Automatic judgment system and method for mechanical threat level of power transmission line

Abstract

The invention provides an automatic judgment system and method for a mechanical threat level of a power transmission line, belongs to the technical field of safety operation and maintenance of the power transmission line, and constructs an automatic judgment system comprising a data acquisition module, a dimension decomposition and data preprocessing module, a visual language large model training and reasoning module and a result output and early warning processing module. The method comprises the following steps: acquiring a line image and basic attribute data, forming a structured data packet, analyzing an unstructured image into structured tags of different core dimensions, then completing multi-modal feature extraction and multi-dimensional decision analysis to output a threat level, and finally realizing display, early warning and storage. According to the method, dimension deconstruction is carried out on a complex mechanical threat judgment problem, a visual language large model technology and professional knowledge of power transmission line operation and maintenance are deeply fused, refined grading judgment of power transmission line mechanical threats is realized, the accuracy and interpretability of judgment are improved, and the judgment efficiency is improved. And the automation and intelligence level of operation and maintenance of the power transmission line is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of power transmission line safety operation and maintenance technology, specifically to an automated system and method for determining the mechanical threat level of power transmission lines. Background Technology

[0002] As the core of the power system's energy transmission, transmission lines are critical infrastructure ensuring electricity supply for social production and daily life. Their safe and stable operation directly affects the reliability and continuity of power supply. my country's transmission line network covers an extremely wide area, traversing not only plains and hills but also extending to complex environments such as mountains, forests, and coastlines. Long-term exposure to a complex environment interwoven with natural and human factors exposes them to diverse mechanical threats, which have become one of the main causes of transmission line failures and power grid outages. Specifically, the intrusion of construction machinery is the most significant human-caused mechanical threat. When heavy engineering equipment such as cranes, cement pump trucks, and excavators operate illegally within the transmission line corridor, they are prone to approaching or even touching the conductors due to improper operation, causing short circuits and power outages. In severe cases, it can even cause towers to collapse. Mechanical damage caused by the natural environment is also not to be ignored. Extreme weather such as storms, blizzards, and freezing temperatures can easily cause trees to fall and cover the lines, and tower hardware to loosen and fall off. Meanwhile, foreign objects such as kites, balloons, plastic films, and bird nests can get caught on the conductors, which can damage the insulation performance of the lines, creating a continuous short circuit risk and posing a hidden danger to the safety of the power grid.

[0003] To address these threats, the power operation and maintenance industry has long relied on traditional manual inspections and helicopter patrols for monitoring. However, both methods have insurmountable drawbacks. Manual inspections are the most basic maintenance method, requiring personnel to carry testing equipment and patrol along the lines on foot. This is not only labor-intensive and inefficient, but also significantly limited by geographical conditions. For complex sections such as mountainous areas and areas crossing rivers or seas, manual inspections are often inaccessible, resulting in numerous monitoring blind spots. Furthermore, manual inspections rely on the experience and judgment of maintenance personnel, making them prone to missed detections and misjudgments. They also cannot achieve 24-hour real-time monitoring, hindering rapid response to sudden threats. While helicopter patrols can cover areas inaccessible by humans, increasing the scope and efficiency of inspections, their maintenance costs are extremely high, reaching tens of thousands of yuan per hour for a single helicopter patrol. They are only suitable for regular inspections of key lines. In addition, helicopter patrols are severely constrained by weather conditions; heavy fog, heavy rain, and strong winds can force patrols to be interrupted, resulting in insufficient real-time performance and flexibility.

[0004] In recent years, with the rapid development of computer vision and deep learning technologies, image recognition-based methods for detecting hidden dangers in power transmission lines have gradually become a research and application hotspot. This technology uses drones and high-definition cameras mounted at fixed monitoring points to collect images of the power lines and their surroundings. Combined with target detection algorithms such as YOLO and Faster R-CNN, it can automatically identify potential hazards such as external intrusion, fallen trees, and foreign object attachment. Compared to traditional methods, it offers advantages such as high efficiency, automation, and the ability to provide real-time early warnings via edge computing. For example, Sun et al., based on an improved YOLOv8 model, achieved high accuracy and recall rates in identifying foreign objects such as balloons, kites, and bird nests, demonstrating the application potential of image recognition technology in power transmission line monitoring. Xu et al. systematically summarized the integrated application of drone platforms and image recognition technology, proposing an autonomous drone flight and image acquisition scheme suitable for power transmission line inspection.

