterized in that, The weighted scoring and grading algorithm first presets a base score based on the severity of the defect, then adjusts the weights by weighting the equipment importance and the environment, and finally grades the defect according to the grading criteria. The grading criteria are as follows: a total score > 90 points is a "critical defect" and will be processed within 24 hours; 60-90 points is a "serious defect" and will be processed within 7 days; and < 60 points is a "general defect" and will be processed within 30 days.

10. The IoT-based drone intelligent AI power line defect identification system according to claim 4, characterized in that, The intelligent dispatching algorithm calculates the matching degree based on the skill tags, geographical location, and current workload of the handlers. The matching degree is calculated as: skill matching × 0.6 + distance × 0.3 + workload × 0.

1. The algorithm then automatically assigns the task to the optimal handler.

Images