[0005] However, existing research and applications of image recognition technology still face many key bottlenecks, making it difficult to meet the actual needs of refined and intelligent power operation and maintenance. Firstly, most research remains at the binary level of determining the "presence" or "absence" of potential hazards, failing to achieve a graded assessment of the severity of threats. In actual operation and maintenance, the risk levels of different types and states of threats to transmission lines vary significantly: low-level threats such as bird nests and small kites can be addressed periodically in routine operation and maintenance plans, while high-risk threats such as construction machinery approaching conductors or fallen trees about to touch lines require immediate activation of emergency response procedures. The existing binary judgment model cannot distinguish threat levels, easily leading to wasted operation and maintenance resources or untimely handling of high-risk risks. Secondly, insufficient data collection has become a core issue restricting model performance. In real-world scenarios, high-risk mechanical threats to power transmission lines are sporadic and dangerous, making them difficult to reproduce under experimental conditions. This results in existing publicly available datasets being limited in size and having an imbalanced category distribution. Furthermore, samples of high-risk hazards such as construction machinery intrusion and tower damage are severely lacking. This data imbalance leads to model training bias towards common samples, resulting in insufficient ability to identify high-risk hazards and significantly increasing the risk of missed and false detections. Thirdly, existing models often employ end-to-end black-box processing, only outputting detection results without explaining the basis and logic of the judgments. Maintenance personnel struggle to understand the model's decision-making process and cannot perform targeted maintenance optimizations based on the judgment results. These shortcomings make it difficult for existing technologies to be truly implemented and support intelligent management of mechanical threats to power transmission lines. Therefore, constructing an automated method and system for determining the level of mechanical threats to power transmission lines with hierarchical judgment capabilities, high generalization, and interpretability has become a major technical challenge that needs to be addressed in the power operation and maintenance field. Summary of the Invention

[0006] The purpose of this invention is to provide an automated method and system for determining the mechanical threat level of power transmission lines. The system transmits images of the power transmission lines back to a processing center through an information acquisition system. The processing center analyzes the images and decomposes the mechanical threat level determination mechanism into dimensions. A large visual language model is then trained and used to perform multi-dimensional decision analysis and judgment of the threat level of the power transmission lines. Finally, by aggregating the states from multiple dimensions, the system automates the determination of the mechanical threat level of the power transmission lines and automatically notifies maintenance personnel to take appropriate action.

[0007] To achieve the above objectives, the present invention employs the following technical solutions.

[0008] An automated system for determining the mechanical threat level of power transmission lines, comprising: The data acquisition module is used to collect image data and basic attribute data of transmission lines and form structured data packets; The dimensional decomposition and data preprocessing module receives the structured data packets from the data acquisition module, maps them to multiple preset core dimensions, and generates standardized structured data after data cleaning. The visual language large model training and inference module performs model training iterations based on the standardized structured data, and uses the trained visual language large model to perform multimodal feature extraction and multidimensional decision analysis to output the mechanical threat level of the transmission line. The results output and early warning handling module receives the threat level determination results, performs visualization, tiered early warning push, and full-link data storage.

[0009] Furthermore, the data acquisition module includes an image acquisition unit and a line basic information unit: The image acquisition unit is deployed on the transmission line tower to capture images of the line and surrounding scene at different times and under different weather conditions, and has a built-in wireless communication module. The line basic information unit is used to store basic attribute data of the transmission line. The basic attribute data includes at least one of electrical attributes, tower attributes, conductor / ground wire attributes, and section management information. The line basic information unit is also configured with an API interface for connecting to the power operation and maintenance management system.

[0010] Furthermore, the electrical attributes include the line voltage level, number of circuits, and phase sequence; the tower attributes include the tower number, model, and GPS precise coordinates; the conductor / ground wire attributes include the conductor model, splitting method, and rated sag; and the section management information includes the section to which the line belongs, the operation and maintenance unit, and the contact information of the responsible person. The data packet is encapsulated in JSON format, and the specific fields include tower_id, image_data, voltage_level, tower_type, location, and timestamp.

[0011] Furthermore, the core dimensions in the dimensional decomposition and data preprocessing module include spatial positioning analysis, power facility positioning, target object positioning, and mechanical feature recognition; the target detection model transforms image data and basic attribute data into label information for the corresponding dimensions; the data cleaning includes removing at least one of the following: non-vertical transmission line images and blurry images with pixel sizes below a preset threshold.

[0012] Furthermore, the spatial positioning analysis dimension is used to output the relative positional relationship between the target machinery and the transmission line protection zone, including the area within the protection zone, the near zone of the protection zone, and the area outside the protection zone; the power facility positioning dimension is used to identify and locate the transmission towers and line routes in the image to form the power system protection zone; the target object positioning dimension is used to detect external mechanical objects that interact with or pose a threat to the transmission line, and output the object category and bounding box coordinates; the mechanical feature recognition dimension is used to analyze the working status of the external mechanical objects based on the recognition results of the target object positioning dimension.

[0013] Furthermore, the visual language large model training and inference module includes a model continuous training unit and a multimodal inference unit: The continuous model training unit includes a dataset management submodule, a model training submodule, a hyperparameter optimization submodule, and a model validation submodule. The dataset management submodule is used for uploading and labeling new training data. The model training submodule is used to provide different training tasks. The hyperparameter optimization submodule is used to determine the optimal parameter combination using an automated search method. The model validation submodule is used to evaluate the performance of the trained model on a reserved test set. The multimodal inference unit is used to deploy a trained visual language model, receive standardized structured data, and infer the threat level by combining the power transmission line mechanical threat level determination criteria.

[0014] Furthermore, the results output and early warning handling module includes a visualization display unit, an early warning push unit, and a historical data storage unit; The visualization unit, based on a GIS map, is used to color-code and render the power transmission network and threat levels, with different color identifiers corresponding to different threat levels. The early warning push unit is used to trigger a tiered early warning mechanism based on the threat level. The tiered early warning mechanism includes at least one of the following: system records, SMS notifications, APP push notifications, voice calls, emergency resource association, and dispatching orders. The historical data storage unit is used to store structured threat event archives formed by collected data, dimensional analysis results, threat levels, and handling results. It is also equipped with a data query engine and reporting tools for multi-condition combined queries and data export.

[0015] An automated method for determining the mechanical threat level of power transmission lines includes the following steps: S1. Collect image data and basic attribute data of transmission lines, and associate them to form structured data packets; S2. Map the structured data packet to multiple preset core dimensions, transform unstructured image data into structured label information corresponding to each dimension, and generate standardized structured data after data cleaning; S3. Based on the standardized structured data, train a large visual language model, and use the trained model to perform multimodal feature extraction and multidimensional decision analysis to output the mechanical threat level; S4. Visualize the threat level, push tiered early warnings, and store full-link data.

[0016] Furthermore, step S1 also includes: S11. The image acquisition unit automatically acquires panoramic images of the transmission line at preset time intervals, the preset time interval being configured to be 5-15 minutes / time; the acquired image data is transmitted to the back-end processing center in real time via a high-speed wireless network through a wireless communication module; the line basic information unit synchronizes data with the power operation and maintenance management system in real time or near real time through an API interface. S12. The data acquisition module associates the image data from the image acquisition unit with the basic attribute data from the line basic information unit to generate a structured data packet.

[0017] Furthermore, during the data cleaning process in step S2, images of power transmission lines that are not oriented longitudinally and images of mechanical objects with a pixel size of less than 20×20 are removed. The threat levels mentioned in step S4 include Level A, Level B, Level C, and Level D; among them, Level D threats only trigger system records; Level C threats trigger SMS notifications; Level B threats trigger triple warnings via SMS, APP push, and voice calls; Level A threats trigger the triple warnings and generate disposal orders by associating with emergency resources around the line.

[0018] The advantages of this invention are: This design proposes a multi-dimensional decomposition paradigm for mechanical threats to power transmission lines and a structured data generation method for model-based reasoning. It pioneers a dimensional decomposition paradigm for mechanical threats to power transmission lines, systematically deconstructing the complex threat assessment task into multiple computable and analyzable independent dimensions. The dimensional decomposition and data preprocessing modules integrate various computer vision models to parse and transform raw, unstructured image data into multi-dimensional, standardized structured data representations. Protecting this unique dimensional decomposition paradigm, the predefined set of dimensions, and its corresponding structured data generation mechanism prevents others from circumventing the core innovation of this patent by using different dimension definitions or simple end-to-end models, ensuring the exclusivity of the entire judgment system's logical framework.

[0019] Domain-Adaptive Training Method and Dedicated Inference Mechanism for Determining Mechanical Threats to Transmission Lines: This scheme proposes for the first time to train and apply a large-scale visual-language model specifically for the security of transmission lines. By constructing a large-scale domain-specific dataset containing image-text pairs and designing a dimensionality-aware cue word to supervise and fine-tune the model, it gains the ability to determine the level of mechanical threats. Protecting this domain-customized model training method and its inference process in threat determination prevents others from circumventing infringement by simply replacing similar models or using similar inference methods.

[0020] Enhance the interpretability of judgment results and improve the generalization ability and accuracy of the model: Transform "black box" detection into multi-dimensional transparent analysis. The analysis results of each dimension are traceable, making it easier for operation and maintenance personnel to understand the judgment basis. High-quality structured data is generated through dimensional decomposition, and combined with the continuous training mechanism of the model, the problem of data imbalance is alleviated. High level of automation and intelligence: The entire process requires no human intervention, achieving closed-loop automation from data collection, analysis, judgment to early warning and handling. It combines the reasoning ability of the visual language big model to simulate expert decision-making thinking, improve the accuracy of judgment, adapt to transmission lines of different voltage levels and different scenarios, support model iteration and function expansion, and the historical data storage function provides data support for operation and maintenance optimization, making it easy to implement in engineering. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the automated system for determining the mechanical threat level of power transmission lines according to the present invention. Figure 2 This is a schematic diagram of the data acquisition module structure of the present invention; Figure 3 This is a schematic diagram of the dimensional decomposition and data preprocessing module of the present invention; Figure 4 This is a schematic diagram of the structure of the visual language large model training and inference module of the present invention; Figure 5This is a schematic diagram of the structure of the result output and early warning handling module of the present invention; Figure 6 This is a flowchart of the automated method for determining the mechanical threat level of power transmission lines according to the present invention. Detailed Implementation

[0022] 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.

[0023] Example 1 This embodiment discloses an automated system for determining the mechanical threat level of power transmission lines. Please refer to... Figure 1 It includes: a data acquisition module, a dimensionality decomposition and data preprocessing module, a visual language large model training and inference module, and a result output and early warning handling module.

[0024] Please refer to Figure 2 The data acquisition module is the system's information input terminal, used to collect and determine the necessary image data in real time. It includes an image acquisition unit and a line basic information unit. The image acquisition unit's functional design fully considers the complexity, diversity, and real-time requirements of transmission line scenarios. It mainly uses a group of monitoring cameras on the transmission towers to capture images of the line and surrounding scenes at different time periods and under different weather conditions. The cameras automatically acquire a panoramic image at preset time intervals (such as every 15 minutes) and transmit the image data to the back-end processing center via 4G / 5G networks.

[0025] The line basic information unit is responsible for storing and managing all inherent attribute data related to the transmission line itself, such as line voltage level, tower number and type, conductor type, and the section where the transmission line is located. It synchronizes data with the power company's existing operation and maintenance management system in real-time or near real-time via API. The stored data types include: Electrical attributes: line voltage level (e.g., 35kV, 110kV, 220kV, 500kV), number of circuits, phase sequence, etc.; Tower attributes: tower number, tower model (e.g., straight tower, tension tower, angle tower), material, nominal height, GPS precise coordinates, etc. Conductor / ground wire attributes: conductor type, splitting method, rated sag, safety factor, etc.; Section management information: The section to which the line belongs, the operation and maintenance unit, and the contact information of the responsible person, etc.

[0026] The data acquisition module correlates the data from the two units to form a structured data packet. This structured data packet is an encapsulated data packet formed by the data acquisition module associating basic line attribute data with image metadata. It only contains structured metadata such as tower identification, image storage path, voltage level, and geographic coordinates, while the image itself remains unstructured pixel data. The complete data packet is then sent to the processing center to complete one data acquisition task.

[0027] Data packets, for example: {"tower_id": "T001", "image_data": "(storage path)", "voltage_level": "500kV", "tower_type": "(tower type)", "location": "[longitude, latitude]", "timestamp":"2023-10-27 10:30:00"} Referring to Figure 3, the dimensionality decomposition and data preprocessing module receives raw images and line information from the data acquisition module. Its core task is to use a computer vision model to extract structured information with clearly defined dimensions from unstructured images, perform secondary structuring processing on the unstructured pixel data, and ensure that the data input to the subsequent decision module is high-quality and effective. This module includes a dimensionality mapping unit and a data cleaning unit.

[0028] Dimension Mapping: The original images are deeply analyzed using a computer vision model. The received data is then processed by an object detection model to map the acquired images and line information to four core dimensions: spatial positioning analysis, power facility positioning, target object positioning, and mechanical feature recognition. This transforms pixel information into business-meaning dimensional labels. In the dimension of power facility positioning, target detection models (such as YOLO and Faster R-CNN) identify and accurately locate transmission towers and power lines in images, and output the direction of transmission towers and lines as the protection zone of the power system; In the target object localization dimension, the model detects and localizes all external mechanical objects in the image that may interact with or pose a threat to the power transmission line. This function focuses on non-power facility objects such as cranes and cement pump trucks, outputs the target category, and includes its bounding box coordinates. In the spatial positioning analysis dimension, the relative positional relationship between the target machinery and the transmission line is analyzed and judged through a rule-based method, and the relative position results are output, including those within the protection zone, the near zone of the protection zone, and outside the protection zone. In the dimension of mechanical feature recognition, following the recognition results of the target object positioning dimension, the model is used to identify and analyze the working status of the mechanical facility itself, and output the recognition results (e.g., boom raised, boom not raised).

[0029] The data cleaning unit ensures that only high-quality, valid image data is included in subsequent analysis and judgment processes. This unit filters the data in two ways: 1) Power facilities: filtering out images of power transmission lines that are not oriented longitudinally; 2) Mechanical objects: filtering out images containing mechanical objects with poor clarity (pixel size <20×20). After data cleaning, standardized structured data is generated.

[0030] Please refer to Figure 4 The Visual Language Large Model Training and Inference Module, by combining the multimodal understanding capabilities of the Visual Language Large Model, can comprehensively analyze image and text information to make accurate and interpretable intelligent judgments on the mechanical threats to power transmission lines. This module includes a model continuous training unit and a multimodal inference unit.

[0031] The continuous model training unit includes a dataset management submodule, a model training submodule, a hyperparameter optimization submodule, and a model validation submodule. Dataset management: Stores, version-controls, and labels datasets used for training. Operations personnel can upload new line images, associate them with corresponding dimension mapping data, and label threat levels.

[0032] Model Training: Provides a management interface for training tasks, allowing operations and maintenance personnel to select different base models for training, specify training datasets, set training configurations, and start training tasks with one click; Hyperparameter optimization: Provides automated hyperparameter search functions (such as grid search and Bayesian optimization) to automatically find the best combination of parameters such as learning rate and batch size to maximize model performance and reduce the threshold and cost of manual parameter tuning; Model validation: After training is complete, the model performance is automatically evaluated on a reserved test set.

[0033] The multimodal inference unit applies the capabilities of the trained model to real-time data analysis: Model deployment and loading. Smoothly deploy validated models from the training environment to the production inference environment.

[0034] Multidimensional decision analysis and mechanical threat level determination: This function is the core capability of VLM. Based on the multidimensional information provided by the dimension decomposition module, VLM combines the industry standards for mechanical threat level determination to determine the mechanical threat level and output the results.

[0035] Please refer to Figure 5The results output and early warning response module receives the judgment results from the upstream module and transforms them into intuitive visual information, tiered early warning actions, and traceable data records. This module is directly geared towards maintenance personnel, with the ultimate goal of improving the efficiency and safety of transmission line operation and maintenance. This module includes a visualization unit, an early warning push unit, and a historical data storage unit.

[0036] The visualization unit is based on a GIS map and dynamically renders the distribution of the entire power transmission network. The system uses color coding to classify different line segments and towers based on real-time mechanical threat level data.

[0037] The early warning push unit triggers different early warning mechanisms based on the threat level. Level D threats are only recorded within the system; Level C threats are notified to the maintenance team via SMS; Level B threats are alerted via SMS, APP push, and voice call; Level A threats, in addition to the above methods, are automatically associated with emergency resources around the line (such as the nearest maintenance site and emergency vehicles) to generate a disposal order.

[0038] The historical data storage unit constructs a complete threat event archive, storing all data from the occurrence to the closure of an event, and provides a powerful data query engine and reporting tools, supporting maintenance personnel to flexibly combine queries and export data by time, line, tower, threat level, machine type, handling status and other conditions.

[0039] Example 2 This embodiment, based on the system structure of Embodiment 1, discloses an automated method for determining the mechanical threat level of transmission lines. Please refer to [link / reference]. Figure 6 This includes the following steps: S1. At a preset acquisition frequency of 5 minutes per acquisition, a group of cameras on the power transmission line periodically acquires images of the transmission line and simultaneously calls the API of the line basic information unit database to combine and associate the transmission line images with the basic information, transmitting the data to the processing center via a 4G / 5G network. The processing center then forms the received data into a structured data packet as follows: {"tower_id": "T001", "image_data": "(storage path)", "voltage_level": "500kV", "tower_type": "(tower type)", "location": "[longitude, latitude]", "timestamp":"2023-10-27 10:30:00"} This complete data packet is then transmitted to the downstream module to complete a data acquisition task.

[0040] S2. Receives and standardizes structured data packets. The standardized images are then simultaneously fed into multiple pre-trained computer vision models for multi-dimensional analysis: In the power facility location dimension, YOLO model 1 is used to depict the transmission towers and line routes, forming a power system protection zone; in the target object location and mechanical feature recognition dimension, YOLO model 2 is used to detect mechanical objects and their states; in the spatial location analysis dimension, geometric analysis rules are used to calculate the relative position of the center point of the mechanical object to the protection zone. After dimensional decomposition, the module calculates the detection results for power facility location and target object location, removing images containing non-vertical power lines and mechanical objects with poor clarity (e.g., pixel size <20×20). Finally, the module associates the identified targets, features, and spatial location information and inputs them to downstream modules.

[0041] S3. Operations personnel upload newly acquired line images and multi-dimensional information parsed by upstream modules through the front-end interface, which also provides an annotation interface. They select the base model, training dataset version, and set training parameters. This unit executes the training task on the GPU cluster, continuously monitors the loss function and evaluation metrics, and automatically evaluates the model on the test set after training is complete.

[0042] The system loads a specified Virtual Machine Module (VLM). This unit receives standardized JSON data packets from the upstream module, containing images and dimensional information. It combines the structured dimensional information into natural language prompts, inputting the images, constructed prompts, and threat level determination criteria into the VLM. The VLM internally performs feature extraction, fusion, and inference, ultimately determining the mechanical threat level. This unit outputs the final determination result, including the threat level and multi-dimensional analysis, to downstream components for generating work orders, triggering alarms, etc.

[0043] S4. Receives multi-dimensional information and machinery threat level data packets from upstream modules in real time via API. The visualization unit parses the data packets, extracting the pole location, threat level, and machinery type, and immediately changes the corresponding pole icon on the GIS map to the appropriate color. The early warning push unit routes the event to different processing branches based on the machinery threat level field: Level D: Process completed, only data recording completed.

[0044] C / B level: Call the corresponding notification interface (SMS gateway, APP push service) to immediately send an alert to the operation and maintenance personnel.

[0045] Level A: Send an alert as with Level C / B and generate an emergency response work order.

[0046] After processing, the historical storage unit will convert the original images, line information, multi-dimensional analysis results, mechanical threat level, and handling results into structured data and store it in the MySQL database.

[0047] Example 3 This embodiment addresses the unique challenges of blurred images and high background noise in power transmission line systems during heavy rain by implementing adaptive optimization of the dimensional decomposition and data preprocessing modules. After system startup, the structured data packets transmitted by the data acquisition module exhibit issues such as water mist obscuring the image data and high pixel noise due to the heavy rain.

[0048] The dimensional mapping unit first uses the Retinex dehazing algorithm to preprocess the unstructured image, improving image clarity. Then, it performs dimensional mapping using a YOLOv8 model: In the power facility positioning dimension, the model's feature enhancement branch strengthens the edge features of transmission towers and lines, accurately delineating the power protection zone and avoiding positioning errors caused by rain and fog. In the target object positioning dimension, addressing the weakening of color and outline features of construction cranes during heavy rain, the model detects and locates the cranes, outputting the machinery category and bounding box coordinates. In the spatial positioning analysis dimension, binocular vision ranging corrects distance errors caused by rain and fog, accurately determining the relative position of the machinery to the protection zone as near the protection zone. In the machinery feature recognition dimension, it identifies the crane as being in an un-raised state.

[0049] The data cleaning unit performs a quality assessment on the processed images. Since the pixel size of the mechanical object after defogging is 50×60, which is greater than the 20×20 threshold, and the power transmission line is longitudinal, the data is deemed valid. Finally, standardized structured data containing the crane in the vicinity of the protected area that has not raised its boom is generated and transmitted to the downstream module.

[0050] The multimodal inference unit of the visual language large model training and inference module loads the deployed visual language large model, converts the structured label information into a natural language prompt: "Under heavy rain, there is a crane in the near zone of the protection area of ​​the 500kV line T005 tower that is not in the boom state." The prompt, the original defogging image, and the industry standard for determining the mechanical threat level of transmission lines are input into the model.

[0051] The model uses multimodal feature extraction and fusion, combined with multidimensional information about cranes not raising their booms in the vicinity of the protected area, to infer the threat level of the machinery and ultimately determine it to be Level C.

[0052] The judgment result is then sent to the result output and early warning handling module: the visualization unit marks the T005 tower in yellow on the GIS map, intuitively presenting the location of the C-level threat; the early warning push unit calls the SMS gateway to send early warning SMS messages to the maintenance team responsible for the line section; the historical data storage unit stores the original blurred image of this event, the dehazing image, the analysis results of each dimension, the threat level and early warning records into the MySQL database, forming a complete threat event archive, providing data support for the optimization of the threat judgment model under subsequent rainstorm weather.

[0053] Example 4 This embodiment addresses the issue of low efficiency in hyperparameter tuning during model training by employing a Bayesian optimization algorithm to automate hyperparameter optimization. Operations personnel upload a domain dataset containing 100,000 power transmission line images and label them with threat levels through the front-end interface of the continuous model training unit. Hyperparameters for the basic visual language large-scale model training task are set: learning rate, batch size, and number of training epochs. The continuous model training unit initiates a Bayesian optimization hyperparameter search, using the model's F1 score as the optimization objective, and determines the optimal hyperparameter combination after 50 iterations. The training task is then started based on the optimal parameters, and supervised fine-tuning is performed on a GPU cluster. The convergence of the loss function is monitored in real time during training. After training, the model performance is verified on a test set to ensure it meets deployment requirements.

[0054] Example 5 This embodiment addresses the issue of insufficient high-risk hazard samples by employing incremental training to update the model and improve its ability to identify rare threats. For example, a novel mechanical threat—drone snagging on power lines—first appeared in section T020 of the transmission line. The data acquisition module captured relevant images and formed structured data packets. Maintenance personnel labeled these images as Level B threats and uploaded them to the dataset management module of the model's continuous training unit as new small sample data. The model's continuous training unit loaded the deployed large visual language model and adopted an incremental training strategy: freezing the model's bottom feature extraction layer and only fine-tuning the upper classification layer to prevent the model from forgetting existing knowledge. After training, the model's accuracy in identifying drone snagging threats reached a set threshold, allowing maintenance personnel to determine that this new threat could be included in the scope of detection, thus enabling the model to adaptively update for new scenarios.

[0055] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for automated determination of the level of mechanical threat to a power line, characterized in that, The method comprises the following steps: A data acquisition module is used to acquire image data and basic attribute data of a power transmission line to form a structured data package; A dimension decomposition and data preprocessing module receives the structured data package from the data acquisition module, maps it to a plurality of preset core dimensions, and generates standardized structured data after data cleaning; A visual language large model training and reasoning module performs model training iteration based on the standardized structured data, and uses the trained visual language large model to perform multi-modal feature extraction and multi-dimensional decision analysis, and outputs a power transmission line mechanical threat level; A result output and early warning disposal module receives the threat level determination result, and performs visual display, hierarchical early warning push, and full-link data storage.

2. The system for automated determination of the level of mechanical threat to the power line route according to claim 1, characterized in that, The data acquisition module comprises an image acquisition unit and a line basic information unit: The image acquisition unit is arranged on a power transmission line tower to capture line and surrounding scene images under different time periods and different weather conditions, and is internally provided with a wireless communication module; The line basic information unit is used to store basic attribute data of the power transmission line, and the basic attribute data comprises at least one of electrical properties, tower properties, conductor / ground wire properties, and section management information.

3. The system for automated determination of the level of mechanical threat to the power line route according to claim 2, characterized in that, The electrical properties include line voltage level, loop number, and phase sequence, the tower properties include tower number, model, and GPS precise coordinates, the conductor / ground wire properties include conductor model, splitting mode, and rated sag, and the section management information includes line belonging section, operation and maintenance unit, and contact information of the person in charge.

4. The system for automated determination of the level of mechanical threat to the power line route according to claim 1, characterized in that, The data package is packaged in JSON format, and the specific fields include tower_id, image_data, voltage_level, tower_type, location, and timestamp.

5. The system for automated determination of the level of mechanical threat to the power line route according to claim 4, characterized in that, The core dimensions in the dimension decomposition and data preprocessing module include spatial positioning analysis, power facility positioning, target object positioning, and mechanical feature recognition.

6. The system for automated determination of mechanical threat level of power line route according to claim 1, characterized in that, The data cleaning includes at least one of excluding non-longitudinal power transmission line images and fuzzy images with pixel size below a preset threshold. The spatial positioning analysis dimension is used to output the relative position relationship between the target machine and the power transmission line protection zone, and the relative position relationship includes the protection zone, the near protection zone, and the outside of the protection zone. The power facility positioning dimension is used to identify and position the power transmission tower and line direction in the image to form a power system protection zone. The target object positioning dimension is used to detect external mechanical objects that interact with or threaten the power transmission line, and output object categories and bounding box coordinates. The mechanical feature recognition dimension is used to analyze the working state of the external mechanical object according to the recognition result of the target object positioning dimension. The visual language large model training and reasoning module comprises a model continuous training unit and a multi-modal reasoning unit: The model continuous training unit comprises a dataset management submodule, a model training submodule, a hyperparameter optimization submodule, and a model verification submodule; the dataset management submodule is used for uploading and labeling new training data, the model training submodule is used for providing different training tasks, the hyperparameter optimization submodule is used for determining an optimal parameter combination in an automated search manner, and the model verification submodule is used for evaluating the performance of the trained model on a reserved test set; The multi-modal reasoning unit is used to deploy the trained visual language large model, receive standardized structured data, combine the transmission line mechanical threat level determination standard, and output the threat level.

7. The system for automated determination of mechanical threat level of power line route according to claim 1, characterized in that, The result output and early warning disposal module comprises a visualization display unit, an early warning pushing unit, and a historical data storage unit; The visualization display unit is based on a GIS map and is used for color coding and rendering of the transmission network and threat level, with different threat levels corresponding to different color identifiers; the early warning pushing unit is used to trigger a hierarchical early warning mechanism according to the threat level, the hierarchical early warning mechanism comprising at least one of system record, SMS notification, APP pushing, voice call, emergency resource association, and disposal order; and the historical data storage unit is used to store a structured threat event archive formed by the collected data, dimension analysis result, threat level, and disposal result, and is configured with a data query engine and a report tool for multi-condition combined query and data export.

8. A method for automated determination of the threat level of a power line mechanism, characterized in that The method comprises the following steps: S1. Collecting image data and basic attribute data of the transmission line and associating to form a structured data package; S2. Mapping the structured data package to a plurality of preset core dimensions, converting unstructured image data into structured label information corresponding to each dimension, and generating standardized structured data after data cleaning; S3. Training a visual language large model based on the standardized structured data, performing multi-modal feature extraction and multi-dimensional decision analysis using the trained model, and outputting a mechanical threat level; S4. Visualizing the threat level, pushing a hierarchical early warning, and storing full-link data.

9. The method of claim 8, wherein the method further comprises: Step S1 further comprises: S11. The image acquisition unit automatically acquires panoramic images of the transmission line at a preset time interval, the preset time interval being configured as 5-15 minutes / time; the acquired image data is transmitted to the backend processing center in real time through a wireless communication module via a high-speed wireless network; and the line basic information unit synchronizes data with the power operation and maintenance management system in real time or quasi-real time through an API interface; S12. The data acquisition module associates the image data of the image acquisition unit with the basic attribute data of the line basic information unit to generate a structured data package.

10. The method of claim 8, wherein the method further comprises: During the data cleaning process of step S2, images of the transmission line that are not longitudinally oriented and images of mechanical objects with a pixel size less than 20x20 are removed. The threat level in step S4 includes A level, B level, C level, and D level; wherein, the D level threat only triggers system internal record; the C level threat triggers short message notification; the B level threat triggers triple early warning of short message, APP push, and voice call; the A level threat triggers the triple early warning, and generates a disposal order associated with the line surrounding emergency resources.

Images